Stochastic bilevel models for revenue management in the hotel industry

In this thesis, we develop and solve a stochastic bilevel model for the hotel industry, which is nowadays considered as a mature industry marked by an intense competition and by a complex inventory management. We noticed that over the last 30 years, Hotel Revenue Management research has not proposed and solved models that consider simultaneously inventory assignments, price, length of stay, quality of service and uncertainty. Therefore, the purpose of this doctoral research is to develop a new model for Hotel Revenue Management that is inspired from bilevel pricing models and from the Two-stage Stochastic Models and that allows hotel's managers to account with useful data for pricing decision and assignment allocation, based on a better understanding of consumers' behavior and market uncertainty.;In order to introduce uncertainty information, we have developed a two-stage model: in the first stage the leader set its prices with the goal of maximizing profits in the upper level, and each users' group chooses the least expensive inventory considering the attributes previously defined by them (distance and quality of service), at the lower level. In the second stage, we introduce uncertain information about competitors' prices and demand, and thus the leader must set again its prices and inventory allocations, which also implies changes in users' group distributions. The stages are tied by price variation in each inventory through an absolute and proportional constraint.;We consider that uncertainty can be modeled with the support of random vectors that follow a known distribution function. This information might come from historical data or from the empirical knowledge of the distribution function, and that is close to the true unknown uncertainty. We assume that the random vectors have a finite number of realizations, which in our case corresponds to the scenarios.;In order to solve our model, we developed not only exact strategies but also heuristics. The exact strategy consisted in transforming the basic problem into a Mixed Integer Program problem using the Karush-Kuhn-Tucker Conditions conditions , through the use of big constants and auxiliary binary variables. The main achievement in terms of heuristics is the development of our greedy heuristic, which was able to solve the problem efficiently. This heuristic consisted in copying competitors' prices and re-optimizing in favor of the leader. To keep a global search, the exploration process was followed by a Mixed Integer Program restricted problem that took as origin the solution provided by our heuristic. Finally, the exact strategy supported by heuristics consisted in adding to the Mixed Integer Program original problem a heuristic that looks for integer solutions directly in the branch and bound (B&B) tree.;Once the model and the heuristics were developed, a data generation process was designed. The procedure sought not only to generate realistic instances for the industry but also to avoid unfeasible situations. To do this, we modeled price and demand fluctuations through the use of uniform random variables and we developed an analytical process that allowed us to disregard quickly atypical situations. The numerical results are presented for the two previous strategies, being the most performing the one based on our heuristic complemented with the Mixed Integer Program restricted problem. Moreover, the obtained results performed as expected in terms of its economic behavior. Depending on having or not a competitive advantage with respect to the location of its hotels, the leader has a more or less predatory behavior with its competition. In a situation under a competitive advantage, the leader seeks to imitate the price of its competitors in order to attract users' groups that provide the highest revenue. If the leader is not in an advantageous position, it set lower prices than the competition to compensate users' groups more sensible to distance. At the same time, it set competitive prices to attract users' groups that are more sensitive to quality of service than to distance, which implies that the leader reallocates its inventories and disregards users' groups providing lower revenues.;First, we introduced stochasticity on price and demand simultaneously and then, we added more complexity by varying the capacity of the industry. The heuristic was able to obtain a result, which was again behaving economically as expected.;Therefore, the main contributions of this research are to provide a elaborated model for Hotel Revenue Management, to solve small and large instances in a reasonable computing time, to obtain good results through the use of our heuristic (although we cannot assure it is the optimal solution), and to provide very useful results such as: pricing information, users group distribution in inventories, users group revenue contributions, sensitivity to capacity parameters, for decision making in the hotel industry. (Abstract shortened by UMI.).