The capacity market is a marketplace for trading generation capacity, and is viewed by its proponents as a mechanism to ensure power system reliability. Based on practice at NYISO, we propose optimization models for capacity markets and analyze outcomes. We find that with capacity markets, more generators are profitable, especially the ones with a lower net cost of new entry. Also, it is possible for a generator to earn more by exercising market power. We conduct case studies on NYISO dataset.

The seller wants to sell an item to n i.i.d. buyers and only the support [a,b] are known; a/b quantifies relative support information (RSI). The seller either minimizes worst-case regret or maximizes worst-case approximation ratio. We show that i) with low RSI, second-price auctions (SPA) is optimal; ii) with high RSI, SPA is not optimal, and we introduce a new mechanism, the "pooling auction" (POOL), which is optimal; iii) with moderate RSI, a combination of SPA and POOL is optimal.

We study resale platforms, a growing type of online marketplaces in developing countries. These platforms allow their users to earn profits by selling products to their contacts, who do not typically shop online. We analyze data from a major resale platform in India and leverage our empirical findings to develop a model of the platform. We provide insights into key design aspects of the platform, such as the structure that determines resellers’ margins and the product ranking algorithm.

In traditional assortment optimization, profits may be improved by making some products unavailable as part of the lost demand flows to higher margin products. The refined assortment optimization problem takes a subtler approach to improving profits without necessarily making some unavailable. This is done by judiciously making some products less appealing thereby steering demand to more profitable products. We develop a theoretical framework and provide tight revenue bounds.

We study efficient real-time data collection approaches for an online retailer that dynamically personalizes the assortment offerings based on customers’ attributes to learn their preferences and maximize revenue. We study the structure of efficient exploration in this setting, prove a performance lower bound, and propose efficient learning policies. We illustrate the superior performance of our policies over existing benchmark using a dataset from a large Chilean retailer.

We investigate assortment problem under distributionally robust and robust satisficing formulations. We generalize the optimality of revenue-ordered set for both distributionally robust assortment and robust assortment satisficing problems. We further discuss the choice model based on worst-case distribution and derive managerial insights by theoretical analysis. Moreover, we discuss the two robust formulations of assortment problem with cardinality constraint and develop efficient algorithms.

We consider an assortment optimization problem where the customer may purchase multiple products and possibly more than one unit of each product. We adopt the customer consumption model based on the multiple-discrete-choice (MDC) model. We present an algorithmic framework that delivers near-optimal algorithms for different variations of the assortment problem.

Essentially every choice involves an information collection or search phase prior to a decision-making phase. To study how these phases interact, we embed the Exponomial Choice model (Alptekinoglu and Semple 2016) in a classical model of sequential search with perfect recall (Weitzman 1979). We derive the search path and final choice probabilities in closed form and develop all the analytical tools to enable joint optimization of assortment and prices efficiently.

Online discussion platforms facilitate remote discussions between users. This paper explores the impact of consensus on engagement and proposes algorithms to optimize rankings. Consensus is identified as a crucial engagement driver, and our proposed algorithm outperformed current approaches in an experiment. Promoting debate over echo chambers, consensus is essential for user engagement and platform design.

In this work, we study the product replacement problem. This problem arises in the fast fashion industry, where the company's headquarters frequently send new items to stores, expecting store managers to remove current items to make room for the new items and display them. Our study addresses a general case, including misaligned incentives among decision-makers and cross-item effects. We present optimal strategies for the headquarters, but also investigate different store managers' behaviors.

We study the impact of wait time information distortion on customer join decision and the service provider’s revenue in virtual queues. We show that the optimal revenue is first increasing and then decreasing in the distortion level. A zero distortion is optimal only when the service capacity is sufficient and the service value is high. Customer joining rate could be either increasing or decreasing in the distortion level.

TBD

We study stochastic online resource allocation: a decision maker needs to allocate limited resources to i.i.d. sequential requests generated from an unknown distribution in order to maximize reward. Online resource allocation has been studied extensively in the past, but prior results crucially rely on the assumption that the total number of requests (the horizon) is known to the decision maker in advance. In this work, we develop online algorithms that are robust to horizon uncertainty.

We consider an online advance reservation problem under customer heterogeneity. The decision maker decides on admitting a customer based on the customer's class and the current system information. Our problem model captures applications in medical appointment and hotel room booking, where a resources are to be scheduled for customers' uses in advance. We derive a novel rejection price based algorithm, and we quantify the theoretical guarantee of our algorithm in terms of its competitive ratio.

TBD

Our paper studies how to determine the personalized incentive for each customer to learn her true preference in the assortment optimization and product design problem. We provide algorithms that sequentially decide the personalized incentives based on the evaluated customer's contribution to the revenue increase. We show that compared to the naive supervised learning algorithm that provides fixed incentives, our algorithm can reduce the total cost significantly while achieving high revenue.

We introduce the concept of randomization into the robust assortment optimization liter-ature. We show that the firm can improve its worst-case expected revenues by selecting an assortment randomly according to a prudently designed probability distribution. We demonstrate this potential benefit of randomization both theoretically in an abstract problem formulation and empirically across three choice models: the multinomial logit model, the Markov chain model, and the preference ranking model.

We study dynamic joint assortment and pricing over a finite time period. The goal is to maximize the expected cumulative revenue. We propose a new doubly high-dimensional contextual bandit model to formulate the problem. We developed a computationally tractable algorithm and provide a convergence rate for it. The numerical results of applying our method to a real-world online retail data set demonstrate the efficiency of our method, which is further supported by extensive simulations.

We study privacy protection mechanisms inspired by recent regulatory regimes, limited data retention and customer self protection. Privacy protection affects the estimation of demand model and thus the charged prices. We find that the change of the resulting price and which customer groups benefit from the protection depend on the product type. A real dataset of online auto loans validates our theoretical findings. We also extend the framework to nonlinear demand functions and duopoly.

Personalized recommender systems face privacy concerns. We explore optimal design under local differential privacy constraints. Our findings suggest a coarse-grained threshold policy for recommendations. Pursuing privacy comes at an economic loss but may benefit consumers. Our study guides algorithm design and informs regulators on privacy policies.

TBD

We study an extension of the cardinality-constrained knapsack problem where each item has a concave piecewise-linear utility structure. For the offline problem, we present a fully polynomial-time approximation scheme and show that it can be cast as the maximization of a submodular function with cardinality constraints; the latter result allows us to derive a greedy (1 − 1/e)-approximation algorithm. For the online problem in the random order model, we present a 6.027-competitive algorithm.

TBD

TBD

We study an adversarial online assortment problem with reusable resources and customer-dependent usage durations. We propose a novel online algorithm which features rejection durations filtering out unprofitable products. We show that it achieves a competitive ratio within a constant factor from the best possible one with large capacities. This is the first work that derives a non-trivial performance guarantee for such problem. We further extend our algorithm framework to other reward functions.

We consider an assortment optimization problem for a class of online video games. Our paper is the first to study assortment optimization for the gaming industry under discrete choice models; it is also the first to devise solution approaches for the constrained mixture-of-nested-logit model with performance guarantees. Numerical experiments show that our approaches perform well across a variety of settings. Our work provides guidance to online video game stores for effective revenue maximization.

We study a new class of stock-dependent pricing policies for reusable resources that set prices based on the stock at hand. To find the optimal policy in this class, we introduce a reformulation that is convex. We provide a sharp characterization of the regret of this policy class via matching upper and lower bounds and show they can significantly improve upon static pricing. A simple two-price policy that changes prices when the stock is below a threshold can achieve the optimal rate of regret.

We consider the problem of optimal assortment design for a platform by imposing fairness constraints that guarantees exposures of all products to consumers. We show that optimal solution can be found in polynomial time and the algorithm is easy to implement in practice. We investigate the welfare change of imposing the fairness constraints and find that it is possible to achieve a win-win-win solution where the platform, sellers, and consumers are all better off.

Problem: Allocating perishable resources online with non-linear utilities and complementarities. Methodology/Results: An algorithm achieves optimal envy-efficiency Pareto frontier by adapting to perishing order predictions and desired envy bounds. Numerical performance is demonstrated using data from a food bank. Managerial Implications: Accurate perishing predictions are crucial for fairness in practical non-profit settings.

We study the marketplace platform's problem that decides the assortment of sellers and how much fee to apply to each seller. In our model, each seller has its own outside option, which depends on customer traffic and product preference. The seller's compensation is determined based on the competition with the other joining sellers. We characterize the platform's optimal fee and assortment decisions and further study how the different fee contracts affect the optimal assortment.