Google trends and the predictability of precious metals

In this paper, we examine how Google trends can influence the predictability of the most traded precious metals, namely Gold, Palladium, Platinum and Silver. We construct a predictive model that simultaneously accounts for conditional heteroscedasticity, due to the use of high frequency data; endogeneity bias due to probable exclusion of important predictors and persistence due to the dynamic behaviour of economic agents. We also allow for asymmetry in Google trends (G-trends hereafter), to explain positive and negative worded news in the predictability of precious metals. The G-trends series appears to, positively and significantly, impact the returns accruable to the considered precious metals and its inclusion in the predictive model outperforms the random walk model. Also, additional information from other macroeconomic variables, as well as the negative and positive partial sums of G-trends, significantly improve the predictive performance of the single predictor G-trends-based model for precious metals. On robustness, while we find consistency of model forecast performance both in the in-sample and out-of-sample forecast periods, and across global factors employed, our results seem to be dependent on the choice of precious metal being predicted.