Researches On Feature-Based Probability Forecasts And Resource Allocation In Revenue Management

Revenue management (RM) is one of the most sucessful application areas ofoperations research, and it is also a hot spot of research on management science andoperations research. What is RM? RM is the process of understanding, anticipatingand reacting to consumer behavior and uncertain environment in order to maximizerevenues by allocating resources to different demands. As the rapid development ofeconomy globalization, and diversification of consumer behavior, supply chain forproducts is prolonged, and life cycle of products is shortened. In this environment, itis a critical problem faced bythe firms to study how to maximize the revenue withlimited resources. Revenue management is an effective pattern for addressing fiercemarket competition, and is a convenient way to allocate resources for managingdifferent demands. Theory and practice of revenue management has gained attentionrecently in many application areas, such as airlines, retailing, hotels, rental car, andenergy management.In this growing competitive environment, the firms face two importantchallenges: one is how to recognize rapidly changing uncertain environment; antoheris how to utilize limited resources to meet diversified market demands. The majorityof revenue management decision-making problem is stochastic optimization model,and the input of the model relates to the probability distribution of decision elements.This needs to forecast the overall distribution of the uncertain environment elements.The result of probability forecast is the the basis of solving the decision-makingmodel, and the quality of probability forecast has direct influence on the decisions,thus it is important to improve probability forecast, and reseach on probabilityforecast issues becomes a fundamental issue in revenue management. Thedecision-making problems in revenue management are essentially resourceallocation problems, in which the limited resources are usually allocated throughquantity or price decisions. Resource allocation decisions directly affect corporate'srevenues, and the improvement on resource allocation decisions is particularlyimportant for maximizing the revenues, therefore the study on resource allocation problems is a core issue in revenue management.The major difficulty of probability forecast in revenue management is, thatelements to be predicted show some specific features, such as intermittent,nonstationary, incomplete and asymmetry. These features make that most traditionalforecasting methods are generally difficult to be used or their forecast accuracy islow. These features generally relate to the the problem background of revenuemanagement. It is helpful to analyze the formation mechanism of these features forcapturing these features and improving the accuracy of probability forecasts. Most ofresource allocation problems in revenue management are large-scale nonlinearoptimization problems, which is not easily solved by applying the generaloptimization techniques. The resource allocation problems in revenue managementoften have some special features, such as concavity or convexity, separability of theobjective function, and so on. It can be used to develop solution methods for theresource allocation problems through analyzing these features and establishing therelationship between the optimal solution and these structural features. This thesisattempts to seek solution ideas and methods by taking full advantage of the problemstructural features for addressing several probability forecast problems and resourceallocation problem in RM, and some results are obtained.Main contributions of this thesis are summarized as follows:(1) Studies on several probability forecast problems. Demand and price are twomajor uncertain elements in revenue management. As the wider use of informationtechnology in revenue management, it is an urgent requirement to improve theaccuracy of probability forecast by exploring and using data features. For threerevenue management problem with typical features (intertimment demand, censoreddemand, and asymmetry price), we study probability forecast problems by extractingand modeling the inherent mechanism of these features. Intermittent demand can beattributed to its self-discipline and the impact of external factors, for example,intermittent sales data may come into being due to normal sales and large-scalepromotional activities; the intermittent spare parts demand is affected by the life ofspare parts and equipment maintenance operations. In airlines or retail industry,censored demand caused by shortages has smoothing demand distributions. The price change of stocks and bonds has some asymmetries due to the impact of riskpreferences of investors.This thesis studies intrinsic features of different RM problems, and investigatestheir formation mechanism. Based on these analyses, the ideas and methods, whichcan be used to capture these features, are investigated, and several feature-basedprobablity forecasting methods are proposed. By analyzing the impact of externalfactors on intermittent demand, we propose two probability forecasting methods tointegrate auto-correlated process and the relationship between explanatory variablesand the nonzero demand, for predicting intertimment demand with stationary andnonstationary nonzero demand. By analyzing the estimation bias of product limitestimator on special points, and using the smoothing demand feature, we developthree distribution completion methods to forecast demand distribution from censoredobservations. By analyzing and capturing the asymmetry features in economic timeseries, we propose three density forecasting methods under two-piece normaldistribution. Based on the studies of probability forecasting methods, we also studyhow to evalute probability forecast, and a general definition for reliability ofprobability forecasts and three reliability-oriented measures are proposed.Some of the above studies have been internationally recognized, and the relevantresults have been pubished on the following international journals: EuropeanJournal of Operational Research (2008, 185(2): 716-725), Journal of theOperational Research Society (2007, 58(1): 52-61), Applied Mathematics andComputation (2006, 181(2): 1035-1048), Expert Systems with Applications (2007,33(2): 434-440). Some researches have developed into application software systems,which achieved significant economic benefits in several enterprises.(2) Studies on several resource allocation problems. It has an important value tostudy the resource allocation problems in RM by analyzing and utilizing theirstructural features. Resource allocation problems in RM can be divided into twocategries: quantity-based RM and price-based RM. Most of quantity-based resourceallocation problems in revenue management are large-scale convex optimizationproblem. These problems can be divided from different perspectives into: continuousor integer variables, linear or nonlinear constraints, separable or nonseparable objective. Price-based RM studies how to adjust price for allocating resources, sodynamic pricing probem is another important problem in RM.This thesis studies several resource allocation problems with different structuralfeatures, and propose several feature-based solution methods for these problems, byanalyzing the relationship between the solution and the structural features. Forcontinuous convex seperable multi-product resource allocation problems with linearand nonlinear constraints, we propose several methods for solving their optimalsolution by utilizing their structural features. For large-scale NP-complete integralresource allocation problems, we propose a variable reduction technique to reducethe size of problem without sacrificing optimality. Based on the structural features,we also propose several efficient heuristics and approximate solution methods. Forthe case that the market size and customers' response to price are both unknown, andthere is no chance of inventory replenishment during the sales season, we study anactive demand learning pricing strategy for dynamic pricing problem, and compare itwith other pricing strategies.Part of these studies has been internationally recognized, and the relevant resultshave been pubished (or accepted for publication) on the following internationaljournals: European Journal of Operational Research, International Journal ofProduction Economics, Computers & Operations Research (33(3): 660-673),Applied Mathematical Modelling. The proposed methods can solve the optimal orappoximate solutions to the resource allocation problems quickly.Finally, in addtion to the above achivements in RM, the thesis discusses somechallenging topics which deserve further research in future.