1 May 2022 in Management Science
By providing quick and easy access to credit, online lending platforms may help borrowers overcome financial setbacks and/or refinance high-interest debt, thereby decreasing bankruptcy filings. On the other hand, these platforms may cause borrowers to overextend themselves financially, leading to a “debt trap” and increasing bankruptcy filings. To investigate the impact of online lending on bankruptcy filings, we leverage variation in when state regulators granted approval for a major online lending platform—Lending Club—to issue peer-to-peer loans. Using a difference-in-differences approach, we find that state approval of Lending Club led to an increase in bankruptcy filings. A complementary instrumental variables analysis using loan-level data yields similar results. We find suggestive evidence that the ease of receiving a Lending Club loan causes some borrowers to overextend themselves financially, leading to bankruptcy. Our results suggest that recent initiatives from online lending platforms to control how borrowers use loans, such as Lending Club’s “balance transfer loans” that send loan funds directly to creditors, can help these platforms provide safe and affordable credit. Our study adds to the literature that examines how online platforms influence society and the economy; it contributes to the literature that examines how financial products, services, and regulations influence bankruptcy filings; and it has policy implications for online lending design and regulation. This paper was accepted by Lorin Hitt, information systems.
1 July 2024 in Management Science
Matching makes or breaks peer-to-peer (P2P) platforms. As the platform-based P2P markets shift from their grassroots nature toward elite offerings, will differentiating suppliers by quality increase matching performance? Can P2P markets differentiate certain suppliers without marginalizing others? We seek answers to these questions by leveraging an empirical opportunity on Airbnb, which differentiates listings that meet high quality standards from others through its Plus program in several U.S. cities. Our findings are threefold. First, we find a sizable increase in market-level matching performance after the program launch, that is, a 15.8%–16.2% increase in the number of booked nights. Second, the rise in matching performance is attributed to reduced search frictions, evident in markets with higher discovery and evaluation costs. Third, the Plus program benefits all listing tiers through reduced search frictions, yielding the most benefits to the Plus and non-Plus listings in terms of increased bookings and prices. Regular listings also receive increased bookings because of competitive prices. Our findings hold across multiple robustness checks and offer important insights for platform design and supply management in P2P markets. This paper was accepted by Hemant Bhargava, information systems. Supplemental Material: The web appendix and data are available at https://doi.org/10.1287/mnsc.2020.03920 .
1 January 2026 in Management Science
Human–algorithm collaboration is becoming increasingly prevalent in the economy and society. However, this collaboration is not always fruitful, and in extreme cases, people become human borgs or totally averse to algorithms. The key to collaborative value is whether humans and algorithms can complement each other in decision making, but it is challenging for humans to disagree with algorithmic recommendations at the right time (i.e., to disagree when algorithms are wrong and not disagree when algorithms are right). To understand the centric role of disagreement in human–algorithm collaboration and examine when and how it benefits, we conducted a field experiment in which human evaluators and algorithms worked together to evaluate loan applications under four scenarios, that is, limited/rich information and with/without disclosure of algorithm rationale. Our results show that human–algorithm collaboration decisions outperformed human- or algorithm-only decisions, and these collaborative values varied across the four scenarios. We further propose a theoretical framework for mechanism examination that centers on the formation and effectiveness of disagreement. We validate the framework empirically and come to the following findings: (1) disagreement exhibits a sizable and nonlinear influence on collaborative value, (2) the differences between human evaluators and algorithms in decision making contribute to disagreement but not to collaborative value, (3) algorithm self-contradiction triggers disagreement and helps human evaluators disagree with algorithms at the right time. These findings provide valuable theoretical insights on how collaborative value is achieved and managerial insights on how to manage disagreement in human–algorithm collaboration. This paper was accepted by Hemant Bhargava, Special Issue on the Human-Algorithm Connection. Funding: Y. Zhang appreciates the support from the National Natural Science Foundation of China [Grant 72272003]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.03844 .
1 August 2023 in Management Science
Do political differences, which are becoming increasingly acute among Americans, inhibit market transactions? We study this by examining whether the perceived political distance between investors and borrowers in an online lending market affects whom investors choose to fund. Using two complementary empirical approaches (a gravity model and a difference-in-differences analysis), we find a nuanced effect: Investors from comparatively conservative states consider political distance when making lending decisions, whereas investors from comparatively liberal states do not. Lending activity drops by as much as 11.6% when the investor’s state is more conservative than the borrower’s state. We also find that political distance between investors and borrowers reduces the likelihood that a borrower’s listing will be funded, thereby limiting the ability of the market to fulfill its function. However, political distance does not predict loan performance, which is consistent with another finding: The relationship of political distance to lending activity is not significant for experienced investors. It may be that investors stop considering political distance after they learn from experience that it does not predict loan performance. We find evidence for two mechanisms underlying our results: (1) a preference-based mechanism, in which investors from conservative states have a general preference for borrowers from conservative states, and (2) a rationality-based mechanism in which investors from conservative states use political ideology as a signal of creditworthiness (rightly or wrongly). Our results contribute to the literatures on online frictions and political (in)tolerance and have implications for the design of online lending (and other) markets.This paper was accepted by Kartik Hosanagar, information systems.Supplemental Material: The data files and online appendix are available at https://doi.org/10.1287/mnsc.2022.4575 .
14 January 2026 in Management Science
A common challenge faced by two-sided marketplaces is to expand the market while ensuring satisfactory matching of the two sides. We study a new platform growth strategy that increases market thickness with free entry while allowing suppliers to discretionarily signal product quality to buyers, thus forming a tiered market structure that maintains matching performance. We consider such an example with nonfungible token (NFT) markets where suppliers could choose to delay the entry cost of the NFT creation fee until an NFT is sold (lazy minting) or pay the fee up front at the time of NFT creation (gas minting). We use the difference-in-differences strategy to identify the impact of this lazy-minting policy on market matching performance. We find that the impacts of this policy are multifaceted: Although the introduction of lazy minting reduces the average matching likelihood of an NFT by 98.2%, it increases the gas-minting segment’s matching likelihood by 112.3% and first-sale price by 126.5%. Overall, this new growth strategy benefits the platform because of increased total sales and higher revenues. We unravel two mechanisms: the market thickness effect and the quality signaling effect. For the latter, we introduce a systematic framework to empirically establish the quality signaling mechanism with a separating equilibrium, where suppliers prefer gas minting for their high-quality creations. This effectively sends a quality signal that buyers could rely on to recognize high-quality NFTs. This new platform growth strategy provides generic implications for platforms endeavoring to balance market thickness and matching performance. This paper was accepted by Anindya Ghose, information systems. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2022.03573 .
10 September 2026 in SSRN Electronic Journal
Traditional human-subject surveys are a cornerstone of Information Systems (IS) research for testing theoretical frameworks involving socio-technical and psychological constructs, yet their reliance on researcher-elicited and self-reported data introduces well-known limitations. Is there an approach to utilize the rich observational data to examine the carefully developed, multi-construct theoretical framework that is often tested by a traditional survey? Leveraging the semantic reasoning capabilities of large language models (LLM), we develop the method of LLM-based observational survey (LOS), a complementary research design to traditional survey design that operationalizes theoretically grounded constructs from naturally occurring digital traces. Whereas emerging LLM-based annotation approaches primarily use LLMs to classify or measure individual constructs, LOS moves LLMs from a tool for annotating data to an integral component of research design. Specifically, LOS translates the entire theory-testing architecture of a traditional survey (including sampling logic, construct definitions, measurement items, temporal structure, hypothesized relationships, and empirical estimation) into an observational design based on naturally occurring digital traces. By interpreting heterogeneous natural-language expressions against theoretically specified construct definitions and measurement semantics, LLMs enable multi-construct theory testing that would be difficult to accomplish using conventional text-analytic approaches. LOS thus provides a distinct and complementary pathway for examining theoretical frameworks without directly eliciting responses from research participants. We demonstrate LOS by examining how generative AI (GenAI) usage influences users' broader life outcomes. Drawing on technology dependency theory, we theorize two mediating pathways: functional dependency, reflecting reliance on GenAI for task execution and decision support, and emotional dependency, reflecting psychological attachment to GenAI for social and emotional support. We test the same theoretical framework using a two-wave human-subject survey of 290 GenAI users and an aligned LOS based on naturally occurring posts and comments from Reddit users. The two designs yield largely convergent evidence despite fundamentally different sampling and measurement processes. Our demonstration validates the potential of LOS to complement traditional surveys with observational evidence and reveals important tradeoffs in construct observability and LLM-based measurement. Our study broadens the methodological toolkit for theory testing in IS research and illustrates a new form of inquiry made possible by human-AI collaboration.