Econometric studies of firm-level behavior are gaining acceptance among some industrial organization economists. This is a potentially useful tool for detecting noncompetitive behavior. Policy makers and antitrust enforcement officials are interested in the results of these studies as they are applied to specific industries to help enforce current antitrust regulations and develop new policies.

These New Empirical Industrial Organization (NEIO) econometric behavioral studies typically require detailed price, quantity, and cost data regarding the industry being studied. The models used are derived from the profit maximization problem of individual firms. In spite of this fact, many previous studies have relied on publicly available industry aggregate data, often also aggregated over time to the quarterly or yearly-level. This study investigates the sensitivity of empirical estimates of market power obtained from econometric conjectural variations studies to the level data aggregation used for the analysis. In addition, the sensitivity of the results to model specification is also explored.

The focus of this study is on measurement of oligopsony power in the U. S. beef packing/processing industry. Using Monte Carlo techniques, weekly plant or firm-level data are simulated to be representative of the U. S. beef packing industry in two broadly defined geographical procurement regions. To broaden the scope of the experiment, the assumed underlying technology of the beef packing industry is varied across a broad range of possibilities. In addition, alternative assumptions regarding the conduct of industry participants in the live cattle procurement market are imposed on the data generation process. The disaggregate data sets are aggregated over plants and firms to weekly industry aggregates, and over time to quarterly industry aggregates. At each level of aggregation, the data are tested using 3 alternative specifications of an NEIO econometric market power testing model, that differ by functional form. Results of the tests are compared across aggregation levels, and across model specifications.

The results reveal that in general the actual size of the test of the null hypothesis of no market power is much higher than the chosen nominal size of the test. The power of the test for market power is quite high. Data aggregation tends to bias the results of tests for market power. In addition, an adequately flexible functional form must be specified to capture the underlying technology of the industry when using econometric methods to test for market power. Therefore, in order to be useful for antitrust policy enforcement, econometric behavioral studies must make use of detailed firm (or plant )-level disaggregate data, and must use carefully specified models.