7th Annual INFORMS Pricing and Revenue Management Conference Universitat Pompeu Fabra, June 28-29 2007, Barcelona, Spain FINAL SCHEDULE TRACK A: ROOM 40.002 TRACK B: ROOM 40.010 TRACK C: ROOM 40.012 KEYNOTES: AUDITORI LUNCHES & BREAKS: PATIO GALA DINNER: HOTEL ARTS (SEE ATTACHED MAP) THE LAST SPEAKER IN EACH SESSION IS THE SESSION CO-ORDINATOR June 28 Day 1 Planning 9:00-9:30 Opening/Welcome Andreu Mas-Colell, Professor of Economics UPF 9:30-10:00 10:00-11:30 Break STRATEGIC ISSUES + APPLICATIONS DEMAND MODELS + CB+ FORECASTING PRICING & RM OPTIMIZATION Session A1 Competition in RM Session B1 Strategic Buyers: Models and Empirical Evidence Saving Seats: Reservation Policies with Strategic Consumers, Lariviere Session C1 Network RM 1 Consumer Behavior in Waiting for Lastminute Discounts, Ovchinnikov, Skurnik Venkatesan Milner Cargo Capacity Management on a Network, Levin Nediak McGill Profit Loss due to Price Competition in Markets with Differentiated Products, Farahat Perakis Learning to Price Airline Seats under Competition, Collins Thomas 1 Asymptotic Optimality of a Network RM Model Based on a Fixed Point Approximation, Yao Ye A Stochastic Multiple Leader Stackelberg Model, de Miguel Xu 11:4512:45 Large Scale Markdown Optimization for Retail, Gulcu Sanli Keynote Xavier Vives, Professor of Economics and Financial Management IESE 12:45-2:15 2:30-4:00 Lunch Session A2 Strategic Issues in Service RM Capacity and Value Based Pricing for Professional Services Including Bundled and Non-Bundled Engagements, Wardell, Wynter Helander Competition in Innovation and Pricing for Short Life Cycle Product, Arslan Kachani Shmatov Competition in Prices & Service Level Guarantees, Johari Weintraub Session B2 Consumer Behavior in RM: Models and Empirical Evidence A Quantification of the Internet Price Effect, Brunger Price Variation Antagonism and Firm Pricing Policies Pagliero Courty Revenue Management for Fans, Popescu Rudi 4:00- 4:30 4:30-6:00 RLP Revisited, Talluri Session C2 RM under Customer Choice Top-Down vs. Bottom up Approaches to Behavioral Network RM, Savard Marcotte Bigras Cote Rancourt Schoeb RM under Choice Models, Gallego Li Ratliff A Column generation Algorithm for choice Based RM, Vulcano Bront Diaz Break Session A3 The Value and Future of Pricing & RM Session B3 RM with Endogenous Costs 2 Session C3 Network RM 2 On the Value of Dynamic Pricing, Lobo RM in e-fulfillment, Agatz Perspectives on the Future of Pricing, Boyd Integrated Pricing & Inventory Control with Reference Effects, Gimpl-Heersink Rudloff Taudes Dynamic Pricing of Inventories with Menu Costs, Feldmann 32 Capacity Reservation in a Make-to-Order System with Uncertain Due-Date Preferences, Yadav Pibernik 6:05-7:00 Capacity-Dependent Bid-Prices in Network RM, Topaloglu Kunnumkal Network RM using Outer Aproximation Algorithms Philpott Van Ryzin Keynote Bruno Matheu, Air France Cocktail Dinner Hotel Arts 7:30-… 29 June 2007 9:00-10:30 Session A4 Supply Chain Issues and Contracts Competing for Shelf Space, Martinez de Albeniz Roels Competitive Supply Chain Contracting and Tour Revenue Management, Anderson Markus Capacity Contracts for Cargo Carriers and Forwarders, Cooper Amaruchkul Gupta Session B4 RM with Consumer Behavior Joint Memory Dependent Pricing and product introduction for multiple generations Kachani Arslan Shmatov Optimal Pricing of Services with Switching Costs, Liu van Ryzin Consumer Search Behavior when Product Quality is Uncertain, van Ryzin Debo 3 Session C4 Robust RM 2 Dynamic Revenue Management Policies with Limited Demand Information Ball Gao Karaesmen Robust Newsvendor Competition, Jiang Netessine Savin 10:30-11:00 11:00-12:30 Break Session A5 Cooperation & Competition in RM Session B5 Demand Models and Forecasts A Comparison of Static vs. Dynamic Schemes for Alliance RM, Shumsky Wright Groenvelt Retracing Choice Process Among SNCF Clients, Wiesel Riss Brotcorne Savard The Strategic Role of Capacity in a Joint Inventory Management and Pricing Game, Adida Perakis A Pricing DP game, Walczak Forecasting Models and Scenario Analysis, Castejon Leadtime-Variety Tradeoff in Product Differentiation, Alptekinoglu Corbett Demand Modeling: Insights Based on an Empirical Study, Zeevi Besbes Phillips Capacity Investment and Pricing with Multiple Products and Production Priorities, Transchel Minner Pike 12:30-2:00 2:00-3:30 Session C5 Pricing & Inventory Decisions for Differentiated Products A Stochastic Dynamic Programming Approach to RM in a Make-toStock Production System, Quante Meyr Fleischmann Lunch Session A6 RM in Service Applications RM in Facilities where Customers Wait for Service, Giloni Troy RM for Online Advertising, Fridgeirsdottir Araman Session B6 Overbooking Data Mining Based No Show and Cancellation, Wang Romero Value of Traffic Mixbased Overbooking J. Lancaster Munich Graff 4 Session C6 RM with Limited Information Sequential Prediction Under Imperfect Monitoring, Lugosi Dynamic Pricing with Learning, Kleywegt Access Pricing Services with Positive Externalities, Kumar Johari Robust Revenue Management: Overbooking and FareClass Allocation Karaesmen Ball Lan 3:30-4:00 4:00-5:00 Break Session A7 The Right Approach to RM Risk-Sensitive Capacity Control, Barz Is RM Accessible to the Small Hotelier? Colomines William Session B7 Healthcare Pricing-Capacity Tradeoffs in Rehab Healthcare Networks: Models and Open Problems, Millhiser The Performance of Nonlinear Pricing Schemes in Medical Service in Korea, Kwak Paik Nam Ha 5 Dynamic Pricing Using Adjustable Optimization, Lobel Perakis Session A1: 28 June 10:00-11:30 Room 40.002 Session Title: Competition in RM 6 Learning to Price Airline Seats under Competition A.J.Collins, Lyn Thomas University of Southampton Applied Game Theory is criticised for not being able to model real decision making situations. A game’s sensitive nature and the difficultly in determining the utility payoff functions make it hard for a decision maker to rely upon any game theoretic results. Therefore the models tend to be simple due to the complexity of solving them (i.e. finding the equilibrium). In recent years, due to the increases of computing power, different computer modelling techniques have been applied in Game Theory. A major example is Artificial Intelligence methods e.g. Genetic Algorithms, Neural Networks and Reinforcement Learning (RL). These techniques allow the modeller in incorporate Game Theory within their models (or simulation) without necessary knowing the optimal solution. After a warm up period of repeated episodes is run, the model “learns” to play the game well (though not necessary optimally). This is a form of simulation-optimization. The objective of the research is to investigate the practical usage of RL within a simple sequential stochastic airline seat pricing game. Different forms of RL are considered and compared to the optimal policy, which is found using standard dynamic programming techniques. The airline game and RL methods displays various interesting phenomena, which are also discussed. For completeness, convergence proofs for the RL algorithms are being constructed. Session A1: 28 June 10:00-11:30 Room 40.002 7 Session A1: 28 June 10:00-11:30 Room 40.002 8 Session B1: 28 June 10:00-11:30 Room 40.010 Session Title: Strategic Buyers: Models and Empirical Evidence Saving Seats: Reservation Policies with Strategic Customers Martin A. Lariviere, Northwestern University We consider a service provider such as a restaurant that offers reservations. The firm serves two customer segments. The size of both segments is random. The first market segment demands reservations. These customers contact the service provider before the date that service is required and will only patronize the firm if they are given a reservation that guarantees access to the provider’s limited capacity.The second segment is made up of walk-in customers who can only access the firm at the time that service is required. Walk-in customers face a fixed cost in reaching the service provider and cannot determine whether capacity is available before incurring the fixed cost. Consequently, whether a walk-in customer actually attempts to patronize the firm depends on the amount of capacity she expects to be free. Hence, the number of reservations made available to the first segment determines the amount of demand from late arriving walk-in customers. We examine this trade off and show that the firm may find it optimal to commit to turning away early arriving customers even when these customers offer a higher margin. 9 Session B1: 28 June 10:00-11:30 Room 40.010 10 Large Scale Markdown Optimization for Retail Altan Gulcu, Tugrul Sanli altan.gulcu@sas.com 100 SAS Campus Drive, Cary, NC, 27513 USA Tight profit margins and intense competitive pressures require that retailers maximize the returns from their end-of-season and end-of-life merchandise. Taking markdowns too often, at the wrong time, on the wrong items or at the wrong location can be costly. Moreover, such markdowns are often priced too low, resulting in lower margin; or too high, resulting in lost revenues, leftover inventory and higher labor costs. While retailers often view markdowns as a necessary activity to clear merchandise, when executed properly, they can deliver significant lifts in demand and profitability. SAS Markdown Optimization provides retailers the ability to determine which items should be marked down, by how much they should be marked down, when, and in which markets, stores or store clusters. Retailers of all kinds—apparel, mass, specialty, hard goods, food, drug—can base this analysis on critical factors such as inventory levels, base sales volume, price elasticity and local demand or preferences. It helps retailers develop and implement optimal pricing strategies at the local market level to maximize revenues and profitability while meeting end-of-season inventory goals. We model the markdown optimization problem with multiple objectives, incorporating business rules and solve it via dynamic programming and a customized search algorithm. The model allows the retailer to specify and force uniform pricing rules at store clusters. Our presentation will focus on the modeling and formulation of the business problem, input required for and the output generated by the model and solution methodologies used. Session B1: 28 June 10:00-11:30 Room 40.010 11 Session C1: 28 June 10:00-11:30 Room 40.012 Session Title: Network RM 1 12 Capacity Management on a Network Yuri Levin, Tatsiana Levina, Jeff McGill, Mikhail Nediak Queen’s University We consider a cargo shipping problem on a network of locations connected by flights operated on a fixed periodic schedule. Bookings for cargo shipping between origindestination pairs are made in advance. However, cargo capacity availability as well its utilization by each package are unknown at the time of booking. The problem is to maximize the expected present value of profit by controlling accept/reject decisions for booking requests and dispatching accepted packages through the network. We discuss the structural properties of the optimal policy and numerical approximation schemes. RLP Revisited Kalyan Talluri ICREA & Universitat Pompeu Fabra Empirically, tighter upper bounds to the dynamic programming value function seem to lead to better bid-prices – i.e., better revenues. So obtaining tighter upper bounds is a worthwhile research goal. The Randomized Linear Programming method for Network RM is a very simple and effective method that is based on the Perfect Hindsight (PHS) upper bound obtained from simulating the forecasts. In this work we related the PHS upper bound to recent work by Adelman and the Lagrangian bound of Topaloglu. We also extend our work to choice network RM. Session C1: 28 June 10:00-11:30 Room 40.012 13 Session A2: 28 June 2:30-4:00 Room 40.002 Session Title: Strategic Issues in Service RM 14 COMPETITION IN INNOVATION AND PRICING FOR SHORT LIFE-CYCLE PRODUCTS Hasan Arslan* Soulaymane Kachani Kyrylo Shmatov Fast-changing consumer preferences and high pace of product and process innovation together with intense competition force companies to introduce new products quickly at ever-increasing rates. This leads to proliferation of products with short life-cycles in the apparel and high-tech (consumer electronics, PC, software) industries. Firms in these industries become more vulnerable to competition as product life-cycles decrease. Introducing the right product at the right time and applying the right pricing mechanism are extremely challenging tasks for these companies. While some firms focus on catering to consumer tastes by quickly substituting new products and phasing out the old products, others focus on capturing consumer surplus through frequent price discounts. Firms that have quick turnaround on products are likely to attract more innovation- or fashion- sensitive consumers, whereas firms that execute price discounts over products' life-cycles tend to attract more price-sensitive consumers. In this research, we consider the competition among firms that produce and sell short life-cycle products to consumers who are segmented in terms of price sensitivity and their purchasing behavior towards new and innovative products. Our objective is to derive insights on how pricing and product innovation strategies impact these firms' market positioning. We assume that the market demand for a newly introduced short life-cycle product responds to two main factors: deviation of the product's own price from a market-wide reference price, which is essentially an exponentially-weighted moving average of past prices, and deviation of the product's innovation (fashion) level from the current market average. To the best of our knowledge, our research is the first to incorporate the joint memory effects both in pricing and the product innovation level. In terms of market positioning, we consider three types of firm strategies: (i) price discounts and promotional sales, (ii) high product introduction rate to gain a competitive edge by developing new products and restocking faster than others, and (iii) a hybrid strategy that embeds both approaches. We study the competition among firms under a variety of market conditions through a differential game. This differential game-theoretic model accounts for different types of consumer segments and different types of pricing and product innovation strategies of competing firms. Firms producing short life-cycle products may not have enough time to update their measures of market reference price and reference product innovation level. Therefore, we argue that open-loop strategies in practice fit well to such firms. We first derive open-loop Nash equilibria for an oligopoly game and show that the solution is a linear combination of exponentials. We then extend our analysis to the corresponding closed-loop Nash equilibria. We expected that open-loop and closed-loop Nash equilibrium solutions would not differ much. We confirm this intuition and quantify, both analytically and numerically, how the difference between the two equilibria depends on the demand sensitivity parameters, and how this difference is small for reasonably low values of these parameters. * Sawyer Business School, Suffolk University, Boston, MA, harslan@suffolk.edu IEOR Department, Columbia University, New York, NY, kachani@ieor.columbia.edu APAM Department, Columbia University, New York, NY, kis2101@columbia.edu 15 We focus on characterization of distinct classes of firms in the market in terms of their pricing and product innovation strategies. Specifically, we consider two main classes: discounter firms that control their pricing policy while keeping their product innovation level constant (no improvement of offered products and no new generation development), and high product introduction rate firms that assume a non-trivial product innovation policy while executing a constant pricing policy. We analyze specific competition among different firms that adhere to different types of pricing and product innovation strategies, and derive the corresponding Nash equilibria in these settings. To derive insights on competition among different market positioning strategies, we conduct a real-life case study to compare actual profit figures, achievable under different competition scenarios in different types of market, against the historically realized figures. This shows the level of improvement that may be achieved if more informed pricing and product innovation strategies are adopted. The case study is based on the example of competing retailers in the apparel industry - one that gains a competitive edge through high product introduction rate (e.g., Inditex) and its competitor (e.g., H&M), which competes largely through promoting price discounts. Session A2: 28 June 2:30-4:00 Room 40.002 16 Session A2: 28 June 2:30-4:00 Room 40.002 17 Session B2: 28 June 2:30-4:00 Room 40.010 Session Title: Consumer Behaviour in RM: Models and Empirical Evidence A Quantification of the Internet Price Effect Bill Brunger, Case Western Customers who use Internet/Online Travel Agencies to purchase “clearly leisure” trips (i.e. trips which were booked at least 14-days in advance, where the passenger stayed over the weekend) pay significantly less (11.5% in our sample, displayed below) for similar itineraries in the same markets than those who purchase through traditional travel agencies even though the fares and inventory offered by the airline are identical because of contractual obligations and strategic policy. The purpose of this study is to examine this Internet Price Effect, explanations for which may lie in secondary market segmentation, differences in characteristics of trips chosen, and search efficiencies. A Comparison of Average Fare Paid for Comparable Itineraries by Leisure Passenger (Airline proprietary data) Fare Paid for "clearly LEISURE" Itinerary (Net of all agency fees charged to the customer) February '06 / EWR-RDU EWR-PHX EWR-LAX EWR-ORL IAH-SEA IAH-ORD IAH-LGA CLE-SFO Traditional TA CLE-LAS EWR-RDU EWR-PHX EWR-LAX EWR-ORL IAH-SEA IAH-ORD IAH-LGA CLE-SFO CLE-LAS Internet Agency June '06 Specifically, relying on the Literatures of Customer Behavior/Internet Search and the techniques and precedents of previous price and price dispersion studies, this study will use detailed passenger-level trip data (previously unreported in the Literature) to examine differences across the distribution channels, while controlling for customer, market and flight characteristics and seat value, in order to explore and attempt to quantify the residual impact of Internet search. At the individual level, the Regression Equation should be of the form: FP= ß0 + ß1*DC + ß2*TC+ ß3*TD + ß4*MS + ß5*OpV + ε Where FP is the Fare Paid (as a percent of the mean fare paid for a similar trip in the same market), DC is Distribution Channel, TC, TD and MS are matrices of variables related to Trip Characteristics, Traveler Demographics and Market Structure, OpV is the Opportunity Value of the seat used and ε is an error term. 18 Reserve University will present his research design and preliminary findings, and solicit thoughts on how to proceed with the study. Session B2: 28 June 2:30-4:00 Room 40.010 19 Session B2: 28 June 2:30-4:00 Room 40.010 20 Session C2: 28 June 2:30-4:00 Room 40.012 Session Title: RM under customer choice 21 Revenue Management under Choice Models Guillermo Gallego, Lin Li and Richard Ratliff New revenue management heuristics are described to control capacity availability for the multi-fare, single-leg problem under time dependent choice models. The model is rich enough to allow for downsell, upsell, competition effects and market segments with different sensitivities to price and fare attributes. Both dynamic and static heuristics are described. The heuristics are computationally as simple as applying EMSR-b with buy up, but perform nearly as well as solving the full dynamic programming problem A column generation algorithm for choice-based network revenue management Gustavo Vulcano (New York University, USA) Juan Jose Miranda Bront and Isabel Mendez Diaz (University of Buenos Aires, Argentina) In the last few years, there has been a trend to enrich traditional revenue management models built upon the independent demand paradigm by accounting for customer choice behavior. This extension involves both modeling and computational challenges. One way to describe choice behavior is to assume that each customer belongs to a segment, which is characterized by a consideration set, i.e., a subset of the products provided bythe firm that a customer views as options. Customers choose a particular product according to a multinomial-logit criterion, a model widely used in the marketing literature. In this paper, we consider the choice-based, deterministic, linear programming model (CDLP) of Gallego et al (2004), and the follow-up dynamic programming (DP) decomposition heuristic of van Ryzin and Liu (2004),and focus on the more general version of these models, where customers belong to overlapping segments. To solve the CDLP for real-size networks, we need to develop a column generation algorithm. We prove that the associated column generation subproblemis indeed NP-Complete, and propose a simple, greedy heuristic to overcome the complexity of an exact algorithm. Our computational results show that the heuristic is quite effective, and that the overall approach has good practical potential and leads to high quality solutions. Session C2: 28 June 2:30-4:00 Room 40.012 22 Session A3: 28 June 4:30-6:00 Room 40.002 Session Title: Value and Future of Pricing and RM 23 Perspectives on the Future of Pricing Andy Boyd, PROS Every company that exists sells something, and in doing so sets a price for what it sells. This gives rise to a variety of ways to “price” and “sell,” many of which have received little research attention. We begin by taking a look at demand curves (price response functions) and their practical and theoretical limitations. We then survey some business models where the use of demand curves is inappropriate. Special attention is given to actual methods of pricing that can benefit from the application of mathematical modeling. Capacity Reservation Levels in a Make-To-Order System with Uncertain DueDate Preferences Richard Pibernik and Prashant Yadav MIT-Zaragoza International Logistics Program We consider a Make-to-Order manufacturer facing random demand from two classes of customers. We develop an integrated model for reserving capacity in anticipation of future order arrivals from high priority customers and setting due dates for incoming orders. Our research exhibits two distinct features: (1) we explicitly model the manufacturer’s uncertainty about the customers’ due date preferences for future orders and (2) we utilize a service level measure for reserving capacity rather than estimating short and long term implications of due date quoting with a penalty cost function. We identify an interesting effect (“t-pooling”) that arises when the (partial) knowledge of customer due date preferences is utilized in making capacity reservation and order allocation decisions. We characterize the relationship between the customer due date preferences and the required reservation quantities and show that not considering the t-pooling effect (as done in traditional capacity and inventory rationing literature) leads to excessive capacity reservations. Numerical analyses are conducted to investigate the behavior and performance of our capacity reservation and due date quoting approach in a dynamic setting with multiple planning horizons and roll-overs. One interesting and seemingly counterintuitive finding of our analyses is that under certain conditions reserving capacity for high priority customers not only improves high priority fulfillment, but also increases the overall system fill rate. Session A3: 28 June 4:30-6:00 Room 40.002 24 Session B3: 28 June 4:30-6:00 Room 40.010 Session Title: RM with Endogenous Costs 25 Session B3: 28 June 4:30-6:00 Room 40.010 26 Dynamic Pricing of Inventories with Menu Cost Gunnar Feldmann, University of Massachusetts The exchange of goods and service is affected by pricing policies. There are two broad categories: posted-price mechanisms (take-it-or-leave-it pricing), and pricediscovery mechanisms (auction pricing). In the past companies fixed prices of a product or service for a relative long time period; i.e, the prices were considered static. The reason for this strategy is found in the absence of accurate demand information, high transaction costs associated with changing prices, and huge investment in software and hardware to implement a dynamic pricing strategy. Although, dynamically posted prices are also take-it-or-leave-it prices the seller can dynamically change prices over time. The goal here is to balance demand and supply. Early adopters of dynamic pricing methods are commonly found in industries where the short-term capacity is difficult to change such as airlines, cruise ships, and hotels. Most of these industries operate in a centralized fashion which allows prices to be changed at little or no cost. Contrasting to the former are retail like industries, short-term supply is more flexible but price changes are costly. Hence the focus has been on improving inventory management practices. Only recently there has been an increased interest in adaptive pricing strategies. What has not been completely addressed in the literature though is the cost of changing prices in conjunction with inventory management. Costs associated with changing production rates (set up costs, new equipment, etc.) have been extensively studied. However, there is often also a cost associated to changing prices, for example, costs to update computer databases, to change posted price in the store, to send out advertising fliers, to buy radio advertising time, and so on. This type of cost is commonly referred to as menu cost. In order to compare the change in profit resulting from dynamic pricing policies, menu cost must be considered in the total cost formulation. In this paper we are particularly interested in the optimization of net profit in settings of constraint capacity or supply chains with high order costs which do not allow for responsive production/order changes in the face of demand uncertainty. Therefore, the motivation of this research effort is to make progress towards developing an effective method of demand adaptive pricing with menu cost. Our approach is similar to an(S,s) policy where prices are changed once the inventory goes above or below a certain threshold value. Session B3: 28 June 4:30-6:00 Room 40.010 27 Session C3: 28 June 4:30-6:00 Room 40.010 Session Title: Network RM 2 28 Network revenue management optimization using outer approximation algorithms Andy Philpott (Joint work with Garrett. van Ryzin and Michael Frankovich) We study dynamic bid-pricing policies in network revenue management problems. The policies are derived by computing an outer approximation to the optimal value function of the deterministic linear program that arises from considering the problem over the remainder of the decision horizon. The outer approximation results in an algorithm that triggers a re-optimization of the deterministic linear program when a threshold condition is reached. Using numerical simulations, the trigger policy is compared with a multi-stage stochastic programming approach to computing an optimal bid-pricing policy. Session C3: 28 June 4:30-6:00 Room 40.012 29 Session A4: 29 June 9:00-10:30 Room 40.002 Session Title: Supply Chain Issues and Contracts 30 Session A4: 29 June 9:00-10:30 Room 40.002 31 Session A4: 29 June 9:00-10:30 Room 40.002 32 Session B4: 29 June 9:00-10:30 Room 40.010 Session Title: RM with customer behavior JOINT MEMORY-DEPENDENT PRICING AND PRODUCT INTRODUCTION FOR MULTIPLE GENERATIONS Hasan Arslan* Soulaymane Kachani Kyrylo Shmatov Fierce competition and rapid innovation in technology intensive and fast-changing markets force many firms to introduce new generations of products at ever increasing rates. Firms adopt multiple generation product development framework, and compete to extend their product offerings through advancement of their product or process technology. As the rate of new generation product introductions increases, product life cycles in many industries such as electronics, personal computers, cellular phones, software, and toys become shorter, which magnifies the importance of managing both the development and sales of new generation products for manufacturing firms. Therefore, manufacturing firms face important strategic questions of when to introduce the new generation product, when to phase out the old generation product, what pricing mechanism to follow for the old generation product before and after the introduction of the new generation, and what pricing mechanism to follow for the new generation product before and after phasing out the old one. Answers to these questions significantly impact the survival and success of the firms. Since manufacturing firms benefit from introducing multiple generations of products, it is very helpful to understand the dynamics of product transition from one generation to the next, and to develop a strategic plan on when to bring new generation products to the market and how to price them over their product lifecycles. In our research, we analyze how to efficiently manage the development and sales of multiple generation products. In particular, we look at two mechanisms jointly: product introduction timing mechanism and product pricing mechanism. To provide insights into the dynamics of managing multiple generation products, we consider a manufacturer that introduces successive generations of products in a market where consumers form the reference price through their personal shopping experience and exposure to price information. A reference price can be defined as the internal price, to which consumers compare the observed price (Fibich et al., Operations Research 2003). As customers visit the store, they develop price expectations in the form of a reference price that becomes the benchmark, against which customers compare the current price. Namely, the reference price effect implies that differences between the reference price and the current shelf price affect the demand for the product. Since reference price is formed through past price exposures of consumers, the problem of optimal price strategy for the manufacturer results in an optimal control problem in a monopolistic environment, and a differential game in a competitive setting. Given the impact of the formed reference price on the demand for product generations, we develop * Sawyer Business School, Suffolk University, Boston, MA, harslan@suffolk.edu IEOR Department, Columbia University, New York, NY, kachani@ieor.columbia.edu APAM Department, Columbia University, New York, NY, kis2101@columbia.edu 33 optimization models for the manufacturer to determine (i) the optimal pricing strategies for the newly introduced product generation and the incumbent product generations and (ii) the optimal introduction times for the newly developed product generations. We analyze the manufacturer both in a monopoly and a duopoly environment. The main contributions of our research are threefold. First, we characterize pricing and introduction time strategies explicitly for each successive product generation. This helps us carry out sensitivity analysis for the optimal strategies to gain new insights into the problem. Second, we generalize our model to the case of an arbitrary number of successive generations. We show that this model can be solved much more efficiently compared to the equivalent dynamic programming model. We develop an efficient solution procedure to determine optimal pricing and introduction decisions, and use it to derive important insights into the problem’s nature. Third, we deduce the equilibrium pricing and introduction time strategies in a duopoly game between two manufacturers. We analyze both symmetric and asymmetric manufacturers in terms of the impact of reference price on demand. We demonstrate the value of taking the reference price effect into account in developing both pricing and product life-cycle decisions. 34 Consumer Search Behavior when the Product Quality is Uncertain Laurens G. Debo, Carnegie Mellon University Garrett J. van RyzinColumbia University Observed inventory levels may influence customer purchasing, especially when there is uncertainty on the quality of the product. Customers may infer from low inventory levels that other customers have bought the product and therefore the product may be of high quality. We analyze a model in which the equilibrium customer search and purchasing behavior depends on the costs of information gathering, the prior of the quality, the strength of the private information, the noise on the inventory observations and the replenishment lead times. We discuss implications for design of distribution networks for new, innovative, fashion or other products with uncertain quality. Session B4: 29 June 9:00-10:30 Room 40.010 35 Session C4: 29 June 9:00-10:30 Room 40.012 Session Title: Robust RM2 36 Robust Newsvendor Competition Houyuan Jiang, Judge Business School, University of Cambridge Serguei Netessine, The Wharton School of Business, University of Pennsylvania Sergei Savin, Graduate School of Business, Columbia University We generalize the analysis of newsvendor competition to the setting in which the competitors only possess information about support of demand distribution but do not know the distribution itself. We analyze and compare several approaches relying on (relative or absolute) regret minimization or the worst-case criterion instead of the traditional approach in which the expected profit is maximized. For a game with an arbitrary number of players we find the unique Nash equilibrium solution for these robust newsvendor games. We further obtain closed-form expressions for the optimal order quantities which lead to intuitive insights into the problem. Numerical analysis indicates that, among different robust approaches, the absolute regret minimization offers the most sensible alternative when demand distribution is unknown. Session C4: 29 June 9:00-10:30 Room 40.012 37 Session A5: 29 June 11:00-12:30 Room 40.002 Session Title: Competition in RM 38 Session A5: 29 June 11:00-12:30 Room 40.002 39 Pricing DP Game Darius Walczak, PROS Revenue Management Abstract: We investigate two airlines in the following competitive setting. Each airline tries to maximize expected revenue from an initial, finite inventory of seats over a finite time horizon. The price point structure is the same for both. Customers seek the lowest price in the market; when the prices match they make a random choice. We allow a nonzero probability that a customer does not buy. All necessary probabilities are known to each airline and the price points can change at each customer request. The model is related to Dudey (1992) and Talluri (2003). We analyze conditions needed for the existence of pure strategy equilibria in the sense of Nash. We present realistic numerical examples including cases without pure strategy or with multiple equilibria. We compare the equilibrium pricing to the cooperative solution and several heuristics, focusing on those that make use of market information modeled as an independent process (Walczak 2006). Session A5: 29 June 11:00-12:30 Room 40.002 40 Session B5: 29 June 11:00-12:30 Room 40.010 Session Title: Demand Models and Forecasts Retracing choice process amongst SNCF clients E. Wiesel1&2, M. Riss1, L. Brotcorne2, G. Savard3 SNCF, Innovation & Research Department, Paris, France Université de Valenciennes, LAMIH-ROI, Valenciennes, France 3 Polytechnique Montréal, Montréal, Québec, Canada 1 2 The price bracket of the SNCF, the French National Railway Company, is quite extended and offers various products to its customers. At the current situation, the databases handled by the company, hold information concerning actual booking transactions but very few details concerning the clients who carry them out. Nevertheless, understanding the demand is a vital element for the success of the company which must react to the rapidly developing market of transportation. Needless to say that the demand concerns various activities of the company: marketing, strategy, the commercial optimization centre, client service etc. As it concerns Revenue Management, the development of a high-performance revenue optimization tool requires mastering forecasting and estimation techniques. In recent models of Revenue Management it is acknowledged that the fundamental unit of demand is the client1. It is admitted that demand is the result of a decision process involving the clients and the available alternatives. Therefore, we must shift our focus to models of client preferences and behaviour. This talk will present a framework which attempts to reconstruct the choice process made by the clients, taking account of the accessible data, historic booking transactions and marketing surveys. Our model is based on discrete choice methodology, a theory relying on the principle of utility maximization. Preferences are represented by the probability a client choose a given alternative within his choice set. These probabilities are evaluated given the choice sets of the clients and their respective decision protocols. We pay a specific attention to latent structures in the choice process: the correlation between the products offered by the company, the choice set formation and the decision protocols. We refer to generalized random utility models, a family of models which relies on the concept of hierarchical choice trees that goes hand in hand with a more explicit consideration of the factors that intervene in the decision process. According to surveys carried out by the marketing service of the SNCF and taking account of the particularities of the company’s price bracket, we propose two kinds of latent classes: we define different customer profiles related to the availabilities of the transportation products (for instance, card holders vs. non card holders) and segments (leisure, business, etc.) which refer to the decision mode. 1 G. J. van Ryzin, Demand Models, In Fifth Annual INFORMS Revenue Management and Pricing Section Conference, Cambridge, MA, June 2005, Graduate School of Business, Columbia University. 41 Tuning up the concepts of profiles and segments we can take under consideration the characteristics of the clients and other latent variables. In order to obtain a tractable model we consider a linear by parameters utility function concentrating on the attributes of the alternatives (tariff, duration, quality of service, etc.) and choose a Logit kernel which yields a closed form probability function. Howbeit, the structure of the model remains general as it can treat non linear relations between the attributes and together with the other components of the model we can simulate other distributions such as the Probit kernel. The originality of this model lays in it capacity to adapt to the SNCF environment and specificities, in particular, the data handled by the company. Nevertheless, this is a generic model that can be easily adapted to other settings, not necessarily in the transportation industry, and could be used in various fields. Moreover, this model provides an application of discrete choice theory for the analysis of demand on a microeconomic level. Forecasting Models and Scenarios Analysis FERNANDO CASTEJÓN IBERIA L.A.E. – REVENUE MANAGEMENT A review about the different techniques used by Iberia to provide (each week) information about the main traffic parameters: - ASK - RPK - LF - Yield - Revenue - RASK Also, a tool to study the different results (scenarios) when changes are done in price or capacity will be showed. Session B5: 29 June 11:00-12:30 Room 40.010 42 Session B5: 29 June 11:00-12:30 Room 40.010 43 Session C5: 29 June 11:00-12:30 Room 40.012 Session Title: Pricing and inventory for differentiated products 44 Session C5: 29 June 11:00-12:30 Room 40.012 45 Session C5: 29 June 11:00-12:30 Room 40.012 46 Session A6: 29 June 2:00-3:30 Room 40.002 Session Title: RM in Service Applications Revenue Management for Facilities Where Customers Wait for Service Avi Giloni, Sy Syms School of Business, Yeshiva University Philip M. Troy, Ph.D., Decision Support Systems Analyst, Les Entreprises TROYWARE An important component of managing revenue generating facilities where customers wait for services is to select an implementable pricing mechanism that maximizes revenues. Since waiting can take away from the benefit customers receive for services, customers will generally try to ensure that the benefit they receive exceeds the waiting costs they expect to incur plus any tolls they are charged. We assume that customers can compute their expected waiting costs provided they can see the jobs in front of them and the distribution of time that those jobs will take to be serviced, or that they know the parameters characterizing arrivals and job servicing. There have been several pricing mechanisms suggested in the literature. One common pricing mechanism is to charge a state independent toll, i.e. a toll that does not vary with the number of jobs in the facility. A less commonly used pricing mechanism is to charge a state dependent toll that does vary with the number of jobs in the facility. The latter mechanism has the advantage of being better able to respond to different levels of waiting, arriving customers incur. Unfortunately, both types of pricing mechanisms are limited by their inability to fully collect as revenue the value that customers receive in excess of their costs. With respect to state dependent tolls, this is a well known result (see Chen and Frank, 2001), and it can be easily demonstrated for state independent tolls as well. This limitation is most obvious when there are just two types of customers using the facility, where the first type has a low arrival rate but high job benefits, and the second type has a high arrival rate and low job benefits. An obvious approach to addressing this limitation is applying revenue management principles to the toll selection process. Surprisingly, to the best of our knowledge, this has not been done. In context of this, our first goal is to apply and characterize revenue management approaches to state dependent and independent toll mechanisms, in context of facilities in which customers are served on a first come first served basis. We do so both with and without assuming that facilities have the ability to perfectly differentiate customer groups, the former to determine what is possible, and the latter to determine what is plausible. For the purpose of analysis we rely on an M/M/S/I queuing model, and the assumption that customer waiting costs are strictly non-decreasing in the expected time they wait. Our approach permits us to relax the frequently used assumptions that revenue is continuously differentiable and concave as a function of the job submission rate. We also allow multiple servers. Keeping in mind that state dependent pricing mechanisms appear to be more controllable, due to their ability to specify a toll for each state, our second goal is to characterize circumstances in which it is possible to achieve similar results using state independent tolls as can be obtained using state dependent tolls. We have three results. The first is we demonstrate how to apply revenue maximization techniques to facilities where customers wait for service for both state dependent as well as state independent tolls. Surprisingly, when using state independent tolls and allowing customers to see the jobs in front of them, the 47 resulting optimization problem appears too complex to solve exactly, but can be approximately solved using a heuristic. Our second result is that the revenue management approach for state dependent tolls, the approach that appears more complex, is more effective, simpler, and easier to compute than for the state independent tolls. Our third result is that the state independent toll mechanisms can obtain similar results to those obtainable using the state dependent toll mechanism when the increase in expected waiting costs between states is very small. This can occur not only when waiting cost rates are low, but also when there are multiple servers. This in turn suggests that managers wishing to use a state independent revenue management scheme consider implementing their facility with multiple servers. 48 REVENUE MANAGEMENT FOR ONLINE ADVERTISING Victor Araman Kristin Fridgeirsdottir The Internet is a fast growing advertising medium. Companies are increasingly taking advantage of the Internet to reach out to more customers and are allocating an increasing portion of their marketing budget towards online advertising. We consider a web host that generates revenues from displaying advertisements on its website. The problem faced by the web host is quite complex with multiple ad spaces on many webpages to be assigned to different advertisers. Each advertiser brings a set of requirements including number of hits, size of the ad, length of the ad campaign, types of customers to display the ad to etc. In this paper we focus on the operational problem of the web host of effectively matching demand with supply through pricing. The web host faces two main uncertainties. One driven by the demand, where advertisers approach the web host requiring a number of impressions (number of viewers to see the ad) to be met in a certain period of time. The other one driven by supply, which consists of the viewers visiting the web site. The web hosts seeks to determine the advertising price that attracts the right number of advertisers given the stream of visitors to the website so that the advertising revenues generated are maximized. We model the web host operation as a multi-class queueing system with multiple servers. We assume both the advertisers’ arrivals and the viewers’ stream to be Poisson processes. This system has interesting features different from those of classical queueing systems. We use the model to analyze the problem faced by the web host of determining the optimal price to charge the advertiser. Moreover, the models developed in this paper provide a new tractable framework for analyzing the operations of a web host. Session A6: 29 June 2:00-3:30 Room 40.002 49 Session A6: 29 June 2:00-3:30 Room 40.002 50 Session B6: 29 June 2:00-3:30 Room 40.010 Session Title: Overbooking 51 VALUE OF TRAFFIC MIX-BASED OVERBOOKING J. Lancaster Munich Graff, JDA Software Overbooking is the most financially successful and oldest of revenue management practices. Although overbooking is considered a solved problem in most service industries, air cargo faces additional challenges because of the dramatic variation between the quantities booked and the quantities actually shipped. The effectiveness of air cargo overbooking models is heavily dependent on the quality of the show-up rate forecasts, which are traditionally produced from leg level history. Because show-up behavior is a characteristic of the shipper, however, rather than the leg, intuition suggests that forecasting improvements can be achieved by segmenting demand and modeling at customer rather than leg level. This study tests that intuition against data provided by a U.S. air carrier, examining the convergence rates of a simple leg-based forecaster following shifts in traffic mix. The results are particularly interesting for revenue management, because the very purpose of a revenue management system is to change the traffic mix. With originand-destination controls and a leg-based forecasting solution, this results in an openloop system with no effective feedback between traffic mix optimization and show-up rate forecasting. Session B6: 29 June 2:00-3:30 Room 40.010 52 Session B6: 29 June 2:00-3:30 Room 40.010 53 Session C6: 29 June 2:00-3:30 Room 40.012 Session Title: RM with learning Sequential prediction under imperfect monitoring Gabor Lugosi, ICREA & UPF We survey some on-line forecasting techniques for sequential prediction when the forecaster does not have access to the sequence to be predicted but rather receives a feedback signal. We investigate when the forecaster can perform almost as well as the best predictor defined in hindsight. We describe some stylized applications for dynamic pricing problems. Pricing and Buyer Learning Anton Kleywegt, Georgia Tech Many traditional dynamic pricing models such as the ones widely used in revenue management assume that the demand at each point in time depends on the price at that point in time only, that is, it is independent of prices at other points in time. Recently some models of so-called strategic customer behavior have been studied, in which buyers' purchasing decisions at a point in time depend on the prices at other points in time, for example, on the sellers' pricing policies. Many new questions are associated with such models. One question is how the buyers can be expected to obtain and process all the information necessary to make such complicated decisions. We study several models in which buyers learn quantities that are simpler than the pricing policies of the sellers. We investigate the convergence of the buyers' estimates, and compare the limits with outcomes associated with full information- 54 Session C6: 29 June 2:00-3:30 Room 40.012 55 Session A7: 29 June 4:00-5:00 Room 40.002 Session Title: Right Approach to RM 56 Is Yield Management Accessible to the Small Hotelier? By Robin Williams, Lluis Colomines, Aïda Díodeplus Contrary to popular belief within some areas of the Industry, the most successful approach to yield is a step by step development of techniques and knowledge which may or may not lead to an inversion in software. The starting point has to be the acquirement of knowledge, understanding the basis of yield is what will provide any hotelier with the foundation on which to build an effective Yield Management process. Once an hotelier understands those principles he will be able to put in place the necessary process to ensure that the accurate information required is obtained on a regular basis. Many hoteliers have had great success by using simple Excel or Access based calculations with existing although reduced historical information and this has given them the encouragement to continue improving these processes gradually but effectively. Based on this idea of starting small and building on strong foundations it would seem that what the majority of hoteliers need today is good strong tutoring in the principles of Yield and the possibilities available to them internally and externally, without the need for excessive resources. Similarly, one can also consider the idea of punctual actions or can Yield only be applied on a continual and constant basis? Session A7: 29 June 4:00-5:00 Room 40.002 57 Session B7: 29 June 4:00-5:00 Room 40.010 Session Title: RM in Healthcare Pricing-Capacity Tradeoffs in Rehabilative Healthcare Networks: Models and Open Problems William P. Millhiser, PhD Assistant Professor of Management Baruch College, City University of New York A United States network of rehabilative care facilities bases the decision to accept patients for prescheduled surgeries on bed availably and expected reimbursement. With uncertain recovery times, there are weeks when all beds are occupied, prescheduled surgeries are delayed and new surgery requests are rejected. Using queueing network with blocking models, we explore relationships between a facility’s capacity and the reimbursement thresholds used for patient acceptance/rejection. The Performance among Nonlinear Pricing Scheme in Medical Service in Korea Dr. Youngsik Kwak (Jinju National University, Jinju, Korea) Dr. Sookyung Paik (Inje University Paik Medical Center, Seoul, Korea) Dr. Brian Y. S. Nam (Samsung OpenTide China, Beijing, China) Dr. Hwanho Ha (Jinju National University, Jinju, Korea) Nonlinear pricing abounds in practice because it is a potentially powerful pricing method to explore consumer surplus. The various forms of nonlinear pricing are feasible within a given industry. In this context, it is important for manufacturers and retailers to understand which nonlinear pricing scheme is appropriate to apply in their specific situation and which nonlinear pricing schedule is the most profitable in their market situation. Although the merits of nonlinear pricing are well documented, the attempt to apply nonlinear pricing in medical service in South Korea has been relatively rare. The researcher aims to try to full this gap by applying a practice-oriented simulation model to health examination data. The model consists of two parts; the procedure of calibrating the price response function for each client firm and the simulation calculating the price points for maximizing sales volume and profit for each nonlinear pricing scheme. The researchers apply the model to Inje University Paik Medical Center’s health examination data for the first time in Korean medical market. We compare the sales volumes among nonlinear pricing scheme such as n-block tariff, two-part tariff and all unit discount price schedule. We found that n-block tariff outperforms two-part tariff, all unit discount price schedule, and uniform pricing. 58 Directions to Hotel Arts (DINNER ON 28 June 7:30 PM) Take a right on Ramon Trias Fargas when you exit the University Keep walking on Ramon Trias Fargas (towards the beach) till you see this building (hotel Arts) on your left (5 to 10 minute walk) Cocktail hour is in the garden Dinner is in Sala Pau Casals 59