A Sufficient Statistics Approach for Aggregating Firm-Level Experiments

We consider a dynamic economy populated by heterogeneous firms subject to generic capital frictions: adjustment costs, taxes and financing constraints. A random subset of firms in this economy receives an empirical "treatment", which modifies the parameters governing these frictions. An econometrician observes the firm-level response to this treatment, and wishes to calculate how macroeconomic outcomes would change if all firms in the economy were treated. Our paper proposes a simple methodology to estimate this aggregate counterfactual using firm-level evidence only. Our approach takes general equilibrium effects into account, requires neither a structural estimation nor a precise knowledge on the exact nature of the experiment and can be implemented using simple moments of the distribution of revenue-to-capital ratios. We provide a set of sufficient conditions under which these formulas are valid and investigate the robustness of our approach to multiple variations in the aggregation framework.