Many industries use traditional revenue management (RM) systems to forecast demand for products and to determine product prices and availabilities. However, traditional RM systems were designed before the internet was a dominant distribution channel and during a period in which customers had limited knowledge of all of their product offerings. Choice-based RM systems use discrete choice models to forecast demand in a way that better reflects today's purchasing environment. Although much has been done on the optimization of choice-based models, little research has been undertaken on how to estimate the model using demand data from a single firm. This research focuses on estimation of parameters for choice-based RM systems using marginal versus expected log likelihood functions. The marginal formulation has a similar mathematical structure as limited information maximum likelihood (LIML) estimators, which have previously been developed in the transportation planning field but have not yet been applied to RM. Sequential estimation techniques such as LIML show strong promise as a framework that enable researchers to solve open research questions related to the incorporation of more realistic product substitution patterns and competitors product information. The primary research goals are to: (1) integrate nested logit and choice-based RM models and incorporate competitive information; (2) understand data requirements for choice-based RM models; and, (3) validate these new models on industry datasets.
If successful, the results of this research will help shape the vision for the next-generation RM systems by advancing theories needed to estimate RM choice parameters. Through testing and validating models on industry datasets, this research will be able to develop recommendations on data requirements and relevant industry settings needed to successfully estimate choice-based RM parameters. The results of this research will lead to increased profits for firms and better product and service offerings for customers. Previous research has suggested incremental revenue gains ranging from 1-10 percent may be possible with choice-based RM systems; thus, potential impacts of this research are substantial. The results of this research will also be applicable to other disciplines that face censored data, e.g., when retail sales are censored due to stock-outs.
