This study aims to predict Taiwan stock returns through gold and crude oil futures prices using monthly data from TAIEX and 19 stock sector indexes from January 1996 to December 2020. By using a 60-month rolling window horizon, we compare the forecast performances of various regression models, where the forecast performances are measured by MAE (mean absolute error) and (out-of-sample R-square). In addition to using spot returns and the first principal component of futures returns on gold and crude oil, five traditional financial variables (dividend to price ratio, earnings to price ratio, market price to book value ratio, long-term yield, and short-term yield) are added to the regression model to explain and predict stock returns. Given that the regression models have included these traditional financial variables, the empirical results reveal that adding gold or crude oil price information to the model substantially improves its explanatory ability. Additionally, except during periods of high stock returns, the forecast ability of crude oil price information on stock returns is significantly better than traditional forecast variables. Furthermore, although gold prices are not as accurate as crude oil prices in predicting stock returns, their predictive capabilities are often better than the traditional financial variables.

Stock return; Gold futures; Crude oil futures; Principal component analysis; Forecast performance