Antitrust market definition using statistical learning techniques and consumer characteristics

Market definition is the first step in an antitrust case and relies on empirical evidence of substitution patterns. Cross-price elasticity estimates are preferred evidence for studying substitution patterns, due to advances in IO econometric modelling. However, the data and time requirements of these models weigh against their universal adoption for market definition purposes. These practical constraints — and the need for a greater variety of evidence — lead practitioners to rely on a larger set of less sophisticated tools for market definition. The paper proposes an addition to the existing toolkit, namely an analysis of consumer characteristics for market definition purposes. The paper shows how cluster analysis can be used to identify meaningful groups of substitutes on the basis of homogeneity of their consumer profiles. Cluster analysis enforces consistency, while recent bootstrap techniques ensure robust conclusions. To illustrate the tool, the paper relies on data from a recently concluded radio merger in South Africa.