Modeling global real economic activity: Evidence from variable selection across quantiles

Abstract

We conduct an open search of predictors of global real economic activity. To this end, we apply a predictive quantile regression framework, using four alternative proxies of global real economic activity during February 1997–August 2019 and building on a combination of machine learning algorithms to identify their predictors out of 23 candidate explanatory variables. The contemporaneous level of global real economic activity, the Asian and US financial stress are found the most robust predictors. The effect of US financial shocks appears asymmetric, as they undermine global economic growth when the latter is below the median, but do not matter much when the world economy expands fast. Besides, US shadow interest rates are significantly and positively linked to global real economic activity. This effect holds in a high-growth regime of the world economy and suggests that rising US policy rates, contrary to the conventional wisdom, entail US dollar depreciation rather than appreciation. A weaker US dollar stimulates dollar-denominated cross-border bank inflows to the countries other than the USA, leading to a rise in real investment worldwide and industrial output growth. Thus, our empirical findings inform policymakers which indicators should be monitored more closely to predict future shifts in global economic growth and also provide certain insights about optimal policy responses to such shifts.