The Integration of the Indian Wheat Sector into the Global Market May 2008 Ashish Shenoy† Department of Economics Stanford University Stanford,CA 94305 Under the direction of Professor Peter Hansen

Abstract World food prices have risen rapidly in recent months. The trend has brought up concerns about how markets in developing nations respond to international conditions. In this paper, I try to determine whether the price of wheat in India converges to the world level. Using monthly prices from the United States of America, Canada, Australia, Argentina, and India over a period of thirteen years, I look for evidence of cointegration among the series. Cointegrated series follow a common stochastic process, and thus can be said to move together. I first test for cointegration without restrictions to identify the number of cointegrating vectors and common trends, and then impose restrictions to see how quickly markets adjust to disequilibria. I find evidence that the world wheat trading centers are integrated, with Australia being the most dominant. The Indian wheat price does not converge with the other four. I next use the Granger Representation Theorem to model the adjustment of the markets to shocks. I find that the Indian market adjusts more slowly to a new equilibrium, but the total magnitude of adjustment is greater. Possible explanations include poor infrastructure, regional segmentation within India, and high levels of government intervention.

Keywords: India, wheat, food price, cointegration †

I would like to thank Professor Hansen for his patient mentoring and guidance, without which this project would not have been possible. I am also grateful to Professor Wally Falcon, Professor Roz Naylor, Professor Nick Hope, Dr. Charan Singh, Smt. Asha Kannan, Dr. Ramesh Golait, Pankaj Kumar, and Professor Geoffrey Rothwell for their help in various stages of my research. Any errors that remain are solely my own.

Contents 1 Introduction

2

2 Literature Review 2.1 Price Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 World Wheat Trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Developing Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6 6 9 13

3 Indian Wheat 3.1 Production . . . . 3.2 Market Structure . 3.3 Government Policy 3.4 Recent History . .

. . . .

14 14 16 17 18

4 Data 4.1 Data Selection and Treatment . . . . . . . . . . . . . . . . . . . . . . 4.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . .

20 20 23

5 Methodology 5.1 Maximum Likelihood Model 5.2 Unrestricted Estimation . . 5.3 Restricted Estimation . . . . 5.4 Adjustment . . . . . . . . .

. . . .

24 25 27 28 29

6 Results 6.1 Unrestricted Estimate . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Restricted Estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31 32 33 37

7 Discussion 7.1 Interpretation of Results . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Possible Explanations . . . . . . . . . . . . . . . . . . . . . . . . . . .

42 42 44

8 Conclusion

46

References

49

CONTENTS

Shenoy 1

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

1

Introduction

Food prices worldwide are skyrocketing. The repercussions of higher prices can be felt in every part of the globe. Over the last three years, food prices have risen by an average of 83%. Cereals seem to be hit the hardest, experiencing an increase of over 20% in the last two months. This recent jump caps off a three year period in which cereal prices almost tripled (World Bank, 2008). In the United States, consumers have responded to the crisis and expectations of further price increases by raising their foodgrains purchases, driving many stores to strat rationing rice and wheat. The fallout abroad has been worse, ranging from food riots to political upheaval in countries from Haiti to Bangladesh, all in response to the the declining availability of basic staples. Officials worry about the lasting impacts of the crisis as it is unclear how soon prices will stop climbing and whether they will return to their previous levels. Some of the factors leading to the current crisis seem temporary. A drought in Australia coupled with sub-par wheat harvests in the United States and Europe drove down global supply, causing prices to rise. Poor policy has compounded the problem as many countries that had in the past held large buffer stocks in case of shortages let stocks dwindle, expecting to rely on imports to make up shortfalls in production (Krugman, 2008). Without a buffer crop to combat rising prices, many large grain producers, including India and Thailand, have severely limited exports in order to protect domestic consumers. Such export restrictions contract supply even further, putting more pressure on prices. High energy costs have exacerbated the current situation. Farmers in countries that export large amounts of food tend to use machinery for production, meaning that gas prices are directly represented in farm production costs (Krugman, 2008). Mechanization of agriculture, use of synthetic fertilizers, and transportation from farm

Introduction

Shenoy 2

to market all feed into price transmission from energy to food. As the cost of energy increases, demand for foodgrains rises as well (Naylor et al., 2007). High gas prices make alternatives such as biofuels more attractive, meaning consumers must compete with biofuel producers in the market. Although only a few crops are currently used for energy alternatives, food supply in other sectors decreases as farmers substitute into production of the more lucrative fuel crops. Energy costs generate both supply and demand effects that raise the price of food. Financial capital invested in an attractive food market may be amplifying the problem. Some investors coming out of shrinking housing and derivatives markets turned to food as a new financial instrument. These traders added to the competition for goods normally sought by corporations for resale and distribution to households. It is difficult to identify exactly how much of this investment is simply response to economic incentives, but some experts believe that speculative trading has driven prices to some extent (see Timmer, 2008). The financial pressures on food prices are somewhat nominal because commodities are generally traded in dollars. The declining value of the currency has caused inflated the apparent affect of other pressures. Dollar devaluation is only a small part of the story, however, and food costs have seen significant real appreciation. Although many of these determinants seem temporary, some structural factors indicate that higher prices may be more permanent. Even taking into account the current weather patterns, growth in annual food production has consistently outpaced population growth; the problem lies in food demand and not food supply (Timmer, 2008). While financial pressures may inflate food demand, alternative energy and changing diets compose the factors underlying rising food demand. Scientists seeking new sources for fuel have turned to foods as a way to generate energy. The price of oil drives this process to some extent, but environmental concerns surrounding fossil fuels would inspire the preference for renewable energy even in the presence of cheap

Introduction

Shenoy 3

oil. Subsidies in developed nations promoting the use of ethanol and other food-based fuels support the industry, guaranteeing that producers of biofuel will compete with consumers for grains and crop land. Global poverty reduction has a feedback effect on food prices as well. This effect is most pronounced in East and South Asia, where incomes are rising rapidly in nations with very high populations. As incomes grow, consumption preferences among the population shift. More people are able to afford more food, which directly augments food demand. People also shift away from cheaper grains and starchy roots into dairy, poultry, and meat (Bennett, 1941). Calories from these luxury goods require a higher amount of grain used as feed for animals than calories from grains directly, so substitution into these goods increases the effective amount of food people seek to consume. While poverty reduction on the whole increases access to food for the lowest income households, it drives up prices in world commodity markets. As a result of these structural factors, even if next year brings good harvests, there is still reason to believe that food prices will remain high in the future. This climate of high food prices has left many concerned about international stability. It has already sparked violence in some regions, and prices are expected to rise further in the near future. Medium-term forecasts estimate that the cost of food will remain much higher than its 2004 level until at least 2015 (World Bank, 2008). High prices fall disproportionately hard on low-income families, who are forced to spend a much larger portion of their income on food in order to survive (Timmer, 2000). Their effective wage will be lower, and many will not be able to cover the added costs. Officials worry that without significant policy intervention, food price increases could cause millions to starve and wipe out decades of poverty relief (Moon, 2008). Popular resistance to a more harsh economic environment has already inspired contempt for incumbent governments, a reaction that will continue to threaten political stability around the world.

Introduction

Shenoy 4

Given these concerns, it is important to understand how nations respond to international prices on an individual level. In this paper, I focus on the interaction between the wheat market in India and international wheat markets. Of India’s total agricultural land, 50% is devoted to either rice or wheat. The prominence of wheat in the Indian agricultural sector places the nation among the top three wheat producers in the world. Agriculture in India provides employment to almost 60% of the Indian workforce (Organization for Economic Cooperation and Development, 2007). The price these people receive for the commodities they produce determines the profitability of farming, which in turn drives the decision to stay in agriculture or move into a more productive industrial and service sector. Indian consumers are also extremely sensitive to food prices. On average, Indians spend 50 to 70% of their income on food. Wheat and rice alone compose 15% of the average Indian citizen’s total expenditures (Reserve Bank of India, 2006). Because such a high percentage of earnings goes into buying food, changes in price have significant impacts on the Indian population. The combined prevalence of wheat as a source of income and as a consumer staple has placed it at the forefront of political and economic discourse throghout the nation. In this paper, I seek to analyze how well the Indian wheat market is integrated with other world markets. To do so, I will test whether the price of wheat in India converges to the export price prevalent in the United States of America, Canada, Australia, and Argentina. In Section 2, I review some of the relevant literature regarding analysis of price series, the world wheat trade, and the Indian situation. In Section 3, I undertake a brief overview of the Indian wheat sector. Section 4 describes the data I use in my analysis, and Section 5 covers the methodology used to analyze the data. In Section 6 I present my results, with a discussion and some proposed explanations following in Section 7. I close with some concluding remarks in Section 8.

Introduction

Shenoy 5

2

Literature Review

2.1

Price Integration

Wheat markets, like other commodity markets, display a high degree of volatility. In the recent past, the standard deviation of prices has ranged from 10% to 50% in the United States (see, e.g., Crain and Lee, 1996). Among these price movements, it is possible to isolate related effets across markets. Analysis of commodity prices is complicated by the fact that the series tend to be nonstationary. The term nonstationary describes a subset of autoregressive series, or series in which each observation can be represented as a function of previous observations with a stochastic component. An autoregressive process that is stationary has a mean that is either fixed or follows a deterministic trend over time. In contrast, nonstationary series do not revert to a predicted mean, which causes the estimated variance to increase with the length of the series. If a series is nonstationary, but is made stationary after differencing d times, it is said to be integrated of order d, written as I(d). Time series that do not revert to a predictable mean require special treatment in econometric models (Box, 1970). Granger and Newbold (1974) find that ordinary least squares(OLS) regression on nonstationary series is unreliable. They note that regression with time series often produces estimators with very high R2 values but strong evidence of serial autocorrelation in the residuals. Such serial autocorrelation indicates that the model is misspecified and the result does not necessarily hold true. The authors identify the nonstationary component of many time series as the cause of the unreliable results. Specifically, OLS regression using nonstationary series produces incorrect estimates of regression coefficients and changes the critical values for tests of significance. When nonstationarity is present in the data, regression is not adequate to draw conclusions about a model.

Literature Review

Shenoy 6

Several authors have proposed tests to determine the order of integration of time series (Dickey and Fuller, 1979; Said and Dickey, 1984; Phillips and Perron, 1988; Kwiatkowski et al., 1992). These tests model a series of zt as an autoregressive function of the form zt = ρzt−1 + a + bt + εt

(1)

where (a + bt) represents a linear deterministic trend and εt is an independent and identically distributed stochastic process. In (1), ρ determines the persistence of past shocks in the series. A value of ρ such that |ρ| < 1 guarantees that any stochastic shock to the process will dissipate over time, meaning that the process centers on the deterministic trend. In contrast, |ρ| = 1 means that shocks remain persistent over time, and the process does not necessarily revert to an underlying trajectory. The presence of a unit root, |ρ| = 1, in an autoregressive series indicates that the series is nonstationary. Even though OLS regression is inappropriate for analyzing nonstationary data, it is possible to find stable relationships among multiple nonstationary series. Engle and Granger (1987) identify a condition in which series revert to a long-run equilibrium, known as cointegration. For a group of time-dependent variables that are I(1), linear combinations of these variables will tend to be I(1). If there exists one or more linear combinations of these variables that are I(0), then the variables are said to be cointegrated. The vector of coefficients in this linear combination, called the cointegrating vector, describes the nature of the long-run relationship among the variables. Cointegration can be generalized to I(d) variables with linear combinations that are I(b) where b < d. The authors propose a two-step method of identifying cointegrating vectors. In the first step, they use OLS regression to estimate potential vectors. They then test the possible cointegrating vectors by plugging them into a model involving first differences.

Literature Review

Shenoy 7

When more than two nonstationary series are present, multiple cointegrating vectors may exist in the data. For a group of p series, the presence of r unique cointegrating vectors indicates that there are p − r stochastic trends among the p series (Stock and Watson, 1988). If r = 0, meaning no cointegrating vectors exist, then each series follows its own random walk and the variables are unrelated. If r = p−1 cointegrating vectors exist, then a single stochastic trend guides all p series. If 0 < r < p − 1, the series are partially integrated, but more than one common trend exists in the data. The presence of p cointegrating vectors indicates that the model is misspecified and the series are actually stationary. The method of finding cointegrating vectors described by Engle and Granger runs into problems when more than two series are present. For multiple markets, each pair must be analyzed independently. Furthermore, which variable is chosen to be the dependent variable affects the results and the method may fail to uncover any cointegrating vectors that are not revealed by regression. Johansen and Juselius (1990) propose an alternate method to identify cointegrating vectors based on a Vector Auto-Regressive(VAR) model. In a VAR, each variable is represented as a function of the lagged values of all variables in the system. I explore this method in more detail in Section 5. In this paper, I use cointegration tests to evaluate market integration. The definition of market integration comes from Ravallion (1986), who notes the distinction between integration and efficiency. Integration simply refers to the comovement of prices; it is still possible that single actors hold monopoly power or are able to manipulate prices through other means. Cointegration provides appropriate information about the comovement of prices by identifying common trends. The number of trends that appear in a group of prices describes the degree to which the prices move together. This result makes no determination regarding the efficiency of resource allocation within the market.

Literature Review

Shenoy 8

Bernard and Durlauf (1995) add to this definition in a study of whether the Law of One Price(LOP) holds among OECD countries. Formally, they define the LOP as

lim E(Pi,t+k − Pj,t+k |It ) = 0

k→∞

(2)

where It is the information available at time t. The definition in (2) implies that for two price series to converge, they must not only be cointegrated, but must have a cointegrating vector of [1, −1]. Such a restriction suggests that the two series move together and respond to shocks by reverting to price equivalence in the long run. The authors bring up the possibility of other cointegrating vectors of the form [1, ξ] where ξ 6= 1 that indicate a weaker type of integration. Bernard and Durlauf posit that cointegrating vectors of the latter form describe situations where two markets react to the same shocks, but respond in different magnitudes. I discuss such a possibility and its implications in more depth in Section 5.

2.2

World Wheat Trade

Researchers employ cointegration tests to describe the price linkages that exist within world commodity markets. Ardeni (1989) applies the Engle-Granger two-step method to multiple world commodity markets. The author identifies markets trading a commodity and tests every possible pair for cointegration. Among wheat markets, he finds evidence that the prices in the U.S.A., Canada, and Australia are linked. Estimation also reveals cointegrating coefficients very close to 1, supporting the LOP. Ardeni finds rejects the cointegration hypothesis for any other commodity, and concludes broadly that trade in commodities does not lead to convergence among world prices. Bukenya and Labys (2005) apply a more complicated set of tests to multiple commodity markets. In the wheat market, they consider prices from the United

Literature Review

Shenoy 9

States, Argentina, Australia, and Canada on an annual basis from 1950 to 1998. They first look for correlation among the data, breaking the sets into subperiods in order to overcome problems with stationarity. With this approach, they find varying results and little evidence for price convergence. The authors then test for cointegration. Cointegration tests suggest that world wheat prices generally trend toward certain equilibrium conditions, though the authors do not identify the number of cointegrating vectors. They use variance decomposition and estimate impulse response functions to produce a more detailed picture of the world market. They find that the United States acts as a price leader, accounting for between 18 and 76% of the variance in price in other markets. With annual prices used to analyze such a volatile market, however, the relevance of the variance decomposition is questionable. These two studies focus on commodity markets in general and not specifically on wheat, so the authors do not explore the market with much depth. Authors that have looked specifically at the world wheat market have found a more complex set of relationships. Goodwin (1992) follows up on Ardeni by applying the maximum likelihood model described by Johansen and Juselius. The author takes export prices from the U.S., Canada, and Australia and import prices from Japan and Rotterdam on a monthly bases from January 1978 to December 1989. In his first estimation, he finds no sign of price convergence. Goodwin adjusts the study by accounting for freight rates at import markets and reruns the model. In the second pass, he finds one cointegrating vector, indicating four stochastic trends among the five markets. He postulates that more cointegrating vectors exist, but cannot be confirmed due to the low power of the test. Goodwin interprets his findings as an indication that the LOP holds in world wheat markets. Mohanty et al. (1996) employ a similar method to analyze the price dynamics among the five largest export centers of wheat. They analyze monthly prices from 1981 to 1993 in the United States, Canada, Australia, the European Union, and

Literature Review

Shenoy 10

Argentina. Like Goodwin, the authors find one cointegrating vector among world markets, which they use to define a long-run relationship. They then estimate adjustment coefficients to evaluate the rate of correction in each market. The authors find that prices in Canada, Australia, and Argentina move more slowly than those in the U.S. and the E.U. The Canadian price drives all others while only responding to changes in the Australian price. The American price affects prices in every market except Canada, and adjusts in response to both the Canadian market and the Australian market. Among the remaining three markets, the authors do not see interaction except through the United States or Canada. They conclude that no distinct price leader exists, but prices are determined primarily in Canada, the United States, and Australia. Updated versions of the previous studies, using similar methods with newer data, find stronger evidence of integration among wheat markets. Yavapolkul et al. (2006) apply maximum likelihood estimation to world wheat and rice markets from April, 1996, to February, 2002. In their analysis of wheat prices, they include the American, Canadian, Australian, Argentine, and Indian markets. They find evidence of two cointegrating vectors among the five markets, meaning that the markets are connected, but the LOP does not hold perfectly. The authors interpret their findings as describing one equilibrium relationship among developed countries and another between developed and developing countries. Analysis of individual responses reveals that the Indian and Argentine markets react strongly to price changes in the other markets without much causality going the opposite direction. The Canadian market appears to be the most exogenous, responding only to the U.S. market. Overall, the authors suggest that America and Canada feature most prominently in world wheat price determination. Although wheat is generally treated as a homogeneous good in the literature, this may not provide an entirely accurate picture of the market. Wheat is classified by

Literature Review

Shenoy 11

protein content, and different types of wheat have different end uses. Soft wheat has the lowest protein content and is best suited for cakes and pastries; medium and hard wheats have higher protein contents that support bread-making; and durum wheat has the highest protein content, best for pastas and semolina flour. There is evidence to suggest that the market for wheat ought to be broken down by class and end use before it can be analyzed (see Wilson, 1989; Larue, 1991). Goshray and Lloyd (2003) take this level of complexity into account when looking for wheat price convergence. The authors identify eleven different types of wheat traded in the United States, Canada, the European Union, Australia, and Argentina and categorize them into three different classes. Using monthly prices from July, 1980 to December, 1999, they test every possible pair for cointegration. They find evidence that every price series is cointegrated with every other series except for Australian prime hard wheat, which appears to act independently. In the second step of analysis, the authors examine the variance of the cointegrating vectors to identify which are structural and which are derived form interaction with intermediary products. They conclude that cointegrating relationships among the same wheat class have the least variance, meaning they represent the strongest trends. For estimation of the world market, they divide the system into four hard wheats, four medium wheats, and three soft wheats. They find evidence of three cointegrating vectors and one common trend among the hard wheats, one cointegrating vector and three common trends among the medium wheats, and two cointegrating vector and one common trend among the soft wheats. They conclude that Canadian prices lead the hard wheat market, Australian and Argentine prices lead the medium market, and American prices lead the soft wheat market. All markets are well integrated, especially hard and soft wheat markets, and prices show signs of responding to common trends.

Literature Review

Shenoy 12

2.3

Developing Markets

Although not many studies have addressed the role of India in the international sector, it is possible to gain some insight from the literature. Previous research disagrees on whether and to what extent price signals are transmitted between the developed and developing world (see Mundlak and Larson, 1992; Quiroz and Soto, 1993; Morisset, 1998). The conflicting evidence has led some to put forth the idea that a different price for commodities prevails among developing countries than among developed countries. Yang et al. (2000) address this possibility in the context of soybeans. Looking at soy prices from the United States, the United Kingdom, Argentina, and Brazil, the authors find evidence of three cointegrating vectors among the four markets. Their result clearly shows that the LOP holds among developed and developing nations; two different world prices do not exist. The estimated adjustment coefficients indicate that prices in the developing nations of Argentina and Brazil react quickly to developed markets, but causality is weak in the other direction. Although this finding suggests that the Indian market will be similarly integrated, the authors emphasize that their results can only be generalized to cases with little government intervention. Reduced government interference in commodity markets does not necessarily follow from broader economic reforms. Baffes and Gardner (2003) consider the case of several developing nations that reformed their economies in the 1980’s and ’90’s. The authors consider cases of economic liberalization that were driven by identifiable political decisions followed by implementation to some extent at major trading centers. They look at 31 different commodity prices spanning eight nations and find very little evidence that economic reforms increase integration into world markets. Fewer than half of the cases even provide estimators with the correct sign, and fewer than half of those are statistically significant. Overall, only Argentina undertook changes that unambiguously increased commodity market integration, and only two

Literature Review

Shenoy 13

other nations saw a higher degree of integration in any commodities after reform. The authors hypothesize that even in the context of economic liberalization, political incentives to protect commodities remain very high. Domestic conditions provide an alternate explanation for the findings: inertia within a national market may prevent price transmission even in the absence of restrictive policy. I consider how these two components factor into the Indian wheat market in the next section.

3

Indian Wheat

Before I begin my analysis, I provide a brief description of the wheat sector in India. A qualitative understanding of the market helps put the findings into context.

3.1

Production

The Indian wheat market is primarily composed of small farmers holding little land. The average farmer operated on 1.55 hectares of land in 1995, and more than two thirds currently farm less than 1 hectare. Fewer than 1% of agricultural landholders in India own more than 10 hectares (Organization for Economic Cooperation and Development, 2007). This dispersal of land holdings and productive capacity creates an agricultural supply sector composed of many actors with little market power. Among the cereals, almost two thirds of production is consumed within the household, meaning many in agriculture produce at a near subsistence level (Persaud and Rosen, 2003). Limitations in rural credit, infrastructure, and market information further hinder farmers’ ability to respond to economic signals. As a result, the price elasticity of supply in the nation is very low, estimated at below .1 for wheat production (Jha et al., 2007). These producers enjoyed large gains in yield from green revolution technology in the 1960’s and early 70’s. Improvements in irrigation and input usage supported

Indian Wheat

Shenoy 14

Figure 1: Wheat is produced in the north, with high concentration in Punjab and Haryana. Figure from U.S. Department of Agriculture (2000). rising production in the late ’70’s and ’80’s. During this period, India’s wheat output grew at an average rate of 5% a year, almost 4% of which came from increases in yield. Since then, growth in productivity has tapered off. In recent years, output has grown at a more modest 2% a year, three fourths of which is increased area planted (Jha et al., 2007). The current yield in India is slightly below the world average of 2.7 metric tonnes per hectare despite the fact that 87% of the area sown to wheat is irrigated. For comparison, yield in China is more than double the Indian figure even though only half of Chinese wheat is irrigated (Government of India, 2004). Some see this poor performance as an indication of room for improvement with better resource management and technical knowedge. Others fear that soil degradation and climate conditions keep the nation’s peak possible yield low. The wheat grown in India is primarily hard or durum wheat; it has a high protein content that makes it more suitable for breads than for pastries. The farmers that grow it employ a wheat/rice crop rotation that keeps the land productive throughout the year. The wheat season extends from planting in November to harvest in April.

Indian Wheat

Shenoy 15

Production is dominated by the northern states, from where the crop is transported to mills around the country for processing. Uttar Pradesh leads the country, accounting for over a third of annual production, followed by Punjab and Haryana at 20% and 10% respectively (U.S. Department of Agriculture, 2000). Figure 1 shows intensity of wheat production by region.

3.2

Market Structure

To organize the many actors in commodity markets, the Indian government has set up wholesale market centers known as mandis, usually yards or warehouses inside which transactions take place. State level agricultural marketing boards are responsible for organization and administration of mandis. Farmers bring their crop to the local center to be registered and classified according to type and quality, then negotiate a sale price and sell the crop to a licensed trader. The trader immediately turns the lot over to the buyer, taking a small commission from the transaction. Once produce has been registered at one mandi, farmers are free to take the produce to other mandis to seek a better price, but they must pay the transportation and carrying costs associated with moving from market to market. Wholesale trade in agricultural commodities is generally not allowed outside of the mandi system. In total, more than 750 mandis exist across the nation, governing markets in 140 different crops (Thomas, 2003). Given the segmentation of the Indian wheat sector into several trading centers, a few studies have tried to determine whether the entire national market can be accurately described by a single price. They question whether high transportation costs, poor infrastructure, and incomplete market information isolate spatially separated markets, leading to wheat traders in different parts of the country facing different prices. Ghosh (2003) uses the maximum likelihood method of cointegration to test price series from 22 markets in 5 states over the 13 year period from 1984 to 1997. He

Indian Wheat

Shenoy 16

finds that within and across states, markets are integrated despite their geographical separation, meaning that price signals are transmitted from one market to another. However, the study finds evidence of more than one common stochastic trend among markets, even among markets within the same state. This result suggests that prices in different markets may diverge due to local pressures, and one price may not adequately describe all wheat markets in the nation, even after accounting for transaction costs. The author concludes that, despite the presence of multiple stochastic trends, markets throughout India respond to related pressures, so it makes some sense to view the country as a single market rather than several separate ones. Using a similar method, Srinivasan (2007) analyzes monthly wholesale prices in 11 different markets from April 1997 to June 2003. His results confirm those of the earlier study: There exists evidence of cointegration among markets, but with multiple common trends. The presence of multiple trends indicates that prices do not necessarily converge, but the cointegration result implies that price signals are transmitted across markets. The author attributes the lack of perfect integration to a combination of high transportation and transaction costs, poor methods of communication, poor contract enforcement, and varying government intervention by state. I discuss the implications of regional segmentation in Sections 4 and 7.

3.3

Government Policy

Indian government policy is motivated by several different goals. According to the Ministry of Agriculture, the government seeks to establish environmentally, economically, and socially sustainable food security through its actions in the agricultural sector (Roy, 2007). This overarching philosophy generates a patchwork of often contradictory policies with varying objectives. In an effort to promote production and limit reliance on food imports, most state governments offer water, fertilizer, and

Indian Wheat

Shenoy 17

electricity to farmers at subsidized rates. These subsidies aim to support yields and keep pace with the growing population. In addition to subsidies, farmers receive price supports to ensure that their business remains profitable. At the start of each marketing season, the Ministry of Agriculture sets a minimum support price (MSP) for key agricultural goods such as wheat and rice. The MSP acts as a price floor: If the market price drops too low, farmers have the option of selling their produce to the government at the MSP. Using subsidies and price supports, the Indian government tries to ensure revenues for its agricultural population while maximizing production. On the demand side, the central government of India actively participates in food distribution. The Food Corporation of India(FCI), a state-run enterprise, is responsible for procuring foodgrains and distributing them publicly under various schemes. The largest distribution scheme is the Public Distribution System(PDS), revised to be more effective in 1998, through which grains are sold to low income families at a discounted rate. The FCI also administers several welfare food-distribution programs, including the Food for Work program and Midday Meal program. In a more general role, the FCI is responsible for maintaining national buffer stocks of wheat and rice. In the event of a food shortage, it can use these stocks to meet demand and keep prices low. All wheat procured by the FCI is either bought domestically at the MSP or imported at the international rate. In total, the Food Corporation of India deals with 20% of the nation’s wheat product every year (Jha et al., 2007). The Indian government subsidizes both production and consumption of cereals in an effort to meet its food security and rural revenue targets.

3.4

Recent History

Over the last 13 years, production of wheat in India has ranged from a low of 62 million tonnes in 1995 to a peak of 76 million tonnes in 1999. Most of the variations

Indian Wheat

Shenoy 18

in production can be attributed to annual weather patterns; area sown and yield per hectare have remained steady over the recent past (Reserve Bank of India, 2006). The relative stagnation in production combined with rapid economic and population growth has periodically placed upward pressure on wheat prices. In the recent past, India has fluctuated between being a net exporter and net importer of wheat. At the tail end of the reforms of the early 1990s, the Indian government opened the wheat market to allow exports. Wheat traders took advantage of the opportunity to export large amounts of wheat, leading the government to place quantitative restrictions on exports in April of 1996. The nation continued to export in decreasing amounts until 1997, when successive low-yield years led to a shortage in government stocks and domestic prices started to climb. From 1997 to 1999, India became a net importer of wheat: In an effort to control prices and meet demand, the Ministry of Commerce banned wheat exports and the State Trading Corporation(STC), a branch of the FCI, imported large amounts of wheat. At this time, all imports were canalized through government agencies such as the STC, leaving private traders unable to act in world markets. Good weather led to increased yields in 1998 and 1999, which eased the pressure on the government. By the end of 1999, high yields drove many farmers to sell their produce to the government, causing the FCI to procure well above the required amount for public distribution. To avoid excessive carrying costs from storage of extra wheat, the Indian government decided to export the surplus, subsidizing licensed exporters to sell wheat on world markets. In July of 2001, the Indian wheat market was once again opened to export as the government cut back subsidies and lifted all quantitative restrictions on the sale of wheat abroad. Imports were still canalized through government agencies in this period, but high stocks left little reason to import wheat. The nation enjoyed a period of high exports. India continued to be a significant exporter of wheat until 2005, when depletion

Indian Wheat

Shenoy 19

of buffer stocks led to concerns that the FCI would not meet its procurement and distribution targets. As wheat production failed to keep pace with growth in demand, the STC floated a tender to import wheat in June of the following year. At the same time, the Indian government lifted restrictions on wheat imports in the private sector, allowing private traders to freely import wheat for the first time. Prices continued to rise after the import decision, however, and the government of India banned exports of wheat in February, 2007. High prices in the domestic market have confounded government procurement efforts, so the FCI has turned to further imports to meet its public distribution and buffer stock targets.

4

Data

4.1

Data Selection and Treatment

In this paper, I look for evidence of transmission of wheat prices across markets in India and major world trading centers. The Indian wheat price is obtained from the Office of the Economic Advisor(OEA). The OEA compiles the prices reported at each mandi into a national figure and publishes weekly, monthly, and annual wholesale price index(WPI) data for every commodity. These prices reflect the revenue received by farmers and do not necessarily correlate to the cost of wheat to consumers, although there is reason to believe that the two are fairly well linked (see Persaud and Rosen, 2003). Consumer price data is not published for individual commodities so the WPI provides the best approximation possible of the effect of international markets on Indian consumers. Considering the mixed evidence for a single national price detailed in Section 3, it is unclear whether my results can be generalized to the entire nation. The method of data collection used by the OEA is heavily biased toward the large northern wheat

Data

Shenoy 20

producing states of Punjab, Haryana, and Uttar Pradesh, where more mandis trading in wheat exist. Unfortunately, records at the Indian Ministry of Agriculture are not complete enough to put together a consistent time series for any individual state or region, so I am unable to compute a more complex regional analysis. For the purposes of this study I convert the national monthly WPI price published by the Indian government to a spot price, using a basket of prices from 33 different markets in 8 major wheat-trading states for normalization. I use monthly prices in the selected markets taken from April, 1994 to September, 2007. Although a larger time series may produce more significant results, two constraints in the Indian data prevent me from extending my analysis further into the past. First, the Indian method of collecting and recording price data changed in March, 1994, meaning that including earlier figures may may generate an inconsistent series (Sharma). More importantly, the Indian economy underwent significant reforms in the period from 1991 to 1993. These reforms began with the government unpegging the Indian rupee in 1991 to avoid a balance of payments crisis, leading to full currency convertibility by March, 1994. The new monetary regime brought with it more openness to market solutions instead of reliance on government licensing and mandates. By removing protections for most firms and industries, the Indian government rolled back much of the anti-agricultural bias that had developed in its policy over the previous forty-five years (Ahluwalia, 2000). These reforms changed the nature of trade in India at both the macroeconomic and microeconomic levels and reshaped the attitude of the country with respect to commerce. As a result, it does not make sense to seek a single, persistent measure of market integration before and after the reforms. Exports of wheat in the rest of the world are highly centralized, with five producers accounting for over 80% of exports. To represent world prices, I take data from the United States of America, Canada, Australia, and Argentina. These four nations

Data

Shenoy 21

account for about 25%, 15%, 15%, and 10% of annual exports. I use the FOB price of U.S. No. 2 hard red winter wheat at Gulf ports, Canadian western red spring wheat at St. Lawrence, Australian soft white wheat set by the Australian Wheat Board, and No. 2 Trigo Pan Argentine wheat reported by the Secretary of Agriculture (U.S. Department of Agriculture, 2008). These varieties all represent medium to hard wheats as Austalian white wheat has a higher protein content than the aveage soft wheat (Australian Wheat Board, 2006). I exclude the European Union, the other major exporter, from my study because of the nature of its product. Exports from the European Union mainly comprise soft wheat, which is a poor substitute for the medium, hard, and durum wheats favored in India. Imports of soft wheat have historically come almost exclusively from the medium-protein soft wheat of Australia (Government of India, 2007). I am also forced to exclude nations in the Former Soviet Union despite their growing contribution to the world wheat market due to unavailability of data. Wheat in international markets is traded in dollars, so I convert the Indian wheat price to dollars using spot exchange rates. For analysis, I seasonally adjust all series using the X-12-ARIMA software from the U.S. Census Bureau (see Time Series Staff, 2007). Seasonal adjustment removes the effects of the annual production cycle from the wheat data, limiting the extent to which data are related simply through time of year. I take the natural logarithm of each series to eliminate variation in movement due to level differences. Figure 2 plots the five series in logarithmic terms over the thirteen year span. The American, Canadian, Australian, and Argentine prices appear to follow the same general trend. The Indian price occasionally displays similar movement, but with less consistency.

Data

Shenoy 22

Figure 2: Price of wheat in logarithmic terms for the five markets included in the study.

4.2

Descriptive Statistics

Descriptive statistics for the adjusted data are provided in Table 1. Indian prices show the least variance, and Argentine prices the most variance, over the period of analysis. Previous research has found that commodity prices are nonstationary with first differences that are stationary. To confirm that this is the case with my data, I apply two different unit root tests to the value and first difference of each price series. I first apply the Augmented Dickey-Fuller (ADF) test, which takes the presence of a unit root as its null hypothesis and tests the alternate hypothesis of no unit root present (Said and Dickey, 1984). I also apply the Kwiatkowski, Phillips, Schmidt, and Shin (KPSS) test, which tests the null hypothesis of no unit root present against the alternate hypothesis that a unit root exists (Kwiatkowski et al., 1992). Because the

Data

Shenoy 23

Descriptive Statistics

Max Min Median Mean Standard Deviation

U.S.A. (PU S ) 5.80 4.62 5.02 5.03 0.221

Canada (PCAN ) 5.61 4.93 5.21 5.20 0.156

Australia (PAU S ) 5.85 4.88 5.16 5.19 0.179

Argentina (PARG ) 5.57 4.54 4.93 4.99 0.233

India (PIN D ) 5.38 4.78 5.00 5.02 0.120

Table 1: Descriptive statistics of the five price series.

two tests take opposite null hypotheses, their agreement provides a strong argument for or against stationarity. The ADF test fails to reject the null hypothesis of nonstationarity in level with even 90% confidence and rejects the null hypothesis of nonstationarity in first differences with over 99% confidence for all five of the series. The KPSS test rejects the null hyothesis of stationarity in level with over 99% confidence for the United States, Canada, and Australia; and with over 95% confidence for Argentina. While the null hypothesis of stationarity cannot be rejected for Indian wheat prices at the 95% confidence level, the results of the tests are consistent with a non-stationary price series. The KPSS testfails to rejet the null hypothesis of stationarity in first difference for all five series. Unit root test results are reported in Table 2. Given these results and the findings of previous literature regarding wheat prices, I continue under the assumption that all series are I(1).

5

Methodology

Because I am dealing with nonstationary series, I test for evidence of cointegration. I choose this method over OLS regressions on the first differences because of the information lost by differencing. Cointegration can address simultaneous causality, accounts for the influence of lags, and provides information regarding both the long-

Methodology

Shenoy 24

Unit Root Tests Series USA level difference Canada level difference Australia level difference Argentina level difference India level difference Critical Values: Level:10% 5% 1% Difference:10% 5% 1%

ADF

KPSS

-1.61 -10.21***

0.262*** 0.125

-1.18 -11.42***

0.271*** 0.077

-1.68 -10.06***

0.230*** 0.135

-2.82 -8.27***

0.182** 0.075

-2.54 -9.79***

0.141* 0.108

-3.14 -3.44 -4.02

0.119 0.146 0.216

-2.58 -2.88 -3.47

0.347 0.463 0.739

Table 2: Lag length selected using the Akaike Information Criterion (Akaike, 1974). Critical Values derived from MacKinnon (1996). *, **, *** denote rejection of H0 at the 90%, 95%, and 99% confidence levels.

run conditions and the speed of adjustment. The optimal number of lags is determined using the AIC (Akaike, 1974). All estimation of cointegration coefficients is coded by me using Ox (see Doornik, 2007).

5.1

Maximum Likelihood Model

I use the maximum likelihood method of testing for cointegration to analyze the selected price series (Johansen and Juselius, 1990). The method starts by modeling the system in a VAR. If Pt is a vector containing p prices (P1,t , . . . , Pp,t ) at time t for

Methodology

Shenoy 25

t = 1, . . . , T , the VAR representation of the system using k lags can be written as

Pt =

k X

Πi Pt−i + µ + δt + εt

(3)

i=1

where µ is a constant, δ describes a linear trend, and ε is a stochastic error term. In (3), each Πi is a p × p matrix of coefficients for its respective vector of lags. The VAR can be rewritten in its error-correction form (ECM) by differencing the price terms: ∆Pt = ΠPt−1 +

k X

Γi ∆Pt−i + µ + δt + εt

(4)

i=1

In the ECM form, the left-hand side is stationary because it is simply the first difference of an I(1) process. All of the ∆P terms in the right-hand side are I(0) for the same reason. For the equation to be balanced, ΠPt−1 must also be I(0). The matrix Π represents a linear combination of the variables contained in Pt−1 , and thus carries information about the cointegrating relationships. The number of cointegrating relationships, r, corresponds to the rank of Π. For a value of r where 0 < r < p, Π can be broken down into p × r matrices α and β such that Π = αβ ′ . Each column of β represents a cointegrating vector, meaning each column describes a long-run relationship exhibited in the data. Formally, a column vector in β of < β1,1 , . . . , βp,1 > indicates a long run equilibrium of

β1,1 P1,t + · · · + βp,1 Pp,t = c

(5)

where c is a stationary process. Each column of α represents a vector of adjustment coefficients; the values describe the rates at which the variables adjust when the system is not in the equilibrium described by the corresponding column of β.

Methodology

Shenoy 26

5.2

Unrestricted Estimation

I start by estimating a VAR without any restrictions on α or β. Johansen and Juselius consider multiple possible specifications for the estimation problem. In my analysis, I include a constant to account for possible differences across markets due to taxes, transportation, or other transactions costs. I also include a linear trend restricted to the cointegrating space. Because my prices are reported in nominal values and are not adjusted for inflation, I must allow for the possibility that the equilibrium relationship has some drift. Formally, the ECM form under this specification is written as

∆Pt = αβ ′ (Pt−1 ) +

k X

Γi ∆Pt−i + µ0 + µ1 t + εt

(6)

i=1

where µ1 t can can be broken down into the matrix α and the vector ϕ such that µ1 = αϕ′ . The model describe in (6) can be expressed in the form

Yt = ΠXt + ΨZt + εt

(7)

where Xt represents the nonstationary component of the ECM equation with the linear trend and Zt represents the stationary component and the constant. Π can still be broken down into α and β, where Π = αβ ′ . A procedure to estimate coefficients in (7) is reported by Hansen (2008). Given the unrestricted estimation, the rank of Π is determined by solving the eigenvector problem from Johansen and Juselius (1990). For p eigenvalues, ordered such that λ1 > · · · > λp , the authors present a likelihood ratio statistic, or trace statistic, calculated by

Trace Statistic (λ-trace) = −T

p X

log (1 − λi )

(8)

i=r+1

Methodology

Shenoy 27

The trace statistic tests the null hypothesis of at most r cointegrating vectors against the alternate hypothesis of more than r cointegrating vectors. The statistic is applied starting with r = 0 and is reapplied using r + 1 cointegrating vectors until it fails to reject a null hypothesis of at most r cointegrating vectors. β is composed of the eigenvectors corresponding to the r largest eigenvalues λ1 , . . . , λr accepted by the trace statistic. The unrestricted estimate reveals the number of cointegrating relationships and common stochastic trends present in world wheat markets.

5.3

Restricted Estimation

To better understand the relationship among wheat prices, I add restrictions to the VAR estimation. I first limit β so that cointegrating vectors only operate on pairs of markets. The presence of three or more markets in a single cointegrating vector would signify that the difference in prices between any two of the markets could grow arbitrarily large, as long as the third market moved in a way to balance the equation derived from the cointegrating vector. It is more reasonable to test pairs of markets to see if prices converge. I only need to test p − 1 unique pairs of markets because cointegration between any other pair can be expressed as a linear combination of the existing cointegrating vectors. As defined by Bernard and Durlauf, price convergence between two markets requires that the difference between prices approach zero over time, described in (2). Although the authors consider the possibility of a cointegrating vector of [1, ξ] where ξ 6= 1, such a cointegrating vector would make little sense in the context of wheat markets. The vector does not attest to a disparity in the magnitude of adjustment to common trends, but rather to the nature of the long-run equilibrium. It would indicate that a price shock drives a permanent wedge between the two markets that increases in size as the markets stray further from their original level. This price wedge

Methodology

Shenoy 28

may never close because nonstationary series are not, by definition, mean-reverting. Any cointegrating vector other than [1, −1] introduces the possibility that markets respond to common trends but carry permanent price differences. Because this idea of a permanent, increasing wedge does not seem plausible, I preserve the assumption that integrated markets follow a [1, −1] relationship. In the restricted problem, the ECM retains its form in (6), but with a fixed β. For instance, if r = 4, then the conditions on β are such that  1 1 1 1 −1 0 0 0    0 −1 0 0 β=   0 0 −1 0  0 0 0 −1 

(9)

The restricted value of β and free parameters in the cointegrating space from ϕ can be expressed as vec(β, ϕ) = Hϕ + h where ϕ contains the free parameters, and H and h are fixed. The estimation procedure for the restricted problem involves a two-step iterative process based on (7) and is described by Hansen (2002).

5.4

Adjustment

With restrictions placed on β, I can predict how the three markets respond to differences in price across trading centers. The estimated values of α describe the rates at which markets adjust to disequilibrium conditions. Specifically, a [1, −1] cointegrating vector yields an adjustment in period t of 



ˆ i (Pi − Pj )t−1  α ∆Pt =   α ˆ j (Pi − Pj )t−1

Methodology

(10)

Shenoy 29

Figure 3: The vector α represents the adjustment made by the two markets i and j, and β⊥ characterizes their long-run relationship. The projection of α onto β gives the magnitude of the correction. for any pair of markets i and j. The total rate of adjustment is calculated as the sum of the the two individual vectors projected onto the line perpendicular to the long-run relationship. Figure 3 provides a graphical interpretation of the vectors α and β. The values of α corresponding to a cointegrating vector describe the rate at which a shock disappears and the individual reaction of each independent market. After estimating the cointegrating relationships and adjustment coefficients in the data, I try to predict how a shock to the variables affects the entire system. I use the Granger Representation Theorem from Hansen (2005) that defines a system of I(1) variables as a function of the residuals. The equation makes use of the concept of an orthogonal component to a matrix A, written as A⊥ , defined so that A′⊥ A = 0 and the determinant of the square matrix |(A, A⊥ )| = 6 0. With this notation, the long-run impact on prices predicted by the Granger Representation Theorem is

′ ′ Γβ⊥ )−1 α⊥ εt P = Cεt = β⊥ (α⊥

where Γ = (I −

Methodology

Pk

i=1

(11)

Γi ) from (6). The matrix C describes the impact of a shock

Shenoy 30

in any one price on the final values of all four prices. The high volatility of wheat markets implies that prices may be set on a new trajectory in response to a new shock before reaching the equilibrium created by an old shock, so the impacts predicted by C may never fully materialize. The Granger Representation Theorem also provides a method for modeling how prices move toward the new expected equilibrium. For t = 1, . . . , T , the change in the value of the prices at time t + h due to a shock in period t can be expressed as a portion of the final value of the shock Pt∗ (Hansen and Lunde, 2006). The size of the change is given by ∂Pt+h ′ ′ = (C + Ch ) × Ωα⊥ (α⊥ Ωα⊥ )−1 α⊥ Γβ⊥ ∂Pt∗ I estimate Ω using the average of the residuals, formally Ω =

(12)

1 T

PT

i=1

εi ε′i . Ch is

defined recursively starting with C1 as

∆Ct = αβ ′ Ct−1 +

k X

Γi ∆Ct−i

(13)

i=1

with the convention that C0 = I − C and C−1 = · · · = C−k = −C. As t increases, Ch drops to 0, so the series of (C + Ch ) converges at C. Some algebra using (12) reveals limh→∞ ∂Pt+h = ∂Pt∗ β⊥′ . This final result can be interpreted as the new equilibrium condition. In response to a stochastic shock, the series trends toward a new equilibrium point in the direction of β⊥ , with the new equilibrium described by C and the adjustments leading to the equilibrium described by the series of Ch .

6

Results

In all models, the first coefficient corresponds to PU S , the second corresponds to PCAN , the third to PAU S , the fourth to PARG , and the fifth to PIN D . The AIC selects

Results

Shenoy 31

Trace Test Results Model: r=0 r≤1 r≤2 r≤3 r≤4

µ0 = 0, µ1 = 0 Value Concl. 85.25 Reject (59.46) 56.50 Reject (39.89) 32.39 Reject (24.31) 9.23 Accept (12.53) 1.78 Accept (3.84)

µ0 = αρ′ , µ1 = 0 Value Concl. 81.95 Reject (34.40) 52.98 Reject (28.14) 28.92 Reject (22.00) 8.42 Accept (15.67) 2.70 Accept (9.24)

µ0 6= 0, µ1 = 0 Value Concl. 78.54 Reject (33.46) 49.92 Reject (27.07) 25.88 Reject (20.79) 6.35 Accept (14.07) 1.77 Accept (3.76)

µ0 6= 0, µ1 = αρ′ Value Concl. 85.25 Reject (37.52) 56.56 Reject (31.46) 32.39 Reject (25.54) 9.29 Accept (18.96) 2.75 Accept (12.25)

Table 3: Trace Test from (8) to determine the rank of Π. Critical values are taken from Osterwald-Lenum (1992). All specifications point to the existence of three cointegrating vectors and two common trends.

2 lags for estimation in every case.

6.1

Unrestricted Estimate

I start with an unrestricted VAR to determine the number of cointegrating relationships present in the data. The number of cointegrating relationships, r, is computed as the rank of Π in (6). I estimate r using the trace statistic, given by (8). I start by taking r = 0 as my null hypothesis and testing against the alternate hypothesis of r > 0. I then update H0 to r ≤ 1 and test against the alternate of r > 1, and I continue in this manner until I fail to reject H0 . After computing results with an unrestricted constant and a restricted trend, I check the cases of an unrestricted constant and no trend, a restricted constant, and neither a trend nor a constant. I find that regardless of the specification I use, there is evidence of at least three cointegrating vectors in the data, but I cannot reject the null hypothesis that r ≤ 3. Results of the trace test are presented in Table 3. I estimate an unrestricted VAR with three cointegrating vectors and present the

Results

Shenoy 32

Unrestricted VAR Estimators Model µ0 6= 0, µ1 = αρ′ (log L = 2771.79)

µ0 6= 0, µ1 = 0 (log L = 2769.88)

0.007 0.24 0.038 0.13 -0.014 0.006 0.23 0.038 0.13 -0.023

α ˆ 0.024 0.079 0.28 -0.23 -0.08 0.013 0.049 0.28 -0.20 -0.064

-0.039 -0.059 -0.048 0.20 -0.006 -0.038 -0.051 -0.048 0.20 0.004

0.77 -1 0 0 0.009 0.76 -1 0 0 0.18

βˆ 0.77 0 -1 0 0.19 0.77 0 -1 0 0.10

0.89 0 0 -1 0.015 0.91 0 0 -1 -0.39

µ ˆ0 -0.005 0.29 0.13 0.19 -0.05 -0.075 0.041 0.13 0.38 -0.05

µ ˆ1 4.5e-5 1.7e-4 8.8e-7 -1.2e-4 2.3e-5 – – – – –

Table 4: Unrestricted estimation of the VAR. I first include an unrestricted constant and a restricted trend (top table), then drop the trend and estimate again (bottom table).

results in the first part of Table 4. Estimation with an unrestricted constant and a restricted trend yields coefficients for µ1 that are very close to zero, so I remove the restricted trend and estimate the VAR again. Likelihood for the estimation decreases slightly under the new specification, but not enough to indicate a significant loss of fit. For the rest of this paper, I continue estimation with an unrestricted constant and no trend. This specification can be written as (6) with the condition that µ1 = 0. The second part of Table 4 contains the results from the new VAR evaluation.

6.2

Restricted Estimate

Unrestricted estimation of the VAR reveals the optimal specification and the appropriate number of cointegrating relations. I next codify the LOP condition from (2) and try to identify which markets are integrated. I run the model multiple times with different restrictions on β to find the one with the best fit. Three cointegrating vectors among the five series implies that wheat prices are guided by two common stochastic trends. This result most likely describes a situation in which four markets

Results

Shenoy 33

Test of Possible Market Structures Market Excluded logL P1 P2 log L P1 P2 log L

U.S.A

Canada

Australia

Argentina

India

2758.45 PU S PCAN 2758.15 PCAN PARG 2757.22

2757.46 PU S PAU S 2756.49 PCAN PIN 2757.63

2755.61 PU S PARG 2756.26 PAU S PARG 2757.42

2757.19 PU S PIN 2757.76 PAU S PIN 2755.59

2758.57* PCAN PAU S 2756.19 PARG PIN 2756.44

Table 5: Likelihood values with different restrictions on β. The top row tests cases in which a single market is excluded, and the bottom two rows consider a dual economy with two markets in one relationship and three in the other. * indicates the maximum value.

are integrated and one operates independently. It is also possible but unlikely that one trend guides two markets and the other three markets follow the other trend, meaning there are actually two international markets for wheat that set prices independently and do not converge in the long run. Possible reasons for this kind of structure include differentiation by type, traditional trade patterns, or transportation costs that limit available trade routes. Table 5 presents all potential market conditions with a four-one or three-two structure. I find that the highest likelihood value occurs when the Indian market is excluded from world trade. The result suggests that wheat markets in the United States, Canada, Australia, and Argentina follow the LOP. It is important to note that the likelihood value achieved by excluding the Indian market is not very much higher than that achieved by excluding the American market. Further analysis is sensitive to this decision of which market to omit, and the evidence far from conclusively selects the Indian market as the excluded market. However, this finding is consistent with other studies discussed in Section 2 so I continue with my analysis under the assumption that the Indian market operates independent of the others.

Results

Shenoy 34

Restricted VAR Estimators Coefficients

Variable PU S PCAN PAU S PARG PIN D PU S PCAN PAU S PARG PIN D

−0.004 (0.005) 0.19∗∗∗ (0.005) −0.013∗∗∗ (0.004) 0.10∗∗∗ (0.006) 0.001 (0.002) 0 0.19 (0.003) -0.013 (0.002) 0.10 (0.005) 0

α ˆ −0.07∗∗∗ (0.007) −0.18∗∗∗ (0.007) 0.024∗∗∗ (0.005) −0.23∗∗∗ (0.009) 0.037∗∗∗ (0.003) -0.07 (0.004) -0.18 (0.004) 0.024 (0.003) -0.23 (0.007) 0.037 (0.001)

β -0.004 (0.003) -0.002 (0.003) -0.001 (0.003) 0.19∗∗∗ (0.004) 0.002 (0.002)

1

1

1

-1

0

0

0

-1

0

0

0

-1

0

0

0

0

1

1

1

0

-1

0

0

0

0

-1

0

0.19 (0.004)

0

0

-1

0

0

0

0

µ ˆ0 −0.005∗∗∗ (0.001) 0.29∗∗∗ (0.001) 0.13∗∗∗ (0.001) 0.1∗∗∗ (0.001) −0.05∗∗∗ (0.0005) -0.01 (0.007) 0.01 (0.001) 0.004 (0.0004) -0.02 (0.001) 0.01 (0.0001)

Table 6: Restricted estimation with the Indian market excluded. Estimates of all variables are presented in the top table and estimates after dropping insignificant variables are presented in the bottom table. All remaining terms are significant at the 99% level

The maximum logL value of the VAR with restrictions is 2758.57, a full 11.2 points lower than the likelihood of the unrestricted VAR. (2 ∗ log L) is distributed as χ2 in VAR estimation, giving a χ2 statistic of 22.4. My model has six degrees of freedom after normalization, so the loss of fit from restricted estimation is significant at the 99% confidence level. Despite this significant decrease in the likelihood of estimation, I continue with restrictions because the of the prediction made by the LOP condition. After my initial restricted estimation, I drop all of the insignificant terms and estimate the system once again. The decrease in log L from dropping insignificant terms is less than 0.04, signaling an insignificant loss of fit. Full results

Results

Shenoy 35

Figure 4: Price spreads between the PAU S and other prices. PIN D shows the least evidence of mean reversion.

from the restricted estimation with the Indian market excluded are given in Table 6. Tests of cointegration have low power and rely as much on the total length of a data series as on the number of observations. It could be the case that I falsely accept the null hypothesis of r ≤ 3 due to the fact that I do not cover a long enough time span in my analysis (Hakkio and Rush, 1991). If this is the case, then r = 4 and all five markets are integrated. Although I cannot test for this possibility, graphically analyzing the spreads between markets suggests that it is unlikely. Figure 4 shows the difference between PAU S and prices in other markets. I select the Australian price because the Australian market is the least likely candidate for exclusion from the cointegrating relationships. If markets are cointegrated with a [1,-1] relationship, the spreads should appear stationary. While none of the spreads are very clearly meanreverting, the difference between the Australian and Indian price varies the most with

Results

Shenoy 36

the least apparent underlying trend. The graph supports the statistical inference that the Indian market is not integrated with other world trading centers.

6.3

Adjustment

Restricted estimation implies that the price of wheat in the United States, Canada, Australia, and Argentina converge in the long run. Markets respond to short-run disequilibria by adjusting at the rates given by the column of α corresponding to the column of β that describes their long-run conditions. The total rate of adjustment in two markets is calculated by projecting the sum of the coefficients in α onto β. Because the long-run equilibrium lies on the 45◦ line, the projection problem resolves cleanly into α2 − α1 . The first column in α corresponds to the relationship between PU S and PCAN . The Canadian price responds rapidly to disequilibria, adjusting by almost 20% of the price disparity every month. In contrast, the United States price does not show any significant movement at all. The Australian and Argentine prices also respond to the U.S.-Canada relationship, which indicates tht the Canadian price has a role in setting prces in those markets. The total speed of adjustment between the U.S. and Canada is 0.19, accounted for entirely by Canada. The speed of correction means a shock that pulls the markets out of equilibrium has a halflife of 3.3 months. A visual representation of the adjustment over three years is given in the first graph in Figure 5. The second column in α corresponds to the relationship between PU S and PAU S . In this relationship, both prices move slowly, with the American price adjusting more than the Australian price. These adjustment rates indicate that the Australian market acts as a price leader, especially given the adjustment values corresponding to the other markets. All wheat prices show evidence of correcting in the direction of the

Results

Shenoy 37

Figure 5: Adjustment rates between pairs of markets out of equilibrium. The graphs only indicate the speed of adjustment relative to the size of the price difference, and the magnitude of the units have no meaning.

Results

Shenoy 38

Australian price when it is not in equilibrium, meaning it has a strong role in setting the world price. The total speed of adjustment in the two markets is 0.094, which suggests that shocks to the system last longer, with a halflife of 7.02 months. A visual representation of the adjustment over three years is given in the second graph in Figure 5. The third column in α corresponds to the relationship between PU S and PARG . Like the Canadian market, the Argentine market reacts quickly to price disparities, while the American market shows no signs of moving at all. No other markets respond to this relationship, meaning Argentina is isolated as a price taker. The nation receives price signals from other trading centers, but domestic fluctuations do not affect other world prices. The total rate of adjustment between the U.S. and Argentine market is just 0.19, and any shocks that draw markets out of equilibrium have a halflife of 3.3 months. A visual representation of the adjustment over three years is given in the third graph in Figure 5. The adjustment coefficients paint a picture of a world in which the Australian price leads the market. All markets respond to both PU S and PAU S , but they generally correct in the direction of Australia. PCAN has an impact on the Argentine market and a slight impact on the Australian market as well. The Argentine price has no influence on any other market and does all of the adjusting to meet the world equilibrium. The Indian market is not integrated with the other four so it does not necessarily revert to any equilibrium condition, but it is not entirely independent. The coefficient on PIN D in the second relationship is significant, which means that world prices have an effect on the Indian price. The equilibrium price of wheat in India responds to the world price but does not necessarily settle at the world level. Using the adjustment vectors, I can compute the orthogonal complements to calculate the long-run impacts of a stochastic shock. With β fixed, it is easy to define its orthogonal complement as

Results

Shenoy 39

 1 1  β⊥ =  1 1 0

 0 0  0  0 1

(14)

This matrix places the four integrated markets together in one relationship and the Indian market apart in a separate relationship. The matrix α⊥ is generated using α ˆ from the restricted estimation. With the orthogonal complements, I estimate the matrix C from (11). The matrix describes the change in the long-run value of the prices given a shock in the current period. The four integrated prices exist in a [1, −1] equilibrium, so a shock to any one price will affect all four in exactly the same way. The Indian market operates independently in this system so I cannot model the impact of a shock in India on the world equilibrium or vice versa. Because of the low frequency of my data, it makes sense to adjust Cˆ to include correction in between periods. When a shock occurs in time t, it has already trickled into other markets by time t + 1. I compute the matrix C˜ to describe projected longrun responses as the product of Cˆ and a matrix of expected errors. The expectation matrix has 1 along its diagonal and δi,j at every other value, where δi,j = E[εi |εj = 1]. This adjusted matrix estimates the change in a long-run relationship caused by a shock in a variable taking into account the correction that occurs before the next observation. Cˆ assumes no effect on εi,t from εj,t over the course of a month and C˜ assumes that εi,t can be explained entirely by εj,t. The actual long-run impact of a ˜ stochastic shock lies somewhere between Cˆ and C. The two matrices show that shocks in Australia have the largest impact on the world system, and those in the United States have the second largest impact. This result is expected given the prominent role of the two markets predicted by α. ˆ The Argentine market has the smallest predicted impact, and the coefficients from α ˆ

Results

Shenoy 40

Long-Run Impacts Effect of: εU S εCAN εAU S εARG εIN D

ˆ Minimum impact on C: PU S PCAN PAU S PARG PIN D 0.48 0.48 0.48 0.48 – 0.059 0.059 0.059 0.059 – 0.93 0.93 0.93 0.93 – 0 0 0 0 – – – – – 1.27

PU S 1.12 0.82 1.50 0.57 –

˜ Maximum impact on C: PCAN PAU S PARG PIN D 1.12 1.12 1.12 – 0.82 0.82 0.82 – 1.50 1.50 1.50 – 0.57 0.57 0.57 – – – – 1.28

Table 7: Long-run equilibrium response to shocks in period t. The left table presents lower bounds and the right table presents upper bounds.

suggest that the actual value is at the lower bound of 0. Cˆ and C˜ also predict that shocks in the Indian market have a feedback effect that leads to a new domestic equilibrium farther away from the old one. The a shock in time t is multiplied by almost 1.3 in the long run. Full results of Cˆ and C˜ are presented in Table 7. When markets face a shock that drives them toward a new equilibrium, they respond at different speeds. The adjustment a market makes in period t + i in response to a shock in period t can be given as a portion of the total long-run adjustment as described by (12). The United States, Canadian, Australian, and Argentine markets adjust very quickly; they all reach to within 10% of their total adjustment in three months and to within 5% in eleven months. The American and Argentine markets actually overshoot the final equilbrium early on and then change direction of movement to meet the Canadian and Australian markets at the equilibrium point, although the overshooting is small and may be an artifact of noisy data. In contrast, the Indian market moves much more slowly. After an initial jump to 55% of the final equilibrium, it takes another 18 months for the price of wheat to reach within 10% of the final value. After three years, the price of wheat in India has still only covered 97% of its total predicted movement. A graphical representation of the adjustment speeds of the five markets is given in Figure 6. It is important to note that even though the Indian market appears on the same graph as the others, it does not follow the same

Results

Shenoy 41

Figure 6: Adjustment rates in response to a new equilibrium. The Indian market responds much more slowly than the other four.

trend. The four export markets all move concurrently toward a common equilibrium, but the Indian market operates separately.

7

Discussion

I find significant differences between the way prices behave in the Indian market and the way they behave in other world markets. These differences might be a result of Indian government policies or of inefficiencies in the Indian domestic market.

7.1

Interpretation of Results

My main finding is that the price of wheat in India does not converge to the international level. There is evidence of two common trends among the markets I study, and the most likely interpretation is that one trend guides export prices in the United States, Canada, Australia, and Argentina while the other leads the Indian market. Among the world export centers, Australia seems to have the most dominant role;

Discussion

Shenoy 42

other prices adjust to match the Australian price when the system is out of equilibrium. The American wheat price also drives prices in other markets, though it adjusts more rapidly than the Australian price to eliminate discrepancies between the two. Interestingly, the price of wheat in Canada shows evidence of influencing prices in Argentina and, to a lesser extent, in Australia, but not in America. Argentina acts as a price taker in this market and corrects in response to price differences, but has no influence on other nations. Estimation reveals a complex system of linkages guiding the trade of wheat. Although the Indian market does not show evidence of convergence with other markets, it does react to changes in the international system. When the the American and Australian markets, the two leading price setters, are out of equilibrium, prices in India adjust. Periods of disequilibrium among the large centers correspond to rising or falling world prices, when world trade places the most pressure on the Indian market. I find evidence that the Indian wheat market adjusts much more slowly to shocks than the other four markets, taking almost ten times as long to close 90% of a gap between current price and expected equilibrium price. Despite the slow pace at which the Indian price moves, the total adjustment in response to a shock is greater than the adjustment of the world markets in response to shocks in any individual market. A short-run fluctuation in the Indian market is inflated 130% in the long run, more than the lower bound of responses to shocks in any of the other markets and more than the upper bound in all markets except Australia. Overall, the wheat price in India does not converge to the international level, but it remains susceptible to changes in the world market and adjusts significantly more in response to shocks.

Discussion

Shenoy 43

7.2

Possible Explanations

A couple of factors within India may account for the behavior of the wheat price. First, government policy limits the role of market forces. The philosophy behind government action motivates its insulating effect. To support low-income consumers every year, the FCI sets a procurement target given requirements for public distribution and maintenance of stocks, and then enters the market to meet its target. The institution also acts in defense of producers by creating a price floor for wheat sold at mandis. Because the state deals in such a significant portion of the annual wheat crop, state action almost certainly influences the domestic price. On top of its activities in the market, the Indian government sets import and export restrictions to stop prices from getting too high or too low. Such limitations prevent traders from taking advantage of arbitrage opportunities arising from price differences, meaning that price signals may not be effectively transmitted through the market. Quantitative restrictions tend to be immediate responses to political pressure and not long-term strategies for price control. Once the immediate backlash to rapid movements lightens up, the Indian government has backed away from trade regulations. This immediate policy response causes the Indian price to move more slowly toward new equilibria. The intervention comes at a high cost, as the government directly spends more than $4 for every dollar of grain distributed to consumers on top of the additional burden of taxation and distortion of incentives (Persaud and Rosen, 2003). In the future, such intervention will become increasingly difficult as higher prices raise the opportunity cost of not exporting and make state imports more expensive. Thus far, however, the government has demonstrated its willingness to take on the costs of protecting the domestic market. In the autumn of 2007, for instance, the STC signed a deal to import wheat at almost 20% more than the prevailing domestic price in order to prop up domestic supply. As long government

Discussion

Shenoy 44

officials continue to intervene in this manner, the Indian wheat market may remain separated from the rest of the world. The behavior of the government in response to price changes helps explain the slow adjustment shown by the Indian market to new equilibrium conditions. The state manipulates buffer stocks and foreign trade to ease the pace of price movement. These policies can influence the rate at which the market approaches a new equilibrium, but they have little influence over the new equilibrium itself. In the absence of other stabilizing markets to dilute the impact of shocks, the Indian wheat price faces the full effect of a short-term movement and any feedback it may have. Lack of regional integration within the Indian market can contribute to the slow movement of prices. Markets in the nation are not well integrated, so price signals from foreign trade may halt at the border without penetrating further into the country. The price of wheat in India may be slower to respond to new equilibrium conditions simply because market information takes more time to disseminate from one region to another. A new national price would not prevail until the signal has had time to trickle down through a significant portion of the country. This gradual pace of adjustment would also limit evidence of integration with international markets. If the Indian price is responding slowly to volatile export prices in other nations, then evidence of that response and correction will be difficult to uncover. The inertia in the Indian wheat market due to segmentation has its roots in the nature of production. As discussed in Section 3, the majority of the sector is composed of farmers with few resources and little market power. These farmers often lack both the means and the information to respond effectively to price signals. Many Indian producers do not have the capacity to store their crop or transport it over long distances, so their options for sale are limited. They are forced to sell at the local mandi to whatever buyers appear on the day they bring their harvest to market regardless of potential opportunities elsewhere. Limited availability of market

Discussion

Shenoy 45

information adds risk to the sale process. If farmers do not know the price prevailing at various trade centers, then they face the potential of wasting transportation costs on a market with lower prices. Factors limiting the ability of producers in India to make economic decisions with ample information may explain the lack of integration between Indian and world wheat markets.

8

Conclusion

In this paper, I identify three stylized facts that characterize the Indian wheat market. First, the price of wheat in India does not converge to the world level. There is evidence that Indian price still responds to world prices, however, so the market is not completely insulated from international pressures. Second, the Indian price adjusts to changes in equilibrium much more slowly than does the price in the primary export centers of wheat. Third, despite slow price movement, the overall level of adjustment in response to stochastic shocks is much higher in the Indian market than in other world markets. These characteristics are most likely consequences of high government intervention and regional market segmentation. As a result, there is reason to expect different behavior in the future. As Indian income grows and food prices rise, the political and economic costs of policies that separate the Indian market from the rest of the world multiply. Although the state has shown a willingness to bear the costs of its policy up to now, it may not be able to do so for long. Regional segmentation may decrease in the future as well. New communication technologies facilitate the transfer of information in rural areas, and their absorption into the Indian countryside will dictate the decision-making capabilities of farmers. Investments in rural infrastructure that accompany economic development will also give wheat producers more freedom to follow market signals.

Conclusion

Shenoy 46

My findings indicate that situations like the current world food price spike are not felt as sharply in India as they are in the rest of the world. Indian producers and consumers enjoy a degree of protection both from the state and from inefficient markets, although both of these insulators carry costs as well. Government policies place high financial burdens on the population in relation to the benefits they create and poor market structure limits the basic ability of producers and consumers to respond to their environment and make economic decisions. Even though the immediate impacts of price spikes are lower in India, the final correction may be greater if the conditions persist. There is reason for concern that if the world enters a new regime of high prices, the effect on the Indian market may be greater than initial movements suggest. It is important to note that government policy is motivated by price stability, so the relationship between India and the world market may change if this turns out to be the case. Future research in this direction can focus on modeling the exact impact of world prices on the Indian price, which I am unable to do without an equilibrium condition. Such studies might include other domestic price determinants to develop a more complete understanding of how the Indian wheat market behaves. It would also be useful to incorporate the role of the government to identify what portion of my findings can be attributed to decisions by policymakers and how much is inherent in the structure of the market. My results can most likely be generalized to the Indian rice market, which is treated similarly by the state under the purvey of the FCI, but other crops are produced and handled differently. Knowledge of the differences in behavior by crop is required for a more clear picture of the Indian food economy. Further study can also focus on the price linkages between the world wheat market and prices in other developing countries. With a basket of countries and varying levels of integration and price response, it may be possible to determine what factors are most significant with regard to the behavior of prices.

Conclusion

Shenoy 47

This study has implications for Indian policymakers, outside agencies, and Indian consumers. Politicians in India craft laws in the food sector with the intent of preserving a robust domestic supply, limiting volatility, and preventing prices from settling at too high or too low a level. I find that they are somewhat successful in limiting short-term volatility and rapid change in response to external shocks, but cannot address long-run effects under current market structure. It is up to officials to determine whether these achievements justify the costs of Indian wheat policy. Outside actors may be interested in price adjustments to know where to focus attention in the aftermath of a price spike or other shock. My result suggests that immediate attention is best paid to regions more responsive to foreign markets. External actors should not ignore India entirely, however, because the ultimate magnitude of a shock is larger than the initial movement or rate of adjustment may lead one to believe. Finally, actors within the Indian market may have reason to behave differently from actors in other markets. With the current government attitude and market structure, there is low risk of rapid fluctuation and price movements tend to persist for longer periods of time. These facts about the behavior of the Indian wheat price should factor into economic decisions.

Conclusion

Shenoy 48

References Ahluwalia, M. (2000). India’s economic reforms: An appraisal. In J. Sachs and N. Bajpa (eds.), India in the Era of Economic Reforms, New Delhi, India: Oxford University Press, pp. 26–80. Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control, 19 (6), 716–723. Ardeni, P. (1989). Does the law of one price really hold for commodity prices. American Journal of Agricultural Economics, 71 (3), 661–669. Australian Wheat Board (2006). Community education: Commodity overview. URL: http://www.awb.com.au/aboutawb/communityeducation/industryoverview/ Accessed August 10, 2007. Baffes, J. and Gardner, B. (2003). The transmission of world commodity prices to domestic markets under policy reforms in developing countries. Policy Reform, 6 (3), 159–180. Bennett, M. (1941). Wheat Studies of the Food Research Institute. Tech. Rep. 12 and 18, Stanford University, Stanford, CA. Bernard, A. and Durlauf, S. (1995). Convergence in international output. Journal of Applied Econometrics, 10, 97–108. Box, G., G.E.P.and Jenkins (1970). Time Series Analysis: forecasting and control. San Francisco, CA: Holden-Day. Bukenya, J. and Labys, W. (2005). Price convergence on world commodity markets: Fact or fiction? International Regional Science Review, 28 (3), 302–329. Crain, S. and Lee, J. (1996). Volatility in wheat spot and futures markets, 19501993: Government farm programs, seasonality, and causality. The Journal of Finance, 51 (1), 325–343. Dickey, D. A. and Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association, 74 (366), 427–431. Doornik, J. (2007). Ox: An Object-Oriented Matrix Language. London: Timberlake Consultants Press. Engle, R. F. and Granger, C. (1987). Cointegration and error correction: Representation, estimation, and testing. Econometrica, 55 (2), 251–276. Ghosh, M. (2003). Spatial integration of wheat markets in india: Evidence from cointegration tests. Oxford Development Studies, 31 (2), 159–171.

REFERENCES

Shenoy 49

Goodwin, B. K. (1992). Multivariate cointegration tests and the law of one price in international wheat markets. Review of Agricultural Economics, 14 (1), 117–124. Goshray, A. and Lloyd, T. (2003). Price linkages in the international wheat market. In Proceedings of the 25th International Conference of Agricultural Economists, Durban, South Africa: Document Transformation Technologies. Government of India (2004). Agricultural Statistics at a Glance. New Delhi, India: Arihant Offset. Government of India (2007). Export import data bank. http://commerce.nic.in/eidb/ecomcntq.asp Accessed July, 2007. Granger, C. and Newbold, P. (1974). Spurious regressions in econometrics. Journal of Econometrics, 2, 111–120. Hakkio, C. and Rush, M. (1991). Cointegration: how short is the long run? Journal of International Money and Finance, 10, 571–581. Hansen, P. R. (2002). Generalized Reduced Rank Regression. Economics Working Paper 02-02, Brown University. — (2005). Granger’s representation theorem: A closed-form expression for i(1) processes. Econometrics Journal, 8, 23–38. — (2008). Reduced-Rank Regression: A Useful Determinant Identity. Research Paper 2008-2, Center for Research in Econometric Analysis of Time Series. — and Lunde, A. (2006). Realized variance and market microstructure noise. Journal of Business and Economic Statistics, 24, 127–161. Jha, S., Srinivasan, P. and Landes, M. (2007). Indian Wheat and Rice Sector Policies and the Implications of Reform. Economic Research Report 41, United States Department of Agriculture Economic Research Service. Johansen, S. and Juselius, K. (1990). Maximum likelihood estimation and inference on cointegration - with applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52 (2), 169–210. Krugman, P. (2008). Grains gone wild. The New York Times, editorial Column, April 7. Kwiatkowski, D., Phillips, P., Schmidt, P. and Shin, Y. (1992). Testing the null hypothesis of stationarity against the alternative of a unit root. Journal of Econometrics, 54, 159–178. Larue, B. (1991). Is wheat a homogenous product? Canadian Journal of Agricultural Economics, 39 (1), 103–117.

REFERENCES

Shenoy 50

MacKinnon, J. (1996). Numerical distribution functions for unit root and cointegration tests. Journal of Applied Econometrics, 11, 601–618. Mohanty, S., Peterson, E. and Smith, D. (1996). Relationships between u.s. and canadian wheat prices: Cointegration and error correction approach. Canadian Journal of Agricultural Economics/Revue, 44 (3), 265–276. Moon, B. K. (2008). Address to the Special High-level Meeting of ECOSOC with the Bretton Woods Institutions, the WTO and UNCTAD. Morisset, J. (1998). Unfair trade? the increasing gap between world and domestic prices in commodity markets during the past 25 years. World Bank Economic Review, 12 (3), 503–526. Mundlak, Y. and Larson, D. (1992). On the transmission of world agricultural prices. World Bank Economic Review, 6 (3), 399–422. Naylor, R., Liska, A., Burke, M., Falcon, W., Gaskell, J., Rozelle, S. and Cassman, K. (2007). The ripple effect: Biofuels, food security, and the environment. Environment, 49 (4), 30–43. Organization for Economic Cooperation and Development (2007). Agricultural Policies in Non-OECD Countries: Monitoring and Evaluation. Paris, France: OECD Publishing. Osterwald-Lenum, M. (1992). A note with quantiles of the asymptotic distribution of the maximum likelihood cointegration rank test statistics. Oxford Bulletin of Economics and Statistics, 54 (3), 461–472. Persaud, S. and Rosen, S. (2003). India’s Consumer and Producer Price Policies: Implications for Food Security. Food Security Assessment GFA-14, United States Department of Agriculture Economics Research Service. Phillips, P. and Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75 (2), 335–346. Quiroz, J. and Soto, R. (1993). International Price Signals in Agricultural Markets: Do Governments Care? Working Paper 088, Georgetown University. Ravallion, M. (1986). Testing market integration. American Journal of Agricultural Economics, 68 (1), 102–109. Reserve Bank of India (2006). Handbook of Statistics on the Indian Economy 2005-2006. Mumbai, India: Reserve Bank of India. Roy, D. (2007). Wheat imports and food security. The Hindu, august 10. Said, S. and Dickey, D. (1984). Testing for unit roots in autoregressive-moving average models of unknown order. Biometrika, 71 (3), 599–607.

REFERENCES

Shenoy 51

Sharma, G. (). Compilation of consumer price indices (cpi) and wholesale price index (wpi) in india. Srinivasan, T. (2007). ”market integration, transaction costs, and the indian wheat market - a systematic study. Stock, J. and Watson, M. (1988). Testing for common trends. Journal of the American Statistical Association, 83 (404), 1079–1107. Thomas, S. (2003). Agricultural Commodity Markets in India: Policy Issues for Growth. Tech. rep., Indira Gandhi Institute for Development Research, Mumbai, India. Time Series Staff (2007). X-12-ARIMA Reference Manual. Washington, DC: U.S. Census Bureau. Timmer, P. (2000). The macro dimensions of food security: Economic growth, equitable distribution, and food price stability. Food Policy, pp. 283–295. — (2008). Asian rice crisis puts 10 million or more at risk, q and A Session. U.S. Department of Agriculture (2000). India wheat: Producing regions - gis applications. URL: http://www.fas.usda.gov/remote/india/wheat2000/wht map/wht map.html Accessed: July, 2007. U.S. Department of Agriculture (2008). Wheat Yearbook. United States Department of Agriculture. Wilson, W. (1989). Differentiation and implicit prices in export wheat markets. Western Journal of Agricultural Economics, 14 (1), 67–77. World Bank (2008). Rising food prices: Policy options and world bank response, informational document. Yang, J., Bessler, D. and Leatham, D. (2000). The law of one price: Developed and developing country market integration. Journal of Agricultural and Applied Economics, 32 (3), 429–440. Yavapolkul, N., Gopinath, M. and Gulati, A. (2006). Post-uruguay round price linkages between developed and developing countries: the case of rice and wheat markets. Agricultural Economics, 34, 259–272.

REFERENCES

Shenoy 52