Winds of Change: Estimating Learning by Doing without Cost or Input Data
Published in Working Paper, 2022
We measure how much learning has reduced costs in the wind turbine industry. As in many industrial settings, we do not observe costs, so we infer them using standard demand system data. To measure wind farm developer preferences, we embed a simple but physically realistic model of how wind turbine characteristics, like rotor size, relate to power production, into a standard discrete choice demand system. Next, we use a standard oligopoly model to invert these preferences and recover manufacturing costs, and their dependence on cumulative manufacturing experience. Because current sales increase future experience, manufacturers have dynamic incentives when setting prices. We account for these dynamic markdowns using methods developed in Berry and Pakes (2000), which allow us to control for dynamics without computing the equilibrium of a dynamic game. We find that a doubling of manufacturing experience reduces manufacturing costs by 14 to 29 percent. Only 1 to 2 percent of experience spills over to other turbine models produced by the same firm, and spillovers to turbines produced by other firms are on the order 0.1 to 0.6 percent. Though inter-firm spillovers are small, in aggregate, they are responsible for significant cost reductions over time. These results are consistent with policymaker motivation for generously subsidizing the industry.
Join work with Richard Sweeney