Forecasting precious metal returns with multivariate random forests

We use multivariate random forests to compute out-of-sample forecasts of a vector of returns of four precious metal prices (gold, silver, platinum, and palladium). We compare the multivariate forecasts with univariate out-of-sample forecasts implied by random forests independently fitted to every single return series. Using univariate and multivariate forecast evaluation criteria, we show that multivariate forecasts are more accurate than univariate forecasts.