Agricultural Commodities: Risk Management for Exporting Countries

Abstract

We consider three aspects of agricultural risk management: volatility modelling of commodity returns for several agricultural commodities, convenience yield modelling for various commodities and weather risk in Thailand, a supplier of rubber, sugar and rice. To model the volatility of commodity returns, we extend the GJR-Generalized Autoregressive Conditional Heteroskedasticity (GJR-GARCH). The inclusion of seasonal patterns, composed of a day-of-the-week effect (representing investor behaviour) and a yearly effect (representing harvest yields) are important in providing more accurate models of volatility. To capture fat tails, Standardised-t and the generalised error distribution (GED) are employed in estimations and compared with Gaussian error distribution. The Value-at-Risk (VaR) of the optimal volatility model's forecasting performances are used to determine the accuracy of these models. The second study examines convenience yield modelling and heteroskedasticity. The analysis of seven agriculture net convenience yields clearly shows that the benefit of net convenience yield exists only in the short term, otherwise it converges to zero. Both the current tests and a new proposed test confirm the existence of heteroskedasticity. An autoregressive model is used to model convenience yield: GARCH (1,1) and Standardised-t are more accurate than the alternatives considered. Finally, the net convenience yield is investigated in the context of the international market. We found that the depreciation of exchange rates of the leading exporting countries eliminates the benefit of holding the agricultural product. In the final study, we devised a weather insurance model, a hybrid error distribution that captures two important daily rainfall characteristics, to assess the weather risk to the Thai agricultural products. Our probability distribution combines a seasonal-Weibull distribution, which captures moderate rainfall, with an Extreme Value distribution, which captures extreme to heavy rainfall. Monte Carlo methods are used to simulate the daily data in order to compare our model performances with the actual data and estimate the rainfall insurance premium.