Learning-by-Doing and Cannibalization Effects at Multi-Vintage Firms: Evidence from the Semiconductor Industry

Abstract

Previous studies on the measurement of learning-by-doing emphasize the importance of accounting for multi-vintage effects having an impact on firms' production costs through economies of scope. This study shows that accounting for cannibalization effects on the demand side is equally important for the adequate measurement of learning. Since multi-vintage firms anticipate the demand-side cannibalization effects in their production optimization, a previously omitted incentive to decrease production is captured having an impact on the measurement of learning by doing. We derive an empirical model from a dynamic oligopoly game of learning-by-doing and allow cannibalization effects to enter from the demand side. Using quarterly firm-level data for the dynamic random access memory semiconductor industry, we find support for cannibalization effects entering firms' pricing relations resulting in higher estimated learning effects. Abstract Previous studies on the measurement of learning-by-doing emphasize the importance of accounting for multi-vintage effects having an impact on firms' production costs through economies of scope. This study shows that accounting for cannibalization effects on the demand side is equally important for the adequate measurement of learning. Since multi-vintage firms anticipate the demand-side cannibalization effects in their production optimization, a previously omitted incentive to decrease production is captured having an impact on the measurement of learning by doing. We derive an empirical model from a dynamic oligopoly game of learning-by-doing and allow cannibalization effects to enter from the demand side. Using quarterly firm-level data for the dynamic random access memory semiconductor industry, we find support for cannibalization effects entering firms' pricing relations resulting in higher estimated learning effects. JEL: C1, L1, L6, O3. KEYWORDS:

Heterogeneous Merger Impacts on Competitive Outcomes

“…The estimate confirms that firms set prices in the elastic portion of the demand function, which is supported by oligopoly theory. The magnitude of the price elasticity is also similar to the results from previous studies (see, e.g., Brist and Wilson (1997), Siebert (2010), and Zulehner (2003)), which further confirms the reliability of our estimation results.…”

“…The production yield rate for a new product generation (defined as the percentage of wafers that successfully pass all production stages) starts at around 20 percent and drastically increases throughout the life cycle. Hence, the manufacturing process for each generation is characterized by significant learning-by-doing (or dynamic economies of scale) effects (see also Fudenberg and Tirole (1983) and Siebert (2010)). …”

“…Generations are typically considered to be homogeneous within a generation and heterogeneous across generations (see also Irwin and Klenow (1994), Siebert (2010), andZulehner (2003)). Figure 1b shows that the prices of every generation decline (relatively monotonically and continuously) over time.…”

Dynamic Model of Markets of Successive Product Generations

“…The DRAM market is one of the most closely watched markets of all integrated circuit categories. Several papers studied the technological evolution of DRAMs, trying to understand the dynamics of memory chip generations from different economic and technological perspectives [28][29][30][31]. We want to discuss here the empirical mean price and unit sales evolution of the DRAM market and compare them with the presented model.…”

“…As a proxy for µ i (t) we take advantage from empirical data of the average selling price (ASP) displayed in Fig. 1 in a half-logarithmic plot [22,29]. Also displayed is a fit of Eq.…”