Time-Series Models for Estimating Economic Damages in Antitrust (and Other) Litigation: The Relative Merits of Predictive versus Dummy-Variable Approaches

Economic damages can arise from a variety of bad acts: a breach of contract, a patent violation, an oil-spill, or a price-fixing conspiracy, for examples. The calculation of economic damages involves the description of a “but-for world.” The but-for world implies a set of economic values-such as profits or quantities or prices-that would have prevailed but-for the alleged bad acts. These but-for values become the basis for calculating damages.

A but-for world cannot be demonstrated with certainty because it never occurs; it is a hypothetical entity. Expert economists play a crucial role in the disputes over economic damages because of their ability to model and estimate what would have happened in a particular economic environment if one aspect of that environment had been different. What would profits have been if a contract or a patent had not been violated? What would fish harvests have been if not for an oil-spill? What would prices have been absent a price-fixing conspiracy?

This exercise often involves the comparison of relevant economic values across two distinct time periods, the damage period and the benchmark or clean period. And the basic exercise, describing and analyzing a but-for world, often involves time-series data-empirical observations that are generated at regular or irregular intervals over time. For example, in a price-fixing conspiracy, regression analysis is typically employed to estimate the historical relationship between actual transaction prices and various explanatory variables, such as production costs and the level of demand for the product. That econometrically derived relationship is used to generate but-for prices within the damage period.

Here I examine the properties of two alternative approaches to implementing time-series damages models. The two approaches will be referred to as “predictive” and “dummy-variable.”The comparison between the two approaches also serves as a reminder of the potential problem to which time-series models are susceptible: the crucial but sometimes forgotten “spurious regression” problem.