David McCarthy

Researchers modelling firms usually describe productivity through a linear AR(1) process with Gaussian innovations. Using Portuguese manufacturing data with firm-level prices and quantities, we show that empirical productivity deviates from an AR(1). First, the tails of the productivity distribution are wider. Second, firms at the tails face substantially more volatility than the median firm. Third, large shocks are more frequent than predicted by an AR(1) process, and some of them manage to erase a firm’s accumulated history (memory-wiping). Methodologically, we develop an algorithm to iteratively estimate the production function and the demand system, based on the control-function approach, while allowing productivity to follow a nonlinear process. We embed the estimated productivity process in a simple model of labour demand with adjustment costs. Increased tail dynamics reduce the magnitude of estimated frictions. The model becomes able to match the distribution of firm growth, including large changes in growth rates that characterise business dynamism.