Extracting airline and passenger behavior from online distribution channels: applications using online pricing and seat map data

Although the airline industry has drastically changed since its deregulation in 1978, publically available sources of data have remained nearly the same. In the U.S., most researchers and decision-makers rely on government data that contains highly aggregated price information (e.g., average quarterly prices). However, aggregate data can hide important market behavior. With the emergence of online distribution channels, there is a new opportunity to model air travel demand using detailed price information. This dissertation uses online prices and seat maps to build a dataset of daily prices and bookings at the flight-level. Several research contributions are made, all related to leveraging online data to better understand airline pricing and product strategies, and how these strategies impact customers, as well as the industry in general. One major contribution is the finding that the recent product debundling trend in the U.S. airline industry has diluted revenues to the U.S. Airport and Airways Trust Fund by at least five percent. Additionally, several new behavioral insights are found for one debundling trend that has been widely adopted by U.S. airlines: seat reservation fees. Customers are found to be between 2 and 3.3 times more likely to purchase premium coach seats (with extra legroom and early boarding privileges) when there are no regular coach window or aisle seats that can be reserved for free, suggesting that the ability of airlines to charge seat fees is strongly tied to load factors. Model results are used to explore optimal seat fees and find that an optimal static fee could increase revenues by 8 percent, whereas optimal dynamic fees could increase revenues by 10.2 percent. Another major contribution is in modeling daily bookings and estimating price elasticities using ordinary least squares (OLS) regression without correcting for price endogeneity and two-stage least squares (2SLS) regression, which corrects for endogeneity. Results highlight the importance of correcting for price endogeneity (which is not often done in air travel applications). In particular, models that do not correct for endogeneity find inelastic demand estimates whereas models that do correct for endogeneity find elastic demand estimates. This is important, as pricing recommendations differ for inelastic and elastic models. A set of instrumental variables are found to pass validity tests and can be used to correct for price endogeneity in future models of daily flight-level demand.Ph.D