Contemporaneous interactions among fuel, biofuel and agricultural commodities

Energy Economics 58 (2016) 1–10

Contents lists available at ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneeco

Contemporaneous interactions among fuel, biofuel and agricultural commodities Adrian Fernandez-Perez, Bart Frijns ⁎, Alireza Tourani-Rad 1 Department of Finance, Auckland University of Technology, New Zealand

a r t i c l e

i n f o

Article history: Received 3 September 2015 Received in revised form 2 May 2016 Accepted 11 May 2016 Available online 14 June 2016 JEL classification: C32 Q16 Q42 Keywords: Contemporaneous interactions Biofuel Structural VAR

a b s t r a c t This study examines the contemporaneous interactions among energy (oil and ethanol) and agricultural commodities (corn, soybean, and wheat) in the United States during the period 1 June 2006 to 22 January 2016. Since traditional VAR analysis is not able to capture the contemporaneous interactions among these commodities, we employ a structural VAR analysis in combination with the identification through heteroskedasticity approach. The empirical results indicate that i) the contemporaneous interactions are important, asymmetric, and have implications for impulse response functions; ii) crude oil has a unidirectional contemporaneous impact on the agricultural commodities, and the agricultural commodities (corn and soybean) – mostly used in the biofuel production – have a unidirectional contemporaneous impact on ethanol; and finally, iii) these contemporaneous relations depend on the price level of crude oil in that there are stronger effects from crude oil (agricultural commodities) to agricultural commodities (ethanol) in high crude oil price states. © 2016 Elsevier B.V. All rights reserved.

1. Introduction In recent years, strong linkages among energy and agricultural markets have attracted widespread attention from practitioners, academics and policy makers. Various theories related to the food crises, legislative policies on biofuels, global demand for and supply of agricultural products and the financialization of commodities have been put forward and examined empirically to explain these linkages. Most of the empirical studies rely on lead–lag dynamics to explain these linkages. However, given that many of these commodities are also actively traded in their respective futures markets, we expect the interactions among them to occur almost immediately. Commodity futures markets incorporate information quickly (see e.g. Lehecka et al., 2014; Chatrath et al., 2012) and information about one commodity affects other commodities swiftly, especially nowadays where markets are highly integrated (see, e.g., Hong and Yogo, 2012; Hou and Szymanowska, 2015; Boons et al., 2014, among others). As such, correctly identifying the interrelations among energy and agricultural markets is vital for properly understanding the dynamics of these commodities (and determining the relation ⁎ Corresponding author at: Department of Finance, Auckland University of Technology, Private Bag 92006, 1142 Auckland, New Zealand. Tel.: +64 9 921 9999x5706; fax: +64 9 921 9940. E-mail address: [email protected] (B. Frijns). 1 Alireza Tourani-Rad acknowledges the financial support by the Czech Science Foundation as part of the project no. 14-02108S “The nexus between sovereign and bank crises”.

http://dx.doi.org/10.1016/j.eneco.2016.05.014 0140-9883/© 2016 Elsevier B.V. All rights reserved.

between oil and food prices), and developing trading and risk management strategies. As previous literature has predominantly relied on reduced-form models, such as vector autoregressive (VAR) or vector error correction (VEC) models,2 these models are not able to capture the contemporaneous relations among energy and agricultural commodities. In these models, the contemporaneous relations, which can be interpreted as causal relations, are generally left in the residuals of the model, and are therefore unidentified. Saghaian (2010), for instance, investigates the interrelationship between crude oil, ethanol, corn, soybean and wheat and concludes that while there is strong evidence of contemporaneous correlations among these commodities, the evidence on causality is mixed. In this study, we aim to resolve the problem of contemporaneous correlations, by implementing a novel technique known as identification through heteroskedasticity which was originally developed by Rigobon (2003). This technique uses a structural VAR (SVAR) approach to break up the contemporaneous relationships into causal relations. These contemporaneous relations differ from Granger causality, which captures the causal effect of one lagged variable on the current value of another variable. Through implementation of this model, we attempt to assess the price transmission between energy commodities (crude oil and ethanol), and agricultural commodities (soybean, corn and wheat). Furthermore, given that data aggregation over time would only increase 2 For an overview of the price transmission literature that focuses predominantly on VAR and VEC models, see Serra and Zilberman (2013).

2

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

the issue of contemporaneous correlations, we estimate our model using daily data on the futures contracts of these commodities for the period from 1 June 2006 to 22 January 2016. We employ futures contracts traded in the US markets. These markets are all in the same time zone and have extensive overlapping trading hours. This makes the modeling of contemporaneous interactions among these commodity futures even more important. Investigating the contemporaneous interactions among fuel, biofuel and agricultural commodity futures is important for, at least two reasons. First, information is transmitted rapidly among markets these days and may no longer be captured by investigating the impact of lagged variables on current variables as in standard VAR models. Second, the contemporaneous interactions are not only important for describing short-term commodity price behavior, but also long-run behavior, as captured by impulse-response functions. We thus expect the contemporaneous relations among the commodity futures contracts in our sample to be strong, as nowadays energy and agricultural markets are interconnected, and information is transmitted swiftly across markets. We document several important findings. We show that traditional VAR analysis based on lead–lag relations does not capture much of the co-movement among the commodities in our sample. Through the application of SVAR analysis, we are, however, able to identify the directional contemporaneous effects, which in many cases are significant and asymmetric, i.e. the contemporaneous effect of one commodity on the other can be of a different magnitude than the other way around. For instance, we document that crude oil has a direct impact on the agricultural commodities, while there is no direct impact in the opposite direction. We also note that corn and soybean have a unidirectional impact on ethanol and there are bi-directional effects between the pairs soybean–corn and corn–wheat. These dynamics are not observed in the traditional reduced-form VAR. Through impulse-response analysis, we demonstrate that appropriately accounting for the short-run relations has important consequences for the long-run as well. We observe that shocks applied to a traditional reduced-form VAR lead to very different outcomes than shocks applied to the SVAR. Finally, we show that the contemporaneous relations are dependent on the price level of crude oil, with stronger contemporaneous effects from crude oil on agricultural commodities, and likewise, stronger contemporaneous effects from agricultural commodities on ethanol in high price states. The remainder of this paper is organized as follows. Section 2 provides an overview of the relevant literature. Section 3 outlines the identification through heteroskedasticity approach. Section 4 describes the data and in Section 5, we present and discuss the empirical results. Section 6 concludes the paper.

2. Literature review The literature on the relations between fuel–biofuel/feed crop commodities has expanded rapidly in the last decade.3 Part of this literature addresses the question of whether the usage of biofuel has caused a stronger linkage between fuel prices and food prices. The idea of production of large amounts of crop-based biofuels to reduce dependence on fossil fuels has proven to be controversial, particularly because of the sharp upsurge in food prices, known as the “food crisis” (see, e.g., Du et al., 2011; Harri et al., 2009; Ji and Fan, 2012) which is attributed to price spill-over from crude oil to other markets, especially the agricultural ones. As the price of crude oil increases, demand for biofuel increases, and since biofuels are mainly extracted from agricultural commodities that are normally used in food production, a higher crude oil price encourages farmers to substitute food production by 3 Three overview articles on various aspects of biofuel-related price transmission literature have appeared recently. Janda et al. (2012) consider the technological, social, environmental and policy aspects. Serra and Zilberman (2013) focus on biofuel related time-series literature and Zilberman et al. (2013) provide a general overview of biofuel (and fuel) and commodity food prices.

energy-related commodity production (known as the substitution effect) which results in higher food prices (see, e.g., Tyner, 2010; Chen et al., 2010; Vacha et al., 2013). However, these arguments have been contested. Wetzstein and Wetzstein (2011), for instance, go as far as calling the hypothesis of a strong connection between crude oil and agricultural commodity prices a myth. They argue that investment in biofuel is subject to adjustment costs, irreversibility, and uncertainty. Hence, biofuel production and consequently the demand for agricultural commodities may be less responsive to energy prices than has been assumed. Meyers et al. (2014) find that, while over short and intermediate time horizons the comovement between energy and agricultural prices are strong, in the long-run agricultural prices tend to be determined predominantly by agricultural supply conditions and the non-biofuel demand for agricultural feed-stocks. Furthermore, energy prices play an inconsequential role in setting long-run agricultural prices. In addition, the huge increase in demand for raw materials and agricultural commodities by the growing Asian economies, in particular China and India, has led to persistently high crude oil and agricultural commodity prices in the past decade (Hamilton, 2009; Kilian, 2009; Wolf, 2008). Therefore, the observed co-movement in crude oil and agricultural commodity prices could be due to higher global demand for agricultural commodities driven by economic activities, rather than price spill-over from crude oil to these markets. Meyers et al. (2014) attempt to distinguish oil-specific shocks from aggregate demand shocks. They observe that oil shocks can explain a small fraction of agricultural commodity price variations before the food crisis in 2006–2008, whereas in the post-crisis period their explanatory abilities become much more significant. After the first food crisis of 2006–2008, the contributions of oil-specific factors to variations in agricultural commodity prices are far greater than those of aggregate demand shocks. Meyers et al.'s (2014) findings are generally in line with those in related studies that clearly show much stronger oil–agriculture linkages after 2006 (see, e.g., Kristoufek et al., 2012; Nazlioglu, 2011; Nazlioglu et al., 2013). Besides global market conditions, biofuel prices are also affected by government policies and regulations. In the U.S., for example, the main policy instruments are subsidies such as the Renewable Fuel Standard and the Clean Air Act as of May 2006, and the import tariff introduced on ethanol (see, e.g., Taheripour and Tyner, 2008). Generally, the proposed standards require motor fuels to contain a minimum amount of fuel generated from renewable sources, such as ethanol, solar or wind energy. In reality, so far only ethanol has become a viable substitute to comply with the new standard, further accelerating the linkage between energy and commodity markets. According to Avalos (2014), the higher the crude oil prices, the greater the incentives for gasoline producers to bring to market blends with higher levels of ethanol. From an empirical point of view, the linkages between energy and commodity markets have been investigated by various studies. Ciaian and Kancs (2011a), for instance, show that the prices of nine agricultural commodities are cointegrated with crude oil prices over the period of 2005–2010. Ciaian and Kancs (2011b) and Kristoufek et al. (2012) examine the relations among the prices of energy and agricultural commodities before and after the first food crisis of 2006–2008 and report that the connections have become much stronger in the post-crisis period. More recently, Kristoufek et al. (2016), utilizing a continuous wavelet framework, find that the prices of ethanol feedstock both in Brazil and the US lead the prices of ethanol and not the other way around. Finally, a number of studies argue that agricultural commodity prices are not affected by the price of crude oil and hence support the neutrality of agricultural commodity markets. Zhang et al. (2010), using the Johansen trace test, find that there are no direct long-term price relations between crude oil and agricultural commodity prices, and that there are only limited direct short-term relationships. Reboredo (2012), employing copula models, investigates the conditional dependence between world oil prices and agricultural commodity prices, and finds a weak dependency between food and oil. Finally,

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

Qiu et al. (2012), using a structural VAR together with a direct acyclic graph, show that crude oil, gasoline and ethanol price shocks do not spill-over into agricultural commodity markets and that market forces of demand and supply are the main drivers of food prices. Economic theories, due to the complex interrelationship of supply and demand factors of energy and agricultural markets, technological advances in both markets, government policies, substitution among commodities, competition, etc., do not provide a clear evidence regarding the causal structures among energy and commodity prices. This issue has, for example, been recognized by Saghaian (2010) who argues that any inference on the causal structures between energy and commodity prices requires a careful investigation of contemporaneous correlations among corresponding innovations. Saghaian, using traditional time-series analysis and Granger causality augmented with a directed graph approach, attempts to identify the linkages and plausible contemporaneous causal structures among energy and commodity markets. He states that correlations among energy and agricultural markets are substantial, but when it comes to causation, the results are mixed. He observes that crude oil prices Granger cause corn, soybeans, and wheat prices. The directed graph results, however, do not show structural causation between the energy and agricultural commodities. The abovementioned empirical studies generally show that there are spill-over effects among the energy and agricultural markets. However, none of them capture the contemporaneous relations between these markets other than acknowledging that there are correlations among them. By not taking these contemporaneous relations into account, these studies miss an important piece of the puzzle in explaining the causal relations among commodity prices. In this paper, we aim to uncover the contemporaneous relations through the application of identification through heteroskedasticity framework and demonstrate the importance of these contemporaneous effects.

3. Identification through heteroskedasticity The main objective of this paper is to address the contemporaneous effects between energy and agricultural commodity futures contracts. This issue is essentially the same problem that one faces in simultaneous equation models, in that the contemporaneous causal effects (or structural parameters) cannot be identified due to endogeneity. One solution to this problem is to use orthogonal structures of residuals (such as Cholesky factorization) or imposing identifying restrictions. However, these restrictions are usually based on ad hoc assumptions about the direction of causality. As an alternative solution to this problem, Rigobon (2003) proposes a methodology that uses heterogeneity in the data to identify the contemporaneous relations in a simultaneous equation model. Provided that there are non-proportional shifts in volatility, these shifts can be used to identify the structural parameters of the model.4 In this section, we detail the model to assess the contemporaneous relations among three agricultural commodities futures: soybean (henceforth soy), corn and wheat, and the two energy commodities futures: ethanol and crude oil (henceforth oil). We follow the identification through heteroskedasticity approach of Rigobon (2003), where

4

The proposed identification through heteroskedasticity approach has recently been applied to investigate contemporaneous spill-over among various markets and assets. Rigobon (2003) originally applied his technique to measure the contemporaneous relationships among Argentinean, Brazilian and Mexican sovereign bonds using a regimeswitching model for the volatility process. In subsequent studies, the technique is employed to study the reaction of monetary policy to stock market movements (Rigobon and Sack, 2003a) and the impact of monetary policy on stock prices (Rigobon and Sack, 2004). Ehrmann et al. (2011) apply the identification through heteroskedasticity approach to examine international financial transmissions among different asset classes (cash, bonds, foreign exchange and stocks) and across different markets (the US and Euro zone). Rigobon and Sack (2003b) assess the spill-over effects between US short- and longterm interest rates and stock prices, while Andersen et al. (2007) study real-time price discovery in money, bond and stock markets in the US, UK and Germany using this technique.

3

our main goal is to identify the structural parameters in matrix A of the following structural VAR AΔyt ¼ c þ ϑðLÞΔyt−1 þ εt ;

ð1Þ

where Δyt is a (5 × 1) vector containing the log price changes on the five commodities in our study, i.e., 1 ΔSoyt C B ΔCornt C B C Δyt ¼ B B ΔWheat t C, @ ΔEthanolt A ΔOilt 0

where Soyt is the change in the log price of soy, etc., c is a vector of constants and ϑ(L) is a matrix polynomial in the lag operator. The (5 × 5) matrix A captures the contemporaneous effects of one variable on another where the main diagonal is normalized to 1, i.e., 0

1 B α 21 B A¼@ ⋮ α 51

α 12 1 ⋱ ⋯

⋯ ⋱ ⋱ α 54

1 α 15 ⋮ C C: α 45 A 1

ð2Þ

The off-diagonal elements in this matrix capture the contemporaneous effects of one commodity on another. For instance, α12 measures the contemporaneous impact of the returns on corn on the returns on soy, whereas α21 measures the contemporaneous impact of the returns on soy on the returns on corn. All other elements of A are defined analogously. It is important to note that the condition of α12 = α21 is not necessary, i.e. contemporaneously, the directional effects on one variable on another can differ. The traditional literature that relies on a reduced form VAR is not able to identify the structural parameters in A. However, studies that have aimed to estimate the structural parameters either impose coefficient restrictions, sign restrictions or make certain assumptions on the long-run impact of shocks. These assumptions are often quite restrictive or ad hoc in that they either assume a direction of causality, or whether a variable has a long-run impact on another variable or not. In addition, Granger causality does not explain the contemporaneous relationships which could be substantial due to the rapid information transmission between markets. In this paper, we impose a much less restrictive set of assumptions to achieve identification. Following Rigobon (2003), we assume that εt, the residuals in Eq. (1), represent structural shocks to the model and that these structural shocks are uncorrelated with each other. The second assumption is that the series display heteroskedasticity (a feature commonly observed among these assets), while the parameters in A are constant across these different heteroskedasticity regimes, i.e. all heteroskedasticity comes from the structural residuals. These two conditions allow us to identify the structural parameters in A. To achieve identification, we rewrite Eq. (1) as a reduced form VAR, i.e. Δyt ¼ A−1 c þ A−1 ϑ ðLÞΔyt−1 þ ηt ;

ð3Þ

where the reduced form residuals ηt = A−1εt are now correlated, and this correlation is determined by the structural relations between the commodities. Hence, in the reduced form VAR, the contemporaneous relations are absorbed into the residuals, and will show up as a correlation between the elements of ηt. Identification of the structural parameters in A can be achieved by focusing on the reduced form residuals ηt. In the case where there would be no heteroskedasticity in εt, there would also be no heteroskedasticity in ηt. As such, we can define the covariance matrix of the reduced form residuals as Var(ηt) = A−1E[εtεt′]A− 1′ = Ω, which contains 15 unique elements, 5 variances and 10 covariances. Likewise, we can define the covariance matrix of the structural residuals

4

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

as Var(εt) = Σ, which is diagonal by assumption. Given that A contains 20 parameters and Σ contains 5 parameters that need to be identified, full identification of Eq. (1) cannot be achieved with the reduced form VAR (we observe only 15 moments, the elements of Ω, but have 25 parameters to identify). In the case where there is heteroskedasticity in the residuals, we can identify additional volatility regimes, i.e. we could introduce a second, say, high volatility regime. For example, we could have two covariance matrices in the reduced form VAR, Ωlow and Ωhigh. Combined, these two matrices provide 30 moments for estimation. Likewise, we have 30 parameters to be identified, i.e. 20 structural parameters in A and 10 variances in Σlow and Σhigh. Hence in the case of two regimes the model would be exactly identified. One could introduce more regimes to achieve overidentification. As Rigobon (2003) points out, this approach of identification is robust to misspecification of the actual conditional volatility process. All that is required are non-proportional shifts in the volatility of the residuals. In our empirical setting, we implement the identification strategy in a way similar to Ehrmann et al. (2011). This strategy involves several steps. First, we estimate the reduced form VAR in Eq. (3) and collect the residuals ηt. These residuals contain the contemporaneous effects (i.e. ηt = A−1εt). Second, we need to determine different volatility regimes for the identification of the structural parameters in A. To do this, we calculate the variance of the residuals using a 22-day rolling window (roughly a 1-monthly variance) over the entire sample period, and assign the data of a particular observation to a high volatility regime if its variance is higher than its mean plus one standard deviation. Third, of all possible regimes that can be identified, we restrict the analysis to the regimes where one of the assets in our sample sits in a high volatility regime while the rest of the series are in the low volatility regime. This provides us with 5 high volatility regimes. In addition, we use one tranquil regime where all series are in a relatively low volatility regime. The use of these 6 regimes to identify the parameters in A ensures that we will have non-proportional shifts in the volatility of one asset versus the others, which is a requirement for identification.5 Finally, we estimate the parameters by the generalized method of moments (GMM) method of Hansen (1982) by solving the problem: ming′g, where g = A−1ΣiA−1′ − Ωi with i = 1,…, 6 regimes.6 To assess the significance of the coefficients in A, we follow a blockbootstrap procedure comparable to Ehrmann et al. (2011). We implement the bootstrap in the following way. In the estimation of the model, we have identified 6 unique regimes. For each regime, we simulate pseudo-residuals that have the same covariance structure as the actual residuals. Making use of the simulated pseudo-residuals, we employ the estimated coefficients of the reduced form VAR to compute the pseudo-data y⁎t . With this pseudo-data, we re-estimate the VAR and keep the residuals, η⁎t . We use these bootstrapped residuals to identify new regimes and estimate the matrix A⁎ based on these bootstrapped residuals. We repeat this procedure 1000 times and store the estimated coefficients in each procedure. These iterations provide the p-values for the point estimates of A.

on these three agricultural commodities because of their role in the production of biofuel (corn mainly being used in the production of ethanol and soybean mainly being used in the production of biodiesel) and because these commodities are substitute grains for livestock and compete in terms of acreage usage (Chen et al., 2010; Saghaian, 2010; Avalos, 2014; Serra and Zilberman, 2013; Han et al., 2015). For oil, we obtain data on the light crude oil futures contracts traded on the New York Mercantile Exchange (NYMEX), which is now part of the Chicago Mercantile Exchange Group. Soy and ethanol futures are both traded in the electronic platform of Chicago Board of Trade (ECBOT), while corn and wheat futures are traded on the open outcry market of the CBOT. Using settlement prices, we construct a continuous series from these futures contracts by considering the most active contract, which is rolled over when trading activity is higher in the next contract. As we expect news in one market to spillover rather rapidly to other markets, and data aggregation would obscure this process, data are collected at a daily frequency.8 All the data are obtained from Thomson Reuters DataStream. Avalos (2014) shows that the Energy Independence Act, that came into effect in May 2006, led to a structural break in the relation between corn, soy and ethanol; hence, we concentrate on the period from 1 June 2006 to 22 January 2016. In Table 1, we report summary statistics for the five commodities in our sample, where Panel A presents descriptive statistics, and Panel B presents their correlation coefficients. In Panel A, we observe positive average log returns for soy, corn and ethanol, a roughly zero return for wheat, and a negative log return for oil. The annual standard deviation ranges from 26% for soy to 38% for oil. We observe mixed evidence on skewness of the different commodities; soy, corn and ethanol display negative skewness, while wheat and oil display positive skewness but close to zero. However, all commodities display excess kurtosis. A Jarque–Bera test rejects the null hypothesis of normality in the return distributions for all series. Finally, we observe that the returns on the agricultural commodity futures are very close to random walks with insignificant first-order autocorrelation, while both ethanol and oil display some degree of autocorrelation, positive for ethanol but negative for oil. Accordingly, the ADF unit root test concludes all the return series are stationary.9 In Panel B, we report the correlation coefficients among commodities in our sample. All correlation coefficients are positive and highly significant. The returns of the agricultural commodities are highly correlated with soy and corn having a correlation coefficient of 0.61, and corn and wheat have a correlation coefficient of 0.65. The lowest correlation is between soy and wheat at 0.47. The correlations between the agricultural commodities and energy commodities are lower, but stronger between the agricultural commodities and ethanol (0.45 on average) than between the agricultural commodities and oil (0.30 on average). We also note that the correlation between ethanol and oil is not very high, at 0.32. All these correlations are highly significant, thus highlighting the importance of the contemporaneous relations between the agricultural and energy commodity futures.

4. Data

5. Results

In this paper, we study the dynamic and contemporaneous interactions between three agricultural commodity futures: corn, soy and wheat, and two energy commodity futures: ethanol and oil.7 We focus

5.1. Reduced-form VAR and Granger causality

5 We conduct Breusch–Pagan tests for heteroskedasticity in the residuals of the reduced form VAR, and reject the null hypothesis of homoskedasticity at the 1% level for all series. Hence, our series fulfill the criteria of heteroskedasticity required for the identification of A. 6 An alternative approach would be to use a multivariate GARCH specification to model the volatility process (see Rigobon and Sack, 2003a). While this is theoretically possible, in practice this technique does not work in our application with 5 variables as it requires the estimation of 75 free parameters through quasi-maximum likelihood. 7 Several studies make use of futures prices in analyzing the relations among fuel, biofuel and feedstock commodities, see e.g. Zhang and Reed (2008), Chang and Su (2010), Chen et al. (2010), Zhang et al. (2010), Natalenov et al. (2011) and Ziegelback and Kastner (2011).

We start our analysis by documenting the results for the reducedform VAR, which is the traditional method in which dynamic 8 Serra and Zilberman (2013), in their review article, criticize the extant literature since they are usually calibrated using low frequency data which are unable to assess short-run price dynamics. 9 We also conduct a Johansen test for cointegration among the log price series, as Campiche et al. (2007), Saghaian (2010), Ciaian and Kancs (2011a, 2011b), Serra et al. (2011) and Nazlioglu (2011) document that some of the series we investigate may be cointegrated. We find no evidence of cointegration between these price series at conventional confidence levels, and thus continue our analysis with a VAR in first-differences. Results are available on request.

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

5

Table 1 Summary statistics.

Panel A: Descriptive statistics Annual mean Annual std. dev. Skewness Kurtosis ADF ρ(1) Jarque–Bera Observations

Soy

Corn

Wheat

Ethanol

Oil

7.81% 25.86% −0.4287 5.7265 −35.02*** 0.0120 823.70*** 2420

5.78% 31.75% −0.0306 4.8405 −35.28*** 0.0265 341.96*** 2420

0.00% 35.34% 0.0820 4.8622 −35.34*** −0.0108 352.38*** 2420

3.69% 30.84% −0.6494 6.5061 −33.76*** 0.1020** 1409.59*** 2420

−10.11% 38.26% 0.0169 7.4181 −35.18*** −0.0383* 1968.33*** 2420

Panel B: Matrix correlations Soy Corn Wheat Ethanol Oil

1 0.6140*** 0.4661*** 0.4421*** 0.3594***

1 0.6481*** 0.5282*** 0.2921***

1 0.3824*** 0.2471***

1 0.3181***

1

Note: This table presents descriptive statistics for the daily log returns on the futures contracts of soy (soybean), corn, wheat, ethanol and oil (light crude oil) in Panel A, and Pearson correlation coefficients in Panel B over the sample period 1 June 2006 to 22 January 2016. We indicate significance at the 10%, 5%, and 1% level by *, **, and ***, respectively.

interactions among variables are examined. To determine the appropriate lag length, we use several diagnostic tests to assess the autocorrelation structure in the data and in the residuals of the VAR (Portmanteau test and LM-test). In addition, we compute various information criteria (Akaike, Schwarz–Bayesian, and Hannan–Quinn). These tests and information criteria suggest that a lag of 1 is sufficient to remove the autocorrelation in the residuals. Hence, we present all our results for a VAR model using 1 lag.10 In Table 2, we present the results for the reduced-from VAR(1), where we report coefficients, their White corrected t-statistics in parentheses, and R2 adjusted in percentages. Numbers in bold are those for which a Granger causality test is significant at the 5% level. There are a few noteworthy observations. First, we observe that the lagged value of oil has an impact of similar magnitude on current values of the three agricultural commodities. However, we do not observe any lagged effect of agricultural commodities on oil, nor do we find persistence in oil itself. We also find no evidence of ethanol having a dynamic effect on any of the commodities, but there is a lagged effect of ethanol on itself (in line with the summary statistics reported in Table 1). Finally, we do not observe any other impact of lagged values of the agricultural commodities. The results for the Granger causality test are broadly in line with the t-tests, i.e. significant Granger causality from oil to soy, corn and wheat, but not the other way around. The findings of mostly insignificant lead–lag dynamics could be expected, since these futures contracts are traded actively in financial markets and pricing of futures is typically fairly efficient. Thus analysis based on lead–lag dynamics would not be very informative. While there is some evidence of lead–lag dynamics among these commodities and some evidence of Granger causality, the results are not very strong (the adjusted R2s reported in Table 2 are all below 2%). However, as shown in Table 1, the contemporaneous correlations among the various commodities are quite high. These findings clearly suggest that the analysis based on a reduced form VAR is not going to be very effective in fully describing the contemporaneous interactions among the five commodities in this study.

5.2. Contemporaneous interactions In this section, we analyze the contemporaneous interactions among the five commodities. As described in Section 3, the contemporaneous interactions among our five commodity futures are obtained from the structural VAR. These contemporaneous effects are captured by the structural parameters of matrix A. 10 In unreported results, we have used alternative lag lengths up to 9 lags. These results are qualitatively very similar.

In Table 3, we report the parameters of A along with their bootstrapped p-values in parentheses, where the contemporaneous effects run from the variables in the columns to the variables in the rows. Since A sits on the left-hand side of Eq. (1), it is important to note that the coefficients change sign when they are moved to the right-hand side of the equation. Hence, a coefficient with a negative sign implies a positive relation, and vice versa. From Table 3, we can make several important observations. The lack of Granger causality in Table 2 contrasts starkly with the significant contemporaneous effects reported in Table 3. To highlight these contemporaneous interactions between the different commodities, we state the contemporaneous relation by only reporting the coefficients that are significant at the 10% level or better below: ΔSoŷt ¼ 0:0003 þ 0:2492  ΔCornt þ 0:0841  ΔOilt þ … ΔCorn̂t ¼ 0:0002 þ 0:3170  ΔSoyt þ 0:1249  ΔWheat t þ … ΔWheat̂t ¼ 0:0000 þ 0:4738  ΔCornt þ 0:0795  ΔOilt þ … ΔEthanol̂t ¼ 0:0001 þ 0:1849  ΔSoyt þ 0:2443  ΔCornt þ 0:0744 ΔOilt þ … ΔOil̂t ¼ −0:0004 þ 0:0860  ΔEthanolt þ … These contemporaneous relations clearly demonstrate that the returns on soy are most strongly affected by corn, but we also observe that oil has a direct impact on soy. For corn, we observe that soy has a strong and direct effect on corn, suggesting that there are bi-directional contemporaneous interactions between this pair of commodities. The effect appears to be stronger from soy to corn than vice versa, but the difference in the effect is not statistically significant.11 We further observe that wheat has a contemporaneous effect on corn. For wheat, the results suggest that corn has the strongest contemporaneous impact, but we also observe a weak effect of oil on wheat. Considering the corn–wheat pair, we note that there are bi-directional contemporaneous effects, but the effect of corn on wheat is much stronger than the other way around, a finding that is in line with Saghaian (2010).12 The asymmetry in their bi11 In unreported results, we perform a test for asymmetries (H0: αij = αji for each i ≠ j) using the bootstrap procedure explained in Section 3. Specifically, for each run in the bootstrap, we calculate the difference between parameter values (αij − αji) and compute empirical p-values. This test fails to reject the null hypothesis of no asymmetry. 12 Arshad and Hameed (2009) focus on the indirect effect of acreage competition between wheat and corn. However, that is probably not a major factor for wheat (as it is for soybeans) since wheat and corn have limited land overlap. There would exist two main transmission mechanisms from higher corn prices to higher soybean prices: first, the competition for planting acreage, since both crops share quite similar soil and climatic requirements. Moreover, corn and soybeans share several industrial uses (e.g. as animal feedstock) and substitution from pricier corn to soybeans could be another factor weighing on the latter's demand and, eventually, price (Avalos, 2014).

6

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

Table 2 Reduced-form VAR dynamic coefficients.

Soyt Cornt Wheatt Ethanolt Oilt

Soyt − 1

Cornt − 1

Wheatt − 1

Ethanolt − 1

Oilt − 1

c

R2(adj) in %

0.0477 (1.53) 0.0341 (0.94) 0.0098 (0.24) −0.0375 (−1.16) 0.0296 (0.71)

0.0069 (0.24) 0.0332 (0.90) −0.0025 (−0.06) −0.0063 (−0.20) 0.0048 (0.12)

−0.0258 (−1.25) −0.0386 (−1.44) 0.0148 (0.41) −0.0138 (−0.58) −0.0395 (−1.33)

0.0008 (0.04) 0.0382 (1.46) −0.0366 (−1.35) 0.1313*** (3.95) −0.0425 (−1.59)

−0.0466*** (−2.78) −0.0440** (−2.15) −0.0573*** (−2.77) −0.0150 (−0.82) −0.0269 (−0.75)

0.0003 (0.78) 0.0002 (0.44) 0.0000 (−0.08) 0.0001 (0.34) −0.0004 (−0.88)

0.33 0.28 0.32 1.06 0.16

Note: This table reports the coefficients for the reduced form VAR(1). We report the coefficients with White corrected t-statistics in parentheses. We indicate significance at the 10%, 5%, and 1% level by *, **, and ***, respectively. Figures printed in bold indicate those cases where we find significant evidence of Granger causality at the 5% level.

Table 3 Contemporaneous interactions.

Soy Corn Wheat Ethanol Oil

Soy

Corn

Wheat

Ethanol

Oil

1 −0.3170*** (0.0000) −0.0533 (0.4620) −0.1849*** (0.0000) −0.1346 (0.1880)

−0.2492*** (0.0000) 1 −0.4738*** (0.0000) −0.2443*** (0.0000) −0.0304 (0.7660)

−0.0066 (0.9180) −0.1249** (0.0280) 1 −0.0279 (0.4800) −0.0762 (0.2040)

−0.0454 (0.1500) −0.0692 (0.1060) 0.0192 (0.6800) 1 −0.0860* (0.0940)

−0.0841** (0.0340) −0.0420 (0.2060) −0.0795* (0.0760) −0.0744*** (0.0020) 1

Note: This table reports the results for the contemporaneous interactions between the different commodities. We report the coefficients of the contemporaneous relations matrix A, along with the bootstrapped p-values in parentheses. The table reports the contemporaneous effect of the variable in each column on the variable in each row. We indicate significance at the 10%, 5%, and 1% level by *, **, and ***, respectively.

directional effects is statistically significant. For ethanol, we observe that soy, corn and oil have a contemporaneous impact. We note that ethanol has a contemporaneous effect on oil, which may be due to a substitution effect, but this effect is statistically weak and should be interpreted with caution (Saghaian, 2010 reaches a similar conclusion in his analysis using directed graphs to uncover contemporaneous relations). Finally, ethanol has no immediate effect on the agricultural commodities.

5.3. Cumulative impulse response functions The previous section demonstrated the contemporaneous effects of one commodity on the others. Although these contemporaneous effects could be thought of as very short-run effects, they also affect the longrun impact of shocks to different commodities. We capture these longrun effects by considering cumulative impulse-response functions (IRF). In this section, we demonstrate that the IRFs based on a traditional reduced-form VAR yield very different results that those based on our structural VAR. The issue with IRFs from a reduced form VAR is that it is often not clear what shock needs to be applied. One could apply a unit shock to a series, but this ignores the correlation among the series, and would not give a realistic reflection of the impact of a shock to one series on the others. Likewise, taking the correlation into account when applying the shock ignores the structural relations among the series. Hence, not knowing the structural relations among the variables will make the application of a correct shock challenging, if not unreliable. Several resolutions are often proposed in the literature ranging from either assuming the structural relations among the variables based on theory to Cholesky decompositions of the covariance matrix of the residuals of the reduced-form VAR or making assumptions on long-run impacts. However, the outcomes of these IRFs will depend on what assumptions are made. A more generic approach is to use generalized impulse response function (GIRF) as developed by Koop et al. (1996) and Pesaran and Shin (1998). These GIRFs are not affected by the ordering of the variables in the VAR (in contrast to IRFs based on a Cholesky

decomposition), but still incorporate the correlation structure of Ω. However, the GIRFs do not use the actual contemporaneous relations between the variables to define the shocks in the impulse response function, as matrix A remains unidentified in the reduced form VAR. Since the identification through heteroskedasticity approach is able to uniquely identify the contemporaneous structural relations in A, we know the structure of the covariance matrix of the reduced form residuals, Ω = A−1ΣA−1′. Given that the structural shocks in Eq. (1) are uncorrelated, we can apply a unit structural shock, which translates into the following initial shock vector for the reduced form model,   E ηt jε jt ¼ 1 ¼ A−1 e j ;

where ej is the (5 × 1) structural shock vector of which the jth element is equal to one and all other elements are zero. We refer to these IRFs as the structural impulse response functions (SIRFs). In Fig. 1, we plot the cumulative impulse response functions up to 10 steps ahead, where we show the GIRF (left column) and the SIRF (right column), and the appropriate shock is applied to the series labeled above each plot. The first plot shows the impact of a shock to soy. According to the SIRF a shock in soy has a substantially lower impact on all other commodities than what the GIRF would attribute. Next, we analyze the impact of shocks in corn. Again, we observe remarkable differences in the outcome of the IRF. The impact of a shock in corn is, in general, less strong in the SIRF than in the GIRF. The GIRF shows a strong impact of corn on soy and wheat of the same magnitude, while the SIRF shows that the impact on soy is reduced almost by half. For the SIRF, we observe some impact of a shock in corn on ethanol, but the impact on oil is virtually zero. For shocks applied to wheat, we observe substantial differences in the outcomes according to the GIRF and the SIRF. According to the GIRF, shocks in wheat have a strong impact on corn and a moderate impact on soy and the energy commodities. However, the SIRF shows those shocks have a small impact on all the commodities. With regards to the two energy commodities, the GIRF shows a moderate

Fig. 1. Cumulative impulse response functions. The graphs show the (standardized) impacts of unit shocks the series specified above the graph. The left column plots the cumulative generalized impulse-response function (GIRF) based on the reduced form VAR, whereas the right column plots the cumulative structural impulse-response function (SIRF) based on the structural VAR.

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

7

8

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

5.4. Contemporaneous relations and crude oil prices states

Table 4 Long-run impact of shocks. Soy

Corn

Wheat

Ethanol

Oil

Panel A: Cumulative generalized impulse response Soy 1.0152 0.6425 0.4412 Corn 0.6113 1.0312 0.6277 Wheat 0.4425 0.6458 0.9911 Ethanol 0.4422 0.5719 0.3436 Oil 0.3015 0.2632 0.1783

0.4570 0.5617 0.4050 1.1186 0.3367

0.3394 0.2596 0.2006 0.2774 0.9632

Panel B: Cumulative structural impulse response Soy 1.0375 0.2814 0.0303 Corn 0.4183 1.0334 0.1192 Wheat 0.2315 0.4817 1.0092 Ethanol 0.2876 0.3385 0.0761 Oil 0.1960 0.1097 0.0525

0.0838 0.1537 −0.0020 1.1455 0.0563

0.0737 0.0682 0.0744 0.1225 0.9683

Note: This table reports the long-run impacts of shocks, measured by the cumulative impulse response function after 100 steps. Unit shocks are applied to the commodity listed in a column and the long-run effect of that shock is reported in each row. Panel A reports the results for the long-run impact of the cumulative generalized impulse response functions, whereas Panel B reports the results for the cumulative impulse responses based on the structural VAR.

impact of both ethanol and oil shocks on all commodities, while the SIRF shows very little impact on any of the commodities. The results from the impulse-response functions demonstrate clear differences in the impact of shocks on other commodities when basing the shocks on a structural versus a reduced form VAR. In Table 4, we report the results for the long-run impact of a unit shock to each series by reporting the 100-step ahead cumulative impulse response function. To make a comparison between the two approaches, we report the results for the GIRF in Panel A, and the results for the SIRF in Panel B. In each column of Table 4, we show the results of a unit shock to the variable in this column on all assets under consideration. Overall, Table 4 demonstrates that the long-run cumulative impacts are larger according to the GIRF. However, once we properly incorporate the contemporaneous interrelations among the five commodities, the impacts are much lower. Among all the commodities, the average impact of shocks in wheat on all the other commodities is more than 6 times bigger according to GIRF than that of the SIRF. Likewise, the impact of a shock in ethanol on wheat is 0.41 according to the GIRF but there is no impact according to the SIRF.

So far, we have shown that there are strong contemporaneous relations between agriculture and/or energy commodities with asymmetric effects between different commodity pairs. However, there is some evidence that both non-energy and energy commodity prices may present nonlinear price behaviors, specifically where demand and supply functions are non-linear (see, e.g., Kristoufek et al., 2014). Such non-linear demand and supply functions could lead to price-dependent comovements between assets. Kristoufek et al. (2014) indeed demonstrate the existence of the price-dependent transmissions. Likewise, Chen et al. (2010) provide evidence of positive short-run links between crude oil and grain prices, which are attributed to the influence of biofuels, showing that such a link is especially relevant during high crude oil price periods. In a related study, Tyner (2010) argues that oil and corn prices moved together tightly over the 2006–2008 period because higher (lower) oil and gasoline prices induced rapid changes in ethanol production which put upward (downward) pressure on corn prices. To get an idea of whether the contemporaneous effects could be price-dependent, in Fig. 2, we plot the daily evolution of cumulative returns of agriculture and energy commodities where the shaded areas represent days where the price of oil is higher than the full sample mean (US$ 81.24). The graph provides some indication about the evolution in cumulative returns of agricultural commodities. We observe that they are similar in periods when the oil price is high compared with periods when the oil price is low. This graph suggests that there could be different dynamics between agricultural and energy commodities depending on whether the price of energy, i.e. crude oil, is high or low. Given the evidence of non-linear, price-dependent relations between fuel, biofuel and agricultural commodities based on the extant literature and our observation from Fig. 2, we analyze whether the contemporaneous relations differ in states where the price of oil is high or low. We define two states in oil prices, one state of high oil prices, where the price of oil is higher than the sample mean price and a state of low oil prices, where the price of oil is lower than the sample mean price. We estimate the structural VAR for both states separately, and obtain the matrix A in each of the two states. In Table 5, we report the contemporaneous relations of matrix A in the high and low oil price states, along with their bootstrapped p-values reported in parentheses. We observe several interesting results. First, we find that the contemporaneous effect of oil on the

Fig. 2. Commodities and oil prices states. The graphs show the cumulative log returns of soybeans, corn, wheat, ethanol and crude oil starting in the beginning of our sample where the shaded areas represent days where the oil price is higher than the full sample mean price from 1 June 2006 to 22 January 2016.

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

9

Table 5 Contemporaneous interactions in high versus low oil price states. Corn

Wheat

Ethanol

Oil

Panel A: High price in crude oil Soy 1 Corn −0.3143*** (0.0000) Wheat −0.0395 (0.6520) Ethanol −0.2059*** (0.0020) Oil −0.1430 (0.2320)

Soy

−0.3117*** (0.0000) 1 −0.5538*** (0.0020) −0.4388*** (0.0000) 0.1474 (0.3880)

0.0325 (0.7700) −0.0722 (0.3560) 1 −0.0738* (0.0700) −0.0832 (0.6020)

−0.0262 (0.5760) −0.0540 (0.1680) 0.0557 (0.2600) 1 −0.0409 (0.3120)

−0.0923 (0.3240) −0.2633* (0.0740) −0.1343 (0.2960) −0.1211** (0.0220) 1

Panel B: Low price in crude oil Soy 1 Corn −0.2772** (0.0200) Wheat 0.0338 (0.8580) Ethanol −0.0450 (0.4420) Oil −0.3156 (0.1140)

−0.3914*** (0.0000) 1 −0.3669*** (0.0060) −0.2857*** (0.0020) 0.0194 (0.9180)

0.0236 (0.7100) −0.2722*** (0.0000) 1 0.0088 (0.9180) −0.1448 (0.1560)

−0.0435 (0.2780) 0.0056 (0.9360) −0.0262 (0.6820) 1 −0.0920 (0.2380)

−0.0813** (0.0200) −0.0082 (0.9560) −0.0190 (0.6980) −0.0932*** (0.0000) 1

Note: In this table, we report the coefficients in matrix A for states when crude oil prices are higher than the full sample mean (Panel A) and when they are lower than the full sample mean (Panel B). We report coefficients, along with their bootstrapped p-values in parentheses. The table reports the contemporaneous interactions of the variable in each column on the variable in each row. We indicate significance at the 10%, 5%, and 1% level by *, **, and ***, respectively.

agricultural commodities is much stronger in the state where the oil price is high. Second, the effects of the agricultural commodities on ethanol are stronger in the high price state.13 We can explain the findings in Table 5 in two ways. First, high oil prices can cause an escalation in the costs of fertilizers, food transportation, and industrial agriculture. As a consequence, agricultural prices are more sensitive to oil price movements when the oil price is high. Second, a high oil price leads to higher fuel price, which increases the demand for biofuel as a substitute for gasoline (see, e.g., Chang and Su, 2010) or can encourage governments to promote biofuel production in an attempt to reduce their dependence on oil. Consequently, ethanol prices would be affected by movements in agriculture commodity prices. Both episodes have been blamed as some of the main causes of the 2006–2008 and 2010–2011 global food price crises. Other contemporaneous effects reported in Table 5 show that the bidirectional interactions between soy and corn are maintained in both oil price states. We also observe a notable difference in the contemporaneous interactions between corn and wheat. In high oil price state, the contemporaneous effect strongly runs from corn to wheat, while in a low oil price state there is bi-directional spillover. A possible explanation for this observation is that in a high oil price state, corn will have high usage for ethanol production, and is not considered much as a substitute for wheat as a feed crop. Instead, in a low oil price state, corn will be used more as a feed crop and hence price shocks in wheat will affect the price of corn and vice versa. 6. Conclusions In this paper, we use the identification through heteroskedasticity approach to examine the contemporaneous relations between energy (crude oil and ethanol) and agricultural (soybeans, corn and wheat) commodities. We show that traditional VAR analysis based on lead– lag relations does not fully capture the co-movements among these commodities, whereas the structural VAR can identify the contemporaneous relations. We document that these contemporaneous effects are important and in several cases asymmetric. Empirically, we show that crude oil has a direct impact on other commodities, while we do not observe contemporaneous effects in the opposite direction. We further note that corn and soybean have a greater impact on ethanol than the other way around, and that there are bi-directional effects between the pairs of soybean–corn and corn–wheat. We also show, through impulse-response analysis, that correctly accounting for the short-run relations has important implications for the long-run too, where shocks applied to a traditional reduced form VAR lead to very different 13 We do note a considerable reduction in statistical significance of the coefficients in both sub-samples, which could be due to the reduction in sample size when the sample is split.

outcomes than shocks applied to the structural VAR. Finally, we report that the contemporaneous relations are dependent on the price level of crude oil. Stronger contemporaneous effects from crude oil to agricultural commodities and from agriculture commodities to ethanol are observed in high oil price state. Overall, our study highlights the importance of taking the contemporaneous interactions among energy and agricultural commodities into consideration. We demonstrate that not properly accounting for these interactions could lead to erroneous interpretations about the relations among the commodities in our study. For example, the knowledge of a unidirectional effect from the agricultural commodities (corn and soybean) – mostly used in the biofuel production – on ethanol, could be useful in designing an appropriate hedging strategy. Based on a traditional VAR analysis, traders in corn and soy would hedge against shocks in ethanol. However, our analysis shows that the causality runs from soy and corn to ethanol, and thus such a hedge would not be necessary. Thus, the results of our paper should be of interest to commodity investors who trade in these products, or any other market participants, with an interest in knowing the interrelations among these energy and agricultural futures.

References Andersen, T.G., Bollerslev, T., Diebold, F.X., Vega, C., 2007. Real-time price discovery in global stock, bond and foreign exchange markets. J. Int. Econ. 73, 251–277. Arshad, F.M., Hameed, A.A., 2009. The long run relationship between petroleum and cereal prices. Glob. Econ. Finance J. 2, 99–100. Avalos, F., 2014. Do oil prices drive food prices? The tale of a structural break. J. Int. Money Financ. 42, 253–271. Boons, M., de Roon, F., Szymanowska, M., 2014. The price of commodity risk in stock and futures markets. Working Paper. Nova School of Business and Economics. Campiche, J., Bryant, H., Richardson, J., Outlaw, J., 2007. Examining the evolving correspondence between petroleum prices and agricultural commodity prices. Paper Presented at the American Agricultural Economics Association Annual Meeting, Portland, OR, July 29–August 1, 2007. Chang, T.H., Su, H.M., 2010. The substitutive effect of biofuels on fossil fuels in the lower and higher crude oil price periods. Energy 35, 2807–2813. Chatrath, A., Miao, H., Ramchander, S., 2012. Does the price of crude oil respond to macroeconomic news? J. Futur. Mark. 32, 536–559. Chen, S.T., Kuo, H.I., Chen, C.-C., 2010. Modeling the relationship between the oil price and global food prices. Appl. Energy 87, 2517–2525. Ciaian, P., Kancs, A., 2011a. Food, energy and environment: is bioenergy the missing link? Food Policy 36, 571–580. Ciaian, P., Kancs, A., 2011b. Interdependencies in the energy–bioenergy–food price systems: a cointegration analysis. Resour. Energy Econ. 33, 326–348. Du, X., Yu, C.L., Hayes, D.J., 2011. Speculation and volatility spillover in the crude oil and agricultural commodity markets: a Bayesian analysis. Energy Econ. 33, 497–503. Ehrmann, M., Fratscher, M., Rigobon, R., 2011. Stocks, bonds, money markets and exchange rates: measuring international financial transmission. J. Appl. Econ. 26, 948–974. Hamilton, J.D., 2009. Understanding crude oil prices. Energy J. 30, 179–206. Han, L., Zhou, Y., Yin, L., 2015. Exogenous impacts on the links between energy and agricultural commodity markets. Energy Econ. 49, 350–358. Hansen, L., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50, 1029–1054.

10

A. Fernandez-Perez et al. / Energy Economics 58 (2016) 1–10

Harri, A., Nalley, L., Hudson, D., 2009. The relationship between oil, exchange rates, and commodity prices. J. Agric. Appl. Econ. 41, 501–510. Hong, H., Yogo, M., 2012. What does futures market interest tell us about the macroeconomy and asset prices? J. Financ. Econ. 105, 473–490. Hou, K., Szymanowska, M., 2015. Commodity-based consumption tracking portfolio and the cross-section of average stock returns. Working Paper. Ohio State University. Janda, K., Kristoufek, L., Zilberman, D., 2012. Biofuels: policies and impacts. Agric. Econ. 58, 367–371. Ji, Q., Fan, Y., 2012. How does oil price volatility affect non-energy commodity markets? Appl. Energy 89, 273–280. Kilian, L., 2009. Not all oil price shocks are alike: disentangling demand and supply shocks in the crude oil market. Am. Econ. Rev. 99, 1053–1069. Koop, G., Pesaran, M.H., Potter, S.M., 1996. Impulse response analysis in nonlinear multivariate models. J. Econ. 74, 119–147. Kristoufek, L., Janda, K., Zilberman, D., 2012. Correlations between biofuels and related commodities before and during the food crisis: a taxonomy perspective. Energy Econ. 34, 1380–1391. Kristoufek, L., Janda, K., Zilberman, D., 2014. Price transmission between biofuels, fuels, and food commodities. Biofuels Bioprod. Biorefin. 8, 362–373. Kristoufek, L., Janda, K., Zilberman, D., 2016. Comovements of ethanol-related prices: evidence from Brazil and the USA. Glob. Change Biol. Bioenergy 8, 346–356. Lehecka, G.V., Wang, X., Garcia, P., 2014. Gone in ten minutes: intraday evidence of announcement effects in the electronic corn futures market. Appl. Econ. Perspect. Policy 36, 504–526. Meyers, R.J., Johnson, S.R., Helmar, M., Baumes, H., 2014. Long-run and short-run comovements in energy prices of agricultural feedstocks for biofuel. Am. J. Agric. Econ. 96, 991–1008. Natalenov, V., Alam, M.J., McKenzie, A.M., Van Huylenbroeck, G., 2011. Is there comovement of agricultural commodities futures prices and crude oil? Energ Policy 39, 4971–4984. Nazlioglu, S., 2011. World oil and agricultural commodity prices: evidence from nonlinear causality. Energ Policy 39, 2935–2943. Nazlioglu, S., Erdem, C., Soytas, U., 2013. Volatility spillover between oil and agricultural commodity markets. Energy Econ. 36, 658–665. Pesaran, H.H., Shin, Y., 1998. Generalized impulse response analysis in linear multivariate models. Econ. Lett. 58, 17–29. Qiu, C., Colson, G., Escalante, C., Wetzstein, M., 2012. Considering macroeconomic indicators in the food before fuel nexus. Energy Econ. 34, 2021–2028.

Reboredo, J.C., 2012. Do food and oil prices co-move? Energ Policy 49, 456–467. Rigobon, R., 2003. Identification through heteroskedasticity. Rev. Econ. Stat. 85, 777–792. Rigobon, R., Sack, B., 2003a. Spillovers across U.S. financial markets. NBER Working Paper vol. 9640. Rigobon, R., Sack, B., 2003b. Measuring the reaction of monetary policy to the stock market. Q. J. Econ. 118, 639–669. Rigobon, R., Sack, B., 2004. The impact of monetary policy on asset prices. J. Monet. Econ. 51, 1553–1575. Saghaian, S.H., 2010. The impact of the oil sector on commodity prices: correlation or causation? J. Agric. Appl. Econ. 42, 477–485. Serra, T., Zilberman, D., 2013. Biofuel-related price transmission literature: a review. Energy Econ. 37, 141–151. Serra, T., Zilberman, D., Gil, J.M., Goodwin, B.K., 2011. Nonlinearities in the US corn–ethanol–oil–gasoline price system. Agric. Econ. 42, 35–45. Taheripour, F., Tyner, W.E., 2008. Ethanol policy analysis — what have we learned so far? Choices 23, 6–11. Tyner, W.E., 2010. The integration of energy and agricultural markets. Agric. Econ. 193–201. Vacha, L., Janda, K., Kristoufek, L., Zilberman, D., 2013. Time–frequency dynamics of biofuel–fuel–food system. Energy Econ. 40, 233–241. Wetzstein, M., Wetzstein, H., 2011. Four myths surrounding U.S. biofuels. Energ Policy 39, 4308–4312. Wolf, M., 2008. Life in a tough world of high commodity prices. Financial Times, March 4. Zhang, Q., Reed, M., 2008. Examining the impact of the world crude oil price on China's agricultural commodity prices: the case of corn, soybean and pork. Paper Presented at the Southern Agricultural Economics Association Annual Meetings, Dallas, Texas, February 2–5, 2008. Zhang, Z., Lohr, L., Escalante, C., Wetzstein, M., 2010. Food versus fuel: what do prices tell us? Energ Policy 38, 445–451. Ziegelback, M., Kastner, G., 2011. European rapeseed and fossil diesel: threshold cointegration analysis and possible implications. Paper Presented at the 51st Annual Meeting of GEWISOLA, Corporative Agriculture: Between Market Needs and Social Expectations. Zilberman, D., Hochman, G., Rajagopal, D., Sexton, S., Timilsina, G., 2013. The impact of biofuels on commodity food prices: assessment of findings. Am. J. Agric. Econ. 95, 275–281.