A boosting approach to forecasting the volatility of gold-price fluctuations under flexible loss

We use a boosting approach to study the time-varying out-of-sample informational content of various financial and macroeconomic variables for forecasting the volatility of gold-price fluctuations. We use an out-of-sample R2 statistic to evaluate forecasts as a function of the shape of a forecaster's loss function. We show that, when compared to an autoregressive benchmark forecast, those forecasters tend to benefit from using predictions implied by the boosting approach who encounter a larger loss when underestimating rather than overestimating the future volatility of gold-price fluctuations. We use a simulation experiment to study the significance of this benefit.