Agricultural Production and Yield Estimation. Two distinctive aspects of Brazilian Agriculture and a perspective on world food problems.

Diogo Bardal - Department of Economics University of Siena Italy E-mail: diogobardal@gmail.com

Abstract Brazil has been increasing its importance on agricultural markets. The reasons are well known to be the relative abundance of land, which allow for a fast response to price stimulus and a fast paced technological change, leaded by research of new varieties and mechanization. We estimate a response Soybean and Corn area to increases in expected return. Regarding yield, we have observed a large variation in the yearly rates of growth in yield, climate being probably the main source of this variation which result in good and bad years. In South America, a special attention should be given to the El Nio and La Nia phenomena, both said to have important effects on rainfalls patterns and consequently in yield. We examine the influence on El Nino and La Nina in historical data and some ways of estimating the impact of climate on yield of Soybean and Corn markets are purposed. Possible implications of climate change may apply.

Keywords: Soybean, Corn, Land Elasticity, El Nio, Yield Estimation

Nov/2012

Introduction
On the 60s Brazil was a country that poorly could produce food to supply its own people. After a wave of reforms, the foundation of the Empresa Brasileira de Pesquisa Agropecuria (EMBRAPA) and the mechanization financed by the state, the country started building its path towards auto sufficiency and to become one of the greatest food exporters of the world (WILLIAM, 1984). In the last five years planted area for grains has increased by more than 3.4 million hectares, now at 50,8 million (Conab. 2012). The growth has been followed by an increased participation in export markets, in which Brazil should surpass the U.S as leading Soybean exporter in the next years (USDA, 2012). The expansion of land cultivation is connected to a few commercial crops: Soybeans, Corn, Cotton and Sugar. Its safe to say that the export market is commanding expansion in the main areas of production of Corn and Soybeans (in Mato Grosso, the leading state in these two crops, 60% of production goes to export markets) (SECEX, 2012). Soybeans and Corn are main ingredients of animal feed, and both prices are connected to the increase in meat and oil consumption due to the economic development of Asia. The conversion of crops into biofuels is also an issue that changes the structure of Agricultural Market increasing its pegging the price of agricultural commodities to oil prices. The recent spike in international commodity prices in the first decade of the 2000s put Brazil and South America on a central position regarding food production. In order to understand better the dynamics of the Brazilian expansion, we would like to explain two characteristics that illustrate how (and how fast) Brazil can contribute to the increase of agricultural output in the years to come.

Two Distinctive Characteristics of Brazilian Agriculture

Land Elasticities

Land availability is huge in Brazil. According to the last Agricultural Census in 2006, just looking at pasture land that could easily be converted into commercial crops, without considering forests and other protected areas, the availability amounts to at least 100 million hectares (IBGE, 2006). The Issue of land abundance and land concentration is deeply rooted in historical factors. Firstly, in Brazilian Expansion to the West (the occupation of states of the center-western regions and the South), there wasnt any similar legislation as the Homestead Act, which allowed for an equal distribution of lands and the formation of a land market (Dias and Amaral, 1990). As a consequence the territory was first occupied by squatters that later lost or sold their lands to bigger owners, which had their property recognized later by exercising political influence. Although we wont enter too much into detail, these structural factors explain several things of Brazilian Agricultural sector, from the existing land conflicts to the highly concentrated and efficient agribusiness sector of today.

The abundance of land and the existence of highly mechanized, capitalized agricultural properties may allow for a fast response of output to changes in prices. Land Elasticities try to evaluate how much do farmers (and the agribusiness sector) respond on the increase of planted area to increases in expected returns, (i.e. if the elasticity is 0.1 it means that a 1% increase in expected returns would lead to a 0,1% increase in planted area). To build an expected return function, we need first to characterize costs and revenues for the farmer. Babcock and Barr (2010) construct the expected return function from the future harvest time prices on the commodity multiplied by expected yield, minus expected costs (in which they use actual costs). Babcock and Barr calculate discrete elasticity, by a period of 2 years (1997-1999 to 2001-2003, 2004 to 2006 and 2006 to 2009). Exp. Return = Exp. Revenues Exp. Costs Exp Revenue = Fharvest prices*Exp.yield Elasticity of land use (for several products) to expected returns in Brazil and in the U.S was then calculated, remaining around 0.17 for Brazil and 0.03 in the U.S, which is in an interesting result, because it shows the difference in price response between the two countries. However, for forecasting purposes the measure seems not very accurate. If we would apply this elasticity for this year, where returns from soybean market in Brazil were estimated at 46.5% (Safras & Mercados) we would expect an increase in soybean area from 25 million ha to 30 million ha in 2012/2013 season, while the planting intentions released by CONAB this September showed only an area of 26.8 million hectares. Another objection we may raise is to this way of constructing the expected returns function. Once yield depends on the level of technology employed, there may be correlation with price levels and expected costs (i.e. a farmer may want to intensify production (and spend more) when prices are high in order to obtain higher yield). The second objection is that by taking an aggregate measure of land elasticity one may ignore the fact that especially for Corn, Soybeans and Cotton, which basically uses the same type of land, there has to be some substitution mechanism between the three crops area depending on their relative price (which imply in the existence of cross elasticities). So the elasticities of each crop might depend on the crop relative prices. We may leave however this matter aside for now. We shall estimate land elasticities for only Soybeans and Corn. For the time being we will define differently the expected returns function in a more convenient way. We will construct it as an expected operational margin in local currency per unit of output: Exp. Return= (Exphprice/ExpCost)*Exrate The expected prices are taken from pre-planting period from CEPEA/ESALQ and is the price in US$ per bag (60kg) of Soybeans and Corn, sold in

Paranagu (a Port city, in hold has the biggest volume of business in agricultural sector). The expected cost we use is not the current year cost, as Babcock and Barr (2010) uses, but an average of fertilizer price in dollars from USDA in April, before planting, when farmers tend to begin purchasing inputs. As Brazil imports more than half of its fertilizers (ANDA, 2012) and they are the most variable component of cost (and an important one, for it constitutes 25% of total cost), it seems like a good measure of operational cost. Our regression analysis of expected return on the first differences of planted area (all in log) found an elasticity for all the period (1997-2011), that is around 0.1, sensibly lower than what was found by Babcock and Barr (2010), but still higher than the elasticity in the U.S, which does not contradict the main conclusion about the differences between Brazil and in the U.S. For forecasting purposes the model tends to fit better in out of sample forecasting, and all residual tests and information criteria point in a good direction. The same analysis for the corn crop has given similar elasticity values (around 0.11) but weaker p-values, which lead to a rejection of the null hypothesis at 10%. The results are probably due to a greater participation of corn in the domestic market, making it less sensible to changes in exchange rate but more sensible in changes to relative prices of other crops, for example. A model that valuates elasticity of land dedicated to corn using just the soybean/corn relative prices tend to predict better and be more coherent with the latest planting intentions. By this brief analysis of Brazilian empirical data on area we want to demonstrate that Brazil has not only been an ever important player in agricultural markets, but will tend to increase its market share in the next years especially in a scenario of rising prices. Its planted area can respond relatively fast to price stimulus, and with technology advance can become within some years the biggest producer of food supply. A similar response may only be obtained from African countries, once their agricultural production is better organized. The evidence goes in the same direction of the basic assumptions of the economics of international trade (and referring to the Stopler-Samuelson theorem) which states that the most efficient country and with the greater factors endowment (in this case land) will be a net exporter of the good in which it is factor intensive. Of course we may look at this also with great concern for the future of food production. Firstly that this great abundance of land and its fast paced utilization in order to supply food consumption in Asia has important environmental consequences. Lead by increased demand, price increases of the major crops is also linked to deforestation as they increase land use for pastures. The mechanism may work as follows: as crop land is expanded by the conversion of low productivity pastures (due to high prices of animal feed and ethanol), prices of livestock and meat increases along, which add to the rate return of cattle farming and the conversion of forests into pastures. This indirect mechanism combined with poor environmental control can seriously damage the protected areas, although they are not necessarily needed for further expansion of output.

The second reason for concern is if this huge increase in animal-feed production is the most efficient way of feeding the world. We know for a fact that to produce one kilo of pork meat we need about five kilos of animal feed, and to produce one kilo of poultry we need about three. The change in consumer habits (as average income increases, per capita consumption of meat tends also to increase) may add an important bottleneck for the availability of food (Baker, 1977). However, we will also put aside the considerations about the consumer market Let us now examine the second factor that in a way makes Brazil and South America an especial case, which are their climatic patterns and its connections with yield values. Technology and the trend in yield in Brazil The second distinctive aspect of Brazilian Agriculture lies in its connection to two important phenomena of the Pacific Ocean, called El Nio and La Nia. Looking at the general trend in yield in soybeans (and also the same would apply for the corn crop), it seems that generally yield has increased through the years, however not at a constant rate of growth. Not only that, generally there is an alternation between positive growth and negative growth, a history made of good years and bad years. This is actually not new to Agriculture. Since the most ancient times people speak of fat cows and thin cows and try to predict them by interpreting dreams of the Pharaoh, using traditional knowledge and so on. Figure 1: Soybeans yield and first log differences (1976-2011)
8.2 8.0 7.8 7.6 7.4 7.2 7.0 1980 1985 1990 1995 2000 2005 2010

.4 .3 .2 .1 .0 -.1 -.2 -.3 -.4 1980 1985 1990 1995 DIF 2000 2005 2010

LYIELDS

Source: Conab, 2012 Let us analyse further the rate of growth. It has been highly variable as we look at the data (standard deviation is 0.05) and a great deal of this variation of the rate of growth is probably related not only to climate but to input utilization (like fertilizers), which partially depend on grain and input prices in the international markets (as we mentioned before Brazil Imports more than half of

its fertilizers 19 million tons, with a consumption of 24 million tons in 2011) (ANDA, 2012). The rate of growth of fertilizer use is also variable, as so it is the use of machines for example. Theoretically to avoid problems of consistency of estimators (because climate expectations affect prices, which affect degree input utilization) one could think of yield as depending of the technology (which could be seen as just a drift in a trend stationary series as we treat here) and climate. Climate could then be seen simply as stochastic term, which in average is zero. In that case:

Another way is to use an ARMA representation. We used it in our initial analysis choosing a most suitable autoregressive process to obtain the model with the lowest possible information criteria and white residuals. We then arrived at an ARMA (4,2) as a possible representation of the growth yield rates of both soybean and corn yields. However we are not very sure that this may be the accurate way of representing the process for forecasting. More importantly, it is non-theoretical analysis, in other words, it offers no economic explanation whatsoever to the phenomena (Johnston & Dinardo, 1999). We propose here two alternative methods of estimation which are useful to explore some insights on how to treat climate and technology. As we mentioned in the beginning of the session, it is largely noticed by climatologists and by the agricultural markets as well, that the El Nio and La Nia phenomena have an important role in the agricultural output of South America. In 2011, for example, La Nina was blamed for provoking losses all over the southern Brazil and Argentinian crops, strongly affected by a disastrous drought. Several articles in the Financial Times and Reuters have reinforced these expectations which have been followed by dramatic increase in International prices given the tight stocks in the U.S (Financial Times, 2012). On this year, even the possibility of a weak El Nio has inflated expectations on production in South America so analysts speak of an increase in production from 66 million tons to 82 million in just a year. According to the National Oceanic and Atmospheric Administration (NOAA) the El Nio is characterized unusually warm temperatures in the equatorial Pacific Ocean while the La Nia by unusually cold temperatures on the same sites. Ocean temperatures affect precipitation regularities, fish reproduction cycles in various parts of the world producing ambiguous effects depending on the region. In Brazil La Nia is connected to drought in the south and El Nio is connected with good rainfalls in the South and Midwest. But does it really affect yield? Or does it even affect positively yield, once we know for a fact that excessive rainfalls during maturation phases should instead be prejudicial ? Data We have used the data set for monthly oceanic temperatures and anomalies (measures of how many degrees hotter or colder than normal) in

the coast of Peru (known as Nino 12 region) produced by (NOAA, 2012). There are a variety of indexes and temperature measures for the oscillations in oceanic temperature. We tried a few of them, which led to non-contradicting results, so we kept the initial one. We have also calculated the 3 month moving average of these temperatures. For yield data, we used the dataset from the Companhia Nacional de Abastecimento (Conab) in national terms and later for only the Southern region. For fertilizer data we relied on Instituto Brasileiro de Geografia Estatstica (IBGE) to get the total sales of fertilizers. The data included about 36 observations for each variable, but only 18 for fertilizer sales. Some of the more interesting tables with estimated values can be found on the appendix.

Methodology
We separated the yearly values for each month as independent variables, making a regression analysis of the log of the first differences of soybean and corn yields. As harvest of the summer crop starts in Feb/March, we used the ocean temperatures from the previous year, as they are likely to affect crops during the growing period (Sept-Mar):

In the above equation, X is the vector of regressors, and B the vector of its respective coefficients (yearly average temperatures in January, March, until March of the year of harvest or the three month average i.e. Jan-Feb-Mar, FebMar-May and so on). As we thought the influence of fertilizer consumption on yield may be important, we later included that variable, but only on a two-stage least square estimation (2SLS) and Generalized Method of Moments (GMM), using the climate anomalies as instruments to estimate Fertilizer consumption (as we suspect that expectations about climate may influence the degree of input utilization).

Results
After dropping the months which were non-significant at 10% value we reached into a model where we see yield clearly explained by movements in January, March May, June, October anomalies, months before the planting period, which, by the way, may be interesting to the expectation formation processes. When we take the moving average of a three month period we see similar results (in terms of sign and magnitude): Jan-Feb-Mar, Feb-Mar-Apr, Apr-May-Jun, Jun-Jul-Aug, Jul-Aug-Sep, all presented very low p-values (<0.003). For corn yields of course the relevant months were different. It makes sense because both crops have different critical periods, where soil moisture is more relevant for yield for example, added to the fact that we cultivate also a winter crop of corn grown in May to July in some states. For the corn yield the relevant months were Mar-Apr-May, May-Jun-Jul, Jun-Jul-Aug and Jul-AugSep.

We consider these results not to completely be satisfactory once they are just slightly better than the ARMA models (both in Information criteria and forecasting power) that we developed in the beginning. Still one can derive important insights about this possible influence on the development of crops. Firstly, the sign of the coefficients is ambiguous: some of them are positive and others are negative, which at least initially was surprising as we expected that a consistent increase in anomalies would lead to an El Nio, and consistently negative anomalies to a La Nia, influencing positively or negatively the growing of crops. One explanation is that in order to have a positive influence on the rate of growth, its needed that total predicted effect of the anomalies be positive (or that the months positively correlated with yields be higher than the negative ones). In fact, this is what we observe when we have strong El Nio years, such as 1982-1983 or 1997-1998 and strong La Nia years like 1999-2001. The total effect of the anomalies seems to point out in an increase/decrease in yields. In fact, 1983 and 1998 were good years (rates of growth of (11.8% and 3.6%), and 2002 was a bad year (rates of growth 0f -6.5%, with a negative accumulated effect). La Nia was also strong in 1988-1989 but we havent had a reduction in yields, on the contrary in 1989 where there was surprisingly an increase of 14% in soybean yields. The explanation may rely in fact the total predicted effect was positive for all models for this year. That might explain why just estimating dummies for strong El Nio and La Nina years give such a bad outcome (p-values are above 0.8). Similarly in 2001 we should also have expected a negative yield growth as it were supposedly a year of La Nia, but as we calculate values we find a dominant positive effect. There was an increase of 13.06% in yield. But the reason may rely also in the influence of input utilization. In the year before the harvest fertilizer sales increased by 16% (or 2.7 million tons) while in 2002, the increase was modest of only 4% (or 600 thousand tons). Therefore we should partially refute the common sense idea that Every El Nio year is good for South American grain production while every La Nia year is bad. One can easily see that by estimating correctly the impacts of ocean temperatures and anomalies on yield. It is a surprise though for us that for forecasting purposes, although standard errors were too high to produce a credible out of sample yield forecast, this model could predict for more than 70% of the years if the season was going to be good (positive rate of growth) or bad (negative rate of growth), regardless of the existence of an El Nio or La Nia, and just using totally exogenous climate data. The final implication of our model is that there is a significant room for adjustment given by the level input utilization, and definitely this is an interesting topic of future research. It is very likely looking at our data that farmers may respond with a larger utilization of inputs when climate variables indicate a bad year ahead. This is not only interesting from the point of view of the rationality of agents (and the possibility of convergent expectation formation, but for analysis in price formation mechanisms and planting decisions, in which the level of technology applied is endogenous to the system).

Conclusion
There is no other country today that is more capable of expanding its agricultural output than Brazil. Although the possible bottlenecks may exist in infrastructure and logistics, the scarcity of land and technology certainly is not an issue for the time being. To understand better the future role of South America on food production one may study better its connection to El Nio and La Nia phenomena, and although there is plenty of research of this topic among climatologists, there hasnt been much in economics. We outlined some possibilities of estimation. We also raised some issues regarding the possible influence that climate may have on farmers decision regarding costs of production and consequently in the calculus of return. Our preliminary investigations, relating the anomalies on temperature of the coastal Peru and fertilizer sales point out in this direction. Although the model that we developed does not produce a sufficiently small confidence interval for forecast, we have been able to predict the good and bad years on soybean and corn production. We have also demonstrated that the common sense belief that El Nio is necessarily good and La Nia indicate bad years may be mistaken. We suspect that by an improvement of a model that includes the level of input utilization and a better investigation of the use of technology in agriculture may yield better forecasting of agricultural yield.

References: Associao Nacional Para a Difuso de Adubos (ANDA). Eseatsticas. Available in: <http://www.anda.org.br/estatistica//principais_indicadores_2012.pdf>. 2012.
Barr K. J., Babcock B. A., Carriquiry M., Nasser A., Harfuch L. (2010) Agricultural Land Elasticities in the United States and Brazil. CARD Working Papers, Iwoa State University. Available in: <http://www.card.iastate.edu/publications/dbs/pdffiles/10wp505.pdf>. Baker, C. B. (1977). U.S perspectives on World Food Problems. Illinois Agricultural Economics, Vol 17. No 2. Pp 1-6. Companha Brasileira de Abastecimento (Conab). Crop estimates. Available in <www.conab.gov.br>. 2012. Dias, Guilherme. L. S. e Cicely M. Amaral " Mudanas Estruturais na Agricultura Brasileira", in Brasil: Uma Dcada em Transio, org. Renato Baumann, Campus, 1999. Financial Times. Lower El Nio Prospects to Hit crop yield. Available in:<http://www.ft.com/intl/cms/s/0/70c2fb9c-1225-11e2-b9fd00144feabdc0.html#axzz2CHAoh5R4>. Instituto Brasileiro de Geografia Estatstica (IBGE). Censo Agrcola 2006. Available in: <www.ibge.gov.br>; Johnston J. Dinardo, J. Econometric Methods. McgrawHill, 1999, 4ed. NOAA. El Nio Souther Oscillation. Current Conditions and Data. Available in: < http://www.cpc.ncep.noaa.gov/products/precip/CWlink/MJO/enso.shtml> Roberts, M. J., Schlenker, W. (2011). Is Agricultural Production Becoming More or Less Sensitive to Extreme Heat? NBER. Available in: < http://www.nber.org/chapters/c12162>. 2012. SECRETARIA DE COMERCIO EXTERIOS. Estatsticas de Exportao. Available in:< http://aliceweb.desenvolvimento.gov.br/> United States Department of Agriculture (USDA). (2012). World Agricultural Supply and Demand Estimates. Available in: <www.usda.gov>. Williams, G. W. Brazilian Soybean Policy: The International effects of Intervention. Amer. J. Agr. Econ. 66(November 1984b). p. 488-498.

Appendix Table 1 Forecasting Results and Probabilities

Forecast. Prob Growth Soybeans 0,287915 0,620097 0,971972 0,722773 0,381702 0,989402 0,120801 0,875909 0,345797 0,783126 0,923059 0,53135 0,104555 0,323791 0,612147 0,935146 0,424694 0,883118 0,704432 0,587052 0,936513 0,574641 0,218012 0,928164 0,013095 0,838354 0,170716 0,614531 0,93904 0,863346 0,368536 0,173348 0,821529 0,205117 0,154301 Percentage TRUE Positive Growth=1; Neg Growth=0 0 1 1 1 0 1 0 1 0 1 0 1 0 0 1 1 1 1 0 1 1 0 1 1 0 1 0 0 1 1 0 0 1 1 0 77,14% True;False; 50% Treshhold T T T T T T T T T T F T T T T T F T F T T F F T T T T F T T T T T F F Forecast. Prob Growth Corn 0,466066 0,831598 0,850902 0,677438 0,798896 0,904215 0,30364 0,684576 0,357527 0,683802 0,911173 0,561417 0,083088 0,534032 0,787685 0,942358 0,580131 0,897977 0,622632 0,698397 0,929125 0,467927 0,4005 0,966958 0,175495 0,811299 0,047797 0,354253 0,869785 0,77459 0,395935 0,440373 0,752886 0,684494 0,406552 Percentage TRUE Positive Growth=1; Neg Growth=0 0 1 1 1 0 0 1 1 0 1 1 1 0 0 1 1 0 1 0 1 1 0 0 1 0 1 0 0 1 1 1 0 1 0 1 71,43% True;False; 50% Treshhold T T T T F F F T T T T T T F T T F T F T T T T T T T T T T T F T T F F

Year 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Source: estimation by the author.

Table2 Regression Output Coefficients log (ycorn)-log(ycorn(-1)) 35 Obs (1977-2012) Coefficien Std. Variable t Error t-Statistic C 0,029797 0,017685 1,6849 MAM(-1) -0,046932 0,018044 -2,6010 MJJ(-1) 0,188615 0,050401 3,7423 JJA(-1) -0,260263 0,071915 -3,6190 JAS(-1) 0,117876 0,037891 3,1109 log(ysoy)-log(ysoy(-1)) 35 Obs (1977-2012) JFM(-1) 0,077643 0,036018 2,1557 FMA(-1) -0,143951 0,052342 -2,7502 AMJ(-1) 0,127668 0,038171 3,3447 JJA(-1) -0,200269 0,061145 -3,2753 JAS(-1) 0,146533 0,044465 3,2955 GMM - log(ysoy)-log(ysoy(-1)) 18 Obs LOG(FERT(-1))LOG(FERT(-2)) 0.725011 0.218964 3.311.098 Instrument list: LOG(JUN(-1)) JUL(-1)) SEP(-1))

Prob. 0.1024 0.0143 0.0008 0.0011 0.0041 0.0393 0.0100 0.0022 0.0027 0.0025

0.0041