Runshan Fu, Yan Huang, and Param Vir Singh. 2021. Crowds, Lending, Machine, and Bias. Information Systems Research, (32:1): 72-92. [Runner Up, Best Paper Published in Information Systems Research, 2021]
The first question people may have when deciding whether to use machine learning algorithms for decision-making is whether machines can improve upon humans’ decisions. In this paper, we answer the question in the context of peer-to-peer lending, where investment decisions are traditionally made by a crowd of investors. We carefully design a contraction procedure to perform a fair comparison between the decisions made by the crowd and by the machine in terms of borrower and investor welfare. We show that the machine learning algorithm can simultaneously improve investors’ payoff irrespective of their risk preferences, and provide more lending opportunities to borrowers with fewer alternative funding options. We also find that the machine is biased in gender and race. We propose a general debias method and show that the debiased algorithm can still increase investor and borrower welfare while minimizing the bias.
Qiaochu Wang, Yan Huang, Stefanus Jasin, and Param Vir Singh. Algorithmic Transparency with Strategic Users. Accepted at Management Science. [Best Student Paper Award, CIST 2019]
Recently, there has been an increased demand for algorithmic transparency to combat algorithmic bias and discrimination, and improve accountability and trust in algorithms. However, firms have been reluctant to make their algorithms transparent, arguing that transparency opens the door to user gaming. In this paper, we theoretically study the impact of algorithmic transparency in a hiring context where a firm is using an algorithm to distinguish between more desirable (high-talent) and less desirable (low-talent) job applicants. Most arguments concluding that gaming by strategic users will negatively affect the prediction accuracy of transparent algorithms are based on the assumption that only the less desirable individuals will game. We show that under certain conditions, algorithmic transparency can increase the incentives for the more desirable individuals to improve on the features that are costly to mimic to further differentiate themselves from the less desirable individuals, leading to a higher prediction accuracy.
Qiaochu Wang, Yan Huang, and Param Vir Singh. Algorithmic Lending, Competition, and Strategic Information Disclosure. Under review at Marketing Science.
We investigate how competition among algorithmic lenders affects their decisions to provide approval odds to consumers. We show that competitive pressures between lenders can undermine the disclosure incentives. Lenders use asymmetric disclosure of approval odds strategically to soften the competition when their algorithms are fairly accurate. The asymmetric disclosure of approval odds endogenously creates product differentiation and allows lenders to focus on different segments of consumers softening the competition on the interest rates. We find that consumer surplus is highest when both lenders provide approval odds and lowest when neither provides approval odds. However, our analysis also shows that any policy that mandates all lenders to provide personalized approval odds to consumers may not necessarily improve consumer surplus.
Xiyang Hu, Yan Huang, Beibei Li, and Tian Lu. Human-Algorithmic Bias: Source, Evolution, and Impact. Under review at Management Science. [Finalist, Best Student Paper Award, CIST 2021]
In this study, leveraging a unique repeat decision-making setting in a high-stakes micro-lending context, we aim to uncover the underlying source, evolution dynamics, and associated impacts of bias. We first develop and estimate a structural econometric model of the decision dynamics to understand the source and evolution of potential bias in human evaluators in microloan granting. We find that both preference-based bias and belief-based bias are present in human evaluators' decisions and are in favor of female applicants. Through counterfactual simulations, we quantify the effects of the two types of bias on both fairness and profits. The results show that the elimination of either of the two biases improves the fairness in financial resource allocation, as well as the platform profits. The profit improvement mainly stems from the increase in the approval probability for male borrowers, especially those who would eventually pay back loans. Furthermore, to examine how human biases evolve when being inherited by machine learning (ML) algorithms, we then train a set of state-of-the-art ML algorithms for default risk prediction on both real-world datasets with human biases encoded within and counterfactual datasets with human biases partially or fully removed. By comparing the decision outcomes in different counterfactual settings, we find that even fairness-unaware ML algorithms can reduce bias present in human loan-granting decisions. Interestingly, while removing both types of human biases from the training data can further improve ML fairness, the fairness-enhancing effects vary significantly between new and repeat applicants. Based on our findings, we discuss how to reduce decision bias most effectively in a human-machine learning pipeline.
Yan Huang, Param Vir Singh, and Kannan Srinivasan. 2014. Crowdsourcing New Product Ideas under Consumer Learning. Management Science (60:9): 2138-2159. [Finalist, 2019 INFORMS TIMES Best Paper Award]
In this paper, we empirically study idea contributors’ behavior in crowdsourced ideation initiatives, where individuals post ideas for the firm to implement and vote on other’s ideas, and the firm then decides which ideas to implement. Although popular, crowdsourced ideation faces increasing criticism as the number of ideas generated often declines over time. We show that the downward trend in the number of ideas contributed is in fact a sign of market efficiency rather than failure. We develop and estimate a dynamic structural model that captures the economic mechanisms shaping individuals’ idea contribution behavior. We find that in the initial stages of the crowdsourcing process, individuals tend to underestimate the cost the firm incurs to implement their ideas, but overestimate the potential of their ideas. Hence, the idea market is initially overcrowded with low-potential ideas. However, individuals learn about their abilities to come up with high-potential ideas from peer voting, and learn about the firm’s cost structure from its responses to all consumer-generated ideas. Over time, contributors of low-potential ideas eventually drop out, while contributors of high-potential ideas remain active, and the crowdsourcing market becomes more efficient.
Yan Huang, Param Vir Singh, and Anindya Ghose. 2015. A Structural Model of Employee Behavioral Dynamics in Enterprise Social Media. Management Science (61:12): 2825-2844. [Finalist, Best Paper Published in the IS Department of Management Science in 2014-2016]
This paper empirically examines employee behavioral dynamics in enterprise blogs, which have been widely adopted by firms as new platforms for employees to freely share knowledge and expertise. Firms that adopt enterprise blogs often find excessive leisure-related posts on their blogging platform, and are concerned that such posts will undermine the very goal of enterprise blogging. We analyze employees’ social media content creation and consumption behavior on the enterprise blog site of a large IT service firm, and find a significant positive readership spillover from leisure-related posts to work-related posts. Therefore, a policy that prohibits leisure-related activities will have a detrimental effect on employees’ work-related content creation and consumption.
Zhaohui (Zoey) Jiang, Yan Huang, and Damian R. Beil. The Role of Feedback in Dynamic Crowdsourcing Contests: A Structural Empirical Analysis. Accepted at Management Science. [Finalist, IBM Best Student Paper Award for INFORMS Service Science, 2019]
This paper empirically investigates the impact of performance feedback (e.g., star-ratings) on the outcome of crowdsourcing contests, and finds that, contrary to the common belief that frequent and timely performance feedback is helpful in improving the contest outcome, delaying feedback until the later stage of the contest is optimal. We develop a dynamic structural model to capture the economic processes that drive contest participants’ behavior, and estimate the model using a data set collected from a major online crowdsourcing design platform. The model captures key features of crowdsourcing contests, including a large participant pool, entries by new participants throughout the contest, exploitation and exploration behaviors (quantified with a computer vision algorithm) by contest incumbents, and participants’ strategic choice among these entry, exploration, and exploitation decisions in a dynamic game. Using counter-factual simulations, we show that the late feedback policy leads to a better overall contest outcome. This is because performance feedback helps guide creators’ exploration and exploitation decisions, but can have a discouraging effect on entries and incumbents’ follow-up actions. The late feedback policy attains the former benefit while mitigating the latter problem, by only giving feedback after many creators have had a chance to enter.
Zhaohui (Zoey) Jiang, Yan Huang, and Damian R. Beil. 2021. The Role of Problem Specification in Crowdsourcing Contest for Design Problems: A Theoretical and Empirical Analysis. Manufacturing & Service Operations Management (23:3): 637-656.
This paper studies the impact of problem specification in crowdsourcing creative contests. At first sight, one may think seekers should specify their problems in the most thorough manner possible. However, we show that an overly specified problem may backfire, especially when the seeker is not careful about the types of information he/she provides. We develop a game-theoretic model featuring different types of information (categorized as “conceptual objectives” or “execution guidelines”) conveyed in problem specifications. We show theoretically and verify empirically that, with more conceptual objectives disclosed in the problem specification, the number of participants in a contest eventually decreases; with more execution guidelines in the problem specification, the trial effort provision by each participant increases; and the best solution quality always increases with more execution guidelines, but eventually decreases with more conceptual objectives. Therefore, to maximize the best solution quality in crowdsourced design problems, seekers should always provide more execution guidelines, but only a moderate number of conceptual objectives.
He Huang, Yan Huang, Zhijun Yan, and Han Zhang. Join to Sell? An Empirical Analysis of Seller Participation within an Online Social Network. Accepted at Management Information Systems Quarterly.
In this paper, we empirically study how a focal seller’s effort and sales performance are affected by the effort and sales performance of other sellers she is connected to (i.e., inviter and invitees) and the commissions she has received. We find evidence for social influence, competition, and free riding among connected sellers: Inviter effort has a positive impact on the focal seller’s effort, indicating social influence; inviter performance has a negative impact on the focal seller’s effort and performance, indicating competition. Interestingly, invitee effort/performance has no significant effect on the focal seller. Therefore, social influence and competition among connected sellers are one-directional and happen in the “top-down” manner. We also find that higher commissions from invitees’ sales performance reduce the focal seller’s effort, indicating free riding.
Ajit Sharma, Yan Huang, and M.S. Krishnan. Learning in a Disruptive Tablet-Based Customer Engagement Platform: An Empirical Analysis in the Banking Industry.
In this paper, we empirically study the learning dynamics of sales officers when opening bank accounts using a tablet-based banking application at a large private bank in an emerging market. Our results show that immediately after sales officers switch from the traditional system to the tablet-based system, account opening efficiency decreases. As sales officers learn to use the tablet-based system, the account opening efficiency improves and stabilizes at a level higher than the level under the traditional system. Our results also reveal that although initially, high performers in the traditional system continue to maintain their edge in the new mobile platform, low performers learn faster, and the gap between high and low performers reduces over time. Surprisingly, neither a higher degree nor an educational background in science or technology results in better performance or faster learning in the new tablet-based system. Our findings have important managerial implications for managing the transition to disruptive mobile customer engagement platforms and policy implications for technology related education and employment in emerging-markets.
Brett Danaher, Yan Huang, Michael D. Smith, and Rahul Telang. 2014. An Empirical Analysis of Digital Music Bundling Strategies. Management Science (60:6): 1413-1433.
This paper examines the pricing and bundling strategies in digital music markets where both single songs and albums are available to consumers. We collaborate with a major music label to conduct a pricing experiment. We use data from this pricing experiment to empirically estimate a structural model of consumer choices, and then evaluate welfare under various policy-relevant counterfactual scenarios. Our results show that tiered pricing (setting different prices for songs in different popularity tiers) coupled with reduced album pricing increases revenue while also increasing consumer surplus.
Yan Huang, Stefanus Jasin, and Puneet Manchanda. 2019. “Level Up”: Leveraging Skill and Engagement to Maximize Player Retention in Online Video Games. Information Systems Research (30:3): 927-947. [INFORMS eBusiness Best Paper Award, 2017]
In this paper, we investigate the evolution of online video game players’ (gamers’) latent engagement state and its implications for gamer matching strategies. We first build a Hidden Markov Model (HMM) to capture the evolution of gamers’ latent engagement as a function of their game-play outcome, and estimate the HMM using a longitudinal data set obtained from a major international video gaming company, which contains detailed game-play history of a large random sample of gamers. The estimation results show that high-, medium- and low-engagement-state gamers respond differently to motivational factors such as achievement and challenge. We use the estimated HMM to develop a matching algorithm that learns each gamer’s current engagement state and exploits that learning to match each gamer to a game round that maximizes his/her game-play volume. We show that our matching algorithm increases game-play, leading to economically significant revenue gains for the company.
Manqi (Maggie) Li, Yan Huang, and Amitabh Sinha. 2020. Data-Driven Promotion Planning for Paid Mobile Applications. Information Systems Research (31:3): 1007-1029.
In this paper, we empirically estimate a demand model for paid mobile apps and quantify the effect of price promotions on download volume. The estimation results reveal two interesting characteristics of the demand of paid apps: (1) the magnitude of the direct promotion effect changes with time within a multi-day promotion; and (2) due to the visibility effect (i.e., apps ranked high on the download chart are more visible to consumers), a price promotion also has an indirect effect on download volume through affecting app rank, and this effect can persist after the promotion ends. Based on the empirically estimated demand model, we propose a promotion planning heuristic and show that the proposed policy can increase the app lifetime revenue by around 10%.
Manqi (Maggie) Li, Xiang Liu, Yan Huang, and Cong Shi. Integrating Empirical Estimation and Assortment Personalization for E-Commerce: A Consider-then-Choose Model. [Finalist, MSOM Data Driven Research Challenge, 2018]
This paper proposes a framework for online retailers to determine the optimal search result set, which involves estimating a structural model of consumer choices and solving a “virtual” assortment optimization problem. We first estimate a two-stage Multinomial Logit (MNL) based consider-then-choose model that captures how consumers form their consideration set and make purchase decisions. To maximize the expected revenue, we compute the optimal target assortment set based on each consumer’s taste. Then we adjust the display of items to induce this consumer to form her consideration set that coincides with the target assortment set. We formulate this consideration set induction process as a nonconvex optimization, for which we provide the sufficient and necessary condition for feasibility. We provide a simple closed-form relationship between the viewing cost and the number of items a consumer is willing to consider. To mitigate computational difficulties associated with nonconvexity, we develop an efficient heuristic to induce the optimal consideration set. We test the heuristic and show that it yields near-optimal solutions.
ARTICLES IN EDITED BOOKS/VOLUMES:
Runshan Fu, Yan Huang, and Param Vir Singh. 2020 Artificial Intelligence and Algorithmic Bias: Source, Detection, Mitigation and Implications. INFORMS Tutorials in Operations Research, INFORMS.
Yan Huang, Stephanie Lee, and Yong Tan. 2019. Structural Econometric Models. INFORMS Tutorials in Operations Research, INFORMS.
SELECTED WORK IN PROGRESS
"Privacy discrimination under GDPR," with Qiaochu Wang and Param Vir Singh.
"The Where, When, and Why of Movie Piracy," with Helen (Shuxuan) Zeng, Gordon Burtch, and Michael Smith.
"Product Rankings, AI Pricing Algorithms and Collusion," with Liying Qiu and Param Vir Singh.
"Algorithms, Artificial Intelligence, and Simple Rule Based Pricing Algorithms," with Qiaochu Wang and Param Vir Singh.
"Credit Risk Modeling without Sensitive Features: An Adversarial Deep Learning Model for Fairness and Profit," with Xiyang Hu, Beibei Li and Tian Lu.