Limits and Uses of Price Tests for Market Definition

Various quantitative tools have been developed for defining markets in competition law investigations. Econometric tests on price co-movement represent one such set of tools: two regions or products are considered part of the same market if their prices co-move. However, price co-movement tests, especially the more advanced econometric tests, have been criticized in the competition policy literature. Critics claim that price tests focus only on price linkages and, furthermore, that some of the advanced price tests are misleading in small samples. This paper applies a range of price tests, including correlation analysis, Granger-causality tests, unit root tests and the recent autoregressive distributed lag (ARDL) bounds test, to data from the 2006-2008 competition investigation into business practices in the South African dairy industry. We argue that the different price tests ask different questions and that it is not useful to dismiss an advanced price test if it suggests a different market than that identified by a simple correlation statistic. We also consider the criticism of poor small-sample performance of price tests: many conventional tests have long been shown to suffer from small-sample power and size problems, but critics fail to account for recent improvements in this regard. The paper concludes that the combination of various price-test results offers a rich picture useful for market definition purposes, especially if they are employed as exploratory tools rather than confirmatory ones.

L40, L11, C32, D4

market definition, price correlation, unit root, bounds test, law of one price