What drives food price volatility? Evidence based on a generalized VAR approach applied to the food, financial and energy markets

The aim of this study is to investigate sources of food prices volatility. The analysis uses daily series for volatility of corn, soybean, wheat, rice, US dollar, crude oil, and SP500 futures spanning the period January 4, 2000 to April 1, 2017. The authors employ the generalized vector autoregressive framework in rolling sample approach in order to capture the time-varying nature of volatility spillovers. The results reveal that: volatility spillovers measures change over time; most of the volatility spillovers are observed within the two groups of markets: food markets and “non-food” markets; corn market is net volatility transmitter. JEL Q17 G15 C58


Introduction
The 2007-2008 and 2010-2011 surges in food prices were not concentrated in a market for a single agricultural commodity but resulted from developments in the whole range of markets for commodities grown in different places. That is why supply-side factors could not have been the only reason for the price co-movement. Alternative explanations for food price upsurges have been put forward in the literature and include e.g.: financial speculation in commodity futures markets, global economic growth (increased demand), trade restrictions, macroeconomic shocks to money supply, the US exchange rate movements (Abbott et al., 2009(Abbott et al., , 2011Gilbert, 2010;Roache, 2010), competition for land (Harvey and Pilgrim, 2011), countries' aggressive stockpiling policies, and tightening relations between food prices and energy prices.
In the literature on food price volatility, however, two reasons are investigated most frequently. The first one is "finacialization" of commodity markets which results from the development of future market trading. Deregulation of commodity markets, initiated in the beginning of the 21 st century, induced an increase in the inflow of capital to commodity futures (Christoffersen, 2014). Crucial for co-movement of commodity futures is the large inflow of commodity index investment (Tang and Xiong, 2012). The prices that underlie agricultural commodity indices are more strongly correlated with the oil price than those that are not included in the indices. The increase of correlation between futures prices of agricultural commodities and oil after 2004, as observed by Tang and Xiong (2012), resulted from significant index investments which started to flow into commodity markets. The second most frequently investigated reason of increased food price volatility concerns the relation between food and energy prices. The relation is bidirectional. On the one hand, modern food production requires more and more energy, e.g. to power agricultural machinery, to heat greenhouses, to power irrigation systems, to produce fertilizers, etc.. On the other hand, some agricultural products are used as a source of energy (biofuels). In the United States corn is used as the main feedstock to produce ethanol. This has resulted in tighter competition for the harvested area: the area used for biofuels (corn) production increased, as fuel ethanol production grew eight-fold from 233 trillion Btu in 2000 to 1,938 trillion Btu in 2014 (EIA), and the land on which it was grown could not be used for other crops.
The objective of the paper is to identify the main sources of food price volatility. Apart from developments in other food markets, the likely sources of food price volatility can include the US stock, energy and foreign exchange markets. We focus on the food prices volatility in the 21 st century, i.e. the period when many developing economies -and among them food exporters -have built tighter links with the world economy. Moreover, this period is of particular interest because it has witnessed both tranquil times and financial turbulence, as the Great Moderation was disrupted by the global financial crisis in the late 2000s.
The study is based on daily series for volatility of futures prices of corn, wheat, soybean 1 , rice, the US dollar, crude oil and the SP500 spanning the period January 4, 2000 to April 1, 2017. We base our analysis on forecast-error variance decompositions from a generalized vector autoregressive (VAR) framework, as proposed by Diebold and Yilmaz (2012). This framework allows us to estimate total, net, directional and pairwise volatility spillovers for markets considered. The generalized VAR is also used to obtain the impulse response functions that can be considered a complementary description of how volatility of the US stock, energy and foreign exchange markets affect food price volatility. Both a whole-sample approach and a rolling-sample approach are used in order to capture the time-varying nature of volatility spillovers. Bearing in mind that the number of parameters to be estimated in comparison to the number of observations is large, we use lasso estimation methods in a single iteration.
Our study is not the first one which examines the role of the stock and energy markets as drivers of food markets volatility (see: Diebold and Yilmaz, 2012;Chevallier and Ielpo, 2013;Jebabli et al., 2014;Awartani et al., 2016;Grosche and Heckelei, 2016). The advantage of our approach, however, is that it is among the most general ones.
Our contribution to the literature is visible in three aspects. First, we use a relatively large number of markets: the set we analyse in our study includes volatility of four main crops produced in the United States, which helps understand interrelations between them and volatility of the US stock, energy and foreign exchange markets. Second, we apply lasso estimation techniques and compare the results with those obtained with the ordinary least squares (OLS). The advantages of the lasso-based approach over the OLS-based approach become evident when the number of regressors is large. 2 What is more, the lasso-based approach seems to be more sensitive to differences in the evolution of the total volatility spillover index in turbulent and tranquil times. Third, apart from using the spillover indices, we extend the analysis by applying such tools as forecast error variance decompositions and 1 Christoffersen et al. (2014) show that corn, soybean, and wheat are among the most heavily traded commodity futures. In the period 2004-2013 there were more than 60 million transactions of soybeans and corn futures and more than 34 million for corn. 2 The issue is particularly important for large VAR models, when effectiveness of estimators seems to be crucial. It is well known (see Green, 2003, p. 49), that when a number of parameters in the model is large in comparison to the number of observations, the variance of OLS estimator increases. generalized impulse responses in order to uncover the direction and strength of volatility transmission. To illustrate the responses of food markets to different impulses we construct the heat plots.
Our findings include four novel results. First, we find that volatility spillovers are observed mostly within the group of food markets and within the group of other markets and much less between these groups. Second, the susceptibility of food markets to volatility spillovers from "non-food" markets, i.e. the stock, energy and foreign exchange markets, is larger during crisis periods. Third, the market for corn seems to be the most important source of volatility within food markets, as it is found to be the net volatility transmitter in most of the analysed subperiods. One may conjecture that the reason for this is that a large part of corn output is used to produce biofuels, and that there is an indirect relation between the food and energy markets. Fourth, the price of rice is detached from the developments in other markets, i.e. the sources of its volatility can hardly be found outside the market for rice.
The paper consists of the following sections. Section 2 presents the most important findings of the studies investigating volatility spillovers between the agricultural commodity, energy and financial markets. Section 3 describes the methodological approach, section 4 presents the data, and section 5 reports and comments on the empirical results. The paper ends with the conclusions.

Literature review
The two main methodological approaches are employed in the studies devoted to the issue of shocks transmission and volatility spillovers between energy markets, agricultural commodity markets (food markets) and financial markets (stock market).
Three main strands can be identified in the literature on the sources of food prices volatility. The first one focuses on relations between energy prices (including biofuel prices) volatility and food prices volatility. The second one examines volatility spillovers between 4 financial markets and commodity markets including both food and energy markets. 3 The third one considers relations between volatility within the food markets.
The results obtained within the first strand of literature suggest that in general volatility is transmitted from food markets to energy markets. Zhang et al. (2009) study price transmissions and volatility spillovers between weekly U.S. ethanol, corn, soybean, gasoline and oil prices and notice volatility transmissions from agricultural commodity prices to energy prices. In contrast, Trujillo- Barrera et al. (2012) show volatility transmission from crude oil to corn and ethanol markets and from corn market to ethanol market. Gardebroek and Hernandez (2013) show significant volatility spillovers from corn to ethanol prices but not the opposite. Mensi et al. (2014) investigate volatility spillovers across international energy (i.e. WTI, Brent, heating, and gasoline) and cereal commodity markets (i.e. wheat, corn, sorghum, and barley) and find that the correlations between the energy and cereal commodity futures evolve through time and are highly volatile, particularly since the subprime mortgage crisis. Serra et al. (2011) analyse Brazilian agricultural markets and confirm that ethanol price volatility is affected by shocks in the oil and sugar markets. Cabrera and Schulz (2016) find no volatility transmission between food, energy and biodiesel markets in Germany. Abderladi and Serra (2015b) consider food and biofuel prices in Spain and find bidirectional and asymmetric volatility spillovers between biodiesel and refined sunflower oil prices.
The results obtained in the second strand of literature reveal, in general, limited volatility transmission between food markets and financial markets (see, e.g., Silvennoinen and Thorp 2013;Chevallier and Ielpo 2013;Awartani et al. 2016), which, however, changes in time.
Volatility transmission increases during turbulent periods. Creti et al. (2013), Mensi et al. (2013), Silvennoinen and Thorp (2013), Diebold and Yilmaz (2012) show the strongest relationship between financial markets and food markets volatility during the global financial crisis. Jebabli et al. (2014) find that during the global financial crisis stock markets are a net transmitter of volatility shocks while a crude oil market is a net receiver. Kang et al. (2017) examine spillover effects among six commodity futures markets (gold, silver, WTI, corn, wheat, and rice) and find that both gold and silver are net volatility transmitters to other commodity markets, while the remaining four commodity futures (i.e. WTI, corn, wheat, and rice) are net receivers of volatility during the recent financial crises. Grosche and Heckelei (2016) reveal the strongest volatility spillover within the agricultural commodities in comparison to other markets.
The third strand of literature is devoted to the analysis of volatility transmissions within the agricultural commodity markets and shows different roles of a particular commodity. Beckmann and Czudaj (2014) argue that potential speculation effects on the corn futures market may be contagious for the cotton and wheat futures markets. Gardebroek et al. (2016) reveal that the markets for wheat and corn are major sources of volatility that spills over the market for soybean. Hamadi et al., (2017) find significant bidirectional volatility spillovers between markets for corn, wheat, soybeans and soybean oil, although a stronger spillover effect is observed from soybeans and soybean oil markets to corn and wheat markets. Pesaran and Shin (1998), building on the work of Koop et al. (1996), introduced the generalized impulse response function (GIRF) and the generalized forecast error variance decomposition (GFEVD) for unrestricted vector autoregressive (VAR) and cointegrated VAR models. Unlike the traditional IRF and FEVD, their approach does not require orthogonalization of shocks and is invariant to the ordering of the variables in VAR models.

Methodology
Since it is rarely possible to justify one particular ordering of variables under consideration, the methods promising to circumvent this restriction are of great interest to the scientific community.
Diebold and Yilmaz (2009) introduced a volatility spillover measure based on the traditional FEVD and focused on total spillovers (from/to each market, to/from all other markets). Later Diebold and Yilmaz (2012), building on the work of Pesaran and Shin (1998), used the GFEVD to introduce a spillover measure which is invariant to the variable ordering.
Additionally, Diebold and Yilmaz (2012) extended their previous work by introducing directional spillovers (from and to a particular market).
This study employs both the spillover indices as introduced by Diebold and Yilmaz (2012) and the GIRF analysis of Pesaran and Shin (1998). The spillover indices are constructed by performing a rolling-window generalized forecast error variance decompositions. This approach enables us to identify time-varying patterns. While the static GFEVD classifies the variables of the study into transmitters and receivers, the dynamic GFEVD may identify episodes when the role of transmitters and receivers of spillovers is interrupted or even reversed. The GIRF is also calculated within the rolling-window approach.
It is assumed that volatility is fixed within periods (in this case days), but can vary across periods. Following Alizadeh et al. (2002), daily high and low prices are used to estimate volatility 4 . The proxy we use is the logarithm of the difference between the highest and lowest log price: where t refers to a particular moment (day).
According to Alizadeh et al. (2002), the log range is superior as a volatility proxy to log absolute or squared returns as it is more efficient, and the log range distribution is closer to normality. This is particularly appealing as the generalized impulse responses and generalized forecast error variance decompositions require normality.
Our empirical strategy includes inference both from the whole sample and from the rolling windows. Each time, we repeat the following steps.
In the first step, the conventional VAR model is estimated. It takes the following form: is a vector of independently and identically distributed disturbances. All VARs are estimated using the lasso regression proposed by Tibshirani (1996). The lasso is a shrinkage method for a linear regression. It minimizes the sum of squared errors, with a bound on the sum of the absolute values of individual regression coefficients. Particularly in the rolling-window approach estimation degrees of freedom are substantially limited, so the application of pruning and shrinkage is quite appealing (Diebold and Yilmaz, 2015).
In the second step, total and directional spillover indices are obtained by generalized forecast error variance decompositions of the moving average representation of the VAR model. Variance decompositions allow for parsing forecast error variances of each variable into parts which are attributable to various system shocks. They allow for assessing the fraction of the H-step-ahead error variance in forecasting one variable that is due to shocks to another variable. The moving average representation of the VAR is: where the × coefficient matrices j A obey the recursion of form The H-step-ahead generalized forecast error variance decomposition invariant to the variable ordering is defined as: The index represents the average contribution of spillovers of volatility shocks across all the markets considered to the total forecast error variance.
Directional spillovers received by market i from all other markets j are defined as: Analogously, directional volatility spillovers transmitted by market i to all other markets j are defined as: The difference between directional volatility spillovers transmitted by market i to all other markets j ( → ( )) and directional spillovers received by market i from all other markets j ( ← ( )) is defined as net volatility spillovers of market i.
In the third step, GIRFs are calculated. An impulse response function depicts the time profile of the effect of shocks on the expected future values of variable in a dynamic system.
The scaled generalized impulse response function is calculated as The GIRs are calculated for a predefined time point ( = 1). In order to visualize GIR results, the R package 'superheat 0.1' for generating extendable and customizable heatmaps developed by Barter and Yu (2017) is used.

Data
We examine the volatility spillovers between the US stock, energy, foreign exchange markets and food markets using daily data spanning the period from January 4, 2000 to April 1, 2017, which yields 4239 observations. In particular, we examine the S&P 500 index futures contract traded on the CME (SP500) 1). In the second case, outliers can be observed in the period after 2008. Fortunately, the empirical strategy adopted in our study uses rolling-window approach, thus, in most cases, the assumption of normality is fulfilled.

Empirical results and discussion
Empirical strategy applied in the study consist of several steps. First, the volatility spillover table for the entire sample is estimated. Next, rolling windows analysis is carried out. The aggregated volatility spillover measures ("from", "to", "net") for each market are estimated. The aim of this step is to find markets that are net volatility transmitters or net volatility receivers in different windows. Then, the analysis moves on to the issue of sources of food prices volatility. Thus, some measures based on volatility spillover tables, which estimate the amount of volatility transmitted to food markets (from the stock, energy and foreign exchange markets, and food markets as well), are calculated. In particular, contributors to forecast error variance of food prices volatility are determined. Finally, the responses of food prices volatility to different shocks are studied within the VAR models with 5 lags that are estimated in rolling samples of 250 daily observations (about one year). The parameters of the models are estimated using the lasso method. The results obtained for the OLS-based approach are presented in the Appendix.

The full-sample results
We calculate the connectedness table based on variance decomposition for the full sample using the lasso estimation and report the results of in Table 2. Its ij-th entry denotes the estimated contributions to the forecast error variance of market i coming from innovations to market j. Therefore, the off-diagonal column sums (labeled 'to others') or row sums (labeled 'from others') are the "to" and "from" directional spillovers, respectively, and the "to minus from" differences are net directional volatility spillovers. The penultimate row in Table 2 reports the contribution of a volatility shock in a particular market to volatility observed in all other markets (stock, energy, foreign exchange and food). The volatility spillovers from all other markets to volatility in a given market is tabulated in the last column. The table of volatility spillovers may be viewed as the "input-output" decomposition of the total volatility spillover index.
As Table 2 demonstrates, the percentage of other markets in the US stock market (SP500) forecast error variance decomposition is 10.6%. At the same time, the US stock market (SP500) transmits about 13.7% of volatility to other markets. The difference between the amount of volatility transmitted to other markets and the amount of volatility received by the stock market is 3.1%, which means that the stock market is a net volatility transmitter. It is worth noting that the stock market transmits only 1.9% of volatility to food markets and 11.7% to other markets and the stock market receives only 1.9% of volatility from food markets and 8.7% from other markets. This means that the connectedness between stock market and food markets is more or less the same in both directions and weak if not negligible.

Similar calculations for the energy market (WTI) and the foreign exchange market (USD)
reveal that both markets are net volatility receivers with indices -0.3% and -2,5% respectively.
The contribution of the energy market to the food markets volatility and "non-food" markets volatility is 2.5% and 7.7%, respectively, and the energy market receives 2.8% of volatility from the food markets and 9.8% from other markets. Similarly, the contribution of the foreign exchange market to food markets volatility and other markets volatility is 3.4% and 8.8%, respectively, while the foreign exchange market receives 2.8% of volatility from the food markets and 9.7% from other markets.
Therefore, we may conclude that there exists some volatility spillovers from the stock, energy and foreign exchange markets to these markets, but not to the food (corn, soybean, wheat and rice) markets. We find that volatility spillovers between the stock, energy and foreign exchange markets and food markets are weak. Such a weak impact of the energy market on the food markets is also found by Awartani et al. (2016).
When the food markets are taken into account, the corn market seems to be the most important, as it is the net volatility transmitter with net volatility spillover index 2.5%. The corn market transmits as much as 34.5% of volatility to other markets, however, most of this volatility (about 32.6%) is transmitted to other food markets (mainly the soybean and wheat markets). The corn market is also the main receiver of volatility (32%) which comes from the food markets (30.2%). Other food markets are net volatility receivers: the net volatility index is -2.2%, -0.4% and -0.2% for the soybean, wheat, and rice markets, respectively. The rice market is specific in this respect, because it transmits about 3.1% of volatility to other markets and receives also only 3.3% of volatility from other markets. So, the rice market seems to belong neither to the food markets (which is surprising) nor to the "non-food" markets (which is natural). The different nature of the rice market can result from unique conditions required for rice production, which makes the problem of competition for land invalid, as no other crop can be grown on the same land that is used for rice.
In order to check if our results are robust, we calculate the connectedness table based on variance decomposition for the full sample using the OLS estimation. The results of the direction of implied volatility spillovers are presented in Table 2A in the Appendix. The comparison of the two methods applied reveal that in case of the lasso method there are less volatility transmission in the system. What is common for both approaches is that the same two marketsthe stock and corn marketsare the only net volatility transmitters.
As another robustness check we analyse the connectedness table estimated for the system extended by adding the VIX index, following Basak and Pavlova (2016). The results are reported in Table 1A in the Appendix. The results of the direction of implied volatility spillovers show that only the connectedness between the stock market (SP500) and the VIX index changes. The stock market transmits 33.4% of volatility to other markets, out of which 27.7% spills over the VIX. At the same time, the stock market receives 11.0% of volatility from other markets. So, in this case the stock market becomes the net volatility receiver (-22.3%) and the VIX becomes the net volatility transmitter to all markets in the system (29.5%). The remaining elements of the connectedness table do not change significantly.
It can be observed, however, that the food markets (the corn, soybean and wheat markets) and the "non-food" markets (the stock, energy and foreign exchange markets) constitute two separate clusters. The volatility transmission within such clusters is significantly larger than between clusters.

Total volatility spillover index
Many changes took place in the food markets during the sample period, i.e. between 2000 and 2017. Some of them affected the relations studied gradually, e.g. the increase in capital mobility, the importance of electronic trading and hedge funds, and a shift in the distribution of crops related to increased production of biofuels, while other changes, like surges in food prises in 2007-2008 and 2010-2011, exerted a sudden impact on the markets. This suggests that a dynamic approach should be used to analyse the relations. Following Diebold and Yilmaz (2012), we estimate our models using a rolling-window approach. Each window includes 250 days, approximately the number of working days in a calendar year. The underlying VAR model, estimated using the lasso method, has five lags, and the forecasting horizon is 10 days. We divide the sample period into subperiods in order to assess the evolution of the total volatility spillover index. The results obtained by the lasso method are presented in Fig. 2.
The index is about 14% for the initial subperiods. In the Appendix in Fig. 1A the total volatility spillover index obtained with the lasso method is compared with the one estimated with the OLS method. Two conclusions can be drawn from this comparison. First, the total spillover index estimated with the OLS is larger (its value ranges from 17% to 43%) than the one obtained with the lasso. Second, the difference between the two indices decreases during turmoil periods. This observation implies that lasso estimation is more appropriate to distinguish normal from extreme market conditions. Fig. 3 presents the directional volatility spillovers from the others to each of seven markets which are calculated using the lasso method (corresponding to the "from others" column in Table 2). 6 The results presented reveal different patterns of volatility transmitted by the food markets and the stock, energy and foreign exchange markets (SP500, WTI, USD).

Gross and net volatility spillovers
The volatility spillovers from the food markets to other markets display a cyclical behaviour.
In the entire period there are subperiods in which more (in comparison to other subperiods) volatility is transmitted (e.g. windows covering 2009, 2011 or 2016) and subperiods in which less volatility is transmitted (2004,2010,2015). The rice market is specific, as it transmits much less volatility than any other market. In case of the stock, energy and foreign exchange markets, the situation is quite different. In the initial superiods (up to 2007) little volatility is transmitted to other markets. Then, the volatility transmission increases significantly. Finally, for windows beginning since 2014 the amount of volatility transmitted decreases.  others which are calculated using the lasso method (corresponding to the "to others" row in Table 2). 7 In general, the results are quite similar to the ones presented in Fig. 3. Once again, there is a clear distinction between volatility received by the food markets (the corn, soybean and wheat markets) and the "non-food" markets (SP500, WTI, USD). What is more, the periods in which the largest amount of volatility is received by particular markets are similar to the ones obtained when the transmission of volatility is recorded. Once more, the rice market receives less volatility than other markets.

Sources of food prices volatility
The net volatility spillovers between pairs of markets in which one element belongs to the stock, energy and foreign exchange markets and the second element belongs to the food markets are estimated. The results are presented in Fig. 6. 9 In each case there are short periods when the food market dominates over other markets, i.e. is the net volatility transmitter (negative values in Fig. 6), and short periods of the opposite relation, i.e. the food market is the net receiver of volatility (positive values in Fig. 6). The volume of net volatility spillovers, however, is low for most subperiods. This suggests that the relations between volatility in the stock, energy and foreign exchange markets and volatility in the food markets are not very strong. In this respect our results are similar to those reported in many other studies (see, e.g., 9 Fig. 5A in the Appendix shows the pairwise net volatility spillovers between the US stock (SP500), energy (WTI) and foreign exchange (USD) markets and the food market obtained with the OLS method. In the remaining subperiods the rice market is the net volatility receiver.  In case of the FEVD of the soybean and wheat markets, apart from the importance of their own shocks, the second most important factor is the corn market. The share of the corn market in the FEV exceeds 20% in many subperiods. The third most important factor is the wheat market for the soybean market and the soybean market for the wheat market. The share of one of these agricultural commodities in the FEV of the other ranges from 4 to 10%.
Again, the role of stock, energy and foreign exchange markets is not significant. The FEVD of the rice market demonstrates that no market transmits significant amount of volatility to the rice market. It is worth mentioning that around 2011 the FEV of the rice market is accounted for by other factors in about 20%.
Apart from the volatility spillover indices, we intend to uncover the response of food volatility to shocks originating from the stock (SP500), energy (WTI) and foreign exchange (USD) markets and from the food markets. Thus, we calculate the generalized impulse response function for all rolling windows and different horizons. Fig. 9 presents, however, the response of food volatility to one standard deviation shock at one-day horizon. The colours The results illustrated in Fig. 9 can be summarized with three observations. 12 First, in most windows, food volatility increases as a result of shocks originating from all markets (warm colours dominate the heatplot). There are, however, subperiods in which the response of food volatility to other markets shocks are negligible or even negative (for example, the response of the corn market to the stock market (SP500) shock for windows covering 2006, or the response of any food market to the energy market (WTI) shocks in the same subperiod).
Second, the strongest, positive response of food volatility is observed for shocks generated on other food markets (the soybean and wheat markets response to the corn market shocks, or the corn market responses to the wheat or soybean markets shocks). The rice market seems to depend only weakly on other food shocks, as the response of the rice market to other shocks is moderate. Third, the strongest response of food volatility to shocks generated from the stock, energy and foreign exchange markets appear between 2007 and 2012. It suggests that during the global financial crisis and food crises food volatility was more sensitive to information coming into the market, and the responses of food volatility were more significant. Fig. 9. Response of the food markets to shocks originating from other markets (the lasso method).

Conclusion
The objective of the study is to examine volatility spillovers in the food markets and the "non-food" markets. Unlike in previous studies, we compare the volatility transmission within different food markets and between the food markets and the stock, energy and foreign exchange markets, which allows us to assess the importance of volatility of these markets in triggering volatility of the food markets. Our main findings can be summarized in the following way.
The total volatility index has increased over time, which means that more volatility is transmitted between the markets. The largest values are observed in two food crises (2008 and between 2011-2012). The results obtained for rolling directional spillover (from and to) reveal, however, a cyclical behaviour of the food markets and the increase of volatility spillover in the stock, energy and foreign exchange markets since 2007.
Most volatility transmissions are observed among the same categories of markets. We identify two groups which are interrelated in terms of volatility spillovers, i.e. the US stock, energy and foreign exchange markets and the food markets (including corn, soybean and wheat). A typical food market transmits much more volatility to other food markets than to other markets. This can result from heterogeneity of the stock, energy, foreign exchange markets. Volatility of the market for rice does not seem to depend on other markets and is not transmitted to other markets, with the exception of one episode, i.e. 2009, when the price of rice reached a record high and the volatility shocks spilled over soybean and wheat markets.
The sources of the forecast error variance of food markets volatility vary for different food markets and for different supberiods. The corn market, however, seems to be the most important agricultural commodity, as it transmits vast volatility to other food markets. The corn market is the net volatility transmitter for the soybean and wheat markets and is the second most important source of the FEV of these two kinds of food market, representing up to 20% of the FEV.
The results of the generalized impulse response functions suggest similar conclusions. The strongest response of food markets volatility results from shocks originating from another food market (with the exception of the rice market). Much smaller, but still a positive response of the food markets volatility to the shocks in the "non-food" markets can be observed. Finally, food markets volatility is more sensitive to shocks from different markets during the global financial crisis and surges in food prices.
The most general conclusion of the paper is that the role of the financial and energy markets in creating the food markets volatility is limited. In particular, volatility of energy prices appears to be insignificant for food prices. Interestingly, the corn market seems to be the most important food market, as it is the net volatility transmitter to the soybean, wheat and rice markets. Since the share of corn production used for biofuels (ethanol) has risen significantly during the analysed period, it can be concluded that the relations between energy and agricultural commodities markets have become tighter, although in an indirect way, i.e. via the market for corn.