Notes on estimating experience curves

Experience curves have for some time provided a conceptual framework for forecasting unit costs of new products or production activities. On a practical level, however, statistical estimation of experience curves involves a number of subtle model specification issues which if ignored can render estimated curves totally inappropriate. Several of these issues are explored in a heuristic setting. A major finding of the analysis undertaken is that many estimates of experience curves are estimates of models seriously flawed due to specification errors. First, price experience curves and the problems encountered in estimating these functions due to missing variables are treated. Then, attention is focused on cost experience curves. It is argued that estimates of cost experience curves are generally biased due to both missing variables and missing equations; the latter is usually termed simultaneous equation bias.