Exchange Rate Movements and The Mexico-United States

EXCHANGE RATE MOVEMENTS AND TH E MEXICO-UNITED
STATES AGRI-FOOD TRADE: A COINTEGRATION ANALYSIS
A T hesis
Presented to
T he Faculty of Graduate Studies
of
T he University o f Guelph
by
JOSE LUIS JARAMILLO VILLANUEVA
In partial fulfilment of requirements

for the degree of
Doctor of Philosophy
May, 2006
Jose Luis Jaramillo-Villanueva, 2006
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Library and
Archives Canada
Bibliotheque et
Archives Canada
Published Heritage
Branch
Direction du
Patrimoine de I'edition
395 W ellington Street

Ottawa ON K1A 0N4
Canada
395, rue W ellington

Ottawa ON K1A 0N4
Canada
Your file Votre reference
ISBN: 978-0-494-20930-1
Our file Notre reference
ISBN: 978-0-494-20930-1
NOTICE:
The author has granted a non
exclusive license allowing Library
and Archives Canada to reproduce,
publish, archive, preserve, conserve,
communicate to the public by
telecommunication or on the Internet,
loan, distribute and sell theses
worldwide, for commercial or non
commercial purposes, in microform,
paper, electronic and/or any other
formats.
AVIS:
L'auteur a accorde une licence non exclusive
permettant a la Bibliotheque et Archives
Canada de reproduire, publier, archiver,
sauvegarder, conserver, transmettre au public
par telecommunication ou par I'lnternet, preter,
distribuer et vendre des theses partout dans
le monde, a des fins commerciales ou autres,
sur support microforme, papier, electronique
et/ou autres formats.
The author retains copyright

ownership and moral rights in
this thesis. Neither the thesis
nor substantial extracts from it
may be printed or otherwise
reproduced without the author's
permission.
L'auteur conserve la propriete du droit d'auteur

et des droits moraux qui protege cette these.
Ni la these ni des extraits substantiels de
celle-ci ne doivent etre imprimes ou autrement
reproduits sans son autorisation.
In compliance with the Canadian

Privacy Act some supporting
forms may have been removed
from this thesis.
Conformement a la loi canadienne

sur la protection de la vie privee,
quelques formulaires secondaires
ont ete enleves de cette these.
While these forms may be included

in the document page count,
their removal does not represent
any loss of content from the
thesis.
Bien que ces formulaires

aient inclus dans la pagination,
il n'y aura aucun contenu manquant.
i*i
Canada
A BSTRACT
EX C H A N G E RATE M O V EM EN TS AND M EX IC O -U N ITED STATES
A G RI-FO O D TRA D E: A C O IN TE G R A TIO N ANALYSIS
Jose Luis Jaramillo Villanueva

University of Guelph, May 2006
Advisor
Dr. Rakhal Sarker
The favourable Mexico-United States exchange rate has been cited as an important factor
for the growth in exports of agri-food products from Mexico to the United States. Some
analysts have also emphasized the importance of NAFTA and the exchange rate volatility
in the performance of the Mexico-US agri-food trade. An important policy question is the
extent to which changes in the Mexico-US exchange rate and its volatility have
contributed to the growth in agri-food trade between these two countries. The purpose of
this study is to quantify the effects of the Mexico-US exchange rate changes and its
volatility on Mexico-United States agri-food trade flows using the Maximum Likelihood
Cointegration analysis.
To guide the empirical analysis, an expected utility maximization model has been
developed in this study. The comparative statics results derived from this model show that
while changes in exchange rate have a positive effect on imports, exchange rate volatility
has a negative influence on trade flows. A set of reduced form equations was developed
and estimated to determine empirically the effects of changes in exchange rate, its
volatility, and other relevant factors on trade flows of tomato, maize, sorghum, and milk
powder from 1989 to 2004.
The results from cointegration analysis show that while changes in exchange rate
have a positive effect on trade flows, volatility of the exchange rate has a negative effect
on trade flows. The specification of the exchange rate volatility does matter because
different estimates are obtained when alternative measures of volatility are used. The
results indicate that the volatility measure generated from a GARCH (1,1) model provides
more consistent results in terms of signs and sizes of the estimated coefficients than those
from other volatility measures.
While VEC models provide results supporting the findings of the cointegration
results on the effects of exchange rate and its volatility on trade flows, the short-run
elasticities are smaller than the corresponding long-run elasticities. So, the results satisfy
the Le Chatelier Principle. The free trade agreement seems to have played a significant
role in shaping the trade flows of tomato and maize, but not in the case of sorghum and
milk powder.
A C K N O W LE D G EM E N TS
First of all I thank God for helping me to complete this dissertation. I would like to
extend sincere thanks to my wife Adriana Palacios and my daughter Adriana Jaramillo
who supported me emotionally and took care of most family issues during my doctoral
studies at the University of Guelph. Your endless assistance, love and care are very much
appreciated. Without your help, I could not have accomplished my degree. I would also like
to thank my parents, Roberto and Maria, for their love and teachings of being hard working
and perseverant to be successful in life.
I express my gratitude to my advisor, Dr. Rakhal Sarker, who was not only an
advisor, but also a friend to me. I thank him for always being attentive, for his guidance,
encouragement and support in carrying out different phases of my PhD studies at Guelph. I
would also like to extend special thanks to the members of my advisory committee, Dr.
John Cranfield, Dr. Maury Bredhal, and Dr. Thanasis Stengos for their helpful comments at
various stages of this research. Thanks Dr. to Alfons Weersink for chairing my final oral
examination. Also thanks to Dr. Karl Meilke and Dr. Stephen Clark for serving as internal
and external examiners.
I wish to thank my friend Juan Cabas, a fellow PhD student for helpful discussions
and support. Finally, I am very grateful to my sponsors, El Consejo Nacional de Ciencia y
Tecnologia (Conacyt) and my Institution, El Colegio de Postgraduados, for their generous
support without which it would not have been possible to complete my PhD at the
University of Guelph.
TABLE OF CONTENTS
Page
CHAPTER 1
INTRODUCTION
1.1 The Economic Problem
1.2 Economic Research Problem
1.3 Purpose
1.4 Specific Objectives
1.5 Hypothesis
1.6 Contributions of this Research
1.7 Outline of the Study
CHAPTER 2
LITERATURE REVIEW
11
2.1 Introduction
11
2.2 Policy Reform in Mexico and the Agrifood-Sector
11
2.2.1 Transition Agricultural Policy
13
2.2.2 The Mexican Agricultural and Trading Sector
15
2.3 Literature Review on Exchange Rate Movements and Trade

2.3.1 Theoretical Studies on Exchange Rate Volatility and Trade
19
20
2.3.1.1 Risk Aversion and Risk Neutrality
20
2.3.1.2 Non-Competitive Trading Firm
24
2.3.1.3 Forward Markets and International Trade
25
ii
Page
2.3.1.4 Time Horizon of the Model
26
2.3.2 Empirical Studies on Exchange Rate Movements on Agricultural Trade
26
2.3.3 Modeling the Exchange Rate Volatility Movement on Agricultural Trade
28
2.3.4 Modeling Exchange Rate Volatility in Agricultural Trade
32
2.3.5 Exchange Rate Treatment in the Economic Model
32
2.3.6 Empirical Studies on Exchange Rate Volatility and Trade
35
2.3.6.1 Specification of the Volatility Measure
35
2.3.6.2 Nominal or Real Measures of Exchange Rate Volatility
38
2.3.6.3 Aggregate, Bilateral or Sector Trade Data
40
2.3.6.4 Estimation Technique Used
43
2.3.7 Empirical Studies on Exchange Rate and Mexican Agricultural Trade
44
2.4 Summary
47
CHAPTER 3
CONCEPTUAL FRAMEWORK
49
3.1 Introduction
49
3.2 Introduction to Exchange Rate Economics
50
3.3 Theories of Exchange Rate Determination
52
3.3.1 Flexible-Price Monetary Model
52
3.3.2 Sticky-Price and Real Interest Differential Monetary Models
53
3.3.3 Portfolio Balance Model
54
3.4 International Trade Models
56
3.4.1 The Hecksher-Ohlin Model of Trade
iii
57
Page
3.4.2 The Armington Trade Model
58
3.4.3 The Gravity Model
60
3.5 Exchange Rate and International Trade
62
3.5.1 Graphical Representation of a Trade Model
63
3.5.2 Algebraic Formulation of the Trade Model
65
3.6 Characteristics of the Commodity Industries in Mexico
69
3.6.1 Fresh Vegetables Industry and Exports
70
3.6.2 Grain Industry and Agricultural Policies in Mexico
74
3.6.3 Milk Powder Production and Policy
79
3.7 Economic Framework
83
3.7.1 Measures of Risk Aversion
83
3.7.2 The Theory of Input Demand Under Uncertainty
84
3.7.2.1 The Supply Function
87
3.7.2.2 The Import Demand Function
88
3.7.3 Derivation of the Import Demand and Export Supply
89
3.7.4 Comparative Statics for the Model
93
3.7.4.1 The Respond of Demand Function to Changes in Price of Inputs
94
3.7.4.2 The Respond of Import Demand Function to Outputs Price and
96
Exchange Rate
3.7.4.3 The Respond of Prices and Output to Changes in Exchange Rate
3.7.5 Empirical Implementation
97
98
3.8 Empirical Estimation of Equations
100
3.9 Summary
108
iv
CHAPTER 4
Page
FORMULATION OF THE EMPIRICAL MODEL

4.1 Introduction
110
4.2 Unit Roots and Cointegration Analysis
112
4.3 Empirical Specification and Cointegration Analysis
121
4.4 Short-Run Dynamic Relationships an Error Correction Model (ECM)
130
4.5 Exchange Rate Volatility Measures
132
4.5.1 Modeling Exchange Rate Volatility with ARCH Model
134
4.5.2 Modeling Exchange Rate Volatility with Nonparametric Technique
141
4.5.3 Variance-Standard Deviation Volatility Measures
145
4.5.3.1 Moving Sample Standard Deviation of the Growth Rate of the
145
Exchange Rate
4.5.3.2 Absolute Percentage Change of the Exchange Rate
145
4.6 Data Description and Sources
146
4.7 Summary
149
CHAPTER 5
RESULTS OF UNIT TOOT TEST AND VOLATILITY MEASURE
5.1 Introduction
150
5.2 Unit Root Test for Individual Time Series
150
5.2.1 Selection of the Lag Length
152
5.2.1.1 Results of Unit Root Test for Macroeconomic Variables
152
5.2.1.2 Results of Unit Root Test for Tomato Models
154
Page
5.2.1.3 Results of Unit Root Test for Com and Sorghum Models
155
5.2.1.4 Results of Unit Root Test for M ilk Powder Models
155
5.3 Exchange Rate Volatility Results
157
5.3.1 Volatility Results from a Standard Deviation Models
157
5.3.2 Volatility Results from an ARCH and GARCH Model
159
5.3.3 Volatility Measure Results from a Non-Parametric Model
168
5.4 Summary
172
CHAPTER 6
MAXIMUM LIKELIHOOD COINTEGRATION RESULTS
6.1 Introduction
174
6.2 Econometric Modeling Strategy
174
6.3 Optimal Lag Length Selection Results
176
6.4 Maximum Likelihood Cointegration Results
181
6.4.1 Results of Cointegration for the Tomato Models
182
6.4.1.1 Cointegration Results for Tomato Model Using V I
182
6.4.1.2 Results of Cointegration for Tomato Model Using V2
185
6.4.1.3 Results of Cointegration for Tomato Using VG
187
6.4.1.4 Results of Cointegration for Tomato Using VN
188
6.4.1.5 Relationship between Tomato Models
189
6.4.2 Results of Cointegration for Maize Model
191
6.4.2.1 Cointegration Results Using Volatility VI
vi
191
Page
6.4.2.2 Results of Cointegration Analysis for Maize Using V2
192
6.4.2.3 Results of Cointegration for Maize Using VG
193
6.4.2.4 Results of Cointegration for Maize Using VN
193
6.4.2.5 Relationship between Maize Models
195
6.4.3 Results of Cointegration for Sorghum Model
195
6.4.3.1 Cointegration Results Using Volatility V I
195
6.4.3.2 Results of Cointegration Using V2
197
6.4.3.3 Results of Cointegration for Sorghum Using VG
198
6.4.3.4 Results of Cointegration for Sorghum Using VN
199
6.4.3.5 Relationship between Sorghum Models

6.4.4 Results of Cointegration for Milk Powder Models
200
2001
6.4.4.1 Cointegration Results for Milk Powder Using Volatility V I
201
6.4.4.2 Results of Cointegration for Milk Powder Using V2
203
6.4.4.3 Results of Cointegration for Milk Powder Using VG
204
6.4.4.4 Results of Cointegration for Milk Powder Using VN
206
6.4.4.5 Relationship between Milk Powder Models
207
6.4.5 Comparison of Long-Run Results
208
6.5 Testing Hypothesis on the Cointegration Vectors
211
6.5.1 Testing the Significance of Individual Coefficients
212
6.5.2 Testing the Significance of a Set of Coefficients
214
6.5.3 Test on the Equality of the Coefficients of Prices and Exchange Rate
215
6.6 Summary
217
vii
CHAPTER 7
Page
RESULTS OF THE VECTOR ERROR CORRECTION MODEL

7.1 Introduction
219
7.2 An Overview of the Vector Error Correction Model (VECM)
219
7.3 Results of the VECM: The US Import Demand of Fresh Tomatoes
223
7.3.1 Main Finding for Tomato Models

7.4 Results for the VECM: The US M exicos Imports Demand for Maize
7.4.1 Main Finding for Maize Models
228
230
235
7.5 Results of the VECM: The M exicos Imports Demand for Sorghum
7.5.1 Main Findings for Sorghum Models
7.6 Results of the VECM: The M exicos Import Demand for Milk Powder
7.6.1 Main Findings for Milk Powder Models
7.7 Hypothesis Testing
236
240
240
244
245
246
7.7.2 Testing the Significance of a set of Coefficient
247
7.7.3 Test of the Equality of the Coefficient of Foreign Prices and Exchange
248
Rate
7.8. Comparison de Short-Run Results
249
7.9 Summary
252
viii
CHAPTER 8
Page
CONCLUDING REMARKS
8.1 Introduction
254
8.2 Modeling Strategy
257
8.3 Major Findings
259
8.4 Policy Implications
264
8.5 Contributions of This Research
265
8.6 Limitations and Suggestions for Future Research
266
References
268
Appendices
285
ix
LIST OF TABLES
Page
Table 2.1
Mexican Total Exports and Imports by Economic Activity, (Million
12
US dollars)
Table 2.2
Mexican Agricultural Exports by M ain Commodities (Million US
17
Dollars)
Table 2.3
Mexican Agricultural Imports by Main Commodities (Million US
18
Dollars)
Table 3.1
Main Supplies of the Tomato US Market
74
Table 3.2
Description and Specification of Variables for Tomato Equation
102
Table 3.3
Description and Specification of Variables for Com Equation
104
Table 3.4
Description and Specification of Variables for Sorghum Equation
106
Table 3.5
Description andSpecification of Variables for M ilk Powder Equation 108
Table 4.1
Comparison of Johansen, ECM and Engle-Granger procedures for
122
Testing Cointegration
Table 4.2
Raw Data, Description and Sources
Table 5.1
Unit Root Test Results for US and Mexico Macroeconomic Variables 153
Table 5.2
Unit Root Test Results for the Tomato-Models
155
Table 5.3
Unit Root Test Results for the Com and Sorghum-Models
156
Table 5.4
Unit Root Test Results for the Milk Powder-Models
156
Table 5.5
Statistic to Test for ARCH Errors
161
Table 5.6
Results for the GARCH (1,1) Measure of The Volatility
164
Table 5.7
Results for the ARCH LM Test Using One Lag
167
Table 5.8
Results for the ARCH LM Test Using Sixteen Lags
167
Table 5.9
Results from OLS Regression to Get Residuals
169
Table 5.10
Results from an OLS Third Regression to Get Conditional Variance
171
Table 6.1
Optimal Lag Length Determination for Tomato Models
177
Table 6.2
Optimal Lag Length
178
Table 6.3
Optimal Lag Length Determination for Sorghum Models
179
Table 6.4
Optimal Lag Length Determination for Milk Powder Models
180
Determination for Maize Models
148
Page
Table 6.5
Summary of Long-Run Cointegration Results for Four Commodities
211
Using Different Exchange Rate Volatility Measures
Table 6.6
Lon-Run Test Results on Significance of Individual Coefficients on
213
Mexico-US Trade Flows

Table 6.7
Lon-Run Test Results on the Joint Significance of Exchange Rate
215
and Volatility of the Exchange Rate on Mexico-US Trade Flows

Table 6.8
Long-Run Test Results on Equality of Coefficients of Exchange Rate 216

and Foreign Price on Mexico-US Import Demand
Table 7.1
Error Correction Model Results for the US Tomato Import Demand
225
Table 7.2
Error Correction Model Results for the US Tomato Import Demand
226
Table 7.3
Error Correction Model Results for Mexicos Import Demand of
231
Maize
Table 7.4
Error Correction Model Results for the Mexicos Import Demand for 233
Maize
Table 7.5
Error Correction Model Results for the Mexicos Import Demand of
237
Sorghum
Table 7.6
Error Correction Model Results for the Mexicos Import Demand for 239
Sorghum
Table 7.7
Error Correction Model Results for Milk Powder Import Demand
242
Table 7.8
Error Correction Model Results for Milk Powder Import Demand
243
Table 7.9
Short-Run Test Results on Significance of Individual Coefficient on
246
Mexico US Trade Flows

Table 7.10
Short-Run Test Results on the Joint Significance of the Exchange
247
Rate and Volatility of the Exchange Rate on Mexico-US Trade Flows

Table 7.11
Short-Run Test Results on Equality of Coefficient of Exchange Rate
248
and Foreign Price on Mexico-US Import Demand

Table 7.12
Results of the VECM for Four Commodities using Different

Specifications of Volatility of the Exchange Rate
xi
251
LIST OF FIGURES
Page
Figure 2.1
International Trade Flows (Not-Oil) for Mexico (1985-2003)
12
Figure 3.1
Effects of Exchange Rate on a Particular Agricultural Commodity
64
Figure 3.2
Production-Marketing Process for Exports of Mexican Tomatoes
73
Figure 5.1
Exchange Rate Volatility Measures, Moving Average Standard
158
Deviation (V I) and Percentage Change of the Exchange Rate (V2)

Figure 5.2
Exchange Rate Data Expressed as First Difference of the Logarithm
163
of the Real Mexico Exchange Rate (direr)

Figure 5.3
Mexico-US Exchange Rate Volatility from GARCH (1,1)
165
Figure 5.4
Results of the Normality Test for GARCH (1,1) Model
166
Figure 5.5
Volatility from the Semi-Parametric Estimation (VN)
172
xii
CHAPTER 1
INTRODUCTION
Mexico is an important partner in the North American Free Trade Agreement (NAFTA).
However, unlike Canada and the United States, agriculture in Mexico is less developed.
Agricultural transformation is still underway and domestic production of major staples
such as white com and beans are still dominated by small and mid-sized farmers. The
agri-food sector employs about 25 percent of the Mexican population. About 50 percent
of farmers in Mexico produce for self-consumption and 80 percent of grain and oilseed
farmers possess less than 5 ha. (Yunes-Naude, 2002).
Since the early 1980s, Mexico has initiated significant re-orientation of various
economic sectors and has re-instrumented its development policies. Chronic imbalances
in public finances, growing threats of inflation and skyrocketing external debt during the
early 1980s forced the government of Mexico to initiate economic liberalization (Bank of
Mexico, 1996). This economic liberalization was reinforced by other developments. In
fact, the Mexican economy and its agricultural sector were influenced by four key
factors: (i) the unexpected devaluation of the Mexican peso in 1982; (ii) progressive
liberalization of trade with the accession of Mexico to the General Agreement on Trade
and Tariffs (GATT) in 1986; (iii) the conclusion and implementation of North American
Free Trade Agreement in 1994; and (iv) the Mexican peso crisis of 1995 (Mora-Flores,
et. al., 2002).
The decline in oil prices during the early 1980s coupled with rising foreign debt
eventually led to the Mexican peso crisis in 1982. This crisis forced the Mexican
Government to adopt a new development strategy focused on opening and liberalizing
most economic activities and trade. The new trade policy was implemented in 1985 and it
was reinforced institutionally in 1986 when Mexico formally joined the GATT as a
member (Bank of Mexico, 1996).
The importance of trade in the Mexican economy has been growing since it joined
the GATT in 1986. The trade sector received a significant boost in 1994 with the
successful implementation of NAFTA. The share of agri-food trade relative to total
agricultural supply in Mexico almost doubled from about 19 percent during 1990-93 to
35 percent during 1994-2002. Agricultural trade between Mexico and the United States
increased significantly since the implementation of NAFTA. W hile the volume of
Mexican exports of major fruits and vegetables grew by 75 percent to 100 percent under
the NAFTA, imports of grains such as com, soybean, wheat and sorghum increased by
over 80 percent (Yunez and Barceinas, 2002).
Due to the growing dependence on trade, the Mexican agri-food sector is
becoming more vulnerable to changes in exchange rate of Mexican peso relative to the
United States dollar. Also important in this context are the changes in exchange rate
volatility which have been documented by the Bank of Mexico since 1995 (Bazdrech,
2002).
The importance of the effects of exchange rate changes on agriculture and
agricultural trade was first emphasized by Schuh (1974) who argued that changes in
exchange rates have a large impact on the price received for farm products and the price
paid for imported tradable inputs. His analysis drew attention to the importance of an
overvalued exchange rate on agricultural exports from the United States, during the
1970s. A number of empirical studies measuring the effects of changes on exchange rate
in agricultural trade followed. Kost (1976) and Chambers and Just (1981) found that
exchange rate fluctuations have significant impact on exports and domestic use of
commodities in the United States. Coleman and Meilke (1988) found that the devaluation
of the Canadian dollar led to a significant increase in net exports of beef and a small
increase in net exports of pork from Canada to the United States.
Since the beginning of the floating exchange rate regime the volatility of the
exchange rate has increased and the potential impact of exchange rate volatility on trade
was recognized. Economic theory suggests that unexpected changes in the exchange rate
volatility adversely influence the decisions made by risk-averse commodity traders.
Ethier (1973) showed theoretically how uncertainty about an exchange rate would reduce
the volume of trade, and Hooper and Kohlhagen (1978) tested empirically the effects of
volatility on trade. The results, however, are mixed.
1.1 The Economic Problem

The majority view among economists in the early 1970s was that a floating exchange rate
system was the appropriate way to avoid exchange rate misalignments (i.e. the departure
of nominal exchange rates from their long-run equilibrium level or from the market
fundamentals). It is now recognized, about three decades later, that the new international
monetary system may have generated some new problems. Casual empirical observation
suggests that under the floating exchange rate system, the movement of nominal
exchange rates does not fully reflect the movements of economic fundamentals between
countries, especially, in the short-run (Frankel 1991). Deviation of nominal exchange
rates from monetary fundamentals has been substantial and persistent. It is believed that
while the misalignment problem has not been dramatically mitigated under the new
system, significant increases in volatility created additional problems (Bazdrech, 2002).
Since the economic and trade liberalization took place in the mid 1980s in Mexican
agricultural trade, both exports and imports have grown dramatically in terms of volume
and value.
It has been argued that exchange rate played an important role in the performance
of Mexican agricultural trade since the 1980s. Some studies of United States-Mexico
agricultural trade have generally emphasized the importance of NAFTA in the dramatic
increase in exports and imports (Rosenzweig, 1996; Schwentesius, et. al., 1996; and
USDA-ERS, 1999). Others have emphasized the role played by exchange rate changes in
explaining Mexican trade performance (Diaz-Garces, 2002; and Mora-Flores et. ah,
2002). Since the shift from the fixed to a floating exchange rate system in 1995, the
Mexico-United States exchange rate has also been characterized by unexpected periods
of calm followed by episodes of high volatility.
While the border between Mexico and its important trade partners, the United
States and Canada, became increasingly open due to NAFTA, Mexican agri-food trade
has also been influenced by changes in exchange rate and exchange rate volatility.
However, it is yet to be determined the extent to which NAFTA, the continuous
devaluation of the Mexican peso, and the exchange rate volatility contributed to the
growth in Mexican agri-food trade during the last decade.
Unexpected changes in exchange rates influence the decisions made by riskaverse traders and may act as an impediment to international trade (McKenzie 1998). In
the short-run, exchange rate risk may be hedged. In the long-run, however, international
traders are exposed to higher and possibly unhedgable exchange rate risk.
The available literature on Mexican exchange rate and agricultural trade
(Espinoza-Arellano 1998; Malaga et al 2001; Diaz-Garces, 2002; and Mora-Flores 2002)
provide evidence that changes in exchange rate do have significant effect on trade. This
is surprising given the gradual devaluation of Mexican peso since the 1980s, moreover,
no research has been conducted to investigate the effects of exchange rate volatility on
Mexican agri-food trade. This issue is also very important for Mexican traders, agri-food
policy makers and agri-food producers in Mexico since evidences exist that a floating
exchange rate regime has been characterized by unexpected changes in exchange rate,
which may have increased the levels of risk traders face (Bazdrech, 2002). If individuals
are risk-averse, they would be willing to incur an added cost to avoid this risk. So, if risk
is not hedged properly, it acts as an implicit cost.
A study on the extent to which changes in exchange rate and its volatility affect
Mexican
agricultural
trade
would
provide
valuable
information
to
producers,
international traders and policy makers in Mexico. They might use such information to
devise more appropriate marketing strategies to reduce risk and improve competitiveness
of Mexican agri-food products in the international market.
1.2 Economic Research Problem

The export oriented economic policy initiated by the Mexican government in the 1980s
and the trade liberalization process that has taken place since the accession of Mexico to
the GATT have placed Mexico in a highly competitive global market. The performance
of Mexico in this context has been remarkable not only in manufacturing but also in agri
food products.
Krueger (1999) and Konno et. al. (2003) investigated whether NAFTA
significantly influenced bilateral trade between the United States and Mexico. They
found that NAFTA had no additional impact on the long-run trade relationship between
these two countries because a gradual switching was in progress since 1985. They argued
that as market integration under the NAFTA continues, Mexican exports would become
even more sensitive to changes in economic conditions of the United States economy in
the future.
Since the end of the 1970s when the world markets became increasingly open,
several theoretical and empirical studies have focused on the effects of both changes in
exchange rate and changes in exchange rate volatility on agri-food trade flows.
Unfortunately the results have been mixed and controversial as both positive and negative
effects of exchange rate on trade have been found in the trade literature.
Empirical studies on the effects of exchange rate movements on Mexico-Unites
States agri-food trade have either theoretical or empirical limitations. Existing studies
considered just estimation of either an import demand or an export supply function to
determine the effects of exchange rate movements on the volume and the value of trade
flows and produced negative results. They focused on the short-run effects of exchange
rate on aggregate trade and ignored the effects of exchange rate volatility on trade. The
previous studies have ignored the time series properties of the macroeconomic data used
in the estimation. Therefore, the results from these studies could be spurious. In view of
these limitations, an attempt is made in this research to perform a more comprehensive
study of the effects of exchange rate changes and its variability on Mexico-United States
agri-food trade. In particular, this study deals with an investigation of the nature and
extent of the influence of changes in exchange rate and exchange rate volatility on
Mexican agri-food trade flows with the Unites States the last decade.
1.3 Purpose
The purpose of this study is to quantify the effects of changes in the Mexico-United
States exchange rate and its volatility on the Mexico-United States agri-food trade flows
using the Maximum Likelihood Cointegration analysis.
1.4 Specific Objectives
The specific objectives of this study are as follows:
1.
To develop an analytical framework capable of analyzing the effects of changes in

exchange rate and its volatility on the Mexico-Unites States agri-food trade paying
due attention to commodity specific institutional features.
2.
To determine the long-run and short-run effects of changes in exchange rate and its
volatility on Mexico-United States agri-food trade. The Johansens Maximum
Likelihood Cointegration Analysis is used to examine the long-term effects of
exchange rate variability on trade. However, the short-term variability is assessed
using the vector error correction model (ECM).
3.
To determine if real exchange rate volatility has an adverse effect on the volume of
the Mexico-Unites States agri-food trade flows using a set of four volatility
measures.
4.
To discuss the policy implications of the results for improved competitiveness of

Mexican agri-food trade.
1.5 Hypothesis
To accomplish the above objectives, the following set of hypothesis stated in null form
was tested in this study:
1.
The changes in Mexico-Unites States exchange rate have an insignificant effect on

the Mexico-Unites States agri-food trade flows in the long run.
2.
The changes in Mexico-Unites States exchange rate have an insignificant effect on

the Mexico-Unites States agri-food trade flows in the short-run.
3.
Exchange rate volatility has an insignificant effect on the volume of the MexicoUnites States agri-food trade flows in the long-run.
4.
Exchange rate volatility has an insignificant effect on the volume of the MexicoUnites States agri-food trade flows in the short-run.
5.
The coefficients of exchange rate and foreign price is equal (i.e., it is not
appropriate to use exchange rate as a separate variable in the trade model).

This research attempted to make disciplinary as well as policy contributions. First,
institutional features related to specific commodities are highlighted and incorporated
into a trade model. In doing so this research extends the model of Appelbaum and Kohli
(1997) by including Commodity-specific information into the traders profit function. A
set of comparative statics results is derived from the model to determine the direction of
the effect of changes in exchange rate and exchange rate volatility on trade flows. This
represents the analytical contribution of this research.
This research has used Johansen Maximum Likelihood Cointegration analysis to

generate long-run effects of exchange rate changes and its volatility on Mexican
agricultural trade flows of selected commodities. On the other hand, Error Correction
Model is used to generate relevant the short-run effects of these variables on trade flows.
This represents the empirical contribution of this research.
Since no consensus exist in the trade literature on the appropriate exchange rate
volatility measure to be included in a trade model, an attempt is made in this research to
identify the most appropriate volatility measure from a set of alternatives using different
econometric techniques. This also makes an important contribution to our existing stock
of knowledge.
Finally, generating valuable information about the extent to which exchange rate
movements have contributed to Mexican trade performance in the agri-food sector, this
study makes an important policy contribution. The stakeholders in the Mexican agri-food
and trade sector can use the results to better assess exchange rate risk and devise more
appropriate marketing strategies to deal with such risks. This is likely to improve the
competitiveness of Mexican agri-food products in the international market.
1.7 Outline of the Study

The study is divided into eight chapters. Chapter 2 provides background information
about the Mexican agri-food sector and its institutional features. It also provides a critical
assessment of the literature on exchange rate economics with especial emphasis on
exchange rate effects on agri-food trade. Chapter 3 provides a framework for analyzing
the effects of exchange rate changes on trade flows. It includes the theory of exchange
rate determination and the relationship between exchange rate and international trade.
The theoretical trade model developed in this research is also presented in this chapter.
Chapter 4 is an exposition of the empirical model and data issues. Specifically, the
equations that make up the econometric model, specifications of variables, sources of
data, procedure to test for the presence of unit root in each data series, the Johansens
Maximum Likelihood Cointegration approach and the specification of the Error
Correction Model (ECM) are all outlined in this chapter. The last section of this chapter
deals with the procedures followed to derive four alternative measures of exchange rate
volatility used in this study. Chapter 5 presents the empirical results of unit root tests
pertaining to individual series as well as the empirical results from the four different
approaches to estimate the volatility of the exchange rate. Chapter 6 presents and
discusses the results of the Maximum likelihood cointegration analysis. Chapter 7 deals
with the results of the short run dynamic relationships estimated from the Error
Correction Models. Chapter 8 provides a summary of the major findings of this study,
highlights some limitations and offer suggestions for further research in this area.
10
CHAPTER 2
LITERATURE REVIEW
2.1 Introduction
This chapter provides a brief overview of the Mexican economy and its agri-food sector
and a critical assessment of the relevant published literature on the effects of exchange
rate movements and trade flows. First a background of the most recent developments in
the Mexican economy, Mexican agri-food sector and trade policy is provided. This is
followed by a revision of the theoretical literature on the relationship between exchange
rate movements and trade flows. The third section deals with the empirical published
research on the effects of exchange rate movements on international trade flows. This
section is divided in two sub-sections: the effects of changes in exchange rate on
agricultural trade and the effect of changes in exchange rate volatility on trade flows. The
fourth section is devoted to studies of Mexico-United States agri-food trade. The chapter
is concluded with a summary of the main points and synthesis.
2.2 Policy Reform in Mexico and the Agri-food Sector

The overall economic and trade liberalization undertaken by the Mexican government has
its origin in the debt crisis of 1982. This crisis resulted in a depreciation of 95 percent of
the Mexican peso against the United States dollar during 1982-1983. It forced Mexican
Government to adopt a new development strategy focused on opening and liberalizing
most economic activities including trade. The new trade policy was implemented in early
1985 and reinforced institutionally in 1986 when Mexico joined the GATT as a member.
Additionally the Mexican government established an economic reform program to
11
maintain price stability and economic growth. This reform together with the above
changes positively affected the growth of the Mexican economy and the international
trade performance of Mexico (See figure 2.1).
Figure 2.1 International Trade Flows (Not-Oil) for Mexico (1985-2003)
35000 -i
30000 O 25000 ^
20000
15000 -
**
10000
5000 LO
CO
CD
CO
hCO
CO
CO
O)
CO
O
o>
*
05
CM
co
a>
0) 0
o
0
o
CM
cm
CM
CM
o o
o
o
Years
Total Exports
Total Imports
During the 1987-1994 period the exchange rate regime evolved from a fixed rate
until 1988, to a transitory crawling peg and eventually, from January 1991 to 1994, to the
adoption of a band that was widened gradually.
During the early 1990s large capital
inflows and financial liberalization took place in Mexico, but the current account deficit
grew quickly from 6 billion dollars in 1989 to 20 billions dollars in 1993 and the
overvaluation of the domestic currency, gave away to the balance of payments problem
and the financial crisis of 1994-1995.
12
2.2.1 Transition Agricultural Policy

As a consequence of the economic crisis of 1982 the Mexican Government initiated a
radical change in the direction of its development policies, from a strategy of import
substitution to a model of outward orientation with declining state interventions in the
economy. A phase-out of government interventions in agriculture started at the end of the
1980s and deepened during early 1990s with the implementation of the North American
Free Trade Agreement or NAFTA in 1994.
La Compania Nacional de Subsistencias Populares (CONASUPO) was the main
institution through which government interventions in the Mexican agricultural sector
took place. Before the reforms of the 1980s, the programs of CONASUPO covered most
of the grain and oilseed crops produced and consumed in Mexico. By supporting the
prices of these crops, by processing, storing, and distributing the crops and by regulating
trade through direct imports, CONASUPO exerted significant control over crucial
components of agri-food production, marketing and distribution.
Taylor et al (2004) stated that by 1996, most of the agencies and financial
activities of CONASUPO were dismantled or privatized, and by 2000, the liquidation of
CONASUPO was complete. To gradually substitute some of the CONASUPOs activities
in the early 1990s, a marketing agency called Apoyos y Servicios a la Comercializacion
Agropecuaria (ASERCA) was created. The operations of ASERCA are directed towards
marketing of basic crops, but the agency does not buy or store commodities, as
CONASUPO did. Another important function of ASERCA is that it operates the
Procampo program through which Mexican Government transfer direct cash income to
Mexican farmers (PROCAMPO).
13
A major reform in Mexican state intervention in staple production was the

elimination of guaranteed prices that CONASUPO traditionally awarded to the producers
of basic crops. Starting in 1995, the government took further steps towards a more
liberalized food chain that led to the final decision to liquidate CONASUPO in 2000.
After these changes, the PROCAMPO program, a decoupled income support program, is
the only income support program available for farmers producing basic crops. The
purpose of this program is to facilitate the transition of farmers from price supports to a
freer and more open international market. PROCAMPO is planned to last until 2008,
when full liberalization under NAFTA will be reached.
In addition to ASERCA and PROCAMPO, in 1995 the Mexican Government
created Alliance for the Countryside. The main objective of this initiative is to increase
agricultural productivity by making additional capitals available for farmers which they
can invest in sanitary and other investment projects. A purpose of the Alliance is to
promote farming efficiency through crop substitution (mainly from basic crops to fruits
and vegetables) for farmers who have a potential comparative advantage in producing
such crops in the context of an open economy.
As part of the policy reform, the Mexican government signed NAFTA which
came into effect in 1994. Under NAFTA, some agricultural commodities were liberalized
in January 1994; others (the ones considered politically more sensitive by the signing
governments) were subject to a process of year-to-year liberalization, so that full free
trade was reached for some in January 2003. For a few commodities will be full
liberalization attained in January 2008. For the latter group of commodities, tariff rate
quotas (TRQs) and/or seasonal tariffs are used to protect domestic producers. While
14
Mexico can impose TRQs on the imports of barley, dry edible beans, maize and
powdered milk, this right was never exercised for maize. The United States imposed
seasonal tariffs as well as TRQs for several fresh vegetables and fruits imported from
Mexico. Quota levels were established based on 1989-91 trade flows between Mexico
and its two North American partners. In 1994, the TRQs were set at 2,500,000 metric
tons (Mts.) for United States maize and 1,000 Mts. for Canadian maize. Although the
above-quota base or consolidated tariff on maize from both countries was fixed at 215
percent (or 206.4 US$/Mt.), Mexico did not impose any TRQ on maize imported from
the United States.
2.2.2 The Mexican Agricultural and Trading Sector

The agricultural sector is an important component of the economy in Mexico, not only in
terms of employment, income generation and supply of food, but also in terms of its
contribution to the national GDP. The agricultural sector employs about 25 percent of the
Mexican labour force and contributes 5 percent to total national GDP. As a consequence
of the recurrent economic instability and devaluations, the agricultural sector was
severely harmed, public investment in infrastructure declined sharply, subsidies to
agricultural inputs were reduced for most group of crops and guaranteed prices for most
commodities were eliminated (Valdes 1993).
With major domestic policy reforms, such as the elimination of all guaranteed
prices for commodities, the promotion of exports and affective trade liberalization,
Mexico faces a global market in which trade pattern will be shaped largely by
fundamentals affecting the comparative advantage of the country. In particular, the
15
movement of the exchange rate will have a strong influence on trade outcomes. After the
trade liberalization, the participation of Mexico in the international market has increased
remarkably, not only in manufactured products but also on agricultural commodities
(Table 2.1).
Table 2.1 Mexican Exports and Imports by Economic Activity (Million US dollars)
1992
1994
1995
1998
2002
1985
1990
2 000
Exports
Agriculture and Forestry
1184.5
1720.7
1679.3
2220.9
3323.4
3435.7
3615.4
3285.6
Cattle and Apiculture
224.4
441.7
433
457.3
692.8
360.9
601.6
606.3
Mining
13818.9
9537.5
7775.7
6994.4
7974.8
6865.4
15427.4
13509.9
Manufactures
6427.9
14861.2
36168.8
51075.3
67382.9
106550.4 146497.4
143185.2
Not Classified prod.
8.1
277.4
138.6
134.3
167.6
247.4
183.7
Total FOB
21663.8
26838.5
46195.6
60882.2
79541.6
117459.4 166454.9 160770.4
313
Imports
Agriculture and Forestry
1296
1829.9
2402.1
2993.3
2478.8
4280.6
4304.9
4872
Cattle and Apiculture
311
241.1
456.3
378.1
164.9
492.2
493.9
503.4
Mining.
212.8
388.8
520
438
600.5
916
1325.6
1966.9
Manufactures.
12581.9
28523.1
58235.2
74424.8
67500
116431.2
165135.6
160613.9
Not Classified prod.
131.4
289
515.9
1111.6
79345.9
1708.9
3252.8
3196.7
713.7
72453
125373
174457.8 168678.7
Total FOB
14533.1
31271.9
Source: Ministry o f Economy-Bank o f Mexico
62129.3
As a consequence, Mexican trade flows have increased dramatically. The value

of exports rose by more that seven times from 21,663.8 million of dollars in 1985 to
160,770.4 million of dollars in 2002. Also imports rose by more that eleven times, from
14,533 million dollars to 168,678.7 millions in 2002. Trade balance became increasingly
negative from a surplus of 7,130.7 million dollars in 1985 to a deficit of 7,908.3 million
dollars in 2002.
Trade flows of agricultural commodities increase in both exports and imports in
the same direction. Exports increased by 2.7 times from 1408.9 million dollars to 3891.9
million. In the 1990s Mexico became a net importer of agricultural products from a
surplus of around 200 millions dollars to a deficit of 1356.4 million in 2002. An
16
exception was 1995 which Mexico reached a trade surplus of 844.6 millions. According
to Calva (1996), this surplus was due to depreciation of the Mexican peso, as it changed
from $Mex 3.5/$US in 1994 to $Mex 7.5/$US in 1995.
Historically most of the Mexican trade occurred with United States and the
volume of trade has increased since the economic liberalization of the 1990s. The share
of total exports to North America rose from 85.6 percent in 1993 to 90.7 percent in 2002,
while Mexican exports to other destinations decreased slightly during the 1990s. The
most important agricultural commodities exported by value are fresh vegetables, coffee,
dry chickpeas, strawberries, watermelon and other fresh fruits. Malaga et al, (2001)
reported that vegetables consist of about 15 percent of total agricultural production and
employ nearly 18 percent of the agricultural labour force. Fresh vegetables are by far the
leading agricultural exports accounting for 48 to 65 percent of the value of Mexican
agricultural exports. The case of tomato is remarkable since tomato exports rose by 252
percent between 1990 and 2002 (Table 2.2).
Table 2.2 Mexican Agricultural Exports by Main Commodities (Million US Dollars)

Commodity
1991
1992
1994
1995
1996
1998
2000
2001
2002
Coffee
368.1
258.1
359.7
706.2
677.2
617.3
613.7
211.2
176.7
Strawberries
20.7
12.2
31.5
43.9
53.4
68.5
50.3
42.6
79.7
Dry checkpeas
32
35.5
44.5
72.9
103.3
67.2
105.8
127.9
84.7
Tomatoes
261.8
166.8
394.4
585.7
539.9
589.3
462.3
532.5
661.1
Fresh Vegetables
489.3
550.9
689.4
929
742
1095.7
1352.2
1421.6
1348
W ater melon
142.3
89.4
89.1
114.4
128.3
135.6
156.7
157.9
129.1
Other Fresh Fruits 283.6
319.4
350.2
401.7
418.3
481
486.3
486.9
530
Others
1432.3
1958.8
2853.8
2662.4
3054.6
3227.3
2980.6
3009.3
Total
1876.8 1679.3 2220.9 3323.4
Source: Secretaria de Economia-Bank of Mexico
3197.3
3435.7
3615.4
3325.9
3408.1
1597.8
17
In case of imports, Mexico decreased their imports in term of value from United
States from 71 percent in 1993 to 66 percent in 2002. Imports from the rest of the world
increased slightly. Fresh vegetables, live-cattle, dry chick peas, avocados, tobacco, coffee
and sugar account for about 85 percent of total agricultural exports in the 1990s, while
com, wheat, feed grains, potatoes, cotton seed and milk powder accounted for about 80
percent of total agricultural imports into Mexico. It is worth noting the remarkable
increase in imports of com and soybean since the 1990s. Imports of com increased from
178.5 million dollars in 1991 to 840 million in 2002, while the imports of soybean
increased from 348 million in 1991 to 891.8 millions in 2002. These two commodities
accounted for 35% of total agricultural imports into Mexico during the 1991-2002 period
(Table 2.3).
Table 2.3 Mexican Agricultural Imports by Main Commodities (Million US Dollars)
Commodity
1990
1992
1994
1995
1996
1998
2000
2001
2002
Com
426.1
155.7
337.3
361.0
1051.0
610.2
539.0
628.2
622.5
Soybean
217.4
512.1
640.4
542.3
897.5
861.4
782.9
851.2
891.8
Wheat
46.2
163.5
189.2
217.1
427.1
339.4
333.6
422.6
467.4
Sorghum
331.4
542.1
394.8
254.3
331.2
348.9
469.4
514.5
506.8
Milk Powder
295.4
271.3
194.7
211.4
299.5
166.4
233.0
308.6
200.0
Fresh Meat
69.9
236.0
269.8
81.28
144.8
446.1
712.9
831.6
953.3
Potatoes
2.06
6.50
16.42
13.12
15.10
26.27
43.68
48.89
57.79
Cotton Seed
8.51
34.68
31.61
19.39
23.03
29.71
46.67
50.31
44.16
2828.6
3161.6
3656.2
3743.9
Sub Total
1397.3
1922.0 2074.4
1700.1
3189.6
Source: Mexican Ministry of Economy and Bank of Mexico
Since 1994 Mexico adopted a floating exchange rate system and peso depreciated
significantly against the United States dollar. It has been argued in different studies
(Malaga et al, 2001; Mora-Flores, 2002) that depreciation of the Mexican peso has
18
contributed to the Mexican trade performance. Schwentesius and Gomez (1996) pointed
out that in Mexico the performance of agricultural trade can be better explained
considering the effects of devaluation on production cost and domestic and international
commodity price level. Furthermore, the agricultural imports of grains, meats and dairy
products increased in volume and value after 1995.
2.3 Literature Review on Exchange Rate Movements and Trade

The shift from a fixed to a floating exchange rate system in the early 1970s by major
trading nations signalled the beginning of a new debate in international economics. Some
economists believed that this new system, characterized by high level of exchange rate
volatility, has had a negative effect on international trade flows, and so, an adverse effect
on global welfare. Empirical observations show that daily or monthly nominal exchange
rate movements have become much more volatile since 1973. Liang (1998) examined the
volatility of the real exchange rate using alternative data sets and different econometric
methods over the period 1880-1997. She found strong support in favour of the hypothesis
of non-neutrality of nominal exchange rate regime on real exchange rate volatility. That
is, flexible exchange rate periods have been associated with higher real exchange rate
volatility than the fixed exchange rate period.
While there is a general agreement among economists that flexible exchange rate
period has been characterized by a high level of exchange rate volatility, the theoretical
and empirical research so far has provided contradictory evidence related to the effects of
exchange rate volatility on international trade flows (Cho, et al. 2002). Studies by
Cushman (1983); Kenen and Rodrick (1986); Akhtar and Spencer-Hilton (1984);
19
Thursby and Thursby (1987); Cushman (1988); Anderson and Garcia (1989); Lastrapes
and Koray (1990); Chowdhury (1993); Qian and Varangis (1994); and Cho et al, (2002)
have found a negative relationship between exchange rate volatility and international
trade flows. On the other hand, studies by Franke (1991), Asseery and Peel (1991) and
Kroner and Lastrapes (1993) suggest that a positive relationship exists between exchange
rate volatility and trade flows. Controversial empirical findings of the effects of exchange
rate volatility on trade may be associated with the estimation method employed, the use
of nominal versus real exchange rates, and the specification of the volatility measures
used in these studies.
2.3.1 Theoretical Studies on Exchange Rate Volatility and Trade

2.3.1.1 Risk Aversion and Risk Neutrality
The negative hypothesis that exchange rate risk could reduce trade flows is based on the
theoretical notion that risk-averse exporters would reduce their output when faced with an
increase in exchange rate risk. This view was supported formally in the literature by
Ethier (1973); Clark (1973); Kawai (1983); Brodsky (1984); De Grauwe (1988); and
Demers (1991).
Clark (1973) analyzed a model of an exporting firm with no hedging possibilities
and fixed production that invoices its exports in the currency of the importer and thus
faces an exchange rate risk. An exporting risk-averse firm will produce less than a risk
neutral exporter, given the same market price, when there is a rise in the exchange rate
variability. To produce the same output, a risk-averse firm must receive a price higher
20
than a risk-neutral firm. As a consequence, the export supply curve will shift up and to
the left and this movement implies a reduction in the volume of trade.
Ethier (1973) found support for the negative effect of exchange rate volatility on
trade flows. He examined the effect of the forward exchange rate on international trade
and the relationship between the market for goods and the forward exchange market.
Ethier represented the exchange rate risk as the standard deviation of the spot exchange
rate and modeled the decision of a risk-averse importers regarding both the amount of
forward exchange rate cover and the volume of goods to be imported. He argued that (i)
to the extent that firms know how their revenues depend on the future exchange rate;
exchange rate uncertainty influences only the degree of forward cover and not the level of
trade; (ii) the pattern of trade will be inefficient to the extent that the forward rate does
not accurately reflect traders expectations concerning the future behaviour of the spot
rate; and, (iii) uncertainty about how the firms revenue depends upon the future
exchange rate will cause the volume of trade to become sensitive to exchange
uncertainty, which reduces the level of trade and will increase the terms of the trade of
expected profit for a reduction in risk.
While the Ethier (1973) and Clark (1973) models are static in nature and assume
fixed production, Kawai (1983) studied the behaviour of a competitive firm in a dynamic
context. The firm is engaged in domestic production and foreign trade and financial
transactions in domestic and foreign currencies. Since the firm faces uncertainty about
the future exchange rates and future prices of output and inputs, it attempts to hedge
against exchange rate risk if it is risk-averse. He argued that a risk-averse firm hedges in
the forward exchange market against not only the exchange risk but price risk as well. If
21
the forward exchange market is unavailable, the firm can still use the international
financial market for hedging. If the firms expectations about the future spot exchange
rate are different with respect to the forward exchange rate or interest rate parities, its
decisions on production and investment becomes inefficient. If the firm has access
neither to the forward exchange market nor the international financial market, an increase
in exchange rate uncertainty will curtail the firm s optimal investment and foreign trade
in the long run.
Demers (1991) showed basically the same results as Ethier (1973) but instead of
risk-averse traders, he assumed a risk neutral competitive firm which is uncertain about
its demand but has prior beliefs which it updates each period by Bayes law after
receiving information signals. He builds upon Arrow (1970) by demonstrating that the
irreversibility of investment in physical capital in conjunction with the anticipation of
receiving information and of learning the true state of demand lead to (i) lower
investment levels than otherwise since the firm can not disinvest if market conditions turn
out to be less favourable than anticipated; (ii) a time varying risk premium, or marginal
adjustment cost; (iii) a gradual adjustment of the capital stock to the desired level. He
argued that in an uncertain world the irreversibility of investment in physical capital lead
to reduced production levels and so quantity traded over time.
Frankel (1991) also analyzed the exporting strategy of a risk neutral firm. The
firm was assumed to operate in a monopolistically competitive environment and
maximize the net present value of expected cash flows from exports, which was specified
as an increasing function of the real exchange rate. The export strategy was determined
by the transaction costs incurred whereby the firm weighs the entry/exit costs associated
22
with entering/abandoning a foreign market against the profits/losses created by exports.

When the cash flow function is convex in the exchange rate, the present value of cash
flow grows faster than that of entry and exit costs and the firm benefits from increased
exchange rate volatility. In this scenario, the model predicts that firms will enter a market
sooner and exit later when exchange rate volatility increases and that the number of
trading firms will also increase.
De Grauwe (1988) captures the essential ambiguity of the debate by modeling a
competitive producer who must decide between selling in the domestic or foreign
markets. Both domestic and foreign prices are fixed and so the only source of risk for this
firm is the local currency price of exports. Given these assumptions, the response of this
producer to an increase in exchange rate risk, depends on either the expected marginal
utility of export income or the concavity of the exchange rate function. Where producers
exhibit only a slight degree of risk aversion they will produce less. For exporters, the
higher exchange rate risk reduces the expected marginal utility of export revenues.
However, where producers are extremely risk-averse, they will worry about the worst
possible outcome. This means that an increase in exchange rate risk will raise the
expected marginal utility of export revenue steam. This model does not allow the
diversification of exchange rate risk. De Grauwe concedes that the introduction of a
capital market would enrich the analysis, but not remove the basic ambiguity identified.
Dellas and Ben-Zion (1993) provide additional evidence as to the uncertain nature
of the impact of volatility using a standards asset portfolio model. However, unlike most
other studies, which focus on the variance of the exchange rate to represent volatility, the
authors specify unanticipated fluctuations in the exchange rate as constituting risk. The
23
asset in their model is a nominal unhedged trade contract, which contains a risk element
in the form of an exposure to changes in the exchange rate. Their analysis examines a
single individual who consumes as well as imports and exports both available goods (no
production is incorporated in the model). The results of this model indicate that an
increase in the riskiness of the return on these assets depends on the nature of the risk
aversion parameter assumed. Assuming a convex function, an increase in risk increases
the level of exports. This result is found to be robust to the presence of a forward market
with non-zero transaction costs and the introduction of production.
2.3.1.2 Non-Competitive Trading Firm

Most theoretical models choose to focus on the decisions of a single trading firm.
However, Broil (1994) recognized the increasing importance of multinational trading
firms in the global trading environment and focused on the economic behavior of a riskaverse multinational firm that produces in a foreign country and sells that output abroad.
It was assumed that the multinational firm has monopoly power in the foreign market and
faces exchange rate uncertainty. Exchange rate risk in this model was specified as the
difference between the spot exchange rate and the expected exchange rate. Where that
exchange rate risk is not diversifiable, production is shown to decline in the foreign
country as a result of exchange rate uncertainty. However, where mature foreign currency
forward markets are accessible, the probability distribution of the spot exchange rate
bears no impact on the decision function of the multinational enterprise with respect to
foreign investment and foreign labor demand.
24
On the other hand W olf (1995) considered a utility maximizing and risk-averse
competitive import agent. This agent is assumed to face uncertainty about not only
changes in the exchange rate during the period in which decisions are made (measured as
the variance of the exchange rate) but also uncertainty about the prices of imported
commodities. Faced with a multiplicative risk structure, W olf proves explicitly that the
variance of the exchange rate has a negative effect on the level imports. It is also shown
that demand by the importing firm of inputs and hedging instruments is less compared to
derived demand when the input price is known.
In a similar fashion, Gagnon (1993) modeled the response function of traders who
buy goods in one country and sell in another to maximize discounted expected future
utility. These traders are assumed to be risk averse as well as face a convex cost structure
in response to adjusting the level of trade because of contracting and marketing costs. In
this model, both exchange rate variability (measured as the unconditional variance) and
exchange rate uncertainty (measured as the conditional variance) are shown to reduce
trade flows. If adjustment is assumed costless, then the former relationship is lost but the
latter remains with no changes.
2.3.1.3 Forward Markets and International Trade

De Grauwe (1988) hypothesized that capital markets would not change the basic
ambiguity about the impact of volatility on trade flows. Other authors formally tested this
notion. Viaene and De Vries (1992) formally incorporated a mature forward market into
their analysis. Assuming that individual merchants operate in a competitive world
market, Viaene and De Vries (1992) begin by proving the preposition that in the absence
25
of forward markets, an increase in exchange rate volatility reduces both imports and
exports. A forward market was then incorporated into the equations of this model. The
results show that the effect of a change in exchange rate volatility on imports and exports
are opposite to each other. This result is attributed to the fact that importers and exporters
are by definition, on opposite sides of the forward market as is their exposure toward
movements in the exchange rate. Hence, increased exchange rate volatility may enhance
or reduce trade flows depending on the net currency position of that country.
2.3.1.4 Time Horizon of the Model

Sercu et al (1992) introduce friction to the decision function of a producer, rather than
entry and exit costs as in Franke (1991), standard duties and transports costs are assumed.
Further, a short-term market period perspective is adopted as risk neutral firms produce
in the current period for sales in the next period. The decision of whether or not to export
is made in the next period once the end of period exchange rate is known. In this setting,
an increase in exchange rate volatility measured as an increase in the conditional standard
deviation of the exchange rate is shown to increase production, prices and trade whether
the producer is competitive or a monopolist.
2.3.2. Empirical Studies on Exchange Rate Movements and Agricultural Trade

The exchange rate is the price of one currency in terms of another. Exchange rates are
important economic variables because they are used to convert foreign prices into
domestic currency and vice versa. These relative prices determine which goods are traded
and where they are shipped or sourced. Being able to convert one currency into another at
26
the prevailing exchange rate is crucial to international business and decision-making. The
difference in relative prices determines the flow of agricultural products and the patterns
of trade. Based on trade theory, the real exchange rate is viewed as the relative price of
traded to non-traded goods. Real exchange rate movements accommodate changes in
technology, income levels, or borrowing from abroad that require either higher or lower
relative price of non-traded goods to clear those markets.
Rosson (2001) points out that proponents of the current flexible exchange rate
system argue that since floating exchange rates are allowed to fluctuate according to the
supply and demand with a minimum of government intervention then they respond
quickly to changes in market conditions. Market disequilibria are less likely to occur
under a floating rate system and business and resource allocation decisions are made in a
timely, efficient manner. Flexible exchange rate regimes have been criticized because
they may be subject to destabilizing speculation. Some analysts argue that floating
exchange rates result in additional risk and uncertainty for market participants and that
while most of this risk can be hedged transactions costs are higher than under a fixed
exchange rate system.
The trade literature related to the effects of exchange rate movements on
agricultural trade focused on whether exchange rate matter to agricultural trade flows,
and if it matters, what is the magnitude of this effect on agricultural trade. Among the
proponents that exchange rates do not matter for trade performance are Batten and
Belongia (1986), Fuller et al. (1992), Bessler (1986), and Greenes (1975). Contrary to
this view, others have argued that exchange rate does matter and that it is a key variable
in explaining trade performance. These studies include Cho et al. (2002), Anderson and
27
Garcia (1989), Chambers and Just (1981), Thraen, et al. (1992), Carter, et al. (1990),
Mora-Flores (2002), Collins, et al. (1980), and Guzel and Kulshreshtha (1995). M ost of
the research supporting that exchange rate matter has found a positive effect of exchange
rate on trade flows. As the theoretical and empirical literature on this topic evolved a
number of issues have come to the forefront (see appendix 1 to 4).
Schuh (1974) called attention to the relationship between exchange rate and
agricultural products. He argued that the overvalued dollar caused the decline in
agricultural exports due to their relative expense in other countries. The overvalued dollar
led to depressed prices and lower farm profits causing an under valuation of farm
resources and excess supply of output. He argued that changes in monetary policy
induced international capital flows, which in turn caused changes in the value of the
dollar. These changes in the value of the dollar had an impact on the level of imports and
exports. The net result of these changes was that agriculture, as an export-oriented
industry, must bear the majority of the burden caused by changes in monetary and fiscal
policies.
2.3.3 Modeling the Exchange Rate Movements on Agricultural Trade

Order and Fackler (1986) pointed out that the earliest attempts to evaluate Schuhs
argument empirically were conducted in a partial equilibrium modeling framework and
focused on assessing the elasticities of price transmission and of supply and demand that
affected trade. The partial equilibrium assessments were able to attribute only a small part
of the substantial relative price movements in the early 1970s to the exchange rate; results
which are consistent with Schuhs long-run claim. Models that offered a more complete
28
linkage of real exchange rate movements to underlying causes accounted for market
equilibrium of traded and non-traded goods and provided somewhat more support for real
exchange rate effects on agriculture.
Kost (1976) developed a simple but consistent theoretical framework to assess the
effects of exchange rate changes (devaluation or appreciation) of a countrys currency on
commodity trade. He traces the effects of changes in exchange rates on commodity
production, consumption, trade levels, and price for any two trading partners. Examining
the effects of a devaluation of the exporters currency which has the same effects as that
of an appreciation of the importers currency, Kost stated that either of these actions
cause an increase in quantity exported and prices in the exporter country, which causes an
increase in production and decrease in consumption in the exporting country.
Kost points out that there is an upper limit on how much price and quantity can
change in response to an exchange rate change. The maximum that price and quantity can
change is by the same percentage of the exchange rate change, and the price maximum
would only occur if the export supply curve is perfectly inelastic, while the quantity
maximum would ocuur only if the export supply curve in perfectly elastic. The impacts
of an exchange rate change on imports and exports depend on the magnitude of the
exchange rate change.
Vellianitis-Fidas (1976) tested the hypothesis that exchange rate changes have a
significant effect on the demand for Unites States agricultural exports. To test her
hypothesis, a cross-sectional study of the demand for Unites States agricultural exports
was performed for the period 1971-1973. Additionally, the exchange rate changes in
importing countries were examined to determine if changes in these rates explained
29
variations in imports over time, both from the Unites States and the world in the period
1954-1969. Her results supported the hypothesis that special characteristics of the
agricultural sector negate the effect of exchange rate changes in demand for Unites States
agricultural exports. Exchange rate was not significant in the wheat equation and not
important in the com and soybean equations. Almost none of the variation in changes in
quantities exported for 1971 to 1972 and 1972 to 1973 was explained by the variation in
the exchange rates.
An alternative approach to empirical modeling adopted the time-series analysis to
seek causal relationships and dynamic impacts from monetary indicators to agriculture.
Sims (1980) pioneered the use of small dynamic models without too many a priori
restrictions as an alternative to over-identified structures imposed either by traditional
Keynesians or by the neoclassical rational expectations models. W hile it was appealing to
think that monetary effects on agricultural prices and trade could be measured easily in
small dynamic models where they were important, it turned out to be a difficult task.
Orden and Fackler (1986), using the VAR approach, detected little effect of
money supply on real Unites States agricultural prices or export values. Shocks to
financial market variables such as a short-term interest rate or the exchange rate had
larger impacts. These shocks explained about 20 percent of forecast error variance for
exports and 10 percent for real agricultural prices one year ahead, and over 50 percent
and 25 percent respectively, for a three-year forecast horizons. An increase in the interest
rate or appreciation of the dollar had a depressing effect on agriculture.
Dorfman and Lastrapes (1996) implemented time-series methods to measure the
effects of monetary variables on agriculture. They imposed the long-run restriction of
30
monetary neutrality to identify policy shocks, and used Bayesian techniques to

investigate sensitivity of their results to various model specifications. Their identifying
restriction ensures that the price level, sectoral prices and money rise equiproportionately in the long run. This is an appealing constraint. They also found plausible
short-run monetary policy impacts on interest rates, output and the price level. While
monetary shocks raised real agricultural prices in the short run, but explaind only a small
fraction of crop and livestock relative price variability.
Fabiosa (2002) modeled a representative Canadian pork exporter using a
structural YAR approach to examine the impacts of the exchange rate volatility on
Canadian exports to United States and Japan. The pork export supply equation is
expressed as a function of the expected level of real exchange rate and a time-varying
variance of real exchange rate. An AR(p) model was used to represent the expected real
exchange rate, and a GARCH(p, q) model is used to generate the time-varying variance
which was used as a measure of the real exchange rate volatility. Fabiosa (2002) favoured
the use of a structural VAR approach arguing that in contrast with the standard VAR, the
former allows one to build a theoretical structure to explain the relationships of the
relevant variables. Both the contemporaneous relationships of variables and the dynamics
are captured in a structural model. The results suggest that the domestic pork price in
Canada has a negative impact on pork trade while the U.S. pork price has a positive
effect. The level of the Canadian exchange rate relative to the U.S. dollar has a positive
impact on pork exports. The volatility of the Canadian exchange rate has a negative effect
on pork trade.
31
2.3.4 Modelling Exchange Rate Volatility in Agricultural Trade.

Ian Sheldon (2003) analyzed the effect of medium to long-run exchange rate uncertainty
on agricultural trade and compared it to the impact in other sectors. A gravity model is
applied to bilateral trade flows for 10 developed countries for the period 1974-1995,
allowing various cross-country determinants of trade such income, distance, and
membership of customs unions, common borders and exchange rate uncertainty in the
model. Sheldon emphasized that in studying the effects of exchange rate uncertainty on
trade, a critical issue is to distinguish between short- and medium/long-term changes in
exchange rates. A common argument against using short-run variability is that exchange
rate risk can be readily, at some cost, hedged with appropriate short-term risk
management instruments. Sheldon concluded that compared to trade in other sectors,
agricultural trade has been more adversely affected by medium to long-run uncertainty in
real exchange rates. Anderson and Garcia (1989) and Cho et al (2002) are studies that
have included a volatility measure in their trade model. They found that exchange rate
volatility exerts a nnegative effect on agricultural trade and that disaggregated trade is
more susceptible to uncertainty than the aggregate trade.
2.3.5 Exchange Rate Treatment in the Economic Model

Chambers and Just (1979) noted that while some research found that exchange rate plays
a role in agricultural exports, others found that the exchange rate has relatively small
impact on the agriculture sector of the economy. They critiqued the treatment of
exchange rates in agricultural trade models. They argued that the approach to deal with
the exchange rate in agricultural models has been overly restrictive.
32
Collins, Meyers, and Bredahl (1980) used a simple analytical method to analyze
the impact of multiple exchange rate changes, rates of inflation, and trade restrictions on
real Unites States commodity prices. Their study included two steps: (i) an expression of
the short m n Unites States commodity price changes caused by both nominal exchange
rate changes and adjusted exchange rate and (ii) calculated annual changes in Unites
States prices of wheat, com, soybeans, and cotton attributed to exchange rate changes and
inflation rates in 1971-1977. They compared these estimated changes with observed
changes in prices to determine how large the exchange rate impact was on Unites States
agriculture. The authors noted that there were some other issues involved. The size of the
exchange rate impact depends on crop, year, country, governmental influence in markets,
elasticities, measured price variables and alternative prices considered. They argued that
if the exchange rate changes reflect only differential rates of inflation then under free
trade, nominal commodity prices change but the underlying supply and demand do not. If
the exchange rate is fixed, differential inflation rates cause supply and demand changes
and the use of nominal price insulation policies increase the impact of inflation and
exchange rate changes on Unites States export demand and real commodity prices
increase significantly.
Using monthly data for 1973-mid- 1980s, Chambers (1981) found some evidence
of a causal relationship between money supply and agricultural imports and exports,
while there was little evidence of a causal relationship between the interest rate and
agricultural exports and imports. He also tested the null hypothesis that money supply did
not cause the level of wheat exports for the period of 1892-1952 and found that there is
limited evidence of causality in this instance as well.
33
Batten and Belongia (1986) provided empirical support for the exchange rates do
not matter position. They argued that the real stimulus for export demand comes from
income enhancements in importing countries. Looking at a changing monetary policy,
they found that neither monetary policy nor a government deficit has an impact on the
value of the Unites States dollar. Focusing on the factors that affect changes in real
exchange rate, they failed to discover any evidence that monetary policy or budget
deficits have had effects on the real value of the dollar.
Schwartz (1986) compared the effects of changes in exchange rate in a simple
competitive versus a noncompetitive market for wheat. In the simple competitive case,
under a floating exchange rate, a change in exchange rate in one country will cause a
short run adjustment in price, output, trade flows, market share of exports and export
volume for two countries competing with one another. The country that experiences
depreciation in currency value will see a decrease in exports and the other country will
see an increase in exports. The more the exchange rate fluctuates, the more variable
short-run changes in domestic prices and trade shares are.
Bradshaw and Orden (1990) tested the Granger Causality of exchange rates on
agricultural prices and exports. They examined the impact of the real agricultural tradeweighted exchange rate for monthly forecasts of real cash prices and export sales
volumes comparing out-of-sample forecasting performance of univariate models to
bivariate model. Bradshaw and Orden found that their bivariate models outperformed the
univariate models in statistically significant ways. They argued that model specification
and the choice between in sample and out of sample tests are important in determining
Granger Causality among the exchange rate, prices and exports sales.
34
Dorfman and Lastrapes (1996) disaggregated Unites States agricultural data into
crop and livestock for the period 1952 to 1993 and used a Bayesian approach to test the
responses of agricultural prices to money supply shocks. The results suggest short to
medium-mn benefits to both sectors from an expansionary monetary policy.
2.3.6. Empirical Studies on Exchange Rate Volatility and Trade

While the theoretical literature on exchange rate volatility and trade support the
hypothesis that exchange rate uncertainty has an adverse effect on trade flows, the
empirical literature thus far, has uncovered both negative and positive effects of exchange
rate volatility on trade flows.
As the empirical literature on this topic has evolved, a number of critical issues
have come to the surface. The most important issue reported in the empirical research
relates to the measurement of exchange rate volatility and the focuses on the effects of
volatility in the short-run or in the long-run. Concerns have also been raised about the
nominal or real measure of exchange rate volatility, sample period, frequency of data,
aggregated versus disaggregated data, the nature of trade flows and the estimation
technique used. The following section reviews specification issues related to exchange
rate volatilities.
2.3.6.1 Specification of the Volatility Measure

It is generally agreed that uncertainty constitutes exchange rate risk. However, no
generally accepted measure to quantify such risk exists. Volatility of the exchange rate
has been specified in a number of ways in the existing literature. The published literature
35
reports nine different measures of volatility that have been empirically tested. Most
published research has tested a variance or standard deviation-based volatility measure,
specifically the moving average of the standard deviation (Cushman, 1983; Akhtar and
Spence-Hilton, 1984; Gotur, 1985; Bailey et al, 1987; Kenen and Rodrik, 1986; BiniSmaghi, 1991). Belanger et al (1992) proposed a non-parametric measure of volatility.
They argued that nonparametric measures are robust to assumptions regarding the
distribution of the data and offer additional insights than measures that assume Gaussian
errors. Recently ARCH and GARCH models have also been used to generate a measure
of exchange rate volatility (Mckenzie, 1998; Kroner and Lastrapes, 1989; Qian and
Varangis, 1994).
Akhtar and Spence-Hilton (1984) analyzed multilateral United States-Germany
trade over the floating rate period using a nominal volatility measure specified as the
standard deviation of daily exchange rate over a three month period. Their results reveal
that exchange rate risk reduces the volume of trade.
In contrast, some recent studies have rejected the hypothesis that exchange-rate
volatility has had a systematically adverse impact on trade. Gotur (1985) tests the
robustness of Akhtar and Hilton's empirical results and finds their methodology to be
flawed in that it included a nominal volatility measure. Gotur stated that over the medium
term, the real exchange rate is the more relevant measure because the effects of
uncertainty on a firms revenue and costs arise from fluctuations in the exchange rate are
likely to be offset in large part by movements in costs and prices.
Kroner and Lastrapes (1989) estimate the reduced form impact of exchange rate
volatility on international trade quantities and prices using a joint estimation technique in
36
the context of a parameterized model of conditional variance. The empirical model

allows joint estimation of the relationship between volatility and trade and how past
information is related to perceived volatility. The model imposes rationality on the
variance forecasts of market participants by estimating a multivariate GARCH model.
This model restricts the variance that affects trade to be the same as that generated by the
data. Their approach avoids the arbitrariness of the conventional tests by using the data to
specify the variance forecast model. Tests are performed for five industrialized countries
using monthly data. They found that the GARCH conditional variance has a statistically
significant impact on the reduced form equations for all countries. For most of the
countries, the magnitude of the effect is stronger for export prices than quantities. In
addition, the estimated magnitude of the impact of volatility on exports is not robust to
using the conventional estimation strategy.
Belanger et al (1992) examined United States-Canada quarterly trade flows
(1974-1987). They used a carefully selected set of data to ensure the compatibility of
price and quantity measures across borders, a level of disaggregation in the data to make
it possible to identify differences in attitudes towards risk or hedging habits between
trade sectors. The authors measured exchange rate risk by using the observed forecast
errors on the forward exchange market. Then, they used non-parametric instruments in
the estimation in order to take into account the pitfalls associated with the use of a proxy
for the risk variable. It is argued that a more appropriate measure should be based on the
observed forecast errors in the forward market. Because this measure rests on marketbased source of errors this approach is likely to be superior to measures based on the
specification of an ad hoc statistical model to mimic the forecasting decisions of the
37
agents. The results showed no stable economically and statistically significant negative
impact of exchange rate risk on Unites States imports from Canada. It was true for all the
sectors examined except for capital goods that accounts for only 12 percent of imports
from Canada. They results seem to challenge results found in some recent bilateral trade
studies such as Maskus (1987), Thursby and Thursby (1987), and Cushman (1988).
Peree and Steinherr (1989) estimated export equations on aggregate and bilateral
basis; the general functional form was quite traditional. They used data for the period
1960-1985 for five developed countries. Their paper focuses on the problem of
approximating
meaningfully
exchange
rate
uncertainty
without
tackling
other
shortcomings. In view of the interest in long run uncertainty they proposed a measure for
projections several years into the future. Variances over past periods are of very limited
relevance for appreciating uncertainty over periods of several years in the future.
Consequently, two measures of medium term exchange rate uncertainty were constructed.
One is a weighted function of the magnitude of past movements in nominal exchange
rates and the current deviation of the exchange rate from equilibrium. The second
depends on both the duration and the amplitude of misalignment from equilibrium
exchange rates. The empirical evidence suggests that when exchange rate uncertainty was
defined over a medium term period it does adversely affect trade flows of the country
under review, with the exception of the Unites States.
2.3.6.2 Nominal or Real Measures of Exchange Rate Volatility

One important question in the debate on the impact of exchange rate volatility is whether
it is real or nominal exchange rate volatility which enters into the decision making
38
process of the traders. Some of the theoretical and empirical research in this area focused
on the use of nominal exchange rate volatility included Ethier (1973), Hooper and
Kohlhagen (1978), Koray and Lastrapes (1989), Anderson and Garcia (1989), and
McKenzie and Brooks (1997), while others focused on the use of real measures of
exchange rate volatility (such as Cushman (1986), Caballero and Corbo (1989), Assery
and Peel (1991), Chowdhury (1993), Lastrapes and Koray (1990), McKenzie (1998), and
Fabiosa (2002)). Under both specifications negative and positive effects of exchange rate
volatility on international trade have been reported.
Qian and Varangis (1994) provided evidence which suggests that the nominal and
real exchange rates have moved closely together during the floating period. Hence, the
authors argued that the distinction between real or nominal volatility should make no
difference to the results derived. Some authors have tried to test this proposition by
retesting their equations using alternate volatility measures. Thursby and Thursby (1987)
find that their results for nominal exchange rate volatility are indistinguishable from those
based on real volatility.
Bini-Smaghi (1991) supports this view of nominal measure of exchange rate
volatility by arguing that risk should be regarded as nominal rather than real exchange
rate risk as the latter depends not only on the variance of the nominal exchange rate but
on that of relative prices. The proxy for exchange rate risk was based on the standard
deviation of the effective nominal exchange rate index for each quarter. In the estimation
procedure, the author included up to eight lag periods on both the real exchange rate and
the risk proxy. The results indicate that exchange rate variability did serve to reduce
39
German bilateral trade in general and also Unites States exports. Other volatility
measures were also tested and the results were found not to change significantly.
2.3.6.3 Aggregate, Bilateral or Sectoral Trade Data

An examination of the published literature suggests that most of the studies in this area
have chosen to analyze aggregate trade flow data. There are arguments supporting the use
of either aggregate or sectoral trade flows. McKenzie and Melbourne (1999) argued that
by using aggregate trade flow data one implicitly assumes that the impact of exchange
rate volatility is uniform across commodities.
Hooper and Kohlhagen (1978) made attempts to test the impact of exchange rate
volatility on trade flows and prices using quarterly multilateral and bilateral trade flow
data for G7 countries for the period 1965-1975. W hen aggregate trade data was the focus
of the model, no significant impact of volatility on volume or price was found. For
bilateral trade flows, the authors did report limited evidence of a relationship to price,
although the direction of this association was conflicting.
Klein (1990), Bini-Smaghi (1991) and McKenzie and Melbourne (1999) argued
strongly for sectorally disaggregated estimation of the trade-risk relationship and
demonstrate that desegregation uncovers significant intersectoral variation in the effect of
exchange rate volatility on trade flows. For example, some sectors, such as agriculture,
may be far more sensitive to exchange rate risk than other sectors (Pick 1990). A related
aggregation issue concerns the frequency of the data used in estimation. Due largely to
data limitations, most studies employ lower frequency, quarterly or annual, to examine
the trade and risk relationship (McKenzie and Melbourne 1999). However, temporal
40
aggregation necessarily dampens exchange rate variability, which may make identifying
any true trade-risk relationship more difficult (Wang and Barret 2002). Moreover, where
different sectors have different conventions for contracting and delivery or payment lags,
intersectoral and intertemporal aggregation together could necessarily mute real traderisk effects.
The literature indicated that many of the studies in this research area have
undertaken an examination of aggregate trade flows. These studies implicitly assume that
the impact of exchange rate volatility is uniform between countries and across
commodities both in terms of direction and magnitude. If this assumption is incorrect,
then, the examination of aggregate trade data is likely to dilute the true nature of the
relationship and lessen the probability of deriving a significant result. Recognising this
possibility researchers in recent years have adopted trade models which focus on
disaggregated trade data in the form of bilateral and sectoral trade flows.
In this spirit, Klein (1990) analyzed the effects of real exchange rate volatility on
specific categories of bilateral exports by the United States over the period 1978 to 1986.
The volatility measures used the standard deviation of the monthly percentage change in
the bilateral real exchange rate. The importance of the nature of cost and demand
functions in determining the effect of exchange rate volatility on trade suggests
undertaking an analysis disaggregated by industrial sector while differences across
destination markets in the behavior of exchange rate volatility calls for an analysis of
bilateral, as opposed to multilateral, trade flows. Kleins results supported the hypothesis
that volatility affects differentially by category of exports. It is found that real exchange
rate volatility significantly affects the value of United States exports for six of the nine
41
sectors studied and in five of these sectors the effects are positive. These results suggest
that real exchange rate volatility may stimulate export supply by risk-neutral firms
through its effects on their expected profits.
Hooper and Kohlhagen (1978) adopted the Ethier (1973) model of market
equilibrium for traded goods and extended it by relaxing the assumption of infinitive
elasticity of the export supply function. Hooper and Kohlhagen formally tested the
impact of volatility on trade flows and prices using quarterly multilateral and bilateral
trade flow data for Germany, Japan, Canada, France, the United States and the U.K. for
the period 1965-1975. When aggregated trade data were the focus of the model, no
significant impact of volatility on volume or price was found. However, for bilateral trade
flows, the authors found evidence that exchange rate uncertainty has a significant impact
on prices but not significant impact on volume of trade.
McKenzie (1998) analyses the impact of exchange rate volatility on both
aggregate and sectoral trade flow data for the Australian economy over the period 19881995. ARCH models were used to generate a measure of exchange rate volatility and
then volatility was tested in a standard model. In this Model he specified imports or
exports as a function of prices, income, the exchange rate and its volatility. He concluded
that in terms of aggregate, volatility of the exchange rate negatively affected Australian
imports. These results indicates that the aggregate nature of trade data may dilute the
effects of volatility and also that the effects of volatility may potentially be sensitive to
the nature of the industry producing the traded commodity.
42
2.3.6A Estimation Technique Used

Studies on the effects of exchange rate movements on trade flows have used a wide
variety of empirical techniques. Most of the research from the 1970s and 1980s used
traditional econometrics techniques (OLS, 2-SLS). One possibility is to take a cross
sectional approach and compare the level or rate of growth of trade across countries at a
given point in time. Cross-section studies were undertaken by Kenen (1986) and Thursby
and Thursby (1987). Others have chosen different forms of time series analysis; among
them are (Chowdhury (1993), Lastrapes and Koray (1990), Thraen et. al. (1992) and
Assery and Peel (1991) who used the vector auto regression (VAR) and GARCH-in mean
models.
Empirical studies have in general been unable to establish, systematically,
significant negative links between measured exchange rate volatility and the volume of
international trade, whether on an aggregate or on a bilateral basis (Kenen and Rodrik,
1986). One possible reason for the non robust results is that in previous empirical work it
has not been recognized that real exports and some of its proposed determinants such as
real world trade, are, a priori, potentially non-stationary integrated variables. No
recognition of the above issue implies that the inferences made concerning income and
price elasticities as well as the impact of exchange rate volatility on exports are
potentially misleading (Engle and Granger, 1987). In agricultural trade there is some
research which use cointegration techniques such as Orden and Fackler (1986), Jennings
et al. (1990), Sarker (1993), Smith (1999), and Cho et al. (2002).
43
2.3.7 Empirical Studies on Exchange Rate and Mexican Agricultural Trade

While the border restrictions to free trade between Mexico and the United States have
been gradually reduced and dismantled over time to make agri-food trade between
Mexico and its NAFTA partners more liberalized, the changes in peso-dollar exchange
rate and exchange rate volatility may have also contributed to the expansion of Mexican
agri-food trade in recent years. The relevant role of Mexican currency since 1980 in the
trade performance of Mexico is not well known as the literature on the topic has been
rather scarce. An attempt is made here to review the relevant trade literature concerning
the relationship between Mexican exchange rate movements and trade flows.
Espinoza-Arellano et al (1998) attempted to determine the primary economic
forces influencing M exicos competitiveness in the Unites States winter melon market.
Price linkage equations were used to link retail and farm level prices in the Unites States,
Mexico, and Caribbean Nations. The price transmission equations included tariffs and
exchange rates. Per capita melon demand was estimated as a function of own-retail price,
other melon prices and per capita income. Supply equations were estimated with an
acreage and yield functions. Mexican and Caribbean prices were specified as a function
of Unites States prices, real exchange rate, and applicable tariffs for the December to
May periods of 1970-1994. The 3-stage least squares technique was used to estimate the
model equations and then it was validated using in-sample simulation. A baseline was
created in order to compare results across scenarios. The results show that the effect of
the 1994-1995 peso devaluation had important short run affect on Mexican melon exports
to the Unites States Melon exports increased 36, 18, and 4 percent, respectively relative
to the baseline, due to the peso devaluation. As a result, M exicos share of Unites States
44
imports increased by 11, 7, and 2 percent respectively. They argued that exchange rates
do have an important effect on trade, in particular, the weakening of the exporters
currency increases export opportunities in the short run.
Kapombe and Colyer (1998) used a multiple equation structural time series model
as the theoretical framework for supply and demand equations to analyze Unites States
broiler exports. They used quarterly data from 1970-1995 for Unites States broiler
production, demand, inputs, exports, meat prices and annual values for broiler import
equations for Mexico, Japan, Hong Kong and Canada. The results indicated that Mexican
broiler demand is negatively influenced by the exchange rate, as a one percent increase in
the peso-dollar exchange rates will result in a 0.58 percent decrease in import demand.
Japanese demand for Unites States broiler exports were negatively influenced by the
exchange rate, a one percent increase in the yen-Unites States dollar exchange rate will
cause a 0.96 percent decrease in import demand. Also Hong Kong import demand is
negatively influenced by the exchange rate.
Diaz-Garces (2002) examined the performance of the Mexican trade flow for the
period 1980 to 2000, which includes important institutional changes such as accession of
Mexico to GATT and the implementation of the NAFTA agreement. Dias pointed out
that the remarkable performance of the Mexican trading sector has been frequently
discussed in political sectors but serious studies on the topic have been rather scarce. He
estimated long-run import and export demand functions for the period 1980-2000 using
co-integration analysis. He found that the export equation, as a function of the Unites
States industry output and the real exchange rate, is stable for the period 1990-2000. The
import demand function, as a function of the Mexican industrial output, real exchange
45
rate, and exports, showed instability for the same period. It was not possible to find a
unique cointegration vector for the whole sample period because of the structural change
in 1989-1990.
Konno and Fukushige (2003) assesses the long-run bilateral trade relations in the
import function between the US and Mexico by applying the Dynamic Ordinary Least
Squares (DOLS) method proposed by Stock and Watson (1988) together with a gradual
switching model for each countrys long-run import function. They used domestic
income, domestic price, foreign price and exchange rate as determinants of bilateral
import functions. The general foreign export price is used as the foreign price because
exact export prices for each country were not available. The data used in this paper was
quarterly data from 1981(1) to 1999(4). Their results suggest that no additional impact
was caused by NAFTA in either import functions when NAFTA came into effect. There
was a gradual shift in the long-run relation for each import function much earlier than the
operation of NAFTA. The estimated result also suggest that Mexican exports would
become more sensitive to the condition of the Unites States economy through gradual
increases in income elasticity and the domestic price elasticity of the Unites States
imports, while Mexican exports to the Unites States still account for about eighty percent
of the total Mexican exports.
Mora-Flores et al (2003) assessed the effect of changes in export prices on
Mexican crop and livestock gross value. Changes in quantities due to variations of prices
were modeled by means of a Cobb-Douglas price function while the total change in
exports gross value was estimated using the logarithmic differential between both
variables. They used monthly data from 1998 to 1996. A direct relationship between
46
export price and export quantities was found. The response for cattle exports was greater
than the response for other agricultural commodities, in terms of price elasticity. A
devaluation of the exchange rate implies higher prices for importers and higher gross
value for exports.
2.4. Summary
The economic and trade liberalization undertaken by the Mexican government during the
1980s and 1990s positively the growth of the economy and the Mexico-United States
trade flows of agri-food products has grown twice since 1990. To explain the
performance of the Mexican agricultural trade, different hypothesis have been advanced
such as the contributions of NAFTA and the peso devaluation of 1994-1995 but no
consensus exist on this respect. The review of the published literature on the effects of
changes in the exchange rate and its volatility on agricultural trade reveals some
interesting features; first, most of the previous studies have ignored the time series
properties of data. Thus, these studies may be subject to spurious relationship bias
(Granger (1981), and Engel and Granger (1987)). Second, different data sets have been
used by different studies, so the results of the effects of exchange rate on trade are not
comparable among studies. Third, some previous studies have concentrated on aggregate
trade flows that essentially mask the effects of exchange rate on a particular sector and no
attention on disaggregated trade flows or particular groups of commodities. Finally, the
focus of most empirical studies has been on the specification of a standard deviation or
variance-based measure of exchange rate volatility.
47
The review of the literature also revealed that a number of empirical studies
dealing with effects of exchange rate changes on trade flows have explicitly focused on
the Mexico-Unites States trade but no conclusive evidence is provided to explain the
Mexico-Unites States trade performance. Finally, none of the studies investigated the
effects of exchange rate volatility on Mexican trade flows and the specification of the
exchange rate volatility remains an unsolved issue.
48
CHAPTER 3
CONCEPTUAL FRAMEWORK
3.1 Introduction
The purpose of this chapter is to develop a theoretical framework to be used to explain
the effects of changes in exchange rate and its volatility on international agri-food trade
flows. This framework is then used to develop an empirical specification of the trade
model used to determine empirically the effects of changes in exchange rate and its
volatility on the Mexico-United States agri-food trade flows.
The first part of this chapter presents a brief overview of the modem theory of
exchange rate determination based on the monetary approach. In this framework,
exchange rates are determined as part of the general theory of asset price determination.
This is followed by a short discussion of the main trade models in the literature and the
role of the exchange rate on international trade. The second part of this chapter consists
of two sections: the first section describes the microeconomic foundations of the
theoretical model and the derivation of the import demand functions. The second section
deals with the derivation of comparative statics results from the theoretical model used in
this research and interpretation of these results. The third part describes observed
peculiarities
in
specific
commodities
(i.e.
specific
institutional
and marketing
characteristics of each commodity industry) and how those could be accommodated in

the empirical implementation of the theoretical model. This is followed by an attempt to
specify trade equations for the selected commodities and provide data descriptions. The
final section provides a summary of main results and guides the reader to the next
chapter.
49
3.2 Introduction to Exchange Rate Economics

Since the early 1970s, private market forces within a floating exchange rate system
determined the exchange rate. Exchange rate behaviour can be characterized by relatively
small and continuous changes that occur quickly in response to new information. Thus,
the search for an explanation of exchange rate behaviour in a setting of an integrated
international financial market has been an important issue in recent years.
Prior to the 1970s, the dominant international macro model was an open
economy Keynesian model developed by Mundell and Fleming (1962) which integrates
asset markets and capital mobility into an open-economy model to determine exchange
rate between two currencies. An extension of the Mundell-Fleming framework by
Dombusch (1976) introduces rational expectations and output price dynamics into
Mundell-Fleming framework and provides a good understanding of the effects of
exchange rate variability in the economy.
The distinguishing feature of exchange rate models developed during the 1970 is
that they are based on consideration of stock equilibrium in the international financial
markets (Taylor, 1995). The major insight of the model from the 1980s is that foreign
exchange is a financial asset. The implication is that current spot exchange rate reflects
expected values of future exogenous variables discounted back to the present period and
that the price of a currency is determined by the demand for it as a financial asset relative
to the demand for other currencies (Frenkel and Mussa, 1985).
The influential paper of Obsfeld and Rogoff (1995) initiated a new line of
research on exchange rate dynamics and on the international transmission mechanism of
fiscal and monetary policies by applying an intertemporal general equilibrium model with
50
nominal rigidities in the presence of monopolistic competition. This model substantially

complement the traditional Mundell-Fleming (1962) and Dombusch (1976) framework
with an explicit micro foundation. Together with the endogenous determination of longrun supply and the incorporation of lifetime-utility maximizing households, the new class
of models can be characterized as a micro foundation dynamic version of the MundellFleming model (Fendel, 2002). This microstructure literature provides an understanding
of the mechanism generating deviations from the market fundamentals. In this literature
researchers focused on the behavior of market agents and market characteristics rather
than on the influence of macro fundamentals (Taylor, 1995).
The 1980s and 1990s have seen an enormous growth in the studies on exchange
rate economics. Given the importance attributed to the exchange rate in the success or
failure of an open economy, it is not surprising that exchange rate economics is one of the
most heavily researched areas of the discipline (Macdonald and Taylor, 1992). The
period since the shift to the floating exchange rate in 1973 has generated rich sets of data
on exchange rate and on related variables, giving researchers an opportunity to test a
number of propositions relating to foreign exchange markets. Despite this extensive
research, a large number of issues remain unresolved and exchange rate economics
continues to be a challenging area of research.
Exchange rate economics comprises a set of broad areas of research such as
studies on exchange rate determination (exchange rate is determined by supply and
demand of money), the relationship of exchange rate and good prices (the law of one
price, exchange rate pass-through and pricing to market) and the effects of exchange rate
on international trade. The primary objective of this research is to investigate the effects
51
of changes in exchange rate and its volatility on Mexican agri-food trade. However, for
the sake of better understanding, this section provides a brief exposition of two main
theories of exchange rate determination; (i) the monetary approach (the flexible-price
model and the sticky-price differential formulations) and (ii) the portfolio balance
approach.
3.3 Theories of Exchange Rate Determination

3.3.1 Flexible-Price Monetary Model
Since an exchange rate is defined as the price of one countrys money in terms of that of
another, it makes sense to analyze the determinants of that price in terms of the demand
for two monies. This is the basic rationale of the monetary approach to the exchange rate
(Frenkel, 1976; Kouri, 1976; and Mussa, 1976; 1979).
The early, flexible-price monetary model relies on the assumptions of both,
purchasing power parity (PPP) and the existence of stable money demand functions for
the domestic and foreign economies. The demand for money is assumed to depend of real
income, y, the price level, p and the level of the interest rate r (an asterisk denotes foreign
variables). Monetary equilibria in the domestic and foreign country, respectively, are
given by:
m s, = p t + (pyt - Xrt ,
(3.1)
m f = p* +</>* y* - A* r * t
(3.2)
Equilibrium in the traded goods market ensures that there are no further profitable
incentives for trade flows to occur. This occurs when prices in a common currency are
equalized and PPP holds. That is, st = p t - p *. Where st is the logarithm of the nominal
52
exchange rate. Thus, if PPP holds continuously, the logarithm of the real exchange rate
q, say (qt = st - p t + p * )is a constant. The world price, p *, is exogenous to the
domestic economy, being determined by the world money supply. The domestic money
supply determines the domestic price level and hence, the exchange rate is determined by
relative money supplies. Algebraically, substituting equations (3.1) and (3.2) into the PPP
equation yields:
st = (m s - m s*)t - <pyt + <p* y* + Art - X * r*
(3.3)
Equation (3.3) is the basic flexible-price monetary model. The above equation says that
an increase in the domestic money supply, relative to the foreign money stock will lead to
a rise in st - that is, a fall in the value of the domestic currency in term of the foreign
currency. An increase in domestic output, as opposed to the domestic money supply,
appreciates the domestic currency. Similarly, a rise in domestic interest rates depreciates
the domestic currency.
3.3.2 Sticky-Price and Real Interest Differential Monetary Models

A problem with the early flexible-price variant of the monetary approach is that it
assumes continuous PPP. Under this assumption, the real exchange rate (the exchange
rate adjusted for differences in national prices levels) cannot vary by definition. Yet, a
major characteristic of the 1980s experience with the floating system has been the wide
gravitations in the real rates of exchange between many of the major currencies, bringing
with them the very real consequences of shifts in international competitiveness. Clearly
therefore, the simple, flexible-price monetary approach does not fit the observable facts.
An attempt to rehabilitate the monetary model led to the development of a second
53
generation of monetary models, beginning with Dombusch (1976). The sticky-price

monetary model allows for substantial overshooting of both the nominal and the real
price adjusted exchange rates beyond their long-run equilibrium (PPP) levels, because the
jump variables in the system- exchange rates and interest rates- compensates for
sluggishness in other variable such as goods prices.
The intuition behind the overshooting results in the sticky-price monetary model
is as follows. Given the effects of a cut in the nominal money supply of a country, sticky
prices in the short run imply an initial fall in the real money supply and a consequent rise
in interest rates in order to clear the money market. The rise in domestic interest rates
then leads to a capital inflow and an appreciation of the nominal exchange rate.
A short-run equilibrium is achieved when the expected rate of depreciation is just
equal to the interest differential (uncovered interest parity holds). Since the expected rate
of depreciation must then be nonzero for a nonzero interest differential, the exchange rate
must have overshot its long-run equilibrium (PPP) level. In the medium run, however,
domestic prices begin to fall in response to the fall in money supply. This alleviates
pressure in the money market (the real money supply rise) and domestic interest rates
begin to decline. The exchange rate then depreciates slowly in order to converge at the
long-run PPP level.
3.3.3 Portfolio Balance Model

A common aspect in the flexible-price and sticky-price monetary models with the
portfolio balance model is that the level of the exchange rate is determined by supply and
demand in the markets for financial assets. The exchange rate however, is a principal
54
determinant of the current account of the balance of payments. A surplus (deficit) on the
current account represents a rise (fall) in net domestic holdings of foreign assets, which in
turn affects the level of wealth, the level of asset demand, both of which affect the
exchange rate. Thus, the portfolio balance model is an inherently dynamic model of
exchange rate adjustment, which includes asset markets, the current account, the price
level and the rate of asset accumulation.
The portfolio balance model, like the sticky-price model, allows one to
distinguish between asset market short-run equilibrium and the dynamic adjustment to
long-run equilibrium. Unlike the sticky-price model, it also allows for the full interaction
between the exchange rate, the balance of payments, the level of wealth and stock
equilibrium. In this model the exchange rate is determined solely by the interaction of
supply and demand in the asset market in the short run.
The key feature of this model is that it assumed imperfect substitutability between
domestic and foreign assets. In this model, the net financial wealth of the private sector
(W) consists of money (M), domestically issued bonds (B) and foreign bonds
denominated in foreign currency (B*). Then, the definition of wealth is W= M + B +
SB*. With foreign and domestic interest rates given by i and i* and Se the expected rate
of depreciation of the domestic currency, the demand functions are defined as follow:
M = M (i, i*, + Se) W,
(3.4)
B = B (i, i*, + Se)W ,
(3.5)
SB =B *(i, i*, + Se)W ,
(3.6)
B* = T (S/P) + i*B*,
(3.7)
55
The rate of change of B* given by equation (3.7), the capital account, is equal to
the current account which is in turn equal to the sum of the trade balance, T(/) and net
debt service receipts, i*B*, which assumes static expectations and an open market
purchase of bonds by government. In order to induce agents to hold more money and
fewer bonds the domestic interest rate falls and, as agents attempt to compensate for the
reduction of their portfolios of domestic assets by buying foreign bonds, the exchange
rate will depreciate, driving up the domestic currency value of foreign bonds.
3.4. International Trade Models

International trade is traditionally thought to consist of each country exporting the goods
most suited to its factor endowment, technology and climate while importing the goods
least suited for its national endowment characteristics. Such trade is called inter-industry
trade because countries export and import the products of different industries. On the
other hand, intra-industry trade represents international trade within industries rather than
between industries (Ruffin, 1999).
The modeling of international commodity flows has been a heavily studied
economic area since the beginning of economics as a scientific discipline. One of the
most successful approaches in the economic literature that account for inter-industry
trade is the Hecksher-Ohlin (H-O) model. It links a country's endowments of land, labour
and capital to its exports and imports of these factors as embodied in goods, which is the
leading explanation for trade patterns.
56
3.4.1 The Hecksher-Ohlin Model of Trade

The H -0 model describes a world economy in which trading countries faces the same
technological frontiers and has productive factors with the same qualities. The only
difference between countries is in terms of the physical quantities of the factors of
production, so that the H -0 model is an account of trade based on factor endowments.
This theory has three fundamental features. First, each country exports goods that are
intensive in the countrys relatively abundant factors. The second feature is that trade
based on factor endowments benefits abundant factors and hurts scarce factors. The last
feature of the H -0 model is that international trade results in a tendency toward factor
price equalization.
The H-O model assumes that the differences between countries are differences in
the relative endowments of factors of production. When trade occurs, it will be nationally
advantageous and will have notable effects on prices, wages and rents when the trading
nations differ in their relative factor endowments and when different industries use
factors in different proportions. In the H-O model, economies export the services of their
abundant factors and import the services of their scarce factors. But empirical
investigations of the H -0 model have not had much success (Trefler 1995).
The importance of intra-industry trade was widely recognized in the 1960s, thus,
most of the developments in trade modelling have focused on this type of international
trade. Among the most heavily used models are the Armington model and recently the
gravity model. The following section describes briefly these two modelling approaches.
57
3.4.2 The Armington Trade Model

The Armington (1969) model is based on the differentiation of products with respect to
their origin and the imperfect substitution in demand between imports and domestic
supply. The theoretical derivation of the Armington model gives a linear parsimonious
specification of the import demand system. This approach models the demand system for
differentiated goods in terms of the origin of exports (Yang and Koo, 1992). This model
assumes that goods imported from different countries (or regions) create different
consumer utility, i.e., the elasticities of substitution among imported goods from different
places are not infinite. Since this simple linear specification is consistent with utility
maximization and economizes the degrees of freedom in empirical application, the model
has been widely used for international trade analysis.
The Armington model contains two major assumptions induced by the constant
elasticity of substitution (CES) utility function. These are the single CES and
homotheticity assumptions. The single CES assumption is that the elasticity of
substitution between any two products is independent of the quantity demanded and is the
constant across all pair of commodities. This assumption restricts responses of the import
demand for each product to the price change (relative to the price index for the good) to
be the same for all products. This assumption would be too restrictive if the elasticity of
substitution between any pair of products is not the same. Even under the single CES
circumstance, effects of relative price changes on market share are not likely to be the
same (Yang and Koo, 1992). The homotheticity assumption in the Armington model
implies that the size of market does not affect each exporting countrys relative market
58
share and that expenditure elasticities are the same and unitary. These assumption are too
restrictive for meaningful empirical analysis.
A concern in using the Armington model is whether the restrictive assumption
should be imposed on data that do not support the assumptions. If the data do not satisfy
these assumptions, the estimated results would be biased. A number of studies re
specified the Armnigton model to derive more sensible results.
In general, a two-stage budgeting procedure assumes that consumers allocate their
total expenditures in two stages. In the first stage, total expenditure is allocated over
broad groups of goods, while in the second stage group expenditures are allocated over
individual commodities. It is well known that weak separability of the direct utility
function over broad groups of goods is a necessary and sufficient condition for the second
stage of a two-stage budgeting procedure (Deaton and Muellbauer, 1980). However,
weak separability imposes restrictions on consumer behaviour. First, the marginal rate of
substitution between two goods from the same group is independent of the consumption
of goods in other groups. Second, the substitution effects between goods in different
groups are limited. A price change of a commodity in one group affects the demand for a
commodity in another group only through the group income effect. Third, separability
implies a restrictive relationship between price and income effects.
The first two-stage of Armingtons framework is, in general, equivalent to that of
two stage budgeting process. That is, in the first stage the importer decides how much of
a particular commodity to import. In the second stage, given the total amount to be
imported, the importer decides how much to import from each supplier. Thus, the
59
implications of weak separability apply to the possible substitution effects among

commodity groups. The Armington model uses a CES within-group specification, hence;
r p \ (X~c)
W- = b
\P j
where wt is the marker share of import from source i,bi is a constant, Pt is the price of
the commodity from the ith source, P is the import price index depending only on the
within-group prices and o is the constant elasticity of substitution parameter. The CES
specification implies weak separability between different import sources.
The Armington approach also assumes homotheticity of the sub-utility or withingroups utility functions. This implies that importers market share is independent of
group expenditures. Consequently, all expenditure elasticities within a group are equal
and unitary and import market shares change only in response to relative price changes.
Thus, the Armington framework implies that in the second stage (within-groups
allocations) market shares do not vary with expenditures and that different import sources
are separable (Alston et al., 1990).
3.4.3 The Gravity Model

Gravity models were first applied to international trade by Tinbergen (1962) who
proposed that the volume of trade could be estimated as an increasing function of the
national incomes of the trading partners and a decreasing function of the distance
between them.
Although the gravity model became popular because of its empirical
success, it was also criticized because it lacked a coherent theoretical foundation. These
theoretical foundations were subsequently developed by Anderson (1979) and Bergstrand
60
(1985), who derived gravity models from models of monopolistic competition and
Deardorff (1998), who demonstrated that the gravity model can be derived from the
Heckscher-Ohlin framework. Typical gravity models contain the following components:
(i) economic factors affecting trade flows in the origin countries, (ii) economic factors
affecting trade flows in the destination countries, and (iii) natural or artificial factors
enhancing or restricting trade flows.
A gravity model is a reduced form equation generated from a partial equilibrium
model. The demand equation for the specific commodity is derived from utility
maximization using a constant elasticity of substitution (CES) utility function subject to
an income constraint. The supply equation is derived from the firms profit maximization
approach in the exporting countries with resource allocation assuming a constant
elasticity of transformation (CET) during the production process (Bergstran, 1989). The
commodity-specific gravity model, under market equilibrium conditions of demand and
supply systems, can be derived as follows:
X, = c c ^ Y f C f T ^ P ^ P f E f e ^
i = 1,2,
, N and j = 1,2,...., Af
where X tj is the volume of commodity traded from country i to country j; Yi(Yj )

represents income of country i(j); CtJ is transport cost between trading countries i and j;
Ttj represents any other factors either aiding or resisting trade between countries i and j;
Pi (P j) is the price of the commodity at country i s export port (country j s import
port); E tj is the spot exchange rate (the value of country j s currency in terms of country
i s currency); and etj is the random error term.
61
The income of the exporting country can be interpreted as the countrys

production capacity, while an importing countrys income is the countrys purchasing
power. It is expected that trade flows are positively related to the exporting and importing
countries income. Transportation costs and tariffs, which serve as trade barriers, should
be negatively related to the volume of trade flows. The prices of a commodity in
exporting and importing countries are important in determining trade flows, thus, they
enter the gravity equation and are negatively related with traded quantity.
A commodity moves from a country in which the price of the commodity is low
to countries in which the price is high. Trade flows are hypothesized to be positively
related to changes in export prices and negatively related to changes in imports prices.
Exchange rate is also an important factor affecting the trade flows. Appreciation of a
countrys currency against other currencies reduces the countrys exports and increases
imports and depreciation stimulates the countrys exports. Gravity models can be used to
analyze either aggregate or single commodity trade flows (Koo et ah, 1994).
3.5 Exchange Rates and International Trade

Exchange rate and international trade is a heavily studied field since the 1970s. It
became relevant after the shift from the fixed to the floating exchange rate regime. To
assess the effects of changes in exchange rate and its variability on international trade
flows several models have been proposed in the literature (e.g., Ethier, 1973; Kost, 1976;
Kawai, 1973). The model proposed in this section starts with a graphical representation
of Kosts exposition of a simple two-country trade model for a single commodity. An
algebraic formulation of this trade model follows the graphical representation and finally
62
the theoretical model based on the Batra and Ullah (1974) model is developed and fully
described.
3.5.1 Graphical Representation of a Trade Model

The simplest way to explain how exchange rate plays an important role in trade is to
consider, for illustrative purposes, the following one commodity-two countries trade
model. While K osts model is a very simple one, it allows us to incorporate the effects of
exchange rate changes on trade volume and prices. This model does not accommodate the
effects of changes in the prices of related goods, income and volatility on trade flows.
To incorporate the effects of these factors on trade volume and prices, import
demand and export supply equations are derived from an expected utility maximization
process. Since volatility of the exchange rate is an important issue in this research, a
separate section is devoted to discuss the pros and cons of different volatility measures
used in the literature. Following Kost (1976) a two-country partial equilibrium trade
model for a single commodity is often used to describe the effects of an appreciation and
depreciation of a currency on prices, production, consumptions and trade flows. In this
model, the commodity traded between two countries is assumed to be a homogeneous
good. Mexico is the exporting country while the foreign country (U.S) is the importing
country. Assuming that both the U.S. and Mexico to be large countries, the interaction of
demand and supply functions in each country yields the closed economy equilibrium
price and quantity produced and consumed. The intersection between the excess supply
from Mexico and excess demand from U.S. in Figure 3.1 determines the equilibrium
price and quantity of the commodity to be traded between these two countries.
63
Assuming initially that the exchange rate between these two countries is at par
(i.e. one Mexican peso equal to one US dollar), as shown by the 45 line OA in panel (c)
of Figure 3.1, the volume of trade is equal to the distance OQo. From panel (a) the initial
price, Po is higher than the autarky price in Mexico. This creates an excess supply of a
of this good in Mexico. Since exchange rate is at par and that Po is lower than equilibrium
price in the U.S., the United States is willing to import from Mexico the same amount of
the commodity (a). If a change in exchange rate causes the value of the Mexican peso
to depreciate, the OA rotates to OA, the excess demand curve rotates upward from ED to
ED in panel (b). The new equilibrium is attained where ES intersects the new excess
demand (ED) function. The intersection of ES and ED" functions results in an increase
in the exporters price to OPi and an expansion of trade to OQi (panel b).
Fig 3.1 Effects of Exchange Rate on a Particular Agricultural Commodity.
$MX
'Mex
ES
ED
ED
Mex
(A) Mexico (exporter)
(B) The Trade Sector
(C) $MX/$US Exchange
(D) U.S. (importer)
64
The depreciation of the Mexican peso relative to the U.S. dollar causes Mexican
commodities to become cheaper relative to U.S. products. At the new price Pi, U.S. is
now willing to import b units of the commodity from Mexico. As can be seen from
Figure 3.1, the trade equilibrium price lies between equilibrium price of the exporting
country and that of the importing country. The exact location of this price in the two
countries depends on the relative elasticities of the excess supply and demand functions,
which in turn depend on the elasticity of demand and supply in Mexico and in the U.S.
3.5.2 Algebraic Formulation of the Trade Model

The above two-country partial equilibrium trade model is reformulated below as in
algebraic model. It allows one to include the effects of exchange rate changes on
quantities and prices. Competitive markets are assumed in each country. W ith no barriers
to trade and no transportation costs, the model consists of Mexican excess supply and
U.S. excess demand that can be extended to incorporate a rest of the world sector. The
Mexican Market can be characterized as:
D\iex = a - bPMex
Domestic Demand;
(3.8)
SMex = c + dPMex
Domestic Supply; and (no storage),
(3.9)
ES\fex = S[y[ex - Djvlex = (C

Then,
<0 + (b + d)P\iex
(3.10)
(b + d ) > 0
(3.11)
For the importing country

D u s = e - f P us Domestic demand
(3.12)
65
S u s = g + hPus Domestic supply
(3.13)
E D = D us - S us = e - f P u s - ( 8 + hPu s)
ED = ( e - g ) - ( f + h)Pus
(3.14)
dED,
'U S
(3.15)
=
-(f+h)<
dPuS
Pus =
(Price equilibrium)
(3.16)
ES = ED
(Market clearing condition)
(3.17)
where: Pmcx is price in Mexico (Mexican Pesos), Pus is price in U.S. (US dollars)and r is
the exchange rate between Mexico and U.S. currency. Substituting (3.10), (3.14) and
(3.16), into (3.17) gives
ES(PMex) = ED(PMex r ) .
(3.18)
Taking the total derivative of (3.18) results in

dES
V^Mex
dED
dED
dPUa = r dPMex + PUa dr
vPuS
vPuS
im
(3.19)
Equation (3.19) is then solved for the elasticity of the exporters price with respect to a
change in the exchange rate, E p r to obtain
p
Mex-r
dPMex
dr
ur
r
P
M ex
----e E S - e ED '
Similarly, the elasticity of the importers price with respect to a change in the exchange
rate, e p
is
= dP,JS J _
P u s ,r
Pus
,3 2 l)
es
ES
ED
66
where, e ED is the elasticity of excess demand and s ES is the elasticity of excess supply.
The elasticity of Mexican market price with respect to exchange rate (e Pu r ), as
well as that of U.S. market price (s p^ r ), depends on the elasticity of excess demand for
U.S. and that of the excess supply for Mexico. Since the Mexican supply elasticity is
positive and that of the U.S. demand elasticity is negative, it follows that Pu r < 0. The
effects of changes in the exchange rate on the quantity traded is found by taking the total
derivative of the U.S. (excess) demand and substituting result (3.19) result into an
elasticity expression:
dED r
dr ED
dP r
3 ED Pm
(3.22)
dr Pus dPus ED
(3.23)
therefore, s ED r= (e P(js) ( e ED)
(3.24)
Thus, the elasticity of the equilibrium quantity traded with respect to the exchange
rate ( s P r) also depends on the elasticity of excess demand for U.S and that of excess
supply for Mexico. Again, since e ES > 0 and s ED < 0 , then the demand and supply
functions are well behaved, s EDr < 0, (i.e. the elasticity of import demand with respect to
exchange rate will be negative).
From equation (3.20) one can deduce that a devaluation of the Mexican peso,
represented by a decline in r, will increase PmCx ranging between zero and the percentage
change in the exchange rate. But at the same time it will increase the quantity traded.
Given e ES = e ED ^ 0, it is clear from the above trade model that for a given change in
67
the exchange rate there is an impact on both price and quantity traded. This model is
simplistic in that both supply and demand are functions of only the commodity price. In
reality supply and demand are also influenced by other factors such as prices of related
goods, consumers income, tastes and technology. Due to theses limitations the above
model is not suitable for this research. Consequently, a more general model is developed
next.
To determine the effects of changes in exchange rate and its volatility on Mexican
agri-food trade, this study relies on standard international trade theory. To represent the
effects of exchange rate uncertainty, income and other prices on trade flows, it is assumed
that a risk-averse competitive firm is faced with uncertainty in the international market
and that it maximizes the expected utility of its profits (Ethier, 1973; Kawai, 1973; and
Batra andUllah, 1974).
The firm faces exchange rate uncertainty both through the domestic cost of
imports and through the value of exports. Since there is no forward market for foreign
exchange, the firm faces an exchange rate risk for future conversion of its sales revenues
from the importing country into the currency of the exporting country. If the firm is riskaverse, it would be willing to incur an added cost to avoid this risk, so that the un-hedged
risk serves as an implicit cost. Thus, in the presence of risk, the firms supply price at
each quantity would be higher than if there was no risk.
A few previous studies dealing with exchange rate effects on agricultural trade
have included a measure of volatility in their analysis. Inclusion of a volatility measure in
a trade model is likely to minimize bias in the estimated coefficients due to omitted
variables and allows the trade response to changes in exchange rate volatility. There is,
68
however, no generally accepted measure of quantifying exchange rate volatility. This

study draws upon the empirical literature and focuses on a set of volatility measures to
determine the effects of exchange rate volatility on trade flows.
It is recognized in the trade literature that exchange rate movements have different
effects on different goods according to the degree of tradability. Indeed, in the case of
some agricultural products, such as vegetables in Mexico, the exported commodities are
produced with large amounts of imported inputs that are denominated in foreign
currency. So the effect of exchange rate movements also depend on the proportion of
domestic and imported inputs required in the production and export of the commodity.
To better assess the effects of exchange rate movements on Mexican agricultural
trade, the most important commodities traded are categorized in three groups: fresh
vegetables, grains and processed food (milk powder). The most important agricultural
commodities exported, by value, are fresh vegetables (tomato, bell pepper and
cucumber), coffee and live cattle. These commodities account for more than 60 percent
of total agricultural export value in Mexico (see table 2.2 in chapter two). In the case of
imports, com, wheat and feed grains (soybean and sorghum) account for more than 50
percent of total agricultural imports into Mexico (see table 2.3. in chapter two).
3.6. Characteristics of the Commodity Industries in Mexico

The agri-food sector in Mexico contributes to employment and income of about 30
percent of the population and the grain and horticultural sectors are very prominent. It is
to be noted here that for Mexico, the most important exported agricultural product is
tomato while maize and sorghum account for the larger share of imports into Mexico.
69
Mexico is also the world leading importer of non-fat dry milk. The following section is
devoted to examining some distinguishing features of these commodities in Mexico, so
that some institutional features can be incorporated into the model.
3.6.1. Fresh Vegetables Industry and Exports

Vegetable production in Mexico generates almost 20 percent of the total value of
agricultural production in Mexico. Of the approximately 100,000 horticultural producers
in Mexico, an estimated 20,000 take part in exports to some degree, although there is
high concentration of production for exports among a limited number of large growers
and their families (Schwentesius and Gomez, 2001). The horticultural sector grew at an
annual rate of 2.9 percent during the last decade. An important contributing factor to this
growth has been the rapid expansion of domestic demand which currently absorbs about
80 percent of the domestic supply.
While Mexico is a large and growing market for fresh produce, it is expected that
it will remain a major exporter of fresh fruits and vegetables. It has a wealth of
productive land in both tropical and temperate climates. Mexico exports far more produce
to the United States than the United States ships to Mexico. In 2001, Mexican produce
shipments to the United States totaled about $2.5 billion and U.S. produce shipments to
Mexico was about $300 million. Besides relative magnitude, the graphs reveal other
contrasts. Vegetables account for most of M exicos shipments, while fruit constitutes
most of U.S. shipments. And, since 1995, U.S. shipments have been growing much more
rapidly than Mexican shipments, albeit from a lower base.
70
Mexican-U.S. bilateral produce trade is largely complementary. That is, the

United States purchases products from Mexico that are not or cannot be produced
efficiently in the United States; bananas, for example. Similarly, the United States tends
to export produce in which it has a comparative advantage, such as apples and pears.
Much of the complementarity stems from differences in growing seasons.
Mexican exports of fresh vegetables have been growing since the Mexican trade
liberalization in the 1980s. They had grown from 30 percent of the value of all
agricultural exports, in 1980, to over 50 percent in the first half of the 1990s. M exicos
share of the US vegetable market almost doubled from 7.6 percent in the early seventies
to about 15 percent in the mid nineties. Mexico is by far the largest foreign supplier to
U.S. market, typically accounting for around 70 percent of all U.S. vegetable imports.
Favourable winter weather and significantly lower labour costs have made the Northwest
region of Mexico a natural supplier to the U.S winter vegetable market. Specifically the
State of Sinaloa has developed a modem fresh vegetable industry, adapting U.S.
technologies for use almost exclusively on irrigated, middle-sized and large private
farms. Sinaloas vegetable growers apply modem marketing and management techniques
and have developed strong financial and commercial ties with the US distributors.
(Malaga et al., 1996).
Tomato industry is the most important fresh vegetable both in terms of value and
as a generator of labour demand. It represents about 37 percent of value of all vegetable
Mexican exports and 16 percent of the value from all agricultural exports. Tomato
production takes place in all of the country but six states (Sinaloa, Baja California, San
Luis Potosi, Jalisco and Nayarit) contribute more than 70 percent of the value of
71
production. Fresh tomato production has two perfectly differentiated destination markets;
80 percent of the production goes to the domestic market, 70 percent of this production is
marketed through three regional wholesale markets.
United States is the leading external supplier of tomatoes, onions, avocados,
lettuce, oranges and potatoes to Mexico (USDA-ERS, 2003). Mexico is normally
considered a source of these products, not a destination. However, Mexico is an
expanding market for fresh fruit and vegetables. While Mexico will supply much of the
growing demand itself, it will not supply all of it and certainly not all products. Given its
location, the United States is the natural supplier. California and Texas are closer to
Mexico City than to Chicago or New York. And northern Mexico, where demand is
growing the most, is even closer. Shipping produce from the United States to Mexico
should be no more remarkable than shipping produce within States. Indeed, as the two
economies become more closely integrated, channels of distribution have expanded and
barriers to free exchange are being dismantled (USDA-ERS, 2003).
Tomato exports from Sinaloa, Mexico, directly compete with tomato produced in
South Florida. Mexican shippers in Sinaloa produce mainly extended-shelf-life tomatoes
that are harvested as vine-riped that helped them to gain additional market share relative
to Florida growers over the last decade. On the other hand, the export production of Baja
California competes with the production of California that goes to the market from May
to December. Mature-green tomatoes are produced domestically in both Florida and
California. Florida tomatoes are shipped predominantly to eastern and mid-western
markets, while the western half of the country is served primarily by tomatoes from
California and Mexico (Sexton and Zhang, 2003).
72
The challenge to supply seasonal, perishable products year-round has favoured

Mexican exports and increased horizontal and vertical integration among shippers
regionally,
nationally
and internationally
(Cook,
2001).
In the grower-shipper
combination the shipper often joint ventures with growers to obtain the necessary
production and then markets for a fee, often advancing cartons, controlling the harvest
operations and imposing grades and standards to generate a consistent quality. In fact, the
process from production regions in Mexico to consumption centres in the US the
commodity receives marketing services that add value to the final consumers of the
product. While sometimes multinationals act as the grower-shipper, several large
Mexican growers have expanded their operations in this way in recent years (Lacroix, et
al., 2001). To illustrate the above, Figure 3.2 describes the tomato supply chain.
Fig 3.2 Production-Marketing Process for Exports of Mexican Tomatoes

Operation costs (inputs) = imports
Mexican
Growers
Broker/shipper
Exports
Wholesaler
Supermarkets
Terminal market.
(Quality, packing and safety requirements)

Land
Water
Electricity
Machinery
Labour
73
Fewer than a thousand farms dominate fresh tomato production in North America
and fewer than 50 shippers control the first marketing stage as tomatoes move into the
wholesale, retail and food service sectors. Often these shippers are also the growers. The
participation in the U.S. market by the main supplying regions is summarized in table 3.1.
Table 3.1 Main Suppliers of the Tomato US Market

Region
Number of
Size (acreage)
Percent of Regions Tomato
Grower-Shippers
Shipments
California
15
650-6500
80
Florida
800-7500
75
Mexico
14
500-6000
60 (Exports)
Source: Calvin and Barrios, (1998).
3.6.2. Grain Industry and Agricultural Policies in Mexico

While some grain producers in Mexico hold large farms, modem technology and
production practices, a more representative Mexican producer has access to roughly 10
hectares of farmland. Indeed, according to Mexicos 1991 National Agricultural and
Livestock Census, 61 percent of the farms where com was the principal crop were
smaller than 5 hectares (INEGI). Census data also reveal that only 31 percent of all com
farms used improved varieties of com, 35 percent had tractors and 9 percent had access to
irrigation, a critical input to the Mexican com sector (Nadal 2004). Efforts to improve
com production in Mexico have raised yields to about 6 metric tons per hectare on
irrigated land and 2.0 metric tons per hectare on rainfed land during the 1990s.
In the last decades and especially in the last ten years, the production of basic
grains in Mexico has been strongly affected by political, social and economic changes.
74
The growing integration of the economies and the implementation of structural

adjustment programs are the principal events that have influenced the consumption and
production of basic grains. The programs emphasized three types of policies: a decrease
or elimination of tariffs and controls on international trade; a decrease or elimination of
subsidies to consumers, along with an equalization of domestic producer prices with their
world equivalents; and a devaluation of real exchange rates, (Byerlee and Sain 1991).
In particular, the reduction of production incentives, the reduction of trade
barriers, the liberalization of input and product prices, the deregulation of the currency
exchange rate and the control of inflation have determined how basic grains are produced
and how they will be produced in the future (Sainz and Lopez-Pereyra, 1999).
Domestic prices during the 1970s, 1980s and part of the 1990s in Mexico were
maintained above the international prices. This trend was maintained even when
distortions in the exchange rate were decreased. With the second phase of the structural
adjustment programs, subsidies on maize production were eliminated and a system of
price bands was adopted. The system aims to generate greater efficiency in the allocation
of resources through the link between domestic and international prices. The government
minimizes the impacts of variability in the band by setting minimum and maximum
internal prices and by regulating imports. The results of this policy in the domestic
market depend strongly on the trend of maize prices in the international market, (Sainz
and Lopez-Pereyra, 1999).
Basic grains, including maize, are tradable commodities. The international prices
of each crop, as well as the prices of close substitutes in consumption and production and
of the inputs used to produce maize all play an important role in the economy of each
75
crop industry. Many changes in production and consumption trends have their origins in
changes in international prices. The economic policy implemented in the 1980s and
continued in the 1990s has tended to link domestic prices to international prices so that
the latter may serve as a reference in the allocation of resources. According to the above
reasoning, the maize economies cannot be understood without examining what has
happened to world market prices of maize, wheat (a substitute in consumption), sorghum
(a substitute in production) and fertilizer.
The price of maize in real terms fell in the international market during the 1970s
and most of the 1980s. By 1988 prices stabilized and then began to show a slight
increase during the 1990s. W ith respect to the price of maize substitutes, during the 1980s
and 1990s, wheat has become more expensive relative to maize in the international
market while sorghum has remained more or less stable (Sainz and Lopez-Pereyra, 1999).
With respect to maize consumption, total consumption is the aggregate of two
principal components: direct human consumption and indirect consumption as a
component of livestock feed (most in poultry, egg and pork production). White maize is
used for direct human consumption, while yellow maize is used mainly for feeding
livestock in Mexico. Growth in total maize consumption results from growth in both
components.
Given the preferences of people in Mexico for white maize, which is produced in
limited quantities in other regions and traded only in very small quantities on
international markets, growth in industrial processing of white maize may represent a
powerful incentive for increasing maize productivity. The maize price and the prices of
substitute crops and other relevant factors are involved in determining the demand for the
76
final product (that is, for poultry meat, eggs and pork since the demand for grains is
regarded as derived demand). Since maize and sorghum are the main components (in
volume) of concentrated feeds for the poultry industry, demand for both grains increased
proportionally. Maize and sorghum volumes used in manufacturing balanced feeds
depend fundamentally on the domestic availability of sorghum and its relative prices.
Maize and sorghum both provide substantial energy to animal diets.
In Mexico, growth in productivity rose considerably over the last decade.
Together with moderate growth in cultivated area, yield growth contributed to M exicos
high growth rate in maize production. Compared with Latin American countries and the
world in general, maize production and yields in Mexico during the last ten years grew at
a higher rate. Maize production in central and southern Mexico (the states of Jalisco,
Mexico, Oaxaca, Veracruz, Tabasco, Chiapas, Campeche, Yucatan and Quintana Roo)
represents approximately 60 percent of the countrys total production (SAGARPA,
2004). Growth rates vary among the states of centre and southern Mexico, but maize
production in the area as a whole rose by 4.0 percent per year.
While yellow com accounts for the bulk of U.S. com production, white com
dominates production in Mexico. In the feed market, yellow com from the United States
is supplementing Mexican production, which is clearly insufficient to meet growing
domestic demand.
The majority of Mexican imports of com consist of yellow com (around 80
percent), which is primarily used in Mexico as an ingredient in animal feed. From 1998
to 2002, Mexico also imported significant quantities of white com, which is used to make
traditional food. Of total imports, the U.S. supplies about 95 percent of total com
77
imported into Mexico. Mexico has pursued a transitional policy toward the United States
com which is more liberal than that required under the NAFTA. With the loosening of
Mexican trade restrictions, U.S. com exports to Mexico have increased dramatically to
about 7.7 million metric tons in 2003 (USDA 2004). Over the next decade, Mexican com
demand is expected to follow recent trends. The demand for feed com is expected to
increase rapidly with per capita income as Mexicans incorporate more meat into their
diets, while the demand for food com is expected to expand with population growth,
(Zahniser and Coyle, 2004)
Since the NAFTA took effect in 1994, Mexican grain imports have increased,
particularly com imports have increased by 240 percent relative to the average annual
imports from 1984 to 1993. Similar trends are also observed for sorghum, wheat and
soybean imports. To ensure that domestic demand for com is fully met, however, the
Mexican Government has issued additional import permits beyond the amount required
by NAFTA ((Zahniser and Coyle, 2004). In the late 1990s, white com emerged as a
significant component of Mexican imports from U.S. From 1998 to 2002, white com
made up about 15 percent of U.S. com exports to Mexico, compared with roughly 2
percent during 1991- 93 (USDA-ERS, 2004). Mexican com imports follow a different
channel depending on the end-use. The hog and poultry industry import most of the
yellow com as well as sorghum and soybean, while large flour companies account for
most of the white com and wheat imports.
78
3.6.3. Milk Powder Production and Policy

Mexico is one of the world's largest importers of dairy products and the world's largest
importer of non-fat dry milk (NFDM). Changes in the demand for dairy products will
therefore influence world markets. The United States, the European Union and New
Zealand as the largest exporters of dairy products to Mexico. The North American Free
Trade Agreement (NAFTA) and the proximity of Mexico to the US have already
intensified interest in US dairy product exports to Mexico. In the wake of the Uruguay
Round Agreements (URA) of General Agreement on Tariffs and Trade (GATT) and
changing economic conditions in Mexico and the impacts of NAFTA, the demand for
dairy products took increased importance. The Mexican government is the sole importer
of NFDM, 70 percent of which is distributed to the social sector in Mexico, while the rest
is sold to private companies. Most of the imported milk powder is reconstituted or
transformed into milk products such as cheese and yoghurt. The milk subsidy program is
undergoing some privatization, this move and other changes in government policy are
likely to affect future imports of non-fat dry milk considerably. As incomes grow,
Mexico is expected to become a larger market for fluid milk, butter, cheese and yogurt.
Imports of fluid milk, cheese and whey exhibit the fastest growth rates among dairy
products, increasing by 201, lO land 887 percent respectively, since 1990. Imports of
other products such as non-fat dry milk, butter, yogurt and ice cream have also increased.
The dairy sector is an important economic industry since it contributes 23 percent
of the value of the livestock sub-sector, but domestic milk production is not enough to
supply the requirements of milk demand. The Mexican dairy sector is characterized by a
long-standing tradition of government intervention in the form of price policies and
79
subsidies. Agricultural performance was quite favorable in the early eighties, strongly
supported by subsidies as part of the Sistema Alimentaria Mexicana (SAM). In 1982,
following a drought, agricultural GDP fell 2.9 percent and fixed investment, wages and
cultivated land diminished significantly. Also agricultural imports fell by 54.6 percent in
the same year (Villa-Issa, 1990).
In 1982, several government policies were instituted which influenced the dairy
sector. Price controls were put in at every level of milk production and marketing
(production, processing and consumption). Retail prices were fixed despite rising
production costs. This led to a slow expansion in milk production, chronic production
shortfalls and increasing imports.
In 1990, the milk system had a nutritional dependency coefficient of 38 percent
(calculated as imports/disposable milk (imports + production)), which has been steadily
increasing over the past four years. Milk and dairy products as a group were the principal
food imports in 1993 with a volume of 445 000 t and a value of $626.3 million. This
shows a tendency to increase dependency on imports in both volume and value terms
(Munoz et al., 1994). Mexican dairy policy has tended to favour the consumers at the
expense of the dairy producers. Consequently, Mexican milk has the lowest producer
subsidy equivalent (PSE) of -5 6 percent (Munoz and Odermatt, 1993). This negative PSE
reflects a tax on producers rather than a subsidy. The principal factor that explains the
negative PSE is low producer prices.
Until 1988, producers were forced to sell their milk at below market prices.
Decapitalization forced many farmers out of production, especially those who produced
low quality milk. Price controls on milk led to the diversification in the use of fluid milk
80
by using milk for products whose prices were not controlled (such as cheese, yogurt, and
butter), adulteration of products by substituting up to 80 percent vegetable fat instead of
butterfat, distribution of milk through less controlled channels, and increased vertical
integration of dairy farmers. All these decreased the production of milk which led to an
increase in social programs of CONASUPO (Munoz and Odermatt, 1993).
Dairy-product consumption shows distinct patterns in Mexico, where per-capita
consumption of fluid milk is much less when compared to that of the EU or the US.
However, per-capita consumption of NFDM is two to three times more than in the US or
EU. In 1991, per-capita fluid milk consumption was 46 percent of US per-capita
consumption and NFDM consumption was about 288 percent of US consumption. Milk
is not a traditional consumption item and protein source in Mexico. The high
consumption of NFDM occurs because it is easy to store and transport without
refrigeration and also because of its price.
Cheese is a very important item in the daily diet of the average Mexican
household. It is consumed either as an appetizer or topping, or as a main dish. Different
types of cheese are used for different purposes. However, the majority of the Mexican
consumers prefer fresh cheeses to aged ones. Tastes and preferences are also influenced
by income level where fresh cheeses are consumed at lower-income level and aged
cheeses consumed more at middle and high-income level (Munoz and Odermatt 1993). In
1992, 10 percent of total cheese consumption was made up of imports which were mostly
hard or semi-hard cheeses.
CONASUPO has existed in Mexico since 1938 under various names, adopting its
current name in 1965. CONASUPO is involved in the distribution of basic foods to the
81
poor through subsidized programs. In the 1970s, CONASUPO handled 23 products, in

establishing tariffs, licenses for imports and guaranteed prices. By the 1990s, most of
CONASUPO's operations were concentrated on distribution of com, dry beans and
NFDM under guaranteed prices. CONASUPO buys com and sells it at a subsidized price
through public tender. It contracts tortilla factories to produce tortillas and distribute to
the poor. CONASUPO also buys dry beans via warehouses during harvest and sells the
beans at lower than market prices. In 1994, CONASUPO was responsible for the
distribution of 40-50 percent of pasteurized milk and 30 percent of NFDM. About 70
percent of its milk was distributed to the poor. Leche Industrializada CONASUPO
(LICONSA) reconstitutes milk powder and distributes the milk through outlets around
the country. CONASUPO was the only importer of NFDM in 1995. However,
CONASUPO's role as a monopoly or monopsony was changing over time and by 2000
CONASUPO has already privatised its plants.
Prior to 1986, the Mexican dairy sector like the rest of the Mexican economy was
characterized by high tariffs, prevalent non-tariff barriers, such as licensing and extensive
government involvement in import purchases. Since Mexico's entrance into GATT in
1986 and the beginning of a more open economy, trade barriers have been lowered and
import licenses requirement eliminated. In early 1992, CONASUPO began negotiating
direct purchases with individual bids instead of using public tenders as it had previously
(Knutson et al., 1993). These NFDM purchases were held as stocks and in turn sold to
Mexican dairy product producers. Under the NAFFA, Mexico converted its import
license for NFDM to a tariff-rate-quota (TRQ) to be phased out over fifteen years. For the
US, the first 881,840 cwt of skimmed and whole milk powder will enter the Mexican
82
market duty free. Imports over the quota level will be assessed at a tariff of 133.4 percent.
This tariff will gradually be phased out by the year 2008.
3.7. Economic Framework

Batra and Ullah (1974) examined the behaviour of the competitive firm faced with
making input hiring decisions under conditions of price uncertainty. They showed that a
marginal increase in uncertainty stimulates a decline in the firms output, provided the
coefficient of absolute risk-aversion is decreasing. The assumption of decreasing absolute
risk aversion is based on the attractive features of these types of functions to produce
meaningful economic outcomes.
To perform this research the modeling strategy consists of the following steps: to
outline a theoretical model capable of linking the microeconomics of the specific
industries outlined in the above section with the international trade activity. The
theoretical model is used to generate import demand functions. The comparative statics
of the model enable us to study the relationship between trade flows and the exchange
rate and its volatility. The basic economic theory underlying this study is the utility
maximization under uncertainty and the international trade theory.
3.7.1. Measures of Risk Aversion

It has been shown in the literature that if the utility function is to satisfy the Von
Newmann-Morgenstem axioms, it must be bounded from above. This means that as
profits approach infinity, the utility function must become concave with U"< 0 , while it
is U"= 0 when a agent is risk neutral. So, the marginal utility of income is decreasing.
83
Individuals with such utility functions will be risk-averse. Absolute risk aversion captures
the above features. A unique measure of the direction of bending and the rate of change
in slope of the utility function is the absolute risk-aversion defined by Arrow and Pratt as:
U\x)
Hence, the degree of absolute risk aversion is higher as the curvature of the utility
function is larger. The degree of absolute risk aversion may change as the wealth level
changes, i.e. RA may be a function of the level of wealth. In this way, it is possible to
classify utility functions with diminishing absolute risk aversion (DARA), where RA(x) is
decreasing in wealth. In other words, the degree of absolute risk aversion diminishes as
the wealth level increases. When there is constant absolute risk aversion (CARA), RA(x)
is constant in wealth, which means that the degree of absolute risk aversion remains the
same as the wealth level increases. However, with Increasing Absolute Risk Aversion
(IARA), the degree of absolute risk aversion increases as the wealth level increases.
Arrow (1970) showed that with decreasing absolute risk aversion, the absolute
demand for risky investment increases as the wealth level increases. Furthermore, with
increasing absolute risk aversion, the demand for risky investment decreases as the
wealth increases. Because the demand for risky investment should increase as wealth
increases an agents utility function should have a feature of DARA.
3.7.2. The Theory of Input Demand under Uncertainty

The theory of input demand for a competitive firm developed by Batra and Ullah (1974)
is used as a starting point for the economic framework used in this research.
The
assumptions of the basic model are as follows: the firm in this model is not large enough
84
to influence its subjective probability distribution about its sales price, which is assumed
to be a random variable. Therefore, the firm is a price taker. Input prices are given for the
firm. The firms decision regarding input utilization and volume of output must be made
prior to the knowledge of the market price. The firm seeks to maximize expected utility
from profits and its attitudes toward risk can be described by a Von NewmannMorgenstem utility function, i.e., the utility function is such that: U = U (tc), where U (7i)
> 0 and U (7t) is less, equal or greater than zero depending on whether the firm is riskaverse, risk neutral or risk referrer. The firms profit is given by
n = pq - w L - r K ,
(3.25)
where P is the product price which is assumed to be random, with density function f(p)
and the mean E(p) = p. The expected utility of profits is given by
E {U [P f(K,L) - wL - rK]} ,
(3.26)
where E is the expectation operator, f ( ) have isoquants that are convex to the origin, that
is, 2 f Kf Lf K i - f t f x K ~ f x f u . > 0 The firm chooses input quantities so as to maximize
expected utility from profits. The first order conditions for the maximum are
oL
= E[U \ n ) { p f L - w)\ = 0 ,
(3.27)
and
(3.28)
= E [{ /W O * -r)] = 0 .
The second order conditions are given by
= A = E [ u "(K)(pfL - w f + pf LIU w ] < 0
= ^ = E[u
and
- r f + pfa v '(*)] < o ,
85
(3.29)
(3.30)
and W
- B f = D > 0,
where, Bi = 9 f ^ = E[U' \7t)(pfL - w)(pfK - r ) + p f KLU \ k )]

oLoK
(3.31)
In the certainty case these conditions ensure utility maximization assuming that
the production function is strictly concave. Under a random price, assuming risk
aversion, strict concavity of the production function is sufficient but not necessary to
ensure expected utility maximization.
It can be shown that under price uncertainty, the risk-averse competitive firm
demands smaller quantities of inputs and produce lower output than in the case under
uncertainty. The first order conditions (3.27) and (3.28) are
E [ U \ k )P - f L] = E[U' ( x) - w]
and
(3.32)
E [U 'Or) p - f K] = E[ U Or) r] ,
(3.33)
Subtracting E[ U '(tt) ju- f L] from both sides of (3.32) we obtain

E [ U ' ( x ) - f L( p - M ) ] = E [ U ' m w - j u f L].
( 3 -34 )
Applying expectation operator to both sides of equation (3.25) yields the

expression E( k ) ~ / i q - - { w L + r K ) . Substituting the expression for - ( w L + r K ) in
(3.27)
then,
K = E(ju) + ( p - fi) q .
U'(7t) <U'[E{7tj\
for
p > //
Assuming
or
strict
multiplying
concavity
of
through
by
U (-),
then
(p - j u ) f L,
f L( p ~ M ) - U '[E(n:)]fL( p - j u ) for all p. Taking the expectation, noting that

U'[E(7 t)] is a given number and that f i is non-random, then E[ U '(7r)fL( p -
ju ) ] <
0.
Using this result in (3.34) implies that E[U '(7r)(w- jUfL] < 0 and since marginal utility is
86
always positive, w < j l f L. Similarly can be shown that r < jufK . The above exposition
shows that under uncertainty, the expected marginal value product of each factor exceeds
its price or exceeds marginal cost. Thus, the optimal output produced under uncertainty
will be lower than that produced under certainty.
3.7.2.1 The Supply Function

This section shows how the firms output responds to a change in the expected price. Let
price be defined as (p + 0) where 0 is a shift parameter. A change in 0 shifts the
probability distribution to the left or to the right and changes the expected value of price
without altering the shape of the probability distribution. Differentiating (3.27) and (3.28)
with respect to 0 and evaluating the derivative at 0 = 0 yields
A H + A H = - E [ U W ( p f L - w ) - q + f L -U '(*)] = FJ and
off
off
(3.35)
B<B
11 =-E\ V - r > 4
off + ^ off
(336)
+h V W ] =
The solution to this system yields:

dL = A2Fl - B lF2
dff
dK _ AtF2- B t F x
dff~
Since
(3.38)
and A 2 are less than zero, and D > 0 the problem now is to sign the term Bi Fi
and F 2. Using the first order conditions and equation (3.31).

Bt = E [ u (n)(pfK - r ) \ f L / f K) + f KLU \ n ) - p] = E [ u \ n ) { p f L - w ) \ f K / f L) + f KLU \ jc) p ]
87
Using (3.36), the expression for Bi and the expression for Fi and F 2 in (3.37) and (3.38)
are
3L E ( U ' p ) { E ^ W ) H q l h ) E [ U ' ( p f K- r ) } } ( f Kf KL- f J KK)
= ----------- 1------------------------------------------de
d
dK
and
(3.39)
E(U { E(U ) + ( g / f L)E[U \ p f L - w)]] ( / t / -
D
Since E [ U " ( p - f K - r ) ] and E[U"( p- f L- w ) \
are positive under the hypothesis of
decreasing absolute risk-aversion, or increasing absolute risk-aversion and also assuming

that f u < 0 and f KL > 0 , then, dL/ d&> 0 and dK I dO> 0 . These results continue to hold
if the firm is risk neutral so that U = 0. In other words, a rise in the expected price will
induce the firm to hire more inputs. The expression for the change in output can be
obtained by using (3.39) and (3.40). Thus
dq
E W p ^ E i J J d H q l f K) E [ W ( p f K - r j \ } ( 2 f J J KL- f Lf KK)
'
Risk neutrality, decreasing absolute risk-aversion or increasing absolute risk-aversion are

sufficient to ensure a rise in the expected price will induce the firm to increase its output.
3.7.2.2 The Import Demand Function

This section shows how a change in input price affects the firms demand for factors. It is
assumed that there is a change in the wage rate but the price of the other input does not
change. Differentiating the first order conditions in equations (3.29) and (3.30) with
respect to w yields
Aj(9L/ dw) + B](dK / dw) = C, and
By0 L / dw) + A20 K / dw) = C 2 ,
88
where C, = L E[U"{7t){P f L - w)] + E [ U \ n j \ and C 2 = L E\(J"{ k ){P f K - r ) ] .

Following the same procedure as in previous section, then
dL
A2E(U') - f K + L - E ( U ' - p ) E [ U " ( p - f K - r ) l f Lf KK - f Kf KL)
dw
D -fL
dK
(3 .4 2 )
A1E ( U ' ) - f K + [ E ( U ' p ) l L E [ U " ( p - f L - w ) \ + E ( U ' p ) Y f Kf l x - f Lf KL)

D f L
(3.44)
The signs of the expressions given by (3.42), (3.43) and (3.44) depend on the
assumptions about the firms attitude to risk as well as the sign of f KLand f... In the
certainty case we have d L / d w < 0 due to concavity, but the signs of dK / dw and d q / d w
depend on the sign of f KL in addition to concavity. The main result of this discussion is
that with a well-behaved production function and the hypothesis concerning the firm s
risk attitude, a rise in the price of any input induces the firm to reduce the demand for that
input and lower its output. However, the firm s response toward the demand for the other
input is indeterminate. Under risk-aversion, firms utilize smaller quantities of inputs and
thus produce a lower output than a firm operating under certainty, even though the former
minimizes the unit cost of production of whatever output it chooses to produce.
3.7.3. Derivation of the Import Demand function

The estimation of import or export functions for various countries or various
commodities have received a great deal of attention in the empirical literature of
agricultural trade. The demand for traded goods is usually written as a linear or log-linear
89
function of real income and the price of the traded goods relative to the price of domestic
substitutes and other relevant factors. Depending upon whether traded goods are
considered as intermediated goods or finished products, such demand functions for
imports can be derived from conventional production or utility theory. It is clear,
however, that imports can be expressed as a function of output or income and one relative
price only if imports and domestic goods can be aggregated.
In this model production theory is used to theoretically derive the import demand
functions by treating imports as inputs to the domestic technology. It is assumed that
import and export decisions are made by profit maximizing firms which operate under
perfect competition in all commodities and factor markets. Firms choose their optimal
output mix and their input requirements subject to a vector of output and import prices.
These domestic factors are assumed to be mobile between firms and their rental rates are
determined by their marginal products.
The assumption that the commodities in this study (tomatoes, maize, sorghum and
milk powder) are intermediate goods is based on the fact that tomatoes in the United
States and maize, sorghum and milk powder in Mexico go through a value added process
as explained in section 3.6.
Most of the empirical literature dealing with trade under uncertainty has evolved
from finance literature. Thus, most authors regresses some measure of exchange rate
volatility and sometimes the level of the exchange rate on the volume of imports or
exports. In most cases, the functional form is either linear or log-linear. This amounts to
using a rather simplistic specification of import and export functions. Indeed, until two
decades ago, nearly all the empirical work on import and export determination relied on
90
such ad hoc specifications, where the quantities of import and exports were written as
linear or log-linear functions of relative prices and other factors. These approaches have
some disadvantages. In particular, they are based on little or no economic theory. More
seriously, such demand and supply functions are not consistent with well-behaved utility
or production functions, except in very restrictive cases.
For the model used in this research imports are used together with domestic inputs
and services to produce output that can be absorbed at home or exported. In this line of
research, using production theory, Kohli (1982) disaggregated output and modeled the
supply of exports. Appelbaum and Kohli (1979) allowed for non-competitive behaviour
and tested for departures from price taking behaviour. Furthermore, Appelbaum and
Kohli (1997) modeled import demand function under uncertainty. They argued that
uncertainty must be taken into account explicitly if import decisions and its role as a
deterrent of international trade need to be assessed.
In this research, using the same
framework, an attempt is made to model import demand under exchange rate uncertainty.
Exchange rate uncertainty enters the model affecting both firms revenue and firms
expenditure on imported inputs.
Import demand function will now be derived. Let q = f (Xy Xm, K ) be a
neoclassical production function. Assume that f(-) is continuous, non-decreasing, linearly
homogeneous and strictly quasi-concave or concave, (The assumption of a well-behaved
production function hold for both the theoretical and empirical derivations). Where X l is
the input labour, Xm represents the quantity of the imported input needed to produce
output q and K is capital. Output can be absorbed in the domestic market or can be
exported. It is assumed that the only source of risk is the uncertain exchange rate.
91
Therefore the exchange rate is a random variable and by implication, so is the foreign
price and profits. The profit maximization problem of the firm can be represented as:
Max^ . v , { { /[( + 1) / / ( X L, X M,K ) - w LX L-(R+ tfM)w MX M- w KJf]}}
where R - R + 0i the currency price of imports (foreign currency) and 0 is a random

variable distributed according to the density function g(Q), with E(0) = 0, (so that E(R) =
R ) and Var(R)= Var(6) = er2 . It is assumed that U'(-) is a Von-Newman-Morgestem
utility function with /'()> 0- The solution to the firm s problem defines the (dual)
indirect (expected) utility function V which is
MaxytXE{u[(R + 9X)P y - wLX L ~ ( R + 9 M)wMX M - w k k ] : y < F(*)}
(3-45)
where p represents higher moments of g(0) and the random variable 9 is continuous and
convex to the moments (Proofs are in Appelbaum, 1993). The firms demand and supply
function can be obtained from the above indirect expected utility function by applying
envelope theorem. Hence
dV
dw,
| = = - [ /' (*)] y .
Therefore the firms input demand and output supply functions are given by
v /, -a
v
x , = -d
dw (.
ow K
and
(3.46)
yK
92
3.7.4 Comparative Statics for the Model

Comparative statics show how the input demand and output supply functions behave as
input prices, output prices and the random exchange rate change. By maximizing
expected utility of profits the first order conditions are obtained as
E U '(7r)[(R + 0{) P f L( X L, X m,K) - w L] = 0 and
(3.48)
EU \ k ) [ ( I + Qx)P f M(XL, X M,K ) - (R+ 0M) w M] = 0 .
(3.49)
The above conditions can be rewritten as

( ^ + ^ ) P - / l (X l , X m, K) =
w l
and
(3.50)
(R + 91) P - f M( X h, X M, K) = ( R + 9 M)-wM ,
(3.51)
where 9 = Cov [ U (n),v]/E[U(n)]. 9 is positive, zero or negative as U "(-) is positive,

zero or negative (i.e as the agent is risk-averse, risk neutral, or risk lover). The term
wM is the full marginal cost of imports and 9m represents the marginal cost of
uncertainty (risk premium). It follows from (3.51) that the presence of uncertainty will
generally lead the value of the marginal product of imports to deviate from the expected
marginal cost of imported products. Particularly under risk-aversion, 0 ispositive so the
value of the marginal product of imports will exceed their expected market price. Thus,
this implies that the quantity of imports will be smaller under uncertainty. The solution to
the system (3.50) and (3.51) is
X ^X liw ^w ^p J^),
(3.52)
X ,M =X*M(wL,wM, p , R , 9 M).
(3.53)
These relations indicate the amount of each factor that will be hired as a function
of the factor and product price; they are the choice functions of this model. Assuming that
93
it is possible to solve for equations (3.52) and (3.53), it becomes meaningful to perform
comparative statics of the profit maximization model to know the changes in factor
employment due to given changes in prices and exchange rate.
Substituting equations (3.52) and (3.53) back into equations (3.50) and (3.51) results in
(R + 0l ) P - f L(X*L(wL,wM, p ) , X tM(wL,wM, p), K ) - w l = 0 and
(3.54)
(R + 0 J P- f M( X l ( w L,wu , p), X*u (wL,wM, p ) , K ) - ( R + 0 M)-wM =O,
(3.55)
3.7.4.1 The Response of Import Demand Function to Changes in Price of Inputs:

The relations (3.54) and (3.55) are identities in input and output price. Hence, the
assertion that a firm always obeys equations (3.48) and (3.49) for any prices, means that
one must convert these equations to the identities such as (3.54) and (3.55). Being
identities, these equations can be differentiated implicitly with respect to input and output
prices, producing relations that express changes in input and output quantities given
changes in prices. The assumption of a risk-averse agents and a well-behaved production
function hold for the comparative statics. For all Differentiation of equations (3.54) and
(3.55) with respect to wLyields
( R + 0 l ) p . f LL^ L + ( R + 0l ) p . f l M^ L - l = O and
dwL
dwL
(3.56)
( R + 0 , ) P f KL^ + ( R + 0 t ) P f u t f ^
O = O.
owL
dwL
(3.57)
dX
To solve for - multiply equation (3.56) by / mm and equation (3.57) by/uvi to get
dwL
( R + 0 l)P f MMf u , ^ + (R + 0t) P - f lMf m ^ M f m = 0 and

dwL
dwL
94
(3.58)
(R + el) P f L ' ^ + ( R + e i)PfMMfm i ^ - - = o .

dw,L
(3.59)
dw,
'J y y L
Subtracting equation (3.58) from equation (3.59) and factoring yields
= ^ ----------fj m ----------
<Q
(3.60)
This expression is evidently negative under the hypothesis of risk-aversion and a well
behaved production function since denominator is positive. In other words, demand of
labor responds negatively to changes in its price. Differentiation of equations (3.54) and
(3.55) with respect to wm yields
( R + d1) p . f lL^
^
+ ( R + 0 l) p . f lM^ L ^ - O
dwM dR
owM dR
= O and
(3.61)
( + g , ) f - / , a f Ct
+(R + el) P - f m
~(R + e,)= o.
dwM dR
dwM dR
(3.62)
3X *
To solve for - ----- multiply equation (3.61) b y / lm and equation (3.62) by/L L to get
dw2 dR
( R + e , ) P - f u , f u. ^ - ^ - + (R + ^ ) P - f m ^ JL^ S - - 0 = 0
dwM dR
dwM dR
and
( R + e l) P . f l j KL^ . ^ M . + ( R + e l) p - f u. f m ^ - ^ - < . R + e l) f u. = o .
@R
OWj^ oR
(3.63)
(3.64)

dX' d wu
dR
(R+0,)U
<0
(3.65)
( * + 0 ,W u / , - / i )
This expression indicates that demand of imported inputs responds negatively to changes
in its price and the exchange rate.
95
3.7.4.2 The Response of Import Demand to Output Price and Exchange Rate:
Differentiation of equations (3.54) and (3.55) with respect to p yields
(R + et) P - f u. ^ ^ + <,R + 0 J P - f u, Q x - ^ = HR + et) f L and

J dp dR
(3.66)
d x l dP
( R + <?,)P f ML
+ (R + 0,)P
1 dP dR
0-67)
+ )/
d x[ dP
To solve for - multiply equation (3.66) by /mm and equation (3.67) b y / m to get
dP dR
(R + %
+W
- f
(it + 0 , ) P - f L 1 / 1 | + ( R + 4 ) P f u , S mm
f w
f /
~ (* + 3 ) / /
(3 6 8 )
(3.69)
dX'L dP
dP dR
- ( R + f f ,)[/,/ + / , / , J
(R +
- f m)
> 0
This expression is indicates that demand of domestic input responds positively to

changes in output price and the exchange rate. Similarly, differentiation of first order
conditions with respect to P, but solving for
a x ; dp
dp dR
yields
- q + 3 )[/ / + / , / , . ] ^ 0
(+a,)F(/u / - / i )
Again, under the hypothesis of risk-aversion and assuming negative sign for the cross
partial derivative, the demand for imports respond positively to changes in output price
and the exchange rate.
96
3.7.4.3 The Response of Import Prices and Output to Changes in Exchange Rate
To know the effect of changes in exchange rate on the price of imports differentiate
(3.54) with respect to wm , yielding
dwM dR
+ ( R + 0 l) P - f ul
dwM dR
-0 = 0.
(3.72)
Solving (3.66) for , gives an expression that indicates the effect on the input price
dR
of changes in exchange rate
dWM ~
dR
(R + 0 J P
>0
1
dX^
' LL
dwM
(3.73)
dxl
J LM
dwM
Equation (3.73) indicates that the price of imports responds positively to changes in
exchange rate. To know the effect of changes in exchange rate on price of output, solve
dP
equation (3.67) for to obtain expression (3.68), which indicates that the output price
dR
respond positively given a devaluation of the Mexico-US exchange rate
d X M dP _
(R + 9\)\ f m f mm + f t f i M ] > q
dP dR
2 dX L
74 )
dp
Summary on Comparative Statics
The comparative statics results the profit maximization model show how the import
demand function behave as input prices, output prices, the random exchange rate and its
volatility change values. The main results from the comparative statics are as follows:
97
While the import demand (traded volume) responds negatively to exchange

rate uncertainty, it responds positively to changes in exchange rate
The price of imports responds positively to changes in exchange rate
Demand of domestic input responds positively to changes in output price
Demand of labor responds negatively to changes in its price.
Demand of imported inputs responds negatively to changes in its price.
The comparative statics results show theoretically how the import demand function
change due to changes in a number of explanatory variables. These results are used to
guide the empirical analysis presented in chapter six and chapter seven.
3.7.5. Empirical Implementation

Once the theoretical framework has been developed and comparative statics results are
derived, one can proceed with the empirical specification of the import demand functions.
The simplest and most widely used procedure for estimating aggregate import demand is
the use of a demand function relating the total quantity of imports demanded by a country
to the level of its real expenditure or real income and to the price of imports and domestic
substitutes measured in the same currency. The assumption of differentiated products is
made for the empirical analysis. This implies that domestic and imported commodities
create different consumer utility. Appelbaum and Kohli (1997) argued that uncertainty
must be taken into account explicitly if import decisions and its role as a deterrent of
international trade need to be assessed. The exchange rate variable and the volatility of
98
the exchange rate variable are introduced directly into the import demand functions to be
estimated. The general formulation of this imports demand function is:
Qd = F(Y, ER, V, Pc, Ps )
(3.75)
where Y is real per capita disposable income

ER is real exchange rate ($MX/$US)
V is a volatility of the exchange rate measure
Pc is the border price of the commodity
Ps is the price of a substitute or competing product
The import demand function is a function of income, prices, exchange rate and
exchange rate uncertainty. In this research uncertainty of the exchange rate is represented
by the volatility of exchange rate. The random variable is distributed according to the
density function g(6), with E(d) = 0, so that E(R) = R and Var(R)= Var (d)= cr2. Since the
moments of the distribution g(6) are unknown, they need to be estimated in order to
proceed with the estimation of the import demand. The moments of the g(0) distribution
are estimated using a GARCH and a nonparametric method, and two commonly used
procedures.
As for the appropriate specification of import demand, the theory does not provide
any specific direction on the best functional form and the most appropriate measures of
the variables to be used in the analysis. Thursby and Thursby (1984) tested nine most
common specifications of aggregate import demand functions for the United States and
found the appropriate functional form to be log-linear rather. Import demand equation is
specified in logarithmic form also because of its convenience and ease of interpretation of
the results.
99
3.8 Empirical Estimation of Equations

Based on the theoretical framework presented above, the Unites States import demand
function for tomatoes from Mexico and Mexicos import demand functions for maize,
sorghum and milk powder from the Unites States are estimated. For each of the selected
commodities, a reduced-form equation is estimated employing monthly data from 1989:1
to 2004:12.
1. Tomato Equation; For fresh tomatoes the following equation was estimated:
Qt* = 0 0 + Pi USY + p2Pt + p3Pus + p 4W + psER + p 6Vi + p 7D t + et
(3.76)
where:
dQ *
_
>0
dUSY
dQ * _
-^ -< 0
dPt
dQ *
- ^ > 0
dPus
dQ *
- ^ > 0
dW
dQ * _
>0
dER
dQ * _ dQ *
< 0 >0
dVt
dDt
Definition of variables:
Q* is quantity of imports; it represents the volume of fresh tomato (excluding green
house tomato) imported from Mexico to United States. It is measured in metric tons.
USY is United States per-capita income (importers income), the nominal US per-capita
disposable income was deflated using the United states Consumer Price Index (CPI) to
obtain real per-capita disposable income in the United States. W is the ratio of United
States-Mexico farm wage rate (expressed in US dollars per hour and has been deflated
using the United States CPI). It represents a ratio of unit cost of production, Pt is the
border fresh tomato price (in US dollars per kilogram), it is the quotient of the total value
of Mexican exports to US and the volume (in metric tons) of Mexican exports. All border
prices and Unites States values were converted to real terms using the US CPI. Pus is the
100
price of a substitute, in the case of tomatoes is the tomato price in the Unites States (in
US dollars per kilogram). This variable was selected basically because using the domestic
tomato price in the Unites States provides theoretically expected signs. D; is NAFTA and
seasonal dummy variables to indicate the effects of liberalization (the reduction or
elimination of tariffs) and seasonal effects respectively. ER is the Mexico-Unites States
exchange rate (pesos per US dollar). V) are measures of exchange rate volatility. Four
different volatility measures were tested for each commodity. Table 3.2 provides a
description of the variables and a summary statistics.
The expected sign for income variable (USY) is positive indicating that as real
income in the importing country rise so will the quantity imported. The expected sign for
Pt is negative and for Pus is positive indicating an inverse relationship between quantity
demanded and own price and a direct relationship between import demand and related
commodity price. The ratio of wage rates is expected to have a positive coefficient. A
positive coefficient is expected for the foreign exchange variable because it is assumed
that appreciation of the Unites States dollar lead to an increase in imports as Mexican
products become relatively cheaper to those in the United States. Assuming risk-averse
traders, the coefficients of the measures of volatility are expected to be negative.
The trade liberalization dummy variable is expected to have a positive coefficient
because the hypothesis of free trade is that trade liberalization increases trade flows and
wealth of nations.
101
Table 3.2 Description and Specification of Variables for Tomato Equation

Variable
Description
Units
M ean
Variance
St Dev.
Qt*
Quantity o f exports represents the volume

of tomato exported by Mexico to US.
Metric Ton
46,337
98334425
31,358
USY
Importers income is the US per-capita

disposable income.
US$/Person
6401.617
844031.79
918.712
Pt
The border price is the quotient o f the total

value o f Mexican exports to US and the
volume o f these exports.
US$/kg
0.75029
0.07672
0.27699
Pus
Pus is the price of a substitute, in this case is

the tomato price in the Unites States.
US$/Kg
0.77361
0.30482
0.55211
W represents the unit cost of production,

proxied by the ration of the US and
Mexican real farm wage rates.
US$/hour
6.7392
0.9424
0.9708
ER
ER is the Mexican peso per US dollar

exchange rate.
MX$/US$
6.84389
10.2334
3.19897
2. Corn Trade Equation

The specification of the maize import demand function for Mexico includes domestic
per-capita income, foreign price, a price for a related good represented by the price of
domestic maize, exchange rate, exchange rate volatility, a free trade dummy variable and
an additional exogenous variable. For the case of maize imports from the United States
into Mexico the following equation was estimated:
(3.77)
Qc* = p0 + p iM X Y + p 2Pc + p 3Pcd + P 4 N V + fisDi + PeER + p 7Vt + et

where:
3 2 . >o
dMXY
3 g . <()
dQ'
dPc
dPcd
>0
dQ'
>0
dINV
dQ '
dER
>0
dQ *
dV;
<0
102
dQ*
dD,
>0
Here Qc* is quantity of imported maize for Mexico from the Unites States. It represents
the volume of maize that Mexican importers import from the Unites States. It is measured
in metric tons. MXY is importers income, the original series was in constant pesos
(1993=100) so, series was divided by the population to obtain the real Mexican GDP per
person. Pc is the border price (in US$/kg) of the commodity, it is the quotient of the total
value (in US dollars) of Mexican imports from Unites States and the volume (in metric
tons) of Mexican imports. All border pieces and Unites States values were converted to
real terms by using the US CPI. Pcd is the price of a substitute. The substitute of
imported com is domestic com. The variable INV is the inventory of hogs on feed.
Dj
is
NAFTA or seasonal dummy variable to indicate the effects of liberalization (the

reduction of tariffs). ER is the Mexican peso per US dollar exchange rate. V) is a
measures of exchange rate volatility. Four different volatility measures were tested, one
at a time, in each equation (Table 3.3).
The expected sign for income variable (MXY) is positive indicating that as real
income in the importing country (Mexico) rises so the quantity imported. The expected
sign for Pc is negative and for Pcd is positive indicating an inverse relationship between
quantity demanded and consumer price and a positive relationship between import
demand and related price respectively. A positive coefficient is expected for the INV
variable. A positive coefficient is expected for the foreign exchange variable because
appreciation of the US dollar leads to a decrease in imports from US as maize in the US
become relatively expensive to those in Mexico. Assuming risk-averse traders, the
103
coefficients of the measures of volatility are expected to be negative. The dummy

variable is expected to have a positive coefficient.
Table 3.3 Description and Specification o f Variables for Corn Equation

Description
Quantity o f imports represents the
volume of corn imported by Mexico
from US.
Importers income is the MX per-capita
GDP.
Units
M ean
Variance
St Dev.
Metric Ton
328953
61987948
248973.7
MX$/person
2977.21
1869674.5
1367.36
Pc
The border price is the quotient of the

total value o f Mexican imports from US
and the volume of these imports.
US$/kg
0.13667
0.003009
0.05485
Pcd
Ps is the price o f a substitute, in this

case is the price of domestic maize in
Mexico
INV is the inventory o f hogs on feed in
Mexico.
Units
11971
0.1016E+9
10083
Million
Units
6.74318
1.20296
1.0968
ER

exchange rate.
MX$/US$
6.84389
10.2334
3.19897
Di
Dummy variables to represent trade

liberalization
Zero / one
...
Variable
Qc*
MXY
INV
3. Sorghum Trade Equation

The specification of the sorghum import demand function for Mexico includes domestic
per-capita income, foreign price, a price for a related good represented by price of
domestic sorghum, exchange rate, exchange rate volatility, a free trade dummy variable
and an additional exogenous variable represented by inventory of hogs on feed. For the
case of sorghum imports from the United States into Mexico the following equation was
estimated:
104
Qs* = fio + p iM X Y + p 2Ps + p 3Psd + P 4 N V + PsDi + PeER + p 7Vi + t
(3.78)
where:
dQ *
^
dMXY
>0
dQ *
-^-<0
dPs
dQ *
- ^ > 0
dPsd
dQ *

dINV
>
dQ * _
^ > 0
dER
dQ *
<
dVt
dQ * ,,
-^ > 0
dDi
Here: Qs* is quantity of imported sorghum. It represents the volume of sorghum that
Mexico imports from United States. It is measured in metric tons. MXY is importers
income; the original series was in constant pesos (1993=100) so, the series was divided
by the population to obtain the real Mexican GDP per person. Ps is the border price (in
US$/kg) of the commodity; it is the quotient of the total value (in US dollars) of Mexican
imports from United States and the volume (in metric tons) of Mexican imports. All
border pieces and United States values were converted to real terms by using the US CPI.
Psd is the price of a substitute. Because the bulk of sorghum imports go to feed industry,
the substitute of sorghum is the price of domestic sorghum. It is the price of domestic
sorghum in Mexico. D; is NAFTA or seasonal dummy variable to indicate the effects of
liberalization (the reduction of tariffs). ER is the Mexican peso per US dollar exchange
rate. V) is measures of exchange rate volatility. Four different volatility measures were
tested, one at a time, in each equation (Table 3.4).
income in the importing country rises so the quantity imported. The expected sign for Ps
is negative and for Psd is positive indicating an inverse relationship between quantity
demanded and own price and a direct relationship between demand and related price
105
respectively. A positive coefficient is expected for the foreign exchange variable because
appreciation of the US dollar leads to a decrease in imports from United States as
sorghum in the United States become relatively expensive to those in Mexico. Assuming
risk-averse traders, measures of volatility are expected to be negative. The dummy
variable is expected to have a positive coefficient.
Table 3.4 Description and Specification of Variables for Sorghum Equation

Variable
Mean
Variance
St Dev.
Metric Ton
286800
22199414
148994
Importers income is the MX per-capita GDP
MX$/person
2977
1869674
1367.3
The border price is the quotient of the total

value o f Mexican imports from US and the
volume of these imports.
US$/kg
0.114
0.00030
0.01736
US$/kg
0.130
0.0027
0.0521
Million
Units
6.743
1.2029
1.0968
MX$/US$
6.843
10.2334
3.1989
Zero / one
...
...
Description
Quantity o f imports represents the volume
imported by Mexico from US.
MXY
Ps
Qs*
Psd
INV
ER
Di
Ps is the price of a substitute, in the case, it is

the price o f domestic sorghum.
INV is the inventory o f hogs on feed in
Mexico
exchange rate.
Dummy variables to represent trade
liberalization
Units
4. Milk Powder Equation

The specification of the milk powder import demand function for Mexico includes
domestic per-capita income, foreign price, a price for a related good represented by the
price of milk powder imported from Canada into Mexico, exchange rate, exchange rate
volatility, a free trade dummy variable and an additional exogenous variable (MSF). For
106
the case of milk powder from the United States into Mexico the following equation was
estimated:
Qm* = p 0 +P1 M Y +p2Pm +p3Pca +p4M S F +psDi + p 6E R + p 7Vi + et

dQ*
>0
dMXY
dQ *
= <
dPm
dQ *
- ^ >0
dPca
dQ *
- >0
dMSF
dQ *
dER
>
dQ *
^< 0
dVt
(3.79)
dQ *
= >
dDt
Here Qm* is quantity of imported milk powder. It represents the volume of imports into
Mexico from United States. It is measured in metric tons. MXY is importers income, the
original series was in constant pesos (1993=100) so, the series was divided by the
population to obtain the real Mexican GDP per person. Pm is the border price (in
US$/Kg) of the commodity, it is the quotient of the total value (in US dollars) of Mexican
imports from United States and the volume (in metric tons) of Mexican imports. All
border prices and United States values were converted to real terms by the US CPI. Pea is
the price of a substitute, it is represented by the price of milk powder imported from
Canada into Mexico. MSF is the difference between total production and consumption of
milk products in Mexico. D; is NAFTA or seasonal dummy variable to indicate the
effects of liberalization (the reduction of tariffs). ER is the Mexican peso per US dollar
exchange rate. V) are measures of exchange rate volatility. Four different volatility
measures were tested, one at a time, in each equation (Table 3.5).
income in the importing country rises so the quantity imported. The expected sign for Pm
is negative and for Pea is positive. A positive coefficient is expected for the foreign
exchange variable because appreciation of the US dollar leads to a fall in Mexican
107
imports as US products become relatively expensive to those in Mexico or other

exporting countries. Assuming risk-averse traders, measures of volatility are expected to
be negative. The dummy variable is expected to have a positive coefficient.
Table 3.5 Description and Specification of Variables for Milk Powder Equations
Variable
Qm*
Description
Quantity o f imports represents the
volume o f milk powder imported
by Mexico from US.
Units
M ean
Variance
St Dev.
Metric Ton
3842.6
1826667
4274
MXY
Importers income is the Mexico

per-capita GDP.
MX$/person
2977.211
1869674
1367.3
Pm
The border price is the quotient of

the total value o f Mexican imports
from US and the volume o f these
imports.
Ps is the price of a related good; it
is the price of milk powder
imported from Canada.
US$/kg
1.4227
0.2655
0.5153
US$/kg
1.5622
0.2771
0.5490
MSF is the difference between

total consumption and production
o f milk in Mexico.
ER is the Mexican peso per US
dollar exchange rate.
Hundreds
Liters
306.24
20158
141.98
MX$/US$
6.843888
10.2334
3.198
Pea
MSF
ER
Di
Dummy variables to represent

trade liberalization
3.9 Summary
Exchange rate variability has increased since the 1970s after the termination of the
Bretton-woods agreement on fixed exchange rates. The exchange rate variability has been
identified as a source of important negative effects on international trade flows. As a
consequence, it has been a concern for trading nations. This issue, together with changes
in exchange rate, is the main concern of this research. In this chapter, the effects of
108
exchange rate changes on commodity price were illustrated. First, graphically and then,
an algebraic treatment was presented. For the purpose of explanation the graphical and
algebraic treatment is appropriate, but since this approach does not allow for the
treatment of income, related prices and exchange rate volatility, then, a more general
framework was described.
Given the nature of the traded commodities dealt with in this research, the
theoretical framework developed treats import as intermediate good that enter into the
production process, together with domestic inputs, to produce final goods. Uncertainty of
exchange rate enters the model at both exporters revenue and exporters expenditure on
inputs. The comparative statics results show that the direction of the effects of exchange
rate movements on both import demand and export supply. The next step in this research
is to select the econometric techniques to be used for the estimation of import demand
functions and the appropriate treatment of data non-stationary and exchange rate
volatility, all of which are considered in chapter four.
109
CHAPTER 4
FORMULATION OF THE EMPIRICAL MODEL
4.1 Introduction
The purpose of this chapter is two fold. First, to discuss the selection of suitable
econometric methods for estimating the effects of changes in exchange rate and its
volatility on Mexico-US agri-food trade, second, to describe data used and to develop
four alternative exchange rate volatility measures to be used in estimation. The chapter
concludes with the definition and specification of the empirical model based on the
theoretical model described in chapter three.
The determination of the effects of the exchange rate on agri-food trade can be
performed using alternative approaches. The literature reports the use of two different
techniques: the structural approach and the reduced form technique. In the structural
approache a set of equations in each country representing the market for a homogeneous
commodity is formulated. This approach was very popular in the 1970s and 1980s
especially for agricultural products. Ordinary least squares and two stage least squares are
the most common econometric methods employed to estimate structural models.
The structural approach has the advantage that institutional, production and
marketing information about the specific commodity can be incorporated relatively easily
in to the structural model. Since the idea is to represent the underlying economic
conditions relevant for a commodity, this approach is highly demanding on specific data.
One of the principal limitations of using OLS to estimate a trade model is that stationarity
of the macroeconomic data is often ignored.
110
Regression analysis based on time-series data assumes that the underlying timeseries are stationary. They are stationary in the sense that mean, variance and covariance
of the individual series are time-invariant. The classical t-test and F-test are based on this
assumption. This assumption can be verified by determining if the time-series in question
contains a unit-root. A time-series that has a unit-root is known as a random walk, and a
random walk is an example of a non-stationary time-series. It is well documented in the
literature that many macroeconomic time series, like exchange rate, commodity prices,
incomes, etc., follow a random walk process (Banerjee et al., 1986; Meese and Singleton,
1983; and Nelson and Plosser, 1982). W hen non-stationary variables are used in
regression analysis the classical t and F-test are not appropriate and may lead to
misleading conclusions (Hendry, 1986).
Most of the empirical work on the effects of exchange rate and its volatility on
agricultural trade have been done without considering the time series properties of data.
This is particularly true for empirical studies on Mexican trade. No previous study on
Mexican trade performance has made any attempt to test for the existence of nonstationary data and use appropriate econometric analysis in view of data non-stationarity
Consequently, all previous studies may suffer from: a) a spurious regression problem
often characterized by a high R and low value of the Durbin-Watson statistic b)
inconsistent and less efficient ordinary least squares (OLS) parameters estimates unless
the variables are co-integrated (Engle and Granger, 1987); and c) not having a valid error
correction representation (Phillips, 1986).
The reduced form approach uses a time series model to study the economic
relationship between the exchange rate movements and the variables entering the model
111
such as prices,
quantities
and other relevant variables
expressing institutional
information. In the reduced form approach, the endogenous variables, are expressed
explicitly in terms of lagged values of itself and of other variables in the model. The main
advantage of the reduced form model is that meaningful economic hypotheses can be
tested without estimation of structural parameters and it can accommodate non-stationary
data. Also, reduced form model is less demanding in terms of data.
The structure of this chapter is as follows: Section 4.1 discusses the different
approaches proposed in the literature to tests for the presence of a unit root in the data.
Section 4.2 deals with the methods for determining cointegration and discusses
advantages and disadvantages of each technique to finally propose the appropriate one for
this particular study. The Error correction model (ECM) is discussed in section 4.3. Also
it presents an overview of the estimation technique. Four different volatility measures are
developed in section 4.4. The chapter ends with a discussion of the data and data sources
and a conclusion.
4.2 Unit Roots and Cointegration Analysis

In empirical work using economic time series it is important to determine whether the
relevant data are level or difference-stationary. To test the relevance of the difference
stationarity formally, various tests for a unit root have been developed in the econometric
literature, notably Dickey and Fuller (1979), and Phillips and Perron (1988). More
recently, Schmidt and Phillips (1992) proposed a different test for a unit root applying the
Lagrange multiplier (LM) principle to the component representation of time series.
They derived the LM test statistic for autoregressive time series of order one (AR (1))
112
with independent and identically distributed errors, and they modified the statistic nonparametrically along the lines of Phillips (1987) to deal with more general time series
(Oya and Toda, 1995).
The unit root tests proposed in the literature can be classified into parametric and
non-parametric groups. Simulation studies performed by several authors (e.g. Schwert,
1989a; DeJong et al., 1992) reported that the non-parametric unit root test of Phillips
(1987) and Phillips and Perron (1988) do not work well for typical sample sizes available
for economic analysis. The Dickey-Fuller (DF) and Augmented Dickey-Fuller (ADF)
unit root tests are often applied to test whether a time series has a unit root.
Since cointegration analysis is interesting only for non-stationary time series, the
first step in cointegration analysis is to verify that all variables in question are integrated
of the first order (i.e., there is a unit root in the data set).A variable y r is said to have a
unit root in its autoregressive process if it has the following autoregressive representation:
(1- L) yt = Y\ (1-L) Yt-i + .........+ YP(1-L) Yt-p + et
(4.1)
where: = a stationary stochastic process,

E k < i and Lky, = y,-k
A number of test statistics have been proposed in the literature to test for the
existence of unit roots. Many of these are variants of the t-like tests proposed by Dickey
and Fuller (1979). To test for the presence of a unit root in yt, an augmented DickeyFuller (ADF) test can be computed by running the following regression:
(1 - L ) y t = a 0 + a, yr_, +
1=1
(i - L)- + ,
113
(4 -2)
A negative and significant estimate of axis inconsistent with the null hypothesis
of a unit root in y t . The t-ratio on the estimated ax, however, it does not have a standards
t-distribution. The critical value provided by Dickey and Fuller (1979) needs to be used.
A number of econometric studies have found that standard test for a unit root,
such as the ADF (Dickey and Fuller, 1979, 1981) and the Phillips and Perron (1988)
tests, have low power against stationary alternatives in the relatively small samples. (See
Dejong et al., 1992). This is especially true when a series under investigation is a near
integrated process. Since the low power of the univariate unit root test is the primary
problem, it is important to investigate whether or not the null hypothesis of a unit root is
rejected by a more powerful test.
Oya and Toda (1995) developed a new procedure for testing the presence of a unit
root test by combining features of the ADF and the LM tests. Through Monte Carlo
simulations they examined the dependence of the finite-sample distributions of the DF
and LM test statistics on the estimated parameters and initial values of AR processes.
They argued that by combining two different test statistics one may obtain a unit root test
that is more robust to the initial value (and the value of coefficient parameters). Note,
however, there is no unique way of combining the DF and LM tests to produce a more
robust unit root test. It would be interesting to seek an optimal way of combining these
two unit root tests. The asymptotic local power for different methods of combination
could presumably be compared, for example, in a local-to-unity asymptotic framework,
formulating the initial value as an increasing function of the sample size.
To overcome the problem of low power and severe size distortions of
conventional methods, Ng and Perron (2001) propose new model selection procedures
114
and estimation strategy that produce good power and reliable size. They suggested new
modified information criteria (MIC) with a sample dependent penalty factor, along with
the GLS detrending of Elliott et al. (1996) to increase the power of their tests. This test
was applied by Yoon (2003) to Brazilian data. He concluded that the Ng and Perron
procedure for testing unit root yield reliable results. Brazilian inflation rate is not
stationary, while application of the ADF test yields stationarity for the same data set.
It is a well-established empirical fact that standard unit root tests fail to reject the
null hypothesis of a unit root for many economic time series. This was first argued
systematically in the influential article of Nelson and Plosser (1982) who applied DickeyFuller type tests (Dickey and Fuller (1976) and Dickey and Fuller (1979)) to 14 annual
U.S. time series and failed to reject the hypothesis of a unit root in all but one of the
series. These results do not changed by allowing for error autocorrelation using the
augmented tests of Said and Dickey (1984) or the test statistics of Phillips (1987) and
Phillips and Perron (1988). Similar results are obtained for many other macroeconomic
time series. The standard conclusion that is drawn from this empirical evidence is that
many or most aggregate economic time series contain a unit root. However, it is
important to note that in this empirical work the unit root is the null hypothesis to be
tested, and the way in which classical hypothesis testing is carried out ensures that the
null hypothesis is accepted unless there is strong evidence against it. Therefore, an
alternative explanation for the common failure to reject a unit root is simply that most
economic time series are not very informative about whether or not there is a unit root, or
equivalently, that standard unit root tests are not very powerful against relevant
alternatives. Several more recent studies have argued that this is indeed the case.
115
To overcome the above limitations of previous studies, Kwiatkowsky et al (1992)

proposed a test of the null hypothesis that an observable series is stationary around a
deterministic trend. The series is expressed as the sum of deterministic trend, random
walk, and stationary error and the test is the LM test of the hypothesis that the random
walk has zero variance. The asymptotic distribution of the statistic is derived under the
null and under the alternative that the series is difference-stationary. Finite sample size
and power are considered in a Monte Carlo experiment. The test was applied to the
Nelson-Plosser data, and for many of these series the hypothesis of trend stationarity was
not rejected.
In recent years, various Bayesian analyses have been conducted on unit roottesting. Using flat prior Bayesian technique, DeJong and Whiteman (1989) tested the
Nelson-Plosser series, stock prices and dividend data, and postwar quarterly real GNP for
the U.S.A. Their results challenged the classical unit root test results in many cases.
Schotman and Van Dijk (1991) analyzed the random walk hypothesis for real exchange
rates and found more evidence in favor of the trend stationary model than the classical
unit root test. Using a modified information matrix-based procedure, Zivot and Phillips
(1994) considered autoregressive models with fitted deterministic trends allowing for
certain types of structural change. Their results are generally in accord with those of
Phillips (1991a). In addition, their Bayesian analysis also show evidence of trend breaks
in some of the macroeconomic series with break occurring around 1929, partially
supporting the conclusion reached by Perron (1980). An advantage of Bayesian
techniques in nonstationary models with unit roots is that the asymptotic form of the
posterior density is normal under rather general conditions, a result that facilitates large
116
sample Bayesian inference and that contrasts with the non-standard asymptotic
distribution theory of classical estimators and tests.
As can be seen from the above discussion almost no method meets all the
desirable characteristics in terms of power and robustness. In most cases, it is ultimately
the reserchers preferences that determine what test should be applied to actual data.
Thus, the ADF test statistic is used in this research instead of Dickey and Fuller and
Phillips and Perron tests.
Empirical Implementation of DF and ADF Tests

While several unit root tests are available in the literature, this study employs ADF to test
for unit roots, because it produces more consistent results than other tests. Assuming yt is
a time series variable that is integrated of order one without drift, these tests can be
applied by altering the autoregressive process in the following way:
yt = 3V-i + t
(4-3)
where yt and yt./ are the present and the immediate past values of a variable, respectively;
and et is a stationary error term. Equation A .l can be expressed in the following form:
(4.4)
y , = a + 0 )y
where 6 is an arbitrary parameter. W hen 6 =0, then Equation A.2 equals Equation A .I.
After rearranging Equation A.2, the following equation is obtained:
Ay, = yt -y, -l =0y,_l +,
(4.5)
If 6 equals 0 and et is stationary, then yt ~ 1(1), and if - 2 < 6 <0, then yt is a stationary
process. Based on the above modification, Dickey and Fuller (1976) proposed a test of
117
Ho: 9 = 0 against Ho: 6 < 0. If the null hypothesis is accepted then the process is 1(1),
i.e. yt ~ 1(1). Dickey and Fuller considered the following three different equations to test
for the presence of unit roots:
A y ,= 0 y ,-i+ ,
(4.6)
Ay, = a 0 + 6yt_x + e ,
(4.7)
Ay, = a0 +9y,_l + a 2t + e t
(4.8)
The differences among the above regression equations depend on the presence of
ao, constant (drift) and ci2 t, deterministic term (time trend), all of which are called
nuisance parameters. Test results can be based on OLS estimations. The above equations
represent the first order autoregressive process (a process depending only on one lag
value). The test can be extended for higher order autoregressive processes. In conducting
the DF test based on equations 4.6 to 4.8 it was assumed that the error term was
uncorrelated, but in case the errors are correlated, Dickey and Fuller developed a test
known as the Augmented Dickey and Fuller test. It is conducted by augmenting to
equations 4.6 to 4.8 with lagged values of the dependent variable to get the following
equations:
P
(4.9)
j=i
P
(4.10)
Ay, = a 0 + a,y,_! + / M - j + ,
(4.11)
118
where ao and t are the constant and the time trend, respectively. There are t-critical
values related to Equations 4.9, 4.10 and 4.11, respectively. Enders (1995) suggests that a
procedure on Equations 4.9, 4.10 and 4.11 should always be followed to test for unit
roots when it is not known whether that process includes a trend term or drift. Unit root
tests utilize the following three main null hypothesis tests based on Equations 4.9, 4.10
and 4.11:
Ho: a\ = 0 (testing unit roots in Equation 4.9
Ho: a\ = <32 = 0 (testing unit roots with the time trend in Equation 4.10
Ho: a\ = ao = 0 (testing unit roots with the constant term in Equation 4.11.
Analyses are conducted based on the results of DF, ADF tests. A summary of the
procedure of the steps involved in each test is as follows: Check for unit roots in each
variable with the time trend and the constant terms in Equation 4.10. If a null hypothesis
of Ho'. a\ = 0 is not rejected (at the Dickey-Fuller critical value), there are unit roots. If the
null is rejected, then check for the presence of the time trend, ci2 , in Equation 4.10. If the
time trend is significant and if the presence of unit roots is not rejected according to the
conventional t-value, it can be concluded that the process of the variable has unit roots
with the time trend. If both are rejected, then it can be concluded that the process is
stationary.
If there is no time trend, i.e. null is rejected in Equation 4.10, then check for the
unit roots and the constant term in Equation 4.9. First, check the process for unit roots at
the Dickey-Fuller critical value. If there are no unit roots, it can be concluded that there
are no unit roots in the process, which means that the variable is stationary. If a constant
term is significant, then check the results for unit roots. If Ho is not rejected according to
119
the t-value, it can be concluded that the process of the variable has unit roots with the
constant. If Ho is rejected, it can be concluded that the process has no unit roots.
If there is no constant, check the process with neither constant nor time trend in
Equation 4.8. If there are no unit roots, the process is stationary, otherwise it would have
unit roots. If there are unit roots in any of these hypothesis tests, check the variable in
first difference form to check for two unit roots. If there are no unit roots, then it can be
concluded that the variable can be represented by an 1(1) process. Otherwise, the above
steps are repeated for the process in first difference values to test for the presence o f 1(2)
process.
Selection of the Lag Length

An important issue concerning the appropriate lag length arises. First, If a few lags are
included, that means the regression residuals would not behave like white-noise process.
Therefore, the model will not appropriately capture the actual error process so that a{ and
its standard error will not be well-estimated. Second, including too many lags reduces the
power of the test to reject the null of a unit root since the increased number of lags
necessitates the estimation of additional parameters and a loss of degrees of freedom
(Harris, 1995). The degrees of freedom decrease since the number of parameters
estimated has increased and the number of usable observation has decreased. As such, the
presence of unnecessary lags will reduce the power of the Dickey-Fuller test to detect a
unit root. In fact, an augmented Dickey-Fuller test may indicate a unit root for some lag
lengths but not for others (Enders, 2003).
120
It is possible to determine the lag length truncation for each series using an
information criterion such as the Akaike Information Criterion (AIC) or the Schwarz
Bayesian Criterion (SBC). These methods are proposed for this research. O f course, in a
very large sample with normally distributed errors, the methods should all select the same
lag length. In practice, the Schwarz (1978) criterion (SBC) will select a more
parsimonious model than will either the Akaike (1973) Information Criterion (AIC).
4.3 Empirical Specification and Cointegration Analysis

Several methods to estimate cointegrating vectors (long-run equilibrium relationships)
have been proposed in the literature since Granger (1983) introduced the idea of
cointegration. Chronologically they are: ordinary least squares (OLS) by Engle and
Granger (1987), nonlinear least squares (NLS) by Stock (1987), principal components
(PC) by Stock and Watson (1988), maximum likelihood in a fully specified error
correction model (MLECM) by Johansen (1988), instrumental variables (IV) by Hansen
and Phillips (1990), and spectral regression (SR) by Phillips (1991a).
The Engle-Granger Procedure

Engle and Granger (1987) propose testing for cointegration by testing whether the
residuals of a static regression are stationary. The unit root test used is that of Dickey and
Fuller (1981), which is based on a finite-order autoregression. Engle and Grangers
procedure imposes a common factor restriction on the dynamics o f the relationship
between the variables involved. If that restriction is invalid, a loss of power relative to the
ECM and Johansen procedures may well result. Ericcson (2001) compared EG
121
procedure, EG as a restricted ECM and Johansen approach.
The results support that
Johansen approach has better properties than the other alternatives (see Table 4.1).
Table 4.1 Comparison of Johansen, ECM and Engle-Granger Procedures for

Testing Cointegration___________________________________________________
Aspect
Statistic
Johansen
Maximal eigenvalue and trace
statistic
ECM (both types)

-Kd (K) Harbo et al (1998)
statistic
Engle-Granger
Assumptions
Well specifies full system
-Weak exogeneity o f Z for

(3
-Common
restrictions
Advantages
-Maximum likelihood o f full

system
-Determines r (the number of
Starting point for error

correction modeling
Unrestrictive dynamics
Weak exogeinity is often
valid empirically
Robust to particulars of
the marginals process
Super consistent
estimation o f P
cointegrating vectors),
Disadvantages
|3 and a
Full system should be well

specified
Sources
for Johansen (1988, 1995)
critical
values Johansen and Juseluis (1990)
and p-values
Osterwald-Lemus (1992)
_________
McKinnon et al (1999)
Source: Ericcson (1993)
Weak
exogeneity
is
assumed.
Usually r < 1 is imposed
Banerjee et al (1993)
Banerjee et al (1998)
Harbo etal (1998)
Pesaran et al (2000)
factor
Inferences on P are messy

Biases in estimating
Usually r < 1 is imposed
Normalization affects
estimation.
Engle and Granger (1987)
McKinnon (1991, 1994,
1996)___________________
Gonzalo (1994) compared several methods of estimating cointegrating vectors such

as ordinary least squares, Stock and Watson procedure, maximum likelihood in error
correction model and canonical correlation. He found that although all of them are
superconsistent, the estimates obtained from different methods vary considerably.
Examining the asymptotic distribution of the estimators resulting from these methods he
argued that the Maximum Likelihood approach in a fully specified Error Correction
Model (Johansens approach) has better properties than other estimators. A Monte Carlo
122
study indicated that finite sample properties are consistent with the asymptotic results.
This is so even when the errors are non-Gaussian or when the dynamics are unknown
Before performing cointegration analysis, Gonzalo (1994) suggests that it is
important to discuss three elements present in any cointegrated system. First, the
existence of unit roots, second the multivariate aspect and third the dynamics. Not taking
these elements into account may create problems in estimation. In general the coefficient
estimates will be biased in mean and median as well as inefficient. The distribution will
be nonsymmetric and nonstandard and there will be nuisance parameter dependencies.
The appropriate cointegrating method should have the following characteristics:
Incorporate all prior knowledge about the presence of unit roots. This eliminates
median bias, non-symmetry, part of the nuisance parameter dependencies, and
increases efficiency.
Full system estimation. This eliminates the simultaneous equation bias and
increases efficiency.
Flexible enough to capture the dynamics of the system.

Of the five methods, (OLS, NLS, ML in an ECM, PC, and CC), Gonzalo (1994)
compared only maximum likelihood in an error correction model that satisfies these
requirements. This approach ensures that coefficient estimates are symmetrically
distributed and median unbiased and that hypothesis tests may be conducted using
standard asymptotic chi-squared tests. None of the other methods considered has these
properties. Although the above properties are based on asymptotic theory, Gonzalo
(1994) showed, via a Monte Carlo study, that this conclusion is still valid for finite
samples. Based on results of Gonzalo (1994), Phillips (1991) and Hubrich (2001) among
123
others, it is suggested that research on the estimation of a cointegrated system should

proceed with the maximum likelihood method in a fully specified error correction model.
The conclusions obtained agree with the theoretical results in Phillips (1991) who
showed that the best way to proceed in the estimation of cointegrated systems is full
system estimation by maximum likelihood, incorporating all prior knowledge about the
presence of unit roots. This approach ensures that coefficient estimates are symmetrically
distributed, median unbiased, asymptotically efficient and that hypothesis tests may be
conducted using standard asymptotic chi-squared tests.
The simplest procedure is to estimate a fully specified error correction model
(ECM) by maximum likelihood (ML). This is exactly the method proposed by Johansen
and the one which performs better in a Monte Carlo experiment, even when the errors are
non-normally distributed or when the dynamics are unknown.
In cointegration analysis, one branch of the associated literature has been
concerned with finding the number of cointegrating relations in a system of time series
variables. The early tests for the number of cointegration relations are due to Johansen
(1988) and Stock and Watson (1988). These tests are derived under special assumptions
which are not always realistic in applied work. These tests are based on asymptotic
considerations and it was found in simulation studies that their small sample properties
may not be very satisfactory in some situations of interest from the point of view of
applied work.
The performance of the different cointegration techniques are reviewed again by
Hubrich et al (2001). The various sets of assumptions for the asymptotic validity of the
tests are compared within a general unifying framework. The comparison includes
124
likelihood ratio tests, Lagrange multiplier and Wald type test, lag augmentation tests,
tests based on canonical correlations, the Stock-Watson tests and Bierens nonparametric
tests. Asymptotic results regarding the power of these tests and previous small sample
simulation studies were discussed.
Hubrich (2001) compared a wide range of systems cointegration tests with respect
to their asymptotic properties as well as their small sample performance. It is well known
that in applied work, it is important to know the assumptions under which a test
procedure is supposed to work in order to choose the appropriate test. An important
aspect in this context is the assumption about the trend terms of the process and whether
the test statistics depend on these terms. Therefore, he suggested to focusing on analyzing
the sensitivity of the test performance to the trending properties of the data generating
process. He found that a trade-off between the generality of the assumptions for the
validity of a test on the one hand and its local power on the other hand.
The aim of the simulation study performed by Hubrich (2001) was to explore the
strengths and weaknesses of the different systems of cointegration tests to compare on the
basis of some simple DGPs. The idea is that tests which dont work well for a simple
DGP cannot be trusted in more complicated situations in general even if they have been
shown to perform well in specific complicated situations. Using this criterion, it was
found that the lag augmentation tests, the canonical correlation tests, the Stock and
Watson tests as well as the Bierens test perform very poorly even under ideal conditions.
Therefore their use cannot be recommended for applied work. In General likelihood
based methods are found to perform better than other methods.
125
Testing for Cointegration

Given the presence of a unit root in each series, a precondition for the existence of a
stable steady-state relationship is cointegration between the variables. A vector of
variables is said to be cointegrated if each variable in the vector has a unit in its
univariate representation, but some linear combination of these variables is stationary
(Engle and Granger 1987). From the previous discussion, at least five alternatives
approaches for testing cointegration have been advanced in the literature. These are the
two-step procedure developed by Engle and Granger (1987), the dynamic ordinary leas
squares (OLS) procedure developed by Stock and Watson (1988), the system approach
developed by Johansen (1988, 1991), the non-linear least squares of Watson (1987) and
the canonical correlation of Bossaerts (1988). The Engle-Granger procedure is a single
equation, regression residual-based test. Although it is a simple and attractive test for
bivariate models, it does not perform well in a multivariate situation.
The approach developed by Johansen derives maximum likelihood estimators of
the cointegrating vectors for a VAR system. It extends the Engle-Granger procedure to a
multivariate context where there may exist more than one cointegrating relationship
among a set of n variables. Moreover, it provides a likelihood ratio test for testing the
hypothesis that there are r cointegrating vectors. This approach also provides a very
flexible format for investigating the properties of the estimators under various
assumptions about the underlying data-generating process (DGP) and allows for testing
economic hypothesis. Because of these attractive features, Johansens approach will be
used in this research. What follows is a brief description of Johansen Maximum
Likelihood Cointegration Analysis.
126
Johansens Approach: An Exposition

Following Johansen (1988), 1991) and Johansen and Juselius (1990), this approach starts
with a klh order unrestricted VAR representation of Xt such that:
Xt = c + 7iiXt_i + 712X ^2 +
Where:
7ikXt-k + [X+ \)/ TD + Dt + St
(t = 1,..., T)
(4.12)
Xt = a vector of p 1(1) variables,

Dt = eleven seasonal dummies,
71 i = a (p x p) matrices of parameters,
c = a (p x 1) vector of constant terms,

TD = trade dummy variables,
s ~ N ID (0 ,D ).
In general economic time series are non-stationary processes, and VAR system like 4.12
have usually been expressed in first differenced form. Using V =1 - L, where L is the lag
operator, the model in equation 4.12 can be reparameterized as:
v x , = c + r j v x t-i + r 2 v x ,_2 + .... + r K.i v x t.k+1 - n x t.k+ ODt + st
(4.13)
where:
Tj = - 1 +7ii + .......... + 7ii, and
- n = I-7li_7t2- ....- 7Ik;

Vi = 1,2,......., k-1.
It is interesting to note that the reparameterized model is a traditional firstdifference VAR model except for the term II Xt.k. The coefficient matrix of Xt.k, II,
contains information about the long-run relationships among variables in the vector of
data. If II has a full rank, then X is a stationary process. In this case, a non-differenced
VAR model is appropriate. If II has a zero rank, then II is a null matrix and Xt is an
integrated process; only in this case, a traditional first-difference VAR model is
appropriate (Orden and Fisher 1991). If, however, 0 < (rank (II) = r) < p, cointegration
127
holds and n can be represented as the product of two p x r matrices a and (3, such that II
= a(3'. The /?'s are the cointegrating vectors and a s are the weights. In this case, f i fX t
is stationary. The long-run equilibrium is unique only when r =1.
The maximum likelihood estimation of II consists of two sets of regressions; one
set generates the residuals Rot from the regression of VXt on VXt_i, ...., VXt.k+i, and the
other set generates Rkt from the regression of Xt.r on VXt-i,
VXt_k+i (Johansen, 1995)
The concentrated likelihood function in terms of the product moment matrices of the
residuals can be expressed as:
l <0=icufi) r 2=iSoo - s0j{p'skkpr P'sk0r 2qa<01-T/2
(4 . 14 )
where Soo, Sok, Sko and Skk are the product moment matrices of the residuals defined as:
Sij = T ' Z * . * ;
Vij = 0,k
t=1
It is clear from equation 4.14 that maximizing the concentrated likelihood function is
equivalent to minimizing |QA(/?)|. This minimization amounts to solving the following
eigenvalue problem:
l-V -C -X O -C '-'h O
(4.15)
where C is a (p x p) matrix such that Skk = C'C. The vector of eigenvalues is given by X
while the corresponding eigenvectors can be derived as Vi = C'~l ei, where e is are the
eigenvectors from equation 4.15. The estimates of a and Q. can be obtained by using the
estimated value of /?. The null hypothesis that there are r cointegrating vectors is tested
using two likelihood ratio tests called the trace test and the maximum eigenvalue test. If
Hi: is a special case of H 2: for r = p, then the trace statistic is defined as:
128
(4 .1 6 )
2 In(Q; H 2/ H i) = - T ^ In{ 1 - M)
Similarly, the maximum eigenvalue statistic for testing H 2 (r) in H 2 (r+1) can be defined
as:
-2In (Q; r| r +1) = -T in (1 -Ar+1)
(4.17)
The asymptotic distributions of these likelihood ratio tests do not follow the
standard chisquared distribution. They actually represent multivariate versions of the
Dickey-Fuller distribution. The critical values for these tests are generated through
simulation and are reported in Johansen and Juselius (1990), Owterwald and Lenum
(1992) and MacKinnon (1999).
Johansens approach provides a convenient framework for testing linear
hypotheses expressed in terms of the coefficients p., a and (3. In particular, it is possible to
test for the presence or absence of linear trends in the stochastic part of the model. It is
important to choose the appropriate model formulation (i.e., with or without a trend
variable), because the asymptotic distributions of the test statistic and estimators depend
on which assumption is maintained (West 1988). The null hypothesis of the absence of a
linear trend in the model is equivalent to H q: |i. = e t/? ' Notice that when Hq: is true:
+M = uft-k + a f t = a f i* 'X *,_t i ,
where /3* = (/?,/?) (3* = and X*t-k = (Xt.k< 1).
The test statistic can be computed as:
-2 In(Q ;H *2 ( r < 1) | H2 (r < 1)) = -T
In [ (1 -i* ) / (1 - Xt) ]
which is distributed as x 2 with r degrees of freedom.
129
(4.18)
Likelihood ratio tests can also be formulated to test a variety of linear restrictions
on a and p. Theoretical and empirical economic knowledge can be used to formulate
these restrictions. These restrictions essentially limit the space spanned by the r
cointegrating vectors to lie in the s-dimensional space. If s = r, then the cointegration
space is said to be fully specified (Johansen 1991). Let / / 3: y5 = Hg represent a formulation
of a linear restriction on the cointegrating vectors, where H is a p x X matrix of
restrictions designed to restrict the space spanned by /3 to lie in s-dimensional space and 8
is a set of cointegrating vectors (see Johansen 1991 for details). The likelihood ratio test
can be computed as:
-2In (Q; H3 1H2) = -T Er In [(1-X,A3.i)/(l - 1* 0]
(4.19)
This statistic is also distributed as %2 with r (p-s) degrees of freedom.
4.4 Short-Run Dynamic Relationships and Error Correction Model (ECM)

The importance of the ECM in the cointegration literature derives from a demonstration
by Engle and Granger that if two variables are integrated of order one and are
conintegrated, they can be modeled as having been generated by an ECM. This work by
Granger and Engle has attempted to emphasize the use of the long-run equilibrium
relationships from economic theory in empirical models used by time-series analysts to
explain the short-run dynamics. Granger and Newbold (1986) concluded that the
resulting error-correction models should produce better short-run forecasts and will
certainly produce long-run forecasts that hold together in economically meaningful
ways. Furthermore, Engle and Yoo (1987) provide a theoretical analysis to support the
superior forecasting ability of the ECM model over unrestricted VAR models.
130
R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.
Tests for either or both short run or long run causality can be conducted using a
vector error correction model (VECM). Cointegration among the variables implies that
there is a long run error correction process, so, any deviation from long run equilibrium
will be restored by the correction of the equilibrium error back towards its long run
equilibrium.
Having obtained the long-run cointegration relations using the Johansen approach,
it is now possible to reformulate the above model and estimate the VECM with the errorcorrection terms explicitly included in it. Although Engle and Grangers (1987) two-step
error-correction model may also be used in multivariate context, the JohansensVECM
yields more efficient estimators of cointegrating vectors. This is because the VECM is
estimated with the full information maximum likelihood method, which allows for testing
for cointegration in a whole system of equation in one step and without requiring a
specific variable to be normalized. This allows one to avoid carrying over the errors from
the first step into the second, as would be the case if Eangle-Grangers methodology
used. It also has the advantage of not requiring a priori assumptions of endogenity or
exogenity of the variables. Consider a VECM that is of the form:
*-i
(4.20)
k-1
where , = ^ F y AY t_j and a/3'Yl k are the vector autoregressive (VAR) component in
j =i
first differences and error-correction components, respectively, in levels of the VECM..

Y t is a p x 1 vector of variables and is integrated of order one. fj, is a p x 1 vector of
constants, k is a lag structure, while e t is a p x 1 vector of white noise error term. Tj is a
p x p matrix that represents short-term adjustments among variables across p equations at
131
the jth lag. /?' is a p x r matrix of cointegrating vectors, and A denotes first differences.
a is a p x r matrix of speed of adjustment parameters representing the speed of error
correction mechanism.
In estimating the VECM, stationary and unit roots are first checked by performing
the selected test on the variables in levels and first differences. Only variables integrated
of the same order may be cointegrated and the unit root test will help to determine which
variables are 1(1). For this research the Error Correction Model will be used to analyze
the short-run dynamics among the variables given the desirable properties of short-run
forecasting based on this procedure.
The VECM is estimated using OLS since diagnostic test is performed such as
goodness of fit, the F-test, the DW test for first order autocorrelation, the BreuschGodfrey Lagrange multiplier test for serial autocorrelation, Ranseys RESET test for miss
specification and the Jarque-Bera test for normality.
4.5 Exchange Rate Volatility Measures

The aim of empirical research on exchange rate volatility and trade is to uncover any
systematic relationship between exchange rate risk and trade flows. The primary reason
for investigating such a relationship is that increased volatility may discourage
international trade. While it is generally recognized in the trade literature that exchange
rate volatility induces additional uncertainty in trade, no consensus exists on how to
measure it.
A number of statistical measures of variability have been used in the literature.
The measures can be broadly categorized into three groups: (i) Standard deviation and
132
variance-based volatility measures (the standard deviation, variance of the spot exchange
rate around its trend, the difference between previous forward and current spot rates, the
coefficient of variation, the standard deviation of the annual percent change of a bilateral
exchange rate around the mean), (ii) Autoregressive Conditional Heteroskedasticity
(ARCH) model or Generalized Autoregressive Conditional Heteroskedasticity (GARCH)
models, and (iii) non-parametric techniques. Each of these measures has its advantages
and shortcomings. Appendix 5 provides an overview of the most widely used models to
measure volatility in empirical studies.
Economic theory does not provide a clear guidance on features of the most
appropriate volatility measure (Clark, et al 2004). In an attempt to shed light on the
efficacy of different volatility measures on trade flows, this research attempts to test the
effects of four alternative exchange rate volatility measures. These volatility measures on
trade flows are as follows: (i) a volatility measure based on an ARCH or GARCH model;
(ii) a non-parametric-based exchange rate volatility measure, and (iii) two variance or
standard deviation-based volatility measures (moving average of the standard deviation
of the exchange rate and the average absolute difference between the previous forward
and the current spot rate; Hopefully, this comparison help to identify what constitutes a
good measure of exchange rate risk in the context of Mexico-US agri-food trade. In the
next section each of the volatility measures is described in terms of their advantages and
shortcomings, empirical justification and specification and what each measure intends to
capture.
Previous research on exchange rate volatility and trade has failed to recognize that
exchange rates are generated through a stochastic process and that exchange rate data
133
have exhibit certain set of characteristics that violates the assumptions of a standard
regression analysis. As indicated by by Bollerslev et al. (1992), financial time series are
typically heteroskedastic, leptokurtic and exhibit volatility clustering. He suggests that
these features could be handled more succesfully by modeling the volatility of the time
series as conditional on past behaviour. Since their introduction by Engle (1982) ARCH
models and their subsequent generalizations (see Bollerslev, 1986) have proved to be a
very useful technique to measure exchange rate volatility. The GARCH models have
been particularly used to describe the properties of high frequency exchange rate time
series. But also ARCH and GARCH models have been recently used in different studies
of exchange rate volatility using monthly data (McKenzie, 1999; Lastrapes and Koray,
1990; Mckenzie, 1998; Qian and Varangis, 1994).
4.5.1 Modeling Exchange Rate Volatility with ARCH Model

The ARCH model describes the forecast variance in terms of current observable. Instead
of using short or long sample standard deviations, the ARCH model takes weighted
averages of past squared forecast errors. These weights usually assign more influence to
recent information and less to the distant past.
In conventional econometricsmodels, the variance of the disturbance term is
assumed to be constant. However, a large body of literature shows that many economic
time series exhibit periods of large volatility followed by periods of unusual tranquility.
In such cases the assumption of constant variance is not appropriate. To forecast variance
one can choose different approaches. One approach is to explicitly introduce an
independent variable to help to predict the volatility. Consider the following case:
134
iVr+1 = ,+iX
,
(4-21)
where: yt+i = The variable of interest

8t+i = a white-noise disturbance term with variance a
x t = an independent variable that can be observed at period t

If xt = x r_t = x l2 = .... = constant, the {yt } sequence is the familiar white-noise
process with a constant variance. However when the realizations of the {yt } are not
equal, the variance of the yt+i conditional on the observable values of x,
is
Var(yl+1 / x t ) = xfcr2, here the conditional variance of yt+i is dependent on the realized
values of x,. If the magnitude of (jct)2 is large (small), the variance of yt+i will be large
(small) as well. Then, in this way, the introduction of the {xt } can explain periods of
volatility in the {yt } sequence. In fact, equation (4.18) can be generalized as:
Ln (y ,) = a0 + ax ln(xM) + et
(4.22)
where et is the error term, et = ln(r). But a major difficulty with this strategy is that it
assumes a specific cause for the changing variance. Also in this approach {v,}is forced to
affect the mean of the Ln(yt ). Instead of using ad hoc variables choices for x, and/or data
transformation Engle (1982) showed that it is possible to simultaneously model the mean
and the variance of a series.
To elaborate the Engle methodology further, consider the case that the variance of
{ t } is not constant; then, one can estimate any tendency for sustained movements in the
135
variance using an ARMA model. Let \et} be the estimated residual from the following
model; y, = a 0 + axy t__x + e t , so, the conditional variance of yt+i is:
Var(yt+l / y t) = Et [(y(+1 - a0 - axy t) 2] = E,(t+l) 2
(4.23)
where Et(et+1) 2 = o 2. Assuming that this conditional variance is not constant, one
strategy is to forecast the conditional variance as an AR(q) process using squares of the
estimated residuals.
(4.24)
i f - a 0 + axif_x + a 2if_2 + ..... + aqef_q + v,
where vt is a white-noise process. If the values ax, a2, ...., an all equal zero, the
estimated variance is simply the a0. Otherwise the conditional variance of yt evolves
according to the autoregressive process given by (4.24). Then, we can use model (424) to
forecast the conditional variance at t+1 as:
E
, f+ \
2 ^ f - l
q f + \- q
(4.25)
Model (4.24) is the autoregressive conditional heteroskedastic (ARCH) model.

For estimation purposes instead of the linear specification of (4.24), the common practice
in the literature is to use a multiplicative conditional heteroskedastic model. The reason is
that the model for {yt } and the conditional variance are best estimated simultaneously
using maximum likelihood techniques.
According to Engle (2004), the important contribution of the ARCH model is that
the weights could be estimated from historical data even though the true volatility was
never observed. It works as follows: forecasts can be calculated every period. By
examining these forecasts for different periods, a set of weights can be determined that
make the forecasts closest to the variance of the next return. This procedure, based on
136
Maximum Likelihood estimation, gives a systematic way to estimate the optimal weights.
Once the weights are determined, the dynamic model of time-varying volatility can be
used to measure the volatility at any time and to forecast it into the near and distant
future.
The ARCH Model and Exchange Rate Data

The characterization of exchange rate movements, including second-order dynamics,
have important implications for many issues in international trade and finance because
for domestic assets, international portfolio management depends on expected exchange
rate movements through time. Several policy-oriented questions relating to the impact of
the exchange rate on different macroeconomic variables require an understanding of the
exchange rate dynamics.
Traditional time series models have not been able to capture the stylized facts of
short-run exchange rate movements, such as their contiguous periods of volatility and
stability together with their leptokurtic unconditional distributions (see Friedman and
Vandersteel, 1982). The ARCH class of models is ideally suited to model such behavior.
There are many benefits to formulating an explicit dynamic model of volatility.
As mentioned above optimal parameters can be estimated using the Maximum
Likelihood method. Then, tests for the adequacy and accuracy of a volatility model can
be used to verify the procedure. One-step, two-steps, and multi-step forecasts can be
constructed using these parameters. The unconditional distribution can be derived
mathematically and are generally realistic. Inserting the relevant variable into the model
can enable researcher to test economic models that seek to determine the cause of
137
volatility. Incorporating additional endogenous variables and equations can similarly

enable researchers to test economic consequences of volatility.
Modeling Volatility with GARCH Model

GARCH is a time-series technique that allows one modeling the serial dependence of
volatility. It includes past variances in the explanation of future variances. A time-series
is said to show GARCH effects if it heteroskedastic (i.e., its variances vary with time).
Otherwise, the time series is to be homoskedastic (variances remain constant with time).
In many of the applications with the linear ARCH (q) model, a long lag length q
is called for. An alternative and more flexible lag structure is often provided by the
Generalized ARCH, or GARCH(p, q), model in Bollerslev (1986).
o f =co+Ya a if-i + X Pi
1=1
1=1
=co+ a (L )f +
(4.26)
To ensure a well-defined process all the parameters in the infinite-order AR

representation: o f = <p{L)sf =
a { L ) s f must be nonnegative, where it is
assumed that the roots of the polynomial /3(A) = 1 lie outside the unit circle (see Nelson
and Cao, 1991). For a GARCH (1,1) process this amounts to ensuring that both a
and ft
are non-negative. It follows also that e t is covariance stationary if and only if

a( 1) + J3(\) < 1. Of course in that situation the GARCH (p, q) corresponds exactly to an
infinite-order linear ARCH model with geometrically declining parameters.
GARCH model essentially generalizes the purely autoregressive ARCH model to
an autoregressive moving average model. The weights on past squared residuals are
138
assumed to decline geometrically at a rate to be estimated from the data. An intuitively

appealing interpretation of the GARCH (1,1) model is that the GARCH forecast variance
is a weighted average of three different variance forecasts. One is a constant variance that
corresponds to the long-run average. The second is the forecast that was made in the
previous period. The third is the new information that was not available when the
previous forecast was made. This could be viewed as a variance forecast based on one
period of information. The weights on these three forecasts determine how fast the
variance changes with new information and how fast it reverts to its long-run mean.
Another advantage o f GARCH modeling is that it takes into account excess kurtosis (i.e.,
fat
tail
behavior)
and
volatility
clustering,
two
important
characteristics
of
macroeconomic time series. It provides accurate forecasts of variances and covariances of

asset returns through its ability to model time-varying conditional variances. As a
consequence, this method seems suitable for modeling risk as in the case of foreign
exchange rate.
GARCH models can be used to examine foreign exchange markets which couple
highly persistent periods of volatility and tranquility with significant fat tail behavior.
These characteristics of exchange rate are particularly well suited for GARCH modeling
(Bollerslev et. al., 1992). Finally, GARCH model offers a more parsimonious model (i.e.,
using fewer parameters) that lessens the computational burden.
GARCH Limitations; GARCH models are parametric specifications that operate
best under relatively stable market conditions. Although GARCH is explicitly designed to
model time-varying conditional variances, GARCH models often fail to capture highly
irregular phenomena, including wild market fluctuations (e.g., crashes and subsequent
139
rebounds), and other highly unanticipated events that can lead to significant structural
change (Gourieroux, 1997). GARCH models often fail to fully capture the fat tails
observed in asset return series. Heteroscedasticity explains some of the fat tail behavior,
but typically not all of it. To compensate for this limitation, fat-tailed distributions such
as Student's t have been applied to GARCH modeling.
In this research the GARCH technique is applied to the real monthly exchange
rate time series for the period 1989:1 to 2004:12. Utilizing a GARCH model to capture
the conditional variance of real rates is particularly appropriate for the flexible exchange
rate period because this period has generated more volatile real exchange rates than the
fixed exchange rate regime. The next section describes the specification of the ARCH
and GARCH models and how they can be employed to generate exchange rate volatility
measures.
The estimation of GARCH models basically consist of three steps: the first
concerns to the selection of (p, q); the second deals with testing for ARCH effects; c) the
third one is the estimation of the model. To apply the GARCH technique, it is necessary
to find the best fitting time series model for the real exchange rate; an ARIMA model
could perform this task. Once this is done a GARCH (p, q) model to capture the time
varying conditional variance of the real exchange rate is be used, the proposed model will
have the following specification:
a) dLnER =ao+ iet-i+ et,
b) e, 1 1 - N (0, A,).
= a + z m 2- , + 1 ; m - ,
1=1
<4 -27>
1=1
140
where: dLnER is the difference in the log of the real exchange rate with respect to the
previous period.
I t-i is all relevant and available information at time t-I;
et-i is the moving average term;
et is the disturbance term;
ht is the conditional variance.
Equations in (4.27) need to be estimated jointly for the sample period. If
coefficients of e t-i and h t-i are statistically significant, significant ARCH and GARCH
effects are said to exist in the data. The prediction value of equation (c) in (4.27) provides
a measure of real exchange rate uncertainty.
4.5.2 Modelling Exchange Rate Volatility with Nonparametric Technique

To model the changing nature of risk in economic models, most of the previous work has
used a parametric functional form with the exception of Pagan and Ullah (1988), Gallant
and Tauchen (1989) and Pagan and Hong (1991) who used non-parametric approach to
specify risk terms. The non-parametric specification of the risk has the advantage of
being robust to mis-specification of the functional form (Baltagi and Li, 2001)
Pagan and Ullah (1988) and Pagan and Hong (1991) criticized the parametric
formulation of risk terms on the grounds that there is no optimizing theory that indicates
the functional form relating risk premiums to the information available to the economic
agent. Using data from two applications on equity premiums and excess holding yield for
treasury bills, Pagan and Hong (1991) demonstrated that the nonparametric approach
provides a superior method for modeling risk than parametric methods.
141
Pagan and Ullah (1988) suggested an instrumental variable method to estimate the
model with a non-parametrically specified risk term. Pagan and Hong (1991) argued that
one can also use least squares after the plug-in risk term is estimated non-parametrically.
However, Pagan and Hong (1991) did not develop the asymptotic distribution of their
proposed estimator. Baltagi and Li (2001) established the 4 n -normality result of the
Pagan and Hong (1991) estimator with generated regressors. Following Pagan and Ullah
(1988) and Baltagi and Li (2001) the departure of the proposed approach is that in
modeling macro-economic models agents are assumed to optimally choose a variable yt
on the basis of some information set Ft.
E( yt \xt Ft_l) = xta + a f S
(4.28)
where xt e F is a p x l vector variable and cr2 is the variance of some variable say wt,
conditional on
Ft., i.e. a f = E [(w, - E(wf | Ff_, ))2 1Ft_x] = E(wf \ Ft_x) - [E(wt \ Ft_x) f .
<T2 represents a risk term arising from the failure of agents to be able to correctly predict
some variable wt, which may be yt, (Pagan and Ullah, 1988). The size of crfS indicates
the impact of the risky environment upon yt. A test for S = 0 is a test for the significance
of the risk premium. Writing (4.28) as a regression model gives:
yt = x toc+o-*S+t =( xt,<jf)
+ ,
(4.29)
j
s x j + t , with E(t I xt Ft_{) = 0 ,
where
X t = (xcr,2)'
and
y = ( a , S ) . If <j2were observable, regressing yt on
( x , a f ) would given a Vn -consistent estimator of y under quite general conditions

(Pagan and Ullah, 1988).
142
In
practice
functions E(wf \ Ft) and
of ,
is
not
observable because
the conditional
mean
E(wt | F() have unknown functional forms. These unknown
functional forms can consistently be estimated using some nonparametric techniques. In

principle, Ft_i can contain all past information up to period t-1. However, under the
Markovian assumptions on wt and w f, there exists a finite dimensional as follows:
E(wf | Fm ) = E(wf | z,) and E(wt \ Ft_x) = E(w, \ zt)
(4.30)
Assuming that equation (4.30) holds o f = E{w2 | zt) - [ E ( w t \ zt) f = o-2(z,)

Following Pagan and Ullah (1988) and Pagan and Hong (1991), this research
considers the case that (y,w,z) are observable and propose to estimate the unknown
conditional mean functions E(wf \ Ft) and E(wt \ Ft) by the nonparametric kernel method.
Empirical Non-Parametric Estimation of Volatility

In this research Xu and A,2t are used to denote kernel estimator of E(wf \ Ft) and
E(wt | Ft) respectively. Hence,
Xu = (nhqy l^ w sK (^ L^ L)/ f t f,
Itt
h
(4.31)
hx = ( n W y ' Y w i K i ^ ^ ) / f ,
itt
h
(4.32)
where f t is the kernel estimator of f ( z , ) (the marginal density function of zt ) given by
/, = ( n h i y ^ K i ^ ^ )
Y!
h
Therefore, a] - E(wf \ z , ) - [ E ( w t \ zt) f = <t2(z;) is estimated by erf =
143
- i^)2
The above results are applied to monthly observations of bilateral exchange rate
between the U.S. dollar and the Mexican peso for the period of 1989:1 to 2004:12. The
data consist of monthly spot rates, and one-month forward rates. The sample consists of
192 monthly observations. Hai, Mark and Wu (1997) provide convincing evidence that if
the spot and forward rates, St+k and fk,t (k = 1,3) are cointegrated, it is possible to use this
information to obtain a more powerful test. The proposed approach was implemented by
Baltagi and Li (2001) as follows:
st+k- f kj=<Zo + f o t + ,
(4 -33)
Equation (4.33) is exactly the same as considered in Pagan and Ullah (1988)
where they used it to study the risk premium of $U.S.= Canadian exchange market.
Following Baltagi and Li (2001) this research chooses:
wr = ,+t - /* , and z, = (zu z2ty = (sI+k_hst+k_2 - f k t_2y .
Since z, e R 2 (q = 2), condition (A2) in page. 449 of Baltagi and Li (2001) require a
higher order kernel to be used. However, as they argued (in Remark 2.1 page 449) one
can show
that a second order kernel can be used for q=2. Therefore, it is possible to use
a product second order normal kernel function. In the normal kernel, the smoothing
parameters are chosen via ht = czt sdn~U5 where ZjiSd js is the sample standard deviation of
(st+k - f k , )"_* (i = 1,2) and different values of c can be used. The window width h is an
important parameter, and its choice determines the size or the interval around z over
which the observations are averaged. Usually the larger the h is, the less is the variance
and the smoother the curve, but the larger the bias. The parameter h can be calculated as:
h = n~ll(n+q). And use the (zn ,
Z jq )
data scaled with their standard deviations.
144
4.5.3 Variance-Standard Deviation Volatility Measures:

The concerns about the impact of exchange rate volatility on trade originate from the
belief that unexpected changes in exchange rates influence behavior by risk-averse
decision-makers (Brodsky, 1984). Risk-aversion is usually modeled by assuming that
decision-makers maximize objective functions in which unexpected events show-up as
squared deviations from the expected values (Kenen and Rodrik, 1986).
4.5.3.1 Moving sample standard deviation of the growth rate of the exchange rate
For the estimation of the sample standard deviation of the growth rate of the exchange
rate the following formula is used:
1/ 2
(4.34)
where m is the order of the moving average, and Z the (log) relative price of foreign
consumer goods in terms of US consumer goods. This measure have been used in the
literature by Cushman (1983; 1986); Gotur (1985); Kenen and Rodrik (1986); Bailey
Tavlas and Ulan (1987); Koray and Lastrapes (1989); Bini-Smaghi (1991); Chowdhury
(1993). This proxy of exchange rate volatility intends to capture temporal variations in
the absolute magnitude of changes in exchange rate.
4.5.3.2 Absolute Percentage Change of the Exchange Rate

For the estimation of the absolute percentage change of the exchange rate the following
formula is used:
(4.35)
145
where: e is the spot exchange rate and t refers to time.

This measure have been used in the literature by Thursby and Thursby (1985); Bailey,
Tavlas and Ulan (1986). This measure attempts to capture just variances of the future
spot exchange rate. Hooper and Kolhagen (1978) recognized that econometric estimates
based on this volatility measure may understate the actual amount of risk in floating rate
periods since this measure ignore other closely related sources of exchange market
uncertainty such as expectations about exchange rate.
In general, proxies of exchange rate risk such as standard deviation of the level of
exchange rate and standard deviation of the rate of change of exchange rate can be
criticized on the grounds that they lack a parametric model for the time varying change of
exchange rate On this issue Peree and Steinherr (1989) argue that the use of the variance
of ER in the immediate past as a proxy for ER is of limited relevance because it does not
capture ER uncertainty on the basis of historical experience. Moreover, the use of such
measures to proxy ER volatility may not be appropriate because measures of
changeableness fail to fully capture the uncertainty element embodied in changes in
ER.
4.6. Data Description and Sources of Data

The present research will focus on the period 1989-2004. The commodities under study
represent more than 50 percent of the total imported and exported volume and value.
Those products include tomato, yellow com, sorghum, and milk powder. These
commodities were selected because they are important traded items and because they
present very different supply and demand conditions as well as institutional
146
arrangements. So, it is expected that exchange rate movements will affect them in
different manner. Furthermore, more than 90% of the traded value (exports/imports)
takes place with the U.S., so it allows for work with just US-Mexico exchange rate data.
Commodities were identified based on their codes under the Harmonized Commodity
Description and Coding System (Harmonized System, or HS).
Monthly data from January 1989 to December 2003 have been used to estimate
the models. This period was selected based on the availability of data. Also most of the
Mexican trade liberalization took place during this period. Since the focus of this study is
to determine the effects of exchange rate variations on agri-food trade flows, the main
component of the data consists of commodity data and the macroeconomic variables
related to them. Table 4.2 describes the main characteristics of the data and respective
sources of the data used.
All US domestic prices, including the wage rate and border prices were converted
to real values using the general US consumer price index which includes all main cities
and all items. Similarly, all domestic prices in Mexico were transformed to real values
using the general consumer price index in Mexico.
147
Table 4.2 Raw Data, Description and Sources

Raw data
Description
Sources
Exchange Rate
Nominal bilateral monthly M exico-US

exchange rate, end o f period exchange rate.
1989; 1-2004; 12.
USDA-ERS.
http://www.ers.usda.gov/Data/Exc
hangeRate.
US CPI
Monthly general US CPI. All items. 1989;12004;12. Base Period: 1982-84=100.
US department of labor
http://www.bls.gov/cpi/cpifaq.htm
#Question%2015.
US Income
US Personal Disposable Income 1989; 12004; 12. ($US/person).
U.S. Department o f Commerce

http://www.bea.doc.gov/bea/an/ni
paguid.
US W age
Rate
US Farm Labor, Monthly, Dollars/hour.

1989;1-2004;12
US National Agricultural
Statistics Service (NASS).
http://www.usda.gov/nass/pubs/re
portname.htm#Farm Labor.
M exico Wage
Rate
Nominal wage rate (national wage for

agricultural workers). 1989;1-2004;12
Mexican Ministry o f Labor.

http://www.conasami.gob.mx/
M exico CPI
General CPI. Monthly, 1989-2004. (Based Jun
Bank of Mexico, Economic

indicators
http://www.banxico.org.mx.
Bank of Mexico, Economic
indicators
http://www.banxico.org.mx
2002 = 100 ).
Mexico GDP
Quarterly GDP, Millions o f Pesos. 1993=100
Corn Volume
US Monthly exports (MT), HS 4-DIG.

1989-2004. (Metric tons)
USDA-ERS. FAS, U.S. Trade

Internet System
Corn
Price

value of Mexican imports from US and the
volume of these imports. ($US/Ton)

Internet System
Border
The wholesale nominal price was deflated

using the CPI in Mexico.
Claridades Agropecuarias.
http://www.infoaserca.gob.mx/cla
ridades/
Sorghum
Volume
US Monthly exports (MT), HS 4-DIG. 19892004

Internet System
Sorghum
Border
Price

value of Mexican imports from US and the
volume of these imports. ($US/Ton). 19892004
Domestic Wholesale Price, at the Mexico city
Wholesale marker.

Internet System
M aize
Domestic Price
Domestic
Sorghum
Price
Claridades Agropecuarias.
http://www.infoaserca.gob.mx/cla
ridades/.
148
Table 4.2 Raw Data, Description and Sources (continue)

R a w data
D escrip tio n
S o u rces
Tomato
Volume

1989-2004.
USDA-ERS. FAS,
U.S. Trade Internet Svstem
Tomato
Price

value o f Mexican exports to US and the
volume o f these exports ($US/Ton). 89-2004.
USDA-ERS. FAS,
M ilk Powder
Canada
US Monthly wholesale price.

1989-2004. ($US/Ton)
USDA-ERS.
M ilk Powder
Volume

1989-2004
USDA-ERS. FAS,
M ilk Powder
Border Price

value of Mexican imports from US and
the volume o f these imports. ($US/Ton).
1989-2004.
USDA-ERS. FAS,
4.7. Summary
This chapter described the selection of the econometric techniques for the estimation of
the model outlined in chapter three. The selection was guided by an interest of using the
appropriate econometric methods to incorporate data non-stationary and the appropriate
treatment of exchange rate volatility. To this end, Johansen Maximum Likelihood
cointegration analysis has been selected. This econometric technique is a broadly
accepted in the literature as it accommodates non-stationary data in econometric analysis.
This approach also provides ample opportunity to test meaningful economic hypothesis
without estimating structural parameters.
The appropriate specification and estimation of exchange uncertainty has been a
concern in the literature. This concern is addressed in this research using four alternative
measures including a GARCH and a non-parametric technique to estimate exchange rate
volatility.
149
CHAPTER 5
RESULTS OF UNIT ROOT TEST AND VOLATILITY MEASURES
5.1 Introduction
The purpose of this chapter is to present and discuss the results of unit root tests as well
as the estimates of different exchange rate volatility measures. The chapter is divided in
to three sections: the first section deals with the results of unit root test on individual
series and macroeconomic variables for selected commodities (fresh tomato, yellow com,
sorghum and milk powder). The analysis was carried out using SHAZAM econometric
package (version 10 professional edition). The second section presents and discusses the
estimates volatility measures. The final section concludes the chapter.
5.2 Unit Root Test for Individual Time Series

Testing stationarity and co-integration involves determining the order of integration for
each series that enters the equations used to test for the presence of a unit root. Based on
the discussion in chapter four, the Augmented Dicker-Fuller (ADF) test is used to
determine the integration properties of individual series used in each commodity model.
The ADF test on the variables in level form is used to test for the presence of unit root in
the data, while the ADF statistic for the series in first difference form is used to test and
confirm if the series are integrated of order one. ADF test is implemented as follows:
Consider the following pth order autoregressive processes,
150
Ayt =aJt-x+Y yAyt-i+<
c5-1)
Ay, = a0 +a x
Y Y j ^ t - j + t
(5-2)
7=1
P
Ay, = a 0 + a{y M + a2t + Y YjAy t-j + <
(5-3)
i =i
where
ao
and
are the constant and the time trend, respectively, and
yj
are the
parameters. With an ADF test, the null hypothesis of Ho: a, = 0 is tested against the
alternative hypothesis of Ha:
a,
< 0. A negative and significant estimate of ax is
inconsistent with the null hypothesis of a unit root in the data. If the null hypothesis is not
rejected then this is taken as evidence in favour of the presence of a unit root. Standard t
statistics are not valid for testing the null hypothesis. The appropriate critical values were
derived by Dickey and Fuller (1979). There are different critical values associated with
Equations 5.1, 5.2 and 5.3 respectively. The appropriate statistic to use depends on the
deterministic components included in the Dickey-Fuller regression equation. Without an
intercept or trend, the use of T statistic is suggested; with only the intercept, use the
statistic and with both intercept and trend, use the t T statistic.
The unit root test uses the following three null hypotheses depending upon which
of 5.1, 5.2 or 5.3 is used. These are Ho: a\ = 0 equation 5.1, Ho: a\ = <32 = 0 equation 5.2,
Ho: a\ = ao = 0 equation 5.3. The critical value of the ADF test for model 5.3 is -3.44 at
the 5 percent level of significance. Values of ADF test smaller than the critical value
indicates that the series has a unit root while ADF values higher than the critical -3.44
value confirm that the series is integrated of order zero.
151
5.2.1 Selection of the Lag Length

To determine the appropriate lag length, one can use either the Akaike (1973)
Information Criterion (AIC) or the Schwarz (1978) Criterion (SBC). In a very large
sample with normally distributed errors, both methods would select the same lag length.
In a small sample, however, the SBC will select a more parsimonious model than the
AIC. In this study, the SBC test is used to select the lag length for each variable. The
main results for unit root tests for all variables are reported in tables 5.1 to 5.4. The SBC
was used because it provides more consistent results as it works using a more restrictive
selection criterion.
5.2.1.1 Results of Unit Root Test for Macroeconomic Variables

Unit root test results for the macroeconomic variables are reported in table 5.1. Looking
at the ADF statistic, the null hypothesis of a unit root (i.e. non-stationarity) in the
univatiate representation cannot be rejected for any of the individual series in their level
form at the five percent significance level. So, the results are consistent with the
hypothesis that unit root non-stationarity characterizes all the macroeconomic variables
used in this study. The one exception could be the percentage change of the exchange
rate. Additionally, for the exchange rate variable, the Perron test1 to detect the presence
of structural break was performed. The results indicated that no structural break took
place during the sample period of this study.
1 This test is implemented by adding a dummy variable that takes a value o f (0,1) to allow for shifts in the
intercepts assuming a single crash . Critical values are found in Perron, 1990.
152
Table 5.1 Unit Root Test Results for US and Mexican Macroeconomic Variables
Variable in Level Form
Estimated
Coefficient
Lag Length
(Months)
ADF
Statistic*
Real US-Mexico ER
-0.014912
- 1.092
Trade W eighted Exchange Rate**
-0.018528
-0.9442
US Personal Disposable Income
- 0.026235
- 1.076
US Rural W age Rate
-0.192870
12
- 2. 772
M X Real Per-capita GDP
-0.326130
12
-2 .7 1 0
M X Rural W age Rate
-0.184490
- 1.315
Volatility o f the Exchange Rate (VI)
-0.175660
- 3.068
Volatility o f the Exchange Rate (V2)
- 0.966270
- 7.283
Volatility of the Exchange Rate (VG)
- 0.502867
- 2.028
Volatility of the Exchange Rate (VN)
- 0.672479
- 3.087
- 0.97241
- 7. 162
Trade W eighted Exchange Rate
-1.0096
-8.253
US Personal Disposable Income
- 1.59250
- 9. 156
US Rural W age Rate
- 4.48070
12
- 8. 573
M X Real Per-capita GDP
-4.53180
12
- 5.860
M X Rural W age Rate
- 1.01620
-2 9 .1 4
Volatility o f the Exchange Rate (VI)
- 1.70210
- 8.462
Volatility o f the Exchange Rate (V2)
-5.87510
- 8.935
Volatility of the Exchange Rate (VG)
- 1.23845
- 8.089
Volatility of the Exchange Rate (VN)
- 1.37812
-7 .2 1 0
Variable in First Difference

Real US-M exico ER (Dls/Peso)
* Five percent significance level

** Perron Test, Five percent significance level
The unit root test in first difference form is performed to confirm whether or not
the series is integrated of order one. Since the absolute value of all the Augmented
Dickey and Fuller (ADF) statistics are larger than the critical value at five percent level
153
R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited w ithout perm ission.
all the macroeconomic variables included in the analysis are integrated of order one, i.e.,
they are 1(1).
The above is true also for the volatility of the exchange rate variables, moving
average standard deviation of the exchange rate (VI), percentage change of the exchange
rate (V2), GARCH measure of the volatility of the exchange rate (VG) and the non
parametric measure of the volatility of the exchange rate.
5.2.1.2 Results of Unit Root Test for Tomato Models

The unit root test for the variables considered in the equations for tomatoes is shown in
table 5.2. For all these variables, the null hypothesis of a unit root cannot be rejected
when the data are in their level form at the five percent significance level. However, the
null hypothesis of a unit root is rejected for all variables when the data are used in firstdifferenced form. Therefore, it can be concluded that all the variables in the model for
tomatoes are integrated of order one.
5.2.1.3 Results of Unit Root Test for Corn and Sorghum Models
The optimal lag-length and ADF test results for the variables included in the trade
equations for com and sorghum are presented in Table 5.3. The results indicate that the
null hypothesis of a unit root cannot be rejected when all the variables are in level form.
However, they become stationary after first-differencing. Hence, they are integrated of
order one, 1(1).
154
Table 5.2 Unit Root Test Results for the Tomato Models
Estimated
Coefficient
Lag Length
(Months)
ADF*
Statistic
Tomato Export Volume
- 0.27698
12
-2.188
Tomato Border Price
- 0.96221
- 3.068
M X Domestic Tomato Price
- 0.28591
-2.125
Canada Tomato Price
- 0.55501
-2.881
Tomato Export Volume
-5.1510
12
-7.313
Tomato Border Price
- 3.0371
- 9.027
M X Domestic Tomato Price
- 3.0371
- 9.027
Canada Tomato Price
- 3.0234
-9.321
Variable
Level Form
First Differenced Form
5.2.1.4 Results of Unit Root Test for Milk Powder Models

The results of the unit root analysis for the milk powder model indicate that the null
hypothesis of a unit root cannot be rejected for all the variables but one in their level
form. The import unit value of milk powder appears to be stationary in the level form. All
other variables become stationary after first-differencing (Table 5.4).
155
Table 5.3 Unit Root Test Results for the Corn and Sorghum-Models
Variable in Level Form
Estimated
Coefficient
Lag Length
(Months)
ADF Statistic*
M X Domestic Corn Price
-0.1325
-3.164
Corn Im port Volume
- 0.3334
- 3.160
Corn Border Price
-0.3185
12
-2 .6 8 1
M X Domestic Sorghum Price
-0.3058
-3.420
Sorghum Import Volume
-0.1762
12
- 1.958
Sorghum Border Price
-0.1343
-3.139
M X Domestic Corn Price

Corn Im port Volume
- 1.2311
-3.1122
8
5
- 4.941
- 11.53
Corn Border Price
- 4.5644
12
-5.912
M X Domestic Sorghum Price
- 2.5840
-11.59
Sorghum Import Volume
- 3.8964
12
- 6.009
Sorghum Border Price
- 1.2470
- 4.945
*Five percent significance level
Table 5.4 Unit Root Test Results for the Milk Powder-Mod els
Variable
Estimated
Coefficient
Lag Length
(Months)
ADF Statistic*
Milk Powder Volume
-0.31842
- 3.399
Milk Powder Border Price
-0.36167
-4.122
Canada Milk Powder Price
- 0.00964
- 1.597
Milk Powder Volume
-1.7752
- 8.992
Milk Powder Border Price
- 2.4022
- 10.61
Canada Milk Powder Price
-0.38018
-5.964
L ev el
Form
156
5.3 Exchange Rate Volatility Results

One aim of this study is to shed light on the impact of different exchange rate volatility
measures on agri-food trade flows. To this end four alternative volatility measures are
considered, two of them are standard deviation-based volatility measures (moving
average of the standard deviation of the exchange rate and the absolute percentage
change of the exchange rate) and a volatility measure based on an ARCH or GARCH
model, and a non-parametric exchange rate volatility measure. These will now be
presented in turn. The following section deals with these volatility measures.
5.3.1. Volatility Results from a Standard Deviation Models

The volatility of the exchange rate traditionally has been approximated by standard
deviation or variance-based measures. The existing literature reports at least eight of
these measures. Two standard deviation-based exchange rate volatility measures were
selected based on the availability of data and ease of interpretation. The first is the
moving average of standard deviation of the growth rate of the exchange rate defined as:
1
1/ 2
(5.4)
v ,= . ( rn
- ) Ei=1< Io8 z - . - | o8 z -2)2
where m is the order of the moving average, and Z is the exchange rate variable. The
application of this formula to the data, using three period moving average (three months)
yields a series that is integrated of order one. The second volatility measure is the
absolute percentage change of the exchange rate defined as;
(5.5)
157
where e is the spot exchange rate and t refers to time. The resulting series from the
application of these formulas to the exchange data are shown in appendix 5. The plot of
the standard deviation-based volatility measures reveals that there was a period of
tranquility, from 1989 to 1994. This period corresponds to the government-managed
period of exchange rate determination in Mexico while the period of extremely high
volatility in exchange rate is observed during the Mexican peso crisis in 1994-1996. After
this episode, the peso is floated and the Mexico-US exchange rate is characterized by
high but manageable volatility periods (Figure 5.1). Figure 5.1 also shows that there are
no appreciable differences in the plot between both measures of volatility during the first
sixty months, from 1989 to 1994, but from 1995 to 2004, the variance of the volatility
measure V I is larger that the variance of the volatility measure V2.
Figure 5.1 Exchange Rate Volatility Measures: Moving Average Standard

Deviation (VI) and Percentage Change of the Real Exchange Rate (V2)
40.0
35.0
30.0
25.0
20.0
15.0 4
io .o
5.0
0 .0 -f
-5.0 -r
-
10.0
Jan-1989
Dec-2004
Months
rerceniage Change of ER; V2
Moving Average Std Deviation; VI
158
5.3.2. Volatility Results from an ARCH and GARCH Model

Previous research on exchange rate volatility and trade has not taken into account the
notion that exchange rates are generated through a stochastic process and as a
consequence, exchange rate data exhibit certain characteristics that violates the
assumptions of a standard regression analysis. As indicted by Bollerslev et al. (1992),
financial time series are typically heteroskedastic, leptokurtic and exhibit volatility
clustering. He suggests that these features could be handled more successfully by
modeling the volatility of the time series as conditional on past behaviour. Since their
introduction by Engle (1982) ARCH models, and their subsequent generalizations by
Bollerslev (1986) to GARCH models, proved to be very useful in measuring exchange
rate volatility.
The general form of the GARCH (p, q), model proposed by Bollerslev (1986) is as
follow:
a ) = (0 + Y .^ e li +
;= i
1 = co+ a (L )? + P (L )a f ,
(5.6)
1=1
where: <7? is the conditional variance,

t.j is the moving average term;
et is the disturbance term;
The literature indicates that many forms of volatility can be represented using the
particular GARCH (1, 1) specification (Bollerslev 1986; Enders 2003). The (p, q) in
parentheses is a standard notation in which the first number refers to how many
autoregressive lags or ARCH terms appear in the equation, while the second number
159
refers to how many moving average lags are specified, which is often called as the
number of GARCH terms.
Although this model is directly set up to forecast for just one period, it turns out
that based on the one-period forecast, a two-period forecast can be made. By repeating
this step, long-horizon forecasts can also be constructed. For the GARCH(1,1), the twostep forecast is a little closer to the long-run average variance that is the one-step forecast,
and the distant-horizon forecast is the same for all time periods as long as a + P < 1, i.e.
the coefficients of the ARCH and GARCH terms sum up to less than one. The GARCH
model is mean reverting and conditionally heteroskedastic but has a constant
unconditional variance (Engle, 2001). In this research, the GARCH (p, q) model has been
applied to real monthly exchange rate time series for the period 1989:1 to 2004:12. The
next section describes results from this model.
Following the empirical specification in chapter four, the procedure to calculate
volatility of the exchange rate was as follows: the residuals from a preliminary OLS
estimation were tested for the presence of ARCH behaviour in data. In particular the
following tests were performed.
Tests for Non-Normality. If the normality assumption is used to describe the conditional
error distribution then a property of ARCH is that the unconditional error distribution will
be non-normal with high values for kurtosis;
Test for Autocorrelation. The autocorrelation structure of the residuals and the squared
residuals were inspected. An indication of ARCH is that the residuals will be
uncorrelated but the squared residuals will show autocorrelation. Test statistics are given
160
by Ljung-Box-Pierce Portmanteau tests, Q(20) and Q2(20), for up to twentieth order

serial correlation in the residuals and the squared residuals respectively, table 5.5 shows
these test results.
Table 5.5 Statistic to Test for ARCH Errors

Statistic
p-value
59.7942
27986.7
0.000
Q(20)
31.14
0.053
Q2(20)
7.33
0.995
Test
Excess kurtosis
Jarque-Bera test
The high value for excess kurtosis indicates that the distribution is characterized
by leptokurtosis and an indication that the unconditional error distribution will be non
normal. The standard normal value for kurtosis is three (Engle, 2001). The Jarque-Bera
test statistic provides clear evidence to reject the null hypothesis of normality for the
unconditional distribution of the monthly percentage changes in the Mexico-US exchange
rate at the five percent significance level. Both kurtosis and the Jarque-Bera statistic
suggest the presence of ARCH errors.
Regarding the test for autocorrelation (the Ljung-Box-Pierce portmanteau tests)

'y
Q(20) and Q (20), for up to twentieth order serial correlation in the residuals and the
squared residuals have been used. The sample autocorrelation function of the residuals
shows no autocorrelation while the squared residuals showed autocorrelation, both of
which indicate the presence of ARCH errors.
161
Estimation of a GARCH (p, q) Model

Based on the results of table 5.5, a GARCH model was estimated, the estimated equation
was as follows:
a ; = a>+ a 2t_x + (ia]_x
(5.7)
The estimation of the GARCH model requires the use of a numerical optimization
algorithm . To apply the GARCH technique, it is required to find the best fitted time
series model for the real exchange rate. An ARIMA model could perform this task. Once
it was done, a GARCH (p, q) model to capture the time varying conditional variance of
the real exchange rate can be formulated. The estimated model in this study has the
following specification:
a) ALnER = ao + ait.,-+ eu
b) e, 1 1 - N (0, h, ),
0 h, = A + Z A A , + S A V ,
1=1
(5.8)
1=1
where ALnER is the difference in the log of the real exchange rate with respect of the
previous period, I t-i is all relevant and available information at time t-i. st.j is the moving
average term; et is the disturbance term and hxis the conditional variance. Equations (5.8)
need to be estimated jointly for the sample period. If coefficients on e ,_i and h t-i are
2 The algorithm works as follows: starting values for the parameters o f the mean equation are obtained from
an OLS regression. For an ARCH(q) or GARCH(p,g) process the starting values for the parameters o f the
conditional variance equation are obtained from a regression of the OLS squared residuals on a constant
and q lags. For a GARCH process, the starting values for the parameters on the lagged conditional
variances are set to zero. Pre-sample estimates are required for the squared errors and the conditional
variances (McCullough and Renfro, 1999). An initial Hessian estimate is constructed from the outer
product of the gradient, analytic expressions for the derivatives are used (Bollerslev, 1986). Once initial
values are calculated, the estimation can proceed. SHAZAM uses a quasi-Newton algorithm. A description
of the quasi-Newton method is given in Judge, Griffiths, Hill, Lutkepohl and Lee (1985; pp. 958-960).
162
statistically significant, significant ARCH and GARCH effects are said to exist in the
data. The predicted values from in (c) of equation 5.8 provide a measure of real exchange
rate uncertainty.
Figure 5.2 Exchange Rate Data Expressed as First Difference of the Logarithm of
Real Mexico-US Exchange Rate (direr)
0.20
Months
Different ARCH and GARCH specifications were tested from ARCH (1) to
ARCH (8) and from GARCH (1,1) to different values of (p, q). Only the results from the
model with values p=l and q= l are reported in the text. Based on significance of the
parameters of the lagged squared error term and the lagged forecast variance (conditional
variance), a GARCH (1,1) specification was chosen. For this model, a test for non
normality and a test for autocorrelation were conducted. The data for this model is the
difference in the log of the real Mexico-US exchange rate and it is plotted in figure 5.2
163
shows high variation of the exchange rate particularly during the period of 1994-1996
which represents the Mexican Peso crisis period.
As can be seen in table 5.6, the GARCH term is highly significant. For this
specification, the coefficients of the ARCH and GARCH terms sum up to less than one
and both are positive. Thus, the model is mean reverting and conditionally
heteroskedastic, but has a constant unconditional variance (Engle, 2001).
Table 5.6 Results for the GARCH (1,1) Measure of the Exchange Rate Volatility
Dependent Variable: ALRER
Method: M L - ARCH
Sample: 1989:01 2004:12
Convergence achieved after 48 iterations
Coefficien
Std. Error
z-Statistic
Prob.
0.019748
0.005118
3.858659
0.0001
Variance Equation
0.000589
0.000233
2.531241
0.0114
A R C H (l)
0.003133
0.001578
1.985747
0.0471
G A R C H (l)
0.967598
0.017016
56.86385
0.0000
R-squared
-0.00262
Mean dependent var
0.0176
S.E. o f regression
0.04325
Akaike info criterion
-3.5852
Sum squared resid
0.35166
Schwarz criterion
-3.5173
Log likelihood
Durbin-Watson stat
348.1810
1.7819
The conditional variance function from table 5.6 is specified in the following equation:
h, = 0.00059 + 0.0031e]_x + 0.967V ,
(5.9)
164
The predicted values from this equation, namely conditional variance, provide a measure
of Mexico-US exchange rate volatility. The series of the conditional variance used as a
measure of exchange rate volatility are reported in appendix 8.
The series of the conditional variance is used in the estimation of the import
demand equations (as a measure of exchange rate uncertainty). It is calculated using the
prediction values from equation 5.9. The resulting volatility values are plotted figure 5.4
Figure 5.3 Mexico-US Exchange Rate Volatility from GARCH (1,1)

^
0.007 -|
0.006 -
0.005 0.004 -
0.003 1
0.002
0 .0 0 1
ON
OO
On
ON
On
ON
OO
ON
o
o
Months
Looking at the normality test, the model shows non-normal errors since the
Jarque-Bera test is significant at 5 percent level. Thus the residuals are highly leptokurtic.
The Jarque-Bera statistic rejects the hypothesis of normal distribution. Furthermore, the
165
high values of kurtosis (greater than 3) and negative skewness (see table 5.3) suggest the
presence of ARCH residual in the data.
Figure 5.4 Results of the Normality Test for GARCH (1,1) Model
100
Series: Standardized Residuals
Sample 1989:01 2004:12
Observations 192
80.
60.
40-
Mean
Median
Maximum
Minimum
Std. Dev.
Skewness
Kurtosis
0.036896
0.178402
2.351028
-10.26257
1.000310
-6.178925
61.47810
Jarque-Bera
Probability
28579.24
0.000000
20 .
-10
-8
-2
ARCH LM test is performed to test whether the standardized residuals exhibit

additional ARCH. If the variance equation (5.8) is correctly specified there should be no
ARCH left in the standardized residuals. The test shows a probability of 0.937 using one
lag and 0.997 using 16 lags, which lends to the conclusion that the variance equation is
correctly specified (Table 5.7 and 5.8). If the GARCH (1,1) model describes the data then
the standardized residuals should have zero mean and unit variance and be independently
and identically distributed (Bollerslev, 1986). The results from the estimation of model
(5.8) exhibit a mean value of 0.0085 and a variance of 0.9914 which are fairly close to
the desired values. So, it is an indication that the residuals are independently and
identically distributed.
166
Table 5.7 Results for the ARCH LM Test Using One Lag
F-statistic
Obs*R-squared
0.064932
0.131857
Probability
Probability
Test Equation:
0.937152
0.936198
Dependent Variable: STD_RESIDA2

Sample: 1989:03 2004:12
Variable
Coefficient
Std. Error
t-Statistic
Prob.
0.978232
0.573764
1.704939
0.0899
STD R ESID A2(-1)
0.026198
0.073127
0.358252
0.7206
STD R ESIDA2(-2)
0.002161
0.073125
0.029554
0.9765
R-squared
0.000694
Mean dependent var
S.E. o f regression
7.781392
Akaike info criterion
6.957011
Sum squared resid
11322.86
Schwarz criterion
7.008280
Log likelihood
-657.9161
F-statistic
0.064932
Durbin-W atson stat.
2.000633
Prob (F-statistic)
0.937152
1.006770
Table 5.8 Results for the ARCH LM Test Using Sixteen Lags
ARCH Test using 16 lags
F-statistic
0.273465
Obs*R-squared
4.713546
Probability
Probability
0.997720
0.997004
Test Equation:
Dependent Variable: STD_RESIDA2
Sample; 1990:05 2004:12
Coefficient
Variable
0.908177
C
STD_RESIDA2(-1)
0.018690
STD_RESIDA2(-2)
-0.003214
STD_RESIDA2(-3)
0.149832
0.037721
STD_RESID A2(-4)
STD_RESIDA2(-5)
0.017332
STD_RESIDA2(-6)
-0.032438
STD_RESID A2(-7)
-0.018622
STD_RESIDA2(-8)
-0.004250
STD_RESIDA2(-9)
-0.002147
STD_RESID A2(-10)
0.044182
STD_RESID a2(-11)
0.010525
STD_RESIDA2(-12)
-0.009333
STD_RESIDA2(-13)
-0.021199
STD_RESIDA2(-14)
-0.016366
STD_RESID A2(-15)
-0.012280
STD_RESIDA2(-16)
-0.005713
Log likelihood
-613.7478
Durbin-W atson stat
2.000125
Std. Error
t-Statistic
0.690415
1.315407
0.079302
0.235682
0.079307
-0.040523
0.079297
1.889504
0.080165
0.470548
0.080218
0.216067
0.080225
-0.404341
-0.232236
0.080188
0.080199
-0.052990
0.080200
-0.026771
0.080187
0.550985
0.080223
0.131194
0.080214
-0.116350
0.080161
-0.264451
0.079293
-0.206406
0.079304
-0.154845
0.079296
-0.072052
F-statistic
Prob(F-statistic)
Prob.
0.1903
0.8140
0.9677
0.0606
0.6386
0.8292
0.6865
0.8167
0.9578
0.9787
0.5824
0.8958
0.9075
0.7918
0.8367
0.8771
0.9427
0.273465
0.997720
167
5.3.3 Volatility Measure Results Derived from a Non-Parametric Model

The empirical implementation of the Baltagi and Li (2001) procedures involves four
steps. The first step is to estimate an Ordinary Least Squares (OLS), regressing exchange
rate fundamentals (real GDP, interest rate, money demand, and price level) on real
Mexico-US exchange rate and obtain the residuals. The second step is to form squared
residuals and to regress them on lagged squared residuals using kernel regression . Step
three consists of taking the squared root of fitted values and then divides all variables
entering the initial regression by the square root of the predicted values. Finally, the last
step is performed applying OLS using the resulting variables from step three, the
predicted values from this regression represent a measure of the real exchange rate
volatility. What follows next is a detailed discussion of results obtained at each of these
steps.
Stage 1: Initial OLS Regression to Obtain Residuals

The purpose of the initial regression is to obtain residuals of the Mexico-US exchange
rate. To this end, exchange rate data were regressed on a set of explanatory variables. The
estimated model corresponds to the model of exchange rate determination of Dombush
(1976) and Frenkel (1979). The selected fundamentals are the Mexico-US real exchange
rate (ER); money demand (M l); real output (GDP); interest rate (TB); and consumer
price index (CPI). The specification is as follow:
ER=p0 + PiM l* + p2GDP* + p3TB* + p4CPI* + s
(5.10)
3 This method performs a local averaging o f the observations when estimating the regression function.
168
where ER is the Mexico-US real exchange rate, M l is the money demand, GDP is real
gross domestic product, i is the interest rate and CPI is the general consumer price index.
The star (*) indicates the difference between the fundamentals in Mexico and United
States. The R-square in table 5.10 implies that the selected variables explain a high
proportion of Mexico-US exchange rate variability, and also that all coefficients are
statistically significant at 5 percent.
Table 5.9 Results from an OLS Rgression to get Residuals

R-Square
Variables
Ml
GDP
TB
CPI
CONSTANT
0.9645
R-Square
adjusted
0.9638
Estimated
Coefficient
Standard
Error
t-ratio
P-value
-0.9590
-14.0840
-0.1950
-0.0597
226.2400
0.1199
0.3374
0.0314
0.0220
5.6780
-7.9960
-41.7400
-6.2040
-2.7090
39.8500
0.0000
0.0000
0.0000
0.0070
0.0000
Stage 2: Nonparametric regression with a normal kernel function

In the second stage squared residuals were formed and regressed on lagged squared
residuals using kernel regression4. While there are many specific methods of
nonparametric regression, for simplicity of estimation and interpretation the kernel
regression with a normal kernel function was selected in this study. A central issue in
nonparametric regression is the selection of smoothing parameters.
4 In general, non-parametric regression analysis traces the dependence o f a response variable (Y) on one or
several predictors (Xs) without specifying in advance the function that relates the response to the predictors
169
The selection of the smoothing parameter is always related to a certain

interpretation of the smooth. The literature contains different approaches, but for ease of
calculation, an automatic selection provided by the econometric software based on the
data set was chosen in this study. A good automatically selected parameter is always a
useful starting point (Hardle, 1994). Moreover, an automatically selected bandwidth for
kernel smoothers can facilitate comparison with a standard and can be used for
investigation of high-dimensional regression data (Fox, 2000).
Hardle (1994) argued that the accuracy of the conditional variance depends
mainly on the smoothing parameter. The kernel regression was performed using a
procedure in SHAZAM software that uses a normal kernel function and the
bandwidth/smoothing parameter is set by default. The bandwidth parameter provided by
the software was 0.37011.
Stage 3: Getting the Predicted Values as a Measure of Volatility

To generate an expression of the volatility of the exchange rate using the fitted values
from the kernel regression, five new variables were generated by dividing all the initial
variables by the square root of the fitted values (SQRUHAT) from the kernel regression.
The estimated regression follows,
R = Po + PiM + P2G + P3T + P4 C +
(5.11)
where:
R = ER/SQRUHAT
M = M l/SQRUHAT
G = GDP/S QRUHAT
T = TB/SQRUHAT
C = CPI/SQRUHAT
170
The results of the OLS regression from equation 5.11 are presented in table 5.10. The
above regression generates an R-square of 0.71 which indicates that the selected variables
explain the exchange rate changes fairly well. The estimated parameters of all variables
but one are statistically significant at 5 percent.
Table 5.10 Results from an OLS-Third Regression to get Conditional Variance

R-SQUARE = 0.7104
Name
R-SQUARE ADJUSTED = 0.7042

Coefficient
Std Error
t Ratio
P-Value
1.8800
0.3042
6.1800
-0.9003
0.1151
-7.8190
0.0000
0.0106
0.0929
0.1143
0.9090
0.3991
0.0483
8.2670
0.0000
-0.6115
1.1660
-0.5247
0.6000
CONSTANT
0.0000
The values from the results presented in table 5.10 are taken as a measure of volatility of
the Mexico-US exchange rate. Figure 5.4 shows the behaviour of the exchange rate
volatility generated by this method for the studied period. It is clear that volatility rose as
expected after month 78, which correspond to the beginning of the floating period of the
exchange rate in Mexico.
171
Figure 5.5 Volatility from the Semi-Parametric Estimation (VN)
25.0 0
20.00
15.00 -
10.00
5.00 -
0.00
OO
Ov
CN
M onths
5.4 Summary
This chapter discusses the main results of unit root test and estimated four alternative
measures of Mexico-US real exchange rate volatility. The ADF test was conducted to
determine the integration properties of individual series used in each commodity model.
The ADF statistic from the unit root test for the variables in level form is used to test for
the presence of unit root in the data, while the ADF statistic for the series in first
difference form is used to test and confirm if the series is integrated of order one.
To determine optimum lag length for each series, the Akaike (1973) Information
Criterion (AIC) and Schwarz (1978) Criterion (SBC) were applied. Both criteria selected
the same optimal lag length. The results for the ADF test show that all macroeconomic
172
variables, as well as the price and quantity variables included in each of the four
commodity models are non-stationary, have a unit root in the univariate representation.
The results from the estimation of four different volatility measures show that the
volatility measures based on standard deviation, moving average standard deviation (V I)
and percentage change of the exchange rate (V2) are very similar in tracking the
exchange rate fluctuations. With respect to the GARCH model, it was found that
GARCH(1,1) describes the exchange rate data well based on a set of standard tools .
Finally, the semi-parametric regression was performed using the kernel regression
approach with a normal kernel function to obtain nonparametric measures of the MexicoUS exchange rate volatility. These four volatility measures are incorporated in the
estimation of the import demand functions for the selected commodities in chapter six.
The next chapter reports and discusses the results of the cointegration analysis.
173
CHAPTER 6
MAXIMUM LIKELIHOOD COINTEGRATION RESULTS
6.1 Introduction
The purpose of this chapter is to present and discuss the results of Maximum Likelihood
Cointegration analysis. This chapter is divided into four sections. The first section deals
with an overview of the econometric modeling strategy used to obtain the results. Section
two is devoted to determine the optimal lag structure of each model. The third section
focuses on the determination of the cointegrating vectors and the discussion of the longrun cointegrating relationships. The final section summarizes the main results and
concludes the chapter.
6.2. Econometric Modelling Strategy

To test for the presence of long-run cointegrated relationships among the variables in
each of the trade models for maize, sorghum, milk powder, and fresh tomato, the
Johansen maximum likelihood cointegration method (Johansen 1988; Johansen and
Juselius 1990) is used. This approach involves the following steps:
Determine the optimal lag length for each model;
Determine the number of cointegrating vectors ( r ) for each model;
Normalize of the r cointegrating vectors to obtain the normalized coefficients of
the estimated equation and the corresponding loading vector.
The sequence in which the relevant variables entered the model is determined on the
basis of economic reasoning. For all models, disposable income is considered the most
important variable in explaining movements in exchange rates (Taylor, 1995). This is
174
because a countrys disposable income is expected to influence, in the long-run, the

quantity of imports to a much higher degree than the exchange rate or its variability. The
remaining variables in each model entered the cointegration analysis in a slightly
different way since for each model there is a different set of explanatory variables.
For the fresh tomato model, the sequence in which the relevant variables entered
the cointegration analysis was as follows: The United States personal disposable income
(USY) was first. This was followed by the real Mexico-United States exchange rate (ER),
the volatility of the exchange rate (V), the import price of tomato in the United States
(Pt), the price of a related commodity (the price of tomato in the United States), and the
ratio of wage rates (defined as the United States farm wage rate over the Mexico farm
wage rate). This arrangement generated estimates with theoretically expected signs in
each model. For maize and sorghum models the sequence was to enter real disposable
income in Mexico, exchange rate, exchange rate volatility variables, the change in hogs
inventory (INV), own price and the price of a related commodity, price of domestic
maize and price of domestic sorghum for maize and sorghum models respectively. For
milk powder models, after income, exchange rate and exchange rate volatility, the
sequence was to enter a milk short-fall variable (MSF), price of imports (Pm), and the
price of milk powder imported from Canada into Mexico. The MSF variable is defined as
the difference between domestic consumption and domestic supply of milk and dairy
products in Mexico.
Johansen Maximum Likelihood method essentially uses a Vector Autoregression
formulation. The standard VAR is a reduced form model in which economic theory is
used to guide the empirical analysis. The VAR models are widely used for multivariate
time series analysis (Sims (1980), Liitkepohl (2001), Hamilton (1994), Hendry (1995),
175
Dhrymes (1998) and Clements and Hendry (2002), Harris and Solis (2003)) and are
becoming increasingly popular in time series econometric analysis. The results obtained
using the VAR formulation are sensitive to the lag-length choice and alternative
specifications of the lag structure of a given VAR model generate quite different results
(Haffer and Sheenan, 1991). Therefore, it is important to determine the optimal lag
structure of each of the VAR models.
6.3. Optimal Lag Length Selection Results

For each commodity four models were estimated, each with a different volatility
measure. The appropriate lag length specification for each VAR was determined in two
steps: first the log likelihood values for each of the four models, from one to twelve lags,
were determined. These values are used to construct Sim s (1980) modified likelihood
ratio tests (ML-Test). Sims test allows for testing cross-equation restrictions. It permits
researchers to test the equivalency of models with different lag-lengths and has an
asymptotic x l distribution with degree of freedom d, equal to the number of restrictions
imposed. To examine the entire range of possibilities with respect to trade periods, the
models are estimated under different lag structures, from one period (month) to twelve
periods. Tables 6.1 to 6.4 report a subset of the optimal lag-length choice results, detailed
results of lag-length determination are provided in appendixes 5-20. Although the
selected commodities are quite different from each other, the optimal lag-length obtained
for each of the four commodity models are fairly close to each other as the optimal lag
length varies from nine to ten lags.
176
Table 6.1 Optimal Lag Length Determination for the Tomato Models
Number of
periods lags
Number of Lags
Log Likelihood
M L-Test
LRm X2 *(d>
Test Result
14.02
Reject Lag 8
14.02
Accept Lag 9
14.02
**
TOM ATO MODEL V I

Eight to Nine
Nine to Ten
Ten to Eleven
Eight to Nine
Nine to Ten
Ten to Eleven
8
9
9
10
10
8
9
9
10
10
-39.503
27.967
-40.202
-40.202
11.508
-40.645
-40.645
6.758
TOMATO MODEL V2
-36.938
-37.705
-37.705
-38.015
-38.015
30.675
14.02
Reject Lag 8
8.045
14.02
Accept Lag 9
7.534
14.02
**
TOM ATO MODEL VG

Eight to Nine
Nine to Ten
Ten to Eleven
8
9
9
10
10
-40.907
-41.834
-41.834
-42.434
-42.434
20.403
14.02
Reject Lag 8
3.599
14.02
Accept Lag 9
-332.696
14.02
**
17.413
14.02
Reject lag 9
7.507
14.02
Accept lag 10
-2.866
14.02
**
TOM ATO MODEL VN

Nine to Ten
Ten to Eleven
9
10
10
11
11
-7.318
-7.988
-7.988
-8.614
-8.614
Eleven to Twelve
* df = 7, 5 percent.
** Once an optimal lag is achieved, no further lags are tested
For the tomato models, the optimal lag length using volatility measure V I, V2
and VG is nine lags (%2= 14.02, 83 d f which is significant at the five percent level),
while it is ten using the volatility measure VN. For the maize models, the optimal laglength using V I is ten lags while using V2, VG, and VN it is nine lags (Table 6.2).
177
Table 6.2 Optimal Lag Length Determination for Maize Models (New data)
Number of
periods lags
Num ber of Lags
Log Likelihood
M L-Test
LRmx2 *(d )
Test Result
14.294
14.02
Reject Lag 9
7.586
14.02
Accept Lag 10
-67.062
14.02
**
21.808
14.02
Reject lag 8
13.074
14.02
Accept Lag 9
8.636
14.02
**
17.495
14.02
Reject lag 8
13.403
14.02
Accept Lag 9
9.249
14.02
**
18.308
14.02
Reject lag 8
10.757
14.02
Accept Lag 9
8.080
14.02
**
MAIZE MODEL V I
Nine to Ten
Ten to Eleven
Eleven to Twelve
-37.353
10
-37.902
10
-37.902
11
-38.535
11
-38.535
MAIZE MODEL V2
Eigth to Nine
Nine to Ten
Ten to Eleven
-34.713
-35.258
-35.258
10
-35.761
10
-35.761
MAIZ M ODEL VG
Eigth to Nine
Nine to Ten
-9.830
-10.267
-10.267
10
-10.782
Ten to Eleven
10
-10.782
Eigth to Nine
-4.483
-4.941
-4.941
10
-5.354
10
-5.354
MAIZE MODEL VN
Nine to Ten
Ten to Eleven
The results of the optimal lag-length determination for the sorghum models are
presented in table 6.3. For all sorghum models, the optimal lag-length is ten. For milk
powder models using V I and V2 the optimal lag-length appears to be nine, but for
models using VG and VN it is ten (Table 6.4). While it is tempting to attribute the
differences in optimal lag-lengths to the specifications of volatility measure, a definitive
answer in this regard requires additional investigation which is beyond the scope of this
research.
178
Table 6.3 Optimal Lag Length Determination for Sorghum Models (New data)
Number of
periods lags
Num ber o f Lags
Log Likelihood
M L-Test
LRm x2 *(d)
Test Result
SORGHUM M ODEL V I
Nine to Ten
Ten to Eleven
Eleven to Twelve
-37.335
10
-38.060
10
-38.060
11
-38.649
11
-38.649
18.854
14.02
Reject Lag 9
7.077
14.02
Accept Lag 10
-1.220
14.02
**
14.686
14.02
Reject lag 9
10.516
14.02
Accept Lag 10
-1.473
14.02
**
16.164
14.020
Reject lag 9
6.955
14.020
Accept Lag 10
-1.978
14.02
**
881.0
14.02
Reject lag 9
7.580
14.02
Accept Lag 10
-1.489
14.02
**
SORGHUM MODEL V2
Nine to Ten
Ten to Eleven
Eleven to Twelve
-35.127
10
-35.692
10
-35.692
11
-36.568
11
-36.568
SORGHUM MODEL VG
Nine to Ten
Ten to Eleven
Eleven to Twelve
-10.367
10
-10.989
10
-10.989
11
-11.568
11
-11.568
SORGHUM MODEL VN
Nine to Ten
Ten to Eleven
Eleven to Twelve
-39.403
10
-5.519
10
-5.519
11
-6.151
11
-6.151
179
Table 6.4.0ptimal Lag Length Determination for Milk Powder Models (New data)
Number of periods
lags
Number of
Lags
Log Likelihood
M L-Test
LRmx2 *(d)
Test Result
14.218
14.02
Reject Lag 9
9.269
14.02
Accept Lag 10
-1.290
14.02
**
15.302
14.020
Reject lag 9
10.247
14.020
Accept Lag 10
-1.471
14.020
**
1355.2
14.02
Reject lag 8
12.529
14.02
Accept Lag 9
7.620
14.02
**
30.737
14.02
Reject lag 8
9.235
14.02
Accept Lag 9
10.671
14.02
**
M ILK POW DER M ODEL V I

Nine to Ten
Ten to Eleven
Eleven to Twelve
-31.684
10
-32.231
10
-32.231
11
-33.003
11
-33.003
MILK POW DER MODEL V2

Nine to Ten
Ten to Eleven
Eleven to Twelve
-29.314
10
-29.902
10
-29.902
11
-30.756
11
-30.756
M ILK POW DER MODEL VG

Eigth to Nine
Nine to Ten
Ten to Eleven
-38.502
-4.621
-4.621
10
-5.103
10
-5.103
M. POW DER MODEL VN

Eigth to Nine
Nine to Ten
Ten to Eleven
-32.949
-33.718
-33.718
10
-34.073
10
-34.073
180
6.4. Maximum Likelihood Cointegration Results

Since the reported results in chapter five showed that all relevant variables in different
trade models are integrated of order one, the use of OLS would lead to spurious
regression. An appropriate econometric technique to be used in this situation is
cointegration analysis. The estimated cointegrated relationship represents the stable longrun relationship to which the variables in the system have a tendency to return in the
long-run (Engle and Granger, 1987). Since the variables were log-transformed, the
estimated coefficients represent long-run elasticities (Deaton and Muellbauer, 1980).
Following the Johansen Maximum Likelihood approach, the value of the trace and the
maximum eigenvalue test statistics are used in this study to determine the presence as
well as the number of cointegration relationships in a VAR system.
The trace statistic tests the null hypothesis that the cointegration rank is equal to r
against the alternative that the cointegration rank is k. On the other hand, the maximum
eigenvalue statistic tests the null hypothesis that the cointegration rank is equal to r
against the alternative that the cointegration rank is equal to r+1. When there is only one
cointegrated relationship, which is the case for most of the models in this research, the
relevant cointegration vector is given by the first column of matrix f i (in Equation 4.13
where n = a f i ) under the largest eigenvalue. The cointegrating vector represents the
stable long-run relationship among all variables in a particular model. On the other hand,
the weights (oc's) represent the speed of adjustment of different variables following a
shock to the system in equilibrium. A low absolute value of the coefficient,
a,
would
indicate slow adjustment while a high coefficient indicates rapid adjustment to the long
run equilibrium (Johansen and Juselius, 1990).
181
The critical values of the trace and maximum eigenvalue statistics at the five
percent level of significance reported by MacKinnon (1999) are used in this study
because they provide more refined critical values for the these tests than those available
elsewhere. The next section presents and discusses the results of the Johansens
maximum likelihood cointegration analysis applied to the four commodities. Then, an
attempt is made to bring out informally which volatility measure provides better results
than others and why. The above discussion is conducted taking into account the
properties of different volatility measures as well as consistency of the results with theory
in terms of both signs and magnitude of the long-run coefficients.
6.4.1 Results of Cointegration for the Tomato Models

6.4.1.1 Cointegration Results for Tomato Model Using V I
The full set of results, from the Johansens maximum likelihood cointegration analysis,
for all commodities, is provided in appendix 21-36. For the fresh tomato model using
volatility V I (moving standard deviation of the exchange rate) both, the trace and the
maximum eigenvalue statistics rejects the null hypothesis of no cointegration (r = 0) at
the five percent level of significance. Based on the trace statistic three cointegrating
vectors are present while the maximum eigenvalue test suggests that there are two
cointegrating vectors in this system. For the sake of consistency, two cointegrating
vectors were selected1. Similarly, in all other models analyzed in this study, the number
of cointegrating vectors is selected by employing both test statistics.
1 While the theory suggests that both the trace and maximum eigenvalue test select the same number of
cointegrating vectors, in a small sample, trace test tends to have more than the maximum eigenvalue test. It
is suggested in the literature to employ both tests in empirical work to determine the number of
cointegrating vector in a given system (Lutkepohl et al, 2001).
182
Following the above criteria, it was determined that for tomato model using V I,
there are two cointegraing relationships. The long-run relationships between United
States tomato imports from Mexico and the set of explanatory variables can be
represented as linear functions.
Qt = - 0.687Pt - 2.068Pus +1.008USY + 1.425ER - 0.362V1 + 1.424W1
(6.1)
Pt = -0.603Qt + 0.357USY + 0.737ER + 1.202V1 + 1.287Pus + 0.400W1
(6.2)
The first cointegrating vector, shown in equation 6.1, is normalized using the
coefficient of import quantity while the second cointegrating vector represented by
equation 6.2, is normalized using the coefficient of own price of the commodity. In
equation 6.1 all estimated coefficients but one appear to be consistent with demand
theory. The own price has the expected negative sign implying that if the price of fresh
tomatoes imported into the United States from Mexico increases by 10 percent, the
quantity imported would fall by 6.8 percent. The substitute price does not exhibit the
expected positive sign. A possible explanation could be that the tomato produced in the
United States serves as a complement to the imports of tomato from Mexico by supplying
the eastern seaboard region of the United States.
The elasticity of income exhibits the expected positive sign. This implies that
when per capita personal disposable income in the United States increase, say by 10
percent, imports of fresh tomatoes from Mexico increase also by 10 percent. The high
value of income elasticity could be due to the high quality of Mexican tomatoes imported
183
into the United States and may also reflect the fact that during the winter season high
quality vine-ripped tomatoes from Mexico are highly appreciate by consumers in the
United States.
The positive sign of the exchange rate variable indicates that the bilateral MexicoUnited States exchange rate exerts a positive long run effect on the United States imports
of tomatoes from Mexico. If there were a 10 percent devaluation of the Mexican peso, the
United States import demand for fresh Mexican tomatoes would increase by about 14.2
percent. The exchange rate volatility variable exhibits an expected negative sign,
suggesting that exchange rate uncertainty reduces the volume of tomato imported from
Mexico into the United States. Note that all results are consistent with the expected utility
maximization model outlined in chapter three, so that ^ 0 > 0 and
< 0. These
dER
av,.
results are also consistent with the findings in the literature (e.g. Chowdhury (1993),
Koray and Lastrapes (1989), Qian and Varangis (1994), and Gagnon (1993)) about the
relationship between volatility and trade flows.
The Mexico-United States wage ratio has a positive sign as expected. This
indicates that if agricultural wage rate in the United States increases, imports of fresh
tomatoes from Mexico would also increase. Thus, an increase in the price of labour in the
importing country is followed by an increase in tomato imports from Mexico. Since the
wage rate in the United States was eight to nine times higher than that in Mexico during
the sample period, this result is quite reasonable. Malaga et al (2001) found that wage
rate differential between Mexico and the United States explained about 2 percent of the
growth of Mexico-United States tomato trade flows during 1990 to 1998.
184
The second long-run relationship in equation 6.2 can be interpreted as an inverse

demand function. Note that except for the volatility variable, all other variables have
signs as expected in this equation.
While the estimated weights ( a ' s ) for imports of fresh tomatoes into the US are
low, they suggest that in the event of any shock to the system, exchange rate, the related
price and volatility of the exchange rate will respond faster than any other variable to
bring this system back to the long run equilibrium.
6.4.1.2 Results of Cointegration for Tomato Model Using V2

For the tomato model with the volatility measure V2, the trace and the maximum
eigenvalue statistics reject the null hypothesis of no cointegration at the five percent level
of significance. Based on the trace statistic there are four cointegrating vectors while the
maximum eigenvalue test suggests that only two cointegrating vectors exist in this
system. Thus, it was concluded that there are two cointegrating relationships in this
system. The long-run cointegrating vectors from these models are as follows
Qt = - 0.532Pt + 2.44lPus + 0.596USY + 1.813ER - 0.479V2 + 1.386W1,
(6.3)
Pt = -0.187Qt - 0.226USY - 0.083ER - 0.104V2 + 0.083Pus + 0.127W1.
(6.4)
Equation 6.3 shows that all variables have coefficients with theoretically expected signs.
The negative sign of the own price variable implies a negative relationship between the
185
United States imports of fresh tomatoes from Mexico and tomato import price. Since the
coefficient of own price variable is less than one, it implies that the demand is inelastic.
Thus, for a given increase in import price, the imported quantity will decrease
proportionally less than the change in own price of tomato. Based on the positive sign of
the related price, tomato produced in the United States is a substitute for tomato imported
from Mexico.
The income elasticity of import demand is inelastic. It implies that when percapita personal disposable income in the United States increases, say by 10 percent, the
imports of fresh tomato from Mexico increases by almost 6 percent. The above value of
income elasticity of demand suggests that tomato is a normal good.
The positive value of the exchange rate variable indicates that a devaluation of the
Mexican peso exerts a positive long run effect on the United States imports. The
exchange rate volatility exhibits an expected negative sign. Some studies in the literature
dealing with agricultural trade flows and volatility have reported a negative relationship
between imports and exchange rate volatility (e.g. Anderson and Garcia, 1989; Sarker,
1993; Fabiosa, 2002; and Cho, et.al. 2002).
The elasticity of the Mexico-United States wage ratio shows the expected positive
sign, i.e., as the agricultural wage rate in United States increases, then imports of
tomatoes from Mexico also increases. It is expected that when rural wage rates in the
United States become expensive relative to the price of labour in Mexico, tomato imports
from Mexico would also increase. The positive impact of a lower wage rate in Mexico
with respect to that in the United States were also reported in the literature by Malaga et
al, (2001); Love and Lucier (1996); and Schwentesius and Gomez Cruz (1992) among
186
others. They argued that the United States imports of fresh vegetables were sensitive to
labour cost differentials between Mexico and the United States.
The estimated weights ( a ' s ) for fresh tomato model with V2, suggest that in the
event of any shock to the system exchange rate volatility, the tomato import price as well
as the price of tomato imported from Canada will respond faster than any other variables
to bring this system back to the long run equilibrium.
6.4.1.3 Results of Cointegration for Tomato Using VG

For the tomato model with exchange rate volatility generated by a GARCH model both
the trace and the maximum eigenvalue statistics rejects the null hypothesis of no
cointegration at five percent level of significance. While the trace statistic suggests the
existence of two cointegrating vectors, the maximum eigenvalue statistic indicates that
there is only one cointegrating vector in this system. Based on the above result, one
cointegrating vector was chosen. This vector represents the long-run relationship among
the variables and is given by
Qt = -0.285Pt + 0.844Pus + 0.246USY + 0.589ER + 0.454VG + 0.503W1.
(6.5)
Equation 6.5 shows that all variables but volatility have coefficients with theoretically
expected signs. Since the coefficient of own price variable is less than one, it implies that
the demand is inelastic. Thus, for a given increase in import price, the imported quantity
will decrease proportionally less than the change in own price of tomato. Based on the
positive sign of the related price, tomato produced in the United States is a substitute for
187
tomato imported from Mexico. Also the income elasticity is less than one, which implies
a low response of import demand to changes in income.
The positive value of the exchange rate variable indicates that a devaluation of the
Mexican peso exerts a positive long-run effect on the United States imports. Since the
exchange rate volatility exhibits a positive sign, it implies that a positive relationship
exists in the long-run between Mexico-United States trade tomato flows and volatility of
the exchange rate. The above result is consistent with the hypothesis of risk-averse
importers examined in the theoretical model in chapter three of this study. The sign of the
wage rate ratio is consistent with the theoretical model and with tomato models using V 1
and V2 in this study. The estimated weights ( a 's ) for imports of fresh tomato suggest
that if an external shock affects the long run equilibrium, the price of tomato produced in
the United States respond faster than any other variables to bring this system back to the
long run equilibrium.
6.4.1.4 Results of Cointegration for Tomato Using VN

For the tomato model using exchange rate volatility generated by a non-parametric
method (VN), both the trace and the maximum eigenvalue statistics reject the null
hypothesis of no cointegration at the five percent level of significance. Both tests
statistics indicate that there are two long-run cointegrating relationships in the model.
These cointegrating vectors can be expressed as linear functions
Qt = - l.OlOPt - 1.878Pus + 0.220USY + 0.602ER - 0.229VN + 0.669W1 and
188
( 6 .6 )
Pt = - 1.050Qt - 0.484USY - 1.062ER + 0.414VN + 1.353Pus - 0.4846W1.
(6.7)
______
The above equations show that all parameters exhibit the expected theoretical
sign, the exception being the price of tomato produced in the United States which has a
negative coefficient, the own price elasticity is almost unity in equation (6.7). On the
other hand, the related price has a negative sign, suggesting a complementary relationship
between tomatoes produced in the United States and tomatoes imported from Mexico into
the United States. While exchange rate indicates a positive relationship between trade
flows and Mexico-United States exchange rate, the volatility of exchange rate exert a
negative impact on trade flows. The sign of the wage rate ratio is positive and consistent
with the previous tomato models in this study. The estimated weights ( a ' s ) for imports
of fresh tomato suggest that if an external shock affects the long run equilibrium, the own
price of tomato and the import volume will respond faster than any other variables to
bring this system back to the long run equilibrium.
6.4.1.5 Relationship between Results using Different Exchange Rate Volatility

Measures on Tomato Models
The elasticity of the own price ranges from -0.29 using volatility V I to -1.01 using
volatility VN. Similarly, the cross price elasticity ranges from 0.84 when VG is used to
2.44 when the tomato model includes V2. Regarding the own price elasticity, the
literature reports values ranging from -0.31 to -0.62. For example, Salcedo-Baca (1990)
189
and Gutierrez (1983) found elasticities of -0.31 and -0.44 respectively while Malaga et al
(2001 ) found a value of -0.62.
The magnitude of income elasticity of import demand across models ranges from
0.22 to 1.00. The above values are consistent with income elasticity values found in the
literature, even though the estimation methods and samples were quite different. The
values of income elasticities found in the literature for the United States tomato import
demand range from 0.23 (Mitterhammer, 1978), to 1.47 (Salcedo-Baca, 1990). Malaga et
al. (2001) reported an intermediate value of 0.92 for the elasticity of income.
While the coefficients of exchange rate variable in all four models have the
expected sign, it is not the case for the volatility variable which has an unexpected
positive sign in one model. The positive sign is inconsistent with the theoretical model
outlined in chapter three. The negative sign of the coefficient of the exchange rate
volatility variable is obtained in three out of four cases which are consistent with the
comparative statics results presented in chapter three of this thesis.
It is worthwhile to note that Arellano et al (1998) and Malaga et al (2001) also
found a positive effect of the devaluation of the Mexican peso on the United States
imports of fresh vegetables. They reported that the United States fresh vegetable trade
resulted in higher United States imports in periods of devaluation of the Mexican peso,
particularly when it was allowed to move freely against the United States dollar. Malaga
et al (2001 ) also argued that in the case of fresh tomato, the peso devaluation explained
87 percent of the short-run changes in the United States tomato imports from Mexico
during the period of 1993-1996.
190
6.4.2 Results of Cointegration for Maize Model

6.4.2.1 Cointegration Results Using Volatility VI
For the maize model with moving standard deviation of the exchange rate (VI), the trace
as well as the maximum eigenvalue statistic rejects the null hypothesis of no
cointegration at the five percent level of significance and they indicate the presence of
just one cointegrating vector in this system. The following cointegrating vector represents
the long-run relationship among the variables in this model:
Qc = - 0.995Pc + 1.099Pmx + 0.424MXY + 0.405ER - 1.082V1 - 0.461INV
(6 .8 )
Equation 6.8 can be interpreted as the long-run import demand function for maize in
Mexico. In this equation, all variables exhibit a priori expected signs except for the
variable INV (hogs inventory) which shows a negative sign. The magnitude of the own
price coefficient suggests that the import demand function for the com from the United
States into Mexico is inelastic.
The exchange rate elasticity indicates that the bilateral exchange rate devaluation
exerts a positive long run effect on Mexicos maize imports from the United States.
Given the positive sign of the price of a substitute commodity, the domestic price of
maize produced in Mexico (Pmx) is considered a substitute for maize imported from the
United States. The coefficient of the exchange rate volatility variable exhibits a negative
sign as expected. Thus, in the long-run, uncertainty of the exchange rate exerts a
significant negative effect on maize trade flows between Mexico and the United States.
The coefficient of INV variable shows a negative sign which is contrary to the theoretical
191
expectation because M exicos maize imports consist of yellow com (80-90 percent)
which is mostly used as an input by hog producers, poultry producers, starch
manufacturers and flour millers (Avalos-Sartorio, 1998).
The estimated weights ( a ' s ) for imports of maize suggest that in the event of any
disturbance or shock to the system that affect the long run relationship, the quantity
imported and volatility of the exchange rate will respond faster than any other variables
to bring this system back to the long run equilibrium (Appendix 25).
6.4.2.2 Results of Cointegration Analysis for Maize Using V2

For model for maize using first difference of the real exchange rate (V2), both, the trace
and the maximum eigenvalue statistics reject the null hypothesis of no cointegration at
the five percent level. While the trace indicates the presence of two cointegrating vectors,
the maximum eigenvalue tests suggest the existence of just one cointegrating vector in
this system. The cointegrating vector located under the maximum eigenvalue represents
the long-run relationships among the variables in this system and can be specified as
follows
Qc = - 0.773Pc + 0.958Pmx + 0.339MXY + 0.365ER - 1.062V2 + 0.388INV
(6.9)
Equation 6.9 can be interpreted as the long-run import demand function for maize in
Mexico. In this equation, all variables have coefficients with the expected signs. The
volatility variable shows that a negative effect exists between imports of maize from the
United States into Mexico and exchange rate uncertainty.
192
6.4.2.3 Results of Cointegration for Maize Using VG

For the maize model using exchange rate volatility generated by GARCH model, both,
the trace and the maximum eigenvalue test reject the null hypothesis of no cointegration
at the five percent level of significance. Both statistics suggest the existence of just one
cointegrating vector (see appendix 27). This cointegrating vector represents the long-run
relationship among the variables in this system and can be written as

.
------------------------------------
Qc = -0.804PC + 1.268Pmx + 0.582MXY + 0.631ER - 0.422VG + 0.420INV
(6.10)
Equation 6.10 represents the long-run import demand function for maize in Mexico. In
this equation all the variables have coefficients with the theoretically expected signs. All
elasticity values of import demand for maize are inelastic with the exception of the
elasticity of the related price.
The sign of the exchange rate volatility indicates a negative relationship between
Mexico-US trade flows of maize and exchange rate uncertainty. The estimated weights
(ex's) for imports of maize using VG suggest that in the event of any disturbance
affecting the long run equilibrium, the quantity of maize imports and the import price
variable will responds faster than any other variable to bring this system back to the long
run equilibrium
6.4.2.4 Results of Cointegration for Maize Using VN

For the maize model using exchange rate volatility generated by non-parametric method
(VN), both the trace and the maximum eigenvalue tests reject the null hypothesis o f no
193
cointegration at the five percent level of significance. Both statistics indicate that there is
just one cointegrating relationship in this model. This cointegrating vector represents the
long-run relationship among the variables in this system and can be specified as
Qc = - 1.212Pc + 1.188Pmx + 0.320MXY + 0.481ER + 0.738YN + 0.853INV
(6.11)
Equation 6.11 can be interpreted as the import demand function for maize in
Mexico. The coefficients of all variables but volatility of the exchange rate exhibit
expected signs. Similar to previous maize models, the import demand for maize is
inelastic. The positive sign of the related price suggest that maize produced in Mexico
can be treated as a substitute of maize imported from the United States. W ith regard to
the income variable, the positive sign suggests that as per capita personal disposable
income in Mexico increases, the import demand for maize from the United States will
rise.
Volatility of exchange rate does not have the expected negative sign in this model.
However, the positive sign of the exchange rate coefficient is an indication that, in the
long-run, devaluation of the United States currency relative to the Mexicos currency will
lead to an increase in the volume of maize imported into Mexico from the United States.
The estimated weights ( a ' s ) for maize import demand with VN suggest that in the event
of any disturbance affecting the long run equilibrium, the volume of imports, the INV
variable and the own price of maize will respond faster than any other variable to bring
this system back to the long run equilibrium.
194
6.4.2.5 Relationships between Results using Different Exchange Rate Volatility

Measures on Maize Models
The own price and the price of the substitute commodity have the expected signs in all
four maize models. The price elasticity of import demand ranges from -0.77 using
volatility V2 to -1.21 using volatility VN while the cross price elasticity varies from 0.95
using V2 to 1.26 when the maize model includes VG. In all four maize models, the
income elasticity is between 0.32 and 0.58. These values are considered to be reasonable
for a highly traded commodity such as maize.
The exchange rate has the expected positive sign in all four maize models,
suggesting that a devaluation of the United States dollar will favour imports of maize into
Mexico from the United States and vice versa. The elasticity of exchange rate with
respect to imported volume into Mexico ranges from 0.36 using V2 to 0.63 using VG. On
the other hand, the volatility of exchange rate has a negative coefficient in three out of
four models.
The sign of the coefficient of the INV variable is as expected in all four maize
models. It suggests a positive relationship between changes in hog inventory in Mexico
and imports of maize from the United States. It could be partially due to the fact that the
demand for feed maize imported from the United States into Mexico has been increasing
during the last 10 years (USDA, 2004) as hog production in Mexico expanded gradually.
6.4.3 Results of Cointegration for Sorghum Model

6.4.3.1 Cointegration Results Using Volatility V I
For the sorghum model using volatility V I (as moving standard deviation of the
exchange rate), the trace as well as the maximum eigenvalue statistic rejects the null
195
hypothesis of no cointegration at the five percent level of significance. However, while

the trace test indicates the presence of just one cointegrating vector, the maximum
eigenvalue statistic indicates the presence of two cointegrating vectors in this model. The
long-run relationship is represented by the cointegrating vector located under the
maximum eigenvalue and can be specified as follows:
Qs = - 0.526PS + 0.342Psd + 0.157MXY + 0.142ER - 0.543V1 + 0.738INV
(6.12).
Equation 6.12 can be interpreted as the long-run import demand function for
sorghum in Mexico. The results show that all the variables in this equation have
coefficients with the expected signs. As expected, exchange rate and its volatility had a
positive and a negative effect respectively on import demand for sorghum in Mexico. The
demand is price inelastic, indicating that changes in prices are followed by proportionally
smaller changes in imports of sorghum. This result is true also for the cross price
elasticity. Given the positive sign of the related price (Psd) for Mexicos imports of
sorghum, domestic sorghum is a substitute for sorghum imported from the United States
into Mexico. The coefficient of hog inventory (INV) variable has a positive sign. A
priori, it was expected that given an increase in inventory, import demand of sorghum
would also increase. This is because all the sorghum imports are allocated to animal feed
industries in Mexico (Zahniser and Coyle, 2004).
The estimated weights ( a ' s ) for imports of sorghum suggest that in the event of
any disturbance or shock to the system that affect the long run relationship, inventory of
hogs in Mexico and import volume of sorghum will respond faster than any other
196
variable to bring this system back to the long run equilibrium. These are followed by
volatility of the exchange rate.
6 A 3 .2 Results of Cointegration Using Volatility V2

The results of cointegration analysis for the sorghum model with volatility V2 show that
the trace, as well as the maximum eigenvalue statistics, rejects the null hypothesis of no
cointegration at the five percent level of significance and they indicate the presence of
one cointegrating vector in this system. Thus, there is at most one long-run cointegraing
relationships in this model. The following cointegrating vector represents the long-run
relationship among the variables in this model:
Qs = - 0.590Ps + 0.132Psd + 0.129MXY + 0.138ER - 0.541V2 + 0.867INV
(6.13).
From the above relationship, equation 6.13 can be interpreted as the long-run
import demand function for sorghum in Mexico. In this equation all variables exhibit a
priori expected signs. The negative sign of the own price variable implies a negative
relationship between the Mexican imports of sorghum from the United States and
sorghum import price. The magnitudes of the own price and cross price coefficients
suggest that the long-run import demand function for United States sorghum in Mexico is
inelastic. The exchange rate elasticity indicates that a devaluation of the bilateral
exchange rate exerts a positive long run effect on Mexicos sorghum imports from the
United States. The coefficient of the exchange rate volatility exhibits a negative sign as
expected. Thus, in the long-run, uncertainty of the exchange rate exerts a negative effect
197
on sorghum trade flows between Mexico and the United States. The coefficient of INV
variable shows a positive sign, as expected. A probable explanation is that M exicos
sorghum imports are mostly used as input by hog producers in Mexico (e.g. AvalosSartorio, 1998).
any disturbance or shock to the system that affect the long run relationship, the inventory
and volatility will respond faster than any other variables to bring this system back to the
6.4.3.3 Results of Cointegration for Sorghum Using VG

For the sorghum model using exchange rate volatility generated by the GARCH model,
both the trace and the maximum eigenvalue tests reject the null hypothesis of no
cointegration at the five percent level of significance. However, while the trace statistics
suggest the existence of two cointegrating vectors, the maximum eigenvalue statistic
indicates the presence of one cointegrating vector (see appendix 31). This cointegrating
vector represents the long-run relationship among the variables in this system and can be
written as:
Qs = -0.638Ps + 0.241Psd +0.769MXY + 0.168ER - 0.416VG + 0.516INV
(6.14).
Equation 6.14 represents the long-run import demand function for sorghum in Mexico. In
this equation all variables have coefficients with the expected signs and the magnitudes
indicate inelastic responses.
198
The sign of the exchange rate variable indicates a positive relationship between
Mexico-US trade flows of sorghum and the bilateral Mexico-United States exchange rate.
On the other hand, the sign of the volatility variables indicates a negative relationship
with respect to sorghum import volume, as expected. The estimated weights ( a ' s ) for
imports of sorghum using VG suggest that in the event of any disturbance affecting the
long run equilibrium, the INV and the import volume as well as the own price of the
commodity will respond faster than any other variable to bring this system back to the
6.4.3.4 Results of Cointegration for Sorghum Using VN

For sorghum model using exchange rate volatility generated by a non-parametric method
(VN), both the trace as well as the maximum eigenvalue statistic rejects the null
hypothesis of no cointegration at the five percent level of significance. Similarly, both the
trace and the Maximum eigenvalue statistic indicate the presence of one cointegrating
vector in this model. The relevant long-run relationship is represented by the vector under
the largest eigenvalue and can be specified as:
Qs = - 0.090Ps + 0.141Psd +0.117MXY + 0.103ER - 0.251VN - 0.528INV
(6.15)
Equation 6.15 can be interpreted as the long-run import demand function for
sorghum in Mexico. The results show that all the variables in this equation have expected
signs with the exception of the INV variable. Exchange rate and its volatility suggest a
positive and negative relationship with respect to import demand of sorghum
respectively. The own price elasticity suggests an inelastic demand, indicating that
199
changes in prices are followed by proportionally smaller changes in imports of sorghum.

It is tme also for the cross price elasticity. Given the positive sign of the related price for
Mexicos imports of sorghum, domestic sorghum is a substitute for sorghum imported
from the United States.
The elasticity of the hogs inventory (INV) shows a negative sign. Given an
increase in inventory, it was expected that import demand of sorghum would also
increase. This is because all the sorghum imports are allocated to animal feed industries
in Mexico (Zahniser and Coyle, 2004).
any disturbance or shock to the system that affect the long run relationship, hog inventory
(INV) and imported volume will respond faster than any other variable to bring this
system back to the long run equilibrium.
6.4.3.5 Relationships between Results using Different Exchange Rate Volatility

Measures on Sorghum Models
The own price and the price of the related commodity have the expected a priori sign in
all four sorghum models. The price elasticity of import demand is consistent in three out
of four sorghum models, which range from -0.52 using volatility V I to -0.63 using
volatility VG while the cross price elasticity varies from 0.13 using V2 to 0.34 using
volatility V2. The personal disposable income has the correct sign in all models. In all
four models the value of income elasticity ranges from 0.12 to 0.76. No previous study on
this commodity could be found in the literature to compare the results for this
commodity.
200
The exchange rate has an expected sign in all four sorghum models, suggesting
that a devaluation of the United States will favour imports of sorghum into Mexico from
the United States and vice versa. The elasticity of exchange rate with respect to imported
volume into Mexico is inelastic in all four models and it ranges from 0.10 using VN to
0.16 using VG. The Volatility of the exchange rate has the expected sign in all four
sorghum models. The magnitude of the effects of exchange rate volatility on sorghum
imports into Mexico from the United States ranges from -0.25 using volatility VN to 0.54 using volatility V I and V2. The negative relationship between exchange rate
uncertainty and import volume is consistent with the expected utility maximization
hypothesis assuming risk-aversion.
The sign for the INV variable is the expected in three of four models while model
using VN has a positive sign. The positive relationship between import demand and hogs
inventory is not unexpected given the increasing domestic demand for sorghum in
Mexico to meet the growing demand for feed in the livestock industry. In the case of a
disturbance to the long-run equilibrium, prices, exchange rate volatility and hog
inventory would adjust faster than any other variables to restore the long-run equilibrium.
6.4.4 Results of Cointegration for Milk Powder Model

6.4.4.1 Cointegration Results for Milk Powder Using Volatility VI
For the milk powder model with moving average standard deviation of the exchange rate
(volatility V I), the trace and the maximum eigenvalue statistics reject the null hypothesis
of no cointegration at the five percent level of significance. Based on the trace statistic
there are two cointegrating vectors, while the maximum eigenvalue statistic indicated the
presence of one stable long-run relationships (see appendix 33). Thus, it was concluded
201
that there is at most one cointegraing vector in this system. The following cointegrating
vector represents the long-run relationship among the variables in this system:
Qm = - 1.248Pmk + 0.924Pca + 0.328MXY + 0.252ER - 0.343V1 + 0.144MSF
(6.16).
Equation 6.16 can be interpreted as the long-run import demand function for milk
powder in Mexico. In the above equation all the variables have the expected sign. The
magnitude of the own price coefficient suggests that the import demand function for
United States milk powder in Mexico is inelastic. The coefficient of the related price has
the expected positive sign implying a positive relationship between M exicos import
demand and related price. For milk imports, Mexico has a broad range of sources,
including the United States, Canada, Germany, New Zealand, Denmark and Ireland
among others. So, relatively high price responsiveness of import volume to changes in
prices is expected.
The sign of the exchange rate variable indicates that a devaluation of the bilateral
United States-Mexico exchange rate exerts a positive long-run effect on M exicos milk
powder imports from the United States. On the other hand, the negative sign of the
exchange rate volatility suggests a negative relationship between imports of milk powder
from the United States into Mexico and exchange rate uncertainty.
With regard to the response of import demand to changes in the real exchange
rate, it depends on the degree of elasticity of import demand. For this model, the
exchange rate elasticity is smaller than one, which implies an inelastic demand. The
income elasticity is inelastic, implying that given a 10 percent change in Mexican
income, milk powder import demand would change only by 3.28 percent.
202
The MSF variable (short-fall of milk and milk products in Mexico) has the
expected positive sign, indicating that as internal demand for milk and milk products rise,
imports from the United States also rise. The positive relationship between milk short-fall
and import volume is consistent with the increasing dependency of Mexico for milk and
milk products from the international market, especially from the United States. The
estimated weights ( a ' s ) for imports of milk powder suggest that in the event of any
disturbance to the system, both the own price and imported volume will respond faster
than any other variable to bring this system back to the long run equilibrium.
6A A .2 Results of Cointegration for Milk Powder Using V2

For the milk powder model with volatility V2 generated as first difference of the
exchange rate, the trace and the maximum eigenvalue statistics reject the null hypothesis
of no cointegration at the five percent level of significance. Both the trace and the
maximum eigenvalue test suggest the existence of two cointegrating vectors (see
appendix 34). Thus, it was concluded that there are at most two cointegraing vectors in
this system. The cointegrating vector under the largest eigenvalue represents the long-run
relationships among the variables in this model and can be represented as
Qm = - 0.391Pmk + 0.428Pca + 1.245MXY + 0.455ER - 0.799V2 - 0.205MSF
(6.17).
Pmk = - 0.263Qm - 0.988Pca - 0.252MXY - 0.508ER + 0.527V2 + 0.107MSF
(6.18).
203
From the above long-run relationship, equation 6.17 can be interpreted as the
long-run import demand function for milk powder in Mexico. In the above equation all
the variables, but the MSF variable, have the expected sign. The magnitude of the own
price coefficient suggest that the import demand function for United States milk powder
in Mexico is inelastic. The coefficient of the related price has the expected positive sign
implying a positive relationship between M exicos import demand and related price. The
sign of the exchange rate coefficient indicates that a devaluation of the bilateral exchange
rate exerts a positive long-run effect on Mexicos milk powder imports from the United
States. The negative sign of the exchange rate volatility suggests a negative relationship
between imports of milk powder from the United States into Mexico and exchange rate
uncertainty.
rate, it depends on the elasticity of import demand. For this model the exchange rate
elasticities is less than one, which implies an inelastic demand. The MSF variable (short
fall of milk and milk products in Mexico) has a negative sign which is not as expected.
The estimated weights ( a ' s ) for imports of milk powder suggest that in the event
of any disturbance or shock to the system that affect the long run relationship(s), both
imported volume and volatility will respond faster than any other variable to bring this
system back to the long run equilibrium.
6.4A.3 Results of Cointegration for Milk Powder Using VG

For the milk powder model using exchange rate volatility generated by the GARCH
model (VG), the trace and the maximum eigenvalue statistics reject the null hypothesis of
no cointegration at the five percent level of significance. Based on the trace statistic there
204
are two cointegrating vectors, while the maximum eigenvalue statistic indicated the
presence of just one stable long-run relationship (see appendix 35). Thus, it was
concluded that there is, at least, one cointegraing vector in this system. The following
cointegrating vector represents the long-run relationship among the variables in this
system
^m = -1.007Pmk + 1.433Pca + 0.785MXY + 0.210ER - 0.347VG + 0.777MSF
(6.1i
Equation 6.18 can be interpreted as the long-run import demand function for milk
powder in Mexico. In the above equation, all the variables have the expected economic
sign. The magnitude of the own price coefficient suggests that the import demand
function for United States milk powder in Mexico is unity. The coefficient on the related
price has the expected positive sign implying a positive relationship between M exicos
import demand of milk powder and related price, but the cross price elasticity is elastic.
The sign of the exchange rate coefficient indicates that the bilateral exchange rate
devaluation exerts a positive long-run effect on M exicos milk powder imports from the
United States. Similar to models with V2, the negative sign of the exchange rate volatility
suggests a negative relationship between imports of milk powder from the United States
into Mexico and exchange rate uncertainty.
rate, it depends on the magnitude of the elasticity of import demand. For this model, the
exchange rate elasticity is less than one, which implies an inelastic demand. The MSF
variable (short-fall of milk and milk products in Mexico) has the expected positive sign,
205
indicating that as internal demand for milk and milk products rise, imports from the
United States also rise.
of any disturbance or shock to the system that affect the long run relationship(s), both the
own price and the imported volume will respond faster than any other variable to bring
this system back to the long run equilibrium.
6.4.4.4 Results of Cointegration for Milk Powder Using VN

For the milk powder model with volatility VN, the trace and the maximum eigenvalue
statistics reject the null hypothesis of no cointegration at the five percent level of
significance. Based on both the trace and the maximum eigenvalue statistic there is one
cointegrating vector (see appendix 36). Thus, it was concluded that there is, at most, one
cointegraing vector in this system. The following cointegrating vector represents the
long-run relationship among the variables in this system:
Qm = -1.095Pmk + 1.372Pca + 0.405MXY + 0.208ER - 0.277VN + 0.443MSF
(6.19)
From the above long-run relationships, equation 6.19 can be interpreted as the
long-run import demand function for milk powder in Mexico. In the above equation all
the variables have the expected economic sign. The magnitude of both the own price and
related price coefficients suggest that the import demand function for United States milk
powder in Mexico is elastic. The coefficient of the related price has the expected positive
206
sign implying that milk powder imports from Canada into Mexico can be considered a
substitute for milk powder imports from the United States. The sign of the exchange rate
coefficient indicates a devaluation of the bilateral exchange rate exerts a positive long-run
effect on M exicos milk powder imports from the United States. The negative positive
sign of the exchange rate volatility suggests a negative relationship between imports of
milk powder from the United States into Mexico and exchange rate uncertainty. The MSF
variable (short-fall of milk and milk products in Mexico) has the expected positive sign,
indicating that as internal demand for milk and milk products rise, imports from the
United States also rise.
of any disturbance or shock to the system that affects the long run relationship, both
imported volume and own price of milk powder will respond faster than any other
variable to bring this system back to the long run equilibrium.
6.4A.5 Relationships between Results using Different Exchange Rate Volatility

Measures on Milk Powder Models
The own price and the price of the related commodity have the expected a priori sign in
all milk powder models. The price elasticity of import demand ranges from -0.39 using
volatility V2 to -1.24 using volatility V I while the cross price elasticity varies from 0.42
using V2 to 1.43 when the milk powder model includes VG. In all milk powder models,
income elasticity is inelastic, with the exception of V2, ranging from 0.32 to 1.24. From
the above elasticity values, it can be concluded that the exchange rate variable has a
significant effect on the long-run demand for milk powder imports into Mexico from the
United States.
207
The bilateral exchange rate has a positive sign as expected in all four models,
suggesting that a devaluation of the United States dollars will favour imports of milk
powder into Mexico from the United States. The elasticity of the exchange rate is
inelastic and ranges from 0.21 using VG and VN to 0.45 using V2. On the other hand,
volatility of the exchange rate shows the expected negative sign in all models. The
magnitude ranges from -0.27 for models using VN to -0.79 for using V2. The sign for the
MSF variable is as expected. The positive relationship between import demand of milk
powder and MSF reflects the increasing dependency of Mexico for milk powder from the
international market.
For milk models, the fastest speed of adjustment in the case of a disturbance to the
long-run equilibrium, given by the largest value of the coefficients in the weight matrix
(a), suggest that the import volume of the commodity, own price, and milk short-fall
would adjust faster than any other variable to restore the long-run equilibrium.
6.4.5 Comparison of Long-run Results across Commodities

One of the main objectives of this research is to determine the effects of exchange rate
changes and its volatility on agri-food trade flows. There is no consensus on the
appropriate volatility measure to be used in the trade equations. This section is devoted to
discuss and determine informally which volatility measure provides better results relative
to others. To accomplish the above, a summary of the main economic relationships
obtained through estimating various import demand functions are presented in Table 6.5.
The discussion take into account the consistency of the estimated results with theory, i.e.,
the estimated values are compared with the expected results in terms of signs, magnitudes
208
and the properties of the volatility measures employed. The results are compared among
models using different volatility specification.
Considering the fact that each of the selected commodities has good number of
substitutes and that small proportion of the consumers expenditure is devoted to these
commodities, they can be viewed as necessary goods. For such goods, it is expected that
the magnitudes of price and income elasticities be less than one. It is also expected that in
the long-run, own price, income and related price have stronger influence on import
demand than exchange rate, volatility and wage rate.
For tomatoes, it was found that while models using V I and VN have all the
coefficients with the a priori expected signs, except that of the related price, models V2
and model VG have all coefficients with the expected signs. In terms of magnitude of the
estimated coefficients, only model using VG has coefficients that are less than one while
the other models for tomato have two or three coefficients with magnitude around two,
which are contrary to the above reasoning.
For the maize model using V 1 two of six coefficients are greater than one and its
signs are as expected. For model using V2, all variables have the expected signs and only
the volatility of the exchange rate has coefficient greater than one. Model with VG has all
its coefficients with magnitude less than one with the exception of the coefficient of the
related price. Model using VN shows consistent magnitudes, but two variables have
coefficients greater than one and volatility has a positive sign. For sorghum, the model
using V I has all its coefficients smaller than one. Also all the coefficients have the
expected signs. Model using V2 has all its coefficients less than one and the signs are as
expected. On the other hand, model using VG not only have all its coefficients with the
expected signs but also the magnitudes are reasonable for a normal good. Models using
209
volatility VN has two coefficients greater than one and some of them are smaller in the
long-run with respect to the long-run. It is interesting to notice that model VG has
consistent coefficients in terms of signs and magnitudes.
The literature suggests that the volatility measure generated through the GARCH
method provides more accurate forecasts of conditional variances and co-variances than
those based on variance-standard deviation or non-parametric methods because of its
ability to model time-varying conditional variances. So, this method seems more suitable
for modeling exchange rate risk than others. Based on the above discussion, it can be
concluded that the model using a volatility measure generated by the GARCH method
provides better results than those obtained by using V I, V2 and VN.
Comparing results across the four commodity models, it was found that, for all
import demand functions, models using VG provide more consistent results with theory
than models using other volatility measures. For the four models, all the variables have
the a priori expected signs. Similarly, the magnitudes are reasonable for each of the four
models. The cointegration results suggest that in the long-run, a positive relationship exist
between income and exchange rate with the import volume of the commodity while a
negative relationship exists between own price and the import volume of the commodity.
The above results also suggest that the United States imports of tomato and M exicos
imports of maize, sorghum and milk powder are significantly influenced by the bilateral
Mexico-United States exchange rate.
It is interesting to note that exchange rate volatility (VG) had a negative
relationship with import volume in fifteen of the sixteen models estimated. The above
results imply that volatility of exchange rate have an adverse effect on the Mexico-United
States trade flows of tomatoes, maize, sorghum and milk powder in the long-run.
210
Table 6.5 Summary of Long-run Cointegration Results for Four Commodities using
Different Exchange Rate Volatility Measures
Models for Tomatoes
Variable
Own Price
Related Price
US Income
Exchange Rate
Volatility
W age Rate
Variable
Own Price
Related Price
M X Income
Exchange Rate
Volatility
Inventory
Variable
Own Price
Related Price
M X Income
Exchange Rate
Volatility
Inventory
Variable
Own Price
Related Price
M X Income
Exchange Rate
Volatility
M ilk Shortfall
Expected
Sign
(-)
(+)
(+)
(+)
(-)
(+)
Expected
Sign
(-)
(+)
(+)
(+)
(-)
(+)
Expected
Sign
(-)
(+)
(+)
(+)
(-)
(+)
VI
V2
-0.687
-0.532
+2.441
-2.068
+1.008
+0.596
+1.420
+1.810
-0.360
-0.480
+1.420
+0.380
Models for M aize
VI
V2
-0.772
-0.995
0.958
1.099
0.423
0.339
0.405
0.365
-1.082
-1.061
0.461
0.387
Models for Sorghum
VI
V2
-0.526
-0.590
0.342
0.132
0.157
0.129
0.142
0.138
-0.543
-0.541
0.738
0.867
M odels for M ilk Powder
VG
VN
-0.285
+0.844
+0.246
+0.589
-0.454
+0.503
-1.010
-1.878
+0.220
+0.602
-0.529
+0.669
VG
VN
-0.803
1.267
0.582
0.631
-0.421
0.420
-1.212
1.188
0.320
0.481
0.738
0.852
VG
VN
-0.638
0.241
0.769
0.168
-0.416
0.516
-0.090
0.141
0.116
0.103
-0.251
-0.528
Expected
Sign
VI
V2
VG
VN
(-)
(+)
(+)
(+)
(-)
(+)
-1.248
0.924
0.328
0.252
-0.343
0.144
-0.391
0.428
1.245
0.455
-0.799
-0.205
-1.007
1.433
0.785
0.210
-0.347
0.777
-1.095
1.372
0.405
0.208
-0.277
0.443
1.5 Testing Hypothesisi on the Cointegration Vectors

One of the objectives of this study is to determine if exchange rate and the volatility of
exchange rate have significant effects on agri-food trade flows between Mexico and the
United States. To accomplish this objective in a long-run context, formal hypothesis tests
have been performed on results obtained from various cointegration models. In particular
211
the following hypotheses are tested: (a) changes in exchange rate and its variability have
insignificant effect on trade flows (H0: fier = 0,
H 0:
= 0) in the long-run, (b) the
combined effect of exchange rate and its volatility on trade flows is insignificant (H0: Per
= Pv = 0 ) in the long-run.
Finally, the appropriateness of using exchange rate as a separate variable in the
trade model is tested as H0: p = pPrice. The above hypotheses are implemented using the
general model /3 = Hep2 which reflects relevant linear restrictions imposed on the
cointegrating vectors. The matrix H is a pxs matrix with r < s < p and s equal to the
number of variables minus the number of restrictions.

To test the hypothesis of the significance of individual coefficients, the null hypotheses:
H0: fier = 0 and H0: flVi = 0 are used where fier = estimated coefficient of the exchange
rate and f i Vi = estimated coefficient of the volatility variable. In each case, the restriction
is imposed using a separate H matrix. For example, to test Ho: f3er=Q, the following Hi
matrix has been employed
To test this hypothesis, a likelihood ratio test is used;
LR = T"ZLn1=1
U -i )
(1-4)
, where A y and
are
the eigenvalues from the restricted and unrestricted models respectively. The test statistic is compared to
2
jjfo 9 5 with (r (p-s)) degrees of freedom, where r is the number o f cointegrating vectors in the original
model, p is the number of variables in the model and V is the total number o f variables minus the number
o f restrictions. For example for model for tomato using V I, the number o f degrees of freedom is (2(7 - 6 ))
=
2.
212
1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
1
0
0
0 0
0 0
0 0
0 0
0 0
1 0
0 1
The null hypothesis is rejected for all sixteen commodity-specific models at the
five percent level of significance. Similarly, the null hypothesis related to the effects of
the volatility of the exchange rate on the Mexico-United States agri-food trade flows in
the long-run is rejected for all the sixteen commodity-specific models at the five percent
level (Table 6 .6 ). Thus, both exchange rate and the volatility o f the exchange rate have
significant influence on the trade flows of the selected commodities between Mexico and
The United States in the long-run.
Table 6.6 Long-Run Test Results on Significance of Individual Coefficients on

Critical
Value*
Models
H 0. Per = 0 ,
LR Statistic
pr = o
Per
Pv
Models
LR Statistic
Critical
Value*
Per
Pv
Tomato V I
5.99
16.86
17.25
Sorghum V I
3.84
6.914
19.966
Tomato V2
5.99
14.34
16.72
Sorghum V2
3.84
8.121
19.917
Tomato VG
3.84
18.60
13.44
Sorghum VG
3.84
7.057
17.277
Tomato VN
5.99
13.93
16.28
Sorghum VN
3.84
4.938
13.782
M aize V I
3.84
9.134
13.175
M ilk Powder V I
3.84
5.821
6.671
M aize V2
3.84
9.558
9.755
M ilk Powder V2
3.84
8.067
13.782
M aize VG
3.84
8.041
27.439
M ilk Powder VG
3.84
9.641
18.093
M aize VN
3.84
7.910
11.023
M ilk Powder VN
3.84
6.945
14.981
* X2 (0.95): Critical Values at 5 Percent level o f significance with r { p - s ) degrees of freedom.
213

A joint test on the regression coefficients of the exchange rate and exchange rate
volatility was also conducted to test the null hypothesis that exchange rate and volatility
of the exchange rate jointly have no effect on the Mexico-United States trade flows (i.e.,
the test is that the regression coefficients of exchange rate and volatility are jointly equal
zero). The restrictions are imposed using the following H 2 matrix
1 0
0 0
1 0
0
0
0
0
1 0 0
0 1 0
The null hypothesis is rejected for all sixteen commodity-specific models at the five
percent level of significance (Table 6.7). The results imply that the combined effect of
exchange rate and its volatility on the Mexico-US trade flows of the selected
commodities in the long-run is also statistically significant. Thus, if either variable is
ignored in an empirical analysis, bias due to mis-specification could result.
214
Table 6.7 Long-Run Test Results on the joint Significance of the Exchange Rate and
Volatility of the Exchange Rate on Mexico-US Trade Flows
H 0: Per = Pv 0
Model
Critical Value*
LR Statistic
Critical Value*
LR Statistic
Tomato VI
Tomato V2
9.49
9.49
18.159
21.045
Sorghum VI
Sorghum V2
5.99
5.99
20.847
20.075
Tomato VG
5.99
22.033
Sorghum VG
5.99
18.854
Tomato VN
9.49
19.303
Sorghum VN
5.99
16.848
Maize VI
Maize V2
5.99
5.99
13.208
10.155
Milk Powder VI
Milk Powder V2
5.99
5.99
18.347
12.872
Maize VG
5.99
14.222
Milk Powder VG
5.99
18.219
5.99
14.494
Milk Powder VN
5.99
18.032
Maize VN
Model
* X2 (0.95): Critical Values at 5 Percent level of significance with r ( p - s ) degrees of freedom.
6.5.3 Test on the Equality of the Coefficients of Foreign Prices and Exchange Rate
Despite the recent growth in the trade literature focusing on the effects of exchange rate
on trade, it does not provide conclusive evidence on the appropriate way to incorporate
exchange rate in a trade model. Traditionally, exchange rate has been incorporated into
trade models by expressing all prices in common currency units or using a composite
relative price variable (e.g. Carone, 1996). This implies that both foreign prices and the
exchange rate have equal coefficients. Some analysts argued that this is too rectrictive
and used the exchange rate and foreign prices separately into the trade model.
This study incorporated exchange rate directly into the trade equations to estimate
commodity specific import demand functions. To determine if it was appropriate to do
so, an equality restriction on /? coefficients was formulated to verify that the coefficient
of exchange rate is in fact equal to that of the foreign price. The H 3 matrix, as specified
below, is used to test the null hypothesis of H 0 : J32j - fi5j for i , j = 1,2,3
215
p; where
P 2j = estimated coefficient of exchange rate and J35j = estimated coefficient of foreign

price. In the specific case of the M exicos import demand for maize, as an example, this
restriction is imposed using the following matrix H 3 :
1 0 0 0 0 0
0 1 0 0 0 0
0
0
1 0 0 0
0 0 1 0 0
0 0
0 0 0 0 1 0
0 0 0 0 0 1
1 0 0
The null hypothesis of the equality of foreign price and exchange rate coefficients
is rejected for all sixteen commodity-specific models at the five percent level of
significance (Table 6 .8 ). These results imply that when estimating the import demand
function of tomato, maize, sorghum and milk powder, it is appropriate to include
exchange rate directly into the trade model. Therefore, the approach followed in this
study to incorporate exchange rate as a separate variable into the trade model seems to be
appropriate.
Table 6.8 Long-Run Test Results on Equality of Coefficients of Exchange Rate and
Foreign Price on Mexico-US Import Demand
B -r e s tr ic tio n s : H 0: p 2j = Psj f o r i j = 1 ,2 ,3 ,...., p ;
M odel
Tomato V I
Tomato V2
Critical Value*
LR -Statistic
5.99
5.99
8.721
8.620
Sorghum V I
Sorghum V2
M odel
Critical Value*
LR-Statistic
3.84
3.84
9.329
8.821
Tomato VG
3.84
8.742
Sorghum VG
3.84
9.852
Tomato VN
5.99
6.898
Sorghum VN
3 .84
8.202
M aize V I
M aize V2
3.84
3.84
7.944
9.067
M ilk Powder V I
M ilk Powder V2
3.84
3.84
8.733
9.989
M aize VG
3.84
15.657
Milk Powder VG
3 .84
10.520
3.84
9.671
M ilk Powder VN
3.84
8.922
M aize VN
* X2 (0.95): Critical Values at 5 Percent level o f significance with r ( p - s ) degrees of freedom.
216
6.6 Summary
This chapter presented the main results of maximum likelihood cointegration analysis on
commodity models, each using four different volatility measures. Lag-length selection
results reveal that the optimal lag-length for tomato is nine. Also for maize models the
optimal lag-length is nine, with the exception of model V I. For sorghum models the
optimal lag-length is ten. For milk powder models using volatility V I and V2 the optimal
lag-length is ten while for models using VG and VN is ten. The long-run results for all
commodities are consistent in terms of their expected signs. The only exception is the
coefficient of the hog inventory variable (INV) in one of the sorghum models and the
variable MSF in one of the milk powder models.
Although many factors affect agricultural trade, changes in exchange rate play a
significant role. While the magnitude of the impact of exchange rates on Mexico-United
States trade flows varies by commodity, they are, nevertheless, significant in the longrun. With regard to consistency of the results with theory, it was found that models using
VG seem to provide better long-run results than those provided by models using V I or
V2, both in terms of signs and magnitudes of the estimated coefficients.
Comparing results of the four models for each commodity as well as across the
four commodity models, it was found that, for all import demand functions, models using
VG provide more consistent results with theory than models using other volatility
measures. For the four commodity models using VG, all the variables have the a priory
expected signs with the exception of the volatility of the exchange rate variable which is
positive for maize and milk powder models. Similarly, the magnitudes are fairly
consistent for each of the models. The hypothesis of the significance of the effects of
exchange rate and its volatility on the Mexico-US agri-food trade in the long-run was
217
formally tested. The results suggest that both exchange rate and the volatility of the
exchange rate have significant influence on the trade flows of the selected commodities
between Mexico and The United States in the long-run.
With regard to the speed of adjustment for most import demand equations, the
results suggest that own price, the trade volume, and exchange rate would adjust faster
than any other variable to restore the equilibrium condition. The next chapter presents the
results of the Vector Error Correction models for each of the selected commodities. These
results represent short-run effects of exchange rate and exchange rate volatility and other
variables on Mexico-United States trade flows of the selected commodities.
218
CHAPTER 7
RESULTS OF THE VECTOR ERROR CORRECTION MODEL
7.1 Introduction
The derivation of the long-run relationship is an important step in understanding the
relationship between trade flows and its main determinants. However, a good
understanding of short-run relationships between variables entering the model is also
important for policy analysis. The estimated parameters from the Johansen Maximum
likelihood cointegration analysis are used to formulate the Vector Error Correction Model
(VECM). The estimated parameters from the VECM are interpreted as short-run
elasticities of import demand with respect to each of the independent variables. This
chapter is devoted to accomplish three objectives: (i) the estimation of the VECM; (ii) the
discussion of the short-run results, with particular emphasis on its main explanatory
variables, such as income, prices, exchange rate, and exchange rate volatility; and (iii) a
brief discussion of the policy implications of both short and long-run results. Before
proceeding with the discussion of the VEC results, an overview of how the VEC
mechanism works is presented.
7.2 An Overview of the Vector Error Correction Model (VECM)

Since the introduction of the error correction approach by Sargan (1964) to the economic
literature, the use of error correction models has proved to be a useful tool in linking the
long-run results to the short-run dynamics of an economic process. The importance of the
ECM in the cointegration analysis was demonstrated by Engle and Granger (1997) who
219
showed that if two variables are integrated of order one and are cointegrated, then, they
can be modeled as having been generated by an ECM. Thus, the Engle and Granger (EG)
Representation Theorem established that models with valid ECMs entail cointegration
and, conversely, cointegrated series imply an error-correction representation for the
economic model. In the ECM, the long-run equilibrium relationships are used to impose
constraints on the short-run dynamics in economically meaningful ways.
In cointegration approach, relevant cointegration among the variables implies that
there is an underlying error correction process so that any deviation from the long run
equilibrium will be corrected through changes in the variables and the equilibrium will be
restored by the correction of the equilibrium error. Furthermore, the EG representation
theorem for dynamic modeling ensures that the ECM is not subject to "spurious
regression" problem. Finally, in an ECM derived from the EG Representation
Theorem , all variables are stationary and thus, the OLS is an efficient way to estimate
the parameters of the Vector Error Correction Model (VECM).
Having obtained the long-run cointegration relations using the Johansen approach,
it is now possible to reformulate equation 4.13 presented in chapter four and estimate the
VECM with the error-correction term(s) explicitly included in it. Consider the
reformulation of an unrestricted VAR as a VECM discussed in chapter four, so that,
az ,
= 2 r,A Z _, + nz,_, + ( , + / / + ,
j=1
(7.1)
where n = a(3' , with [3 being a matrix of long-run coefficients and a represent the speed
of adjustment to disequilibrium. For the estimation of the VECM, the number of error
220
correction terms which were determined in chapter six, are incorporated in equation 7.1
to obtain an empirical version of the model as follows:
AZ, = r,.AZr_. + a ( fr lZ t_l) + <l>Dt +ju + t
j=i
_ _
* -i
Where the terms
(7.2)
and
are the vector autoregressive (VAR)
;'= i
component in first differences and error-correction components, respectively. In this

formulation, Z t is a p x 1 vector of 1(1) variables and T . is a p x p matrix that represents
short-term adjustments among variables across p equations at the jth lag. The parameter
represents a (p x r) cointegrating vector present in the system and A denotes first
differences. The a is a p x r matrix of speed of adjustment parameters representing the
speed at which equilibrium error is corrected in this system. Dt is a set of monthly
dummy variables, ju is a p x 1 vector of constant; and e, represents a p x l vector of white
noise error terms. The Error Correction Model as specified in equation 7.2 is estimated to
determine the short-run relationships among the variables included in each model.
The estimation of the error correction model follows Hendrys (1987) "general-tospecific" methodology. Under this approach, the most general form of the model is
estimated first and then the dimension of the parameter space is reduced by sequentially
eliminating variables (lags) with insignificant coefficients. This process leads to a
parsimonious specification of a VECM model unique for each of the selected
commodities. Finally, any potential mis-specification, resulting from this method, is
diminished by formulating the estimated dynamic model in first differences (Brooks,
2002).
221
The initial estimation of the error correction model for each commodity included
twelve lags of the first difference of each relevant variable, a free-trade dummy variable,
a constant term, eleven seasonal dummy variables, twelve lags of the changes in the
dependent variable and the error correction term (ECIY0 lagged one period. The ECT is
derived from the cointegrating vector representing the long-run relationship among the
variables in each of the estimated models and it is through this term that the long-run
relationship is embedded in the ECM. Each of the ECT is obtained through the scalar
multiplication of the cointegrating vector and the Z, vector.
The main results of the Vector Error Correction Models applied to sixteen
commodity models are presented and discussed in the next section. Each model differs
from the others in terms of lag specification and the inclusion of a different volatility
variable. For all models, the estimated coefficients and their corresponding t-statistic are
9
reported along with the R , adjusted R and a set of standard diagnostic statistics. The
estimated coefficients represent short-run elasticities. The coefficient of the errorcorrection term must be negative and statistically significant to ensure dynamic stability
of the model. It also re-reconfirms cointegration relationships obtained in chapter six
(Harris and Sollis, 2003).
Since all VEC models include the same set of explanatory variables but different
volatility measures and the ECT, at least, part of the differences in expected signs and
magnitudes of the estimated short-run elasticity values can be attributed to the
specification of different volatility measures.
222
7.3 Results of the VECM: the US Import Demand for Fresh Tomatoes
To asses the adequacy of each model, the goodness of fit measured by the adjusted R ,
the F-statistic and a set of standard diagnostic tests are used1. For the tomato model, the
empirical results suggest that the statistical fit is quite satisfactory. The R values ranges
from 0.36 for model using VG to 0.46 for the model using V I and all F-statistics are
significant at five percent level. Since the EC model includes a lagged dependant
variable, the conventional Durbin-Watson test can not be used. Instead, one is required to
use the Durbin "h" statistic to test for autocorrelation. While the Jarque-Bera test results
consistently reject the null hypothesis of normality of the residuals at the five percent
significant level for all four models, the null hypothesis could not be rejected at one
percent. The Lagrange Multiplier (LM) statistic indicates that the null hypothesis of no
autocorrelation cannot be rejected at the five percent significant level. Thus, no serial
autocorrelation is present in the VEC models for tomatoes.
The results of the four VEC models for fresh tomato imports into the Unites
States from Mexico are reported in tables 7.1 and 7.2. All own price coefficients exhibit
the expected negative sign and are statistically significant at the five percent level. These
1 One of the key requirements is that the residuals follow a white noise process i.e., residuals have zero
mean, constant variance, and are uncorrelated with explanatory variables. A number o f tests are used to
assess model adequacy such as: (a) the goodness o f fit of the model (R2 and adjusted R2). i.e., what
percentage o f the total variation in the dependent variable is explained by the independent variables); an
F-test is used to test overall significance of the regression, the null hypothesis in this case is Ho: R2 = 0, the
population R2 is zero (all the regression coefficients are zero, excluding the intercept); (b) the DW-h test is
used for detecting first order autocorrelation (which is strictly necessary since all models includes a lagged
dependent variables; (c) the Jarque-Bera test for determining if the residuals are distributed as a normal
distribution. The test is based on the sample skewness and kurtosis of residuals and is a joint test for the
skewness and kurtosis of the disturbances. It is asymptotically distributed as chi-squared with two degrees
of freedom. The null hypothesis is that residuals are normally distributed. For a normal distribution, the
sample skewness should be zero and the value of the kurtosis is three. The Jarque-Bera test determines
whether the sample skewness and kurtosis are unusually different than their expected values, as measured
by a chi-squared statistic; (d) the Hansen test for coefficient instability. It includes a statistic for the
variance and a second one for the joint instability o f the model; and (e) finally, to test for serial
autocorrelation, the LM statistic is employed.
223
results are consistent with the results obtained in the long-run analysis and are also
consistent with the prediction of the theoretical model. The own price elasticity ranges
from -0.092 for the tomato model using VN to -0.15 using volatility V2 and suggest that
an increase in the price of imports leads to a decrease in the quantity of fresh tomato
imported by the Unites States from Mexico.
The price of the related commodity (price of tomatoes produced in the Unites
States) also exhibits the expected positive sign in all four models and it is significant at
the five percent level. The above results suggest than in the short-run, imports of tomato
from Canada into the Unites States are a substitute for tomato imported from Mexico.
The cross price elasticity of the US import demand ranges from 0.195 for tomato model
using volatility VG to 0.536 using volatility V2.
The coefficients of the income variable are all positive and significant suggesting
a direct relationship between income and quantity of fresh tomato imported into the
Unites States from Mexico. The magnitude of the elasticity of income with respect to
import demand ranges from 0.10 for tomato model using VN to 0.206 for tomato model
using V2. All short-run elasticities are smaller than the long-run elasticities, and hence,
are consistent with the Le Chatelier principle.
With regard to the exchange rate variable, all four models yield coefficients with
positive signs as expected. These results support the hypothesis that a depreciation of the
Mexican currency relative to the Unites States dollar makes imports into the Unites States
less expensive resulting in an increase of fresh tomato imports from Mexico. The
elasticity of the exchange rate for the four models ranges from 0.50 for tomato model
using VN to 0.80 for tomato model using V I.
224
Table 7.1 Error Correction Model Results for the US Tomato Import Demand
Volatility V I
Variables
Estimated
Coefficient
T-Ratio
5.921
Volatility V2
Variables
Estimated
Coefficient
T-Ratio
QT t-12
USY mi
ERm
0.178
4.051
0.206
4.515
0.706
0.029
-0.152
2.475
3.802
-3.406
0.536
4.487
0.127
3.399
-3.532
QT M2
0.331
USY mi
0.156
3.939
ERm
0.802
-0.064
2.921
-1.983
V 2
-0.117
-3.428
0.457
3.961
PTM2
PCA mi
V I ,.3
P T ,.12
P C A t.
W l t.8
ECT m
0.295
3.275
-0.048
-3.535
W l,.6
ECT m
-0.072
NAFTA
0.495
2.310
NAFTA
0.475
2.125
M 10
M il
-0.007
-1.078
1.078
M6
M il
-0.076
-0.985
0.116
2.804
CONSTANT
0.000
0.306
CONSTANT
0.078
0.348
2.731
R2
R2 Adjusted
0.459
0.426
R2
R2 Adjusted
0.431
0.399
F-Value
13.68
F-Value
12.17
DW -H Statistic
-0.116
DW-H Statistic
1.225
Skewness
0.271
0.000
Skewness
0.161
0.000
Kurtosis
1.259
3.000
Kurtosis
1.464
3.000
J-B Normal
8.946
9.210**
J-B Normal
6.214
9.210**
Instability Test:
Instability Test
Variance
0.467
0.748
Variance
0.528
0.748
Joint
3.849
3.690
Joint
3.265
3.150
20.780
35.172*
29.746
35.172*
L.M Statistic
L.M Statistic
* Denotes the critical value of chi-squared with 23 degrees of freedom.

** Denotes the critical value of chi-squared with two degrees of freedom.
With regard to the volatility of the exchange rate variable, three out of the four
models yield coefficients with negative signs as expected. Thus, in the short-run volatility
has a negative influence on the Unites States imports of tomato from Mexico. Only the
tomato model with volatility V2 indicates a positive relationship between exchange rate
volatility and the Unites States imports of tomato from Mexico. In all cases, the
coefficients of the volatility variable are significant at the five percent level. The above
results lend empirical support to the comparative statics results presented in chapter three
225
and suggest that risk-averse importers are discouraged by higher volatility episodes of the
Mexico-Unites States exchange rate.
Table 7.2 Error Correction Model results for the US Tomato Import Demand
Volatility VG
Variables
Estimated
Coefficient
QT t-12
USY,.10
0.198
0.110
ER,.i
V G t2
0.663
-0.043
-0.099
0.195
PTt-io
PCAt.
W l,.6
ECT,.!
NAFTA
M3
M il
CONSTANT
Volatility VN
Variables
T-Ratio
3.274
3.750
3.269
-2.671
-3.360
Estimated
Coefficient
QT M2
USY ,.9
0.169
ER,.!
0.499
-0.108
-0.092
VN,.2
0.098
T-Ratio
3.048
9.370
2.350
-2.123
-2.639
3.438
1.508
P T ,. i
PCA ,_!!
0.130
-0.030
0.427
2.936
-3.453
2.081
W l,.6
E C T ,.,
NAFTA
0.416
0.343
-0.041
0.523
0.011
0.070
0.809
0.809
Ml
M2
0.011
0.021
1.103
1.082
0.309
2.365
CONSTANT
0.375
3.152
4.825
-4.068
2.303
R2
R2 Adjusted
0.364
0.332
R2
R2 Adjusted
0.396
0.359
F-Value
9.198
F-Value
10.577
DW -H Statistic
1.677
DW -H Statistic
1.570
Skewness
0.414
0.000
Skewness
0.120
0.000
Kurtosis
2.221
3.000
Kurtosis
2.359
3.000
J-B Normal
8.209
9.210**
J-B Normal
8.875
9.210**
Instability Test
Instability Test:
Variance
0.704
0.748
Variance
0.732
0.748
Joint
2.777
2.890
Joint
3.280
3.150
26.816
35.172*
29.889
35.172*
L.M Statistic
L.M Statistic

The values of the coefficient of the volatility of the exchange rate range from 0.03
for the model using V2 to 0.11 for the model using YN. To the best of my knowledge, no
information exist in the published literature on the relationship between the exchange rate
226
volatility and trade flows in fresh vegetable from Mexico to the Unite States. So, the
results of this study can not be compared to those of a previous study.
The coefficients of the lagged dependent variable are all significant at the five
percent level and the estimated values vary from 0.17 to 0.33. These results illustrate the
importance of previous imports of fresh tomatoes on current import volume and suggest a
stable trade relationship in this commodity (i.e., the Unites States importers of fresh
tomatoes take the volumes they imported in past periods into account when deciding how
much to import in the current period).
Regarding the ratio of the Unites States-Mexico farm wage rate (W l), defined as
farm wage rate in the Unites States over farm wage rate in Mexico, the VECM yields
results in line with those obtained from the long-run analysis as all coefficients are
positive and statistically significant. These results are also consistent with the
comparative static results derived from the theoretical model. These results imply that an
increase in farm wage rate in the United States relative to that in Mexico will lead to an
increase in tomato imports from Mexico. The magnitude of the estimated coefficient
ranges from 0.13 using volatility VG to 0.34 for model using volatility VN. These results
are in line with those of Malaga et al (2001) who showed that increases in M exicos real
wage rates have on Mexicos ability to export fresh vegetables to the United States.
With regard to the Error Correction Term (ECT), the results show that all four
models yield negative and significant coefficients. These results corroborate the findings
of chapter six and reconfirm the presence of cointegrating relationships in the Unites
States import demand function for fresh tomato for Mexico. The negative sign of this
coefficient indicates that the direction of correction is towards the long-run equilibrium
227
while the size indicates the speed of adjustment towards the long-run equilibrium (Carone
1996; Harris and Solis 2003). The small values of the ECT coefficient obtained in all four
models suggest a slow adjustment towards long-run equilibrium.
The coefficient of the free trade dummy variable is positive and significant in all
four models, suggesting that NAFTA had a positive effect on the Unites States imports of
tomato from Mexico. Previous researchers argued that NAFTA has a small positive effect
on the Unites States tomato imports from Mexico and in the short-run, the peso
devaluation rather than NAFTA, explains most of the increase in the Unites States
imports of tomato and fresh vegetables from Mexico. Finally, none of the seasonal
dummy variables is statistically significant. In the spirit of the general to specific
approach each model was run using 12-lags for each variable and the 11 monthly
dummies and then, the variables with insignificant parameters were dropped from the
analysis. In the end, none of the monthly dummy variables was found to be significant in
the tomato import demand functions.
7.3.1 Main Findings for Tomato Models

To summarize, the sign of all explanatory variables are consistent with their theoretical
expectations and with the long-run results. The only exception is the sign of the volatility
variable in one of the four models.
With regard to the exchange rate, all four models have the expected positive sign
and the short-run elasticity of imports with respect to changes in exchange rate varies
from 0.33 to 0.44. These results are consistent with the findings of Arellano et al (1998)
and Malaga et al (2001).
228
The own price elasticity of import demand is inelastic in all four models while the
cross price elasticity is positive but shows greater variability across models. The own
price elasticity of import demand reported in the existing literature varies from -0.31 to 0.62 (Salcedo-Baca, 1990; Gutierrez, 1983; M alaga et al, 2001), which appears to be
much higher than those obtained in this study.
The income elasticity of import demand across all models is less than one. Once
again the short-run income elasticities obtained in this study are smaller than the values
of income elasticities reported in the literature, these values range from 0.23 to 1.47 for
the US imports from Mexico (Mitterhammer, 1978; Salcedo-Baca, 1990; and Malaga et
al., 2001).
The effect of changes in the ratio of wages is found to be positive in all models.
Results in the same direction were found by Love and Lucier (1996); and Schwentesius
and Gomez Cruz (1996). They reported that the United States imports of fresh vegetables
were sensitive to labour cost differentials between Mexico and the United States.
The error correction term is negative and significant in all four models for
tomatoes. These results reconfirm the presence of cointegrating vectors found in each
model discussed in chapter six. However, the small value of the estimated ECT
coefficient implies that the adjustment towards the long-run equilibrium is rather slow.
Finally, NAFTA had a significant positive impact on tomato trade but surprisingly none
of the monthly dummy variables appeared to be significant. In terms of relative size of
elasticities, the short run elasticities are, in general, smaller than their corresponding long
run elasticities, suggesting a relatively inelastic short run import demand for Mexican
fresh tomatoes in the United States.
229
7.4 Results of the VECM: the Mexicos Imports Demand for Maize
All short-run results of Mexican import demand for maize from the United States along
with a set of diagnostic statistics are reported in tables 7.3 and 7.4. The R2 ranges from
0.30 in model using V I to 0.32 in all other models for tomatoes. All F statistics are
significant indicating that the overall goodness of fit is satisfactory. The DW-h statistic
rejects the presence of autocorrelation. W ith regard to the normality test, while the null
hypothesis of normality of the disturbances is rejected at the five percent level, the
hypothesis cannot be rejected at the one percent level of significance. The Lagrange
Multiplier (LM) statistic indicates that the null hypothesis of no autocorrelation cannot be
rejected at the five percent significant level. Thus, the estimated import demand function
for maize seems satisfactory.
The coefficient of the own price of maize exhibits an expected negative sign in all
four models and are all significant at the five percent level. The value of the own price
elasticity ranges from 0.13 to 0.15. The coefficient of the price of the related commodity
(maize produced in Mexico) exhibits an expected positive sign and is significant at five
percent level in all four models. The estimated value of the cross price elasticity ranges
from 0.14 for maize model using volatility VN to 0.18 using volatility V I. The above
results are consistent with the results obtained from the long-run analysis. Maize
produced in Mexico is a substitute for maize imports from the Unites States. Both shortrun own price and cross price elasticity values are smaller than their long-run values. The
small value of the price elasticity reflects the fact that few possibilities of substitution
exist for maize imports from the United States. Other sources of imports of maize into
Mexico are nonviable as long as Mexico enjoys trade preferences because of the NAFTA
230
agreement and lower transportation costs because of the proximity to the United States.
The coefficients of the income variable are all positive and significant suggesting a direct
relationship between income and quantity of maize imported into Mexico from the Unites
States. The estimated values of income elasticity range from 0.08 to 0.23 for all four
maize models.
Table 7.3 Error Correction Model Results: M exicos Import Demand for Maize
V o la tility V I
V a ria b les
E stim a te d
C o efficien t
T -R a tio
V o la tility V 2
V a ria b les
E stim a ted
C o efficie n t
T -R a tio
1.774
0.087
0.086
1.986
3.222
QCt-6
MXY ,.9
0.122
0.234
0.221
- 0.222
0.142
2.222
0.090
- 2.809
1.902
E R t-3
V 2m
DINV tl2
-0.147
0.181
- 2.269
3.081
PCOM
PM X m
CONSTANT
0.909
-4.970
2.725
2.054
2.138
4.327
ECT n
NAFTA
M3
M9
-0.018
0.245
0.493
0.331
QCW
M X Y ,.,
E R t3
V I t-t
DINV n2
PCOM
PM X n
ECT n
NAFTA
-0.155
-0.134
-0.156
-0.170
-0.014
M3
M9
0.379
0.379
0.319
CONSTANT
0.859
8.699
2.499
-3.261
2.210
-2.280
2.621
-5.029
2.613
1.944
2.113
4.234
R2
R2 Adjusted
0.307
0.276
R2
R2 Adjusted
0.327
0.290
F-Value
5.261
F-Value
5.785
DW -H Statistic
-1.292
DW -H Statistic
- 1.269
Skewness
-0.645
0.000
Skewness
- 0.741
0.000
Kurtosis
0.576
3.000
Kurtosis
1.543
3.000
J-B Normal
7.870
9.210**
J-B Normal
8.314
9.210**
Instability Test:
Instability Test
Variance
0.232
0.353
Variance
0.214
0.353
Joint
3.785
2.960
Joint
3.854
3.690
27.17
35.172*
31.106
35.172*
L.M Statistic
L.M Statistic

231
Avalos-Sartorio (1998) reported that M exicos apparent utilization of maize has

risen sharply because demand for meat and meat products have increased in recent years.
Since the domestic production of maize has been remarkably stable, as income rises in
Mexico, the import of maize will continue to grow. The increasing share of maize
demand has been fulfilled by imports, which have risen significantly after the North
American Free Trade Agreement (NAFTA) in 1994.
Regarding the exchange rate variable, all four models yield coefficient with an
expected positive sign and all are significant. This result is consistent with the hypothesis
that a depreciation of the Unites States currency relative to the Mexican peso makes
imports into Mexico less expensive resulting in an increase of maize imports. The
estimated values of the coefficients range from 0.09 for the maize model using V2 to 0.25
for the maize model using VN.
All four maize models yield coefficients with an expected negative sign for the
volatility of the exchange rate variable. This suggests that in the short-run volatility has a
negative influence on Mexican imports of maize from the United States. In all four
models the coefficient of volatility is significant at the five percent level. Thus, riskaverse importers are discouraged by higher volatility episodes of the Mexico-Unites
States exchange rate. As a consequence, exchange rate uncertainty decreases the volumes
of maize imports into Mexico.
The effect of exchange rate volatility of M exicos imports of maize seems to be
important since the estimated values range from 0.06 for model using VG to 0.39 for
model using volatility VN. Even though in the short-run international grain traders faced
with exchange rate uncertainty could employ different tools to reduce uncertainty, such
232
as freight future contracts and commodity future contracts among others. The results
suggest that they have not been effective in Mexico-Unites States maize trade during the
study period.
Table 7.4 Error Correction Model Results: M exicos Import Demand for Maize
Volatility VG
Variables
Estimated
Coefficient
T-Ratio
QC,6
0.081
1.702
QC,-6
MXY ,.9
E R t_3
0.125
0.244
8.376
2.646
MXY ,.9
0.133
8.764
2.895
V G t.i
D IN V tl2
-0.057
0.144
-0.391
-1.980
DINV.12
0.137
1.957
PCOm
PMX n
-0.138
0.153
-3.326
1.993
-2.202
ER t-3
VN,.!
0.259
-0.152
-2.248
ECTu
NAFTA
M3
-0.018
PCOm
PM X,.!
ECT,.i
2.628
-5.002
2.556
Volatility VN
Variables
Estimated
Coefficient
0.070
2.464
-5.338
NAFTA
-0.019
1.808
M3
0.438
1.909
M9
CONSTANT
0.308
0.956
1.999
4.491
M9
CONSTANT
0.883
R2
R2 Adjusted
0.319
0.284
R2
R2 Adjusted
0.322
0.294
F-Value
5.010
F-Value
4.795
DW -H Statistic
- 1.932
4.258
2.092
0.144
1.161
0.427
0.308
1.835
1.988
T-Ratio
2.825
DW -H Statistic
- 1.482
Skewness
- 0.732
0.000
Skewness
-0.813
0.000
Kurtosis
1.815
3.000
Kurtosis
1.135
3.000
J-B Normal
7.891
9.210**
J-B Normal
8.861
9.210**
Instability Test:
Instability Test
Variance
0.233
0.353
Variance
0.302
0.353
Joint
4.082
3.690
Joint
3.631
3.510
29.059
35.172*
29.820
35.172*
L.M Statistic
L.M Statistic

The results suggest that NAFTA is having a significant effect on the Unites
States-Mexico maize trade. This support the findings of Zahniser and Coyle (2004). Since
the agreement took effect in 1994, M exicos imports from the Unites States have
233
increased by 240 percent compared to their average annual level in the pre-NAFTA
period. In regards to seasonality, two seasonal dummy variables, M3 and M9, are found
to be statistically significant. It suggests that imports of maize increase during the winter
season and before the harvest season since the main region devoted to hog and cattle
production in Mexico (the state of Sonora located in the border with the Unites States) is
affected by winter weather conditions.
The coefficient of lagged dependent variable is positive and significant in two of
the four cases. The values range from 0.08 to 0.12. Even though the responsiveness
appears to be small, these results illustrate the influence of previous trade on current
volumes of maize imports into Mexico. In other words, the Mexican maize importers take
into account the volumes they imported in the past to make their decisions on how much
to import in the current period. The lagged dependent variable becomes significant at six
lags for all four maize models. Thus, it implies that it takes about six months for
importers of maize to adjust to changing market conditions.
With regard to the inventory o f hogs (INV), while the VECM yields expected
positive sign for all models, the coefficients are statistically significant only in models
using V2 and VG. Previous research in Mexico does not suggest a strong link between
imports of maize and the domestic feed industry. Results from a regression that
considered the period of 1990 to 998 did not produced a strong link between imports of
whole maize and animal feed production. However, results indicated a strong link
between imports of broken maize and animal feed produced commercially, and between
imports of maize gluten and feed produced by commercial operations (Avalos-Sartorio,
1998). The free trade dummy variables are positive and significant in all four models
234
suggesting that NAFTA has a significant effect on the M exicos imports of maize from
the Unites States.
The coefficient of the Error Correction Term (ECT) is negative and significant in
all four models. These results corroborate the results of the long-run analysis and
reconfirm the presence of cointegrating vectors in the system. The small value of the
ECT coefficient in all four models suggests a slow adjustment towards the long-run
equilibrium.
7.4.1 Main Findings for Maize Models

The signs of the explanatory variables are, in general, consistent with their a priori
expectation and across different exchange rate volatility measures used for maize models.
The estimated coefficient of the hog inventory variable (INV) is positive and significant
in all VEC models. This holds also for the long-run, since all four models yield a positive
coefficient. The above results suggest that the importers of maize in Mexico react to both
short-run and long-run changes in the maize and hogs market conditions.
The error correction term is negative and significant in all maize models, re
confirming the presence of cointegrating vectors reported in chapter six. In all models,
the magnitude of estimated coefficient implies slow adjustment towards the long-run
equilibrium. While NAFTA has been very helpful for significant growth in maize imports
into Mexico, seasonality played a limited role in the trade flows of this commodity.
235
7.5 Results of the VECM: the Mexicos Import Demand for Sorghum
'y
For the sorghum models, the R ranges from 0.36 for model using V2 to 0.38 for model
using V I. All the F-values are statistically significant at the five percent level. The DW-h
statistic indicates that there is no autocorrelation in the estimated models. With regard to
the normality test, the null hypothesis of normality of the disturbances is rejected at the
five percent significant level, but it can not be rejected at the one percent level for all
sorghum models. The Lagrange Multiplier (LM) statistic indicates that no autocorrelation
is present in the models for sorghum.
The results of the four ECM models for M exicos sorghum imports from the
Unites States are reported in tables 7.5 and 7.6. The coefficient of the own price of
sorghum imports exhibits an expected negative sign and is significant at the five percent
level in all four models. These results are consistent with the long-run results presented in
chapter six. For sorghum, the long-run price elasticities are, in general, larger in value
than those in the short-run. The estimated own price elasticity from VEC models range
from 0.10 to 0.15.
The price of the related commodity (sorghum produced in Mexico) exhibits an
expected positive sign and is statistically significant in all sorghum models. The cross
price elasticity of import demand is consistent across four models and range from 0.28
for model using volatility VG to 0.30 using volatility V I. These results are consistent
with the results obtained from the long-run analysis.
The coefficient of the income variable is positive and significant suggesting a
direct relationship between income growth and the quantity of sorghum imported into
Mexico from the Unites States. The estimated value of income elasticity implies that if
236
there is an increase in Mexicos personal disposable income by 10 percent, the imports of

sorghum from the Unites States will increase buy just 0.6 percent.
Table 7.5 Error Correction Model Results: M exicos Import Demand for Sorghum
Volatility VI
Variables
QS t-12
MXY t_2
E R .6
VI
,.3
D I N V .3
Estimated
Coefficient
0.130
1.798
Volatility V2
Variables
Q S t-1 2
Estimated
Coefficient
0.134
T-Ratio
1.778
0.065
0.144
5.547
1.984
MXY ,.2
0.064
ER,-6
0.153
5.377
2.032
-0.181
0.063
-3.553
2.192
-4.084
V 2 ,.3
-0.091
-2.664
0.069
-0.107
0.292
2.165
-2.682
-0.126
0.306
PS ,.7
PSD ,.io
ECT,.!
NAFTA
M4
M7
M il
T-Ratio
5.698
d i n v
,3
PS ,.7
PSD n o
ECT,.,
NAFTA
M4
M7
MU
CONSTANT
4.255
-4.325
-0.011
0.137
-0.288
-0.168
-4.403
0.4228
CONSTANT
-0.259
0.450
-1.601
3.387
R2
R2 Adjusted
0.385
0.351
R2
R2 Adjusted
0.367
0.332
F-Value
8.496
F-Value
7.912
DW -H Statistic
1.233
DW -H Statistic
- 1.045
Skewness
-0.298
0.000
Skewness
-0.322
0.000
Kurtosis
1.904
3.000
Kurtosis
1.834
3.000
J-B Normal
8.997
9.210**
J-B Normal
9.098
9.210**
-1.987
-1.784
Instability Test:
-0.011
0.101
-0.300
-0.196
-0.266
0.442
0.306
-2.039
-2.003
-1.830
3.267
Instability Test
Variance
1.676
0.748
Variance
1.993
0.748
Joint
4.474
3.690
Joint
4.175
4.070
28.740
35.172*
26.900
35.172*
L.M Statistic
L.M Statistic

Regarding the exchange rate variable, all four models yield the expected positive
sign of the coefficients and they are significant at five percent level. This relationship
between exchange rate and import demand for sorghum is consistent with the long-run
237
results obtained in chapter six. The estimated value ranges from 0.12 for model using YG
to 0.15 for sorghum model using V2 and VN.
All four sorghum models yield an expected negative sign for the volatility of the
exchange rate variable, suggesting that in the short-run volatility has a significant
negative influence on M exicos imports of sorghum from the Unites States. Thus, riskaverse Mexican importers are discouraged by high volatility episodes of the MexicoUnites States exchange rate. As a consequence, exchange rate uncertainty decreases the
volumes of sorghum imports into Mexico. The estimated values, however, are small and
range from 0.091 to 0.243.
The coefficients of the lagged dependent variable are all positive and significant
at five percent level. The estimated values range from 0.12 to 0.13. Notwithstanding
small values of the estimates, these results illustrate that the Mexican sorghum importers
take into account the volumes they imported in past periods to make their decisions on
how much to import in the current period. The lagged dependent variable becomes
significant at twelve lags which implies that importers of sorghum takes about twelve
months to adjust to changing market conditions.
Regarding the hog inventory variable (INV), the VECM yields coefficients with
the expected sign and all are statistically significant. The significant coefficient is
reflective of the increasing domestic demand for sorghum in Mexico. Avalos-Sartorio
(1998) found a statistically significant relationship between imports of sorghum and the
domestic production of animal feed. At the present time Mexico is the major sorghum
importer in the world. Mexican imports accounted for about 30 percent of total world
imports of sorghum between 1992 and 2000. However, the countrys domestic
238
production of sorghum has been augmented and thus, its ability to supply the domestic
market is improving.
Table 7.6 Error Correction Model Results: M exicos Import Demand of Sorghum
Volatility VG
Variables
0.123
0.061
Q t-1 2
MXY ,.2
0.119
ER t-6
VG.,3
i n v
Estimated
Coefficient
-0.103
0.068
.,3
-0.154
T-Ratio
T-Ratio
0.128
1.772
5.171
M XY ,.2
0.062
2.189
-5.180
2.221
ER,.6
VN,.3
5.289
2.058
ECT,.!
0.288
-0.012
NAFTA
M4
0.039
-0.297
0.119
-1.977
M7
M il
-0.156
-0.290
-1.629
-1.856
CONSTANT
0.435
3.171
R2
R2 Adjusted
P S D t.jo
Estimated
Coefficient
Q t-1 2
1.748
-2.715
4.278
-4.627
P S , 7
Volatility VN
Variables
i n v
0.150
,.3
-0.243
0.102
-2.664
0.111
-2.988
4.342
P S ,.7
P SD no
E C T ,.,
NAFTA
0.291
-0.012
0.053
3.227
-4.379
0.157
-1.978
-1.864
M4
M7
-0.265
M il
CONSTANT
-0.292
0.420
0.365
0.338
R2
R2 Adjusted
0.362
0.336
F-Value
7.809
F-Value
7.740
DW -H Statistic
-1.790
DW -H Statistic
-1.126
Skewness
-0.451
0.000
Skewness
-0.410
0.000
Kurtosis
1.820
3.000
Kurtosis
1.790
3.000
J-B Normal
8.713
9.210**
J-B Normal
29.897
9.210**
-0.176
-1.866
3.046
Instability Test
Instability Test:
Variance
1.826
0.748
Variance
1.897
0.748
Joint
4.441
3.690
Joint
4.052
3.540
23.750
35.172*
29.917
35.172*
L.M Statistic
L.M Statistic

The Error Correction Terms (ECT) are negative and significant in all models.
Similar to tomato and maize models, the small value of the ECT coefficients in sorghum
models suggests a slow adjustment towards the long-run equilibrium.
239
The free trade dummy variables are positive but not significant in all four
sorghum models, suggesting that the free trade has not been a key driver of M exicos
imports of sorghum from the Unites States. It may be due to the fact that even before the
NAFTA took place, Mexico was already importing significant quantities of sorghum
required for domestic consumption from the United States and the trade was not hindered
by any policy restrictions. In regards to seasonality, sorghum models show that some
seasonal dummy variables are statistically significant in all four models (M4 for all
models and M7 for model using V2). This seasonal pattern may reflect the fact that
importers build stocks for the winter season in the northern states of Mexico where most
of the livestock production is concentrated.
7.5.1 Main Findings for Sorghum Models

The signs of the explanatory variables are consistent with their theoretical expectations.
The estimated values of various short-run elasticities are fairly close to each other across
different volatility specifications. For all sorghum models, import demand functions are
inelastic. Overall, the short-run elasticities are smaller than the corresponding long-run
elasticities. The error correction term is negative and significant in all sorghum models,
reconfirming the presence of cointegration reported in chapter six. In all models, the
small coefficient implies slow adjustment toward the long-run equilibrium.
7.6 Results of the VECM: the M exicos Import Demand for Milk Powder
The VECM for Milk powder in this study yield R values ranging from 0.31 to 0.33. All
F-statistics are significant and the DW-h statistic indicates that there is no autocorrelation
240
in any of the milk powder models. The null hypothesis of normality of the residuals is
rejected at the five percent significance level, but, it cannot be rejected at one percent
level for all four milk powder models. So, the estimated import demand functions for
milk powder seem satisfactory.
The results of the four VEC models for milk powder imports into Mexico from
the Unites States are reported in tables 7.7 and 7.8. The coefficient of the own price of
milk powder imports into Mexico exhibits a negative sign as expected in all four models
and they are significant. The own price elasticity ranges from 0.46 for milk powder
model using V I to 0.61 using volatility VN.
The coefficient of the related price (price of milk powder imported from Canada
into Mexico) exhibits an expected positive sign in all four models and all the coefficients
are significant at five percent level. The cross price elasticities range from 0.66 for milk
powder model using volatility V I to 0.88 using volatility VG. Thus, milk powder
imported from Canada into Mexico is considered a substitute for milk powder imported
from the Unites States.
The coefficients of the income variable are positive and significant in all models
suggesting a direct relationship between personal disposable income in Mexico and
quantity of milk powder imported from the Unites States. The income elasticity of import
demand range from 0.06 for model using V2 to 0.28 for model using VN. Similar to the
previous commodity models, all short-run elasticities are smaller than the corresponding
long-run elasticities.
241
Table 7.7 Error Correction Model Results: Milk Powder Import Demand
Volatility V I
Variables
Estimated
Coefficient
Estimated
Coefficient
T-Ratio
Q M t i2
0 .2 0 0
2.694
M XY ,.3
0.063
2.326
0.484
3.372
-0.133
-2.343
0.443
-0.530
2.659
-2.234
Volatility V2
Variables
T-Ratio
QM t-12
0.181
MXY t3
E R t.7
0 .1 0 1
0.789
4.682
VI . 5
M SFM
PM Km
-0.253
0.430
-1.905
2.479
E R t-7
V 2
M SF t.4
-0.458
-1.999
p m k ,.4
PM C ,.2
ECT t.i
0.661
2.180
0.752
2.141
-0.109
0.744
-4.687
0.983
PM C t 2
ECT t.i
-0.077
-3.923
NAFTA
1.431
1.721
3.271
NAFTA
M3
CONSTANT
2.300
3.421
0.354
1.250
CONSTANT
1.288
2.773
M3
2.508
0.412
R2
0.318
R2
0.314
R 2 Adjusted
0.273
R 2 Adjusted
0.283
F-Value
5.918
F-Value
4.957
DW -H Statistic
- 1.458
DW -H Statistic
-1.231
1.523
Skewness
0.095
0.000
Skewness
0.036
0.000
Kurtosis
1.502
3.000
Kurtosis
1.141
3.000
J-B Normal
8.335
9.210**
J-B Normal
6.612
9.210**
Instability Test
Instability Test:
Variance
Joint
L.M Statistic
0.7147
0.470
Variance
0.689
0.353
2.431
2.890
Joint
2.853
2.890
20.373
35.172*
17.567
35.172*
L.M Statistic

Regarding the exchange rate variable, all models for milk powder yield the
expected positive sign of the coefficients consistent with the results of the theoretical
model for a risk-averse importer. Thus, a depreciation of the Mexican currency relative to
the Unites States dollar makes imports into Mexico more expensive resulting in a
decrease of milk powder imports into Mexico from the Unites States. The estimated
coefficients range from 0.48 for model using V2 to 0.78 for model using V I. The above
242
short-run results support the empirical observations that imports into Mexico from the
Unites States decreased during the Mexican peso devaluation periods.
Table7.8 Error Correction Model Results: Milk Powder Import Demand

Volatility VG
Variables
T-Ratio
Q M .12
MXY ,.3
0.171
0.097
2.046
3.220
QM t-12
M XY t.3
E R ,.7
0.766
2.783
E R ..7
V G ts
M SF t.i
P M K .2
-0.316
-3.368
VN
0.450
2.624
-0.548
0.876
-2.565
2.192
M S F ,.4
p m k ,.2
ECT t-i
NAFTA
CONSTANT
-0.091
1.163
-3.990
1.471
3.144
R2
0.332
R2
0.320
R 2 Adjusted
0.291
R 2 Adjusted
0.290
F-Value
4.910
F-Value
5.737
DW -H Statistic
-0.690
DW -H Statistic
-0.799
Skewness
0.073
0.000
Skewness
0.226
0.000
Kurtosis
1.007
3.000
Kurtosis
1.095
3.000
J-B Normal
5.443
9.210**
J-B Normal
7.766
9.210**
0.456
0.470
1.815
2.890
27.104
35.172*
PM C ,.2
2.008
Instability Test:
PM C ,2
e c t ,.,
NAFTA
CONSTANT
T-Ratio
0.229
0.283
3.737
4.706
0.730
-0.903
2.719
-3.103
0.467
2.859
-2.654
-0.608
0.848
-0.092
0.781
1.288
2.268
-3.948
1.024
2.639
Instability Test
Variance
0.689
0.470
Joint
2.693
2.890
19.950
35.172*
L.M Statistic
Volatility VN
Variables
Estimated
Coefficient
Estimated
Coefficient
Variance
Joint
L.M Statistic

The coefficient of the volatility of the exchange rate variable yields an expected
negative sign for all milk powder models. In all cases, the coefficients of the volatility
variable are significant at five percent level. These results are consistent with the
hypothesis that a risk-averse importer would reduce the volume of imports if faced with
243
uncertainty in relative currency values. The estimated coefficients range from 0.13 for
model with V2 to 0.90 for model with VN.
The coefficient of the lagged dependent variable is positive and significant in all
four models. The estimated values range from 0.17 to 0.22. The lagged dependent
variable becomes significant at twelve lags. So, it is reflective of the time it takes for
importers of milk powder to adjust to changing market conditions.
It is worthwhile to note that the milk short-fall (MSF) variable has a positive and
significant coefficient in all milk powder models. The elasticity of the MSF with respect
to import demand ranges from 0.43 to 0.47. In regard to the Error Correction Term
(ECT), results show that all four models yield negative and significant coefficients.
Similar to previous models, the small value of the ECT coefficient in all four models
suggests a slow adjustment towards the long-run equilibrium.
While the free trade dummy variable has a positive coefficient in all models, none
is statistically significant at the five percent level. This result suggests that free trade had
no significant impact on the M exicos milk powder imports from the Unites States. All
seasonal dummy variables yield statistically insignificant coefficients suggesting that
there is no pronounced seasonal pattern in the quantity of milk powder imports from the
Unites States into Mexico.
7.6.1 Main Findings for Milk Powder Models

The signs all the explanatory variables are consistent with their theoretical expectation
and they are in line with the long-run results obtained in chapter six. The values of shortrun elasticities are fairly close to each other, with the exception of exchange rate
244
volatility. The error correction term is negative and significant in all milk powder models,
reconfirming the presence of cointegrating vectors found in chapter six. In all models, the
small value of the ECT coefficient implies a slow adjustment towards the long-run
equilibrium. A possible explanation for the low adjustment could be the presence of
market interventions in this sector. This hypothesis is plausible for the case of Mexico
since the government maintained price control and other regulation until the 1990s for
milk powder products. While many of the restrictions have been removed, some residual
effects may still persist. The free trade dummy variable appears to be positive and not
significant in all milk powder models confirming that NAFTA did not have a positive
impact on milk powder imports from the Unites States into Mexico. Furthermore, none of
the monthly dummy variables is significant. Finally, the sign of the MSF variable is
positive as expected which reflects the increasing dependency of Mexico for milk powder
from the international market to satisfy growing domestic demand for milk and other
dairy products.
7.7 Hypothesis Testing

To determine if exchange rate and the volatility of exchange rate have significant effects
on agri-food trade flows between Mexico and the United States in a short-run context,
formal hypotheses tests are performed on results obtained from various error correction
models. In particular three different hypothesis are tested: (a) changes in exchange rate
and its variability have insignificant effect on trade flows (H0: fier = 0,
H0: fiv = 0), (b)
the combined effect of exchange rate and its volatility on trade flows is insignificant (Flo:
Per = Pv
= 0). Finally, the appropriateness of using exchange rate as a separate variable in
245
the trade model is tested using H0: /fer = pPrice- The results of these tests are discussed
below.
To perform this, the null hypotheses: H0: fiER = 0 and H0: J3vi = 0 are used where fiER =
estimated coefficient of the exchange rate and f i Vi = estimated coefficient of the volatility
variable and a standard t-test is employed. The null hypothesis is rejected for all sixteen
commodity-specific models at the five percent level of significance. Similarly, the null
hypothesis of the insignificant effect of the volatility of the exchange rate on the MexicoUnited States agri-food trade flows is rejected for fifteen of the sixteen commodityspecific models at the five percent level (Table 7.9). Thus, both exchange rate and the
volatility of the exchange rate have significant influence on the trade flows of the selected
commodities between Mexico and The United States in the short-run.
Table 7.9 Short-Run Test Results on Significance of Individual Coefficients on

H 0:
Critical
Value*
Models
Per
= 0,
pv= o
t-statistic
t-statistic
Critical
Value*
Per
Pv
Models
Per
Pv
Tomato V I
1.96
2.92
1.983
Sorghum V I
1.96
1.984
3.553
Tomato V2
1.96
2.48
3.80
Sorghum V2
1.96
2.032
2.664
Tomato VG
1.96
3.27
2.67
Sorghum VG
1.96
2.189
5.180
Tomato VN
1.96
2.35
2.12
Sorghum VN
1.96
2.058
2.664
M aize V I
1.96
2.222
2.809
M ilk Powder V I
1.96
4.682
1.905
2.343
M aize V2
1.96
2.499
3.261
M ilk Powder V2
1.96
3.372
M aize VG
1.96
2.646
3.326
M ilk Powder VG
1.96
2.783
3.368
M aize VN
1.96
2.895
1.980
M ilk Powder VN
1.96
2.719
3.103
* t Critical Values at 5 Percent level of significance with 180 degrees of freedom.
246

A joint test on the regression coefficients of the exchange rate and volatility was also
conducted to test the null hypothesis that exchange rate and volatility of the exchange rate
jointly have an insignificant effect on the Mexico-United States trade flows (i.e., the test
is that the regression coefficients of exchange rate and volatility are jointly equal zero).
To perform this test, an F-test with (q, n- k) degrees of freedom is used. The null
hypothesis used in this case is: H0: J3ER =
0 . The null hypothesis is rejected for all
sixteen commodity-specific models at the five percent level of significance (Table 7.10).
The result implies that the combined effect of exchange rate and its volatility on the
Mexico-United States trade flows of the selected commodities is also statistically
significant.
Table 7.10 Short-Run Test Results on the joint Significance of the Exchange
Rate and Volatility of the Exchange Rate on Mexico-US Trade Flows
H 0: Per = fiv = 0
Critical Value*
F-Statistic
Sorghum V I
3.00
7.652
6.872
Sorghum V2
3.00
5.161
3 .00
6.965
Sorghum VG
3.00
47.436
Tomato VN
3.00
8.717
Sorghum VN
3.00
6.307
M aize V I
3.00
4.704
M ilk Powder V I
3.00
10.986
Maize V2
3.00
8.076
Milk Powder V2
3.00
7.841
M aize VG
3.00
8.802
M ilk Powder VG
3.00
9.216
M aize VN
3.00
5.969
M ilk Powder VN
3.00
7.940
M odel
Critical Value*
F-Statistic
Tomato V I
3 .00
9.571
Tomato V2
3 .00
Tomato VG
Model
* F-Critical Values at 5 Percent level of significance with two and 180 degrees o f freedom.
247
7.7.3 Test on the Equality of the Coefficients of Foreign Prices and Exchange Rate
In this study, the exchange rate has been incorporated directly into the trade model. To
determine if this was appropriate, linear restriction on the (3 parameter is formulated to
verify that the coefficient of foreign price is in fact equal to the coefficient of the
exchange rate.
The null hypothesis of the equality of foreign price and exchange rate coefficients
is rejected for all sixteen commodity-specific models at the five percent level of
significance (Table 7.11). The results imply that when estimating the Mexico-US import
demand functions, it is appropriate to include exchange rate directly into the econometric
model. Therefore, the approach followed in this study to incorporate exchange rate as a
separate variable into the trade model is appropriate.
Table 7.11 Short-Run Test Results on Equality of Coefficients of Exchange Rate and
Foreign Price on Mexico-US Import Demand
B-restrictions: H 0: p 3j = Psj f o r i j = 1,2,3,.
Model
Critical Value*
F-Statistic
Tomato V I
Tomato V2
3.84
3.84
12.950
6.240
Tomato VG
Critical Value*
F-Statistic
Sorghum V I
Sorghum V2
3.84
3.84
8.241
6.670
Model
3 .8 4
6.220
Sorghum VG
3.84
7.774
Tomato VN
3.84
5.055
Sorghum VN
3.84
7.213
M aize V I
M aize V2
3.84
3.84
7.198
9.084
Milk Powder V I
Milk Powder V2
3.84
3.84
17.660
12.615
M aize VG
3.84
9.540
Milk Powder VG
3.84
12.473
M aize VN
3.84
10.941
Milk Powder VN
3.84
12.104
* F-Critical Values at 5 Percent level of significance with one and 180 degrees of freedom.
248
7.8 Comparison of Short-run Results across Commodities

To perform a comparison of the main results obtained from the short-run import demand
functions, an attempt is made first to informally determine which volatility measure
provides more consistent results than others. To this end, a summary of the results from
various VEC models is presented in Table 7.12.
The short-run results summarized in table 7.12 suggest that the coefficients of all
variables have expected signs in all commodity models. Only in model for maize with
volatility V2, the INV variable has a negative sign. While the magnitudes of the
estimated coefficients vary for each commodity across four different volatility measures,
they are all less than one in absolute value. There is no known benchmark that can be
used to discriminate one set of results from the alternatives. Under these circumstances,
the results from the models using volatility VG are selected based on the consideration
that the volatility measure generated from the GARCH approach is more appealing
theoretically than the alternatives as discussed in chapter four. Despite the apparent
arbitrariness, this choice would also make the comparison of the short-run results across
commodities quite consistent with the comparison of long-run results.
The comparison of the results across models using VG reveals some insights
about the specific institutional features underlying each commodity. The coefficients of
income and own price variables in the United States import demand for tomatoes have
small values, as expected, since it is assumed that consumers in the United Stated have
high disposable income and only a small fraction of income is spent on this commodity.
The coefficient of the exchange rate is higher than the coefficient of the volatility of the
exchange rate variable for tomato models. This fact reflects that this is a highly traded
249
goods and so, traders care more about exchange rate changes than about its volatility. In
the case of the ratio of wage rates, the small value of the coefficient could be an
indication of declining importance wages to total production costs and thus, has a small
effect on tomato trade flows.
Comparing results of M exicos import demand functions for maize, sorghum and
milk powder, the results suggest that the own price variable has higher impact on milk
powder imports than on imports of maize and sorghum from the United States into
Mexico. It could be partly due to the fact that milk powder is a higher value-added good
relative to maize and sorghum and consumer devote a higher income share on milk and
milk products than that on the other two goods. In fact, sorghum models have the smaller
values of the coefficients of own price and income variables. While maize and sorghum
are two of the most subsidized commodities in Mexico, the dairy industry has been
subject to less government subsidy which may also have been reflected in higher values
of the estimated coefficients for milk powder. In fact, Maize has received the highest
levels of support on a per-ton basis than any other commodities in the pre NAFTA era
(World Bank, 2005). The coefficients of exchange rate and volatility are higher for milk
powder model than those for maize and sorghum reflecting the increasing tendency of
dairy imports into Mexico. The value of the elasticity of the exchange rate using YG is in
line with the value of 0.599 reported by Tanyeri-Abur and Rosson (1997). Finally, the
NAFTA has influenced maize and milk powder imports from the United States into
Mexico more significantly than the imports of sorghum. This could be due to the fact that
the levels of sorghum imports were increasing even before the NAFTA agreement as a
results of high growth rates in the hog and poultry industries in Mexico.
250
Table 7.12 Results of the VECM for Four Commodities using Different
Specifications of the Volatility of the Exchange Rate
M odels for Tomatoes
Variable
Lagged Volume
US Income
Exchange Rate
Volatility VI
Own Price
Related Price
W age Rate
ECT
NAFTA
Variable
Lagged Volume
MX Income
Exchange Rate
Volatility VI
Inventory
Own Price
Related Price
ECT
NAFTA
Expected Sign
VI
V2
(+)
(+)
(+)
(-)
(-)
(+)
(+)
(-)
(+)
0.331
0.156
0.802
-0.064
-0.117
0.457
0.295
-0.048
0.495
0.178
0.206
0.706
0.029
-0.152
0.536
0.127
-0.072
0.475
Expected Sign
(+)
(+)
(+)
(-)
(+)
(-)
(+)
(-)
(+)
M odels for Maize

VI
V2
0.122
0.087
0.086
0.234
0.090
-0.155
-0.134
-0.156
0.170
-1.144
0.379
0.221
- 0.222
0.142
-0.147
0.181
-0.018
0.245
VG
VN
0.198
0.663
-0.043
-0.099
0.195
0.130
-0.030
0.427
0.169
0.098
0.499
-0.108
-0.092
0.416
0.343
-0.041
0.523
VG
VN
0.818
0.125
0.244
-0.057
0.144
-0.138
0.153
-0.018
1.161
0.070
0.133
0.259
-0.391
0.137
-0.152
0.144
-0.019
1.808
0.110
M odels for Sorghum
Variable
Lagged Volume
M X Income
Exchange Rate
Volatility VI
Inventory
Own Price
Related Price
ECT
NAFTA
Variable
Lagged Volume
M X Income
Exchange Rate
Volatility VI
Milk Short-fall
Own Price
Related Price
ECT
NAFTA
Expected Sign
VI
V2
VG
VN
(+)
0.130
0.065
0.144
-0.181
0.063
-0.126
0.306
-0.011
0.137
0.134
0.064
0.153
-0.091
0.069
-0.107
0.292
-0.011
0.101
0.123
0.061
0.119
-0.103
0.068
-0.154
0.288
-0.012
0.039
0.128
0.062
0.150
-0.243
0.102
-0.111
0.291
-0.012
0.053
VG
VN
0.171
0.097
0.766
-0.316
0.450
-0.548
0.876
-0.091
1.163
0.229
0.283
0.730
-0.903
0.467
-0.608
0.848
-0.092
0.781
(+)
(+)
(-)
(+)
(-)
(+)
(-)
(+)
Expected Sign
(+)
(+)
(+)
(-)
(+)
(-)
(+)
(-)
(+)
Models for Milk Powder

VI
V2
0.181
0.101
0.789
-0.253
0.430
-0.458
0.661
-0.109
0.744
0.200
0.063
0.484
-0.133
0.443
-0.530
0.752
-0.077
1.431
251
7.9 Summary
The main results obtained from the Vector Error Correction models for the selected
commodities are presented in this chapter. The results indicate that all import demand
functions yield coefficients with theoretically expected signs. In general, short-run
elasticities are smaller than the corresponding long-run elasticities and so the results are
consistent with the Le Chatelier principle. Considering that each regressor in the ECM is
specified in the first-differenced form, the empirical results suggest that the statistical fit
of all import demand functions to data is satisfactory.
Although many factors affect agricultural trade, exchange rates and its volatility
play an important role in the agri-food trade between Mexico and the United States.
While the magnitude of the impact of exchange rates on trade varies by commodity, in
most cases, the responsiveness of agricultural imports to prices is inelastic. The above is
also true for the effect of exchange rate volatility on trade flows. Except for one case, the
exchange rate volatility has a significant negative effect on the trade flows of the selected
commodities. The free trade agreement seems to have played a significant enhancing role
in the trade flows of tomato and maize, but not for sorghum and milk powder.
All error correction terms are negative and significant in all models. As pointed
out by Harris and Sollis (2003), this result supports the validity of an equilibrium
relationship among the variables in each cointegrating equation. This implies that
overlooking the cointegration among the variables would have introduced misspecification in the underlying dynamic structure for each of the four commodities
considered in this study.
252
The change in real imports per period (month in this study) that is attributed to
disequilibrium between the actual and equilibrium levels is measured by the absolute
values of the ECT in each equation. There is a considerable inter-commodity variation in
the adjustment speed to the last periods disequilibrium. In other words, the adjustment of
import volume to changes in its explanatory variables may take about 10-12 months for
tomato and milk powder while up to 28-30 months for maize and sorghum.
The ECM results in this study show that exchange rate volatility has a significant
short-run effect on import demand in addition to its long-run effects. To reinforce the
above finding, the null hypothesis that that exchange rate and volatility have an
insignificant effect on the Mexico-US trade flows was tested formally. The null
hypothesis was rejected using a t-test as well as using the F-test. Thus, ignoring such
variables in the analysis of Mexico-United States trade equations could produce biased
results. It can be argued that traditional import demand studies for commodities that do
not include a variable representing the effects of exchange rate uncertainty are potentially
mis-specified. Finally, a comparison between models for each commodity was made to
informally determine which volatility specification provides more consistent results than
others. It is concluded that models using VG have produced more consistent coefficients
both in terms of signs and magnitudes.
253
CHAPTER 8
CONCLUDING REMARKS
8.1 Introduction
The objective of this chapter is to summarize the major findings of this study, highlights
the key contributions, and identifies certain limitations of this study. It also suggests
some avenues for future research.
Since the initiation of economic and trade liberalization in the mid 1980s in
Mexican agricultural sector, both exports and imports have grown dramatically in terms
of volume and value. It has been argued in policy and academic circles that exchange rate
played an important role in the performance of Mexican agricultural sector during last
two decades. While some studies of the Unites States-Mexico agricultural trade have
emphasized the importance of NAFTA in the dramatic increase in exports and imports
(Rosenzweig, 1996; Schwentesius, et. al., 2001), others have emphasized the role played
by exchange rate changes in enhancing Mexican trade performance in the agri-food
sector (Espinoza-Arellano 1998; Malaga et al 2001; Mora-Flores 2002). Since the shift
from the fixed to a floating exchange rate system in 1995, Mexico-Unites States
exchange rate has been characterized by periods of unexpected calm followed by
episodes of high volatility, which may also has influenced agri-food trade flows between
the two countries.
While the border between Mexico and its important trade partners, the Unites
States and Canada, became increasingly open due to NAFTA, Mexican agri-food trade
has also been influenced by changes in exchange rate and exchange rate volatility. It is
yet to be determined the extent to which the NAFTA, the continuous devaluation of the
254
Mexican peso, and changes in the exchange rate volatility have contributed to the growth
in Mexican agri-food trade during the last decade. Since the Unites States is the most
influential trade partner of Mexico, it is important to investigate the extent to which
exchange rate changes, changes in the volatility of the Unites States-Mexico exchange
rate and the NAFTA have contributed to the expansion of the Mexico-Unites States trade
in agri-food commodities.
While the available literature on the effects of Mexico-Unites States exchange
rate on agricultural trade provide evidence that changes in exchange rate do have an
important effect on trade, no research has been conducted yet to investigate the effects of
exchange rate volatility on Mexican agricultural trade. Empirical studies on the effects of
exchange rate movements on Mexico-Unites States agri-food trade conducted in the past
have ignored the volatility issues and focused only on short-run effects of exchange rate
on trade. Moreover, they concentrated on aggregate trade rather than on trade in specific
commodities. The previous studies have also ignored the time series properties of the data
used in econometric estimation. In view of these limitations in the existing studies and
the fact that they provide inconclusive evidences of the effects of exchange rate
movements on trade flows, efforts are made in this research to undertake a more
comprehensive study of the effects of exchange rate changes and its variability on
Mexico-Unites States agri-food trade flows. Four important traded commodities: tomato,
maize, sorghum, and milk powder, are analyzed for the period from 1989 to 2004.
The economic research problem addressed in this study deals with the nature and
extent of the influence of exchange rate changes and exchange rate volatility on Mexican
agri-food trade with the Unites States during the last decade. The purpose of this study is
255
to quantify the effects of the Mexico-Unites States exchange rate changes and its
volatility on Mexico-Unites States agri-food trade flows using the Maximum Likelihood
Cointegration analysis. This research attempted to accomplish the following specific
objectives:
To develop an analytical framework capable of analyzing the effects of changes in

exchange rate and its volatility on the Mexico-Unites States agri-food trade
paying due attention to commodity specific institutional features.
To determine the long-run and short-run effects of changes in the exchange rate
and its volatility on Mexico-United States agri-food trade.
To test if real exchange rate volatility has an adverse effect on the quantity of the
Mexico-Unites States agri-food trade flows using a set of alternative volatility
measures.
To discuss policy implications of the effects of exchange rate and its volatility on
Mexican agri-food trade flows.
To accomplish the above objectives, the following set of hypothesis stated in null form
was tested in this study:
The changes in Mexico-Unites States exchange rate have insignificant effects on

the Mexico-Unites States agri-food trade flows in the long run.
The changes in Mexico-Unites States exchange rate have insignificant effects on

the Mexico-Unites States agri-food trade flows in the short-run.
Exchange rate volatility has an insignificant effect on the quantity of the MexicoUS agri-food trade flows in the short-run.
256
The coefficient of exchange rate and foreign price is equal (the appropriateness of
using exchange rate as a separate variable in the trade model is tested).
8.2 Modeling Strategy

A critical assessment of the literature revealed that exchange rate and its volatility are two
important factors influencing international trade flows of agri-food commodities.
However, controversies still surround the estimated effects of these variables on trade
flows of individual agri-food commodities. An expected utility maximization model that
incorporates own prices, income, related prices, exchange rate and its volatility has been
developed in this study to shed light on these controversies.
The specification of the economic model was followed by the derivation of a set
of comparative statics to determine the direction of the effects of the key explanatory
variables on import demand. This was followed by developing the empirical specification
of import demand functions for the United States imports of tomato from Mexico as well
as for Mexicos imports of maize, sorghum and milk powder from the United States.
Given the nature of the traded commodities considered, the analytical framework
treats import as an intermediate goods that enter the production process, together with
domestic inputs to produce the final goods.
Uncertainty of exchange rate enters the
model through both exporters revenue and expenditure functions.

The comparative statics results show that while changes in exchange rate have a
positive effect on the import demand function, uncertainty influences the import demand
functions negatively. Other results from the comparative static are as follows: while the
demand for imports responds negatively to changes in output price and the exchange rate,
257
the own price of imports responds positively to changes in exchange rate (i.e. the output
price respond positively given a devaluation of the Mexico-United States change rate).
The empirical model derived from the theoretical framework has been used to
determine the effects of exchange rate changes, its volatility as well as the effects of
income and prices on the trade volumes of specific commodities.
While both the structural approach and the reduced form approach have been used
in the literature to determine the effects of exchange rate on trade flows of agri-food
commodities, a reduced form model has been used in this study. It allows a simpler
means of studying the economic relationship between the exchange rate movements and
the variables entering the model such as prices, quantities and other variables than the
other approaches. The main advantage of the reduced form model is that meaningful
economic hypotheses can be tested without estimation of structural parameters and it can
accommodate non-stationary data. It is widely known that when non-stationary variables
are used in regression analysis, the classical t and F-tests are not appropriate and may
lead to misleading conclusions (Hendry, 1986). The reduced form approach also offers a
convenient framework for addressing data non stationarity issue in econometric
estimation.
The econometric approach used in this research consisted of testing for unit root
in each data series. To this end, the ADF test was employed. The Johansens Maximum
likelihood technique has been used to test for cointegration. This was followed by the
estimation of Vector Error Correction models and hypothesis testing.
While it is generally recognized in the trade literature that exchange rate volatility
induces additional uncertainty in trade, no consensus exists on how to measure it. A
258
number of statistical measures of variability of the exchange rate have been used in the
literature. Economic theory does not provide clear guidance on the features of the most
appropriate volatility measure (Clark, et al 2004). In an attempt to shed light on the
effects of different volatility measures on trade flows, four different volatility measures
were constructed and employed in this study, including a volatility measure based on a
GARCH model, a non-parametric exchange rate volatility measure, and two variance or
standard deviation-based volatility measures (moving average of the standard deviation
of the exchange rate and the average absolute difference between the previous forward
and the current spot rate).
8.3 Major Findings

There is a consensus in the literature that the economic and trade liberalization
undertaken by the Mexican government during the 1980s and 1990s affected positively
the growth of the Mexican economy and the Mexico-Unites States trade flows in agri
food commodities. A critical review of the published literature on the effects of changes
in the exchange rate and its volatility on agricultural trade revealed that most of the
previous studies have ignored the time series properties of data. Thus, these studies may
be subject to spurious relationship bias (Granger, 1981; and Engel and Granger 1987).
Since different data sets have been used by different studies, the results of the effects of
exchange rate on trade are not comparable among studies. Finally, most previous studies
have concentrated on aggregate trade flows that essentially mask the effects of exchange
rate on a particular sector and little attention has been focused on trade flows of specific
commodities.
259
The results from the estimation of four different volatility measures show that the
standard deviation based measures of volatility (V I and V2) closely track exchange rate
fluctuations. With respect to the GARCH model, it was found that GARCH (1,1)
describes well the exchange rate data based on significance of the parameters and related
statistics. Finally, a non-parametric regression was performed using the kernel regression
approach with a normal kernel function to generate the non-parametric volatility measure.
These four volatility measures were incorporated in the estimation of the import demand
function for each of the four commodities considered in this study.
With regard to the unit root test, the ADF test was conducted to determine the
integration properties of individual series used in each commodity equation. Additionally,
for the exchange rate variable, the Perron test was used to determine if there was a
structural break in these series. The results indicated that no structural break took place in
the sample period despite the outlier.
To determine the lag length truncation for each series, the Akaike (1973)
Information Criterion (AIC) and Schwarz (1978) Criterion (SBC) were applied. The
results of the ADF test show that all economic variables included in each of the four
commodity models are non-stationary and they have a unit root in its univariate
representation.
For the Unites States import demand for tomato, cointegration results suggest that
the Unites States personal disposable income and exchange rate have positive impacts on
trade flows in all models using different volatility specifications. These results imply that,
in the long run, devaluation of the Mexican peso relative to the Unites States dollar works
as an incentive for importers to increase the quantity of tomato imported into the Unites
260
States from Mexico. On the other hand, the volatility of the exchange rate produced
mixed results. W hile three tomato models indicated a negative relationship between
volatility and trade flows, one of them suggested a positive relationship.
With regard to the price of tomato, an expected negative relationship is suggested
between own price and trade volume in all models. For the related price, all tomato
models suggest a positive relationship between price of tomato in the Unites States and
Unites States import demand function, but one model yields a coefficient that is not
statistically significant. Thus, the tomato produced in the Unites States is considered a
substitute of the Unites States imports of tomato from Mexico. There is a positive
relationship between farm wage rate ratio and the Unites States import demand. The
increasing wage differential, about 8-9 times, between the Unites States and M exicos
rural wage rates for the analyzed period appear to have enhanced tomato imports from
Mexico.
For M exicos import demand for maize, both the income and exchange rate
variable have significant positive influence on trade flows. The own price of maize and
the price of the related commodity have the expected sign in all models. Thus, M exicos
domestic maize can be considered as a substitute for maize imports. The above result
makes sense since domestic maize is used to complement the requirements for maize
used in feeding animals in Mexico. The inventory variable has a positive sign in all four
maize models. It is interesting to note that exchange rate volatility has a negative impact
on trade volumes in fifteen out of sixteen models. This is consistent with the behavior of
risk-averse importers.
261
An important finding of this research is that the specification of the exchange rate
volatility does matter. Not only different elasticity values are obtained for different
exchange rate volatility measures used, but also, for some models, the magnitudes of
some variables differ across models. While the magnitude of the impact of exchange rates
on trade varies by commodity, in most cases, the responsiveness of agricultural exports to
changes in exchange rate is less than one. In regards to the speed of adjustment, the
magnitude of the estimated weights suggest that in the event of a disturbance to the long
run equilibrium, import volume, own price, and volatility of the exchange rate would
adjust faster than any other variable to restore the equilibrium condition.
The overall results from the ECM indicate that almost all the import demand
models yield theoretically expected signs and magnitudes of relevant variables.
Furthermore, most short-run elasticities are smaller than the corresponding long-run
elasticities. Thus, the Le Chatelier principle holds in almost all of the models.
Considering that each regressor in the ECM is specified in first-differenced form, the
empirical results suggest that the statistical fit of most of the models to the data are fairly
good as indicated by the R , F-statistic, and the set of diagnostic tests employed.
While the magnitude of the impact of exchange rate on trade flows varies by
commodity, in most cases, the responsiveness of agricultural trade flows is less than one
in the short-run. The above is also true for the coefficient of the volatility variable. The
sign of the volatility variable is negative and consistent across all commodities. In
general, the short-run income elasticity of import demand is inelastic, as expected for all
commodities.
262
The free trade dummy variable has a positive impact on trade flows of tomato and
maize, but this is not the case for sorghum and milk powder where the free trade dummy
yields positive but insignificant coefficients. Finally, seasonality influences sorghum and
maize trade flows but not the trade flows of tomatoes and milk powder.
The error correction term is negative and significant in all models. This result
reconfirms the existence of long-run equilibrium relationships among the variables in
each cointegrating equation. It implies that overlooking the cointegration of the variables
would have introduced misspecification bias in the results.
The change in real imports per period (month in this study) that is attributed to
disequilibrium between the actual and equilibrium levels is measured by the absolute
values of the ECT in each equation. There is considerable inter-commodity variation in
the adjustment speed to the last periods disequilibrium. While the adjustment of import
volume to changes in various explanatory variables takes about 10 months for tomato and
20 months for milk powder, it takes about 30 months for maize and sorghum.
Finally, the ECM results in this study show that exchange rate volatility has a
significant effect on import demand, both in the short-run and in the long-run. Thus,
ignoring such a variable could produce biased results. Thus, the traditional import
demand studies that do not include a variable representing the effects of exchange rate
uncertainty are potentially mis-specified.
The results from hypothesis tests indicate that the exchange rate and volatility of
the exchange rate do have a statistically significant effect on the Mexico-United States
trade flows of tomato, maize, sorghum and milk powder. So excluding these variables
from the trade models could lead to model mis-specification and biased results. The
263
results also suggest that for the commodities considered in this study it is appropriate to
include the exchange rate variable directly into the trade model.
Comparing results of the four models across four alternative specifications, it was
found that, for all import demand functions, models using VG provide more consistent
results with theory than those from other models. For the four commodity models using
VG, all the variables have the a priory expected signs and magnitudes. Furthermore, a
comparison of the estimated results for models using VG shows than institutional features
related to each commodity are useful in explaining the magnitudes of trade flows
between Mexico and the United States.
8.4 Policy Implications

The nature of the analysis and the commodities involved makes this study important from
a policy perspective. First, this research provides, for first time, a set of information
regarding how exchange rate changes and changes in the volatility of exchange rate affect
the Mexico-US trade flows of tomato, maize, sorghum, and milk powder.
The positive influence of the peso depreciation on the United States tomato
imports suggest that Mexican exporters have the opportunity to make profits by
allocating different volume through out the year when Mexican peso devaluates with
respect to the United States currency. The above could be possible if Mexican tomato
exporters have the ability to respond quickly to changing market conditions by supplying
seasonal, perishable products year-round using tomato produced in different regions of
Mexico.
264
Secondly, the results from cointegration analysis suggest that trade volumes
between Mexico and the US are sensitive to both short-run and long-run changes to the
bilateral exchange rate. They are also sensitive to exchange rate uncertainty. This means
that assisting the traders to access forward markets could be helpful to mitigate exchange
rate uncertainty, which will have a beneficial effect on trade flows and induce better
allocation of resources in Mexico.

This research makes both disciplinary as well as policy contributions. First, for each of
the four commodities, institutional features related to specific industry were incorporated
in the trade model. In doing so, this research extends Appelbaum and Kohlis (1997)
model with commodity specific information incorporated into the profit function of
traders. A set of comparative statics results is derived from the theoretical model to
determine the effects of exchange rate and exchange rate volatility on trade flows. This
represents the analytical contributions of this research.
This research has used, for the first time, the Johansen Maximum Likelihood
Cointegration analysis to generate long-run effects of exchange rate changes and its
volatility on Mexican agricultural trade flows of selected commodities. Similarly, an
Error Correction Model has been used to generate the short-run effects of these variables
on trade flows. The above constitute an empirical contribution of this research.
Since no consensus exists in the trade literature on the appropriate specification of
exchange rate volatility measure to be included in a trade model, attempts were made to
identify an appropriate measure from a set of four alternatives. It also represents an
265
important contribution to our existing stock of knowledge. This research generated

valuable information about the extent to which exchange rate movements have
contributed to Mexican trade performance in the agri-food sector. Hopefully the results of
this study would assist various stakeholders in Mexico, especially traders of agri-food
products, to better assess exchange rate risk and to take advantage of the opportunities
presented by exchange rate movements. Finally, the exchange rate and exchange rate
volatility not only have long-run effects on trade flows, but also have significant shortrun effects. This empirical finding implies that if import demand functions are estimated
excluding this variable, it could potentially introduce miss-specification biased in the
estimated results.
8.6 Limitations and Suggestion for Future Research

While this research makes valuable analytical, empirical and policy contributions as
indicated above, it does possess few limitations. These are highlighted below.
One of the most common practices in the international trade research is to
estimate trade equations assuming a linear relationship between import volume and a set
of explanatory variables. However, there is always the possibility that a linear
specification may be not fully satisfactory. While it was beyond the scope of this research
to evaluate this issue, it would be useful to explore non linearity issue in the future.
Second, this research makes use of the Vector Error Correction Model to
determine the short-run relationships between the volume of trade and a set of
explanatory variables. Even though the results from the VEC models are consistent with
economic theory, a second econometric approach could be employed to contrast the
266
short-run results of this study. The dynamic response of international trade flows to
exchange rate and its volatility shocks could be evaluated by employing impulseresponse and variance decomposition analysis. Future research along this line will
enhance our understanding of the short-run dynamic effects of changes in exchange rate
and its volatility on agricultural trade flows between Mexico and the United States.
267
References
Akaike, H., (1973) Information Theory and an Extension of the Maximum Likelihood
Principle Proc. Second International Simposium o f Information Theory. 267-281.
Akhtar, M.A and Hilton, S.R., (1984) Effects of Exchange Rate Uncertainty on German
and U.S. Trade FRBNY Quarterly Review!Spring 7-16.
Alston, J.M., Carter, C.A,.Green R, and Pick. D., (1990) Whither Armington Trade
Models? Amer. J. Agr. Econ. 72:455-67.
Anderson, M., and Garcia, P., (1989). Exchange Rate Uncertainty and the Demand for
U.S Soybeans, American Agricultural Economics Associations, August 1989. 721-729.
Anderson, J., E., (1979) A Theoretical Foundation for the Gravity Equation American
Economic Review American Economic Association, vol. 69(1), pages 106-16.
Appelbaum, E., and Kohli, U., (1997) Import Price Uncertainty and Distribution of
Income Review o f Economics and Statistics 79, 620-630.
Appelbaum, E., (1993) Empirical Application of Duality UNDER Uncertainty Paper
presented at The Annual Meetings o f the European Economic Association, Helsinky.
Appelbaum, E., and Kohli, U., (1979) Canada-United States Trade: Test for the SmallOpen Economy Hypothesis Canada Journal Economics 12: 1- 4.
Arellano et al., (1998) Analysis of Forces Affecting Competitive Position of Mexico in
Suplf US Winter Melons International Food and Agribusiness Management Review. 1(4)
495-507.
Arize A. and Steven S. S., (1998) Cointegration, Real Exchange Rate and Modelling the
Demand for Broad Money in Japan Applied Economics, Vol. 25 (June), pp. 717-26.
Armington, P., (1969) A theory of Demand for Products Distinguished by Place of
Production IM F Staff Papers, 16, 159-178.
Arrow, K., and Kurtz, M., (1970) "Optimal Growth with Irreversible Investment in a
Ramsey Model" Econometrica.
Asseery, A., and Peel, D.A., (1991) The Effects of Exchange Rate Volatility on
Exports. Economics Letters 37, 173-177.
Avalos-Sartorio, B., (1998) Structural Developments in the Mexican Grain-Livestock
Subsector: The Grain and Oilseed Subsector. In Loyns, et al.(editors). Economic
Harmonization in the Canadian/U.S./Mexican Grain Subsector. Proceedings of the
268
Fourth Agricultural and Food Policy Systems Information Workshop. Texas A&M
University, University of Guelph, pp. 161-186.
Bailey, M.J., Tavlas, G.S., and Ulan M., (1987) The Impact of Exchange - Rate
Volatility on Export Growth. Some Theoretical Considerations and Empiricals Results
in. Journal o f Policy Modeling vol 9, no 1 22: 225-243.
Bailey, M., Tavlas, S.G., and Ulam, M., (1986) Exchange-Rate Variability and Trade
Performance: Evidence for the Big Seven Industrial Countries," Weltwirtschaftllches
Archiv 122 :466-77.
Baltagi, B.H
& Li, Q., (2001) "Estimation O f Econometric Models With
Nonparametrically Specified Risk Terms," Econometric Reviews, Taylor and Francis
Journals, vol. 20(4), 445-460.
Bank of Mexico (1996) La Economia Mexicana Informe Anual. Mexico.
Bamerjee, A., Doblado J.J., Hendry, D.F., and Smith, G.W., (1986) Exploring
Equilibrium Relationships in Econometric Theory Through Static Models: Some Monte
Carlo Eevidence , Oxford Bulletin o f Economics and Statistics, 48: 253-77.
Batra, R.N. and Ullah, A. (1974). Competitive Firm and the Theory of Input Demand
Under Price Uncertainty The Journal o f Political Economy Vol. 82, No 3, 537-548.
Batten, D.S. and Belongia, M.T. (1986) Monetary Policy, Real Exchange Rates and U.S.
Agricultural Exports. American Journal o f Agricultural Economics 68: 422-427.
Bazdresch, S. And Werner, A., (2005) Regime, Switching Models for the Mexican
Peso Journal o f International Economics, Volume 65, Issue 1 pp 185-201.
Bazdresch, S. (2002) El Cmportamiento del Tipo de Cambio en Mexico y el Regimen de
Libre Flotacion: 1996-2001 Documento de Investigacion No. 2 2002-09, Banco de
Mexico.
Belanger, D., Gutierrez, S., Racette, D., and Raynauld, J., (1992) The Impact of
Exchange Rate Variability on Trade Flows: Further Results on Sectoral U.S. Imports
from Canada. North American Journal o f Economics and Finance 3: 888-892.
Bergstran J.H., (1989) The Generalizad Gravity Equation, Monopolistic Competition,
and the Factor Proportion Theory of International Trade The Review o f Economics and
Statistics, Vol. 71, Issue 1, pages 143-53.
Bergstran J.H., (1985) The Gravity Equation in International Trade: Some
Microeconomic Foundations and Empirical Evidnce The Review o f Economics and
Statistics, 1985, 67, (3), 474-81.
269
Bessler, D. A., and Kling, J.L., (1986) "Forecasting Vector Autoregressions with
Bayesian Priors." American Journal o f Agricultural Economics 68(1): 144-51.
Bini-Smaghi, L., (1991) Exchange Rate Variability and Trade: Why is it so Difficult to
Find any Empirical Relationship? Applied Economics 23, 927-936.
Bollerslev, T., Chou, R. S., and Kroner, K. F., (1992) "ARCH Modelling in Finance a
Review of the Theory and Empirical Evidence" Journal o f Econometrics, 52, 5-59.
Bollerslev, T. (1986) "Generalised Autoregressive Conditional Heteroskedasticity" J.
Econometrics, 51, 307-327.
Bossaerts, P., (1988) Common Non-Stationary Components of Asset Prices Journal
o f Economic Dynamics and Control, 12, 347-364.
Bradshaw, G., and Orden, D., (1990) Granger Causality from the Exchange Rate to
Agricultural Prices and Export Sales. Western Journal o f Agricultural Economics. 15:1
(July), pp. 100-110.
Brodsky, D.A., (1984) Fixed Versus Flexible Exchange Rates and the Measurement of
Exchange Rate Instability Journal o f International Economics 16, pp. 295-306
Brooks C., (2002) Introductory Econometrics to Finance Cambridge University Press
Broil, U., (1994) Foreign Production and Forward Markets Australian Economic
Papers. 62 PP 1-6.
Byerlee, D., and Sain, G., (1991) Relative Food Prices Under Structural Adjustment:
Preliminary Findings from Latin America . Food Policy 16(1): 74-84.
Caballero, J. R., and Corbo, V., (1989) The Effect of Real Exchange Rate Uncertainty
on Exports: Empirical Evidence The World Bank Economic Review, Vol 3 2:263-280
Calva J.L., (1996) La Economia Nacional y la Agricultura de Mexico, A Tres Anos de
O peration
del TLCAN,
NAFTA
and
Agriculture:
Is
the
Experiment
Working Symposium Proceedings.
Carone, G. (1996) "Modeling the U.S. Demand for Imports Through Cointegration and
Error Correction" Journal o f Policy Modeling, Elsevier, vol. 18(1), pages 1-48.
Carter, C.A., Gray, R.S., Furtan, W.H., (1990) Exchange Rate Effects on Inputs and
Outputs in Canadian Agriculture American Journal o f Agricultural Economics. Vol.
72(3).
Chambers, R.G., and Just, R.E., (1981) Effects of Exchange Rate Changes on U.S.
agriculture: a Dynamic Analysis American Journal Agricultural Economics, 61: 249257.
270
Chambers, R.G., and Just, R.E., (1979) A critique of Exchange Rate Treatment in
Agricultural Trade Models. American Journal Agricultural Economics, February 63:
32-46.
Cho, G., Sheldon, M. I., and McCorriston S., (2002) Exchange Rate Uncertainty and
Agricultural Trade American Journal of. Agricultural Economics,. 84 (4) November:
931-942.
Chowdhurry, A.R., (1993) Does Exchange Rate Volatility Depress Trade Flows?
Evidence from Error-Correction Models The Review o f Economics and Statistics, 75:
700-705.
Clark, P.B., Tamarisa, N., and Shang-Jin, W., with Sadikow, A., and Zeng L., (2004)
Exchange Rate Volatility and Trade Flows-Some New Evidence International
Monetary fund, May, 2004.
Clark, P. B., (1973) Uncertainty, Exchange Risk, and the Level of International Trade,
Western Economic Journal 11, September, pp. 302-13.
Clements, M. P., and Hendry, D. F (2002) An Overview of Economic Forecasting in.
(eds.)A Companion to Economic Forecasting, Blackwells. 1- 18.
Coleman, J.R. and Meilke, D.K., (1988) The Influence of Exchange Rates on Red Meat
Trade Between Canada and United States Canadian Journal o f Agricultural Economics
401-424.
Cook, R, (2001) "The U.S. Fresh Produce Industry: An Industry in Transition," Chapter 2
in Postharvest Technology of Horticultural Crops, Adel A. Kader (eds.), University o f
California Division o f Agriculture and Natural Resources, Publication 3311, pp.5- 26,
2001 .
Collins, J.K., Meyers, H.W., and Bredahl, E.M., (1980)Multiple Exchange Rate Change
and U.S. Agricultural Commodity Prices Amer. J. Agr. Econ. 62: 656-665.
Cushman, D.O., (1988) U.S. Bilateral Trade Flows and Exchange Risk During the
Floating Period Journal o f International Economics 24,317-330.
Cushman, D.O., (1986) Has Exchange Risk Depressed International Trade? The Impact
of Third-Country Exchange Risk Journal o f International Money and Finance 5, 361379.
Cushman, D.O., (1983) The Effects of Real Exchange Rate Risk on International Trade
Journal o f International Economics 15, 45-63.
271
Deardorff, A
(1998) Determinants of Bilateral Trade: Does Gravity Work in a
Neoclassical W orld?, in J.A. Frankel, ed. by The Regionalization o f the World Economy,
(Chicago: University of Chicago Press), pp. 7-22.
De Grauwe, P., and De Bellefroid., (1988) Long Run Exchange Rate Variability and
International Trade, Chapter 8 in S. Arndt and J.D. Richardson, ed. by Real Financial
Linkages Among Open Economies, (London: The MIT Press).
De Grauwe, P., (1988) Exchange Rate Variability and the Slowdown in Growth of
International Trade IMF Staff Paper. 35: 63-84.
DeJong, D.N., and Whiteman C.H., (1989) Trends and Random Walks in MacroEconomic Time Series: A Reconsideration Based on the Likelihood Principle Working
paper No. 1989- 4, University of Iowa.
DeJong, D.N., Nankervis J.C., Savin, N. E., and Whiteman, C.H., (1992) The Power
Problem of Unit Root Tests for Time Series with Autoregressive Errors Journal o f
Econometrics, 53: 323-43.
Dellas, H., and Ben-Zion Z., (1993) Real Exchange Rate Volatility and International
Trade: A Reexamination of the Theory. Southern Economic Journal 59 (April): 641-7.
Demers, M., (1991) Investment Under Uncertainty and the Arrival of Information Over
Time The review o f Economic Studies. Vol 58(2) 333-350
Dhrymes, P. J., (1998) Time Series, Unit Roots, and Cointegration . Academic Press,
San Diego California.
Diaz-Garces. D., (2002) Analysis de las Funciones de Importacion y Exportacion de
Mexico, 1980-2000 Documento de Investigacion, Direccion General de Investigacion
Economica. Banco de Mexico.
Dickey, D. A., and Fuller W.A., (1981) Likelihood Ratio Statistics for Autoregressive
Time Series with a Unit Root. Econometrica 49, 1057-1072.
Dickey, D. A., and Fuller, W.A., (1979) Distribution of the Estimators for
Autoregressive Time Series with a Unit Root Journal o f American Statistical
Association, 74: 427-31.
Dickey, D. A., and Fuller W.A., (1976) Distribution of First Order Autoregressive
Estimator" Proceedings o f the Business and Economic Statistocs Section,American
Statistical Association, 278-281.
Dorfman, J. H., and Lastrapes, D.W., (1996) The Dynamic Responses of Crop and
Livestock Prices to Money-Supply Shocks: A Bayesian Analysis Using Long-Run
Identifying Restrictions. American Journal o f Agricultural Economics. 78:3. pp. 530-41.
272
Dombusch, R., (1976) Expectations and Exchange Rate Dynamics Journal o f Political
Economy, Vol. 84, pp. 1161-76.
Elliot, G., Rothenberg T J , and Stock, J.H., (1996) Efficient Tests for an Autoregressive
Unit Root Econometrica, 64: 813-836.
Enders, W., (2003) Applied Econometric Time Series. Publisher: John Wiley and Sons
(WIE), 2003.
Enders, W., ( 1995) Applied Econometric Time Series New York. John Wiley.
Engle, R (2001) "GARCH 101: The Use of ARCH/GARCH Models in Applied
Econometrics," Journal o f Economic Perspectives, American Economic Association, vol.
15(4), pages 157-168.
Engle, R., (2004) "Risk and Volatility: Econometric Models and Financial Practice,"
American Economic Review, American Economic Association, vol. 94(3), pages 405-420.
Engle, R.F. and Granger, C.W.J., (1987) Cointegration and ntegration and Error
Correction: Representation, Estimation and Testing Econometrica, 55: 251-276.
Engle, R.F., and Granger, C.W.J., (1987) Co-integration and Error Correction:
Representation, Estimation and Testing Econometrica 55, 251-276
Engle, R.F. and Yoo B.S. (1987) Forecasting and Testing in Co-Integrated Systems,
Journal o f Econometrics, 35: 143-59.
Engle, R. F., (1982). "Autoregressive Conditional Heteroscedasticity, with Estimates of
the Variance of United Kingdom Inflation" Econometrica, 50, 987-1007.
Ericsson, R. N., (2001) "Forecast Uncertainty in Economic Modeling," International
Finance Discussion Papers 697, Board of Governors of the Federal Reserve System
(U.S.).
Espinoza-Arellano, J.J., Fuller S., and Malaga J., (1998) Analysis of Forces Affecting
Competitiveness of Mexico in Supplying U.S. Winter Melon Market International Food
and Agribusiness Management Review1, 4: 495-507a.
Ethier, W., (1973) International Trade and Forward Exchange Market The American
Economic Review, Vol 63, No.3, 494-503.
Fabiosa, F.J., (2002) Assesing the Impact of the Exchange Rate and its Volatility on
Canadian Pork and Live Swine Exports to the United States and Japan Working Paper
02-WP 355 Center fo r Agricultural and Rural Development, june 2002.
Fendel R., (2002) Taylor-Type Rules and the Term Structure of Interest Rates mimeo.
273
Fleming, J. M., (1962) Domestic Financial Policies Under Fixed and Under Floating
Exchange Rates, IM F Staff Papers 9, 3: 369-79.
Fox, J., (2000) Nonparametric Simple Regression: Smoothing Scatterplots , Thousand
Oaks, CA, Sage Publications.
Fox, J., (2000) Multiple and Generalized Nonparametric Regression. Thousand Oaks,
CA, Sage Publications.
Franke. G., (1991) Exchange Rate Volatility and International Trading Strategy Journal
o f International Money and Finance 10, 292-307.
Frankel, J.A., (1991) The Making of Exchangea Rate Policy in the 1980s Economics
Working Papers 91-157 University of California at Berkeley.
Frenkel J.A., and Mussa, M.L., (1985) Asset Markets, Exchange Rate and the Balance of
Payments. Chapter 14. Handbook o f International Economics. Vol 3. Ansterdam Norht
Holland.
Frankel, J.A., (1979) On the Mark: A Theory of Floating Exchange Rates Based on Real
Interest Differentials" American Economic Review Vol. 69, pp. 610-22.
Frenkel J.A, (1976) A Monetary Approach to the Exchange Rate: Doctrinal Aspect and
Empirical Evidence Scandinavian Journal o f Economics. 200-24.
Friedman, D., and Vandersteel, S., (1982) Short-Run Fluctuations in Foreign Exchange
Rates Journal o f International Economics, 13:171-186.
Fuller, S., Gutierrez, N., and Capps, O., Jr (1992) International Dry Onion Trade:
Factors Affecting Import Demands for U.S, Dry Onions Agribusiness. 8:445-455.
Gagnon, J.E. (1993) Exchange Rate Variability and the Level of International Trade
Journal o f International Economics 34, 269-287.
Gallant, A. R., and Tauchen, G., (1989) Semi Non-Parametric Estimation of
Conditionally Constrained Heterogeneous Processes.Asset pricing Applications
Econometrica 57:1091-1120.
Gonzalo, J., (1994) Five Alternative Methods of Estimating Long-Run Equilibrium
Relationships Journal o f Econometrics 60, 203-233.
Gotur, P., (1985) "Effects of Exchange Rate Volatility on Trade; Some Further
Evidence," IMF S taff Papers 32 /475-512.
Gourieroux, C., (1997) ARCH Models and Financial Applications. New York: Springer
Verlag.
274
Granger, C.W.J., (1983) Co-integrated Variables and Error-Correcting Models

Economics Department Discussion Paper No. 83-13 (University of California, San Diego,
CA).
Granger, C.W.J., and Newbold, P., (1986) Forecasting Economic Time Series, Second
edition, Orlando, Florida, Academic Press.
Granger, C.W.J., (1981) Some Properties of Time Series Data and their Use in
Econometric Model Specification Journal o f Econometrics, 16: 121-130.
Granger, C. W. J, (1969) Investigating Causal Relations by Econometric Models and
Cross Spectral Methods Econometrica 37, 424-438.
Grennes, T. (1975) The Exchange Rate and USA Agriculture: Comment American
Journal o f Agricultural Economics 57: 136-137.
Gutierrez N., (1983) Economic Study of the Winter Vegetables Export Industry of
Northwest Mexico Master Thesis A& M University, College Station, TX.
Guzel, H.A., and Kulshreshtra, S.N., (1995) Effects of Real Exchange Rate Changes on
Canadian Agriculture: A General Equilibrium Evaluation Journal o f Policy Modeling
17: 639-57.
Haffer R.W and Sheehan R.G (1991) Policy Inference Using VAR Models. Economic
Inquiry. January 1991; 29,1.
Hai, W., N., Mark, C., Yangru Wu (1997) Understanding spot and forward exchange
rate regressions Journal o f Applied Econometrics. Vol: 12, No: 6. Pg: 715-734.
Hamilton. J., (1994) Time Series Analysis Princeton University Press,
Hansen B., and Phillips P., (1990) "Estimation and Inference in Models of Cointegration:
A Simulation Study" In Advances in Econometrics, Vol. 8, 1990.
Hardle, W., and Linton, O., (1994) "Applied Nonparametric Methods" in R.F. Engle and
D.L. McFadden, eds., Handbook o f Econometrics, Vol. 4 (North-Holland), chapter 38.
Harris and Sollis (2003) Applied Time Series Modelling and Forecasting Richard
Harris and Robert Sollis, John Wiley and Sons, Chichester.
Harris, R.I., (1995) Using Cointegration Analysis in Econometric Modelling Harvester
Wheatsheaf Prentice Hall.
Hendry, D.F., (1997) The Econometrics of Macro-Economic Forecasting Economic
Journal, 107, 1330-1357.
275
Hendry, D.F. (1995) On the Interactions of Unit Roots and Exogeneity . Econometric
Reviews, 14, 383-419.
Hendry, D.F. (1987) Econometric Methodology: A personal perspective. In T.F.
Bewley (Ed.), Advances in Econometrics, Chapter 10. Cambridge: Cambridge University
Press.
Hendry, D.F., (1986) Econometric Modelling with Cointegrated Variables: An
Overview, Oxford Bulletin o f Economics and Statistics, 48, 201-12.
Hooper, P., and Kohlhagen, S.W., (1978) The Effect of Exchange Rate Uncertainty on
the Prices and Volume of International Trade Journal o f International Economics 8:
483-511.
Hubrich, K., Lutkepohl, H., and Saikkonen, P., (2001) "A Review Of Systems
Cointegration Tests" Econometric Reviews, Taylor and Francis Journals, vol. 20(3), pages
247-318.
Ian Sheldon (2003) Exchange Rate Uncertainty and Agricultural Trade Agricultural
Competitiveness and World Trade Liberalization: Implications of the WTO and
NAFTA, Fargo, ND, May 29-30, 2003.
INEGI (2004). Banco de Information Economica-Sector Extemo. Grupo de Trabajo:
SHCP, Banco de Mexico, Secret aria de Economfa e INEGI.
Jennings S., Adamowicz, W., and Constantino, L., (1990) The Canadian Lumber
Industry and the Macroeconomy: A Vector Autoregression Analysis. Canadian Journal
o f Forest Research. 21: 288-299.
Johansen, S., (1995) Likelihood Based Inference in Cointegrated Vector Autoregressive
Models Oxford University Press.
Johansen, S., (1991) Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Vector Autoregressive Models. Econometrica. 59: 1551-1580.
Johansen, S., and Juselius, K., (1990) Maximum Likelihood Estimation and Inference on
Integration with Applications to the Demand for Money Oxford Bulletin o f Economics
and Statistics. 52: 169-179.
Johansen, S., (1988) Statistical Analysis of Cointegration Vectors. Journal o f Economic
Dynamic and Control. 12: 231-254.
Kapombe, C. M., and Colyer D., (1998) Modeling U.S. Broiler Supply Response: A
Structural Time Series Approach. Submitted to the Review o f Agricultural and Resource
Economics 7:2(Oct 1998):241-251.
276
Kawai, M., (1983) "Price Volatility of Storable Commodities Under Rational

Expectations in Spot and Futures M arkets." International Economic Review.
24(1983):435-458.
Kawai, M., (1973) The Behaviour of an Open-Economy Firm Under Flexible Exchange
Rates Journal o f Political Economy. 20 (23).
Kenen, B.P., and Rodrik, D., (1986) Measuring and Analyzing the Effects of Short-Term
Volatility in Real Exchange Rates The Review o f Economics and Statistics 311-315.
Klein, M.W., (1990) Sectoral Effects of Exchange Rate Volatility on United States
Exports Journal o f International Money and Finance 9: 299-308.
Knutson, K., Meilke, and Yunez-Naude, A.,, (editors 1993) Trade LiberalizationUnder
NAFTA-Report Card on Agriculture. Proceedings o f the Sixth Agricultural and Food
Policy Systems Information Workshop. Texas A&M University, University of Guelph, El
Colegio de Mexico.
Kohli, U., (1982) "Production Theory, Technological Change, and the Demand for
Imports: Switzerland, 1948-1976 European Economic Review 18:369-86.
Konno, T. and Fukushige, M. (2003) Did NAFTA Cause the Structural Changes in
Bilateral Import Functions Between the US and Mexico? Journal o f Policy Modeling 25
(2003) 53-59.
Koo W., Karemera D, Taylor R., (1994) A Gravity Model Analysis of Meat Trade
Policies Agricultural Economics . Volume 10, Issue 1 , January 1994, Pages 81-88.
Koray. F., and Lastrapes, W. D., (1989) Real Exchange Rate Volatility and U.S.
Bilateral Trade: A Var Approach The Review o f economics and statistics 71:708-717.
Kost, W., (1976) Effects of an Exchange Rate Change on Agricultural Trade
Agricultural Economics Research. 28: 99-106.
Kouri, P., (1976) "The Exchange Rate and the Balance of Payments in the Short Run and
the Long Run: A Monetary Approach" Scandinavian Journal o f Economics Vol. 78, pp.
280-308.
Kroner, K. F., and Lastrapes, W.D., ( 1993) The Impact of Exchange Rate Volatility on
International Trade: Reduced Form Estimates using the GARCH-in mean Model Journal
o f International Money and Finance 12; 298-318.
Kroner, K.F., and Lastrapes, W.D., ( 1989) Real Exchange Rate Volatility and U.S.
Bilateral Trade. A VAR Approach . Review o f Economic & Statistics 71(4) 708-712.
Krueger, A., (1999) Are Preferential Trading Arrangements Trade Liberalising or
Protectionist? Journal o f Economic Perspectives, 13, 105-124.
277
Kwiatkowski, D., Phillips, P., Schmidt, P., and Y. Shin, Y., (1992) Testing the Null
Hypothesis of Stationary Against the Alternative of a Unit Root Journal o f Econometrics
54:159-178.
Lacroix, L.J.R., Schwentesius, R.R., Gomez, C.M.A., Echanove, H.F., (2001) Marketing
of Fruits and Vegetables The World Bank Group June 2001.
Lastrates, D.W., and Koray F., (1990) Exchange Rate Volatility and U.S. Multilateral
Trade Flows Journal o f Macroeconomics, vol 12, No 3, 341-362.
Levich, R.M., (1985) Empirical Studies of Exchange Rates: Price Behavior, Rate
Determination and Market Efficiency in: R.W. Jones and P.B. Kenen (eds.), Handbook
o f International Economics, Vol. II, Amsterdam: North Holland, 979-1040.
Liang, H., (1998), Real Exchange Rate Volatility: Does the Nominal Exchange Rate
Regime Matter? International Monetary Fund (working Paper).
Love, J., and Lucier, G., (1996) Florida-Mexico Competition in the US Market for Fresh
Vegetables Vegetables and Specialties. April. USDA-ERS.
Lutkepohl, H., (2001) Comment on Essays on Current State and Future Challenges of
Econometrics Journal o f Econometrics, Elsevier, vol. 100(1), pages 81-82.
Macdonald, R., and Taylor, M. P., (1992) Exchange Rate Economics: A Survey IM F
Staff Papers, March
Mackinnon, G.J., (1999) Critical Values for Cointegration Test in Long-Run Tests
Econometric Theory 15: 367-376
Malaga, E. J., Williams, W.G., and Fuller, W.S., (2001) US-Mexico Fresh Vegetables
Trade: the Effects of Trade Liberalization and Economic Growth Agricultural Economic
26, 45-55.
Malaga, E. J., Williams, W.G., and Fuller, W.S., (1996) Effects of NAFTA and Other
Exogenous Factors on the U.S.-Mexico Fresh Vegetables Trade: A Simulation Analysis
Symposium Nafta and Agriculture: Is the experiment working? San Antonio Texas.
Maskus, K.E., (1987) Exchange Rate Risk and US Trade: A Sectoral Analysis. Federal
Reserve Bank of Kansas City Economic Review, 3: 16-28.
McKenzie, D.M., and Melbourne, T.I.M.R., (1999) The Impact of Exchange Rate
Volatility on International Trade Flows Journal o f Economic Surveys vol. 13, 1: 21-38.
McKenzie, D.M., (1998) The Impact of Exchange Rate Volatility on Australian Trade
Flows Journal o f International Financial Markets 8, 21-38.
278
McKenzie, D.M., and Brooks D.R., (1997) The Impact of Exchange Rate Volatility on
German-US trade flows Journal o f International Financial Markets 7,73-87.
Meese, R. A., and Singleton, Kenneth J., (1983) Rational Expectations and the Volatility
of Floating Exchange Rates International Economic Review, Department of Economics,
University of Pennsylvania and Osaka University Institute of Social and Economic
Research Association, vol. 24(3), pages 721-33.
Mittelhammer, R.C., (1978) The estimation of domestic demand for salad vegetables
using a priori information. Ph.D. Dissertation. W ashington State University.
Mora-Flores,
S.
J.,
Lara-CovarrubiasD.,
Martinez-Damian,
A.M.,
(2003)
Competitiveness and Comparative Data Advantageas of Milk Production System in
Jalisco State, Mexico Agrociencia, vol. 37 Num 1 enero-febrero.
Mora-Flores, S.J., Matus-Gardea, A.J., and Martinez-Damian, A.M., (2002) Effects on
the Gross Value of Agriculture and Cattle Exports Due to Unit Change in their Prices
Agrociencia 36: 389-399.
Mundell, R., (1963) "Capital Mobility and Stabilization Policy under Fixed and Flexible
Exchange Rates," Canadian J. Econ. And Poli. Sci, Nov.. Or as adapted in his
International Economics, 18, pp. 250-262.
Munoz, M., and Odermatt, P., (1993) E1 Sistema Leche de Mexico en el Tratado
Trilateral de Libre Comercio (TLC) In: Juan Pablos (Editors) La Agricultura Mexicana
Frente al Tratado Trilateral de Libre Comercio 2nd edicion, Universidad Autonoma
Chapingo S.A, Mexico.
Mussa, M., (1979) Empirical Regularities in the Behaviour of Exchange Rates and
Theories of Foreing Exchange Market, in K Brunner and A. H. Melttzer eds. CarnegicRochester series on public policy vol 11 (North-Holland Amsterdams).
Mussa M., (1976) "Nominal Exchange Rate Regimes and the Behavior of Real Exchange
Rates: Evidence and implications," Carnegie Rochester Series on Public Policy, Vol. 25,
117-214.
Nadal, A., Wise, A.T., (2004) The Environmental Costs of Agricultural Trade
Liberalization: Mexico-U.S. Maize Trade Under NAFTA Working Group on
Development and Environment in the Americas Discussion Paper Number 4, June 2004.
Nelson, C.R., and Plosser C.I. (1982) Trends and Random Walks in Macroeconomic
Time Series: Some Evidence and Implications, Journal o f Monetary Economics, 10:
139-62.
Ng, S., Perron, P., (2001), Lag Length Selection and the Construction of Unit Root Tests
with Good Size and Power Econometrica 69, 1519-54.
279
Obstfeld, M., and Rogoff K., (1995) Exchange Rate Dynamics Redux Journal o f
Political Economy, 103(3), 624-60.
Orden D., and Fisher, L.A., (1991) Macroeconomic Policy and Agricultural Economics
Research American Journal of Agricultural Economics 73:1348-54.
Orden, D., and Fackler P., (1986) Identifying Monetary Impacts on Agricultural Prices
in VAR Models American Journal o f Agricultural Economics 71: 495-502.
Osterwald-Lenum, M., (1992) "A Note with Quantiles of the Asymptotic Distribution of
the ML Cointegration Rank Test Statistics" Oxford Bulletin o f Economics and Statistics,
54, 461-472.
Oya, K., Toda, H. Y., (1995) Dickey-Fuller Lagrange Multiplier and Combined Tests for
a Unit Root in Autoregressive Time Series Journal o f Time Series Analysis, Vol 19
3:326-347.
Pagan, A. R., and Hong, Y.S., (1991) Nonparametric Estimation and the Risk Premium
in Barnett, William A., James Powell, and George E. Tauchen eds., Nonparametric and
Semiparametric Methods in Econometrics and Statistics: Proceedings of the 5th
International Symposium in Economic Theory and Econometrics, Cambridge University
1991, 51-75.
Pagan, A.R., and Ullah , A., (1988) The Econometric Analysis of Models W hit Risk
Terms Journal o f Applied Econometric.
Peree, E. and Steinherr, A.,(1989) "Exchange Rate Uncertainty and Foreign Trade,"
European Economic Review 33 :1241-64.
Phillips, P.C.B., (1991a) Optimal Inference in Cointegrated System Econometrica, 59:
283-306.
Phillips, P.C.B., and Perron, P., (1988) Testing for Unit Roots in Time Series
Regression, Biometrica, 75 335-346.
Phillips, P.C.B., (1987) Time Series Regression with a Unit Root Econometrica, 55:
277-301.
Phillips, P.C.B., (1986) Understanding Spurious Regressions in Econometrics Journal
o f Economic, 33: 311-40.
Pick, D.H., (1990) Exchange Rate Risk and U.S. Agricultural Trade Flows. American
Journal o f Agricultural Economics 72: 694-700.
Qian, Y., and Varangis, P., (1994) Does Exchange Rate Volatility Hinder Export
Growth? Empirical Economics 19: 371-396.
280
Rosenzweig, P.A., (1996) The Mexican Agriculture in the Context of NAFTA

Agriculture in NAFTA, 1-14.
Rosson, P., Adcock F., and Hobbs A., (2001) Exchange Rate Impacts on U.S.
Agriculture and the Potential Role of Do\\a.r\mUonAmerican Farm Bureau Federation
Research Report, July 2001.
Ruffin, J.R., (1999) The Nature and Significance of Intra-Industry Trade Economic &
Financial Review, October.
Secretaria de Agricultura, Ganaderia y Desarrollo Rural (2004). Several issues.
SAGARPA, Anuario Estadistico de la Production Agricola, Mexico, DF.
Said, S. E., and Dickey, D.A., (1984) Testing for Unit Roots in Autoregressive-Moving
Average Models of Unknown Order Biometrika, 71, 599-607.
Salcedo-Baca. D., (1990) Distributional Effects of the Mexican Agricultural Trade
Policies, The Tomato Case Tesis Doctoral University of Illinois, Illinois, USA.
Sargan, J. D., (1964) "Wages and Prices in the United Kingdom: A Study in Econometric
Methodology" In Hart, P. E., Mills, G., and Whitaker, J. K.(eds.), Econometric Analysis
fo r National Economic Planning, Vol. 16 of Colston Papers, pp. 25-63.
Sarker, R., (1993) A Maximum Likelihood Cointegration Analysis of Canadian Lumber
Exports Canadian Journal o f Agricultural Economics 41: 97-110.
Schmidt, P., and Phillips C.B., (1992) LM Test for a Unit Root Journal ofBussines and
Economic Statistic 7:147-59.
Schotman, P., and Van Dijk, H.K., (1991) A Bayesian analysis of the Unit Root in Real
Exchange Rates Journal o f Econometrics 49, 195-238.
Schuh, G.E., (1974) The Exchange Rate and US Agriculture. American Journal o f
Agricultural Economics. 56: 1-13.
Schwartz, L.N., (1986) Simplicity, Inference and Modeling: Keeping it Sophisticatedly
Simple Lecture derived may 1986 at the JL Kellogg School o f Management
Northwestern University.
Schwartz, L.N., (1978) Self-Financing of an R&D Project American Economic
Review,June 1978.
Schwentesius, R. R., and Gomez, C.M., (2001) NAFTA and the Mexican AgroAlimentary Sector Comercio Exterior, vol 51 No. 6 June.
281
Schwentesius, R. R., and Gomez, C.M., (1996) El Impacto del TLC Sobre las
Exportaciones de Hortalizas Frescas de Inviemo de Mexico a EUA Symposium TriNational NAFTA and Agriculture: Is the Experiment working? San Antonio Texas.
Schwert, G.W., (1989a) Why Does Stock Market Volatility Change Over Time?
Journal o Finance 44:115-1153.
Sercu, P., and Vanhulle C., (1992) Exchange Rate Volatility, International Trade, and
the Value of Exporting Firms. Journal o f Banking and Finance 16: 155-82.
Sexton, R.., Zhang, M., and Chalfant, J., (2003) Grocery Retailer Behaviour in the
Procurement and Sale of Perishable Fresh Produce Commodities Economic Research
Service/USDA.
Sims, C.A., (1980) Macroeconomics and Reality, Econometrica, 48: 1-48
Smith, E.C., (1999) Exchange Rate Variation, Commodity Price Variation and the
Implications for International Trade Journal o f International Money and Finance 18,
471-491.
Stock, J., and Watson, M., (1988) Testing for Common Tyrends Journal o f the
American Statistical Association. 83: 1097-1107
Stock, J., And Watson, M., (1987) Asymptotic Property of Least Square Estimators of
Cointegrating Vectors Econometrica 55: 1035-1056.
Taylor, M.A., and Taylor, P.M., (2004) The Purchasing Power Parity Debate NBER
Working Papers 10607, National Bureau of Economic Research Inc.
Taylor, P.M., (1995) The Economics of Exchange Rates Journal o f Economic
Literature, 33, 13-47, 1995
Thraen, S.C., Hawang, C.T., and Larson, W.D., (1992) Linking of U.S. Monetary Policy
and Exchange Rates to World Soybean Markets Agricultural Economic, 6 (1992) 365384.
Thursby, G.J., and Thursby, C.M., (1987) Bilateral Trade Flows, the Linder Hypothesis,
and Exchange Risk The Review o f Economics and Statistics, 488-495.
Thursby, G.J., (1985) "A Test Strategy for Discriminating Between Autocorrelation and
Misspecification in Regression Analysis: Reply" The Review o f Economics and Statistics,
67(1), February 177-78.
Thursby, G.J., and Thursby, C.M., (1984) "How Reliable Are Simple, Single Equation
Specifications of Import Demand?" The Review o f Economics and Statistics, 66(1),
February 1984, 120-28.
282
Tinbergen, J., (1962) Shoping the W orld Economy Suggestions fo r an International

Economy Policy, The Twentieth Century Fund.
Trefler, D., (1995) The Case of the Missing Trade and Other Mysteries American
Economic Review, American Economic Association, vol. 85(5), pages 1029-46.
US Department of Agriculture, Economic Research Service (2004) Various issues.
Vegetables and Specialties Washington, DC.
US Department of Agriculture, Economic Research Service (2003) Various issues.
Vegetables and Specialties Washington, DC.
US Department of Agriculture, (1999) NAFTA Situation and Outlook Series Report
No. WRS-99-1. Economic Research Service, Washington, DC.
Valdes, A., (1993) World Development Report Washington, DC: US International
Trade Commission. New York: Oxford University Press.
Vellianitis-Fidas.(1976) The Impact of Devaluation on U.S. Agricultural Exports
Agricultural Economics Research 28: Jjuly 107-116.
Viaene, J.M., and C.G. de Vries, (1992) International Trade and Exchange Rate
Volatility European Economic Review 36, pp. 1311-21.
Villa-Issa, M.R., (1990) "Performance of Mexican Agriculture: The Effects of Economic
and Agricultural Policies" American Journal o f Agricultural Economics. 72: August 744748.
Wang K.L., and Barret B. C., (2002) A New Look at the Trade Volume Effects of Real
Exchange Rate Risk Working paper november 2002 WP 2002-41.
Watson, G.A. (1987) A Method for the Chebyshev Solution of an Oveardetermined
system of Complex Linear Equations University o f Dundee Reports NA/ 101.
West, K.D., (1988) Asymptotic Normality When Regressors Have a Unit Root.
Econometrica, 56:1397-1417.
Wolf, A. (1995) Import and Hedging Uncertainty in International Trade. Journal o f
Future Markets 15: 101-110.
Yang , S., Koo W., and Wilson W., (1992) Heteroskedasticity in Crop Yield Models
Journal o f Agricultural and Resource Economic vol 14 no 1.
Yoon, G., (2003) The Time Series Behaviour of Brazilian Inflation Rate: New Evidence
from Unit Root Test with Good Size and Power Applied Economics Letters, 10: 627632.
283
Yunez-Naude, A., (2002) "The Dismantling of CONASUPO, a Mexican State Trader in

Agriculture" The World Economy.
Yunez-Naude A., and Barcenais F., (2002) "Lessons from NAFTA: The Case of Mexico's
Agricultural Sector", Report to the World Bank, Mimeo, December.
Zahniser, S., and Coyle, W., (2004) U.S.- Mexico Com Trade During the NAFTA Era:
New Twists to an Old Story Report from the Economic Research Service
www.ers.usda.gov.
Zivot, E., and Phillips P.C.B., (1994) A Bayesian Analysis of Trend Determination in
Economic Time Series Econometric Review, 13: 291-336.
284
A p p e n d ix 1: Exchange Rate and Agricultural Trade: Exchange Rate Volatility was Included in the Model
Study
Anderson and
Garcia (1989)
Fabiosa (2002)
Major Findings
Remarks
- ER volatility has a negative

effect on trade flows. However
the magnitude varies across
countries depending on: a) access
to forward mk., b) the degree of
market concentration and degree
of risk tolerance by importers.
- Did not take into

account nonstationarity in data
used in the
analysis.
- Considered only
one measure of
volatility.
Exports of
Pork and
Live swine
-The level of ER has a positive

impact on pork exports while its
volatility has a negative impact.
Quarterly data
from 1957 to
1994
OECD
Countries
Imports and
Exports of
Agricultural
And industry
products
- Floating ER did not cause a

significant increase in overall real
domestic currency price variation
for most commodities.
1974-1995
10 developed
Countries
Bilateral
Trade of
Machinery
Chemicals
Agriculture
Other
Manufactures
-Negative effect
-Agricultural trade is more
susceptible to ER uncertainty
than the aggregate trade data
would suggest.
-No stationarity of
the data is taken
into account.
-Structural model
is preferred to
VAR.
- Did not take into
account nonstationary
properties o f the
data.
- It uses data from
the fixed and
floating periods.
- Did not take nonstationarity o f the
data used into
account.
- They use gravity
Model and Panel
data.
Ex. Rate
Treatment
Volatility
Measure
Sample, Data
Countries
OLS
Nominal
Exchange
Rate
Absolute %
change in
quarter to
quarter spot
ER.
Quarterly data
from
1974 to 1985
Bilateral
Trade of
Soybean
Maximum
Likelihood
Real
Exchange
Rate
GARCH
Monthly data
from 1994 to
Estimation
Method
2001
Commodities
U.S
Smith (1999)
Cho, et al
(2002)
OLS
Real
Exchange
Rate
NA
OLS
Real
Exchange
Rate
Standard
dev of the
first
differences
in log ex.
Rate
Appendix 2: Exchange Rate and Agricultural Trade: Volatility of the ER was not Included in the Trade Model
Study
Arellano, et al
(1998)
Carter, Gray
& Furtan
(1990).
Chamber &
Just (1981)
Estimation
method
3-SLS
OLS
3SLS
Ex. Rate
Treatment
Real
Exchange
Rate
Exogenous
Real ER
-Real
ER
Cho, et al.,
(2002)
OLS
Real
Exchange
Rate
Collins
(1980)
Partial
equilibrium
world MK
Model
Structural
time-series
Real
Exchange
Rate
Kapombe and
Kolyer (1999)
Kost (1976)
Algebraic
manipulation
and derivation
Real
Exchange
Rate
Volatility
Measure
None
None
Quarterly data
From
1975 to 1988
Exchange Rate
Pass-through
Inputs and
commodities
None
Quarterly data
From
1969 to 1977
Exports of
Wheat, Com
Soybeans
1974-1995
10 developed
Countries
Bilateral trade of
Machinery
Chemicals
Agriculture
manufactures
Inports and
Exports of
Grains and
oilseeds
Exports of
Broilers
Standard dev
of the first
differences in
log ex. Rate
Short term
ER volatility
None
1970-1977
Annual
37 Countries
70-95
Quarterly
U.S.
None
Exogenous
N.A
Commodities
Sample, Data
Countries
Winter season
data from
1970 to 1994
M exico- U.S.
Caribean
None
Bilateral
Trade of
Melon
Bilateral
Trade of
Agricultural
commodities
Major Findings
Remarks
- Devaluation of the Mexican peso

was significant in increasing melon
exports to U.S. in the short-run
- Yield enhancing technology was
significant in short-run and long-run
in explaining export growth.
Input prices as well as commodity
prices are affected significantly by
exchange rates.
- Complete pass-through of ER for
many inputs used in Canadian
agriculture.
- Exchange rate fluctuations had
significant real impact on exports
and domestic use o f wheat, com and
soybeans in the US.
- Did not take into account

non-stationarity in data
used in the analysis.
Negative effect was found

Agricultural trade is more
susceptible to ER uncertainty than
the aggregate trade data would
suggest
- Exchange rate effects on real U.S.
commodity prices are smallest under
free trade and real commodity price
insulation policies.
Broiler export market is very
sensitive to changes in RER and
trade distortion policies.
- Export market is more price
sensitive than importing market.
- Exchange Rate changes are
expected to have a small impact on
agricultural trade.
- Does not consider nonstationary of the data.

- ER elasticity is not
restricted to lie into (0, 1)
- Did not take nonstationarity of the data
used into account
- They use gravity Model
and Panel data
- Non-stationarity of the
data used was taken into
account
- Did not use the

theoretical model for
empirical estimation.
Ex. Rate
Treatment
Nominal
Exchange
Rate
Time series
Econometric
Methods
Real
Exchange
Rate
N.A
Coleman and
Meilke (1988)
OLS
Nominal
Exchange
Rate
None
Coleman
(1984)
3SLS
Jennings, et al
(1991)
Babula et. al
(1995)
N>
oo
Volatility
Measure
Estimation
Method
Reduced
Form VAR
Study
None
Sample, Data
Countries
Quarterly data
From 1968 to
1987
Canada - U.S.
1978-1989
Montly
US-Row
Quarterly data
From
1972 to 1980
Canada - U.S.
None
-J
Guzel &
Kulshreshtha
(1995)
GAMS
Kaabia and
Gil, (2000)
Cointegration
SVAR Model
Malaga, et.al.
(2001).
3SLS
Nominal
Exchange
Rate
1972-1982
Commodities
Remarks
Lumber
- It was found that exchange rate

changes has a negligible effect on
level of production and exports
- Non-stationary of the
data used was is taken into
account.
Com
- No cointegration was found

between ER, prices, sales and
shipments.
- Only short run connections
between ER and prices was detected
He argued that to took into

getting the ER righ for
agriculture are likely to be
continuosly frustrated
Read Meat
Beef and Pork
The devaluation of the CND has

lead to a significant increase in net
exports of beef and a small increase
in net exports o f pork.
Read Meat (beef

and pork)
- Exchange rate was not a significant

factor in the determination of
production, consumption and trade
of red meats.
OLS is used without

accounting for non
stationary of the data.
- it uses data from both
fixed and floating periods
non-stationary of the data
used.
- It uses data from both
Grains livestock,
dairy and
poultry.
Non-Agric
Products
Services
- Higher prices led to an increase in

overall agricultural output and
exports of all sectors.
- Changes in ER rate and
macroeconomic policy contributed
to agricultural price and income
instability in Canada.
Agricultural variables do not
significantly affect macro variables
in long-term agricultural terms of
trade will deteriorate.
Devaluation was by far the most
important factor behind the SR
increase in Mexican imports but in
LR differentials in growth rates in
prod. Yields explain fut growth.
Real
Exchange
Rate
None
Real
Exchange
Rate
None
Quarterly data
From 1978 to
1995
Agricultural
Output
Real
Exchange
Rate
1974-1993
Mexico-U.S.
Winter season
Fresh vegetables
None
1984
No sample
period
Major Findings
- The volatility of
exchange rate is not
considered.
Non-stationary of the data

is taken into account
The model accounts for

tariffs, ER, Labour costs,
yields, income and
population
to
00
00
Sarker(1993)
Johansens
cointegration
analysis
Nominal
Exchange
Rate
Thraen et al,
(1991)
Structural
Vector
Autoregressio
n (SVR)
Nominal
Exchange
Rate
None.
None
U.S.-Canada
Lumber
Annual data
from 1965 to
1985. Brazil
Argentina
Japan, EC-12
Soymeal
Soy oil
Soybean
- Results suggested that there is only

one stable equilibrium relationship
among variables in the LR.
- ER has a significant positive effect
on lumber exports.
- A week dollar increase exports of
soybean and soymeal significantly
which serves to increase the
equilibrium world price.
- The econometric
procedure used concerns a
small system that omits
other theoretical relevant
variables
- Linkages between
monetary policy ER and
world soybean market are
determined
Appendix 3 Empirical Literature on Exchange Rate Effects on Trade (Exchange Rate Volatility was Included)
Volatility
Measure
Short Term
Volatility
Moving sample
st. dev. of the
growth rates of
RER
Short Term
Moving st. dev
of the growth
rate of RER
Estimation
Method
Vector
Autoregres
sion (VAR)
Ex. Rate
Treatment
Real
Exchange
Rate
Koray and
Lastrapes
(1989)
Vector
autoregresi
on (VAR)
Nominal
Exchange
Rate
Kroner and
Lastrapes
(1989
Cointegrati
on
Nominal
Exchange
Rate
GARCH
Short Term
Volatility
Lastrapes and
Koray (1990)
Vector
autoregresi
on (VAR)
Real
Exchange
Rate
Short Term
Moving st. dev.
Of RER
OLS
Real
Exchange
Rate
Long Term
ARCH to
generate ER
volatility
Study
Chowdhury
(1993)
McKenzie
(1998)
Qian and
Varangis
(1994)
SUR
ARCH-M
Alkthar and
Spence-Hilton
(1991)
Assery and
Peel (1991)
OLS
Error
Correction
Model
Nominal
Exchange
Rate
ARCH Model
Nominal
Exchange
Rate
Real
Exchange
Rate
Short Term
ARCH-in-mean
Short Term
Variability of an
effective nom.
ER index
Squared residual
from the
ARIMA process
Sample, Data
Countries
Quarterly data
From 1973 to 1990
G7 countries
Commodities
Major Findings
Remarks
Aggregate
- ER volatility has a significant

negative, impact on the volume of
exports in all countries in the study
- Market agents react to ER by
favouring domestic to foreign trade
- Non-stationarity in data
used is taken into account.
- Economic agents are
assumed to be risk averse.
Monthly data
1959-1972
1973-1985
Developed
Countries
Monthly data
1973-1990
Developed
Countries
Aggregate
The effect of volatility on imports is

weak although permanent shocks to
volatility do have a neg. impact on this
measure of trade
- The study includes both

used is taken into account.
Aggregate
GARCH has a statistically significant

impact on the reduced form equations
for all countries
1973-1987
DC(US trade)
Montly
Aggregate
The found some evidence for a

statistically significant relationship
between volatility and trade.
used is taken into account
-The model restrict the
variance to be the same as
that generated by the data.
-Reduced form model, it
cannot distinguish between
structural hypothesis.
1969-1995
Quarterly
Australia and DC.
Aggregate
and sectoral
trade data
- The impact of ER volatility does

differ between traded goods sectors.
- It remains difficult to firmly establish
the nature of the relationship.
73-90
Aggregate
Six developed
countries
1974-1981
Quartely
Us-Ger
- ER volatility was found to have a neg

impact on exports for 3 countries, and
a positive impact to another 3
countries.
Aggregate
1972-1987
Quarterly
Five DC
Aggregate
- Exchange rate variability reduced

volume of trade in the floating period.
- N on-stationarity in data
- No structural Model.
- Both ER and ER volatility
enter into the equations.
- ARCH-in-mean, it provides
more efficient coefficient
estimates avoids spurious
regression.
-Explicitly considered the
impact of risk on volume
through prices.
- RER Volatility has a significant

positive impact on exports,
Appendix 3 Empirical Literature on Exchange Rate Effects on Trade (Exchange Rate Volatility was Included)
Study
Bailey et.al,
(1986)
Estimation
Method
OLS
Commodities
Major Findings
Remarks
They found a positive and significant

association between ER variability and
real exports.
- In 3 cases of 33 there was a

negative effect
Ex. Rate
Treatment
Nominal and
Real
Exchange
Rate
Volatility
Measure
Short Term
Absolute%
changes and st.
deviation
Sample, Data
Countries
Quarterly
1962-1974
1975-1985
Nominal
Exchange
Rate
Exogenous
Short Term
St. dev, of
weekly rates of
changes. FER
1976-1984
DCs (intra EMS)
Quarterly
A, B & S
Manufactures
The main source of inconclusive

evidence on significant correlation
between ex. Rate variability and trade
is econometric misspecification
The study includes not only

ER but also inflation and
price insulation policies.
used was not taken into
account
Export equations allowing
for uncertainty are estimated
He argued that third country
ER factors must be included
Aggregate
Bini-Smaghi
(1991)
OLS
Caballero and
Corbo (1989)
OLS and
IV
Real
Exchange
Rate
Exogenous
Long Term
Annual St. desv.
of the RER
Quarterly data
6 developing
countries
Aggregate
A significant negative effect of real

exchange rate uncertainty on exports
Cushman
(1986)
OLS
Real
Exchange
Rate
Short Term
Moving four
quarter st. dev.
Of changes in
ER
1965-1977
1973-1983
DC(US exports
Aggregate
He found significant third country risk

effects. Negative impact on trade flows
from ER risk
Cushman
(1988a)
OLS
Real
Exchange
Rate
1973-1983
DC(US exports)
Aggregate
Firs paper in using RER

based measure of ER risk
De Grauwe
(1987)
SURE
Both
Exogenous
1960-1969
1973-1984
EMS Countries
Aggregate
Significant negative effects of

exchange rate risk on volume of trade in the absence of risk, U.S. imports
would have been about 9% ingher
The greater ex. Rate stability provided
by the EMS has been a source of
growth of intra EMS trade
Gagnon
(1993)
OLS
Theoretical
Model
Significant negative effects on the

level of trade.
OLS
US- 6 ind
countries
1960-1971
1972-1988.
1965-1975
(US - Ger)
Aggregate
Trade
Hooper and
Kohlhagen
(1978)
Real
Exchange
Rate
Exogenous
Nominal
Exchange
Endogenos
Aggregate
bilateral and
multilateral
Ex. Rate uncertainty has a significant

impact on prices but not significant
impact on volume of trade.
The Model is analyzed both

under the assumption of
perfect and imperfect
competition.
Long-run effects should be
considered since they may be
significant w.r.t quantity
Short Times
Bilateral
Exchange Risk
Long Term
Variance of the
yearly% changes
of the bilateral
RER
Uncondditional
variance
Short Term
Variance of the
future spot ex.
Rate.
Setting fiscal policies in an

uncoordinated way within
EMS may contributed to
deflationary bias
Literature on Exchange Rate Effects on Trade (Exchange Rate Volatility was Included)
Study
to
so
Estimation
Method
Ex. Rate
Treatment
Klein (1990)
Pooled
Regression
Real
Exchange
Rate
Klein (1990)
Pooled
Regression
Real
Exchange
Rate
Konno and
Fukushige
(2002)
McKenzie
and Brooks
(1997)
Dynamic
OLS
Real
Exchange
Rate
Nominal
Exchange
Rate
OLS
Perre and
Steinher
(1989)
OLS
Nominal
Exchange
Rate
Thursby and
Thursby
(1987)
OLS
Nominal
Exchange
Rate
Chowdhury
(1993)
Multivariat
e Error
Correction
Model
Koray and
Lastrapes
(1989)
VAR
Real
Exchange
Rate
Nominal
Exchange
Rate
Volatility
Measure
Short Term
Volatility
St.dev. of the
montly %
change in RER.
Short Term
Volatility
St.dev. of the
montly %
change in RER.
Long Term,
Volatility
Short Term
ARCH
Long term
function of the
magnitude of
past movements
in nominal ER
Short Term
Variance of the
Spot ER
Moving sample
st. deviation of
the growth rates
of real ER
Short Term
Moving st. dev.
Of the growth
rate of RER
Sample, Data
Countries
Commodities
Major Findings
Remarks
78-86 Montly(US
exports)
9 Sectors
In 6 of the 9 categories, the volatility

of the exchange rate significantly
affects the value o f exports and in 5
categories the effect is positive
-He showed that volatility

effects differ across
categories of exports.
78-86 Montly(US
exports)
9 Sectors
In 6 of the 9 categories, the volatility

of the exchange rate significantly
affects the value of exports and in 5
categories the effect is positive
-He showed that volatility

effects differ across
categories of exports.
1981-1994
Quarterly
Aggregate
1973-1992
Montly
US-Germany
Aggregate
NAFTA has no additional impact on

import function gradual switching
existed before NAFTA.
They found a significant positive
relationship between volatility and
trade flows
-Unit root tests were

conducted cointegration test
was not conducted
-It would appear that the
volatility in the ER is
sourced solely from NER
1960-1985
Five developed
countries
Aggregate
bilateral
When ER uncertainty is defined over a

medium term it does affect adversely
trade flows of DC but not U.S

Non-stationary of the data
used
1974-1982
D C s (17)
Aggregate
We found support for the Linder

hypothesis, Moreover we found strong
support that variability of ER depress
trade
-Linder hypothesis is tested

in a gravity model.
-Real and nominal measures
of ER risk.
Quarterly data
From 1973 to 1990
G7 countries
Aggregate
Ex. Rate volatility has a significant

neg. impact on the volume of exports
in all countries in the study
- Non-stationary in data
Monthly data
1959-1972
1973-1985
Developed
Countries
Aggregate
The effect of volatility on imports is

weak although permanent shocks to
volatility do have a neg. impact on this
measure of trade
-The study includes both

Appendix 3 Empirical Literature on Exchange Rate Effects on Trade (Exchange Rate Volatility Was Included in the Model)
Estimation
Method
Co
integration
Ex. Rate
Treatment
Nominal
Exchange
Rate
Volatility
Measure
Short Term
GARCH
Lastrapes and
Koray (1990)
VAR
Real
Exchange
Rate
Short term
Moving st.
deviation of real
ER.
McKenzie
(1998)
Co
integration
Real
Exchange
Rate
Long Term
ARCH
Qian and
Varangis
(1994)
SUR
Nominal
Exchange
Rate
Short Term
ARCH
Study
Kroner and
Lastrapes
(1989
Sample, Data
Countries
Monthly data
1973-1990
Developed
Countries
1973-1987
DC(US trade)
Montly
Commodities
Major Findings
Remarks
Aggregate
GARCH has a statistically significant

impact on the reduced form equations
for all countries
-The model restrict the

variance to be the same as
that generated by the data.
Aggregate
The found some evidence for a

statistically significant relationship
between volatility and trade.
-Reduced form model cannot

distinguish between
structural hypothesis.
1969-1995
Quarterly
Australia and DC.
Aggregate
and Sectoral
trade data
-No structural Model Both

ER and ER volatility enter
into the equations
1973-1990
Six developed
countries
A&B
The impact of ex rate volatility does

differ between traded goods sectors. It
remains difficult to firmly establish the
nature of the relationship.
ER volatility was found to have a neg
impact on exports for 3 countries, and
a positive impact to another 3
countries.
-Arch in mean, it provides

more efficient coeff.
-Estimates avoids spurious
regression.
Appendix 4 Empirical Literature on Mexican Agricultural Trade (Exchange Rate Volatility was not Included in the Model)
Study
Estimation
method
Ex. Rate
Treatment
Volatility
Measure
Diaz, (2002)
Co-integration
Error
Correction
Model
Real ex rate
None
Real
Exchange
Rate
Real
Exchange
Rate
Nominal
Exchange
Rate
None
VAR
Real
Exchange
Rate
None
Arellano, et al
(1998)
3-SLS
Real
Exchange
Rate
None
Malaga, et.al.
(2001).
3SLS
Real
Exchange
Rate
None
Mendoza
(1996)
OLS
Mora-Flores, et.,
(2002)
2-SLS
Salas (1980)
OLS
Varella, (20002)
Konno and
Fukushige
(2002)
Dynamic
OLS
Real
Exchange
Rate
None
None
Long Term,
Volatility
Sample, Data
Countries
Yearly data
80-89 and
90-2000
Mexico
Yearly data
1970-1994
Mexico
Yearly data
1988-1986
Mexico
Yearly data
from
1961-1979
Quarterly data
1980-2000
Mexico
Winter season
data from
1970 to 1994
Mexico- U.S.
Caribean
Winter season
data
1974-1993
Mexico-U.S.
Quarterly data
1981-1994
Commodities
Major Findings
Remarks
Aggregate Trade
Flows
Consumption,
capital and in
term goods.
There was a structural change in

the Mexican External sector due
to the liberalization of the 1980s.
Most import and export functions
showed stability for the period
1990-2000.
Dynamic and Non-linear ER
effect on exports was found
- Non-stationary in data used

is taken into account
- Import and exports
equation were estimated
individually
A direct relationship between

export real prices and exported
quantities was found
Structural change in 1978-79
Significant negative effect of ex.
Rate.
- Did not take non-stationary

of the data used into account
Manufactures
Crops and live

stock
Disaggregated
trade
- Did not take non-stationary

of the data used into account
- Not volatility measure was

used
- ER enters just as another
exogenous variable
- Non-stationary in data
Tradables Non
tradables
The effects of RER shocks on

consumption and on the current
account in Mexico are strong
(statistically significant)
Melon
- Devaluation of the Mexican

peso was significant in increasing
melon exports to U.S. in the SR.
Fresh vegetables
Devaluation was by far the most

important factor behind the SR
increase in Mexican imports but
in LR differentials in growth
rates in production

non-stationary in data used
-The model accounts for
tariffs, ER, Labour costs,
yields, income, population
Aggregate
NAFTA has no additional impact

on import function gradual
switching existed before NAFTA.
Unit root tests were

conducted cointegration test
was not conducted

non-stationary in data used
in the analysis.
APPENDICES 5-20
Determination of Optimal Lag Specification for
Selected VAR Models
294
Appendix 5 Optimal Lag Length Selection for Tomato Model VI
M ODEL VI
NLAG S
Log
L ikelihood
-36.841
-37.314
-37.314
-37.841
-37.841
-38.194
-38.194
-38.828
N um ber
V ariables
N*
M LTest
158
58.559
(df=7,
5% )
14.02
Reject Lag 2
151
58.059
14.02
Reject Lag 3
144
33.813
14.02
Reject Lag 4
137
52.050
14.02
Reject Lag 5
130
19.394
14.02
Reject Lag 6
123
21.033
14.02
Reject Lag 7
116
27.967
14.02
Reject Lag 8
109
11.508
14.02
Accept Lag 9
102
6.758
14.02
**
95
-67.053
14.02
L ags
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
ElevenTwelve
-38.828
-39.114
-39.114
-39.503
-39.503
-40.202
-40.202
10
-40.645
10
-40.645
11
-41.208
11
-41.208
T est
R esult
LRm X \
34
41
48
55
62
69
76
83
90
N* Net Number of Observations

** Denotes that once an optimal lag is achieved, no further testing is conducted.
295
Appendix 6 Optimal Lag Lenght Selection for Tomato Model V2

M ODEL V2
Lags
NLAGS
Log
Likelihood
-34.423
N um ber
Variables
N*
M L-Test
158
45.688
14.02
Reject Lag 2
151
42.971
14.02
Reject Lag 3
144
44.432
14.02
Reject Lag 4
137
55.685
14.02
Reject Lag 5
130
19.602
14.02
Reject Lag 6
123
17.630
14.02
Reject Lag 7
116
30.675
14.02
Reject Lag 8
109
8.045
14.02
Accept Lag 9
102
7.534
14.02
**
95
-2.430
14.02
LRm x \
Test Result
(df=7) 5%
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
Eleven-Twelve
-34.791
-34.791
-35.182
-35.182
-35.645
-35.645
-36.324
-36.324
-36.612
-36.612
-36.938
-36.938
-37.705
-37.705
10
-38.015
10
-38.015
11
-38.642
11
-38.642
12
-39.857
34
41
48
55
62
69
76
83
90
97
N* N et Number of Observations
296
Appendix 7 Optimal Lag Lenght Selection for Tomato Model with VG
M ODEL VN
Lags
NLAGS
Log
Likelihood
N um ber
Variables
-37.585
Three-Four
3
3
4
-37.991
-37.991
-38.439
53
-38.439
-39.065
-39.065
-39.972
61
-39.972
-40.423
69
-40.423
-40.907
77
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
ML-Test
LRm
Test Result
(df=7) 5%
Two-Three
Four-Five
N*
7
7
8
8
155
47.917
14.02
Reject Lag 2
147
45.669
14.02
Reject Lag 3
139
53.860
14.02
Reject Lag 4
131
63.517
14.02
Reject Lag 5
123
24.330
14.02
Reject Lag 6
115
18.375
14.02
Reject Lag 7
107
20.403
14.02
Reject Lag 8
99
3.599
14.02
Accept Lag 9
91
-332.696
14.02
**
83
-34.096
14.02
37
45
-40.907
9
9
10
-41.834
-41.834
85
-42.434
93
10
11
-42.434
-9.164
101
11
-9.164
-10.476
109
Eleven-Twelve
12

297
Appendix 8 Optimal Lag Lenght Selection for Tomato Model VN

MODEL VN
Lags
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
NLAGS
2
-38.098
3
3
4
-38.745
-38.745
-39.160
4
5
-39.160
5
6
6
7
7
Eigth-Nine
8
8
Ten-Eleven
Number
Variables
N*
ML-Test
LRm x ]
Test Result
(df=7) 5%
Seven-Eigth
Nine-en
Log
Likelihood
-39.651
-39.651
-40.161
-40.161
-6.428
-6.428
-6.740
-7.318
-7.318
-7.988
-7.988
-8.614
11
12
-8.614
80.231
14.02
Reject lag 2
151
45.684
14.02
Reject lag 3
144
47.156
14.02
Reject lag 4
137
41.792
14.02
Reject lag 5
130
2293.819
14.02
Reject lag 6
123
16.807
14.02
Reject lag 7
116
23.150
14.02
Reject lag 8
109
17.413
14.02
Reject lag 9
102
7.507
14.02
Accept lag 10
95
-2.866
14.02
**
41
48
55
62
69
-6.740
9
9
10
10
11
158
34
76
83
90
Eleven-Twelve
-10.047
97

298
Appendix 9 Optimal Lag Length Selection for Maize Model VI

MODEL
VI
Lags
Two-Three
ThreeFour
Four-Five
Five-Six
Six-Seven
SevenEigth
Eigth-Nine
Nine-en
Ten-Eleven
NLAGS
Log
Likelihood
-33.618
-33.966
3
4
4
5
5
6
6
-33.966
-35.928
-35.928
-36.213
-34.597
Num ber
Variables
-36.213
-36.474
-36.474
10
-37.245
10
-37.245
11
-37.803
11
-37.803
12
-4.197
Test Result
158
43.088
14.02
Reject Lag 2
151
69.469
14.02
Reject lag 3
144
52.785
14.02
Reject lag 4
137
27.610
14.02
Reject lag 5
130
30.220
14.02
Reject lag 6
123
15.401
14.02
Reject lag 7
116
14.411
14.02
Reject lag 8
109
20.064
14.02
Reject lag 9
102
6.693
14.02
Accept Lag 10
95
-67.213
14.02
48
55
-35.484
(df=7) 5%
41
-35.147
-35.484
M L-Test
34
-34.597
-35.147
LRm x ]
N*
62
69
76
83
90
ElevenTwelve
97

299
Appendix 10 Optimal Lag Length Selection for Maize Model V2

MODEL
V2
Lags
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
NLAGS
Log
Likelihood
Num ber
Variables
ML-Test
N*
LRm x l
Test Result
(df=7)
2
-34.260
3
3
4
-34.641
-34.641
34
-35.047
41
5
5
-35.047
-35.437
-35.437
48
5
6
6
7
7
8
8
9
9
10
10
11
-35.806
-35.806
55
-36.226
62
-36.226
-36.744
-36.744
69
-37.335
-37.335
-38.060
-38.060
76
-38.649
90
158
47.252
14.02
Reject Lag 2
151
44.693
14.02
Reject lag 3
144
37.392
14.02
Reject lag 4
137
30.252
14.02
Reject lag 5
130
28.573
14.02
Reject lag 6
123
28.008
14.02
Reject lag 7
116
23.602
14.02
Reject lag 8
109
18.854
14.02
Reject lag 9
102
7.077
14.02
Accept Lag 10
-1.220
14.02
83
ElevenTwelve
11
12
-38.649
-39.259
95
97

300
Appendix 11 Optimal Lag Length Selection for Maize Model VG

MODEL VG
Lags
Two-Three
Three-Four
Four-Five
Five-Six
Number
Variables
NLAGS
Log
Likelihood
-6.083
3
3
4
4
-6.496
34.000
-6.496
-7.302
41.000
5
5
6
-7.302
-7.854
-7.854
48.000
-8.209
55.000
-8.209
8.310
8.310
62.000
Seven-Eigth
6
7
7
Eigth-Nine
8
8
8.700
8.700
9.130
9.130
Six-Seven
Nine-en
Ten-Eleven
9
10
10
11
LRm x ]
N*
M l Test
(df=7) 5%
Test Result
158
51.105
14.02
Reject Lag 2
151
88.675
14.02
Reject lag 3
144
52.975
14.02
Reject lag 4
137
29.184
14.02
Reject lag 5
130
23.319
14.02
Reject lag 6
123
21.060
14.02
Reject lag 7
116
17.200
14.02
Reject lag 8
109
14.560
14.02
Reject Lag 9
102
6.960
14.02
Accept Lag 10
95
-1.360
14.02
69.000
76.000
9.690
9.690
10.270
83.000
90.000
10.270
10.950
97
ElevenTwelve
11
12

301
Appendix 12 Optimal Lag Length Selection for Maize Model VN

MODEL
VN
Lags
Two-Three
ThreeFour
Four-Five
Num ber
Variables
NLAGS
Log
Likelihood
-35.822
-36.133
3
4
-36.133
-36.688
41
4
5
48
Five-Six
Six-Seven
6
6
7
-37.501
-37.501
-37.981
SevenEigth
-37.981
8
8
-38.302
-38.302
69
Eigth-Nine
-4.159
76
9
10
-4.159
-4.886
83
10
11
-4.886
-5.464
90
11
-5.464
TenEleven
M L -Test
(df=7) 5%
Test Result
158
38.493
14.020
Reject Lag 2
151
61.061
14.020
Reject lag 3
144
52.192
14.020
Reject lag 4
137
22.069
14.020
Reject lag 5
130
32.656
14.020
Reject lag 6
123
17.355
14.020
Reject lag 7
116
1365.735
14.020
Reject lag 8
109
18.908
14.020
Reject lag 9
102
6.938
14.020
Accept Lag 10
95
-1.281
14.02
34
-36.688
-37.231
-37.231
Nine-en
LRm X d
N*
55
62
ElevenTwelve
-6.105
97
N* Net Num ber of Observations

302
Appendix 13 Optimal Lag Length Selection for Sorghum Model VI

M ODEL V I
Lags
NLAGS
Log
Likelihood
Two-Three
-34.260
-34.641
-34.641
-35.047
-35.047
-35.437
Three-Four
Four-Five
Five-Six
-35.437
-35.806
Six-Seven
-35.806
-36.226
Seven-Eigth
-36.226
-36.744
-36.744
-37.335
Eigth-Nine
Nine-en
Ten-Eleven
-37.335
10
-38.060
10
-38.060
11
-38.649
11
-38.649
12
-39.259
Num ber
Variables
LRm x \
Test Result
N*
ML Test
(df=7) 5%
158
47.252
14.02
Reject Lag 2
151
44.693
14.02
Reject lag 3
144
37.392
14.02
Reject lag 4
137
30.252
14.02
Reject lag 5
130
28.573
14.02
Reject lag 6
123
28.008
14.02
Reject lag 7
116
23.602
14.02
Reject lag 8
109
18.854
14.02
Reject Lag 9
102
7.077
14.02
Accept Lag 10
95
-1.220
14.02
34.000
41.000
48.000
55.000
62.000
69.000
76.000
83.000
90.000
ElevenTwelve
97.000

303
Appendix 14 Optimal Lag Length Selection for Sorghum Model V2.

MODEL V2
Lags
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
NLAGS
Log
Number
Likelihood Variables
-31.946
-32.315
-32.315
-32.671
-32.671
-33.094
-33.094
-33.507
-33.507
-33.885
-33.885
-34.483
-34.483
-35.127
-35.127
10
-35.692
10
-35.692
11
-36.568
11
-36.568
12
-37.305
LRm X l
N*
Test Result
M L Test (df=7) 5%
158
45.711
14.02
Reject Lag 2
151
39.154
14.02
Reject lag 3
144
40.599
14.02
Reject lag 4
137
33.873
14.02
Reject lag 5
130
25.679
14.02
Reject lag 6
123
32.292
14.02
Reject lag 7
116
25.775
14.02
Reject lag 8
109
14.686
14.02
Reject lag 9
102
10.516
14.02
Accept Lag 10
95
-1.473
14.02
34.000
41.000
48.000
55.000
62.000
69.000
76.000
83.000
90.000
Eleven-Twelve
97.000
N* Net Num ber o f Observations

304
Appendix 15 Optimal Lag Lenght Selection for Sorghum Model VG

MODEL VG
Lags
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
NLAGS
Log
Num ber
-7.266
-7.628
-7.628
-8.215
-8.215
-8.560
-8.560
-8.962
-8.962
-9.447
-9.447
-9.816
-9.816
-10.367
-10.367
10
-10.989
10
-10.989
11
-11.568
11
-11.568
12
-12.557
LRm x \
N*
M l - Test (df=7) 5% Test Result
158
44.988
14.020
Reject Lag 2
151
64.485
14.020
Reject lag 3
144
33.201
14.020
Reject lag 4
137
32.912
14.020
Reject lag 5
130
32.968
14.020
Reject lag 6
123
19.961
14.020
Reject lag 7
116
22.035
14.020
Reject lag 8
109
16.164
14.020
Reject lag 9
102
6.955
14.020
Accept Lag 10
95
-1.978
14.02
34
41
48
55
62
69
76
83
90
Eleven-Twelve
97
N* Net Num ber of Observations

305
Appendix 16 Optimal Lag Length Selection for Sorghum Model VN

MODEL VN
Lags
NLAG
Two-Three
Three-Four
3
3
Number
Log
-36.491
-36.823
-36.823
-37.142
Four-Five
-37.142
5
5
-37.522
-37.522
48
Five-Six
6
6
7
-38.015
-38.015
55
-38.429
-38.429
-4.438
-4.438
62
Six-Seven
Seven-Eigth
Eigth-Nine
8
8
Nine-en
Ten-Eleven
N*
ML
Test
LRm Xd
(df=7) 5%
Test Result
158
41.145
14.02
Reject Lag 2
151
35.032
14.02
Reject lag 3
144
36.564
14.02
Reject lag 4
137
40.381
14.02
Reject lag 5
130
28.182
14.02
Reject lag 6
123
1835.541
14.02
Reject lag 7
116
1398.620
14.02
Reject lag 8
109
881.0
14.02
Reject lag 9
102
7.580
14.02
Accept Lag 10
95
-1.489
14.02
34
41
69
9
9
10
10
11
-39.403
-39.403
76
-5.519
83
-6.151
90
11
12
-6.151
-6.896
97
-5.519
ElevenTwelve
N* Net Number o f Observations

306
Appendix 17 Optimal Lag Length Selection for Milk Powder Model 1
M ODEL V I
Lags
Two-Three
NLAGS
2
3
Three-Four
Four-Five
3
4
4
5
Five-Six
Six-Seven
5
6
6
7
Number
Log
-28.165
-28.567
-29.055
-29.055
41.000
-29.409
-29.409
-29.871
48.000
Eigth-Nine
8
8
9
-30.951
-30.951
-31.684
9
10
-31.684
-32.231
10
11
-32.231
-33.003
11
-33.003
12
-33.648
Nine-en
Ten-Eleven
M L Test
(df=7) 5%
Test Result
158
49.813
14.02
Reject Lag 2
151
53.704
14.02
Reject lag 3
144
33.972
14.02
Reject lag 4
137
37.940
14.02
Reject lag 5
130
37.199
14.02
Reject lag 6
123
29
14.02
Reject lag 7
116
29.331
14.02
Reject lag 8
109
14.218
14.02
Reject Lag 9
102
9.269
14.02
Accept Lag 10
95
-1.290
14.02
55.000
-29.871
Seven-Eigth
N*
34.000
-28.567
-30.419
-30.419
LRm
62.000
69.000
76.000
83.000
90.000
ElevenTwelve
97.000

307
Appendix 18 Optimal Lag Length Selection for Milk Powder Model V2

MODEL
V2
Lags
Two-Three
NLAGS
2
3
Three-Four
Four-Five
Five-Six
Six-Seven
SevenEigth
Eigth-Nine
Nine-en
Ten-Eleven
Log
Likelihood
Number
Variables
-25.915
-26.307
34
-26.307
4
4
5
5
6
6
-26.747
-26.747
-27.068
-27.068
-27.447
-27.447
-28.015
7
8
-28.015
-28.578
8
9
9
-28.578
-29.314
-29.314
10
10
11
-29.902
-29.902
-30.756
11
-30.756
-31.492
LRm x ]
N*
M L Test
(df=7) 5%
Test Result
158
48.619
14.02
Reject Lag 2
151
48.342
14.02
Reject lag 3
144
30.795
14.02
Reject lag 4
137
31.115
14.02
Reject lag 5
130
38.616
14.020
Reject lag 6
123
30.371
14.020
Reject lag 7
116
29.457
14.020
Reject lag 8
109
15.302
14.020
Reject lag 9
102
10.247
14.020
Accept Lag 10
95
-1.471
14.020
41
48
55
62
69
76
83
90
ElevenTwelve
12
97
N* N et Number of Observations
308
Appendix 19 Optimal Lag Length Selection for Milk Powder Model VG

MODEL VG
Lags
Two-Three
Three-Four
Four-Five
Five-Six
Six-Seven
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
NLAGS
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
11
Number
Log
-35.111
-35.620
-35.620
-36.299
-36.299
-36.789
-36.789
-37.292
-37.292
-37.852
-37.852
-38.502
-38.502
-4.621
LRm x l
N*
M L - Test
(df=7) 5%
Test Result
158
63.174
14.02
Reject Lag 2
151
74.613
14.02
Reject lag 3
144
47.030
14.02
Reject lag 4
137
41.300
14.02
Reject lag 5
130
38.089
14.02
Reject lag 6
123
35.090
14.02
Reject lag 7
116
1355.2
14.02
Reject lag 8
109
12.529
14.02
Accept Lag 9
102
7.620
14.02
95
-2.279
14.02
34
41
48
55
62
69
76
-4.621
-5.103
-5.103
83
-5.738
90
ElevenTwelve
11
-5.738
12
-6.878
97
* Net Number of Observations

309
Appendix 20 Optimal Lag Length Selection for Milk Powder Model VN
NLAGS
Log
Likelihood
Two-Three
-30.421
-30.814
-30.814
34
Three-Four
3
3
4
4
-31.215
-31.215
41
Four-Five
-31.576
-31.576
48
Five-Six
5
5
-31.886
-31.886
-32.359
55
Six-Seven
6
6
7
8
8
-32.949
-32.949
69
10
11
-33.718
-33.718
-34.073
-34.073
-34.962
76
90
11
12
-34.962
-35.578
97
MODEL VN
Lags
Seven-Eigth
Eigth-Nine
Nine-en
Ten-Eleven
9
9
10
Num ber
Variables
LRm x ]
Test Results
N*
M L Test
(df=7)
158
48.741
14.02
Reject Lag 2
151
44.103
14.02
Reject lag 3
144
34.650
14.02
Reject lag 4
137
25.444
14.02
Reject lag 5
130
32.191
14.02
Reject lag 6
123
31.861
14.02
Reject lag 7
116
30.737
14.02
Reject lag 8
109
9.235
14.02
Accept Lag 9
102
10.671
14.02
95
-1.232
14.02
62
-32.359
83
ElevenTwelve

310
APPENDICES 21-36
Maximum Likelihood Cointegration Results
311
Appendix 21 Maximum Likelihood Cointegration Results:

United States Import Demand for Tomato
A S ev en -V a r ia b le N in e L a g S ystem U sin g V I
E ig en v a lu es
0.271292
0.22766
0.178369
0.142208
0.122468
0.074314
0.006916
E ig en v ecto rs
USY
-1.0085
-0.5919
-1.1695
4.3220
-0.0067
2.5105
0.1807
ER
-1.4250
-1.2222
-1.0243
-0.0034
0.6317
-2.7650
0.7689
VI
0.3620
-1.9924
-0.6794
-3.3103
0.1215
2.6320
0.8981
Pt
0.6879
1.6578
-2.1761
-1.3886
0.3383
0.2091
-1.3216
Pea
2.0680
-2.1329
-0.1719
2.1181
0.1195
-0.9394
-1.1540
W1
-1.4249
-0.6637
1.0813
-0.7911
0.4424
1.0963
-2.2798
Qt
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eig h ts
USY
0.00023
0.00041
-0.00026
-0.00027
0.00074
0.0002
-0.0002
ER
0.01159
-0.00189
-0.00229
-0.00605
-0.00417
-0.0008
-0.0006
VI
0.07999
-0.06329
-0.06878
-0.04569
-0.02019
0.0423
0.0053
Pt
0.00678
0.06511
-0.05019
0.03072
-0.01794
0.0002
-0.0004
Pea
-0.03253
0.00996
-0.03879
-0.01720
-0.01193
-0.0154
-0.0000
W1
0.01161
-0.01060
-0.00455
-0.00210
-0.00408
-0.0026
-0.0016
Qt
-0.02733
-0.00570
0.01222
-0.00658
-0.04890
0.0187
-0.0037
T estin g th e N u m b er o f C o in teg ra tin g V e cto rs
Trace
Statistic
Trace
(0.95)
MAX (X.)
M AX (X)
(0.95)
*1
II
o
Null
Hypothesis
207.38*
150.40
57.60*
50.51
r< 1
149.78*
117.49
47.02*
44.37
r<2
102.77*
88.59
35.76
38.22
r<3
63.01
63.66
27.92
31.99
r<4
39.09
42.70
23.78
25.68
r<5
15.32
25.64
14.05
19.21
r<6
1.26
12.34
1.26
12.34
* Significance at 5 percent level (MacKinnon, 1999)
312

A S ev e n -V a r ia b le N in e L a g S ystem u sin g V 2
E ig en v a lu es
0.314072
0.228075
0.171309
0.159458
0.11313
0.08900
0.02026
E ig en v ecto rs
USY
-0.5957
-1.2060
0.8849
-4.5951
-0.7619
3.2666
-0.6717
ER
-1.8126
-0.4431
0.1668
-8.9773
0.2224
-3.5346
0.8252
V2
0.4796
-0.5566
-0.7489
-6.0675
0.4413
4.2511
0.8607
W1
-1.3858
0.6769
-0.0568
-1.2528
0.8627
0.6483
-2.4053
Pea
-2.4410
0.4417
0.0248
-5.7788
-0.0811
-1.7996
-1.0284
Pt
0.5323
-5.3284
0.0466
1.9700
0.3984
-0.7687
-0.2096
Qc
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eigh ts
USY
0.00027
-0.00046
-0.00009
0.00029
-0.00073
-0.00023
0.00041
ER
-0.01087
-0.00113
0.00788
-0.00506
-0.00118
0.00150
0.00109
V2
-0.05619
-0.19450
0.10891
-0.28325
0.00974
-0.18688
-0.05002
W1
-0.01223
0.00163
0.00209
-0.01177
-0.00029
0.00276
0.00250
Pea
0.04893
-0.00962
0.02239
-0.02092
-0.00452
0.02318
0.00046
Pt
0.05465
-0.07149
0.00293
0.00373
0.02184
0.01968
0.00376
Qc
0.01444
0.01967
0.03146
-0.00781
0.03945
-0.01271
0.00853
Trace
Statistic
Trace
(0.95)
MAX
(X)
M AX (X)
(0.95)
II
T estin g th e N u m b er o f C o in teg ra tin g V ec to r s
224.05*
150.40
68.61*
50.51
r< 1
155.44*
117.49
47.11*
44.37
r<2
108.33*
88.59
34.19
38.22
r<3
74.13*
63.66
31.61
31.99
r<4
42.51
42.70
21.85
25.68
r <5
20.66
25.64
16.93
19.21
3.73
12.34
3.73
12.34
Null hypothesis
r<6
313

A S ev en -V a ria b le N in e L a g S ystem u sin g V G
E ig en v a lu es
0.2400
0.1805
0.1340
0.1060
0.1013
0.0574
0.0210
E ig en v ec to rs
USY
0.4115
-0.2463
-0.6911
1.8000
1.2848
-4.0166
-0.4092
ER
1.0963
-0.5888
4.4093
0.3389
0.5607
1.6184
-0.6003
VG
-0.4492
-0.4604
-3.1409
1.2229
1.0981
3.1982
0.2617
W1
0.4909
-0.5053
0.3450
-0.8676
-0.1159
-0.9232
2.6691
Pt
-2.4009
0.2850
2.2059
3.1834
-0.1577
-0.0887
0.5835
Pea
-1.6991
-0.8440
1.0792
-3.8290
1.1606
-0.3269
-0.0205
Qt
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eig h ts
USY
0.00048
-0.00058
-0.00066
0.00031
-0.00001
-0.00070
-0.00021
ER
-0.00502
-0.00501
0.00418
-0.00147
0.00428
-0.00355
0.00126
VG
-0.00044
-0.00920
0.00009
0.00080
0.00243
0.00057
-0.00063
W1
-0.00995
-0.00791
0.00829
0.00263
0.00072
-0.00416
0.00000
Pt
0.04810
-0.01693
0.02731
-0.03221
-0.00068
0.00373
-0.00767
Pea
0.10529
-0.02008
0.03870
0.02992
0.01274
0.00086
0.01151
Qt
0.02564
0.04707
0.04380
-0.00188
0.01092
-0.00894
0.00101
Trace
(0.95)
MAX
II
Null hypothesis
166.80*
150.40
49.95*
50.51
r< 1
116.85*
117.49
36.22
44.37
*1
Trace
Statistic
T estin g the N u m b e r o f C o in teg ra tin g V ecto rs
a)
MAX(X)
(0.95)
r<2
80.63
88.59
26.19
38.22
r<3
54.44
63.66
20.40
31.99
r<4
34.04
42.70
19.43
25.68
r<5
14.61
25.64
10.75
19.21
r<6
3.85
12.34
3.85
12.34
314

A S ev e n -V a ria b le N in e L a g S y ste m u sin g V N
E ig en v a lu es
0.2916
0.2325
0.1786
0.1234
0.0984
0.0184
0.0160
E ig en v ecto rs
USY
-0.2204
-0.4611
1.1117
-0.5487
2.2797
-0.6308
-1.8070
ER
-0.6020
-1.0109
0.6143
-0.0140
-3.6494
-0.0701
1.1895
VN
0.2290
0.3944
-0.0229
0.3579
-3.4483
2.4819
-2.0620
Pt
1.0104
0.9522
-0.1465
0.8374
1.1661
-1.1215
-0.5534
Pea
1.8788
1.2893
0.3737
-0.3691
-3.6476
-1.9575
0.0285
W1
-0.6695
-0.4636
-0.1434
0.7855
-1.1315
-2.5702
-0.9714
Qt
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eights
USY
-0.00056
-0.00010
0.00018
0.00053
0.00054
-0.00023
-0.00029
ER
-0.00005
0.00494
-0.00753
0.00299
-0.00037
-0.00176
0.00066
YG
0.00977
0.01970
-0.01950
-0.00861
0.00010
0.01124
-0.00754
W1
-0.00716
0.05219
0.01924
-0.03046
0.00703
-0.00400
0.00337
Pt
-0.06925
0.00500
-0.00098
0.00504
-0.01279
0.00201
0.00459
Pf
0.00070
0.00841
-0.00470
0.00446
-0.00583
-0.00223
-0.00082
Qt
-0.01826
-0.02681
-0.03591
-0.03739
-0.01402
-0.00649
-0.00259
T estin g the N u m b e r o f C o in teg ra tin g V ecto rs

Trace
Statistic
Trace
(0.96)
M AX
(X)
MAX(X)
(0.95)
*1
11
o
Null hypothesis
194.751*
150.41
62.389*
50.51
r<1
132.361*
117.49
47.890*
44.37
r<2
84.471
88.59
36.603
38.22
r <3
48.868
63.66
23.836
31.99
r<4
25.032
42.71
18.742
25.68
r<5
6.290
25.64
3.367
19.21
r<6
2.922
12.34
2.922
12.34
315
Appendix 25 Maximum Likelihood Cointegration Results: Mexicos Import

Demand for Maize
A S ev en V a r ia b le s T en L a g s S y stem U sin g V I
E igenvalues
0.2670
0.2305
0.1379
0.1083
0.0527
0.0263
0.0035
E ig en v ecto rs
MXY
ER
VI
INV
Pc
Pcd
Qc
-1.572
-0.751
1.691
0.154
0.584
-0.554
0.072
-0.171
1.000
0.184
0.776
-2.607
1.155
-47.390
-16.054
-0.136
-0.001
0.085
0.341
5.251
2.214
-1.286
2.266
-32.197
57.211
-95.224
-25.792
0.543
-0.370
-0.514
1.000
1.000
1.000
0.687
1.000
16.437
10.419
0.971
11.198
-16.060
-8.455
5.556
-6.554
-6.660
-0.947
-4.149
0.792
1.000
1.000
0.0005
0.0010
0.0028
0.0003
-0.0003
-0.0041
0.0095
-0.0020
-0.0027
0.0036
-0.0010
0.0001
0.0490
0.0001
W eig h ts
MXY
ER
VI
INV
Pc
Pcd
Qc
0.0012
-0.0003
0.1484
0.0000
-0.0002
-0.0432
0.0363
0.0072
-0.0101
-0.0390
0.0150
0.0031
0.0173
0.1913
0.0006
0.0013
-0.0012
0.0001
0.0014
0.1545
0.0084
0.0068
-0.0791
0.0033
0.0100
0.0007
-0.0206
-0.0197
0.0027
0.0008
0.0405
0.0521
-0.0791
0.0541
0.0006
T estin g T he N u m b er o f c o in te g ra tin g V ecto rs

Trace
Statistic
Trace
(0.95)
Max
M ax X
(0.95)
II
o
**
Null
Hypothesis
166.504*
150.4
56.211*
50.51
r< 1
110.293
117.49
47.429*
44.37
r< 2
62.864
88.59
26.852
38.22
r< 3
36.012
63.66
20.740
31.99
r< 4
15.272
42.7
9.803
25.68
r< 5
5.468
25.64
4.828
19.21
0.641
12.34
0,641
12.34
r<6
316

Demand for Maize
A S ev en V a r ia b le s T en L ags S ystem U sin g V 2
E ig en v a lu es
0.2816
0.2330
0.1343
0.0903
0.0540
0.0280
0.0005
E ig en v ecto rs
MXY
ER
V2
INV
Pc
Pcd
Qc
-1.5078
0.0489
0.8105
-1.7068
-0.1465
3.8035
-5.6579
-0.6609
0.0425
4.4622
0.3479
-0.6384
2.9207
6.1278
1.1106
-0.7290
-0.6720
-1.4157
1.0618
0.8854
3.6993
-0.1382
0.1875
-10.3210
0.0909
-0.3575
1.5111
2.1693
0.2457
0.5568
0.5275
-3.2894
-0.5670
-1.7122
2.3001
-1.3745
0.1448
-0.5379
-1.0283
1.3200
0.0389
0.1777
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eigh ts
MXY
ER
V2
INV
Pc
Pcd
Qc
0.0005
-0.0010
0.0012
0.0005
0.0004
-0.0003
0.0001
0.0010
0.0002
0.0008
0.0019
0.0007
-0.0009
-0.0001
-0.2379
-0.3524
-0.0658
0.1247
-0.0886
-0.0338
-0.0056
0.0017
0.0193
0.1524
0.0049
-0.0581
-0.0052
0.0004
-0.0169
0.0047
0.0060
0.0062
0.0036
0.0015
-0.0003
0.0004
0.0085
-0.0034
0.0104
-0.0004
0.0029
0.0000
-0.1900
0.1333
-0.0231
0.0327
-0.0336
-0.0608
0.0013
T esting The N um ber o f C ointegrating V ectors

Null Hypothesis
Trace
Statistic
Trace
r=0
r< 1
r<2
r<3
r<4
r< 5
r<6
166.375*
150.4
106.512
117.49
58.501
88.59
32.406
63.66
15.278
42.7
5.226
25.64
0.082
12.34
59.862*
48.010*
M ax X
(0.95)
50.51
4437
26.095
17.127
10.052
5.143
0.082
38.22
31.99
25.68
19.21
12.34
M ax
317
Appendix 27 Maximum Likelihood Cointegration Results: for Mexicos Import

Demand for Maize
A Seven V ariables T en L ags System U sin g V G
Eigenvalues
0.234248
0.2032
0.1512
0.1297
0.0872
0.0220
0.0035
-0.5860
-2.3805
3.3061
-11.5885
Eigenvector
MXY
ER
-0.8473
0.9547
-0.1644
-0.5168
2.6810
-0.1812
1.2912
-1.2714
1.6409
20.1984
VG
0.8517
-1.3779
-0.2251
-0.2471
-0.3561
-0.6658
6.9855
INV
-0.3170
2.8230
0.1311
-3.2038
2.6480
0.6237
6.9953
0.3568
-1.4004
-5.7391
-1.0850
3.9987
0.0258
0.6383
-0.6188
-1.3585
-6.4744
1.0000
1.0000
1.0000
1.0000
1.0000
Pc
Pcd
0.3760
-0.4428
-0.6790
7.9606
Qc
1.0000
1.0000
W eights
MXY
ER
VG
INV
Pc
Pcd
0.0007
0.0000
-0.0004
0.0011
-0.0001
-0.0005
0.0003
0.0004
-0.0010
-0.0003
0.0018
0.0014
-0.0007
-0.0003
0.0042
-0.0017
-0.0002
-0.0015
0.0002
0.0010
-0.0001
0.0857
-0.0242
0.0890
0.0926
-0.0807
-0.0008
0.0002
0.0013
0.0143
0.0036
0.0080
0.0054
0.0015
-0.0010
-0.0043
-0.0044
0.0083
0.0033
0.0090
0.0022
0.0002
0.1340
0.0961
0.1315
-0.0662
0.0492
-0.0428
-0.0061
Qc
T e stin g the N u m b er o f C o in te g ra tin g V ectors

Null Hypothesis
r=0
r< 1
r< 2
r<3
r<4
r<5
r<6
Trace
Statistic
Trace
(0.95)
Max
Max X
(0.95)
165.420*
150.4
48.308
50.51
117.110
76.003
117.49
88.59
46.321
63.66
41.113
29.681
25.141
21.179
46.581
42.7
25.64
16.521
44.37
38.22
31.99
25.68
4.027
0.631
19.21
12.34
0.631
12.34
318

Demand for Maize
A Seven V ariables T en L ag System U sing V N
Eigenvalues
0.2610
0.1805
0.1260
0.1096
0.0440
0.0209
0.0028
E igenvector
MXY
ER
VN
-0.1553
0.1438
-1.1758
-2.1803
-0.2343
8.7934
-1.5472
-0.7287
-0.2397
1.2036
-2.2959
-0.9751
2.0795
3.8146
0.7672
0.5498
0.0663
-0.9794
-1.5498
-3.1225
-2.1506
0.7442
INV
Pc
-0.0482
0.0565
-0.5828
8.4852
-0.6842
2.0954
0.2783
0.2783
-3.7386
-1.3613
0.0141
-2.0889
2.0372
Pcd
Qc
-1.9290
-0.6702
-0.7381
1.1919
0.4674
-0.6166
-1.0598
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eights
MXY
ER
VN
INV
Pc
-0.0008
-0.0002
0.0004
-0.0008
-0.0006
-0.0006
0.0002
0.0021
-0.0018
0.0011
0.0002
-0.0013
-0.0005
-0.0002
0.0100
-0.0316
-0.0146
-0.0055
-0.0038
0.0090
0.0015
-0.0340
-0.0480
0.0380
-0.1357
0.0301
-0.0225
-0.0045
-0.0109
-0.0112
0.0104
0.0015
-0.0024
0.0023
-0.0005
0.0096
-0.0015
0.0115
0.0023
-0.0001
0.0021
0.0008
-0.0310
-0.1617
0.0399
0.0988
0.0417
-0.0452
0.0003
Pcd
Qc
T esting the N um ber o f C ointegrating Vectors

Null Hypothesis
II
r<1
r< 2
r<3
r<4
r<5
r<6
Trace
Statistic
Trace
(0.95)
Max X
Max X
(0.95)
148.680*
93.924
57.884
150.4
54.756*
36.039
24.383
21.015
8.152
50.51
44.37
38.22
33.500
117.49
88.59
63.66
12.485
4.332
42.7
25.64
3.823
12.34
0.509
0.509
31.99
25.68
19.21
12.34
319

Demand for Sorghum
A Seven-V ariable Ten Lag System U sing V I

Eigenvalues
0.3122
0.2203
MXY
ER
VI
INV
Ps
Psd
Os
-0.1576
-0.1425
0.5437
-0.7383
0.5262
-0.3429
1.0000
0.3636
-0.4954
MXY
ER
VI
INVI
Ps
Psd
-0.0003
-0.0009
0.1223
-0.2067
-0.0148
Qs
0.0129
0.0931
0.1259
0.0796
0.0300
0.0021
-0.8377
3.3252
7.3889
8.1246
0.4577
1.1488
-1.9248
-3.8282
6.4436
0.4094
0.4412
-13.193
-6.418
-4.106
5.3014
E igenvectors
1.3462
0.4436
0.1505
-2.4206
0.0191
-0.1365
0.3922
-3.7508
-0.2879
-1.4804
-0.6364
4.1565
-0.3562
4.4573
-2.7023
-3.6274
-3.036
-5.572
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
0.0023
0.0008
-0.0006
0.0010
-0.0001
-0.0009
0.0003
0.0014
-0.0002
0.0002
0.0866
0.1085
0.0005
0.0565
-0.0872
-0.0339
-0.0005
-0.0002
0.0104
-0.0029
-0.0140
-0.0011
-0.0028
0.0030
0.0011
-0.0054
0.1667
-0.0039
-0.0054
0.0455
-0.0152
0.0006
1.5540
W eights
-0.0107
-0.0101
0.0497
-0.0966
0.0010
0.0117
-0.0068
-0.0635
T esting the N um ber o f C ointegrating V ectors

Trace
Statistic
Trace
(0.95)
M AX (X)
II
o
Null
Hypothesis
175.709*
150.4
r< 1
r<2
r<3
107.979
62.944
117.49
88.59
63.66
67.730*
45.034*
r<4
r<5
r<6
38.591
20.905
5.894
0.385
42.7
25.64
12.34
M AX (X)
(0.95)
50.51
44.37
24.353
38.22
17.685
31.99
15.011
5.508
25.68
19.21
12.34
0.385
320
0.0000
0.0006
0.0066

Demand for Sorghum
A Seven-V ariable Ten Lag System U sing V2

Eigenvalues
0.2830
0.1938
0.1431
MXY
-0.1299
-0.1389
0.4121
0.2626
-0.2051
-0.2734
0.3113
DINV
Ps
0.5410
0.8675
0.5904
1.2886
0.0921
-1.5099
5.0023
-3.4069
-0.7583
-1.7240
-0.3994
6.3386
0.2252
Psd
-0.1322
0.7131
-7.1816
0.5547
-2.2829
-1.8006
0.0589
-1.6793
-2.4175
Os
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
MXY
ER
-0.0007
-0.0011
0.2772
-0.0023
-0.0003
-0.0004
0.0008
0.0002
-0.0002
0.0000
0.0000
0.1800
-0.0005
-0.0213
-0.0002
-0.0014
-0.2435
-0.0005
-0.1281
0.0016
0.0036
-0.2245
-0.0115
-0.0087
-0.0693
0.0006
0.0058
-0.0763
0.0129
-0.0084
-0.1557
0.0021
0.0010
0.0009
0.0028
0.0000
0.0000
0.0772
-0.0035
-0.0063
0.0163
Ps
Psd
Os
0.0101
-0.0523
-0.0491
0.0518
0.0199
-0.0043
0.0013
0.0876
0.0582
0.0257
0.0001
-1.5373
4.2194
23.0737
-0.5935
-0.1090
-0.1738
4.5946
-0.3373
-26.0202
E igenvectors
ER
V2
-12.3725
-3.3275
-4.1442
-4.2438
W eights
V2
DINV

Null
Hypothesis
r=0
r< 1
r<2
r< 3
r<4
r<5
r<6
Trace
Statistic
Trace
(0.95)
M A X (X )
MAX (k )
(0.95)
159.340*
99.128
150.4
117.49
60.211*
38.988
50.51
44.37
60.139
32.192
15.590
4.736
88.59
63.66
42.7
25.64
27.947
16.601
10.853
4.719
38.22
12.34
0.1706
0.1706
31.99
25.68
19.21
12.34
321
0.0002

Demand for Sorghum________________________________________________
A Seven-V ariable T en L ag System U sing VG
Eigenvalues
0.2752
0.1822
0.2340
0.1021
0.0785
0.0274
0.0065
E igenvectors
MXY
-0.769
0.611
0.385
-0.148
-1.120
-28.220
-1.655
ER
-0.168
-0.357
-0.388
0.675
-1.043
-12.170
4.663
VG
0.916
1.053
0.089
-0.804
0.210
9.015
1.929
INV
-0.516
0.361
-2.511
-1.009
1.221
-5.647
0.214
Ps
0.638
0.180
0.190
-1.314
-3.417
7.732
0.242
Psd
-0.241
0.401
-1.489
1.944
-1.828
7.591
-1.188
Qs
1.000
1.000
1.000
1.000
1.000
1.000
1.000
W eights
MXY
0.0004
-0.0019
0.0007
-0.0008
0.0002
-0.0003
0.0003
ER
0.0007
-0.0016
0.0002
0.0013
-0.0009
-0.0011
-0.0004
VG
0.0018
-0.0008
-0.0020
0.0003
-0.0010
0.0011
-0.0001
INV
-0.0826
-0.1750
-0.1440
0.0778
0.1610
0.0077
-0.0042
Ps
-0.0128
-0.0049
0.0038
0.0066
-0.0075
0.0017
0.0003
Psd
-0.0084
-0.0011
-0.0148
-0.0061
-0.0044
0.0025
-0.0011
Qs
0.0192
-0.0177
0.0495
-0.0552
0.0196
0.0095
-0.0139

Trace
Null Hypothesis
Statistic
Trace
(0.95)
MAX Q,)
M AX (X.)
(0.95)
150.4
58.267
50.51
r< 1
125.166*
117.49
44.004
44.37
r<2
76.911
88.59
36.415
38.22
r<3
40.496
63.66
19.484
31.99
r<4
21.011
42.7
14.793
25.68
r<5
6.217
25.64
5.032
19.21
12.34
1.185
12.34
II
183.433*
r< 6
1.185
322
Appendix 32 Maximum Likelihood Cointegrating Results: Mexicos Import

Demand for Sorghum
A Seven-V ariable Ten L ag System U sing VN
Eigenvalues
0.1071
0.1026
0.0300
0.0014
0.2133
0.2558
-0.3582
0.1601
0.0514
-1.0442
2.0914
4.3032
-77.8048
-13.6504
-30.3623
30.1112
3.9483
-6.9823
-20.0611
-29.0076
-2.2258
-0.8946
7.6610
-1.9824
1.8069
-1.4073
4.3531
-20.1605
1.0000
-0.9866
1.0000
0.8817
1.0000
12.2451
1.0000
-13.1376
6.7477
16.8814
1.0000
5.2322
0.1295
-3.1626
1.0000
0.0001
-0.0009
-0.0005
0.1531
-0.0104
0.0011
-0.0010
-0.0092
-0.0576
-0.0003
0.0004
-0.0004
-0.0016
-0.0002
0.0001
-0.2399
-0.0082
-0.0005
0.0002
0.0343
-0.0777
-0.0065
-0.0044
-0.1506
0.0031
-0.0002
0.0017
-0.0059
-0.0412
-0.0036
0.0237
0.0081
-0.0004
0.0074
-0.0004
0.0004
0.3272
0.1723
0.1290
Eigenvectors
MXY
ER
VN
INV
Ps
Psd
Qs
-0.1169
-0.1030
0.2514
0.5286
0.0904
-0.1414
MXY
ER
VN
INV
Ps
Psd
Qs
-0.0015
-0.0015
1.0000
-4.7967
8.5374
39.8987
W eights
-0.0311
-0.0138
-0.0069
-0.0019
0.0792
0.0029
0.0114
0.0098
0.0031
0.0012
T e stin g th e N u m b er o f C o in teg r a tin g V ectors
Trace
Trace
(0.95)
Max
176.848*
150.4
71.719*
50.51
117.49
105.129
70.890
88.59
45.882
63.66
42.7
25.371
25.64
5.768
0.257
12.34
34.238
25.008
20.510
19.603
5.5108
0.257
44.37
38.22
31.99
25.68
19.21
12.34
Null Hypothesis
r=0
r< 1
r<2
r<3
r<4
r<5
r<6
Max
(0.95)
323
0.0055

Demand for Milk Powder
A Seven-V ariable T en L ag System U sin g V I
E igenvalues
0.3641
0.2447
0.2337
0.1145
0.0364
0.0168
0.0074
E igenvectors
MXY
ER
-0.328
-0.079
-0.501
0.282
-4.829
3.983
5.234
-0.252
-0.203
1.301
0.091
-1.890
16.421
-2.443
VI
0.343
-0.145
-2.359
0.796
2.637
7.687
-0.002
MSF
Pm
-0.144
-0.196
-0.292
1.006
-4.142
-8.210
-3.613
1.248
-0.463
1.028
0.260
0.883
-3.923
2.790
Pmc
Qm
-0.924
-0.630
1.109
0.836
2.836
-0.544
0.788
1.000
1.000
1.000
1.000
1.000
1.000
1.000
W eights
MXY
ER
VI
MSF
Pm
Pmc
Qm
0.0003
-0.0024
-0.0007
-0.0003
0.0008
0.0004
0.0001
0.0007
-0.0010
0.0011
-0.0003
0.0013
-0.0009
-0.0001
0.0037
-0.0275
-0.0906
-0.0732
-0.0101
-0.0147
-0.0110
-0.0833
0.0213
0.0263
-0.0643
-0.0002
0.0028
0.0170
-0.1495
-0.0433
0.0389
-0.0031
-0.0016
0.0002
-0.0031
0.0226
-0.0305
0.0352
-0.0020
-0.0040
-0.0016
0.0004
-0.1621
0.1060
0.1587
-0.1666
0.0476
0.0549
-0.0441
Testing the N um ber o f C ointegrating V ectors

Trace
(0.95)
Max
Max
(0.95)
Trace
"t
II
Null Hypothesis
214.054*
150.41
81.937*
50.51
r< 1
r< 2
132.116*
117.49
40.803
44.37
81.313
88.59
38.189
38.22
r<3
r<4
33.123
63.66
22.002
31.99
11.121
42.71
6.708
25.68
r<5
4.412
25.64
3.059
19.21
r<6
1.353
12.34
1.352
12.34
324

A Seven-Variable Ten Lag System Using V2
Eigenvalues
0.3657
0.2544
0.2266
0.1549
0.0384
0.0172
0.0121
Eigenvectors
MXY
-1.245
0.958
-0.157
0.135
-3.049
6.880
8.964
ER
-0.455
-1.936
-0.002
0.383
-2.154
-5.784
10.904
V2
0.799
2.006
-0.556
0.292
3.762
-0.084
6.211
MSF
0.205
-0.408
-0.500
1.063
-3.169
0.018
-7.952
Pm
0.391
-3.808
0.112
-0.194
0.825
4.159
0.145
Pmc
-0.428
-2.059
-0.324
0.810
3.294
1.353
0.345
Qmk
1.000
1.000
1.000
1.000
1.000
1.000
1.000
Weights
MXY
ER
0.00001
-0.0004
-0.0026
-0.00001
0.0007
0.0000
-0.0003
0.0009
-0.0008
-0.0009
-0.0006
0.0014
-0.0001
0.0007
V2
-0.1918
0.3238
-0.1742
-0.1593
-0.0188
0.0572
0.0491
MSF
-0.1049
-0.0179
0.0102
-0.0705
0.0003
-0.0227
-0.0095
Pm
-0.1317
-0.0750
-0.0097
0.0041
-0.0002
0.0089
0.0035
Pmc
0.0259
-0.0303
-0.0181
-0.0153
-0.0069
0.0008
0.0029
Qmk
-0.1375
-0.0887
0.1919
-0.1948
0.0575
0.0755
-0.0318
Testing the Number of Cointegrating Vectors
Null Hypothesis
r=0
Trace
Trace
(0.95)
Max
M ax
(0.95)
224.952*
150.41
82.395*
50.51
r< 1
142.556*
117.49
53.140*
44.37
r<2
81.415
88.59
36.519
38.22
r<3
42.896
63.66
30.462
31.99
r<4
12.433
42.71
7.088
25.68
r< 5
5.345
25.64
3.140
19.21
r< 6
2.20
12.34
2.20
12.34
325
Apendix 35 Maximum Likelihood Cointegration Results: M exicos Import Demand

For Milk Powder
A S ev en -V a ria b le N in e L a g S ystem U sin g V G
E igenvalues
0.31825
0.23516
0.17705
0.14325
0.08801
0.01900
0.00045
E ig en v ecto rs
MXY
-0.7859
-0.1947
0.3016
-0.6852
0.0788
11.3519
3.2500
ER
-0.2103
-0.5928
0.0905
0.5487
-0.0682
5.6821
-7.7770
VG
0.3470
0.1212
0.1834
-1.2708
1.6822
-2.1786
-2.6896
MSF
-0.7775
0.0378
1.4856
-0.4007
-0.9771
-2.7457
-0.9965
Pm
1.0726
-2.3014
0.3950
-0.2947
0.1152
-1.0433
1.2879
Pmc
-1.4336
-1.1367
0.0966
1.0532
0.8276
-0.4895
1.9529
Qm
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eig h ts
MXY
0.00090
-0.00134
-0.00084
-0.00158
-0.00080
-0.00017
-0.00004
ER
-0.00006
0.00027
-0.00061
-0.00001
-0.00049
-0.00129
0.00008
VG
-0.00114
0.00138
0.00011
-0.00096
0.00346
0.00032
0.00002
MSF
0.02425
0.01683
-0.08250
-0.01240
-0.01971
0.01944
0.00312
Pm
-0.10933
-0.05432
-0.04126
0.00977
-0.00347
0.00664
0.00104
Pmc
0.02793
-0.03368
-0.00219
0.01791
0.01083
-0.00183
0.00058
Qm
-0.09139
0.05827
-0.32480
0.11897
0.01531
-0.00135
-0.01063
T estin g th e N u m b er o f C o in teg r a tin g V ec to r s
Trace
Trace (0.95)
Max
Max
(0.95)
II
o
Null Hypothesis
202.457*
150.41
69.722*
50.51
r<1
132.734*
117.49
41.001
44.37
r< 2
83.942
88.59
35.464
38.22
r<3
48.478
63.66
28.138
31.99
r< 4
20.340
42.71
16.767
25.68
r< 5
3.572
25.64
3.491
19.21
12.34
.081
12.34
r< 6
.081
* Denotes significante at 95% (Mackinon, 1999)
326
Appendix 36 Maximum Likelihood Cointegrating Results: Mexicos Import

A S e v e n -V a r ia b le N in e L a g S y stem U sin g V N
E ig en v a lu es
0.2480
0.1923
0.1885
0.1225
0.0420
0.0124
0.0056
-6.9873
-16.4540
3.5215
E ig en v ecto rs
MXY
-0.4054
ER
VN
MSF
-0.0900
0.5935
0.1149
-0.2082
0.0994
-2.7042
0.2995
0.0547
-31.8890
-2.0118
0.2774
-0.3438
-0.9747
1.5159
-9.4245
9.9088
-1.2855
-0.4434
0.7164
-1.9985
-1.2664
-8.0627
5.7002
-0.9992
Pm
1.0959
-0.2546
-6.1878
-0.5194
0.7021
5.4305
1.4688
Pmc
-1.3727
0.2660
-4.3487
1.0407
4.0610
5.2468
0.6190
Qmk
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
W eig h ts
MXY
0.0011
-0.0012
0.0008
0.0013
-0.0009
0.0003
-0.0001
ER
0.0003
-0.0014
0.0009
-0.0018
-0.0017
-0.0003
0.0000
VN
-0.0122
0.0001
-0.0142
-0.0174
-0.0024
0.0095
0.0013
MSF
-0.0652
0.0690
0.0445
0.0068
-0.0056
0.0064
-0.0173
Pm
-0.0934
-0.0157
0.0855
0.0178
-0.0003
0.0016
0.0037
Pmc
0.0301
0.0044
0.0339
-0.0148
0.0034
0.0015
0.0008
Qmk
-0.1286
0.3126
0.0849
0.1116
-0.0732
-0.0072
0.0325
T estin g th e N u m b er o f C o in te g ra tin g V ecto rs

Trace
Test
Trace
(0.90)
r=0
r< 1
163.616*
150.41
111.738
r<2
72.879
r<3
r<4
r<5
Null Hypothesis
M ax
Max
(0.95)
51.878*
50.51
117.49
38.859
44.37
88.59
38.004
38.22
34.874
63.66
23.790
31.99
11.084
42.71
7.804
25.68
3.279
25.64
2.262
19.21
12.34
1.010
1.010
12.34
r<6
327

Exchange Rate Movements and The Mexico-United States

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Exchange Rate Movements and The Mexico-United States

Uploaded by

Copyright:

Available Formats

EXCHANGE RATE MOVEMENTS AND TH E MEXICO-UNITED

STATES AGRI-FOOD TRADE: A COINTEGRATION ANALYSIS

In partial fulfilment of requirements

Jose Luis Jaramillo-Villanueva, 2006

395 W ellington Street

395, rue W ellington

The author retains copyright

L'auteur conserve la propriete du droit d'auteur

In compliance with the Canadian

Conformement a la loi canadienne

While these forms may be included

Bien que ces formulaires

Jose Luis Jaramillo Villanueva

1.1 The Economic Problem

1.2 Economic Research Problem

1.4 Specific Objectives

1.6 Contributions of this Research

1.7 Outline of the Study

2.2 Policy Reform in Mexico and the Agrifood-Sector

2.2.1 Transition Agricultural Policy

2.2.2 The Mexican Agricultural and Trading Sector

2.3 Literature Review on Exchange Rate Movements and Trade

2.3.1.1 Risk Aversion and Risk Neutrality

2.3.1.2 Non-Competitive Trading Firm

2.3.1.3 Forward Markets and International Trade

2.3.2 Empirical Studies on Exchange Rate Movements on Agricultural Trade

2.3.3 Modeling the Exchange Rate Volatility Movement on Agricultural Trade

2.3.4 Modeling Exchange Rate Volatility in Agricultural Trade

2.3.5 Exchange Rate Treatment in the Economic Model

2.3.6 Empirical Studies on Exchange Rate Volatility and Trade

2.3.6.1 Specification of the Volatility Measure

2.3.6.2 Nominal or Real Measures of Exchange Rate Volatility

2.3.6.3 Aggregate, Bilateral or Sector Trade Data

2.3.6.4 Estimation Technique Used

2.3.7 Empirical Studies on Exchange Rate and Mexican Agricultural Trade

3.2 Introduction to Exchange Rate Economics

3.3 Theories of Exchange Rate Determination

3.3.1 Flexible-Price Monetary Model

3.3.2 Sticky-Price and Real Interest Differential Monetary Models

3.3.3 Portfolio Balance Model

3.4 International Trade Models

3.4.1 The Hecksher-Ohlin Model of Trade

3.4.3 The Gravity Model

3.5 Exchange Rate and International Trade

3.5.1 Graphical Representation of a Trade Model

3.5.2 Algebraic Formulation of the Trade Model

3.6 Characteristics of the Commodity Industries in Mexico

3.6.1 Fresh Vegetables Industry and Exports

3.6.2 Grain Industry and Agricultural Policies in Mexico

3.6.3 Milk Powder Production and Policy

3.7 Economic Framework

3.7.1 Measures of Risk Aversion

3.7.2 The Theory of Input Demand Under Uncertainty

3.7.2.1 The Supply Function

3.7.2.2 The Import Demand Function

3.7.3 Derivation of the Import Demand and Export Supply

3.7.4 Comparative Statics for the Model

3.7.4.1 The Respond of Demand Function to Changes in Price of Inputs

3.7.4.2 The Respond of Import Demand Function to Outputs Price and

3.8 Empirical Estimation of Equations

FORMULATION OF THE EMPIRICAL MODEL

4.2 Unit Roots and Cointegration Analysis

4.3 Empirical Specification and Cointegration Analysis