Dynamic Airline Booking Forecasting

This paper proposes a model for dynamic booking forecasting using a time-inhomogeneous Markov process. The transition probabilities are estimated based on a combination of an empirical and a parametric distribution. This model is applied for flight booking forecasting, where flight forecasts are updated on a daily basis over a time horizon of up to 300 days before the day of departure. The distribution of flight bookings over this time horizon, as well as the expected average flight bookings are determined. Historical data of two years of flights is used in our numerical analysis. The performance of our model is compared with two classical forecasting methods: the additive pickup method and the historical average. We show that our proposed model is up to 8% more accurate than the two classical methods mentioned above. Moreover, by determining the distribution of the flight bookings over a horizon of 300 days before departure, we provide additional information about the uncertainty around the flight
bookings.