Use of Econometrics at the U.S. Department of Justice

• Pages: 131-165

131

CHAPTER VI

USE OF ECONOMETRICS

AT THE U.S. DEPARTMENT OF JUSTICE

A. Introduction

This chapter offers examples of the use of econometrics in the

Antitrust Division of the U.S. Department of Justice (the Antitrust

Division or the Division) in five specific matters. The specificity of the

discussion in this chapter varies according to the extent to which details

of these matters are in the public record. Several points are made. First,

although much discussion of the use of econometrics in antitrust analysis

has centered on the estimation of demand for purposes of market

definition, or for predicting the unilateral price effects of a merger,

econometrics has application to a much wider set of questions that come

before the Division. Second, the role of econometrics at the Division is

not limited to developing evidence that may be introduced in the

courtroom. Econometrics is also employed when deciding whether an

enforcement action is appropriate, and to test whether conclusions

suggested by other types of information, including documents,

testimony, and the opinions of interested parties and their advocates, are

supported by a careful analysis of the available data. It is often the case

that data that could help to resolve a key issue are not available, or that

the available data are imperfect or uninformative. But, if good data are

available, a serious analysis must consider them. A third goal of this

chapter is to provide perspectives on some issues that commonly arise in

the practice of econometrics in antitrust analysis and on the process by

which data and econometric studies are brought to bear on issues in an

investigation.

The chapter begins with a discussion of issues related to demand

estimation. These include the need to have data on factors that are

significant determinants of demand, issues relating to the choice of how

to model demand, and the importance that specifying the error structure

can have on coefficient estimates as well as estimates of standard errors

and other estimates of statistical precision. The discussion does not

address problems of endogeneity and aggregation, the treatment of

promotional activities, and the modeling of consumer dynamics.

132 Econometrics in Antitrust

Some of the general points about data requirements and choice of

functional form are illustrated in discussions of the analysis of the

L’Oreal/Maybelline cosmetics merger and the Philips/Agilent merger

involving medical diagnostic imaging equipment. In the case of the

cosmetics merger, apparent shortcomings in the availability of data

determined the initial model specification and limited the ability to draw

an inference about a key parameter to be estimated. Once the data were

supplemented, it was possible to estimate a more complete model and to

extend the interpretation of results to the parameter of interest. In the

Division’s analysis of the medical equipment merger, data limitations

again made it necessary to specify a relatively restrictive demand model.

It is important to note that the particular demand model chosen for the

estimation can restrict patterns of substitution across products. In this

case, however, there were sufficient data to permit exploration of

alternative specifications that imposed different restrictions, and in this

way it was possible to investigate a richer set of hypotheses about which

products were close substitutes.

A very different use of econometrics occurred during the

investigation of a proposed merger between two financial markets

trading listed options contracts. At issue was whether the competitive

position of the Philadelphia exchange had been in decline. The third

case study describes an econometric analysis of exchange seat prices

intended to shed light on this question. Since the goal was to shed light

on a factual question, an extremely simple but very flexible model was

estimated for use as a benchmark. Comparing actual performance to the

benchmark, it was possible to draw conclusions about changes in the

competitive position of the Philadelphia exchange relative to the market

as a whole. Many of the concerns of demand estimation did not arise in

this setting, because the intended use of the model was very limited.

The last two case studies relate to the provision of direct broadcast

satellite (DBS) services. The first describes an analysis of some work

conducted while investigating a proposed acquisition of high-power DBS

assets by Primestar, Inc. The analysis in this case highlights an emphasis

on careful data preparation. What may appear to be small differences in

data handling sometimes make a large difference in the results, and the

Primestar case provides an example. The case also provides a

framework for thinking about the role of ad hoc econometric models in

antitrust analysis. Indeed, it is sometimes difficult to draw conclusions

Use of Econometrics at the Department of Justice 133

from econometric models when they lack adequate foundations in the

economic theories under consideration.

The final case study returns to the theme of demand estimation, this

time in the context of a proposed merger between two providers of DBS

services—DirecTV and EchoStar. The analysis of the case makes three

points. First, work done in this case highlights the role of modeling

assumptions as a possible substitute for unavailable data—a theme also

addressed in the cosmetics merger. Second, consistent use of a model

can be useful for reaching consistent conclusions without any need to

appeal to separate ad hoc analyses. Third, there is a need for auxiliary

assumptions when calculating standard errors or otherwise assessing

statistical precision. This point is often overlooked in presentations to

the Division.

B. Evaluating Competitive Effects in Mergers: Choosing a Demand

Estimation Model Given Data Limitations

1. Data Requirements and Approaches when Data are Limited

Demand estimation can help answer questions about which products

are in a relevant market and whether particular products in the market are

relatively close substitutes. The closer products involved in a merger are

as substitutes, the greater the possibility that market shares will tend to

understate the extent to which competition may be lessened as a result of

the merger. Once estimates of parameters of the demand model have

been obtained, they can be run through a merger simulation to yield

predictions about the likely effect of a merger on price.1 Even relatively

flexible techniques of estimation sometimes make assumptions that

impose strong restrictions on demand. Such assumptions are usually

necessary to make progress. Although it is easy to criticize these

techniques, their reliability should be measured against the alternative of

nonsystematic analysis of data.

1. See, e.g., chapter XI of this Handbook; Appendix III of this Handbook;

Gregory J. Werden & Luke M. Froeb, Calibrated Economic Models Add

Focus, Accuracy, and Persuasiveness to Merger Analysis, in THE PROS

AND CONS OF MERGER CONTROL 63, 70–79 (Swedish Competition

Authority ed., 2002); Gregory J. Werden, Simulating the Effects of

Differentiated Products Mergers: A Practical Alternative to Structural

Merger Policy, 5 GEO. MASON L. REV. 363 (1997).