Previous studies of learning-by-doing typically impose constant returns to scale and either a constant markup or a specific market structure. In this research I develop and estimate a model allowing for nonconstant returns to scale and variable markups. I apply the model to the semiconductor industry, an industry which often incorporates special industrial polices as a result of learning-by-doing. The data are quarterly observations on firm sales, prices, and characteristics from 1974 through 1993 and cover over 90 percent of total market sales. I estimate the model for multiple generations that exist over the time period. I find there are, generally, increasing returns to scale in the market and that the imposition of constant returns to scale biases estimates of learning to finding greater learning. Policy in this market often relates to dumping allegations. Overstated learning rates may lead to practices that encourage dumping. U.S. firms then lose market share and greater production takes place overseas.
展开▼
