A Bayesian Latent Variable Mixture Model for Filtering Firm Profit Rate

By using Bayesian Markov chain Monte Carlo methods we select the proper subset of competitive firms and find striking evidence for Laplace shaped firm profit rate distributions. Our approach enables us to extract more information from data than previous research. We filter US firm-level data into signal and noise distributions by Gibbs-sampling from a latent variable mixture distribution, extracting a sharply peaked, negatively skewed Laplace-type profit rate distribution. A Bayesian change point analysis yields the subset of large firms with symmetric and stationary Laplace distributed profit rates, adding to the evidence for statistical equilibrium at the economy wide and sectoral levels.