Professional Documents
Culture Documents
Abstract
The objective of this paper is to examine the role of geography in explaining the
inflows using a modified gravity model. The fact that the significance of geographical
variables on FDI still remained even after controlling for the macroeconomic fundamentals, is
in contrast with the standard capital market model. The results can, however, be reconciled if
geographical factors can proxy for information costs, which may in turn explain why country
∗ Acknowledgements: This work was completed as part of the author’s PhD dissertation at Trinity
College Dublin. The financial support from Government of Ireland Research Scholarship in
Humanities and Social Sciences is gratefully acknowledged. The author is grateful to Philip Lane for
invaluable comments at various stages of this work and to the participants of International Economics
workshops in Trinity College Dublin, Spring Meeting of Young Economists 2002, and Annual Meeting
of LACEA 2002 for useful discussions. The author also thanks Shang-Jin Wei for kindly providing the
data on FDI policy and corruption, and to Philip Lane for providing the data on telephone call traffic
(funded by the European Commission as part of a Research Training Network).
1
1. Introduction
The objective of this paper is to examine the role of geography in explaining the
patterns of financial integration among both developed and developing countries. Financial
integration, that is, the integration of countries into international financial markets, has
become an important policy topic in the last few decades. This study examines the role of
geography in the spatial allocation patterns of foreign direct investment (FDI). As FDI has
become an important source of foreign finance, its patterns of allocation can indicate whether
geography matters for a host country to attract financial flows, hence its level of financial
integration.
Foreign direct investment flows have been growing faster than world GDP and
merchandise trade, and the share of foreign direct investment in total private capital flows to
the developing countries has substantially increased. Total net inflows of FDI into developing
countries were US$187 billion in 1997, US$188 billion in 1998, US$222 billion in 1999, and
US$240 billion in 2000 (UNCTAD 2001). Like trade in goods, trade in assets is primarily a
two-way flow between industrial countries. Of total world FDI inflows in 1999-2000, 80 per
cent took the form of two-way FDI flows among developed countries, 17.9 per cent accounted
for FDI inflows to the developing countries. Among the developing countries, only a few
countries in Latin America and East Asia attract the majority of total developing country FDI
flows, while Sub-Saharan Africa performs poorly. Out of US$240 billion in 2000, US$134
billion was invested in Asia and US$95 billion was invested in Latin America. The
distribution of total world FDI inflows among developing countries emphasizes the strong
spatial concentration of FDI inflows: 9.2 per cent was invested in Asia, 7.9 per cent in Latin
The stylized fact that bilateral FDI flows, especially between developed and
2
is in contrast with portfolio diversification theories and with the neoclassical model. 1 The
forces of financial integration, like the forces of global economic integration, are driven by
opportunities and more efficient allocation of capital, and ultimately, increased economic
international financial markets. The ‘home bias’ literature argues that there is considerable
market frictions caused by imperfect information that the countries’ foreign asset portfolios
are biased towards the domestic market and not optimally diversified (French and Poterba
1991: 222-226; Lucas 1990: 92-96; Tesar and Werner 1995: 467-492). Lane (2004)
emphasizes the impressive pace of financial globalization, but argues that behavioural and
informational barrie rs remain significant impediments to a fully unified global capital market.
In this empirical study, we examine the determinants of bilateral FDI flows using a
modified gravity model. Our objective is to answer two questions. First, what host country
characteristics attract FDI flows? Second, which group of flow is more sensitive to changes in
a specific host country characteristic? The first question deals with the direction of impact.
North-North, North-South and South-North FDI flows may all be attracted by similar host
country characteristics. The second question deals with the sensitivity of different groups of
flows to host country characteristics. For example, although both North-North and North-
South FDI are attracted to more developed markets, are North-North flows more sensitive to
changes in income? Among many variables commonly used as the determinants of FDI flows
empirical evidence on the role of geography on FDI is discussed. Section 3 introduces the
benchmark model, explanatory variables and the data. Section 4 reports results from pooled
1
Zebregs (1998) examines the distribution of FDI flows across developing countries and finds that
standard neoclassical models can not explain the concentration of FDI flows in a handful of developing
countries.
3
North FDI inflows. In Section 5, various sensitivity tests are reported on the robustness of the
limited compared to international macroeconomics and trade, but fast growing in number (see
for example Ghosh and Wolf 1998; Eaton and Tamura 1994: 478-510; Di Mauro 2000;
Hausman and Fernandez-Arias 2000; Chunlai 1997; De Menil 1999: 165-201; Wei 1997,
2000: 303-354). Portes and Rey (2000) and Portes et al. (2001: 783-796) use a gravity-style
approach on a panel data of bilateral equity flows and bonds respectively. Their findings
indicate that gross assets flows are negatively correlated with the distance between the two
countries. They interpret the negative sign for distance variable as the existence of
information costs for equity transactions as part of trading costs. Loungani et al. (2002: 526-
543) question why gravity models work for asset flows and attempt to explain the “distance”
puzzle. They conclude that the “distance” puzzle can be reduced by going beyond
There has been an exponential growth on studies examining the role of geography on
bilateral trade flows, especially since the introduction of the gravity model. Such models have
had considerable empirical success in explaining bilateral trade flows, and are also
theoretically well established (see e.g., Anderson 1979:106-116; Bergstrand 1985: 474-481;
Helpman and Krugman 1985; Deardorff, 1995). Tinbergen (1962) was among the first to
present such models. He presented a model in which the volume of trade between any two
countries is determined by gross national product (GNP) in home and host country, and the
Borrowing the gravity model of international trade in this study to explain financial
flows can be based on two observations. First, countries with relatively high bilateral trade
shares tend to have relatively high bilateral FDI shares. Ghosh and Wolf (1998) argue that
4
trade linkages not only lead to short-term financial linkages through trade financing, but also
in the long-term, create demand and supply channels through which domestic firms can
access foreign capital. In other words, trade linkages provide information about the host
countries through growing familiarity, and making it less costly to invest directly in the same
countries.2 Calvo and Mendoza (2000: 79-113) argue that fixed costs are important in
international investment decisions and several authors emphasize at least part of the
information cost is fixed (Ghosh and Wolf, 1998).3 Goldstein and Razin (2002) argue that
there are some fixed set-up costs to investing directly, such as costs of information acquiring.
Bilateral trade linkages may reduce such costs for direct investment.
Second, it is commonly observed that the familiarity effect causes investors to favour
close-by countries with similar characteristics and legal systems over more distant and
institutionally different countries. Senior (1850), in his Political Economy, stresses the
distance, differences in political institutions, language, religion and social customs as barriers
to capital flows. In other words, location matters for financial flows as well as for real
activity. As the distance to the host country increases, the familiarity may decrease, with the
exception of countries that share common historical past (i.e., colonial ties). Another effect of
distance specific to FDI is in relation to control. Distance can reduce investors’ control. FDI is
defined as capital invested with long-lasting interest in an enterprise. Therefore, the concept
of control is inherent in this type of investment. In such a case, distance would play a
significant role in determining which host countries receive the majority of bilateral FDI
flows. On the other hand, international investors may choose FDI over other forms of
investment based on the reason that it gives them control. In that case, investment in more
distant locations through FDI may be preferable to, for example, portfolio investment. By
2
Honohan and Lane (2000: 12-23) show that Ireland’s portfolio investment follows closely its trade
patterns.
3
Hobson (1914) argues that “the greater the amount of capital invested abroad, the easier it is to obtain
information relevant to investments abroad”.
5
being closer to the market, investors’ control is maximized and information asymmetries are
minimized.
the international trade literature, the distance variable is used to proxy for transport costs,
implying reduced trade with increased distance. In this study, we argue that, for FDI flows
distance may be a proxy for information costs, rather than transport costs. The cost of
information gathering would likely increase with distance, as familiarity with the host
country’s investment opportunities, customs and culture decreases. Most North-North FDI
flows are in the form of mergers and acquisitions, implying no physical transfer of assets
except in the form of human capital. 4 The transfer of human capital, may in turn be negatively
affected by distance, as various studies on international migration find that distance affects
costs of migration through increased travel costs, difficulty to get back home and increased
cultural and linguistic differences (see e.g., Karemera et al. 2000:1745-1755; Ximena et al.
2002).
The primary contribution of this study is to examine the determinants of bilateral FDI
flows with a specific focus on geography, and to separate bilateral flows into North-North,
between different groups of FDI flows. As such, the results of this study will contribute to
international finance literature, by incorporating the geography factor in explaining the spatial
concentration of financial flows. Portes and Rey (2000) and Portes et al. (2001: 783-796) have
shown that distance is a significant determinant of bilateral cross-border equity flows and
bonds. Our study examines whether distance, among other geography variables, is also a
there are various factors at play that affect information costs differently for North-North, than
North-South or South-North FDI, and hence the potential effect of geographical variables on
FDI. This study is closest to Loungani et al. (2002: 526-543), where the authors investigate
4
See World Investment Report 1998 (UNCTAD).
6
the “distance” puzzle comparing trade and FDI. Our study further questions the negative
effect of distance on FDI by introducing other geographical variables and also checks the
reducing factors.
Second, this study will have implications on both theory and policy. Horizontal FDI
is motivated by market access and empirical tests have shown that two-way (North-North)
FDI flows among industrial countries can be significantly explained by these models. On the
other hand, FDI between developed and developing countries (North-South) are better
explained by vertical models of FDI as vertical FDI is motivated by cost considerations (Shatz
and Venables, 2000). If indeed, bilateral FDI flows are driven by motives of market access or
cost incentives, the geographical factors should be less significant than size or level of
development of the host country or not significant at all. Ultimately, our results have policy
implications. If geographical variables can explain the patterns of bilateral FDI, financial
The foreign direct investment data are obtained from the OECD’s International Direct
Investment Database. The OECD’s International Direct Investment Database (1999) provides
data on bilateral inflows and outflows of FDI over the period 1980-1998. Each OECD
member country reports bilateral “outflows to” and “inflows from” other member and a
number of partner countries. All values were originally expressed in reporting countries’ own
national currency units, which were then converted into constant 1987 US$ using OECD’s
yearly average exchange rates and US gross domestic product deflator. We categorize all
OECD member countries (excluding Korea, Mexico, Turkey and recently joined Central and
Eastern European countries) as North, and the rest as South. There are 20 industrial countries
in our sample categorized as ‘North’ and 28 developing countries (see Appendix A for list of
7
countries in sample). Other data on the explanatory variables come from various sources (see
Appendix B).
inflows and is estimated by OLS on pooled time-series cross-section data. To answer the
question what characteristics make a host country attractive to FDI inflows, we specify the
following model:
The independent variable, Inflwijt is the ratio of gross bilateral FDI inflows/GDP to
host country i from partner country j at time t.5 In our benchmark regression model, the
explanatory variables are POPit , population of host country i at time t, GDPPCit,, GDP per
capita of host country i at time t, DISTij , distance between country i and j, Zij , a set of control
variables, tt , time dummies, d j , source country dummies, and eijt , is the error term, corrected
While time dummies control for the effect of common global events, source country dummies
control for such source country characteristics that may affect outward FDI. Although the
model is estimated by OLS on pooled time-series cross-section data, it is, in a sense, a quasi-
fixed effects model because some of the unobserved heterogeneity is controlled by source
country dummies. The scale variable (POP), GDP per capita and distance variables are in
logs. The coefficients for the variables in logs can, therefore, be interpreted as semi-
elasticities.7
5
For North-North FDI inflows, the set of source and home countries are identical. Only this group
suffers from mirror statistics discrepancies where the same data is reported by two different sources
(e.g. US reporting inflows from Japan, and Japan reporting outflows to US). For this reason, North-
North FDI flows are calculated as the average of the two values reported by the source and host
countries.
6
Instead of annual data, 5-year averages are used to avoid autocorrelation. Time dummies are for
t1 =1980-84, t 2 =1985-89, t 3 =1990-94, t 4 =1995-98.
7
Alternatively, a double-log specification could have been applied. However, this would mean losing
considerable efficiency since quite a number of negative observations would have been dropped.
8
The set of control variables that are included in the regression analysis are mostly
dummy variables specific to the host-source country pair: BORij and LANGij take the value of
one if the host and source country share a common land border, or a common language, zero
otherwise. Additional explanatory variables include, LATi latitude of the capital city of the
host country i, LANDi dummy, that takes the value of one if the host country has no access to
any seas and oceans, or navigable waters, zero otherwise, RTAij dummy, that is one if the host
and source country are in a regional trade agreement, zero otherwise, CUij dummy, that takes
the value of one if the host and source countries are in a currency union, zero otherwise.
The primary focus is on the geography variables: distance, border dummy, latitude
and landlockedness. The expected sign for distance is negative, implying that as distance
increases information costs will also increase and this will have a negative effect on the
investment. Border is expected to have a positive impact, implying that for a given distance,
countries that share a common border see higher flows.8 Landlockedness is also commonly
used in gravity models. Henderson et al. (2000) argue that being landlocked raises transport
costs by more than 50 per cent. The latitude variable is used in the geography of economic
development literature to examine cross-country inequality of per capita income levels (see
e.g. Gallup et al. 1999). In our study, latitude is used as another proxy for remoteness of the
host country.
Regional Trade Agreements (RTA) and currency union dummy variables can be
transaction costs. Being in a currency union has been shown to have a sig nificant effect on
bilateral trade by Rose (1999). The effect of sharing a common currency is significant even
when exchange rate volatility is taken into account. In this paper, the currency union dummy
Occasionally, there are negative FDI inflows into a host country, which may suggest that the source
country is repatriating earnings back home. In order to avoid losing these observations, the dependent
variable is expressed as a percentage throughout the regression analyses.
8
Parsley and Wei (2000) argue that the national borders have a negative effect on market segmentation
(i.e. the border effect), which may explain the home bias in goods and asset trade.
9
captures a long-term government commitment to integration and decreased exchange rate
uncertainty, and possibly increased financial integration. Common language may also be
grouped together with integration variables, since especially for North-South category, this
Size and the level of income are reported as important host country determinants
especially for market seeking FDI (UNCTAD 1998). As well as a large market to serve to, a
larger economy may also present a more diverse opportunity for investment. For financial
flows, host country size may reflect greater fixed costs of compliance with local regulations.
By GDP per capita variables, we measure the level of development of the host country.
South and South-North FDI inflows are reported in Table 1. This table allows for a
heteroskedasticity-cons istent standard errors are calculated for all regressions. Time and
[Table 1]
Overall, only the geographical factors play a significant role in determining North-
North and North-South FDI inflows. Some results are worth highlighting: (i) distance is
statistically significant even after we control for integration and common language dummy
variables indicating that there are information costs between North and North, and also North
and South countries. For the FDI inflows among developed countries, with largely liberalized
capital markets, advanced infrastructure for communication and high shares of trade, it may
be surprising to see that there are negative effects from distance. This result can be
rationalized if information costs can be internalized to distance, which can in return explain
why country portfolios are still home-biased and financial markets are segmented. For the
9
We have tested the poolability of the cross-sections over the time-series dimension both by examining
individual sub-periods and also by a Chow test.
10
North-South inflows, the geographical factors are economically more significant, indicating
that there are higher information costs between developed and developing countries. (ii) For
North-North FDI, size of the host country is not a significant determinant, whereas for North-
South FDI it is. This result does not support the a priori expectation that the majority of
North-North FDI is horizontal and North-South FDI is vertical. Market size is important in
horizontal FDI. Shatz and Venables (2000) find that FDI in high-income countries (i.e. North-
North) is overwhelmingly horizontal, involving production for sale to the host country
of intermediate stages of the production process. (iii) For North to North bilateral FDI
inflows, familiarity (i.e. sharing a common language) is economically the most significant
determinant. This may explain why UK and US two-way FDI flows dominate the financial
markets. For North to South FDI inflows, although distance is economically more signific ant
compared to FDI between industrial countries, still integration variables matter more.
Our model explains 32 percent of the variation in North-North FDI flows, 73 percent
of North-South and only 7 percent of South-North FDI inflows.10 The poor fitness of our
model to explain South-North FDI flows can be the effect of outliers in the sample. This
result is not surprising considering that majority of South-North FDI is carried out by such
financial centers as Hong Kong, Singapore and Panama. Equity investment in excess of ten
stocks or shares by South financial centers may be driving these results. To formally approve
that these differences are, in fact, statistically significant, we tested the hypothesis that the
coefficients for each group are the same (i.e. ßNN = ßNS = ßSN). The results suggest that, indeed,
the difference in the coefficients for size, GDP per capita and distance are statistically
10
The adjusted R2 decreases to 16 percent for North-South FDI when the currency union dummy is
excluded. Some of the estimated coefficients also increase in economic significance.
11
5. Robustness Tests
In this part of the analysis, a number of additional control variables are entered in
order to check for the robustness of our results. These variables have been found to be
significant determinants of FDI flows, or other types of financial flows, in other studies. Since
the data for these variables are available for much smaller samples (or for only a few years),
their relationship with FDI is examined as robustness tests of our benchmark model.
In Tables 2a and 2b, we examine the effect of government policy (incentives and
restrictions) toward FDI and corruption on FDI flows. While FDI specific incentives and
restrictions of the government policy indicates the general openness and willingness of the
host economy to receive FDI, together with measures of corruption in the host economy, these
variables can proxy for investment costs. The relation between corruption and FDI is
extensively investigated by Wei (1997, 2000: 303-354). In his studies, he finds a statistically
significant, negative and robust relationship between corruption and FDI. Table 2a presents
results from cross-country OLS regressions where average FDI/GDP over 1995-1998 was
regressed on the set of control variables together with two separate indices of corruption and
dummies for government incentives and restrictions. Both the Transparency International (TI)
and Global Competitiveness Report (GCR) indices are expected to have a negative effect on
FDI inflows since higher values indicate higher corruption. The dummies for government
incentives and restrictions range from 0-4, with zero indicating no incentives or no restrictions
and four indicating all incentives/restrictions. Overall, there is no evidence that corruption has
a negative effect on FDI inflows. Both TI and GCR indices are statistically insignificant for
all groups.
[Table 2a]
For comparison, in Table 2b, we test the effect of corruption and FDI policy on FDI,
only this time the dependent variable is the logarithm of FDI inflows. The results are in line
12
with results from Wei (1997, 2000: 303-354). Both TI and GCR corruption indices are
inflows. One unit increase in TI corruption index indicates that host North country receives
18.9 percent less FDI from North source countries and 31.6 percent less FDI from South
source countries.11 GCR index indicates a larger impact of corruption on both North-North
and South-North FDI. There is also evidence that FDI incentives are statistically significant
and positive for North-North FDI (TI index) and for both North-North and South-North (GCR
index).
[Table 2b]
In summary, the results from Table 2b concur the results of Wei, indicating that
corruption has a negative effect on the level FDI inflows but not on the ratio of FDI to GDP.
It is imperative to highlight (i) both in Table 2a and 2b corruption indices and FDI incentives
and restrictions are statistically not significant for North-South FDI inflows, contrary to
general concensus12 (ii) and in both tables, North-South distance and latitude are statistically
significant and robust to the inclusion of these variables. In general, these results indicate that
geographical variables are not robust to the inclusion of corruption and FDI policy variables
for FDI inflows among the developed countries (North-North). Once investment costs are
controlled by these proxies, information costs are no longer significant determinants of North-
In Table 3, the effect of sharing a common legal code is examined. La Porta et al.
(1996) show that different commercial laws have different levels of protection of corporate
shareholders and creditors, and quality of law enforcement. In other words, the financial
development of a country, and ultimately its growth, may be determined by the origins of its
11
Exp (-0.21) – 1 = -0.189; Exp (-0.38) – 1 = -0.316.
12
Local government corruption is one of the important negative factors that affect foreign investment
in developing countries.
13
commercial codes. Feldstein (2000) also argues that global integration of capital markets and
increased volume of FDI spread US-UK forms of corporate culture. There is a growing body
of literature that suggests Anglo-American law (i.e. common law) improve the efficiency with
which capital is invested. Beck et al. (2004) find that a country’s legal origin influences its
firms’ access to foreign finance, and that the firms in countries with French legal origin face
significantly higher obstacles in accessing external finance than firms in common law
countries. In our sample, the countries are categorized into sharing four different legal codes:
English (i.e. the common law), German, French and Scandinavian (i.e. Roman law).
[Table 3]
The common legal code dummy has the expected positive sign for North-North and
South-North FDI inflows, but it is only significant for North-North flows. The host North
countries that share the same legal code as the source North countries receive 0.04 percentage
point more FDI inflows. This result may explain large absolute amounts of two-way FDI
flows between the US and UK. Including legal code as a control does not increase the
explanatory power of the model. Distance and other control variables are robust to the
inclusion of common legal code dummy for all groups. 13 Sharing a common legal origin
provides familiarity with the host country institution, and also reduces costs of investment.
North-South FDI inflows, contrary to North-North FDI, are not affected significantly by
Table 4 presents results of sensitivity checks by including the bilateral trade flows as
a control: trade has been found to be a significant determinant of FDI flows and is highly
correlated with distance and other explanatory variables. Omitting this variable from the
equation would cause the effect of this variable to be wrongfully attributed to distance (and
13
In further detail, when the effect of common law versus Roman law is examined, clear differences
appear. North host countries that have adopted common law receive more FDI inflows from other
North countries, whereas South host countries receive less FDI inflows, compared to host countries that
are of Roman law origin.
14
other variables). Trade is statistically significant for North-North and North-South FDI and it
is positive. This indicates that controlling for all other factors, both North-North and North-
South FDI increase as bilateral trade increases. In other words, trade encourages FDI,
The economic impact of trade (i.e. trade coefficients) on North-North and North-
South FDI are similar. The positive relationship between trade and FDI may indicate to the
On the other hand, the positive relationship for North-South FDI and trade may be driven by
the role of trade in creating familiarity with the host market, and thereby facilitating access of
[Table 4]
Both for North-North and North-South, the distance is still a significant determinant
of FDI although the semi-elasticity of distance decreases, especially for North-North FDI,
when trade is included. This indicates that a large part of the negative effect of distance on
North-North FDI flows can be attributed to trade. Although the semi-elasticity of distance is
reduced, it is not eliminated, suggesting that partly distance does proxy information costs.
Size and GDP per capita become statistically significant for all when trade is introduced to the
model, suggesting that size and level of development do have a statistically significant
relation with FDI independent of their relation through trade. When trade is included in the
model, the R2 s remain similar to the results from Table 1, indicating that the explanatory
5.4. The Effect of Actual Information Flows on FDI: Telephone Call Traffic
Table 5 summarizes the results of robustness checks by the inclusion of telephone call
traffic. By introducing this variable, measured in minutes from country j to i over 1983-95,
we augment the distance variable. Telephone call traffic is a direct measure of information
flows between countries, which has been used in studies by Portes and Rey (2000), Portes et
15
al (2001: 783-796) and Loungani et al. (2002: 526-543) and shown to have a significant and
positive relation with portfolio equity and FDI flows. We expect that the effect of telephone
call traffic be positive on bilateral FDI. The more telephone calls (in minutes) (i.e. the more
[Table 5]
The results indicate that when telephone call traffic is entered into the equation, the
distance variable becomes insignificant for North-North FDI. This might mean that distance is
a proxy for the ability to communicate and actual flows of information for North-North FDI.
In contrast, North-South distance variable is robust and economically more significant even
when telephone call traffic and trade variables are included in the model. This indicates that
for North-South FDI, there are significant information costs and these costs have a negative
effect on FDI. In fact, the explanatory power of the model also increases to 82 percent (Table
5) from 74 percent (Table 4). One problem with the telephone call traffic variable is that it is
Telephone call traffic is negatively correlated with FDI for North-South, and this
result is robust to the exclusion of outliers. As the telephone call traffic decreases, FDI for the
South host country increases, controlling all other factors. One explanation for this may be the
control argument. The investors may prefer FDI over other types of investment in distant
countries where information flows are obstructed.15 By being on location, any potential
information asymmetries are minimized. This result may also be explained by the personal
calls made by emigrants to their families back home. Telephone call traffic variable does not
14
Telephone call traffic variable is unlikely to be a good approximation for telephone charges.
15
Loungani et al. (2002: 526-543) have shown that information infrastructure can also play a
significant role in reducing the distance puzzle.
16
Workers’ remittances can be used as a proxy for personal phone calls; however, it is not available in
bilateral form. In aggregate form, Mexico, Turkey and Brazil are among the highest receiver of
workers’ remittances and also telephone calls (from developed countries).
16
5.5. Other Robustness Tests
As part of our robustness tests, we have also examined the sensitivity of our results to
i) outliers ii) zero observations iii) US dollar movements iv) South sample excluding Israel,
Korea, Hong Kong, Chinese Taipei and Singapore v) the inclusion of a remoteness measure
17
The outliers from Table 4 were excluded from the regression analysis. The
statistical significance of results for North-North flows remains mainly unchanged. 18 The
economic significance of all variables declines. The R2 for North-North FDI improves to 44
percent when the outliers are excluded. The results of North-South FDI flows are generally
robust to the exclusion of outliers; however, they all decline in economic significance.19 The
R2 for North-South FDI is 36 percent compared with 16 percent when US-Panama country
pair is dropped from the sample. For the South-North FDI flows, trade, common border and
landlocked dummy become significant.20 The variables that were significant previously
decline in economic significance. The R2 for South-North FDI is 15 percent when outliers are
Next, the effect of zero observations on the results has been examined. 21 Since our
left-hand side variable is a ratio, not in logs as in most other studies, zero observations are not
dropped out. The regression equations in Table 1 are rerun excluding zero observations. The
single change in the results is that GDP per capita becomes a significant determinant of
17
The residuals that are beyond two standard deviation of the regression are classified as outliers.
18
The outliers for North-North flows include flows to Australia from Japan and US, to Austria from
Germany, to Belgium-Luxembourg from France, Germany, Italy, Japan, Netherlands, and US, to
Canada from US, to Finland from Sweden, to Netherlands from Japan, UK and US, to New Zealand
from Australia, UK and US, to Norway from UK, to Sweden from Finland and to Switzerland and UK
from US.
19
The outliers for North-South flows include flows from Germany to Argentina, from Japan to Hong
Kong, Malaysia and Singapore, from Switzerland to Panama, and from US to Venezuela, Costa Rica,
Chile, Panama and Singapore, and from UK to Malaysia and Singapore.
20
The outliers for South-North flows include flows from Singapore to Australia, Belgium-
Luxembourg, and New Zealand, from Indonesia to New Zealand, from Mexico to Belgium-
Luxe mbourg, from Hong Kong to Belgium-Luxembourg, New Zealand, and from Malaysia to New
Zealand.
21
Zero observations indicate to lack of FDI flows, not to missing data.
17
We have also tested the sensitivity of our results to US dollar movements. As a
robustness measure, we have introduced the US dollar real exchange rate relative to the
Special Drawing Rights (SDR).22 The movement of the dollar with respect to the SDR is one
way of measuring the stability of the dollar (Mundell 2002). The benchmark model in Table 1
and its results are robust even when we control for US dollar movements. For all groups, the
statistical and economic significance of our variables remained unchanged. The reason for
this may be because the effect of large US dollar movements was already controlled for in our
model by time dummies. Some year dummies were also highly collinear with our US/SDR
excluding five countries from the original sample: Israel, Korea, Hong Kong, Chinese Taipei
and Singapore. These countries indeed have achieved high rates of growth in the last decade,
with well functioning market economies.23 The statistical significance of results on the
geography variables are unchanged both for the flows from North to South and from South to
North.
further check the robustness of our geography variables.24 The distance variable is statistically
robust to the inclusion of this variable, but its economic significance is reduced in the North-
North and North-South samples. For the South-North sample, distance becomes statistically
significant at ten per cent and has the expected negative sign. Remoteness variable itself is
22
The US/SDR end-of-period exchange rates are from the IMF’s International Financial Statistics (IFS)
database. The US price level is measured by the US GDP deflator and the foreign price level is
measured by the SDR deflator for 1980-1998, calculated by the author using changing weights of G5
currencies in the SDR basket.
23
Israel is classified as an advanced economy by the IMF and the World Bank. Korea and Singapore
meet most benchmarks of a developed country but resist being classified as such. Chinese Taipei
(Taiwan) and Hong Kong are considered developed by some organizations. However, China, a
developing country claims the land of the first and exercises sovereignty over the latter.
24
Latitude variable is excluded in this specification.
18
Finally, sensitivity of the results to method of estimation was checked by using a
fixed-effects model instead of pooled OLS. Broadly, the results are the same. The details of
the model and results from a two-step fixed-effects model can be obtained from the author
6. Conclusion
In this paper we investigated the determinants of FDI flows with an emphasis on the
role of geography to highlight the factors that affect the spatial allocation of FDI. The
objective of this paper was to examine the role of geography in explaining the patterns of
financial integration among both developed and developing countries. The results indicate
that standard gravity and other control variables have the same impact on each bilateral flow.
The differences arise in the magnitude of the economic effect of these variables. The results
from our benchmark model indicate that geographical factors are significant and robust
It must be highlighted that our results contradict Loungani et al. (2002: 526-543) who
find that the distance puzzle remains even when the actual information flows are accounted
for (by way of including the volume of tele phone traffic). We observe two distinct results for
the North-North and North-South FDI flows. We find that the negative effect of physical
distance (distance puzzle) is no longer significant when we account for the volume of
information flows (i.e. the telephone traffic) and trade between two developed countries. In
contrast, the negative effect of distance is strengthened when the same variables are accounted
for for North to South FDI flows. This, we conclude, implies that there are significant
information costs as part of transaction costs to direct investment between North and South
countries, which in turn may explain why financial integration has been restricted to a few
spatial allocation of FDI. The fact that the significance of geographical variables on financial
19
flows still remained even after controlling for the macroeconomic fundamentals, is in contrast
with the standard capital market model. The results can, however, be reconciled if
geographical factors can proxy for information costs. Further research is required to establish
through which channels the effect of geography works on information costs. Especially, the
distance on financial flows, may also explain how idiosyncratic shocks are spread (i.e.
contagion) to other countries in the same region. If country-specific shocks in one country
result in reversal of capital flows (i.e. net capital outflows) or sudden stops, even if the
the government (i.e. a first-generation model), investors’ lack of confidence may bring on
integration in to the world economy. The results from this study suggest the existence of
significant information costs in financial flows, which may explain why country portfolios are
still home-biased and financial integration is restricted to developed countries and only a few
developing countries. South host countries that have a greater distance to investor countries,
flows.
APPENDIX A
NORTH COUNTRIES: Australia, Austria, Belgium-Luxembourg, Canada, Denmark, Finland, France,
Germany, Iceland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden,
Switzerland, United Kingdom, United States
SOUTH COUNTRIES: Korea, Mexico, Turkey, Algeria, Egypt, Morocco, South Africa, Argentina,
Brazil, Chile, Colombia, Costa Rica, Panama, Venezuela, Kuwait, Saudi Arabia, United Arab Emirates,
Iran, Israel, China, Chinese Taipei, Hong Kong, India, Indonesia, Malaysia, Philippines, Singapore,
Thailand
20
APPENDIX B
LIST OF VARIABLES
Variable Definiti on Source
FDI/GDP FDI Inflows (US$ millions) FDI Inflows – OECD Direct
Investment Statistics, 1999.
References
Beck, T., Demirguc-Kunt, A. and R. Levine (2004). Law and Firms’ Access to Finance.
World Bank Policy Research Working Paper 3194, World Bank, Washington DC.
Calvo, G. and E. Mendoza (2000). Rational Contagion and the Globalization of Securities
∗
The unilateral trade agreements included in sample: EC/Algeria (1976), EC/Egypt (1977), EC/Iceland
(1973), EC/Israel (1975), EC/Morocco (1976), EC/Norway (1973), EC/Turkey (1995), EFTA/Israel
(1993), EFTA/Turkey (1992), US/Israel (1985).
21
Chunlai, C. (1997). The Location Determinants of FDI in Developing Countries.
De Menil, G. (1999). Real capital market integration in the EU: How far has it gone? What
will the effect of the euro be? Economic Policy 28 (April): 167-201.
Gravity Approach. CEPS Working Document 156. Centre for European Policy
Studies, Brussels.
Eaton, J. and A. Tamura (1994). Bilateralism and Regionalism in Japanese and US Trade
Feldstein, M. (2000). Aspects of global economic integration: Outlook for the future.
Cambridge, Mass.
Gallup, J.L., Sachs, J.D. and A. Mellinger (1999). Geography and Economic Development.
Mass.
Ghosh, S. and H. Wolf (1998). The Geography of Capital Flows. mimeo, New
York University.
Goldstein, I. and A. Razin (2002). Volatility of FDI and Portfolio Investment: The Role of
22
Research Department Working Papers 417, Inter-American Development Bank,
Washington DC.
Hobson, C. K. (1914). The Export of Capital. Constable and Company Limited, London.
Honohan, P. and P. Lane (2000). Where do the Irish Invest?. Irish Banking Review
Autumn: 12-23.
Lane, P. (2004). Global financial trade: How far have we come? IIIS Discussion
La Porta, R., Lopez-De-Silan, F., Schleifer, A. and R. Vishny (1996). Law and
Cambridge, Mass.
Loungani, P., A. Mody and A. Razin (2002). The Global Disconnect: The Role of
Lucas, R. E. (1990). Why doesn’t capital flow from rich to poor countries? AEA Papers and
Mundell, R. (2002). Commodity Prices, Exchange Rates and the International Monetary
System. http://www.fao.org//docrep/006/y4344e/y4344e04.htm.
Parsley, D. and S. Wei (2000). Explaining the border effect: The role of exchange rate
variability, shipping costs and geography. NBER Working Paper 7836. National
23
Portes, R.and H. Rey (2000). The Determinants of Cross-Border Equity Flows.
CEPR Discussion Paper 2225. Center for Economic Policy and Research, London.
Portes, R., Rey, H. and Y. Oh (2001). Information and capital flows: The
45 (4-6): 783-796.
Rose, A. (1999). One Money, One Market: The Effect of Common Currencies on
Trade. CEPR Discussion Paper 2329. Center for Economic Policy and Research,
London.
Senior, N. W. (1850). The Political Economy. Richard Griffin and Company, London.
Shatz, H. and A. Venables (2000). The Geography of International Investment. World Bank
Tesar, L. and I. Werner (1995). Home bias and home turnover. Journal of
Wei, S. (1997). How Taxing is Corruption on International Investors. NBER Working Paper
Wei, S. (2000). Local Corruption and Global Capital Flows. Brookings Paper on
Come from, and Why? NBER Working Paper 8998. National Bureau of Economic
Zebregs, H. (1998). Can the neoclassical model explain the distribution of foreign
direct investment across developing countries? IMF Working Paper 139. International
24
Table 1: Pooled Least Squares Estimation of Bilateral FDI Inflows a
N-N N-S S-N
Ln(POP) of host -0.004 -0.06 -0.002
(-1.04) (-2.60)*** (-2.01)**
Ln(GDP-PC) of host -0.002 -0.04 -0.002
(-0.20) (-1.45) (-1.52)
Ln(DIST) -0.09 -0.14 -0.0008
(-7.80)*** (-3.29)*** (-0.75)
Border -0.05 -0.04 0.002
(-1.63)* (-0.14) (0.51)
Language 0.18 0.14 0.009
(4.98)*** (2.98)*** (2.13)**
Latitude -0.002 -0.002 -0.0001
(-3.85)*** (-3.37)*** (-1.56)
Landlocked -0.09 0.003
(-4.14)*** (0.85)
RTA -0.01 -0.18 -0.0008
(-0.89) (-3.52)*** (-0.61)
Currency Union 11.13 0.007
(11.68)*** (1.68)*
Adj R2 0.32 0.73 0.07
S.E. of regression 0.17 0.34 0.02
N 993 992 762
a
Dependent Variable: FDI Inflows as percentage of the GDP of the receiving
country (in 1987 US$ constant). White heteroskedasticity-consistent standard errors and covariances calculated.
t-statistics in parentheses. * indicates 10% significance level, ** 5% significance, *** 1% significance.
25
Table 2a: Controlling for corruption and FDI incentives and restrictions – Cross-section OLS a
26
Table 2b: Controlling for corruption and FDI incentives and restrictions – Cross-section OLS a
27
Table 3: Controlling for legal origin (pooled least squares estimation)a
N-N N-S S-N
ln(POP) of host -0.003 -0.07 -0.002
(-0.83) (-3.08)*** (-1.99)**
ln(GDP -PC) of host 0.0002 -0.04 -0.002
(0.02) (-1.45) (-1.24)
ln(DIST) -0.08 -0.14 -0.0006
(-7.58)*** (-3.38)*** (-0.53)
Common Legal Origin 0.04 0.006 0.0003
(2.92)*** (0.27) (0.25)
Border -0.05 -0.04 0.003
(-1.81)* (-0.60) (0.65)
Language 0.16 0.14 0.009
(4.18)*** (3.14)*** (2.31)***
Latitude -0.002 -0.002 -0.0001
(-3.71)*** (-3.57)*** (-1.56)
Landlocked -0.09 0.003
(-3.81)*** (0.81)
RTA -0.01 -0.18 -0.0006
(-1.13) (-3.58)*** (-0.49)
Currency Union 11.10 0.007
(11.63)*** (1.69)*
Adj R2 0.32 0.73 0.07
S.E. of regression 0.17 0.34 0.02
N 993 961 733
a
White heteroskedasticity-consistent standard errors and covariances calculated.
t-statistics in parentheses. Time and source country dummies are used in all regressions.
* indicates 10% significance level, ** 5% significance, *** 1% significance.
28
Table 4: Controlling for trade (pooled least squares estimation)a
N-N N-S S-N
ln(POP) of host -0.05 -0.11 -0.003
(-4.98)*** (-4.92)*** (-2.06)**
ln(GDP -PC) of host -0.06 -0.10 -0.003
(-3.57)*** (-3.98)*** (-2.03)**
ln(DIST) -0.03 -0.10 0.00006
(-1.70)* (-2.48)*** (0.05)
ln(Trade) 0.06 0.07 0.001
(4.64)*** (3.29)*** (1.59)
Border -0.05 -0.08 0.002
(-1.68)* (-0.35) (0.33)
Language 0.15 0.12 0.008
(3.91)*** (2.65)*** (2.09)**
Latitude -0.0009 -0.003 -0.0001
(-1.89)* (-4.13)*** (-1.55)
Landlocked -0.06 0.003
(-2.49)*** (0.72)
RTA -0.03 -0.17 -0.001
(-2.12)** (-3.41)*** (-1.02)
Currency Union 11.15 0.007
(11.69)*** (1.68)*
Adj R2 0.33 0.74 0.07
S.E. of regression 0.17 0.34 0.02
N 990 946 723
a
White heteroskedasticity-consistent standard errors and covariances calculated.
t-statistics in parentheses. Time and source country dummies are used in all regressions.
* indicates 10% significance level, ** 5% significance, *** 1% significance.
29
Table 5: Controlling for telephone traffic (pooled least squares estimation)a
N-N N-S S-N
ln(POP) of host -0.06 -0.10 -0.004
(-4.76)*** (-3.08)*** (-2.32)**
ln(GDP -PC) of host -0.06 -0.08 -0.004
(-3.12) (-1.97)** (-1.83)*
ln(DIST) -0.02 -0.17 0.002
(-1.08) (-2.24)** (1.01)
ln(Tel) -0.001 -0.06 0.001
(-0.11) (-2.59)*** (2.33)**
ln(Trade) 0.07 0.12 -0.000004
(3.71)*** (4.69)*** (-0.01)
Border -0.05 -0.10 0.006
(-1.55) (-0.49) (2.06)**
Language 0.14 0.20 0.006
(3.28)*** (3.04)*** (1.90)**
Latitude -0.0009 -0.004 -0.00009
(-1.59) (-3.36)*** (-0.99)
Landlocked -0.06 -0.004
(-2.21)** (-2.45)***
RTA -0.03 -0.22 -0.00005
(-1.70)* (-2.49)*** (-0.04)
Currency Union 11.12 0.002
(11.98)*** (0.99)
2
Adj R 0.34 0.82 0.07
S.E. of regression 0.18 0.36 0.02
N 828 531 587
a
White heteroskedasticity-consistent standard errors and covariances calculated.
t-statistics in parentheses. Time and source country dummies are used in all regressions.
* indicates 10% significance level, ** 5% significance, *** 1% significance.
30