Assessing machine learning for forecasting economic risk: Evidence from an expanded Chinese financial information set

Abstract

While data sets used for forecasting can now be greatly improved, expanding data and information size also exposes weaknesses in traditional forecast models. We assess machine learning methods for forecasting monetary policy actions and concomitant macroeconomic risks. We construct an expanded information set on Chinese systemic risk, confirming that this set contains additional information useful for macroeconomic forecasting. We find that machine learning processes offer significant improvement for macroeconomic forecasting, with quantile regression forest exhibiting superior out-of-sample prediction accuracy compared with traditional methodologies. These findings will be of great interest to policy makers and investors.