Choice-based demand forecasting in airline revenue management systems

Jue Wang

Research output: ThesisDoctoral Thesis

644 Downloads (Pure)


[Truncated] Over the last decade, airline markets around the world have been reshaped dramatically by the rapidly growing low-cost carriers and new forms of distribution channel. Significant reduction in searching cost brought by the web-based distribution has made fare product comparison and purchasing an easier task. As a result, traditional demand models based on independent (fare class) demand assumption has been violated. A better understanding of passenger choice behaviour is now needed since the development of new generation revenue management (RM) system requires inputs of demand based on dependent fare classes.

Early studies on dependent demand mainly focused on the buy-up and buy-down behaviour for single-leg flights. With the introduction of discrete choice modelling, more recent studies are beginning to incorporate competitions between flights and carriers into the model. In a discrete choice model, a customer is assumed to weigh up service levels of a fare product against its price. The fare option with the highest satisfaction is the one being chosen. As all the components taken into consideration by a traveller may not be readily at hand for the analyst, the satisfaction or utility of a fare product is measured by way of a systematic component – the observed utility – and a random component – the unobserved utility. As such, the choice decision is modelled up to a probability. Discrete choice models are theoretically sound for fare product demand forecasting, as they directly work on the decision making process of air travellers.

Currently, the most widely applied discrete choice model in revenue management is the multinomial logit model (MNL), within which the unobserved utility of each alternative is independently and identically distributed (IID). Such a structure leads to the independence from irrelevant alternatives or IIA property. That is, the ratio of probabilities for two alternatives is independent from the existence of any other alternative in the choice set. However, the biggest limitation of IIA is the resulting proportional substitution pattern, which suggests that an improvement in the attributes of one alternative reduces the probabilities for all other alternatives by the same percentage. This highly restricted structure is unlikely to hold in the context of real airline markets. This is because the behaviour of compensatory travellers is likely to vary among the population, and to capture these variations advanced DCMs should be applied.

Original languageEnglish
QualificationDoctor of Philosophy
Publication statusUnpublished - 2015


Dive into the research topics of 'Choice-based demand forecasting in airline revenue management systems'. Together they form a unique fingerprint.

Cite this