Endogeneity Problems in Empirical Analyses of Antitrust and Competition

In simple terms, endogeneity is the violation of an assumption in regression analysis that, if not properly addressed, may undermine the reliability of standard empirical analyses and of the inferences based upon them. Endogeneity arises when one or more of the explanatory variables included as controls in a regression model is correlated with the residual of the econometric model (i.e. with unobserved factors). Endogeneity is related to the refrain that “correlation is not causation” in that the presence of endogenous explanatory variables may invalidate causal claims from the estimation. Because well-founded antitrust examination must be based on true causal relationships, a basic understanding of endogeneity is of great importance for antitrust practitioners. In many settings, the observed association between a market outcome of interest (the dependent variable in a regression model) and its assumed determinants (the model’s explanatory variables) reflects a combination of true causation and non-causal correlation (i.e., variables moving together without a causal relationship between them). The presence of endogeneity distorts the naïve estimates, which are contaminated by the portion of the correlation that is non-causal. For example, an observed association between prices and quantity demanded may reflect both a true causal effect, whereby higher prices lead to lower demand, and a non-causal relationship, whereby, for example, expansions in demand coincide with higher prices without a causal relationship (an example of which we illustrate below). The problem can be so severe that it is not uncommon for an analysis that does not properly address endogeneity to erroneously estimate an upward sloping demand curve (i.e. higher prices yield greater demand).While the discussion surrounding endogeneity can quickly get technical, the purpose of this note is to make the topic approachable for non-economists and practitioners that do not have a background in econometrics. We explain the concept and consequences of endogeneity, as well as possible solutions, through a series of examples relevant to antitrust in non-technical and intuitive terms