Forecasting Productivity Using Information from Firm-Level Data

This paper contributes to the productivity literature by using results from firm-level productivity studies to improve forecasts of macro-level productivity growth. The paper employs current research methods on estimating firm-level productivity to build times-series components that capture the joint dynamics of the firm-level productivity and size distributions. The main question of the paper is to assess whether the micro-aggregated components of productivity---the so-called productivity decompositions---add useful information to improve the performance of macro-level productivity forecasts. The paper explores various specifications of decompositions and various forecasting experiments. The result from these horse-races is that micro-aggregated components improve simple aggregate total factor productivity forecasts. While the results are mixed for richer forecasting specifications, the paper shows, using Bayesian model averaging techniques (BMA), that the forecasts using micro-level information were always better than the macro alternative.