Testing dynamic oligopolistic interaction: evidence from the semiconductor industry

This paper analyzes the impact of a dynamic specification on the estimation of the conduct parameter in an oligopolistic market. Various empirical studies have shown that in the semiconductor industry, in particular in the Dynamic Random Access Memory (DRAM) market, one has to account for dynamic elements as learning-by-doing within firms and learning spillovers among them. Therefore this market seems to be appropriate to investigate whether firms behave strategically in a dynamic sense and how open-loop or closed-loop as equilibrium concepts alter the size of the estimated parameters. I apply a structural oligopolistic model of dynamic nonprice competition that incorporates learning-by-doing and spillovers. Theory shows that learning-by-doing and learning spillovers have important consequences for firm behavior. Whether firms in the DRAM industry take the strategic effects of learning-by-doing and learning spillovers actually into account when choosing their output strategies, is answered with empirical evidence. Using quarterly data from 1974-1996 at the firm level, I estimate demand and pricing relations for three different generations of DRAM chips. The empirical results show that the game theoretic specification has an important impact and that firms behave strategically. The assumption of an open-loop specification would underestimate the conduct parameter on average about 50%.