Volatility Modelling of Prices of Cocoa Beans Traded on the Intercontinenteal Exchange (ICE

Abstract

This study compares the performances of ARIMA, ARCH, GARCH, EGARCH and GJR-GARCH on their abilities in modelling and forecasting the volatility of world cocoa bean prices on the London Intercontinental Futures Exchange using weekly averages of prices in Great Britain Pounds from January 1985 through December 2017. It seeks to ascertain whether a model that provides a good in-sample fit necessarily provide the best out-of-sample forecast, in the midst of other models. A number of ARMA combinations were fitted to the first difference of the log of the price series and ARMA(3,0) is selected as the best model based on Akaike information criteria. A residual diagnostic check on the best fitted ARIMA model indicated the presence of ARCH-effects. ARCH-family models are fitted to the residuals with a mean equation of ARMA(3,0). With the help of loss functions, the study finds that GARCH(1,2) model outperforms other models for the in-sample modelling but for the out-of-sample forecast, GJR-GARCH(1,1) emerges as the best model. This study thus concludes that the model that provides a good in-sample fit does not necessarily provide the best out-of-sample forecast. More complex models also proved to be more effective in volatility modelling compared to simple models that do not incorporate asymmetry. Moreover, GJR-GARCH(1,1) is concluded by this study as the most adequate model for capturing volatility of world cocoa bean prices