Demand forecasting and measuring forecast accuracy in general fare structures

Abstract

We present a forecasting methodology that enables forecasting in general fare structures (or, equivalently, demand models), including restricted, semi-restricted, unrestricted and fare families. Our contributions are twofold. First, we re-establish the measurement of forecast accuracy to be applicable to dependent demand models, by introducing a constrained forecast accuracy measure. Second, we provide a parametric forecast model for a general demand model, and we apply the forecast accuracy measure to compute optimal forecast parameters by minimizing the retrospective forecasting error on historical observations. Most of the parameters of our forecast model can be determined analytically in closed form based on historical observations. Appropriate aggregation levels and parsimonious use of parameters ensure robustness and resolve data-sparseness problems. We prove the performance of our approach using real airline data for a fenceless fare structure. Further, we use the Passenger Origin–Destination Simulator to investigate the impact of errors in the forecast parameters on revenue, load factor and forecast accuracy.