Informs Annual Meeting 2017

TB20

INFORMS Houston – 2017

TB18

2 - The Benefit of Scale: Crowdfunding Risks and Returns Mabel Chou, National University of Singapore, 3 60 Pasir Panjang Rd, #01-11, Singapore, 118699, Singapore, mabelchou@nus.edu.sg, Guodong Lyu, Chung-Piaw Teo, Zhichao Zheng, Yuanguang Zhong With the internet becoming more accessible than ever, entrepreneurs nowadays do not need to ask a few people for big sums of money. Instead, they can use the internet to talk to thousands or even more potential funders that each contribute a small amount. A natural question any potential funder would want to ask is: what are the risks and returns for this investment? To attract more funders, this is a question an entrepreneur needs to address properly. In this talk, we will help the entrepreneurs who are looking for crowdfunding to address this important question based on the new insights obtained from an application of Blackwell’s Approachability Theorem. 3 - Signaling to the Crowd: Private Quality Information and Rewards-based Crowdfunding We consider the problem faced by a seller designing a rewards-based crowdfunding campaign via a platform like Kickstarter. The seller solicits donations from contributors, and if total contributions exceed a pre-determined threshold the campaign is a success, the seller receives all donations and each contributor receives a reward; otherwise, contributors are refunded their donations and the campaign is a failure. We determine how the seller should design her crowdfunding campaign when contributors know less than the seller about the value of the reward. 4 - Financing through Advance Selling: the Role of Social Ties S. Alex Yang, London Business School, Regent’s Park, London, NW1 4SA, United Kingdom, sayang@london.edu, Shuang Xiao, Yiangos Papanastasiou Advance selling (the practice of selling a product in advance of its consumption date) has received significant attention in the recent operations literature, and has shown to be beneficial in a variety of industries as a tool for implementing price discrimination, reducing demand uncertainty, and managing capacity more effectively. More recently, start-up companies have been exploring the use of advance selling as a mechanism to help capital-constrained firms secure bank financing. This paper conducts a theoretical investigation into the efficacy of this alternative financing approach. 342B Topics on Service Design Sponsored: Technology, Innovation Management & Entrepreneurship Sponsored Session Chair: Ioannis Bellos, George Mason University, Fairfax, VA, 22030, United States, ibellos@gmu.edu 1 - First Ranked First to Serve: Strategic Agents in Service Contest Konstantinos Stouras, University of Virginia, Charlottesville, VA, United States, StourasK@darden.virginia.edu, Serguei Netessine, Karan Girotra Work-from-home contact centers crowdsource demand to a pool of freelance agents, who are ranked in a predetermined number of priority classes based on their sales performance. The agents endogenously decide whether to participate and provide service on-demand. Agents’ idle time is not compensated and higher ranked agents are utilized more and earn more. In a markovian queueing model with ex ante random capacity, we show that a coarse partition with two priority classes provide better incentives for agents to participate, maximize firm’s profit and asymptotically maximize social welfare. 2 - Balancing Advisory and Service Delivery Efforts in a Nonprofit Organization Priyank Arora, Georgia Institute of Technology, Atlanta, GA, United States, Priyank.Arora@scheller.gatech.edu, Morvarid Rahmani, Karthik Ramachandran This paper studies how a nonprofit organization (NPO), which aims to maximize overall utility delivered to its clients, should balance levels of effort between advisory and service delivery stages. Soudipta Chakraborty, Duke University, 100 Fuqua Drive, Durham, NC, 27708, United States, sc390@duke.edu, Robert Swinney TB20

340B A/B Testing and Experimentation Sponsored: Applied Probability Sponsored Session Chair: Ramesh Johari, Stanford University, Stanford, CA, 94305-4121, United States, ramesh.johari@stanford.edu Co-Chair: David Walsh, Stanford University, San Francisco, CA, 94117, United States, dwalsh@stanford.edu 1 - Should A/B Testers Just be Bayesian? David Walsh, Stanford University, 876 Shotwell Street, San Francisco, CA, 94110, United States, dwalsh@stanford.edu Any large internet company runs thousands of A/B tests each week, because they provide objective feedback on which ideas work. Progress is incremental, so there are usually strong connections between new experiments and those run before. However, the default Frequentist approaches analyze each A/B test in isolation. A Bayesian framework can easily leverage such prior data, enabling more accurate results with shorter experiments. But we examine how it can also be vulnerable to strategic behavior by the experimenter. This undermines the objective feedback loop and enables the A/B tester simply to get the answer they want. 2 - Improving the Expected Improvement Algorithm Russo Dan, dan.joseph.russo@gmail.com The expected improvement (EI) algorithm is a popular strategy for information collection in optimization under uncertainty. To provide rigorous insight into EI, we study its properties in a simple setting where the goal is to choose the best among a finite set of options based on noisy measurements of their quality. In this framework, one can show formally that EI is far from optimal. To overcome this shortcoming, we introduce a simple modification of the expected improvement algorithm. Surprisingly, this simple change results in an algorithm that is asymptotically optimal for problems with Gaussian observation noise, and provably outperforms standard EI by an order of magnitude. 3 - Experimentation and Interference in a Marketplace Nicholas Chamandy, Lyft, San Francisco, CA, United States, nickchamandy@gmail.com Simple “random-user” experiment designs fall short in the face of complex dependence structures. These come in the form of large-scale social graphs or, more recently, spatio-temporal network interactions in a two-sided transportation marketplace. Naive designs are susceptible to statistical interference, which can lead to biased estimates of the treatment effect of interest. We consider the implications of interference for the design and analysis of experiments at Lyft. The literature on causal inference for infectious diseases offers some insights; these are briefly discussed. We motivate the use of large-scale simulation to both filter candidate tests and validate design methodology. 4 - A Personalized BDM Mechanism for Efficient Market Intervention Experiments Imanol Arrieta, Stanford University, Stanford, CA, United States, imanol@stanford.edu The BDM mechanism employs a second-price auction against a random bidder to elicit a consumer’s willingness to pay. It has been recently utilized as a treatment assignment mechanism to evaluate policy interventions while simultaneously measuring the demand for the interventions. We develop a personalized extension of this mechanism using modern machine learning algorithms to predict an individual’s willingness to pay and personalize the “random bidder”. We show through simulations and a mock experiment on Amazon Mechanical Turk that our personalized BDM mechanism results in lower costs and provides better balance over covariates. 342A Operations/Innovative Financing Interaction Sponsored: Manufacturing & Service Oper Mgmt, iFORM Sponsored Session Chair: S. Alex Yang, London Business School, London, NW1 4SA, United Kingdom, sayang@london.edu 1 - The Impact of Demand Uncertainty on Business Interruption Insurance Yuan-Mao Kao, Duke University, Durham, NC, United States, yuan.mao.kao@duke.edu, Kevin Shang, N. Bora Keskin We consider a firm that faces uncertainty about demand and possible business interruptions in advance of a selling season. In this setting, we formulate an optimal insurance design problem and study how different types of uncertainty affect the insurance offering as well as the firm’s operational decisions. TB19

309