Data Driven Choice Model Estimation and Optimization in Revenue Management

Weikun Xu

This dissertation develops data-driven methodologies for estimating and optimizing discrete choice models in revenue management settings. The work addresses three interconnected challenges: estimating choice model parameters from censored transactional data, extending the estimation framework to queueing systems with unobservable balking, and optimizing assortment decisions using neural network-based choice models. Together, these three studies advance both the theoretical foundations and practical applicability of choice-based revenue management. The first study aims to jointly estimate the arrival rate of customers to a market and the nested logit model that forecasts hierarchical customer choices from an assortment of products. The estimation is based on censored transactional data, where lost sales are not recorded. The goal is to determine the arrival rate, customer taste coefficients, and nest dissimilarity parameters that maximize the likelihood of the observed data. The problem is formulated as a maximum likelihood estimation model that addresses two prevailing challenges in the existing literature: estimating demand from data with unobservable lost sales and capturing customer taste heterogeneity arising from hierarchical choices. We characterize conditions under which the model parameters are identifiable, revealing that parameter identification is influenced by the diversity of products and nests. We develop a sequential minorization-maximization algorithm to solve the problem, by which the problem boils down to solving a series of convex optimization models with simple structures, and we show the convergence of the algorithm by leveraging the structural properties of these models. We evaluate the performance of the algorithm by comparing it with widely used benchmarks, using both synthetic and real data. Our findings show that the algorithm consistently outperforms the benchmarks in maximizing in-sample likelihood and ranks among the top two in out-of-sample prediction accuracy. The second study extends the estimation framework to a maritime terminal setting. Forecasting tramp vessel arrivals and the resulting queue length at a terminal is critical for effective terminal planning. Unlike liner shipping, which follows fixed schedules, tramp shipping relies on flexible, market-driven vessel deployments, where charterers may proceed to a terminal when expected waiting times are acceptable but divert to alternative terminals when congestion becomes severe. This behavior induces censoring in transaction data: the operator observes only those vessels that actually enter the terminal and are served, while vessels that turn away are unrecorded. We address this challenge by developing a likelihood-based framework that combines an M/M/1 queue with a multinomial logit model of charterer decisions, enabling joint estimation of the vessel arrival rate, service rates, and the taste coefficients governing charterers' balking decisions. We establish conditions for global identification of the model parameters and develop a sequential minorization-maximization algorithm tailored to the queueing structure, with convergence guarantees that account for the stability constraints of the system. The estimated parameters enable managers to quantify latent demand lost to competitors and evaluate price-congestion trade-offs. We validate the approach using both synthetic data and real Automatic Identification System (AIS) data from Port Everglades and the Port of Vancouver. The third study addresses the assortment optimization problem for checkout product recommendations, where a retailer dynamically selects a contingent assortment of recommended products conditional on the customer's primary shopping cart. We propose a neural network-based basket choice model that captures set-dependent utilities, including cross-selling complementarities and substitution effects among products. We prove that the proposed architecture is a universal approximator of basket choice probabilities and formulate the recommendation problem as a constrained optimization model that maximizes total expected revenue across both primary and recommended sales. Numerical experiments demonstrate the effectiveness of the proposed approach relative to benchmark methods.

Data Driven Choice Model Estimation and Optimization in Revenue Management

Files and links (1)

Abstract

Metrics

Details