Modeling Oil Market Volatility

This paper investigates which model applies best for reflecting volatility of oil market. Modeling oil market volatility is an important subject for both economic and financial analysis, since shocks on oil markets affects economic conditions of oil exporters and oil importers as well as returns on investment portfolios. There are many studies that search best model among various types of GARCH models, apply machine learning or investigate the relation among oil market and other economic variables. Our work extends previous researches by the fact that we investigate wider range of methods, that is first of all, analyzing both different time horizons (1-day, 1-week and 1-month-ahead) and different types of forecasting (rolling and expanding window); then we choose best model among several GARCH types models to which apply ANN, rather than randomly take one; similarly we investigate wider range of additional variables, like stock indexes or exchange rates, however use only those that Granger cause oil market volatility; and, finally, we look both on statistical and economic loss functions comparison. We concluded that among considered GARCH-type models Markov Switching regime model, consisting of two states GARCH and EGARCH, performs significantly better at all time horizons. Moreover, the application of ANN further decreases all loss functions, with reduction becoming more significant at longer horizons. Finally, the longer the forecasting horizon the more additional variables that Granger cause oil market volatility is better to use.