A False Discovery Rate Approach to Optimal Volatility Forecasting Model Selection

Forecasting volatility as a proxy to risk is critical to investment decisions and the study of investment behaviour. Toward this, a growing literature has attempted but not yet been able to identify an optimal volatility forecasting model. In this paper, we show that this is because the optimal model is dynamic. The choice of the optimal model depends on three factors: the asset in question; the period being studied; and the fitted innovation distribution. Therefore, we propose a novel empirical strategy that enables the researcher to select the optimal model as a function of these three factors. Using our Multiple Hypothesis Testing framework with False Discovery Rate approach, we are able to consistently identify significantly more accurate models relative to the literature’s benchmark volatility forecasting models of ARCH(1) and GARCH(1,1).