Predicting exchange rate with commodity prices: New evidence from Westerlund and Narayan (2015) estimator with structural breaks and asymmetries

In this paper, we offer new evidence on the predictability of exchange rate with commodity prices using the Westerlund and Narayan (2015) estimator. The estimator allows us to control for any possible bias in the estimation process due to the presence of persistence, endogeneity and conditional heteroscedasticity effects in our predictors. We further account for the role of asymmetries and structural breaks both for the in-sample and out-of-sample forecasts. In particular, we evaluate whether such considerations matter for the forecast performance of the predictive model for exchange rate. Monthly data of five major tradable currency pairs in the world and disaggregated commodity price indices over the period of 1960–2017 are utilized. We find significant improvements in both the in-sample and out-of-sample forecast performance of the predictive model for exchange rate using this estimator and particularly when asymmetries and structural breaks are accommodated. While the results are robust to multiple currency pairs, commodity price indices, breaks and forecast horizons, however, the choice of data frequency has implications on the forecast outcome. Essentially, the forecast accuracy of the proposed predictive model seems to diminish with lower data frequencies.