Measuring Substitution Patterns in Differentiated Products Industries∗

A very basic empirical question in industrial organization is the following: which products in a differentiated product market are close competitors with one another. This closeness of competition between two products is determined by the degree of consumer substitutability between them. Thus substitution patterns are the key to many supply side questions of interest. For example, the variation in substitution patterns among the products in a market can be used to study firm “conduct”: if there is a high degree of substitutability between the products of rival firms, then markups (and hence prices) should be systematically lower for these products when firms are competing as compared to colluding (Bresnahan 1981, Bresnahan 1987). Furthermore, for any particular hypothesis about firm conduct, substitution patterns drive the effect of counter-factual policy changes on market outcomes, such as mergers, new product introductions, etc. Although substitution patterns are central to empirical work in imperfectly competitive markets, identifying these substitution patterns from market price and quantity data has proven very challenging. The mixed-logit model of demand made famous by Berry et al. (1995) (henceforth BLP for short) can in principle approximate very rich substitution patterns by relaxing the strong ex-ante restrictions that the simple logit places on cross price elasticities (most notably the independence of irrelevant alternatives, aka IIA). This is achieved by allowing consumers to have unobserved taste heterogeneity for observed product characteristics, i.e., random coefficients in utility. While their approach has been hugely influential in providing a framework for studying differentiated product markets, there are very few direct applications (known to us) that have found statistically ∗Preliminary and Incomplete. Please do not cite without the permission of the authors. Gandhi and Houde acknowledge financial support from the NSF.