Forecasting crude oil price volatility out-of-sample using news-based geopolitical risk index: What forms of nonlinearity help improve forecast accuracy the most?

Abstract

We evaluate the degree to which the geopolitical risk (GPR) index suggested in Caldara and Iacoviello (2018) helps improve the accuracy of one-month ahead out-of-sample point (density) forecasts of West Texas Intermediate (WTI) crude oil price log-realized volatility, and what forms of nonlinearity offer the largest degree of relative forecast accuracy gains. The central finding in this study is to demonstrate that contrary to what is perceived in the current literature, the form of nonlinearity that delivers the largest degree of point (density) forecast accuracy gains relative to the benchmark does not have to do with asymmetry (i.e. positive or negative GPR changes) or higher than average GPR increases, but rather GPR increases beyond the maximum GPR value over the last twelve months. In other words, GPR increases help deliver the highest degree of forecast accuracy gains relative to the benchmark if they exceed the maximum value in recent memory. However, GPR decreases below recent minimums do not help deliver relative accuracy gains.