A Data Mining Approach to Market Power Analysis

March 30, 2005

Research conducted by Nghia Tran under advisement by Dr. Sean Warnick, BYU Computer Science Department.

The question about market power is interesting both for business managers and market regulators. Market regulators are interested in maximizing total social welfare in a market by protecting market competition. Business managers want to know the power of different firms or parts of a firm to know their true value in acquisition. The measurement of market power turns out to computation-intensive and requires a very large amount of market data. The well-known measure, the Herfindahl-Hirshman Index, requires on large amount of scanner data to determine the market share of individual firms in markets. This measure also relies on legal definitions of particular markets, which can be very subjective.

We hypothesized that with less data, there are still indicators of market power that can be computed. The measure that we came up with concern the power of coordinating pricing of different products within a firm to produce the most profit. To compute this, we introduced the concept of a product network which contained all products within a firm and their relationships, or cross-price elasticity. We wanted to measure the power of a firm to use their ability of altering the price of different products to raise more profit than when the firm maximizes its profit on individual products.

To study about this ability, we need to use different demand models commonly found in economics literature like linear, log-linear, logit and AIDS (Almost Ideal Demand System) models to know how the market demand behave under price change. However, the linear model seems to fit well to our purpose and it is much simpler than other model. Based on the linear model and assume profit maximization behavior of firms, we come up with a linear dynamics model of how firms change their prices to meet market equilibrium.

With the linear dynamics model, we simulate the systems to know the behavior a hypothetical firm with simulated market data. We found that the profit given by a firm’s product network can be very high compared to that if the firm has no control over the product network but maximizes it profit as if the products are not related. Based on this observation, we define two different measures: 1) the value of cooperation, which is the difference in maximal profit if the firm’s products are related compared to that when the products are not related 2) the relative value of cooperation, which is the percentage of profit of a given firm from the relationships within its product network.

The reasons behind these measure is that the more product lines a firm owns, the more control such firm can have on consumer’s demand. By changing the price of one product, a firm can shift demand to one of its other product, and can earn more profit. If products within a firm have stronger relationships, such firm will have more power over consumer’s demand. One example we can see in the real world is the relationships between printers and print cartridges. A firm can choose to sell very cheap printers but since they have monopolistic power over their print cartridges (each cartridge type works only on certain brand of printers), they can raise the price of print cartridge to very high. The consumers, in the end, are paying more to the same productivity that they receive.

We presented our results in JCIS 2005 and CIEF 2005 and received useful feedbacks from both peer reviewers and also from people who attended our presentations. In our future work, we are building more refined models to have better analyses so thatsswe can study our measure and find other measures that can be indicators of market power. We also want to apply different concept in the study of control of dynamics system so apply on social systems like firms and retails.