eThesis - School of Business electronic theses

This thesis shows that implied volatility is more accurate in volatility prediction in crude oil futures than GARCH models. Using robust loss functions and the adjusted coefficient of determination from regression equations as the comparison criteria, the implied volatility (IV) derived from nearest-to-the-money shortest time-to-expiration options outperforms volatility forecasts from 10 different GARCH specifications in two exchanges, NYMEX and ICE. The time period studied is August 2006 - January 2015 in NYMEX and August 2011 - January 2015 in ICE; the latter is shorter due to data availability. The robustness of the results are confirmed using sub time periods, put and call options separately in the computation of IVs, and including option trading volume and open interest variables in the regression.

The results of this study suggest a transfer of academic interest from GARCH models to implied volatility in crude oil volatility prediction. The mean IV is an unbiased and efficient predictor of future volatility in NYMEX, and an efficient predictor in ICE. Higher option trading volume increases the relative performance of IV further, indicating that options are more "correctly" priced when the market has enough liquidity. No-arbitrage conditions do not seem to fully apply, as call and put options yield different implied volatilities, and as the Black (1976) option pricing model employed is considered suitable for the options studied. Higher option trading volume mitigates the difference between call- and put-derived implied volatilities, and thus reduces violations against the Efficient Market Hypothesis (EMH) and the no-arbitrage conditions.

Nonlinear GARCH models are found more useful than linear GARCH models, though the exact best model varies from time period and market place to another. The limitations of the study are the time-specific nature and poor generalizability of some of the results into other time periods and exchanges than those covered in this study. The contributions of this study include comparing IV with ten different GARCH models in both NYMEX and ICE, which is not done in the literature on crude oil volatility previously. The effect of the 2008-2009 high-volatility period in crude oil returns is shown to have a large weakening impact on the model performance, which hasn't been studied in previous papers. In addition, no prior study in the knowledge of the author includes data on option trading volume and open interest in determining the effect of liquidity on the relative performance of IV and GARCH in crude oil volatility prediction.