Serially Dependent Extreme Events in Agricultural Commodity Futures Markets

Abstract

Extreme price changes have become increasingly common in agricultural commodity futures markets. Many empirical studies have shown that agricultural commodity futures returns are not normally distributed and are heavy-tailed. However, most of the studies do not allow for stochastic dependence of extreme events over time. Statistical tools based on Extreme Value Theory can be utilized to model tail risk in agricultural markets. In this paper, we employ a Bayesian hierarchical model for serially-dependent extreme commodity futures price changes. The model assumes that the distribution of marginal price returns follows the generalized Pareto distribution (GPD), and reflects a serial dependence structure in tail distribution. The model proposed here allows both the parameters in the serial dependence function and the marginal GPD to vary over time. Thus, the model provides important information on changes in the shape of the heavy-tailed distribution. For empirical analysis, we use daily futures prices for corn. Based on our preliminary results, recent years have seen considerable increases in the probability of an extreme price decline in several commodity markets. These results have implications for risk management strategies as well as the design and effectiveness of federal insurance programs.