arXiv subjects

Arun Sundararajan

Publications and source records attributed to Arun Sundararajan.

8 recordsLinked to original sources

arXiv2026

The Illusion of Collusion

Algorithmic agents are used in a variety of competitive decision-making settings, including pricing contexts that range from online retail to residential home rental. We study the emergence of algorithmic collusion when competing agents employ multi-armed bandit algorithms and competition is modeled as a repeated Prisoner's Dilemma game. Notably, agents in our setting perform online learning with no prior model of game structure and have no direct knowledge of competitor states or actions, thus they cannot learn strategies that depend on these factors. These context-free bandits nonetheless frequently learn seemingly collusive behavior, a phenomenon we term naive collusion. Our results reveal that whether naive collusion emerges depends starkly on the choice of behavior policy employed by bandit learners. The mechanism underpinning the emergence of collusive outcomes is synchronicity in agent action plays, where synchronicity captures how often agents play the same action. We show that in the long-run, naive algorithmic collusion never emerges when both agents use a broad class of persistently random algorithms, including the epsilon-greedy algorithm without epsilon decay, sometimes emerges when both agents use greedy-in-the-limit algorithms which feature randomness during exploration but are asymptotically deterministic, and always emerges when both agents use deterministic bandit learning algorithms like those in the well-known upper confidence bound (UCB) family. We highlight market and algorithmic conditions under which one can and cannot predict a priori whether collusion will occur. Our findings have several policy implications: preventing pricing algorithms from conditioning their actions on competitor prices may not preclude algorithmic collusion, symmetry in algorithms may increase collusion potential, and the emergence of algorithmic collusion is path dependent.

econ.GN↗

arXiv2026

Platform Design, Earnings Transparency and Minimum Wage Policies: Evidence from A Natural Experiment on Lyft

We study the effects of a significant design and policy change at a major ridesharing platform that altered both provider earnings and platform transparency, examining how it affected outcomes for drivers, riders, and the platform, and providing managerial insights on balancing competing stakeholder interests while avoiding unintended consequences. In February 2024, Lyft introduced a policy guaranteeing drivers a minimum fraction of rider payments while increasing per-ride earnings transparency. The staggered rollout, first in major markets, created a natural experiment to examine how earnings guarantees and transparency affect ride availability and driver engagement. Using trip-level data from over 47 million rides across a major market and adjacent markets over six months, we apply dynamic staggered difference-in-differences models combined with a geographic border strategy to estimate causal effects on supply, demand, ride production, and platform performance. We find that the policy led to substantial increases in driver engagement, with distinct effects from the guarantee and transparency. Drivers increased working hours and utilization, resulting in more completed trips and higher per-hour and per-trip earnings, with stronger effects among drivers with lower pre-policy earnings and greater income uncertainty. Increased supply also generated positive spillovers on demand. We also find evidence that greater transparency may induce strategic driver behavior. In ongoing work, we develop a counterfactual simulation framework linking driver supply and rider intents to ride production, illustrating how small changes in driver choices could further amplify policy effects. Our study shows how platform-led interventions present an intriguing alternative to government-led minimum pay regulation and provide new strategic insights into managing platform change.

econ.GN↗

arXiv2025

Incentives for Digital Twins: Task-Based Productivity Enhancements with Generative AI

Generative AI is a technology which depends in part on participation by humans in training and improving the automation potential. We focus on the development of an "AI twin" that could complement its creator's efforts, enabling them to produce higher-quality output in their individual style. However, AI twins could also, over time, replace individual humans. We analyze this trade-off using a principal-agent model in which agents have the opportunity to make investments into training an AI twin that lead to a lower cost of effort, a higher probability of success, or both. We propose a new framework to situate the model in which the tasks performed vary in the ease to which AI output can be improved by the human (the "editability") and also vary in the extent to which a non-expert can assess the quality of output (its "verifiability.") Our synthesis of recent empirical studies indicates that productivity gains from the use of generative AI are higher overall when task editability is higher, while non-experts enjoy greater relative productivity gains for tasks with higher verifiability. We show that during investment a strategic agent will trade off improvements in quality and ease of effort to preserve their wage bargaining power. Tasks with high verifiability and low editability are most aligned with a worker's incentives to train their twin, but for tasks where the stakes are low, this alignment is constrained by the risk of displacement. Our results suggest that sustained improvements in company-sponsored generative AI will require nuanced design of human incentives, and that public policy which encourages balancing worker returns with generative AI improvements could yield more sustained long-run productivity gains.

econ.TH↗

arXiv2025

Reasoning and the Trusting Behavior of DeepSeek and GPT: An Experiment Revealing Hidden Fault Lines in Large Language Models

When encountering increasingly frequent performance improvements or cost reductions from a new large language model (LLM), developers of applications leveraging LLMs must decide whether to take advantage of these improvements or stay with older tried-and-tested models. Low perceived switching frictions can lead to choices that do not consider more subtle behavior changes that the transition may induce. Our experiments use a popular game-theoretic behavioral economics model of trust to show stark differences in the trusting behavior of OpenAI's and DeepSeek's models. We highlight a collapse in the economic trust behavior of the o1-mini and o3-mini models as they reconcile profit-maximizing and risk-seeking with future returns from trust, and contrast it with DeepSeek's more sophisticated and profitable trusting behavior that stems from an ability to incorporate deeper concepts like forward planning and theory-of-mind. As LLMs form the basis for high-stakes commercial systems, our results highlight the perils of relying on LLM performance benchmarks that are too narrowly defined and suggest that careful analysis of their hidden fault lines should be part of any organization's AI strategy.

cs.CL↗

arXiv2024

The Rise of Recommerce: Ownership and Sustainability with Overlapping Generations

The emergence of the branded recommerce channel - digitally enabled and branded marketplaces that facilitate purchasing pre-owned items directly from a manufacturer's e-commerce site - leads to new variants of classic IS and economic questions relating to secondary markets. Such branded recommerce is increasingly platform-enabled, creating opportunities for greater sustainability and stronger brand experience control but posing a greater risk of cannibalization of the sales of new items. We model the effects that the sales of pre-owned items have on market segmentation and product durability choices for a monopolist facing heterogeneous customers, contrasting outcomes when the trade of pre-owned goods takes place through a third-party marketplace with outcomes under branded recommerce. We show that the direct revenue benefits of branded recommerce are not their primary source of value to the monopolist, and rather, there are three indirect effects that alter profits and sustainability. Product durability increases, a seller finds it optimal to forgo marketplace fees altogether, and there are greater seller incentives to lower the quality uncertainty associated with pre-owned items. We establish these results for a simple two-period model as well as developing a new infinite horizon model with overlapping generations. Our paper sheds new insight into this emerging digital channel phenomenon, underscoring the importance of recommerce platforms in aligning seller profits with sustainability goals.

econ.GN↗

arXiv2024

Evaluating LLMs for Gender Disparities in Notable Persons

This study examines the use of Large Language Models (LLMs) for retrieving factual information, addressing concerns over their propensity to produce factually incorrect "hallucinated" responses or to altogether decline to even answer prompt at all. Specifically, it investigates the presence of gender-based biases in LLMs' responses to factual inquiries. This paper takes a multi-pronged approach to evaluating GPT models by evaluating fairness across multiple dimensions of recall, hallucinations and declinations. Our findings reveal discernible gender disparities in the responses generated by GPT-3.5. While advancements in GPT-4 have led to improvements in performance, they have not fully eradicated these gender disparities, notably in instances where responses are declined. The study further explores the origins of these disparities by examining the influence of gender associations in prompts and the homogeneity in the responses.

cs.CL↗

arXiv2023

COVID-19 and Digital Resilience: Evidence from Uber Eats

Using order-level data from Uber Technologies, we study how the COVID-19 pandemic and the ensuing shutdown of businesses in the United States in 2020 affected small business restaurant supply and demand on the Uber Eats platform. We find evidence that small restaurants experience significant increases in activity on the platform following the closure of the dine-in channel. We document how locality- and restaurant-specific characteristics moderate the size of the increase in activity through the digital channel and explain how these increases may be due to both demand- and supply-side shock. We observe an increase in the intensity of competitive effects following the economic shock and show that growth in the number of providers on a platform induces both market expansion and heightened inter-provider competition. Higher platform activity in response to the shock does not only have short-run implications: restaurants with larger demand shocks had a higher on-platform survival rate one year after the lockdown, suggesting that the platform channel contributes towards long-run resilience following a crisis. Our findings document the heterogeneous effects of platforms during the pandemic, underscore the critical role that digital technologies play in enabling business resilience in the economy, and provide insight into how platforms can manage competing incentives when balancing market expansion and growth goals with the competitive interests of their incumbent providers.

econ.GN↗

arXiv2006

Evaluating Pricing Strategy Using e-Commerce Data: Evidence and Estimation Challenges

As Internet-based commerce becomes increasingly widespread, large data sets about the demand for and pricing of a wide variety of products become available. These present exciting new opportunities for empirical economic and business research, but also raise new statistical issues and challenges. In this article, we summarize research that aims to assess the optimality of price discrimination in the software industry using a large e-commerce panel data set gathered from Amazon.com. We describe the key parameters that relate to demand and cost that must be reliably estimated to accomplish this research successfully, and we outline our approach to estimating these parameters. This includes a method for ``reverse engineering'' actual demand levels from the sales ranks reported by Amazon, and approaches to estimating demand elasticity, variable costs and the optimality of pricing choices directly from publicly available e-commerce data. Our analysis raises many new challenges to the reliable statistical analysis of e-commerce data and we conclude with a brief summary of some salient ones.

math.ST↗