Changchun Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Changchun Liu has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Xi Xiang
Tsinghua University
13 shared publications Li Zheng
Department of Industrial Engineering, Tsinghua University, Beijing, China
11 shared publicationsChung-Piaw Teo
6 shared publicationsXi Xiang
4 shared publicationsCanrong Zhang
Logistics Engineering and Simulation Laboratory, Graduate School at Shenzhen, Tsinghua University, Shenzhen 518055, P. R. China
3 shared publicationsLixin Miao
3 shared publications
Latest Publications
- Open Access
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Auditable Priority Appointments: Stability, Fairness, and Congestion at a Metro Gate
Public appointment systems for disadvantaged users face an inherent tension between priority protection and system efficiency. We study priority-entry appointments for metro passengers with mobility or assistance needs. Allocating appointments to reduce congestion at general gates can improve overall system efficiency, but may disadvantage the very passengers whom priority access is intended to protect. Conversely, requiring exact stability-so that every unfulfilled group-slot request has a complete service-or priority-based explanation-can leave little scope to improve system-wide congestion. The challenge is to balance these objectives: preserving meaningful priority protection while allowing appointments to be allocated more efficiently. Methodology/results: We model how appointment allocation affects the workload remaining at general gates. Under strict passenger preferences and slot priorities, every stable allocation produces the same modeled congestion, revealing the efficiency cost of exact stability. We introduce block-bounded stability (BB), which relaxes exact stability while retaining reciprocal protection between service and priority: the less service a group receives, the more closely the allocation of the corresponding slot must respect its priority. For every listed group-slot pair, BB guarantees that at least one of the two classical explanations provides coverage of one-half or more, while permitting strict reductions in modeled congestion. In calibrated synthetic experiments with policy-specific uncertainty treatments, BB recovers 42.1% of the gap between exact stability and unconstrained efficiency. The three-month Nanchang Metro deployment recorded 23,485 passenger requests, linking BB-based rosters to notifications and staff assistance. Surveybased calculations implied a 38.7% reduction in mean queueing time, and satisfaction averaged 8.54/10. A two-station comparison showed a 17.9-unit decline after passenger-volume and weekday adjustment. Managerial implications: BB gives operators a transparent way to balance priority access with system-wide efficiency. The less service a group receives at a requested time or better, the less capacity the operator may leave vacant or assign to lower-priority groups in that slot. This protection applies separately to every group-slot pair, preventing congestion reduction from coming at the expense of unbounded departures from the rationale for priority access. At the same time, operators retain flexibility to allocate appointments in ways that reduce pressure on general gates. Compliance can be verified directly from roster and request records without disclosing a demand forecast.
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAuditable Priority Appointments: Stability, Fairness, and Congestion at a Metro Gate
Public appointment systems for disadvantaged users face an inherent tension between priority protection and system efficiency. We study priority-entry appointments for metro passengers with mobility or assistance needs. Allocating appointments to reduce congestion at general gates can improve overall system efficiency, but may disadvantage the very passengers whom priority access is intended to protect. Conversely, requiring exact stability-so that every unfulfilled group-slot request has a complete service-or priority-based explanation-can leave little scope to improve system-wide congestion. The challenge is to balance these objectives: preserving meaningful priority protection while allowing appointments to be allocated more efficiently. Methodology/results: We model how appointment allocation affects the workload remaining at general gates. Under strict passenger preferences and slot priorities, every stable allocation produces the same modeled congestion, revealing the efficiency cost of exact stability. We introduce block-bounded stability (BB), which relaxes exact stability while retaining reciprocal protection between service and priority: the less service a group receives, the more closely the allocation of the corresponding slot must respect its priority. For every listed group-slot pair, BB guarantees that at least one of the two classical explanations provides coverage of one-half or more, while permitting strict reductions in modeled congestion. In calibrated synthetic experiments with policy-specific uncertainty treatments, BB recovers 42.1% of the gap between exact stability and unconstrained efficiency. The three-month Nanchang Metro deployment recorded 23,485 passenger requests, linking BB-based rosters to notifications and staff assistance. Surveybased calculations implied a 38.7% reduction in mean queueing time, and satisfaction averaged 8.54/10. A two-station comparison showed a 17.9-unit decline after passenger-volume and weekday adjustment. Managerial implications: BB gives operators a transparent way to balance priority access with system-wide efficiency. The less service a group receives at a requested time or better, the less capacity the operator may leave vacant or assign to lower-priority groups in that slot. This protection applies separately to every group-slot pair, preventing congestion reduction from coming at the expense of unbounded departures from the rationale for priority access. At the same time, operators retain flexibility to allocate appointments in ways that reduce pressure on general gates. Compliance can be verified directly from roster and request records without disclosing a demand forecast.
- Open Access
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Auditable Priority Appointments: Stability, Fairness, and Congestion at a Metro Gate
\textbf{Problem definition:} Public appointment systems for disadvantaged users face an inherent tension between priority protection and system efficiency. We study priority-entry appointments for metro passengers with mobility or assistance needs. Allocating appointments to reduce congestion at general gates can improve overall system efficiency, but may disadvantage the very passengers whom priority access is intended to protect. Conversely, requiring exact stability—so that every unfulfilled group–slot request has a complete service- or priority-based explanation—can leave little scope to improve system-wide congestion. The challenge is to balance these objectives: preserving meaningful priority protection while allowing appointments to be allocated more efficiently. \textbf{Methodology/results:} We model how appointment allocation affects the workload remaining at general gates. Under strict passenger preferences and slot priorities, every stable allocation produces the same modeled congestion, revealing the efficiency cost of exact stability. We introduce block-bounded stability (BB), which relaxes exact stability while retaining reciprocal protection between service and priority: the less service a group receives, the more closely the allocation of the corresponding slot must respect its priority. For every listed group–slot pair, BB guarantees that at least one of the two classical explanations provides coverage of one-half or more, while permitting strict reductions in modeled congestion. In calibrated synthetic experiments with policy-specific uncertainty treatments, BB recovers 42.1\% of the gap between exact stability and unconstrained efficiency. The three-month Nanchang Metro deployment recorded 23,485 passenger requests, linking BB-based rosters to notifications and staff assistance. Survey-based calculations implied a 38.7\% reduction in mean queueing time, and satisfaction averaged 8.54/10. A two-station comparison showed a 17.9-unit decline after passenger-volume and weekday adjustment. \textbf{Managerial implications:} BB gives operators a transparent way to balance priority access with system-wide efficiency. The less service a group receives at a requested time or better, the less capacity the operator may leave vacant or assign to lower-priority groups in that slot. This protection applies separately to every group–slot pair, preventing congestion reduction from coming at the expense of unbounded departures from the rationale for priority access. At the same time, operators retain flexibility to allocate appointments in ways that reduce pressure on general gates. Compliance can be verified directly from roster and request records without disclosing a demand forecast.
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAuditable Priority Appointments: Stability, Fairness, and Congestion at a Metro Gate
\textbf{Problem definition:} Public appointment systems for disadvantaged users face an inherent tension between priority protection and system efficiency. We study priority-entry appointments for metro passengers with mobility or assistance needs. Allocating appointments to reduce congestion at general gates can improve overall system efficiency, but may disadvantage the very passengers whom priority access is intended to protect. Conversely, requiring exact stability—so that every unfulfilled group–slot request has a complete service- or priority-based explanation—can leave little scope to improve system-wide congestion. The challenge is to balance these objectives: preserving meaningful priority protection while allowing appointments to be allocated more efficiently. \textbf{Methodology/results:} We model how appointment allocation affects the workload remaining at general gates. Under strict passenger preferences and slot priorities, every stable allocation produces the same modeled congestion, revealing the efficiency cost of exact stability. We introduce block-bounded stability (BB), which relaxes exact stability while retaining reciprocal protection between service and priority: the less service a group receives, the more closely the allocation of the corresponding slot must respect its priority. For every listed group–slot pair, BB guarantees that at least one of the two classical explanations provides coverage of one-half or more, while permitting strict reductions in modeled congestion. In calibrated synthetic experiments with policy-specific uncertainty treatments, BB recovers 42.1\% of the gap between exact stability and unconstrained efficiency. The three-month Nanchang Metro deployment recorded 23,485 passenger requests, linking BB-based rosters to notifications and staff assistance. Survey-based calculations implied a 38.7\% reduction in mean queueing time, and satisfaction averaged 8.54/10. A two-station comparison showed a 17.9-unit decline after passenger-volume and weekday adjustment. \textbf{Managerial implications:} BB gives operators a transparent way to balance priority access with system-wide efficiency. The less service a group receives at a requested time or better, the less capacity the operator may leave vacant or assign to lower-priority groups in that slot. This protection applies separately to every group–slot pair, preventing congestion reduction from coming at the expense of unbounded departures from the rationale for priority access. At the same time, operators retain flexibility to allocate appointments in ways that reduce pressure on general gates. Compliance can be verified directly from roster and request records without disclosing a demand forecast.
- Open Access
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Multi-Product Pricing under Rational Inattention: Evidence, Estimation, and Field Deployment from Airline Ancillaries
We study the joint pricing of close substitutes when customer attention is unevenly distributed across products and data on any single (product, context) cell are too thin for fully separate demand estimation. Airline ancillaries, product-line variants, subscription tiers, and hotel room categories all share this structure. The standard multinomial-logit benchmark imposes homogeneous substitution and therefore cannot represent the product-level asymmetries observed in such markets. We develop a generalized rational-inattention demand model with product-specific attention weights. The seller's revenue problem is convex in the induced purchase probabilities and admits a semi-closed-form solution: each product's optimal price decomposes into its effective cost, a product-specific attention markup, and a common demand term that depends on the segment only through a single scalar, the aggregate purchase probability. Pricing therefore reduces to a one-dimensional root search regardless of assortment size. Calibration reduces to a linear program over the KKT residuals of the consumer's problem, with a finite-sample error bound governed by the ℓ 1minimum gain of the design matrix. When per-segment data are sparse, we pool neighboring segments and replace segment-specific intercepts with a linear function of observable features. The total parameter error decomposes into statistical error, merging bias, and approximation bias; the corresponding bias-variance trade-off is operationalized in deployment by a split-sample cross-validation procedure for configuration selection. A revenue-loss bound quadratic in the ℓ 1 parameter error links sample size to revenue and motivates three pricing rules of increasing data appetite: a segment-based rule, a feature-based rule, and a propensity-based rule (deployed as the PCP arm). We evaluate the framework in collaboration with a low-cost airline. Across five checked-baggage tiers, the cheapest weight tier (20 kg) is nearly two orders of magnitude (≈ 80×) more price-sensitive than the highest (40 kg), and cross-tier substitution is small relative to own-price response; both patterns are inconsistent with homogeneous substitution. In a randomized field deployment over 48 origin-destination pairs, the segment-based and feature-based rules each raise baggage revenue by 7-8% (𝑝 < 0.01) relative to uniform pricing, and their posted-price distribution first-order stochastically dominates that of the carrier's incumbent rule: average posted prices fall while revenue rises.
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMulti-Product Pricing under Rational Inattention: Evidence, Estimation, and Field Deployment from Airline Ancillaries
We study the joint pricing of close substitutes when customer attention is unevenly distributed across products and data on any single (product, context) cell are too thin for fully separate demand estimation. Airline ancillaries, product-line variants, subscription tiers, and hotel room categories all share this structure. The standard multinomial-logit benchmark imposes homogeneous substitution and therefore cannot represent the product-level asymmetries observed in such markets. We develop a generalized rational-inattention demand model with product-specific attention weights. The seller's revenue problem is convex in the induced purchase probabilities and admits a semi-closed-form solution: each product's optimal price decomposes into its effective cost, a product-specific attention markup, and a common demand term that depends on the segment only through a single scalar, the aggregate purchase probability. Pricing therefore reduces to a one-dimensional root search regardless of assortment size. Calibration reduces to a linear program over the KKT residuals of the consumer's problem, with a finite-sample error bound governed by the ℓ 1minimum gain of the design matrix. When per-segment data are sparse, we pool neighboring segments and replace segment-specific intercepts with a linear function of observable features. The total parameter error decomposes into statistical error, merging bias, and approximation bias; the corresponding bias-variance trade-off is operationalized in deployment by a split-sample cross-validation procedure for configuration selection. A revenue-loss bound quadratic in the ℓ 1 parameter error links sample size to revenue and motivates three pricing rules of increasing data appetite: a segment-based rule, a feature-based rule, and a propensity-based rule (deployed as the PCP arm). We evaluate the framework in collaboration with a low-cost airline. Across five checked-baggage tiers, the cheapest weight tier (20 kg) is nearly two orders of magnitude (≈ 80×) more price-sensitive than the highest (40 kg), and cross-tier substitution is small relative to own-price response; both patterns are inconsistent with homogeneous substitution. In a randomized field deployment over 48 origin-destination pairs, the segment-based and feature-based rules each raise baggage revenue by 7-8% (𝑝 < 0.01) relative to uniform pricing, and their posted-price distribution first-order stochastically dominates that of the carrier's incumbent rule: average posted prices fall while revenue rises.
- Open Access
- 24 July 2026
- Published by Elsevier in SSRN Electronic Journal
Salient Products in Multi-Product Price Competition: The Price Image Effect
Problem definition. In many markets, a small set of highly visible products plays a disproportionate role in shaping a firm's perceived expensiveness. We study competitive multi-product pricing when such salient products affect not only their own demand but also the demand for the firm's entire assortment through a price-image channel. The central question is how firms should price salient and nonsalient products in equilibrium when a salient price can act as a strategic signal of the firm's overall price position. Methodology and results. We model each firm's price image as a max-affine function of salient product prices and embed it in a multinomial logit (MNL) demand system as a common utility shifter across the firm's products. This formulation captures key-value-item logic and max-gap anchoring: the product that drives price image may change endogenously as prices change. The model shows that salient products create a distinct competitive pricing tradeoff. Raising a salient product's price improves its own margin but worsens the firm's price image and depresses demand for the rest of the assortment; lowering it can function as a targeted competitive instrument whose benefit spills over to nonsalient products. Despite this cross-product coupling, each firm's best response is well defined. Moreover, in the baseline competitive model, a rival's full price vector affects a firm only through the rival's image-adjusted attractiveness, so equilibrium analysis reduces to a scalar fixed-point problem. This reduction clarifies when the market has a unique equilibrium, when multiple equilibria can arise, and how best-response dynamics select among them. Managerial implications. The results show that treating all products symmetrically can lead to systematically wrong competitive pricing decisions. In a simulated choice experiment with 800 observations, a price-image-blind MNL benchmark overprices the salient products and lies 8.7% from the true-model Nash equilibrium, while the estimated salient-product price-image model reduces this gap to 0.3%. We also illustrate the approach using a public four-brand camera conjoint dataset. The estimated model identifies zoom-and-video cameras as the salient product group. In the constructed multi-product pricing game, each brand prices its salient product below its nonsalient products, not because the salient product has low intrinsic utility, but because its price shapes the brand's overall image and therefore affects demand for the whole brand assortment.
- 24 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSalient Products in Multi-Product Price Competition: The Price Image Effect
Problem definition. In many markets, a small set of highly visible products plays a disproportionate role in shaping a firm's perceived expensiveness. We study competitive multi-product pricing when such salient products affect not only their own demand but also the demand for the firm's entire assortment through a price-image channel. The central question is how firms should price salient and nonsalient products in equilibrium when a salient price can act as a strategic signal of the firm's overall price position. Methodology and results. We model each firm's price image as a max-affine function of salient product prices and embed it in a multinomial logit (MNL) demand system as a common utility shifter across the firm's products. This formulation captures key-value-item logic and max-gap anchoring: the product that drives price image may change endogenously as prices change. The model shows that salient products create a distinct competitive pricing tradeoff. Raising a salient product's price improves its own margin but worsens the firm's price image and depresses demand for the rest of the assortment; lowering it can function as a targeted competitive instrument whose benefit spills over to nonsalient products. Despite this cross-product coupling, each firm's best response is well defined. Moreover, in the baseline competitive model, a rival's full price vector affects a firm only through the rival's image-adjusted attractiveness, so equilibrium analysis reduces to a scalar fixed-point problem. This reduction clarifies when the market has a unique equilibrium, when multiple equilibria can arise, and how best-response dynamics select among them. Managerial implications. The results show that treating all products symmetrically can lead to systematically wrong competitive pricing decisions. In a simulated choice experiment with 800 observations, a price-image-blind MNL benchmark overprices the salient products and lies 8.7% from the true-model Nash equilibrium, while the estimated salient-product price-image model reduces this gap to 0.3%. We also illustrate the approach using a public four-brand camera conjoint dataset. The estimated model identifies zoom-and-video cameras as the salient product group. In the constructed multi-product pricing game, each brand prices its salient product below its nonsalient products, not because the salient product has low intrinsic utility, but because its price shapes the brand's overall image and therefore affects demand for the whole brand assortment.
- Open Access
- 16 May 2026
- Published by Elsevier in SSRN Electronic Journal
Perspective Reformulation for Fractional Subset Selection
Retailers, online platforms, and content recommenders face a deceptively simple question: which subset of products, items, or features should be offered? The classical answer-rank items by revenue and offer a revenue-ordered prefix-is optimal under the unconstrained multinomial logit model and remains the industry default. However, real customers experience choice overload, respond to dominated decoys, and differ across preference segments in ways that no single ranking can fully capture. The profit-maximizing assortment is therefore more curated and structurally distinct from revenue ordering: smaller when choice overload dominates, larger when strategic decoys or attraction effects are beneficial, and costly when misidentified. A broad class of revenue-optimization problems share a common form-maximize a ratio of a linear benefit to a denominator capturing competition or saturation-with multi-segment variants extending this to a sum of ratios. We study this structure through the perspective transformation, a classical algebraic substitution that yields three complementary results: an exact polynomialtime reformulation in the linear-over-submodular regime; a Frank-Wolfe heuristic with a computable stationarity certificate and a rounding guarantee in the convex-over-convex regime; and exact liftability of the shared-decision formulation in the sum-of-ratios extension, supporting a first-order heuristic even when the lifted set is nonconvex. The framework's empirical performance tracks these structural guarantees. On moderate-scale benchmarks where exact solutions are computable, FWD recovers near-optimal assortments, empirically validating the rounding guarantee. At practical retail scale, FWD outperforms the quadratic-transform benchmark adapted to the box relaxation while running substantially faster in both single-and multi-segment sweeps. Two real-data studies confirm portability: on the Expedia ICDM-2013 hotel-search panel, the perspective lift identifies a robust dominanceinduced attraction effect and recovers five-hotel assortments that more than double the per-session revenue of revenue-ordering; on the Wisconsin Breast Cancer benchmark, the same machinery matches the best cross-validated SVM accuracy and substantially improves over marginal-gain greedy.
- 16 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPerspective Reformulation for Fractional Subset Selection
Retailers, online platforms, and content recommenders face a deceptively simple question: which subset of products, items, or features should be offered? The classical answer-rank items by revenue and offer a revenue-ordered prefix-is optimal under the unconstrained multinomial logit model and remains the industry default. However, real customers experience choice overload, respond to dominated decoys, and differ across preference segments in ways that no single ranking can fully capture. The profit-maximizing assortment is therefore more curated and structurally distinct from revenue ordering: smaller when choice overload dominates, larger when strategic decoys or attraction effects are beneficial, and costly when misidentified. A broad class of revenue-optimization problems share a common form-maximize a ratio of a linear benefit to a denominator capturing competition or saturation-with multi-segment variants extending this to a sum of ratios. We study this structure through the perspective transformation, a classical algebraic substitution that yields three complementary results: an exact polynomialtime reformulation in the linear-over-submodular regime; a Frank-Wolfe heuristic with a computable stationarity certificate and a rounding guarantee in the convex-over-convex regime; and exact liftability of the shared-decision formulation in the sum-of-ratios extension, supporting a first-order heuristic even when the lifted set is nonconvex. The framework's empirical performance tracks these structural guarantees. On moderate-scale benchmarks where exact solutions are computable, FWD recovers near-optimal assortments, empirically validating the rounding guarantee. At practical retail scale, FWD outperforms the quadratic-transform benchmark adapted to the box relaxation while running substantially faster in both single-and multi-segment sweeps. Two real-data studies confirm portability: on the Expedia ICDM-2013 hotel-search panel, the perspective lift identifies a robust dominanceinduced attraction effect and recovers five-hotel assortments that more than double the per-session revenue of revenue-ordering; on the Wisconsin Breast Cancer benchmark, the same machinery matches the best cross-validated SVM accuracy and substantially improves over marginal-gain greedy.
- 10 January 2026
- Published by Elsevier in Reliability Engineering & System Safety
- Vol. 271
A reliability-oriented conflict robust optimization model for berth allocation under operational disturbances
- 10 January 2026
- Published by Elsevier in Reliability Engineering & System Safety
- Vol. 271
A reliability-oriented conflict robust optimization model for berth allocation under operational disturbances
- 4 March 2024
- Published by Elsevier in Computers & Industrial Engineering
- Vol. 190
Tactical vessel deployment and empty container repositioning considering container turnover times
- 4 March 2024
- Published by Elsevier in Computers & Industrial Engineering
- Vol. 190
Tactical vessel deployment and empty container repositioning considering container turnover times
- Open Access
- 1 January 2024
- Published by Elsevier in SSRN Electronic Journal
Picking Lotto Winning Numbers Using Mixed Integer Programs: Beyond Finite Geometry & Random Pick
Picking lottery numbers has long captivated the public imagination, evolving from a mere game of chance into a significant research challenge that intertwines mathematics, statistics, and even psychology. This study reevaluates betting strategies for the UK 6/59 Lotto game. Our objective is to develop a strategy for creating m tickets to maximize the potential of the highest-performing ticket, increasing the odds of winning. We begin by contextualizing the problem within a moment-based distributionally robust optimization model, and exploit the symmetry in the model formulation to extract the importance of the notion of ``pairwise similarities" to the performance of the strategy. This leads to a mixed-integer programming model to minimize overall similarity while ensuring balanced pairwise similarities among tickets. This new method surpasses both random pick and finite-geometry strategies, boosting the odds of matching three winning numbers by 2-3%, using 27 lotto tickets. Using this new strategy, we demonstrate that it is optimal to buy 15 lotto tickets, to maximize the break-even probability. This challenges the conventional belief that more bets increase losses - while buying one ticket has around 10.83% chances of breaking even, buying 15 tickets in the optimal manner has around 15.88% chances of breaking even. Increasing the size of bets beyond 15 tickets, however, leads to a diminishing return in terms of the break-even probability. This highlights a tipping point where the cost of additional tickets starts to outweigh the proportional increase in winning odds.
- 1 January 2024
- Published by Elsevier in SSRN Electronic Journal
Open AccessPicking Lotto Winning Numbers Using Mixed Integer Programs: Beyond Finite Geometry & Random Pick
Picking lottery numbers has long captivated the public imagination, evolving from a mere game of chance into a significant research challenge that intertwines mathematics, statistics, and even psychology. This study reevaluates betting strategies for the UK 6/59 Lotto game. Our objective is to develop a strategy for creating m tickets to maximize the potential of the highest-performing ticket, increasing the odds of winning. We begin by contextualizing the problem within a moment-based distributionally robust optimization model, and exploit the symmetry in the model formulation to extract the importance of the notion of ``pairwise similarities" to the performance of the strategy. This leads to a mixed-integer programming model to minimize overall similarity while ensuring balanced pairwise similarities among tickets. This new method surpasses both random pick and finite-geometry strategies, boosting the odds of matching three winning numbers by 2-3%, using 27 lotto tickets. Using this new strategy, we demonstrate that it is optimal to buy 15 lotto tickets, to maximize the break-even probability. This challenges the conventional belief that more bets increase losses - while buying one ticket has around 10.83% chances of breaking even, buying 15 tickets in the optimal manner has around 15.88% chances of breaking even. Increasing the size of bets beyond 15 tickets, however, leads to a diminishing return in terms of the break-even probability. This highlights a tipping point where the cost of additional tickets starts to outweigh the proportional increase in winning odds.
- Open Access
- 1 September 2023
- Published by SAGE Publications in Production and Operations Management
- Vol. 32 (9) , 2864-2884
Picking winners: Diversification through portfolio optimization
We develop a general framework for selecting a small pool of candidate solutions to maximize the chances that one will be optimal for a combinatorial optimization problem, under a linear and additive random payoff function. We formulate this problem using a two‐stage distributionally robust model, with a mixed 0–1 semidefinite program. This approach allows us to exploit the “diversification” effect inherent in the problem to address how different candidate solutions can be selected to improve the chances that one will attain a high ex post payoff. More interestingly, using this distributionally robust optimization approach, our model recovers the “evil twin” strategy, well known in the field of football pool betting, under appropriate settings. We also address the computational challenges of scaling up our approach to construct a moderate number of candidate solutions to increase the chances of finding one that performs well. To this end, we develop a sequential optimization approach based on a compact semidefinite programming reformulation of the problem. Extensive numerical results show the superiority of our approach over existing methods.
- 1 September 2023
- Published by SAGE Publications in Production and Operations Management
- Vol. 32 (9) , 2864-2884
Open AccessPicking winners: Diversification through portfolio optimization
We develop a general framework for selecting a small pool of candidate solutions to maximize the chances that one will be optimal for a combinatorial optimization problem, under a linear and additive random payoff function. We formulate this problem using a two‐stage distributionally robust model, with a mixed 0–1 semidefinite program. This approach allows us to exploit the “diversification” effect inherent in the problem to address how different candidate solutions can be selected to improve the chances that one will attain a high ex post payoff. More interestingly, using this distributionally robust optimization approach, our model recovers the “evil twin” strategy, well known in the field of football pool betting, under appropriate settings. We also address the computational challenges of scaling up our approach to construct a moderate number of candidate solutions to increase the chances of finding one that performs well. To this end, we develop a sequential optimization approach based on a compact semidefinite programming reformulation of the problem. Extensive numerical results show the superiority of our approach over existing methods.
- Open Access
- 1 April 2023
- Published by SAGE Publications in Production and Operations Management
- Vol. 32 (4) , 1205-1222
From share of choice to buyers' welfare maximization: Bridging the gap through distributionally robust optimization
We study the product design problem where a decision maker selects the features of a product from a set of feasible options. We focus on two widely studied objectives in this field, that is, the share of choice (SOC) and buyers' welfare (BW). The two objectives are vulnerable to different types of customer preference misspecification, that is, deviation from the nominal utility distribution and effects of outliers, respectively. We formulate a distributionally robust optimization (DRO) SOC maximization model and a winsorized BW maximization model to obtain robust solutions to the two problems. Interestingly, we show that the two robust models are equivalent in a certain sense—for appropriate choices of robustness parameters, both models return the same solution. This observation has important ramifications. For instance, it indicates that a product designed to yield higher BW is more robust to the deviation from the nominal utility distribution for the SOC problem, while that with higher SOC is less sensitive to the effect of outliers for the BW problem. Last but not least, we use the equivalence to develop a new approach to solve the DRO model for the SOC problem, using a version of the winsorized BW model. Extensive numerical experiments demonstrate the superior performance of the proposed approach.
- 1 April 2023
- Published by SAGE Publications in Production and Operations Management
- Vol. 32 (4) , 1205-1222
Open AccessFrom share of choice to buyers' welfare maximization: Bridging the gap through distributionally robust optimization
We study the product design problem where a decision maker selects the features of a product from a set of feasible options. We focus on two widely studied objectives in this field, that is, the share of choice (SOC) and buyers' welfare (BW). The two objectives are vulnerable to different types of customer preference misspecification, that is, deviation from the nominal utility distribution and effects of outliers, respectively. We formulate a distributionally robust optimization (DRO) SOC maximization model and a winsorized BW maximization model to obtain robust solutions to the two problems. Interestingly, we show that the two robust models are equivalent in a certain sense—for appropriate choices of robustness parameters, both models return the same solution. This observation has important ramifications. For instance, it indicates that a product designed to yield higher BW is more robust to the deviation from the nominal utility distribution for the SOC problem, while that with higher SOC is less sensitive to the effect of outliers for the BW problem. Last but not least, we use the equivalence to develop a new approach to solve the DRO model for the SOC problem, using a version of the winsorized BW model. Extensive numerical experiments demonstrate the superior performance of the proposed approach.