Volatility Modeling and Forecasting Efficiency of GARCH Models on Soy Oil Futures in India and The US

Abstract

This paper attempts to fill the gap in volatility studies in the commodity markets by modeling the volatility of the soy oil futures in two interrelated markets of India and the US. The GARCH (1,1), TGARCH (1,1) and EGARCH (1,1) models are tested under the assumptions of normal distribution, Student’s t distribution and general error distribution (GED). The results of our study indicate that there is high persistence of volatility in soy oil futures market and the volatility effect decays slowly with time. The half life for dissipation of volatility spikes in the US market is twice that of the half life in Indian market. The volatility models did not show leverage as the leverage term is found to be insignificant in all cases (p>0.05). A comparative analysis, based on Log Likelihood, AIC and SC criteria, of the three GARCH variants under three alternative distributions shows that refined soy oil returns in India are best modeled by GARCH (1,1) under GED while the bean oil returns in the US are best modeled by EGARCH (1,1). Forecasting efficiency of the GARCH models in the two markets is tested using the RMSE, MAE, MAPE and TIC. For soy oil in India, GARCH (1,1) under GED is best model by MAE and MAPE. For the bean oil in the US, EGARCH (1,1) under Gaussian distribution emerges as the best model based on RMSE and TIC criteria.

Keywords: Volatility, GARCH, Soy oil, Commodity, Forecasting