Zifeng Zhao
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
Zifeng Zhao has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Feiyu Jiang
Fudan University
5 shared publications - Ruxian Wang
Johns Hopkins Carey Business School, Johns Hopkins University, Washington, District of Columbia 20001
3 shared publications - Chenxu Ke
College of Business, Shanghai University of Finance and Economics, Shanghai 200433, China
2 shared publications Chun Yip Yau
Chinese University of Hong Kong
2 shared publicationsKenneth D. West
Department of Economics, University of Wisconsin-Madison, Madison, WI, United States
2 shared publicationsPeng Shi
Risk and Insurance Department, Wisconsin School of Business, University of Wisconsin-Madison, Madison, Wisconsin 53706
2 shared publications
Latest Publications
- 21 August 2026
- Published by Taylor & Francis in Journal of the American Statistical Association
- p. 1-104
Locally Private Nonparametric Contextual Multi-armed Bandits with Transfer Learning
1 IntroductionContextual multi-armed bandit (MAB) (e.g. Lu et al., 2010; Zhou, 2016) is a versatile and general framework for sequential decision-making. It has been widely deployed in various prac...
- 21 August 2026
- Published by Taylor & Francis in Journal of the American Statistical Association
- p. 1-104
Locally Private Nonparametric Contextual Multi-armed Bandits with Transfer Learning
1 IntroductionContextual multi-armed bandit (MAB) (e.g. Lu et al., 2010; Zhou, 2016) is a versatile and general framework for sequential decision-making. It has been widely deployed in various prac...
- 1 July 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 28 (4) , 1112-1132
Data-Driven Pricing for Availability-Based Upgrades Under a Multiple Binary Choice Model with Copula
Problem definition: Intense competition in the travel industry has increasingly shifted focus toward ancillary services, particularly seat upgrades in airlines and room upgrades in hotels. In response to this trend, several innovative solutions have emerged, among which Nor1’s eStandby Upgrade program stands out by offering discounted, availability-based room upgrades. Revenue management for these upgrades is complex because customers may request multiple upgrades, whereas hotels allocate them based on availability and typically grant at most one upgrade per customer. Methodology/results: Partnering with Oracle, which acquired Nor1, we develop a state-of-the-art framework for prediction, pricing, and allocation to maximize total revenue from eStandby upgrades. We first model customer decision making using a novel copula-based multivariate choice model that captures complex dependencies among multiple decisions made by the same customer. Next, we develop efficient pricing and allocation algorithms to address the challenges associated with offering multiple availability-based upgrades and tracking customer requests. Managerial implications: Validated with real-world data and data-driven numerical experiments, our choice model for upgrade requests and algorithms for pricing and allocation demonstrate significant revenue potential by capturing dependencies across customers’ multiple decisions. History: This paper has been accepted as part of the 2025 Manufacturing & Service Operations Management Practice-Based Research Competition. Funding: This work was supported by Oracle Research [Grant CR 4223]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2025.0328 .
- 1 July 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 28 (4) , 1112-1132
Data-Driven Pricing for Availability-Based Upgrades Under a Multiple Binary Choice Model with Copula
Problem definition: Intense competition in the travel industry has increasingly shifted focus toward ancillary services, particularly seat upgrades in airlines and room upgrades in hotels. In response to this trend, several innovative solutions have emerged, among which Nor1’s eStandby Upgrade program stands out by offering discounted, availability-based room upgrades. Revenue management for these upgrades is complex because customers may request multiple upgrades, whereas hotels allocate them based on availability and typically grant at most one upgrade per customer. Methodology/results: Partnering with Oracle, which acquired Nor1, we develop a state-of-the-art framework for prediction, pricing, and allocation to maximize total revenue from eStandby upgrades. We first model customer decision making using a novel copula-based multivariate choice model that captures complex dependencies among multiple decisions made by the same customer. Next, we develop efficient pricing and allocation algorithms to address the challenges associated with offering multiple availability-based upgrades and tracking customer requests. Managerial implications: Validated with real-world data and data-driven numerical experiments, our choice model for upgrade requests and algorithms for pricing and allocation demonstrate significant revenue potential by capturing dependencies across customers’ multiple decisions. History: This paper has been accepted as part of the 2025 Manufacturing & Service Operations Management Practice-Based Research Competition. Funding: This work was supported by Oracle Research [Grant CR 4223]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2025.0328 .
- 1 July 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 28 (4) , 1266-1285
Discrete Choice Models with Piecewise Linear Utility: Modeling, Estimation, and Pricing
Problem definition: This paper incorporates a piecewise linear structure into the utility-price relationship of the classic multinomial logit (MNL) model and studies the associated operations problems such as estimation and pricing. Methodology/results: The derived model is referred to as the piecewise MNL model. We study its model identification and further propose a maximum likelihood estimator (MLE) for its calibration by real data. Because of the presence of inflection points, the log-likelihood function is nondifferentiable, which poses major challenges to both numerical and statistical analyses. We then propose a novel profile-based numerical optimization procedure that locates the MLE efficiently and further establish statistical guarantees for the MLE based on the empirical process theory. We fully characterize the price optimization problem under the piecewise MNL and show that the optimal pricing policy can be quite different from the standard MNL. In particular, the equal-(adjusted-)markup policy is no longer optimal, and the optimal solution may not even be unique. Furthermore, we propose an efficient approximation algorithm for the pricing problem that accounts for consumer heterogeneity within the piecewise MNL framework. Managerial implications: Our extensive numerical experiments on synthetic and real data suggest that the piecewise MNL can provide greater modeling flexibility, improve model fitting and prediction accuracy compared with popular choice models in the literature, and ignoring varying price sensitivities could lead to suboptimal solutions or even substantial losses for firms. Funding: C. Ke acknowledges financial support from the National Natural Science Foundation of China [Grant 72101113]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2023.0169 .
- 1 July 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 28 (4) , 1266-1285
Discrete Choice Models with Piecewise Linear Utility: Modeling, Estimation, and Pricing
Problem definition: This paper incorporates a piecewise linear structure into the utility-price relationship of the classic multinomial logit (MNL) model and studies the associated operations problems such as estimation and pricing. Methodology/results: The derived model is referred to as the piecewise MNL model. We study its model identification and further propose a maximum likelihood estimator (MLE) for its calibration by real data. Because of the presence of inflection points, the log-likelihood function is nondifferentiable, which poses major challenges to both numerical and statistical analyses. We then propose a novel profile-based numerical optimization procedure that locates the MLE efficiently and further establish statistical guarantees for the MLE based on the empirical process theory. We fully characterize the price optimization problem under the piecewise MNL and show that the optimal pricing policy can be quite different from the standard MNL. In particular, the equal-(adjusted-)markup policy is no longer optimal, and the optimal solution may not even be unique. Furthermore, we propose an efficient approximation algorithm for the pricing problem that accounts for consumer heterogeneity within the piecewise MNL framework. Managerial implications: Our extensive numerical experiments on synthetic and real data suggest that the piecewise MNL can provide greater modeling flexibility, improve model fitting and prediction accuracy compared with popular choice models in the literature, and ignoring varying price sensitivities could lead to suboptimal solutions or even substantial losses for firms. Funding: C. Ke acknowledges financial support from the National Natural Science Foundation of China [Grant 72101113]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2023.0169 .
- Open Access
- 30 March 2026
- Published by Taylor & Francis in Journal of the American Statistical Association
- p. 1-16
Contextual Dynamic Pricing: Algorithms, Optimality, and Local Differential Privacy Constraints
- 30 March 2026
- Published by Taylor & Francis in Journal of the American Statistical Association
- p. 1-16
Open AccessContextual Dynamic Pricing: Algorithms, Optimality, and Local Differential Privacy Constraints
- Open Access
Learning Personalized Ad Impact via Contextual Reinforcement Learning under Delayed Rewards
Online advertising platforms use automated auctions to connect advertisers with potential customers, requiring effective bidding strategies to maximize profits. Accurate ad impact estimation requires considering three key factors: delayed and long-term effects, cumulative ad impacts such as reinforcement or fatigue, and customer heterogeneity. However, these effects are often not jointly addressed in previous studies. To capture these factors, we model ad bidding as a Contextual Markov Decision Process (CMDP) with delayed Poisson rewards. For efficient estimation, we propose a two-stage maximum likelihood estimator combined with data-splitting strategies, ensuring controlled estimation error based on the first-stage estimator's (in)accuracy. Building on this, we design a reinforcement learning algorithm to derive efficient personalized bidding strategies. This approach achieves a near-optimal regret bound of $\tilde{O}{(dH^2\sqrt{T})}$, where $d$ is the contextual dimension, $H$ is the number of rounds, and $T$ is the number of customers. Our theoretical findings are validated by simulation experiments.Open AccessLearning Personalized Ad Impact via Contextual Reinforcement Learning under Delayed Rewards
Online advertising platforms use automated auctions to connect advertisers with potential customers, requiring effective bidding strategies to maximize profits. Accurate ad impact estimation requires considering three key factors: delayed and long-term effects, cumulative ad impacts such as reinforcement or fatigue, and customer heterogeneity. However, these effects are often not jointly addressed in previous studies. To capture these factors, we model ad bidding as a Contextual Markov Decision Process (CMDP) with delayed Poisson rewards. For efficient estimation, we propose a two-stage maximum likelihood estimator combined with data-splitting strategies, ensuring controlled estimation error based on the first-stage estimator's (in)accuracy. Building on this, we design a reinforcement learning algorithm to derive efficient personalized bidding strategies. This approach achieves a near-optimal regret bound of $\tilde{O}{(dH^2\sqrt{T})}$, where $d$ is the contextual dimension, $H$ is the number of rounds, and $T$ is the number of customers. Our theoretical findings are validated by simulation experiments. - Open Access
- 1 September 2024
- Published by The R Foundation in The R Journal
- Vol. 16 (3) , 46-66
SNSeg: An R Package for Time Series Segmentation via Self-Normalization
- 1 September 2024
- Published by The R Foundation in The R Journal
- Vol. 16 (3) , 46-66
Open AccessSNSeg: An R Package for Time Series Segmentation via Self-Normalization
- 1 May 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 26 (3) , 1082-1101
Anticipated Wait and Its Effects on Consumer Choice, Pricing, and Assortment Management
Problem definition: We investigate the effects of waiting time, mainly due to production in a make-to-batch-order (MTBO) system, on consumer choice behavior, pricing, assortment, and model estimation. In an MTBO system, the seller/manufacturer first collects orders placed within a certain period of time into a batch and then starts the production process. After the production of all orders in a batch are complete, the products are then shipped and delivered to individual consumers. Because of batch production and shipping, the disutility of the waiting time exhibits negative externality. Methodology/results: We adopt the widely used multinomial logit (MNL) model as a starting point and incorporate the anticipated wait into consumers’ decision making. The derived model, referred to as the MNL with wait model, is a solution of the rational expectation equilibrium and is capable of capturing the effects of negative externality induced by anticipated wait that may change the substitution patterns dramatically. We characterize the multiproduct price optimization problem under the MNL with wait model by establishing a one-to-one mapping between the price vector and the choice probability vector. We find that firms tend to charge higher prices for time-consuming items and charge lower prices for time-saving items compared with the optimal prices under the standard MNL model. In addition to price competition, we also study the Cournot-type competition, in which the decision is the choice probability for each firm and establish the existence of a Nash equilibrium. For assortment optimization, we identify mild conditions under which the optimality of revenue-ordered assortments still holds. However, the assortment problem under the MNL with wait model is generally NP-hard, so we develop approximation algorithms with performance guarantees and provide an easy-to-compute tight upper bound. Moreover, we develop an efficient maximum likelihood-based algorithm for model calibration and further conduct numerical studies to showcase the importance of incorporating disutility due to wait in estimation, pricing, and assortment planning problems. Managerial implications: The MNL with wait model can increase prediction accuracy for consumers’ choice behavior especially when they are aware of the potential wait. Failure to take into account the effects of anticipated wait in firms’ decision making may lead to substantial losses. Funding: The research of C. Ke is supported in part by the National Natural Science Foundation of China [Grant 72101113]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2020.0346 .
- 1 May 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 26 (3) , 1082-1101
Anticipated Wait and Its Effects on Consumer Choice, Pricing, and Assortment Management
Problem definition: We investigate the effects of waiting time, mainly due to production in a make-to-batch-order (MTBO) system, on consumer choice behavior, pricing, assortment, and model estimation. In an MTBO system, the seller/manufacturer first collects orders placed within a certain period of time into a batch and then starts the production process. After the production of all orders in a batch are complete, the products are then shipped and delivered to individual consumers. Because of batch production and shipping, the disutility of the waiting time exhibits negative externality. Methodology/results: We adopt the widely used multinomial logit (MNL) model as a starting point and incorporate the anticipated wait into consumers’ decision making. The derived model, referred to as the MNL with wait model, is a solution of the rational expectation equilibrium and is capable of capturing the effects of negative externality induced by anticipated wait that may change the substitution patterns dramatically. We characterize the multiproduct price optimization problem under the MNL with wait model by establishing a one-to-one mapping between the price vector and the choice probability vector. We find that firms tend to charge higher prices for time-consuming items and charge lower prices for time-saving items compared with the optimal prices under the standard MNL model. In addition to price competition, we also study the Cournot-type competition, in which the decision is the choice probability for each firm and establish the existence of a Nash equilibrium. For assortment optimization, we identify mild conditions under which the optimality of revenue-ordered assortments still holds. However, the assortment problem under the MNL with wait model is generally NP-hard, so we develop approximation algorithms with performance guarantees and provide an easy-to-compute tight upper bound. Moreover, we develop an efficient maximum likelihood-based algorithm for model calibration and further conduct numerical studies to showcase the importance of incorporating disutility due to wait in estimation, pricing, and assortment planning problems. Managerial implications: The MNL with wait model can increase prediction accuracy for consumers’ choice behavior especially when they are aware of the potential wait. Failure to take into account the effects of anticipated wait in firms’ decision making may lead to substantial losses. Funding: The research of C. Ke is supported in part by the National Natural Science Foundation of China [Grant 72101113]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2020.0346 .
- Open Access
SNSeg: An R Package for Time Series Segmentation via Self-Normalization
Time series segmentation aims to identify potential change-points in a sequence of temporally dependent data, so that the original sequence can be partitioned into several homogeneous subsequences. It is useful for modeling and predicting non-stationary time series and is widely applied in natural and social sciences. Existing segmentation methods primarily focus on only one type of parameter changes such as mean and variance, and they typically depend on laborious tuning or smoothing parameters, which can be challenging to choose in practice. The self-normalization based change-point estimation framework SNCP by Zhao et al. (2022), however, offers users more flexibility and convenience as it allows for change-point estimation of different types of parameters (e.g. mean, variance, quantile and autocovariance) in a unified fashion, and requires effortless tuning. In this paper, the R package SNSeg is introduced to implement SNCP for segmentation of univariate and multivariate time series. An extension of SNCP, named SNHD, is also designed and implemented for change-point estimation in the mean vector of high-dimensional time series. The estimated changepoints as well as segmented time series are available with graphical tools. Detailed examples of SNSeg are given in simulations of multivariate autoregressive processes with change-points.
Open AccessSNSeg: An R Package for Time Series Segmentation via Self-Normalization
Time series segmentation aims to identify potential change-points in a sequence of temporally dependent data, so that the original sequence can be partitioned into several homogeneous subsequences. It is useful for modeling and predicting non-stationary time series and is widely applied in natural and social sciences. Existing segmentation methods primarily focus on only one type of parameter changes such as mean and variance, and they typically depend on laborious tuning or smoothing parameters, which can be challenging to choose in practice. The self-normalization based change-point estimation framework SNCP by Zhao et al. (2022), however, offers users more flexibility and convenience as it allows for change-point estimation of different types of parameters (e.g. mean, variance, quantile and autocovariance) in a unified fashion, and requires effortless tuning. In this paper, the R package SNSeg is introduced to implement SNCP for segmentation of univariate and multivariate time series. An extension of SNCP, named SNHD, is also designed and implemented for change-point estimation in the mean vector of high-dimensional time series. The estimated changepoints as well as segmented time series are available with graphical tools. Detailed examples of SNSeg are given in simulations of multivariate autoregressive processes with change-points.
- 2 February 2024
- Published by Taylor & Francis in Journal of the American Statistical Association
- Vol. 119 (548) , 3086-3100
A Composite Likelihood-Based Approach for Change-Point Detection in Spatio-Temporal Processes
- 2 February 2024
- Published by Taylor & Francis in Journal of the American Statistical Association
- Vol. 119 (548) , 3086-3100
A Composite Likelihood-Based Approach for Change-Point Detection in Spatio-Temporal Processes
- Open Access
High-Dimensional Dynamic Pricing under Non-Stationarity: Learning and Earning with Change-Point Detection
We consider a high-dimensional dynamic pricing problem under non-stationarity, where a firm sells products to $T$ sequentially arriving consumers that behave according to an unknown demand model with potential changes at unknown times. The demand model is assumed to be a high-dimensional generalized linear model (GLM), allowing for a feature vector in $\mathbb R^d$ that encodes products and consumer information. To achieve optimal revenue (i.e., least regret), the firm needs to learn and exploit the unknown GLMs while monitoring for potential change-points. To tackle such a problem, we first design a novel penalized likelihood-based online change-point detection algorithm for high-dimensional GLMs, which is the first algorithm in the change-point literature that achieves optimal minimax localization error rate for high-dimensional GLMs. A change-point detection assisted dynamic pricing (CPDP) policy is further proposed and achieves a near-optimal regret of order $O(s\sqrt{Υ_T T}\log(Td))$, where $s$ is the sparsity level and $Υ_T$ is the number of change-points. This regret is accompanied with a minimax lower bound, demonstrating the optimality of CPDP (up to logarithmic factors). In particular, the optimality with respect to $Υ_T$ is seen for the first time in the dynamic pricing literature, and is achieved via a novel accelerated exploration mechanism. Extensive simulation experiments and a real data application on online lending illustrate the efficiency of the proposed policy and the importance and practical value of handling non-stationarity in dynamic pricing.Open AccessHigh-Dimensional Dynamic Pricing under Non-Stationarity: Learning and Earning with Change-Point Detection
We consider a high-dimensional dynamic pricing problem under non-stationarity, where a firm sells products to $T$ sequentially arriving consumers that behave according to an unknown demand model with potential changes at unknown times. The demand model is assumed to be a high-dimensional generalized linear model (GLM), allowing for a feature vector in $\mathbb R^d$ that encodes products and consumer information. To achieve optimal revenue (i.e., least regret), the firm needs to learn and exploit the unknown GLMs while monitoring for potential change-points. To tackle such a problem, we first design a novel penalized likelihood-based online change-point detection algorithm for high-dimensional GLMs, which is the first algorithm in the change-point literature that achieves optimal minimax localization error rate for high-dimensional GLMs. A change-point detection assisted dynamic pricing (CPDP) policy is further proposed and achieves a near-optimal regret of order $O(s\sqrt{Υ_T T}\log(Td))$, where $s$ is the sparsity level and $Υ_T$ is the number of change-points. This regret is accompanied with a minimax lower bound, demonstrating the optimality of CPDP (up to logarithmic factors). In particular, the optimality with respect to $Υ_T$ is seen for the first time in the dynamic pricing literature, and is achieved via a novel accelerated exploration mechanism. Extensive simulation experiments and a real data application on online lending illustrate the efficiency of the proposed policy and the importance and practical value of handling non-stationarity in dynamic pricing.