The impact of passenger mix on reported "hub premiums" in the U.S. airline industry

Southern Economic Journal, Oct, 2005 by Darin Lee, Maria Jose Luengo-Prado

4. Estimation of Hub Premiums

Our approach to assess the effect of hubs on airline prices is to estimate a price equation. In this equation, an appropriate measure of price for a given market and airline is regressed on hub indicators, measures of competition and control variables that may influence the cost and demand characteristics in that market. Thus, the price equation can be thought of as a reduced form specification in which demand and supply characteristics of the market are included as explanatory variables.

In general, the approach used in the literature to estimate airline price equations has been to pool the data of different carriers together and use firm dummy variables to control for firm-specific effects such as differences in costs or quality of service. This approach assumes, however, that the effects of the various right-hand-side variables on prices are constant across all carriers. There may be reasons to expect, however, that specific competitive factors affect carriers in different ways. For example, the degree of head-to-head competition from low-cost carriers varies substantially across the different network carriers, as has the way in which the network carriers have responded to them. Likewise, as discussed earlier, we suspect that there may be differences in how different carriers report unrestricted coach passengers to the Department of Transportation. (21) A Chow test on our data soundly rejects pooling, and thus, we begin by estimating ordinary least-square (OLS) and two-stage least squares (2SLS) models for each of the six large network airlines individually. In addition to the individual airline regressions, we also estimate pooled OLS, 2SLS, and fixed-effects models for the largest 1,000 markets, so that our results may be compared more directly to those in the literature.

Individual Carrier Regressions

We begin by estimating the following equation for each of our six carriers:

(1) ln([P.sub.j]) = [mu] [hub.sub.j][beta] [X.sub.j][delta] [[epsilon].sub.j]

where ln([P.sub.j]) is the natural log of the carrier's average price per mile in market j, p. is a constant, [hub.sub.j] is a matrix of hub dummies, [X.sub.j] is a matrix of controls, and [[epsilon].sub.j] is a random error term assumed to be i.i.d. with mean zero and variance [[sigma].sup.2.sub.[epsilon]]. We want to determine if airlines systematically charge higher prices per mile to passengers originating or terminating at one of their hubs--beyond any potential pricing power conferred by market share--so we include hub dummies in our specification. Specifically, we include two hub dummies: primary hub and secondary hub. Primary hub equals 1 if either endpoint of market j is a primary hub (as enumerated in Table 1) for that airline and 0 otherwise. Likewise, secondary hub equals 1 if either endpoint of market j is a secondary hub for that airline and 0 otherwise. Our approach differs from Borenstein (1989) and Evans and Kessides (1993) in that we control explicitly for the presence of hubs as opposed to using airport market shares. (22) We believe that this approach may be more revealing than the ones used in previous studies, given the high correlation between market share and airport market share in the data. (23)