Garrett J. van Ryzin

The theory and practice of revenue management

Quantity-Based RM.- Single-Resource Capacity Control.- Network Capacity Control.- Overbooking.- Price-based RM.- Dynamic Pricing.- Auctions.- Common Elements.- Customer-Behavior and Market-Response Models.- The Economics of RM.- Estimation and Forecasting.- Industry Profiles.- Implementation.

Revenue Management: Research Overview and Prospects

This survey reviews the forty-year history of research on transportation revenue management (also known as yield management). We cover developments in forecasting, overbooking, seat inventory control, and pricing, as they relate to revenue management, and suggest future research directions. The survey includes a glossary of revenue management terminology and a bibliography of over 190 references.

A Multiproduct Dynamic Pricing Problem and Its Applications to Network Yield Management

A firm has inventories of a set of components that are used to produce a set of products. There is a finite horizon over which the firm can sell its products. Demand for each product is a stochastic point process with an intensity that is a function of the vector of prices for the products and the time at which these prices are offered. The problem is to price the finished products so as to maximize total expected revenue over the finite sales horizon. An upper bound on the optimal expected revenue is established by analyzing a deterministic version of the problem. The solution to the deterministic problem suggests two heuristics for the stochastic problem that are shown to be asymptotically optimal as the expected sales volume tends to infinity. Several applications of the model to network yield management are given. Numerical examples illustrate both the range of problems that can be modeled under this framework and the effectiveness of the proposed heuristics. The results provide several fundamental insights into the performance of yield management systems.

A stochastic and dynamic vehicle routing problem in the Euclidean plane

We propose and analyze a generic mathematical model for dynamic, stochastic vehicle routing problems, the dynamic traveling repairman problem (DTRP). The model is motivated by applications in which the objective is to minimize the wait for service in a stochastic and dynamically changing environment. This is a departure from classical vehicle routing problems where one seeks to minimize total travel time in a static, deterministic environment. Potential areas of application include repair, inventory, emergency service and scheduling problems. The DTRP is defined as follows: Demands for service arrive in time according to a Poisson process, are independent and uniformly distributed in a Euclidean service region, and require an independent and identically distributed amount of on-site service by a vehicle. The problem is to find a policy for routing the service vehicle that minimizes the average time demands spent in the system. We propose and analyze several policies for the DTRP. We find a provably optima...

Stocking Retail Assortments Under Dynamic Consumer Substitution

We analyze a single-period, stochastic inventory model (newsboy-like model) in which a sequence of heterogeneous customers dynamically substitute among product variants within a retail assortment when inventory is depleted. The customer choice decisions are based on a natural and classical utility maximization criterion. Faced with such substitution behavior, the retailer must choose initial inventory levels for the assortment to maximize expected profits.Using a sample path analysis, we analyze structural properties of the expected profit function. We show that, under very general assumptions on the demand process, total sales of each product are concave in their own inventory levels and possess the so-calleddecreasing differences property, meaning that the marginal value of an additional unit of the given product is decreasing in the inventory levels of all other products. For a continuous relaxation of the problem, we then show, via counterexamples, that the expected profit function is in general not even quasiconcave. Thus, global optimization may be difficult. However, we propose and analyze a stochastic gradient algorithm for the problem, and prove that it converges to a stationary point of the expected profit function under mild conditions. Finally, we apply the algorithm to a set of numerical examples and compare the resulting inventory decisions to those of some simpler, naive heuristics. The examples show that substitution effects can have a significant impact on an assortments gross profits. The examples also illustrate some systematic distortions in inventory decisions if substitution effects are ignored. In particular, under substitution one should stock relatively more of popular variants and relatively less of unpopular variants than a traditional newsboy analysis indicates.

Stochastic and dynamic vehicle routing in the Euclidean plane with multiple capacitated vehicles

In 1991, D. J. Bertsimas and G. van Ryzin introduced and analyzed a model for stochastic and dynamic vehicle routing in which a single, uncapacitated vehicle traveling at a constant velocity in a Euclidean region must service demands whose time of arrival, location and on-site service are stochastic. The objective is to find a policy to service demands over an infinite horizon that minimizes the expected system time (wait plus service) of the demands. This paper extends our analysis in several directions. First, we analyze the problem of m identical vehicles with unlimited capacity and show that in heavy traffic the system time is reduced by a factor of 1/m2 over the single-server case. One of these policies improves by a factor of two on the best known system time for the single-server case. We then consider the case in which each vehicle can serve at most q customers before returning to a depot. We show that the stability condition in this case depends strongly on the geometry of the region. Several pol...

On the Choice-Based Linear Programming Model for Network Revenue Management

Gallego et al. [Gallego, G., G. Iyengar, R. Phillips, A. Dubey. 2004. Managing flexible products on a network. CORC Technical Report TR-2004-01, Department of Industrial Engineering and Operations Research, Columbia University, New York.] recently proposed a choice-based deterministic linear programming model (CDLP) for network revenue management (RM) that parallels the widely used deterministic linear programming (DLP) model. While they focused on analyzing “flexible products”---a situation in which the provider has the flexibility of using a collection of products (e.g., different flight times and/or itineraries) to serve the same market demand (e.g., an origin-destination connection)---their approach has broader implications for understanding choice-based RM on a network. In this paper, we explore the implications in detail. Specifically, we characterize optimal offer sets (sets of available network products) by extending to the network case a notion of “efficiency” developed by Talluri and van Ryzin [Talluri, K. T., G. J. van Ryzin. 2004. Revenue management under a general discrete choice model of consumer behavior. Management Sci.50 15--33.] for the single-leg, choice-based RM problem. We show that, asymptotically, as demand and capacity are scaled up, only these efficient sets are used in an optimal policy. This analysis suggests that efficiency is a potentially useful approach for identifying “good” offer sets on networks, as it is in the case of single-leg problems. Second, we propose a practical decomposition heuristic for converting the static CDLP solution into a dynamic control policy. The heuristic is quite similar to the familiar displacement-adjusted virtual nesting (DAVN) approximation used in traditional network RM, and it significantly improves on the performance of the static LP solution. We illustrate the heuristic on several numerical examples.

Overbooking with Substitutable Inventory Classes

This paper considers an overbooking problem with multiple reservation and inventory classes, in which the multiple inventory classes may be used as substitutes to satisfy the demand of a given reservation class (perhaps at a cost). The problem is to jointly determine overbooking levels for the reservation classes, taking into account the substitution options. Such problems arise in a variety of revenue management contexts, including multicabin aircraft, back-to-back scheduled flights on the same leg, hotels with multiple room types, and mixed-vehicle car rental fleets. We model this problem as a two-period optimization problem. In the first period, reservations are accepted given only probabilistic knowledge of cancellations. In the second period, cancellations are realized and surviving customers are assigned to the various inventory classes to maximize the net benefit of assignments (e.g., minimize penalties). For this formulation, we show that the expected revenue function is submodular in the overbooking levels, which implies the natural property that the optimal overbooking level in one reservation class decreases with the number of reservations held in the other reservation classes. We then propose a stochastic gradient algorithm to find the joint optimal overbooking levels. We compare the decisions of the model to those produced by more naive heuristics on some examples motivated by airline applications. The results show that accounting for substitution when setting overbooking levels has a small, but still significant, impact on revenues and costs.

A Randomized Linear Programming Method for Computing Network Bid Prices

We analyze a randomized version of the deterministic linear programming (DLP) method for computing network bid prices. The method consists of simulating a sequence of realizations of itinerary demand and solving deterministic linear programs to allocate capacity to itineraries for each realization. The dual prices from this sequence are then averaged to form a bid price approximation. This randomized linear programming (RLP) method is only slightly more complicated to implement than the DLP method. We show that the RLP method can be viewed as a procedure for estimating the gradient of the expected perfect information (PI) network revenue. That is, the expected revenue obtained with full information on future demand realizations. The expected PI revenue can, in turn, be viewed as an approximation to the optimal value function. We establish conditions under which the RLP procedure provides an unbiased estimator of the gradient of the expected PI revenue. Computational tests are performed to evaluate the revenue performance of the RLP method compared to the DLP.

Estimating Primary Demand for Substitutable Products from Sales Transaction Data

We propose a method for estimating substitute and lost demand when only sales and product availability data are observable, not all products are displayed in all periods (e.g., due to stockouts or availability controls), and the seller knows its aggregate market share. The model combines a multinomial logit (MNL) choice model with a nonhomogeneous Poisson model of arrivals over multiple periods. Our key idea is to view the problem in terms of primary (or first-choice) demand; that is, the demand that would have been observed if all products had been available in all periods. We then apply the expectation-maximization (EM) method to this model, and we treat the observed demand as an incomplete observation of primary demand. This leads to an efficient, iterative procedure for estimating the parameters of the model. All limit points of the procedure are provably stationary points of the incomplete data log-likelihood function. Every iteration of the algorithm consists of simple, closed-form calculations. We illustrate the effectiveness of the procedure on simulated data and two industry data sets.