Scientific Beta

This note describes the volatility forecasting methodology used to produce monthly volatility forecasts for Scientific Beta's Max Volatility Protection and Historical Volatility Adjustment Risk Control options. Our approach seeks to capture the stylised properties of financial returns, including volatility clustering, leverage effects, heavy tails, and structural breaks. We adopt the GJR-GARCH model because it is an asymmetric conditional volatility model that can represent volatility clustering, when high or low volatility tends to persist over time, and leverage effects, when negative returns are associated with higher volatility. The model makes use of student-t innovations to reflect the heavy tails of financial returns. Finally, we average the monthly volatility forecasts from two GJR-GARCH models estimated on differing amounts of daily returns. Our first model makes its estimate on an expanding window, using all available data to improve the precision of estimated parameters. The second model uses a five-year rolling window, which is more useful if structural breaks are present since only relatively recent data is used to fit the model. Overall, we find that our forecasting framework is able to produce reliably accurate monthly volatility forecasts on US equity data compared with benchmarks such as the historical volatility, exponentially weighted moving average and the VIX implied volatility index.