A1609
Title: Firms heterogeneity and aggregate fluctuations: What can we learn from machine learning
Authors: Luigi Pollio - UMBC (United States) [presenting]
Simone Pesce - Boston College (United States)
Marco Errico - Bank of Italy (Italy)
Abstract: The heterogeneous sensitivity of firms to aggregate fluctuations affects business cycle dynamics. The aim is to examine how firms' outcomes (sales, debt, investment, market value) respond to aggregate fluctuations (business cycle, monetary policy, uncertainty, oil shocks) based on eight firm characteristics using the generalized random forest algorithm. Analyzing Compustat micro-level data, three key findings are documented: (1) while linear OLS captures the average effect, there is substantial cross-sectional heterogeneity in firm sensitivities; (2) the importance of firm characteristics varies across outcomes and shocks, with non-financial characteristics explaining more of the heterogeneity; (3) firm sensitivity to aggregate fluctuations shows non-linear patterns based on characteristics. An aggregation theory is developed, and the estimated model is used to generate counterfactual firm-level sensitivities. The findings reveal that (1) heterogeneity in firm sensitivity amplifies aggregate responses to macroeconomic variables; (2) non-linearity at the micro-level has little effect on aggregate responses; (3) non-financial characteristics drive aggregate responses more than financial ones; and (4) state-dependent heterogeneity plays a minor role in influencing aggregate responses.
