0 оценок0% нашли этот документ полезным (0 голосов)
15 просмотров16 страниц
We present strong evidence against the excess-comovement hypothesis-that the prices of commodities
move together beyond what can be explained by fundamentals. Prior studies employ broad macroeconomic
indicators to explain common price movements, and potentially correlated fundamentals
are not controlled for. We use inventory and harvest data to fit a partial equilibrium model that more
effectively captures the variation in individual prices. The model explains the majority of the comovements
among commodities with high price correlation, and all of the comovements among those with
marginal price correlation. Common movements in supply factors appear to play an important role in
the observed comovements in commodity prices.
We present strong evidence against the excess-comovement hypothesis-that the prices of commodities
move together beyond what can be explained by fundamentals. Prior studies employ broad macroeconomic
indicators to explain common price movements, and potentially correlated fundamentals
are not controlled for. We use inventory and harvest data to fit a partial equilibrium model that more
effectively captures the variation in individual prices. The model explains the majority of the comovements
among commodities with high price correlation, and all of the comovements among those with
marginal price correlation. Common movements in supply factors appear to play an important role in
the observed comovements in commodity prices.
We present strong evidence against the excess-comovement hypothesis-that the prices of commodities
move together beyond what can be explained by fundamentals. Prior studies employ broad macroeconomic
indicators to explain common price movements, and potentially correlated fundamentals
are not controlled for. We use inventory and harvest data to fit a partial equilibrium model that more
effectively captures the variation in individual prices. The model explains the majority of the comovements
among commodities with high price correlation, and all of the comovements among those with
marginal price correlation. Common movements in supply factors appear to play an important role in
the observed comovements in commodity prices.
Author(s): Chunrong Ai, Arjun Chatrath and Frank Song Source: American Journal of Agricultural Economics, Vol. 88, No. 3 (Aug., 2006), pp. 574-588 Published by: Oxford University Press on behalf of the Agricultural & Applied Economics Association Stable URL: http://www.jstor.org/stable/3697750 . Accessed: 03/06/2014 21:56 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. . Agricultural & Applied Economics Association and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend access to American Journal of Agricultural Economics. http://www.jstor.org This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions ON THE COMOVEMENT OF COMMODITY PRICES CHUNRONG AI, ARJUN CHATRATH, AND FRANK SONG We present strong evidence against the excess-comovement hypothesis-that the prices of commodi- ties move together beyond what can be explained by fundamentals. Prior studies employ broad macro- economic indicators to explain common price movements, and potentially correlated fundamentals are not controlled for. We use inventory and harvest data to fit a partial equilibrium model that more effectively captures the variation in individual prices. The model explains the majority of the comove- ments among commodities with high price correlation, and all of the comovements among those with marginal price correlation. Common movements in supply factors appear to play an important role in the observed comovements in commodity prices. Key words: commodity prices, comovement, herding. Pindyck and Rotemberg (1990) find that prices of seemingly unrelated commodities move to- gether, even after controlling for macroeco- nomic indicators such as inflation, industrial production (IP), and interest rates. The au- thors regress the price changes of seemingly unrelated commodities (wheat, cotton, copper, gold, crude, lumber, and cocoa) on some im- portant macroeconomic indicators and find the regression residuals to be highly correlated. Their finding, subsequently well known as the "excess comovement hypothesis" (henceforth ECH), calls into question the rationality of commodity markets and flies in the face of the competitive model of price formation. For in- stance, Pindyck and Rotemberg (1990) (PR) suggest that the excess comovements may be due to herding-where traders alternately buy or sell different commodities at the same time, with little economic justification. The ex- cess comovement of prices could impede the decision-making abilities of hedgers and fore- casters, who base their decisions on fundamen- tals, and could imply that countries exporting a portfolio of seemingly unrelated commodities enjoy only limited diversification of revenues. Since PR, several researchers have revis- ited the ECH employing a variety of data and test procedures. Notably, Deb, Trivedi, and Varangis (1996) document that the PR results are sensitive to the neglected structural changes in prices (in the 1970s), and to the con- trols for conditional heteroskedasticity in the price data. The authors suggest that the inap- propriate assumption of normality in the PR regression residuals cause the false appearance of excess comovements. Other researchers have also shown the PR evidence to be sen- sitive to methods. Palaskas and Varagis (1991) employ cointegration analysis to show that ex- cess comovements are the exception rather than the rule in twenty-one pairs of monthly and annual prices. Malliaris and Urritia (1996) employ cointegration analysis to reject the long-term independence of six commodity fu- tures price series. Cashin, McDermott, and Scott (1999) employ a nonparametric mea- sure of comovement, concordance, suggested by Pagan (1999) and find little evidence of syn- chronocity in the turning points in prices of seven commodities over the PR sample period. Taken together these studies seem to suggest that ECH is the artifact of econometric model- ing, and if the right econometric model could be discovered, the evidence of excess comove- ments would disappear. Thus, the research on ECH has focused more on the nature of the comovements, rather than the causes themselves.1 For example, none of the stud- ies explain the poor explanatory powers of the macroeconomic indicators or explore the Chunrong Ai is associate professor, Department of Economics, University of Florida and School of Economics, Hvazhang Univer- sity of Science and Technology, China. Arjun Chatrath is associate professor, School of Business, University of Portland. Frank Song is associate professor, School of Economics and Finance, Hong Kong University. The authors wish to thank the two anonymous reviewers for their useful comments and suggestions. Remaining errors are our own. 1 For instance, Deb, Trevedi, and Varangis (1996) provide results from GARCH models and note that only upon "controlling" for the covariance process of price changes does the covariance in stan- dardized residuals tend to dissipate. Arguably, such a specification captures the symptoms of the underlying covariance in prices, and not the fundamental causes themselves. Amer. J. Agr. Econ. 88(3) (August 2006): 574-588 Copyright 2006 American Agricultural Economics Association This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Ai, Chatrath, and Song Comovement of Commodity Prices 575 possibility that observed comovements are caused by fundamentals beyond these indica- tors. Intuitively, it is unreasonable to expect that broad economic indicators employed in PR will capture or reflect all or even the ma- jority of the supply or demand conditions in individual commodities. Moreover, it is likely that such indicators will reflect demand condi- tions better for some commodities than others. For instance, IP may be more closely re- lated to the consumption demand for lumber or copper than for oats or barley.2 In fact, there is much evidence that macroindicators explain very little of the variation in com- modity prices. Studies on the macroeconomic risk-premiums in commodity prices, including Park, Wei, and Frecka (1988), Bessembinder and Chan (1992), Bailey and Chan (1993), and Bjornson and Carter (1997) find PR-like models to perform very poorly. For instance, the largest adjusted R-squared statistic in Bjornson and Carter (1997) is only 3.3%. In contrast, there is evidence that fundamental factors such as the weather have a relatively large impact on individual commodity price behavior (for instance, Roll 1984; Brunner 1998; Deaton and Laroque 2003). It is worth noting that PR's inference that price comovements may be driven by specu- lative herders can also be challenged by the literature. Given that commodity price co- movements appear to persist during periods of booms and busts, their inference would seem to imply that speculators play a ma- jor role in the behavior of commodity prices. While there is some indication that specula- tive behavior, particularly in futures markets, may result in increased volatility, the weight of the evidence is that speculation has either an ambiguous- or dampening impact on the variation in commodity prices (for instance, see Chari and Jagannathan 1990; Netz 1995; Zulauf and Irwin 1998; Carter 1999; Chatrath and Song 1999; Irwin and Holt 2004). In sum- mary, PR may be premature in concluding that their findings of persisting comovements are "excessive," and implying that herding may be the prominent cause for the comovements. Consequently, further empirical work in this small literature on the relatedness of commod- ity prices is warranted. The primary objective of this article is to ex- amine the extent to which the observed co- movements in the prices of commodities can be explained by the relatedness of their funda- mentals. Specifically, we reexamine the ECH employing commodity-specific data such as production and inventories, in conjunction of the traditional macroeconomic indicators, to more completely control for the relatedness in the demand and supply of the commodi- ties. The main focus of our investigation is on five commodities-wheat, corn, oats, soy- beans, and barley, for which fairly detailed commodity-fundamental information is avail- able. With the exception of wheat, these com- modities are different from those studied in PR. However, as demonstrated shortly, these seemingly unrelated commodities exhibit sim- ilar amplitude of comovement, that is, they have "excess comovement," as defined in PR. The commodity specific (market-level) data that are employed in this study allow us to make improvements on the tests for price comovements in two ways. First, the market- level variables allow us to test for excess co- movements while maintaining a low reliance on presumptions vis-d-vis the relatedness of the commodities. In contrast, PR presuppose the fundamentals for commodities such as wheat, cotton, and cocoa are unrelated beyond the general economic cycles.3 Second, the data allow us to develop relatively direct proxies of demand/supply conditions as are required for effective testing of excess comovements. These data also allow us to make inferences on the relative roles of supply and demand factors in the observed correlations in commodity prices. The relative contributions of supply and de- mand have been studied with respect to indi- vidual commodity price behavior (for instance, Myers and Runge 1985), but not in the context of commodity comovements. The findings in this article are summarized as follows. (i) The correlation of commodity prices remained high for the latter half of the twentieth century. (ii) The macro indicators such as IP and gross domestic product (GDP) fail to explain these correlations, consistent with PR. (iii) The market-level indicators such as inventory and harvest size, in conjunction with the macro indicators, explain a strikingly 2 For instance, in PR, the macro indicators and their lagged val- ues explain less than 10% of the variation in four of the eight commodity prices studied. The indicators were most successful in explaining the variation in gold and crude price changes (adjusted- R2 of 0.24 and 0.21), and least successful for cotton and wheat (0.05 and 0.06). 3 Similarly, the filter employed for unrelated commodities in Deb, Trivedi, and Varangis (1996) is that they are neither jointly produced nor jointly consumed. By their metric, sugar is unre- lated to coffee or cocoa, and lumber and oil are unrelated to each other and to nine other commodities, including, wheat, copper, and cotton. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 576 August 2006 Amer. J. Agr. Econ. large portion of price movements, and ex- plain the majority of correlations in commod- ity prices. (iv) Supply factors appear to play a relatively large role in the observed correla- tions, at least for commodity pairings such as wheat and oats, and soybeans and corn. In the next section we develop a partial equi- librium model (Equilibrium Model), which will accommodate the market-level and macro data. The Data section describes the price and fundamentals data employed in the study. The Empirical Results section begins by presenting correlation results for prices and fundamental factors. The main results are then presented to compare the "Macro Model" similar to PR, and our "Equilibrium Model," one that ac- commodates the market-level data. The com- parison of these two frameworks is broadened to twenty other commodities, agricultural, and otherwise. Finally, an examination into the rel- ative role of supply factors in the observed comovements is undertaken. The final section summarizes the findings. Empirical Implementation Let pi and p1 represent the log price histories of two commodities, i and j, and X a matrix of macroeconomic indicators such as GDP and interest rates. The PR method of testing for pair-wise excess-comovements is based on the residuals (ut) from the regressions (1) pi,t = biXt + ui, Pj,t = bjXt + Uj,t where b is a vector of sensitivities and u, is the regression error term. Typically, first differ- enced prices and economic indicators are em- ployed. In the interest of exposition, we deal in the level series for now. In PR, the ECH is supported if p {uit, ui, t} > 0.4 It is clear that commodity-specific factors, such as inventories or production are not considered in (1), mainly because commodities i and j are presumed en- tirely unrelated in their fundamentals. We refer to (1) as the Macro model. Now consider a framework where individual prices are determined in equilibrium. Because total supply for a nontraded commodity with a single harvest season is inelastic at t, the de- mand function can always be estimated from data. Let the inverse demand function be given by (2) pt = f(D,, Xt) + Et where Dt is the consumption at t, and Et is the unexplained portion of the current price, and f is the function to be estimated. Prices are influ- enced by quantity demanded and the general economic conditions. Thus the macro variables (X,) are modeled as demand shifters. Let zt represent the harvest size and It the inventories for a commodity at time t. The mar- ket clears at Dt + I, = total supply, st = zt + (1 - 8)It-1, where 8 is the per period deterio- ration rate of inventories. From (2) we have (3) pt = f(st-It, Xt) + Et-. Equation (3) is a partial equilibrium formula- tion that considers the effects of both current and expected demand and supply conditions.5 Note that the variable (st - It) generally repre- sents the commodity specific variables missing in (1). In the theory of commodity prices, inven- tories are endogenous (for instance, Williams and Wright 1991; Deaton and Laroque 1992; Chambers and Bailey 1996), and the above demand function alone cannot explain price behavior unless inventory is explained. How- ever, the objective of this article is not to model inventory per se. Our objective is to estimate the demand function employing the observed supply and inventories, and analyze the extent to which these variables explain the common movements in commodity prices. The advan- tage of not modeling inventories is that we do not have to deal with the econometric issue relating to the nonnegativity constraint for in- ventories. In this respect, note that equation (3) holds whether or not the nonnegativity con- straint is binding. The disadvantage of course is that we cannot explain common movements in inventories (if any) across commodities. Be- cause inventory is not explicitly modeled, our 4 Note that zero correlation between the error term in (1) will not represent absolute evidence of a lack of excess comovement as long as all factors have not been controlled for. It is possible that further controls will reveal an underlying relationship that is not evident in the residuals in (2). Naturally, the more compre- hensive the controls, the less the chance of reaching an erroneous conclusion. 5 Large commodity price movements have been associated with both demand and supply shocks. For instance, Gilbert (1989) ar- gues that the dollar's appreciation in the early 1980s magnified the debt obligations of developing countries and induced an increase in the supply of commodities, which resulted in the decline in real dollar prices (also see Deaton and Miller 1996). Dornbusch (1985) argues that an appreciation of the dollar reduces the demand in the rest of the world, and results in a decline in the commodity's real market-clearing price in U.S. dollars. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Ai, Chatrath, and Song Comovement of Commodity Prices 577 study may be considered a partial equilibrium analysis. To estimate equation (3), we must: (i) model the functional form off(.), and (ii) address the endogeneity of the inventory variable. Since our objective is to identify the sources and not the manner of comovement, we do not want our analysis to be impacted by the misspeci- fication of the functional form. To avoid mis- specification error, we adopt the flexible model (4) Pt = to + ai,1(st - It) + ' - + al,n(st - I,)m + O2,1(St - It)Xt + "- " + Z2,n(St - t)kXt + -3Xo + X where the orders of m and k are determined by the data through a cross-validation approach that is common in the nonparametric litera- ture (e.g., Ai and Chen 2003). The endogene- ity of the inventory variable will be addressed by an instrumental variable approach with 1, St, s$,...S"+l, stXt, s2Xt,...sk+ltX as instru- ments. As with equation (1), equation (4) may be estimated with the variables in their levels or in their first differenced forms. For expo- sition, (4) will be referred as the Equilibrium model. A comparison of the explanatory pow- ers of the Equilibrium model and the tradi- tional Macro model will indicate the degree to which prior studies on excess comovements suffer from biases resulting from missing vari- ables. Similar indications may be gleaned from a comparison of the pair-wise correlation of the residuals from the alternate models. Data and Empirical Results Data The majority of the empirical tests in this study are conducted on quarterly data for five commodities-wheat (all), barley (all), corn (for grain), oats, and soybeans, from January 1957 to September 2002. The study spans beyond the sample intervals in PR (1960- 85), and Deb, Trevidi, and Varangis (1960-85 and 1974-92). Our attention is mainly on the five commodities since detailed and lengthy market-specific data are unavailable for other important commodities, such as cotton or lum- ber. Quarterly sampling of prices is employed since commodity-specific data (production, in- ventories, etc.) are sparse for finer intervals. It may be argued that the quarterly sampling makes the rejection of excess comovements more stringent: it is well documented that the correlations in price changes tend to be smaller in higher frequency data (for instance, see PR). The data on U.S. prices, inventories (on- farm, off-farm, total), harvest size, yield per acre, and planted acres for the five commodi- ties are obtained from the U.S. Department of Agriculture.6 These commodities have well known sources of supply, namely carried-over inventories and harvest. We do not directly control for government stockholdings. How- ever, we do indirectly assess the extent to which such stockholdings have altered the nature of commodity comovements by com- paring our results from the full sample with that of the post-1972 sample, over which gov- ernment stockholdings of commodities are known to have become pervasive (for instance, Westcott, and Hoffman 1999; Goodwin, Schnepf, and Dohlman 2005). With the ex- ception of oats, there are no (or very minor) U.S. imports of these commodities. We do not differentiate between net exports and domes- tic disappearance. The macroeconomic indica- tors employed in the paper are IP, the (GDP), consumer price index (CPI), three-month sec- ondary market Treasury bill yield (r), and the broad dollar index (FXR). These data are ob- tained from the Federal Reserve Bank data files. Pindyck and Rotemberg (1990) also em- ploy the S&P 500 Stock index but do not find a significant impact on commodity prices. Finally, our study also employs price histo- ries of twenty other commodities. These data are obtained from the files of the International Monetary Fund and are detailed in the Ap- pendix. While good fundamental data are not available for these commodities, their price histories allow us to comment on the gener- ality of our main findings. In that respect, it is noteworthy that the twenty-five commodities in this study include four commodities studied in Pindyck and Rotemberg (1990)-namely wheat, cotton, copper, and cocoa. Quarterly prices (beginning for March, June, September, and December) for the twenty five commodi- ties are obtained by averaging over monthly prices.7 Commodity Correlations We begin our empirical analysis by examin- ing correlation patterns across five agricultural 6 For oats, inventory data are incomplete between 1986 and 1989. All tests for oats (including bivariate correlations) are conducted with this gap in data. 7 Similar results are obtained when employing prices rather than average prices. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 578 August 2006 Amer. J. Agr. Econ. Table 1. Correlation in Production Variables Wheat Barley Corn Oats Soybeans A. Changes in yield Wheat 1 0.535 0.227 0.521 0.201 Barley 0.537 1 0.431 0.541 0.224 Corn 0.227 0.430 1 0.589 0.828 Oats 0.522 0.543 0.589 1 0.539 Soybeans 0.197 0.218 0.827 0.536 1 B. Changes in planted acres Wheat 1 -0.174 0.412 -0.401 -0.055 Barley -0.178 1 0.226 0.253 0.142 Corn 0.407 0.225 1 -0.649 0.179 Oats -0.425 0.225 -0.633 1 -0.260 Soybeans -0.034 0.137 0.178 -0.277 1 Note: Pearsonian correlation coefficients are reported for changes in yield per acre and planted acres sampled annually for five commodities from 1957 through 2002. Coefficients above the diagonal relate to the de-trended changes. commodities. Our discussion of significance levels for a bivariate correlation coefficient r is based on the statistic t = r/n - 2//1 - which is asymptotically t-distributed with n - 2 degrees of freedom, where n is the sam- ple size. For the full sample, this implies that absolute correlation coefficients of about 0.12 and 0.19 are significant at the 10% and 1% levels, respectively. For price changes and their regression residuals, we also compute Spear- man Rank correlations. Commodity prices are characterized by frequent price jumps so that the price changes are typically fat tailed. The Spearman correlations employ the difference in the ranking of a variable, so that they are relatively immune to extreme outliers. Table 1 reports the cross-commodity cor- relations of some production related funda- mentals. It is apparent that the yield per acre is closely related with correlation ranging from about 0.20 for wheat-soybean to 0.83 for soybeans-corn. One can expect yield changes to be positively related because trends in pro- duction technologies are often related across commodities. However, other factors (such as the weather) seem to be at play. Specifically, the correlations remain high for detrended changes in yield (coefficients presented above the diagonal). On the other hand, the correla- tion coefficient for planted acres ranges from a very negative -0.63 (corn, oats) to a highly positive 0.41 (wheat, corn).8 These patterns continue to be supported for the detrended data. Table 2 (Panel A) reports the correlation co- efficients for the actual supply of the commod- ity. The coefficients for changes in supply are strikingly high for some of the pairings, close to 0.90 or above for wheat-barley, oats-barley, and soybeans-corn. Very negative coefficients are noted for the other pairings. These pat- terns persist for the detrended data. Similarly wide-ranging correlations are seen for changes in inventories (Panel B). The coefficients for changes in disappearance are relatively mod- est, with the exception of pairings of wheat, oats, and barley, possibly reflecting their com- plementarities (Panel C).9 We find the corre- lations in table 1 and 2 are little-changed when the data are restricted to the post-1972 period (these results are not reported in the interest of brevity). Table 3 reports the correlation of quarterly log price changes for the five commodities over the PR sample and the full sample. As the cor- relation coefficients are fairly similar across these samples (as well as the post-1972 sam- ple), we limit our discussion to the results over the longer interval. The relationship between the commodity prices ranges from the very high to the comparatively low. The pair-wise correlation of price changes is between 0.20 (wheat-soybeans) and 0.64 (wheat-barley). All correlation coefficients are significant at the 1% level. The highest coefficients are seen for the pairings involving corn and the small- est for those with soybeans. Rank correlations, discussed shortly, provide similar indications. 8 It should be noted that the relationship between prices and planted acres is likely influenced by government programs such as acreage control. 9 We assume the deterioration rate (8) to be minor and drop it from our framework. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Ai, Chatrath, and Song Comovement of Commodity Prices 579 Table 2. Relatedness in Supply, Disappearance, and Inventories Wheat Barley Corn Oats Soybeans A. Changes in supply Wheat 1 0.946 -0.460 0.733 -0.440 Barley 0.946 1 -0.426 0.887 -0.396 Corn -0.459 -0.426 1 -0.306 0.976 Oats 0.732 0.887 -0.306 1 -0.265 Soybeans -0.440 -0.400 0.977 -0.265 1 B. Changes in Inventories Wheat 1 0.912 -0.285 0.717 -0.246 Barley 0.913 1 -0.250 0.887 -0.205 Corn -0.286 -0.250 1 -0.147 0.941 Oats 0.717 0.888 -0.146 1 -0.100 Soybeans -0.246 -0.202 0.940 -0.095 1 C. Changes in disappearance Wheat 1 0.723 0.119 0.535 -0.055 Barley 0.723 1 0.165 0.741 -0.078 Corn 0.119 0.165 1 -0.059 0.438 Oats 0.535 0.740 -0.050 1 -0.137 Soybeans -0.055 -0.078 0.437 -0.137 1 Note: Pearsonian correlation coefficients are reported for changes in quarterly supply, disappearance, and inventories for five commodities from Q1/1957 through Q4/2002. Coefficients above the diagonal relate to the de-trended changes. Table 3. Correlation of Changes in Prices Wheat Barley Corn Oats Soybeans A. PR sample 1960-1985 Wheat 1 0.697 0.474 0.514 0.151 Barley 0.705 1 0.646 0.793 0.389 Corn 0.490 0.660 1 0.557 0.588 Oats 0.527 0.800 0.573 1 0.389 Soybeans 0.163 0.398 0.595 0.400 1 B. Full sample 1957-2002 Wheat 1 0.638 0.462 0.543 0.198 Barley 0.640 1 0.522 0.581 0.370 Corn 0.470 0.533 1 0.532 0.583 Oats 0.544 0.582 0.536 1 0.377 Soybeans 0.201 0.375 0.589 0.378 1 Note: Pearsonian correlation coefficients are reported for changes in log prices below the diagonal, and changes in log real prices above the diagonal. Figure 1 traces the CPI-deflated prices of the five commodities and demonstrates the relatively weaker relationship for the price of soybeans (the uppermost series), which tends to be more volatile. As is borne out by all the series, the interval 1972-74 witnessed a sharp rise in prices, followed by a compara- bly sharp fall between 1975 and 1977. The co- movements in the price series are especially high over 1972-77. Deb, Trivedi, and Varangis (1996) also note the structural instability of commodities in the early 1970s. High degrees of relatedness across the five commodities per- sist through the 1970s and 1980s, when govern- ment stockholdings started to become more pervasive, and beyond the end of the PR sample (1985), even as commodity prices and price-volatility appear to settle down. Performance of the Macro and Equilibrium Models Table 4 reports the results from the Macro model, where differenced log CPI-deflated prices are regressed on contemporaneous and lagged values of three macro indicators and two sets of {1, 0} dummies aimed at controlling for the oil-crisis period. The macro variables are: the differenced real GDP, differenced dol- lar index, and (as in PR) level interest rate. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 580 August 2006 Amer. J. Agr. Econ. 6 4 0 i L i ii i i ! 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 --- Wheat - Barley - - Corn - - Oats . Soybeans Figure 1. Quarterly real prices of five commodities Q1/1957- Q3/2002 Table 4. OLS Results of Model with Macro Indicators Wheat Barley Corn Oats Soybeans GDP 0.001 (0.21) -0.001 (-0.33) 0.001 (0.69) -0.001 (-0.35) 0.002 (1.01) GDP(-1) -0.001 (-0.07) -0.001 (-0.35) -0.002 (-1.31) -0.002 (-0.79) -0.001 (-0.69) R 0.054 (0.80) 0.042 (0.81) 0.113 (1.63) 0.033 (0.48) 0.129 (1.74) R(-1) -0.058 (-0.83) -0.056 (-1.05) -0.128 (-1.80) -0.044 (-0.62) -0.180 (-2.36) F 0.042 (0.68) 0.006 (0.13) 0.054 (0.85) 0.068 (1.07) -0.051 (-0.76) F(-1) -0.115 (-1.83) -0.069 (-1.41) -0.035 (-0.54) -0.029 (-0.45) 0.030 (0.44) D71-74 0.630 (3.28) 0.650 (4.38) 0.522 (2.66) 0.504 (2.55) 0.520 (2.48) D75-77 -0.644 (-3.16) -0.571 (-3.63) -0.493 (-2.38) -0.443 (-2.12) -0.338 (-1.52) Adj. R2 0.069 0.108 0.053 0.016 0.055 DW 1.950 1.957 1.829 1.925 1.787 Note: The dependent variable is the change in the log of real prices. GDP is changes in real GDP, R is level one-year interest rates, F is the change in the dollar index. D71-74 takes the value of 1 if the quarter falls between 1971 and 1974, and zero otherwise. D75-77 takes the value of 1 if the quarter falls between 1975 and 1977. Coefficients displayed are x 10. Figures in parenthesis are t-statistics. The sample covers Q1/1957-Q4/2002. The macro indicators explain relatively little of the variance of the price changes. The ad- justed R-squared ranges from less than 0.02 for oats to 0.11 for barley. Moreover, only the two dummies (for the 1972-74 and 1975-77 intervals) stand out as statistically significant. As in the PR study, none of the macro indica- tors is consistently important, at least in this specification. Other variables, such as lagged values of the dependent variable, IP, alternate interest rates, and narrower foreign exchange indexes, failed to improve the performance of the model. Prior to the estimation of (4), the vari- ables Dt = (st - It) and st are de- seasoned by regressing them on a constant and three quarterly dummies (Q2, Q3, Q4).10 To estimate the Equilibrium model (4), we must first determine the order of m and k. Once the order of m and k are deter- mined, we then estimate the model using two stage least squares (2SLS) estimation since inventory is endogenous. The 2SLS es- timation uses 1, s,, s ,...,s s, X,, sX,,..., sfX, as instrumental variables for D,,..., D, DXt,..., DXt. The order of m and k are de- termined through a trade-off between overfit- ting the model and obtaining improvements to its explanatory power. We employ a cross- validation approach that involves selecting the order of m and k that minimize a cri- terion akin to the sum of square errors between the fitted and refitted dependent vari- able (see Ai and Chen 2003). Specifically, 10 An alternate estimation that does not employ, the seasonal filters produced only slightly higher adjusted R statistics. By conducting these alternate regressions we assess that the role of seasonality per se in commodity comovements is relatively modest. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Ai, Chatrath, and Song Comovement of Commodity Prices 581 Table 5. Predictive Powers of Macro-indicators and Fundamentals Wheat Barley Corn Oats Soybeans A. Change log real prices 1. Macro model 0.069 0.108 0.053 0.016 0.055 2. Equil model 0.437 0.416 0.542 0.440 0.308 B. Level real prices 1. Macro model 0.580 0.719 0.656 0.661 0.598 2. Equil model 0.895 0.912 0.903 0.873 0.864 Note: The figures are adjusted R2 from the estimation of two specifications. The first is the OLS estimation of the Macro model (1); the second is the 2SLS estimation of the Equilibrium model (4). The sample covers Q1/1957-Q4/2002. we regress Pt, Dt, . . , D, DXt . . , DX, re- spectively on the instruments 1, st, s2,...,s , X,, stXt, ... ,stXt to obtain the fitted val- ues t, Dt ..., DB, DtXt ... ,kXt. We then apply ordinary least squares (OLS) to P, = "lo + Ollb + ? ? ? + -m bm + 21b)Xt + ... 2kbkX + X'3 + ut with the first observation deleted. Then com- pute the fitted value for observation 1 as p = &10 + &11Dib +... + &mDB +&21BX1 +- .. &2k kX1 + X0f3. Similarly, apply OLS with the ith (i = 2,... , N) observation deleted and then com- pute fi. The optimal m, k are those that mini- mize the criterion s(m, k) = z1 (p, - )2 Employing the approach on the differenced model, we find this criterion to decline in the order to at least m = 8, k = 8 for wheat, oats, barley, and soybeans, and up to m = 6, k = 8 for corn. The criterion provides slightly lower orders for the level series. Thus, our selection of m = k= 6 represents a conservative order of polynomial. As before, Xt contains the three current and lagged macro indicators and the two calendar dummies. The Macro and Equi- librium models were estimated for the full sam- ple, for the PR-sample, and for the post-1972 period. The results are similar so that we only report those for the longer interval. Table 5 reports the R-squared coefficients from the Equilibrium and Macro models em- ploying, alternately, level- and first-differenced variables. It is readily apparent that the Equi- librium model far outperforms the Macro model in explaining price behavior. For the first differenced specifications (Panel A), the R-squared from the Equilibrium model is be- tween 0.30 and 0.50 larger than from the Macro model, representing between a fourfold and a twenty-sevenfold increase in explanatory power. For the specifications involving vari- ables in their levels (Panel B), the R-squared from the Macro model ranges from 0.58 (wheat) to 0.72 (barley), while that from the Equilibrium model ranges from 0.86 (soy- beans) to 0.91 (barley). Figures 2 and 3 provide a graphic compari- son of the performance of the Macro and Equi- librium models for the two commodities for which the Equilibrium model performed best (worst), that is, corn (soybeans).11 It is evi- dent that the predicted values from the Macro model captures little more than the general trend in real prices: in not one of the com- modities do the predicted values trace the turn- ing points. On the other hand, the predicted values from the Equilibrium model trace the peaks and valleys remarkably well for corn, and reasonably so for soybeans. To summarize, the Macro model is almost entirely ineffec- tive in explaining commodity price behavior while the Equilibrium model explains a sub- stantial amount of price variation. As both em- ploy only "fundamental" variables, it is clear that the latter will provide a much better op- portunity to investigate the existence of excess comovements. Residual Correlations The residuals from the Macro and the Equi- librium models generally do not appear to be normal, with probability plots that indicate clustering at the tails.12 While a high frequency of outliers will not negate the correlation re- sults per se, it becomes important to get a sense of their influence on the results. Thus, n The graphic fit of wheat, barley, and oats appears closer to corn than soybeans. These figures are not reported in the interest of brevity. 12 While the nonnormality appears stronger in the residuals from the Macro model, the Jarque-Berra chi-square tests reject normal- ity in the residuals from both models. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 582 August 2006 Amer. J. Agr. Econ. (a) 0.4:;-\/> o.6 : i. 0'A.A )\ .. 0.2 . 0 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 (b) 3 2.5 2 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 P(Actual) - - - P(Macro) P(Equil) 1.5 I1 Figure 2. (a) Real corn prices: actual versus predicted and (b) real soybeans prices: actual versus predicted we augment the Pearsonian correlations with Rank correlations. Table 6 reports the correlation coefficients for the change in log real prices (Panel A), for the residuals from the Macro model (Panel B) and for the Equilibrium Model (Panel C). The Pearsonian coefficients for the Macro- model residuals range from 0.13 to 0.58 and the Rank correlations range from 0.22 to 0.65. The comparison of the correlation ma- trices in Panels A and B indicate that the macrovariables explain only a minority of the correlation-even among the comparatively unrelated commodities (for instance, wheat and soybeans). On the other hand, the corre- lation coefficients for the residuals from the Equilibrium model presented in Panel C are relatively small-even for the highly related wheat and barley, and wheat and oats pairings. The Pearsonian coefficients range from -0.02 (barley-soybeans) to 0.165 (wheat-oats), sub- stantially smaller than those in Panels A or B. Similar range coefficients are observed for the Rank correlations, from -0.04 (barley- soybeans) to 0.15 (wheat-oats). In summary, the Equilibrium model appears to capture the fundamental relationships well enough to ex- plain the majority of the correlation in ob- viously related commodities. For less related commodities, for instance the wheat-soybeans pairing, there appears to be no residual corre- lation, in other words, no excess comovements. To perform a wider test on excess comove- ments we examine the correlations between the five agricultural commodities studied thus far, and a wider group of commodities repre- senting both agriculture and manufacturing. Table 7 reports the Pearsonian correlations This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Ai, Chatrath, and Song Comovement of Commodity Prices 583 (a) 0.3 -00l" 1 % ... ... -0.3 -0.5 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 (b) 0.6 0.4 SI -0.2 -0.4 -0.6 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 S Actual - - - Macro Equil Figure 3. (a) Change in log real corn prices: actual versus predicted and (b) changes in log real soybeans prices: actual versus predicted between residuals from the Macro and Equi- librium models for the five commodities and the residuals from the Macro model for the twenty other commodities. The Rank correlations are fairly similar, and are not re- ported. The log real price changes (reported in the first column) for wheat, barley, corn, oats, and soybeans are correlated most positively to the following commodities: raw material in- dex (range 0.15-0.24), cotton (0.16-0.28), co- conut (0.07-0.21), sugar-US (0.09-0.19), and sugar-I (0.07-0.22). The correlation coeffi- cients of the residuals from the Macro model (reported in the second column for each com- modity) are not substantially reduced. They range from raw material index (0.15-0.23), cotton (0.14-0.26), coconut (0.04-0.19), sugar- US (0.05-0.13), and sugar-I (0.04-0.16). On the other hand, the correlation coefficients from the Equilibrium model (third column for each of the five commodities) are sub- stantially smaller for these commodities. They range from raw material (-0.03 to 0.06), cotton (0.01-0.08), coconut (-0.00 to 0.06), sugar-US (-0.00 to 0.05), and sugar-I (-0.01 to 0.06). Importantly, the relationship between wheat and cotton, that is found to persist in Pindyck and Rotemberg (1990), disappears when em- ploying the Equilibrium model. Once more, the Equilibrium model appears to explain the comovements that the Macro model does not A Test of Confirmation and a Note on the Role of Supply The correlations of the residuals from the Macro and Equilibrium models that have been presented thus far suggest that the supply and This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 584 August 2006 Amer. J. Agr. Econ. Table 6. Correlation of Residuals from Macro and Equilibrium Models Wheat Barley Corn Oats Soybeans A. Changes in log real prices Wheat 1 0.561 0.384 0.569 0.258 Barley 0.638 1 0.471 0.534 0.340 Corn 0.462 0.522 1 0.469 0.652 Oats 0.543 0.581 0.532 1 0.309 Soybeans 0.198 0.370 0.583 0.377 1 B. Residuals from macro model Wheat 1 0.532 0.321 0.541 0.219 Barley 0.584 1 0.429 0.507 0.321 Corn 0.408 0.463 1 0.452 0.650 Oats 0.496 0.545 0.506 1 0.336 Soybeans 0.132 0.301 0.558 0.329 1 C. Residuals from equilibrium model Wheat 1 0.133 0.144 0.152 0.044 Barley 0.156 1 0.010 0.066 -0.039 Corn 0.118 0.052 1 0.043 0.100 Oats 0.165 0.097 0.012 1 0.127 Soybeans 0.031 -0.019 0.079 0.096 1 Note: Statistics below and above the diagonal are, respectively, Pearsonian coefficients and Spearman Rank coefficients. The sample covers Q1/1957-Q4/2002. inventory variables (along with the macro indi- cators) explain the majority of common move- ments in commodity prices. Two important questions remain to be reconciled: First, to what extent does the difference in the esti- mation techniques play a role in the results? As described above, the Macro model is es- timated with a linear specification while the Equilibrium model is implemented using 2SLS and a sixth order polynomial. Second, to what degree are the commodity comovements a re- sult of supply factors? Economists have gained an increasing appreciation for the importance of supply shocks as sources of fluctuations in aggregate economic performance in general, and the distribution of price changes in parti- cular (for instance, see Myers and Runge 1985; Balke and Wynne 1996). The relative role of supply factors in commodity price comove- ments remains to be addressed. We address the first question on the role of estimation technique via a more careful com- parison of the empirical results from the Macro and Equilibrium frameworks. We do this by respecifying the empirical framework of the Macro model to more closely match that of the Equilibrium model. We estimate the model (5) pt = o + oX1t + . . . + anXt + Et using OLS, with a sixth-order polynomial on the contemporaneous and lagged macroeco- nomic indicators. We examine the extent to which this specification improves the explana- tory power over the more traditional specifica- tion Macro model, and the extent to which this specification affects residual correlations. We find that the explanatory power of the Macro model is little changed for corn, oats, and soybeans with R-square coefficients of 0.023, 0.025, and 0.005 for the differenced series. For wheat and barley, the polynomial model per- formed better, with R-squared coefficients of 0.172 and 0.161, respectively, which represent improvements of 0.10 and 0.05 over the linear model (from table 4). However, the polyno- mial fit for the Macro model did not provide a noteworthy change in residual correlations for any of the pairings using either Pearsonian or Rank correlations. Thus, it appears that it is mainly the inputs of the two models, rather than their econometric implementation that cause the disparity in the results of residual correlations. We address the second question-on the relative role of supply in commodity comovements by integrating out the inventory, a demand variable. Note that we can always write (6) pt = E{f (st - It, Xt) I st, Xt}+ u = g(s,, X,) + ut where ut = Et + (f(Dt, Xt) - g(st, Xt). Thus, the relative role of supply can be assessed by com- paring the fit of equation (6) with that of the This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Table 7. Relatively Unrelated Commodities: Correlation of Residuals from the Macro and Equilibrium Models Wheat Barley Corn Oats Soybeans AP c(Macro) e(Equil) AP e(Macro) e(Equil) AP ?(Macro) c(Equil) AP ?(Macro) ?(Equil) AP ?(Macro) ?(Equil) Materials Nickel 0.002 -0.012 0.008 -0.057 -0.082 0.042 0.055 0.037 -0.057 -0.060 -0.086 -0.073 0.006 0.008 -0.017 Copper -0.018 -0.062 -0.138 0.050 0.014 -0.056 -0.015 -0.062 -0.104 0.031 0.038 -0.091 0.051 -0.030 -0.080 TSP 0.150 0.015 -0.049 0.136 -0.046 -0.002 0.133 0.028 -0.001 0.088 -0.014 -0.039 0.047 -0.058 -0.049 Rubber 0.158 0.177 -0.070 0.084 0.094 0.046 0.092 0.101 0.091 0.100 0.152 0.031 0.090 0.019 0.072 Raw-material 0.177 0.182 -0.032 0.151 0.154 0.024 0.186 0.209 0.052 0.178 0.227 0.054 0.240 0.204 0.062 Beverages/sugar Sugar (I) 0.160 0.113 0.038 0.217 0.164 -0.007 0.073 0.036 0.032 0.137 0.120 0.057 0.121 0.069 0.022 Sugar (US) 0.124 0.072 0.046 0.192 0.133 -0.004 0.087 0.049 0.040 0.155 0.128 0.020 0.112 0.072 0.053 Coffee (0) -0.018 0.027 -0.038 -0.124 -0.087 -0.035 -0.042 -0.005 -0.026 0.036 0.080 0.003 0.102 0.107 -0.002 Coffee (R) -0.017 0.023 -0.001 -0.061 -0.029 -0.065 -0.098 -0.066 -0.016 -0.040 -0.001 -0.040 0.027 0.016 -0.028 Cocoa -0.001 -0.012 0.015 -0.101 -0.127 0.039 0.031 0.015 -0.036 -0.110 -0.129 -0.017 0.045 0.033 -0.013 Edible oils Palm 0.146 0.122 0.019 0.099 0.062 0.012 0.181 0.146 0.026 0.108 0.081 0.034 0.242 0.212 0.051 Groundnut 0.039 0.012 -0.009 0.028 -0.006 0.024 0.156 0.122 0.029 -0.017 -0.050 -0.035 0.136 0.114 0.032 Coconut 0.160 0.136 -0.002 0.072 0.038 0.041 0.193 0.163 0.020 0.137 0.118 0.037 0.214 0.185 0.065 Meats/hides Hides -0.033 -0.047 -0.007 0.040 0.048 -0.069 0.110 0.116 0.021 -0.010 0.013 -0.018 0.114 0.073 0.037 Lamb 0.091 0.071 0.036 0.155 0.136 0.091 0.060 0.081 0.116 0.093 0.109 0.022 -0.041 -0.061 0.047 Beef 0.095 0.072 -0.095 0.170 0.168 0.043 0.239 0.214 0.052 0.102 0.105 0.066 0.164 0.112 0.063 Miscellaneous Rice 0.116 0.071 -0.018 0.018 -0.049 0.032 0.145 0.105 0.045 -0.048 -0.094 -0.036 0.023 -0.016 0.024 Cotton 0.160 0.140 0.022 0.198 0.182 0.026 0.282 0.263 0.052 0.171 0.169 0.013 0.260 0.216 0.083 Banana -0.001 0.004 0.013 0.107 0.127 0.008 0.083 0.102 0.056 0.209 0.232 0.062 0.056 0.058 -0.011 Fishmeal 0.123 0.121 0.061 0.102 0.095 0.058 0.058 0.038 -0.053 -0.010 -0.024 0.024 0.137 0.120 0.046 Note: AP is the change in log of real prices, E(Macro) is the residual from the OLS estimation with controls for Macroeconomic indicators, and E(Equil) is the residuals from the 2SLS estimation of the Equilibrium model. Coefficients are Pearsonian correlations. The sample covers Q1/1957-04/2002. 0 a 0 This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 586 August 2006 Amer. J. Agr. Econ. Table 8. The Role of Supply in Commodity Comovements Wheat Barley Corn Oats Soybeans A. Adjusted R2 0.618 0.427 0.510 0.535 0.348 B. Correlation of residuals Wheat 1 0.074 0.016 -0.042 0.036 Barley 0.055 1 0.142 0.093 0.016 Corn 0.109 0.151 1 0.091 0.146 Oats -0.022 0.114 0.097 1 0.061 Soybeans 0.048 0.010 0.161 0.085 1 Note: The results are from the OLS estimation of the supply-model [equation (6)J. Correlations below the diagonal are Pearsonian correlations and above the diagonal are Rank correlations for the OLS residuals. The sample covers Q1/1957-Q4/2002. Macro model, and more importantly, the Equilibrium model.13 To avoid possible misspecification error in the functional form, we adopt the same functional form for g(.) as was employed to implement the Equilibrium model: g(st, Xt) = o0 + aOl,lSt + -? - + atl,nS't + O2,1sXt +- " " + -2,nskXt + o3Xt. Equation (6) is estimated employing OLS with s, dea- sonalized as it was in the estimation of the Equilibrium model. Table 8 reports the regression adjusted R-squareds from (6), our "Supply" model. In the interest of brevity, we present results only for first differenced series over the full sam- ple, since results from level series and those for the PR sample and post-1972 sample produced comparable results. While our intention is not to compare the goodness of fit across the 2SLS (Equilibrium) and OLS (Supply) estimations, it is worth noting that the Supply model es- timations produced adjusted R-squared coef- ficients that are relatively high, ranging from 0.35 for soybeans, to 0.62 for wheat. More direct evidence of the role of supply in the commodity price comovements is provided by the residual correlations. These compare favor- ably to the Equilibrium model for which the re- sults were earlier reported (table 6, Panel C). For instance, the correlation coefficients for pairings involving wheat are consistently lower in the Supply model, using either Pearsonian or Rank correlations. Importantly, only a few cor- relation coefficients in table 8 are statistically significant, suggesting that supply-side funda- mental factors (along with the macrovariables) may be sufficient in explaining the majority of the comovements in commodity prices. Thus, while the incomplete controls for fundamental factors (for instance, we only consider the U.S. market) do not allow us to comprehensively distinguish between the impact of demand and supply in the observed comovements, the evi- dence suggests that the supply factors play the larger role. Conclusion This study addresses the important question of whether the observed correlation in the prices of commodities is "excessive," as described by Pindyck and Rotemberg (1990). Our findings suggest that the comovements are not exces- sive. We show that much of the comovements come from common tendencies in demand and supply factors. We fit a partial equilibrium model that controls for commodity-factor cor- relations ignored in Pindyck and Rotemberg (1990). This empirical model explains the ma- jority of the comovements among commodi- ties with high price correlation, and all of the comovements among those that are marginally correlated. How does the evidence in this article help our understanding of commodity price behav- ior? Foremost, our findings provide, in our es- timation, the most convincing evidence against the ECH for commodities. Further, the success of the empirical model that employs commod- ity supply data suggests that commodity funda- mentals are related more closely than assumed in Pindyk and Rotemberg. In particular, the supply side factors appear to play a large role in the observed price comovements. Our re- sults also show that the fundamental factors explain a large portion of the variability in in- dividual commodity prices and price changes, raising doubts on the role of speculation per se in causing the large price movements com- monly observed in commodity markets. Over- all, the results reaffirm the notion that price s3 A caveat in this framework is that the X variables may repre- sent commodity demand, so that we are not fully controlling for demand effects. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions Ai, Chatrath, and Song Comovement of Commodity Prices 587 movements are not a sufficient statistic for un- derstanding commodity markets or developing a commodity price model. [Received September 2004; accepted September 2005.] References Ai, C., and C. Chen. 2003. "Efficient Estimation of Conditional Moment Restrictions Models Containing Unknown Functions." Economet- rica 71:1795-843. Bailey, W., and K.C. Chan. 1993. "Macroeconomic Influences and the Variability of the Commod- ity Futures Basis." Journal of Finance 48:555- 74. Balke, N.S., and M.A. Wynne. 1996. "Supply Shocks and the Distribution of Price Changes." Fed- eral Reserve Bank of Dallas. Economic Review QI:10-18. Bessembinder, H., and K. Chan. 1992. "Time- Varying Risk Premia and Forecastable Re- turns in Futures Markets." Journal of Financial Economics 32:169-93. Bjornson, B., and C.A. Carter. 1997. "New Evidence on Agricultural Commodity Returns Perfor- mance under Time-Varying Risk." American Journal of Agricultural Economics 79:918-30. Brunner, A.D. 1998. "El Nino and World Primary Commodity Prices: Warm Water or Hot Air?" Board of Governors of the Federal Reserve System. International Finance Discussion Pa- pers 608, April. Carter, C.A. 1999. "Commodity Futures Markets: A Survey." Australian Journal of Agricultural and Resource Economics 43:209-47. Cashin, P., C.J. McDermott, and A. Scott. 1999. "The Myth of Comoving Commodity Prices." IMF Working Paper 169. Chambers, M.J., and R.E. Bailey. 1996. "The Theory of Commodity Price Fluctuations." Journal of Political Economy 104:924-57. Chari, V.V., and R. Jagannathan. 1990. "The Sim- ple Analytics of Commodity Futures Markets: Do They Stabilize Prices? Do They Raise Wel- fare?" Federal Reserve Bank of Minneapolis, Quarterly Review 14:1-13. Chatrath, A., and E Song. 1999. "Futures Commit- ments and Commodity Price Jumps." Financial Review 34:95-112. Deaton, A., and G. Laroque. 1992. "On the Be- haviour of Commodity Prices." Review ofEco- nomic Studies 59:1-23. -. 2003. "A Model of Commodity Prices Af- ter Sir Arthur Lewis." Journal of Development Economics 71:289-310. Deaton, A., and R. Miller. 1996. "International Commodity Prices, Macroeconomic Perfor- mance and Politics in Sub-Saharan Africa." Journal of African Economies 5(AERC Supplement):99-191. Deb, E, P.K. Trivedi, and P. Varangis. 1996. "The Excess Comovement of Commodity Prices Re- considered." Journal of Applied Econometrics 11:275-91. Dornbusch, R. 1985. "Policy and Performance Links Between LDC Debtors and Industrial Na- tions." Brookings Papers on Economic Activity 2:303-56. Gilbert, C.L. 1989. "The Impact of Exchange Rates and Developing Country Debt on Commodity Prices." Economic Journal 99:773-84. Goodwin, B.K., R. Schnepf, and E. Dohlman. 2005. "Modelling Soybean Prices in a Changing Pol- icy Environment." Applied Economics 37:253- 63. Irwin, S.H., and B.R. Holt. 2004. "The Impact of Large Fund and CTA Trading on Futures Market Volatility." In G.N. Gregoriou, V.N. Karavas, ES. L'Habitant and E Rouah, eds. Commodity Trading Advisors: Risk, Perfor- mance Analysis Selection. New York: John Wi- ley and Sons, Inc. pp. 151-82. Malliaris, A.G., and J.L. Urritia. 1996. "Linkages be- tween Agricultural Commodity Futures Con- tracts." Journal of Futures Markets 16:595-610. Myers, R.J., and C.E Runge. 1985. "The Relative Contribution of Supply and Demand to Insta- bility in the U.S. Corn Market." North Central Journal of Agricultural Economics 7:70-77. Netz, J.S. 1995. "The Effect of Futures Markets and Corners on Storage and Spot Price Variability." American Journal of Agricultural Economics 182-93. Pagan, A. 1999. "A Framework for Understand- ing Bull and Bear Markets." Mimeo. Canberra: Australian National University. Palaskas, T.B., and P.N. Varangis. 1991. "Is There Excess Co-Movement of Primary Commodity Prices?: A Co-Integration Test." Working Pa- per Series No. 758, International Economics Department, Washington DC: World Bank. Park, H.Y., K.C.J. Wei, and T.J. Frecka. 1988. "A Further Investigation of the Risk-Return Relationship for Commodity Futures." Ad- vances in Futures and Options Research, Vol. 3. EJ. Fabozzi, ed., pp. 357-77. Greenwich, CT: JAI Press. Pindyck, R.S., and J.J. Rotemberg. 1990. "The Ex- cess Co-Movement of Commodity Prices." Economic Journal 100:1173-89. Roll, R. 1984. "Orange Juice and the Weather." American Economic Review 74:861-80. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions 588 August 2006 Amer. J. Agr. Econ. Westcott, P.C., and L.A. Hoffman. 1999. "Price De- termination for Corn and Wheat: The Role of Market Factors and Government Programs." U.S. Department of Agriculture, Economic Re- search Service. Williams, J.C., and B.D. Wright. 1991. Storage and Commodity Markets. Cambridge, U.K.: Cam- bridge University Press. Zulauf, C.R., and S.H. Irwin. 1998. "Market Effi- ciency and Marketing to Enhance Income of Crop Producers." Review of Agricultural Eco- nomics 20:308-31. Appendix Prices (in U.S. dollars) from IMF's International Financial Statistics Database 1. Banana: Average of Chiquita/Del Monte/Dole, U.S., Gulf Delivery. 2. Beef: Australian/New Zealand, Frozen, U.S. import price. 3. Cocoa: Average of three nearest active futures contracts, New York Cocoa Exchange. 4. Coffee, Other: El Salvador and Guatemala, ex- dock New York. 5. Coffee, Robusta: Uganda/Cote d'Ivoire, ex- dock New York. 6. Copper: Grade A London Mercantile Ex- change, c.i.f., European Ports. 7. Cotton: Liverpool A, c.i.f., Liverpool. 8. Coconut oil: Philippines/Indonesian, c.i.f. Rotterdam. 9. Fishmeal: 64/65%, any origin, c.i.f., Rotterdam. 10. Groundnut oil: Any origin, c.i.f., Rotterdam. Prior to 1974, Nigerian bulk, c.i.f., U.K. 11. Hides: Chicago, f.o.b., shipping point. 12. Lamb, New Zealand, Frozen, London Price. 13. Nickel: Melting grade, c.i.f., North Europe. 14. Palm oil: Malaysian/Indonesian, c.i.f., NW Eu- ropean Ports. Prior to 1974, UNCTAD. 15. Sugar, International: Sugar Agreement Price. 16. Sugar, US: Contract #14, c.i.f., U.S. import price. 17. Raw Materials (Agricultural): Raw material index. 18. Rubber: Malaysian, f.o.b., Malaysia/Singapore. 19. Rice: 5% broken, normal price, f.o.b., Bangkok. 20. TSP (Triple Super-Phosphate): bulk, Gulf of Mexico. This content downloaded from 202.185.96.100 on Tue, 3 Jun 2014 21:56:28 PM All use subject to JSTOR Terms and Conditions