Emerging Markets at Risk

Abstract

Policymakers would like to predict and mitigate the risks associated with the post-global financial crisis rise in corporate leverage in emerging markets. However, long-standing advanced-economy bankruptcy models fail to capture the idiosyncrasies that impact the solvency of emerging market firms. We study how a machine learning technique for variable selection, LASSO, can improve corporate distress risk models in emerging markets. Exploring the trade-off between model fit and predictive power, we find that larger models forecast distress with more accuracy during periods of economic stress (when global factors gain relevance), while more parsimonious specifications outperform during normal times.