Forecasting Implied Volatility Alpha Asset Management Roger Kramer Brian Storey Matt Whalley Kristen Zolla
Objective and Methodology Objective Develop a model that forecasts the CBOE Volatility Index (VIX) Methodology Sampling frequency – weekly Extensive variable development Sample size – 313 observations Two 50-week holdout samples
Regression Model Predictive Variables VIX Level, Lag 1 Intuition: Mean reversion Negative correlation Change in 10-Yr U.S. Treasury Yield, Lag 2 Intuition: “Flight to quality” precedes equity market volatility Negative correlation Change in S&P 100 (if positive), Lag 1 Intuition: Volatility affected by momentum effect of equity market Positive correlation
Final Model: Out-of-Sample Performance First out-of-sample 213-week sample; 50-week holdout for validation Correct direction forecast: 64% Second out-of-sample 263-week sample; 50-week holdout for validation Correct direction forecast: 60%
Regression Analysis Summary Statistics
Final Model Results Trading Strategy #1 Continuous trading – long or short every week Correct direction forecast: 57.5% VIX % Change > 0: 72.1% VIX % Change < 0: 43.4% Mean return (weekly): 3.89% Mean return (winning weeks): 11.26% Mean return (losing weeks): -6.14% Standard deviation (weekly): 11.82% Cumulative return (10/95 – 1/03): 2,300,000%
Final Model Results Trading Strategy #2 Trade if absolute forecasted change in VIX exceeds 5% Trade in 73 of 313 weeks (23.3% of the time) Correct direction forecast: 63.0% Mean return if trading (weekly): 5.95% Mean return overall (weekly): 1.39% Standard deviation (weekly): 5.73% Cumulative return (10/95 – 1/03): only 4,591%
Final Model Results
Issues and Recommendations VIX is Not Tradable Develop an options trading strategy using VIX forecasts; i.e. buy/sell OEX straddles Examine the effects of transactions costs Naive Entry/Exit Signals Double moving average crossovers Various thresholds for forecasted VIX changes
Conclusions Our simple model predicts VIX direction with reasonable precision With further research, similar model could be used for profitable trading