Dynamic booking forecasting for airline revenue management

Abstract

Airline revenue management aims to sell the seats on their planes to passengers at a price that is as close as possible to their maximum willingness to pay for a seat. In practice, airlines try to achieve this by creating a fare structure for the seats on their planes after which the seats are grouped under these different fares. Revenue management practice for airlines therefore consists of a pricing part and a seat allocation part. The fares can be updated, however, the fare structure is considered static compared to the seat allocation (also called allocation) control that takes place more dynamically in revenue management systems. Seat allocation control can therefore be considered as the key process within revenue management of an airline. The booking forecasting model plays a central role in seat inventory control and therefore in the revenue management process. In theory, a 100% accurate forecast will result in a maximum revenue figure and inaccurate forecasts will result in a deviation from this maximum achievable revenue. Studies on the impact of forecasting accuracy on passenger revenues have demonstrated this by showing that a 10% improvement in forecasting accuracy can lead to a 0.5% to 3% increase in annual passenger revenue. For a major US airline with high demand flights this comes down to $10 to $60 million. The quality of a revenue management forecast is of outmost importance for an airline. However, forecasting for revenue management has proven to be a challenging task because of the dynamics and complexity of the bookings process which is influenced by many factors such as pricing discounts, special events or defections of passengers from delayed or cancelled flights. As a result, the prediction for the future demand expressed as a single value is mostly wrong. In order to have an accurate forecasting model it is therefore important not only to forecast future demand close to the actual realized values, but also to provide the revenue management controller with information on the degree of uncertainty that is involved in the prediction. From there the research objective is formulated as follows,