The Impact of the Oil Sector on Commodity Prices: Correlation or Causation?
Journal of Agricultural and Applied Economics, 2010):477-485 ?– 2010 Southern Agricultural Economics Association The Impact of the Oil Sector on Commodity Prices: Correlation or Causation? Sayed H. Saghaian The interconnections of agriculture and energy markets have increased through the rise in the new biofuel agribusinesses and the oil-ethanol-corn linkages. The question is whether these linkages have a causal structure by which oil prices affect commodity prices and through these links, instability is transferred from energy markets to already volatile agricultural markets.
In this article, we present empirical esults using contemporary time-series analysis and Granger causality supplemented by a directed graph theory modeling approach to identify the links and plausible contemporaneous causal structures among energy and commodity variables. The results show that although there is a strong correlation among oil and commodity prices, the evidence for a causal link from oil to commodity prices is mixed.
Key Words: ethanol prices, crude oil prices, corn prices, soybean prices, wheat prices, causal structure JEL Classifications: QI 1, QI 3, Q42, Q48 We recently observed several occurrences of ajor importance to the agricultural sector simultaneously: the extreme price hikes in the energy sector, the extreme commodity price variability with wider variation and higher averages compared with the past, and the continuing global financial and economic crisis. Last year’s farm income was the highest recorded in the history of the U. S.
Within this context, the purpose of this article is to examine the extent of energy and agricultural sectors’ interlinkages and their interconnections, and the causal structure of the impact of crude oil and ethanol prices on commodity prices. Sayed H. Saghaian is an associate professor at the University of Kentucky, Department of Agricultural Economics, Lexington, KY. This research was supported by the University of Kentucky Agricultural Experiment Station and is published by permission of the Director as station number 09-04-112. transportation fuel to be blended with gasoline to increase its octane level. Later the role of ethanol was shifted to become an “oxygenate” to help gasoline burn more efficiently through several government mandates. The maximum amount of ethanol that could currently be blended stands at the 10% level. With the current U. S. consumption of gasoline being approximately 140 billion gallons annually, the maximum amount of ethanol blended as Elo is approximately 14 billion gallons (Taheripour and Tyner, 2008). Ethanol production has increased tremendously in recent years.
There were only approximately 50 ethanol plants in the U. S. in the late 1990s, producing approximately one billion gallons annually. The Renewable Fuels Standard Act, which was passed in 2005, targeted 7. 5 billion gallons of ethanol production by the year 2012. Additionally, Congress passed another energy bill in 2007, doubling the Renewable Fuels Standard by the year 2015 to billion gallons. 478 Journal of Agricultural and Applied Economics, August 2010 A large portion of the growth in corn demand is associated with growth in ethanol production, because most ethanol in the U.
S. is made from corn. Higher gasoline prices make ethanol production more viable, increasing the supply of ethanol. More ethanol plants and production translates into more demand for corn, which in turn increases corn prices, ceteris paribus (assuming all other things being equal). l Higher corn prices make corn more profitable to grow, causing ome farmers to shift from other crops to corn production. This will also push food, seed, and industrial users to shift from corn to other commodities, increasing their prices.
The objective of this research is to identify the nature of these links and address how variables such as crude oil and ethanol prices impact prices of different commodities such as corn, soybeans, and wheat. Literature Review Agricultural economists have long studied factors that move commodity prices over time. In the past they were particularly interested in the effects of exchange rate policy on agricultural rices. Schuh (1974) argued that changes in U. S. macro policy could affect the value of dollar, which in turn impacts the competitiveness of U. S. gricultural commodities in the world in this topic was later heightened as a result of the “overshooting hypothesis” (Frankel, 1986; Saghaian, Reed, and Marchant, 2002a). Agricultural price overshooting also affects farm prices and income and could partially explain the observed price variability. For many years, much attention was given to the role of exchange rates and monetary change and its transmission of macro changes to agricultural prices. However, nowadays, the links among oil, ethanol, and commodity prices and the nature of relationships between energy and agricultural sectors have become an important issue. As pointed out by an anonymous reviewer, for 2009-2010 marketing year, increased demand for corn for ethanol was more than offset by a slight increase in acreage, which led to lower prices. Prices also declined from July 2008 to August 2009 along with declining acreage and increasing ethanol demand. One concern is that the integration of agricultural and energy markets could add to the already volatile agricultural prices. According to a report by the Food and Agriculture Organization of the United Nations, food prices increased by almost 40% in 2007 and continued increasing sharply in 2008 (Rosegrant, 2008).
Taheripour and Tyner (2008) showed that a large share of the corn price hikes is the result of the increase in the oil prices. Rosegrant (2008) shows that 30% of the increase in grain prices is estimated to be the result of the increased biofuel demand with corn prices having the sharpest increase with 39% in real prices. In the most recent issue of Choices Magazine, Irwin and Good (2009), who examined hanges in the agricultural commodity prices, showed recent commodity price changes have higher averages and wider variations than previous price changes.
In the same issue, von Braun and Torero (2009), who investigated the commodity price spike of 2007-2008, looked at the role of trade policy changes such as the rise in export barriers and the fall of import barriers as well as the role of speculative activity in the observed price spike in the commodity markets. Baffes (2007) showed among nonenergy commodities, oil price changes have the highest pass-through to food commodities and fertilizers. on Braun et al. (2008) found high energy prices have increased the costs of transportation pesticides, making agricultural production more expensive.
Morehart (2009) investigated the impact of macroeconomic policy on land values. He found that land values are also highly sensitive to macroeconomic conditions. Muhammad and Kebede (2009) argued the emerging ethanol market has integrated oil and corn prices in such a way that the agricultural sector is now importing instability from the oil sector. Conley and George (2008) argue that continuous growth of biofuel industries and the ncreased demand for corn have important implications for the managers of grain farms and agribusinesses.
They conclude that factors such as government macro policies regarding ethanol would cause structural changes not only in the U. S. production and marketing of corn, but also other crops such as soybeans, wheat, and Saghaian: The Impact of the Oil Sector on Commodity Prices possibly even cotton as a result of the rotational nature of crop production. Siebert, Hagerman, and Park (2008) argue that some farmers, who have been interested in ethanol production, have made unnecessary very large downstream investments in the past.
They discuss investment techniques that could improve and enhance methods of investing in the production of ethanol. Obviously these issues are of paramount importance for the future feasibility of ethanol production and profitability of farm operations in the U. S. Econometric Model Development and Empirical Results Most agricultural economists are comfortable with a supply-demand framework in commodity price analysis and such analyses are quite common in the literature. This is natural because there are strong conceptual foundations linking economic variables to producer and consumer decisions.
These foundations have been used for decades and are well understood and accepted in the profession. However, economic theory does not provide sufficient information about the causal structures among energy and commodity prices. The challenge in the analysis of macroeconomic linkages to agriculture is to eliminate the simultaneous (supply-demand) linkages among commodities so that the relationship among individual commodity prices and macroeconomic variables can be isolated.
If daily or weekly grain prices are dominated by revised storage estimates, crop estimates, and weather fears, it is ifficult to isolate the effects of other variables changes. Examining the causal structure of energy and commodity prices can show how they react to crude oil shocks and increased ethanol prices while also taking into consideration the simultaneity among the prices. There have been numerous theoretical and empirical estimates of the effects of macroeconomic variables on commodity prices.
These analyses have progressively improved as a result of theoretical refinements and more powerful time-series techniques (vector autoregressive and vector error correction models) that provide 479 better adjustments for nonstationarity and long un relationships among variables (Crane and Nourzad, 1998; Schmidt, 2000). Recent advances in time-series econometric techniques allow us to use a reduced form of commodity price equations that collapses the structural simultaneity of commodity models and isolates underlying macroeconomic relationships.
The tools are powerful enough that linkages among commodities can be viewed by predicting forward movements in endogenous variables (commodity prices) using time-series techniques. The empirical model underlying this study is built on the existing literature (Saghaian, Ozertan, and Spaulding, 2008). We include monthly prices of five variables: corn prices per bushel, soybean prices per bushel, wheat prices per bushel along with crude oil and ethanol prices per gallon.
Second, we build on Robertson and Orden’s (1990) cointegration approach by using Johansen and Juselius’ (1992) method of estimation. Empirically, the first difference in each variable is represented as a function of its own lagged value, the lagged values of the other variables, and the cointegration equation. Given the nature of the underlying data series, we conduct stationarity tests of the series using the augmented Dickey-Fuller test. Then, we perform a cointegration test to determine whether there exists a long-run relationship among the series in the system.
Third, we specify a vector error correction model and conduct hypothesis testing within this framework. Finally, this is followed by Directed Graph analysis and structures among the variables. Stationarity Testing Monthly time-series data are collected from for the variables. Commodity price data come from the Agriculture Statistics Board. 2 Oil and ethanol data come from the Economic Research Service, USDA, 2008. 2 The assistance of Andrew Mohammad in providing the data used in this study is ratefully acknowledged.
Descriptive statistics of the variables can be found in Table 1. Table 2. The Correlation Matrix of the Variables Variables Correlation Matrix Ethanol Corn Soybeans Wheat 0. 89 0. 45 0. 49 0. 67 The correlation matrix of the five variables as shown in Table 2 indicates a high correlation of 89% between oil and ethanol price series. This is expected because oil and ethanol are nearly perfect substitutes. Also, there is a high correlation among the commodity prices: corn and soybeans 88%, corn and wheat almost 90%, and soybeans and wheat 83%. sed to determine the order of integration of ach univariate series. This test involves running a regression of the first difference of the series against the series lagged one period, lag difference terms, and a constant. The results of the unit-root test are estimated by ordinary least squares and presented in Table 3. The second column of Table 3 summarizes the ADF test results for each original variable, whereas the third column presents the results for the first difference of each series.
Following Enders (1995) and Hendrys (1986) “General to Specific” procedure, we started with an overspecifled ADF regression in which n was relatively large and then used battery of lag length diagnostic tests to refine the specification for each univariate series. We use the Akaike Information Criterion (AIC) and Schwarz criterion to determine the appropriate lag specification (n). In general, the F-statistic of the ADF regression with n 5 2 was statistically significant (p < 0. 01) in each case. In general, partial t-statistics were not significantly different 2. 6 Soybeana 6. 43 Wheata 3. 81 Oilb 40. 44 Ethanolb 1. 58 Dollars/bushel. b Dollars/gallon. 0. 87 1. 97 1. 52 26. 37 1. 53 4. 09 2. 13 11. 28 0. 90 5. 48 13. 30 10. 00 133. 93 3. 58 0. 46 0. 64 0. 88 0. 9 0. 3 from zero beyond two lags. In each case, we failed to reject the null hypothesis of zero firstorder autocorrelation at the 5% level of significance using the Durbin-Watson bound test. As shown in Table 3, the ADF test statistics in absolute value for all series rose after first differencing (right-most column of Table 3).
Thus, we are able to reject the null hypothesis and conclude that each series is stationary after first differencing. Based on this analysis, we use all data as an integrated process of order 1 or I (1). Johansen’s Cointegration Tests Based on the ADF test, a vector error correction VEC) model is more appropriate than a vector autoregression model to characterize the multivariate relationships among the eight series (Engle and Granger, 1987; Enders, 1995).
Cointegration tests were performed using cointegrating rank, r, or the number of cointegrating vectors in the system using the likelihood ratio (LR) test (Holden and Perman, Table 3. Augmented Dickey-Fuller (ADF)a Test Results Table 1. Descriptive Statistics of Variables in the Empirical Model Standard Variables Mean Deviation Minimum Maximum Ethanol Corn Soybeans Wheat for Variables in Levels Test Results for Variables after First-Differencing Soybean 22. 53 21 . 82 21 . 83 22. 01 26. 02b 28. 67b 28. 65b 26. 9b 29. 80b In absolute value and compared with MacKinnon, Haug, and Michelis (1999) critical values. 1994; Vickner and Davies, 2000). Theoretically, the rank, r, can be at most one less than the number of endogenous variables in the model. The LR test in our analysis determines if cointegrating vectors exist among the eight endogenous macroeconomic series. Table 4 presents the results of cointegration tests for each commodity. Each cointegrating equation contains an intercept and a money supply slope coefficient.
At the 1% level of significance for the trace test Oohansen and Juselius, 1992) and Max-Eigen statistics, we reject the null hypothesis that r 5 0; thus, LR tests reveal there exists a stationary, linear combination among corn, soybean, wheat, oil, and ethanol series. Vector Error Correction Model The Johansen’s cointegration test indicates that the series are cointegrated. Therefore, as discussed, the VEC model is appropriate for this study. In this model, the first difference of each variable is represented as a function of its own lagged values, the lagged values of the other ariables, and the cointegrating equations.
In a VEC system, it is difficult to characterize the qualitative relationships among variables and the expected sign of the unknown parameters to be estimated. However, we expect the long-run equilibrium relationships to be positive in the case of the relationship among the commodity prices, oil prices, and ethanol prices because a rise in the crude oil and ethanol prices is expected to increase the level of commodity prices. The speed of adjustment parameters represents overshooting parameters, indicating how Table 4. Johansen Cointegration Test Resultsa Null Hypothesisb 50