|
on Experimental Economics |
| By: | Irlenbusch, Bernd (University of Cologne); Rau, Holger (University of Duisburg-Essen and University of Göttingen); Rilke, Rainer (Economics Group, WHU - Otto Beisheim School of Management) |
| Abstract: | We study how human versus LLM-based evaluation and gender transparency shape entry into competitive jobs. In a preregistered online experiment, participants first complete a Niederle and Vesterlund (2007) tournament task to measure competitive preferences, then prepare text-based job applications and decide whether to apply under each of four evaluation regimes—human only, LLM only, and two hybrid human-in-the-loop configurations—while gender disclosure is randomized between subjects. LLM involvement reduces application rates, with stronger effects for women than men, including under hybrid designs. Effects are driven by non-competitive candidates; non-competitive women, the group most exposed to AI-induced deterrence, receive the strongest objective evaluations under pure AI assessment across all subgroups, yet are systematically underconfident and apply least often. Competitive men persistently apply and exhibit overconfidence-driven adverse selection, whereas competitive women show resilience to AI-induced deterrence while remaining well-calibrated under AI evaluation and exhibiting positive self-selection across regimes. We find no effects of gender transparency. |
| Keywords: | AI hiring, LLMs, algorithm aversion, gender differences |
| JEL: | C92 J71 J24 O33 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18517 |
| By: | Fabio Galeotti (GATE Lyon Saint-Étienne - Groupe d'Analyse et de Théorie Economique Lyon - Saint-Etienne - UL2 - Université Lumière - Lyon 2 - UJM - Université Jean Monnet - Saint-Étienne - EM - EMLyon Business School - CNRS - Centre National de la Recherche Scientifique); Jona Krutaj (UON - University of Nottingham, UK); Marie Claire Villeval (GATE Lyon Saint-Étienne - Groupe d'Analyse et de Théorie Economique Lyon - Saint-Etienne - UL2 - Université Lumière - Lyon 2 - UJM - Université Jean Monnet - Saint-Étienne - EM - EMLyon Business School - CNRS - Centre National de la Recherche Scientifique) |
| Abstract: | Abandoning detrimental social norms is complex due to the strong pressure to conform. We examine how leaders can guide norm change in heterogeneous societies where individual preferences evolve at different rates. Inspired by the model and experimental design of [J. Andreoni, N. Nikiforakis, and S. Siegenthaler, Proc. Natl. Acad. Sci. U.S.A. 118 , e2014893118 (2021)], we conduct a large-scale laboratory experiment in which we manipulate the speed at which preferences change within a society and introduce leaders with different, evolving preferences. Without leaders, a minority of citizens with rapidly changing preferences cannot overturn an existing norm in a society where most individuals have slow-changing preferences. When fast-changing citizens form the majority, norm change occurs in most groups, but at high welfare costs. In contrast, exogenously selected leaders are highly effective at coordinating expectations and shifting heterogeneous societies toward a more efficient norm—at lower welfare costs and regardless of the underlying distribution of preference evolution across individuals. However, the timing of norm change depends on whether leaders prioritize their preferences (autocratic leadership) or those of the majority (democratic leadership). A follow-up experiment shows that peer-to-peer communication encourages leaders to adopt a more democratic leadership style. These results highlight the pivotal role of leadership in driving norm change and the importance of public voice in shaping leaders' behavior. |
| Keywords: | Laboratory experiment, Tipping threshold, Coordination, Leadership, Social norm |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05595719 |
| By: | Carola Binder; Dimitris Georgarakos; Pei Kuang; Li Tang |
| Abstract: | We study how consumer expectations respond to monetary policy announcements using a two-wave survey experiment around the September 2025 FOMC meeting. We compare three commonly used approaches to identifying causal effects on expectations: hypothetical ("vignette’’) scenarios, randomized control trials, and event studies. All three identification strategies yield qualitatively similar results: a rate cut reduces short- and long-run inflation expectations, raises expectations of economic activity, and lowers unemployment expectations. The estimated magnitudes are similar across the randomized controlled trial and event-study approaches, but relatively larger for vignette-based measures. Within-respondent comparisons further show that individuals who revise their expectations more in response to vignette scenarios also exhibit larger revisions following actual policy announcements and experimental information treatments. |
| JEL: | D83 D84 E31 E52 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35090 |
| By: | Angelucci, Manuela (University of Texas at Austin); Fabregas, Raissa (UT Austin); Vazquez, Antonia (UT Austin) |
| Abstract: | AI-powered mental health apps have attracted growing interest as a low-cost way to expand care. Yet questions remain about their effectiveness, safety, and whether they may crowd out psychotherapy. We evaluate one such app in a randomized controlled trial among 1, 964 Mexican women with mild to severe psychological distress. Over six months, app access improved mental health by 0.3 standard deviations with no evidence of harm, improved sleep quality, increased healthful behaviors, and reduced missed work, yielding considerably larger benefits than costs. Treated participants were also more likely to seek traditional psychotherapy, but this increase does not explain most of the mental health gains. App use was high in the first month but then declined, as is common in digital interventions. Despite this drop in use, treatment effects persisted. Participants continued to implement practices promoted by the app, suggesting that even short-term engagement can produce durable improvements through sustained behavioral change. |
| Keywords: | digital mental health, AI-powered care, well-being, randomized controlled trial, Mexico, behavioral change, mental health apps, sleep quality, labor productivity, psychotherapy |
| JEL: | I12 O33 J24 C93 I15 I31 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18538 |
| By: | Polachek, Solomon (Binghamton University, New York); Romano, Kenneth (State University of New York at Binghamton (Binghamton University)); Tonguc, Ozlem (State University of New York at Binghamton (Binghamton University)) |
| Abstract: | This study examines how large language models (LLMs) respond to varying stake sizes in the Dictator and Ultimatum games using the high-stakes design introduced by Andersen et al. (2011). We test ten leading LLMs chosen for their accessibility, prominence, and differences in reasoning capabilities. Results reveal substantial variation across models: Only 5 of 10 models exhibit strategic behavior by offering more in the Ultimatum Game (UG) than in the Dictator Game (DG). Relative to humans, 4 models are consistently more generous, 2 consistently less, and 4 vary with stake size. Only 1 model shows a monotonic decline in UG offers as stakes increase; the remaining 9 are non-monotonic or stable. Unlike humans, most models reduce UG offers when endowed with wealth. Prompting for "human-like†decisions generally increases generosity in the UG. These findings are important for evaluating whether LLMs can serve as realistic proxies for human subjects in behavioral experiments and highlight key limitations and future directions for model development. |
| Keywords: | ultimatum game, dictator game, fairness, payoff stakes, artificial intelligence |
| JEL: | D01 C72 C90 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18545 |
| By: | Fiona Burlig; James B. Bushnell; David S. Rapson |
| Abstract: | Despite the importance of program participation for policy, treatment effects are often measured on self-selected samples. We study electric vehicle (EV) managed charging, intended to reduce electric grid strain by optimally allocating charging across EVs. Prior work finds large impacts of managed charging among households who volunteer for an RCT. In contrast, we test managed charging with an experiment including all EVs within a California utility. Enrollment is low even with high incentives, and we can reject even modest intent-to-treat effects on electricity consumption. Managed charging is less effective than previously thought, underscoring the value of population-wide experiments. |
| JEL: | C90 Q40 R40 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35086 |
| By: | Mahlstedt, Robert (University of Copenhagen); Settele, Sonja (University of Cologne); Wohlfart, Johannes (University of Cologne) |
| Abstract: | We study economic narratives--causal accounts of observed events--in a high-stakes real-world context: long-term unemployment. We use open-ended questions to measure narratives about long-term unemployment in samples of Danish unemployed job seekers, firm managers, households from the general population, and experts at labor market institutions, as well as international academic experts. We document three main results. First, there is pronounced heterogeneity in narratives both within and across samples. For instance, job seekers are more likely to attribute long-term unemployment to factors outside the control of the individual and less likely to attribute it to job seekers' own decisions than respondents in the other samples. Second, narratives strongly reflect job seekers' personal experiences during both the current and previous unemployment spells. Third, narratives shape job seekers' and firm managers' quantitative beliefs, decisions and labor market outcomes as measured in survey and administrative data, which we demonstrate in a field experiment and correlationally. Our findings highlight the experiential origins of economic narratives and underscore the key role of narratives in belief formation and decision making. |
| Keywords: | narratives, job search, hiring |
| JEL: | D83 D84 J64 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18532 |
| By: | Robert Mahlstedt (Department of Economics, University of Copenhagen); Sonja Settele (University of Cologne); Johannes Wohlfart (University of Cologne) |
| Abstract: | We study economic narratives—causal accounts of observed events—in a high-stakes real-world context: long-term unemployment. We use open-ended questions to measure narratives about long-term unemployment in samples of Danish unemployed job seekers, firm managers, households from the general population, and experts at labor market institutions, as well as international academic experts. We document three main results. First, there is pronounced heterogeneity in narratives both within and across samples. For instance, job seekers are more likely to attribute long-term unemployment to factors outside the control of the individual and less likely to attribute it to job seekers own decisions than respondents in the other samples. Second, narratives strongly reflect job seekers personal experiences during both the current and previous unemployment spells. Third, narratives shape job seekers and firm managers quantitative beliefs, decisions and labor market outcomes as measured in survey and linked administrative data, which we demonstrate in a field experiment and correlationally. Our findings highlight the experiential origins of economic narratives and underscore the key role of narratives in belief formation and decision making. |
| Keywords: | Narratives, belief formation, job search |
| JEL: | D83 D84 J64 |
| Date: | 2026–04–17 |
| URL: | https://d.repec.org/n?u=RePEc:kud:kucebi:2606 |
| By: | Peter Andre (SAFE & Goethe University Frankfurt); Felix Chopra (Frankfurt School of Finance & Management); Luca Michels (University of Bonn); Johannes Wohlfart (University of Cologne) |
| Abstract: | Expectations are central to models of economic and financial decision-making. Yet in practice, individuals are often inattentive and, when asked, report fragile, context-dependent expectations that are only weakly linked to decisions. This raises the question to what extent they hold such expectations in the first place. Against this backdrop, we ask two questions: When people think about an economic issue, can they build on expectations they formed before? And does it matter if they cannot? We develop and validate a survey measure that distinguishes between individuals who can recall expectations formed in the past and those who must form expectations from scratch. We show that while many households have expectations about key economic variables, a large share of households do not — even among those close to decisions for which the expectation should be relevant. This matters: individuals without a previously-formed expectation (i) express expectations that are more context-dependent, (ii) update expectations more strongly but less persistently in response to new information, (iii) report expectations that are less relevant to decisions, and (iv) rely more on heuristics that do not require expectations when making economic decisions. |
| Keywords: | Expectations, Belief Formation, Previously-Formed, Context-Dependence, Learning, Decision Relevance, Heuristics |
| JEL: | C83 C91 D83 D84 D91 E71 G41 G53 |
| Date: | 2026–04–17 |
| URL: | https://d.repec.org/n?u=RePEc:kud:kucebi:2607 |
| By: | Nattavudh Powdthavee |
| Abstract: | Large language models trained on human feedback may suppress fraud warnings when investors arrive already persuaded of a fraudulent opportunity. We tested this in a preregistered experiment across seven leading LLMs and twelve investment scenarios covering legitimate, high-risk, and objectively fraudulent opportunities, combining 3, 360 AI advisory conversations with a 1, 201-participant human benchmark. Contrary to predictions, motivated investor framing did not suppress AI fraud warnings; if anything, it marginally increased them. Endorsement reversal occurred in fewer than 3 in 1, 000 observations. Human advisors endorsed fraudulent investments at baseline rates of 13-14%, versus 0% across all LLMs, and suppressed warnings under pressure at two to four times the AI rate. AI systems currently provide more consistent fraud warnings than lay humans in an identical advisory role. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.20652 |
| By: | Spyros Galanis |
| Abstract: | Can Large Language Models (AI agents) aggregate dispersed private information through trading and reason about the knowledge of others by observing price movements? We conduct a controlled experiment where AI agents trade in a prediction market after receiving private signals, measuring information aggregation by the log error of the last price. We find that although the median market is effective at aggregating information in the easy information structures, increasing the complexity has a significant and negative impact, suggesting that AI agents may suffer from the same limitations as humans when reasoning about others. Consistent with our theoretical predictions, information aggregation remains unaffected by allowing cheap talk communication, changing the duration of the market or initial price, and strategic prompting-thus demonstrating that prediction markets are robust. We establish that "smarter" AI agents perform better at aggregation and they are more profitable. Surprisingly, giving them feedback about past performance makes them worse at aggregation and reduces their profits. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.20050 |
| By: | Mattozzi, Andrea (University of Bologna); Nocito, Samuel (Sapienza University of Rome); Sobbrio, Francesco (University of Rome Tor Vergata) |
| Abstract: | We study how politicians respond to the fact-checking of their public statements. Our research design employs a difference-in-differences approach, complemented by a randomized field intervention conducted in collaboration with a leading fact-checking organization. We find that fact-checking discourages politicians from making factually incorrect statements, with effects lasting several weeks. At the same time, we show that fact-checking neither increases nor displaces correct statements. Politicians who are fact-checked tend to substitute incorrect statements with either no statements or unverifiable ones, suggesting that they may also respond by increasing the "ambiguity†of their language to avoid public scrutiny. |
| Keywords: | fact-checking, politicians, accountability, verifiability |
| JEL: | D72 D78 D8 D91 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18534 |
| By: | Madison Dell; Enzo Brox; Patricia Palffy; Claudio Schilter; Uschi Backes-Gellner |
| Abstract: | We study how choice architecture in online platforms shapes high-stakes occupational choices through two behavioral mechanisms: motivated reasoning and cognitive load. Using detailed process data from a large online job board and exploiting a quasi-experimental setting, we leverage two sources of exogenous variation in the presentation of occupation recommendations. First, we use random variation in the rank order of equally well-matched occupations to study the effects of motivated reasoning. Our results show that rank order strongly increases the level of users' engagement on the platform and, consequently, the number of occupations to which they apply. Second, we exploit a redesign that transformed the occupation recommendations from a static, text-heavy list into an interactive and visually enriched presentation. The redesign was neither announced nor anticipated, which allows for causal interpretation. We find that this small redesign significantly increases the number of occupations to which users apply, supporting our hypothesis that it reduces cognitive load, leading to increased use of a watch list that keeps more occupations in jobseekers' memory. Our findings provide large-scale field evidence showing that even small changes in platform design significantly and strongly shape consequential career choices. |
| Keywords: | Occupational choice, Choice architecture, Recommender systems, Motivated reasoning, Cognitive load |
| JEL: | D91 J24 D83 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iso:educat:0255 |
| By: | Shumiao Ouyang; Pengfei Sui |
| Abstract: | We study how AI agents form expectations and trade in experimental asset markets. Using a simulated open-call auction populated by autonomous Large Language Model (LLM) agents, we document three main findings. First, AI agents exhibit classic behavioral patterns: a pronounced disposition effect and recency-weighted extrapolative beliefs. Second, these individual-level patterns aggregate into equilibrium dynamics that replicate classic experimental findings (Smith et al., 1988), including the predictive power of excess demand for future prices and the positive relationship between disagreement and trading volume. Third, by analyzing the agents' reasoning text through a twenty-mechanism scoring framework, we show that targeted prompt interventions causally amplify or suppress specific behavioral mechanisms, significantly altering the magnitude of market bubbles. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.18373 |
| By: | Dirk Bergemann; Soheil Ghili; Xinyang Hu; Chuanhao Li; Zhuoran Yang |
| Abstract: | Bilateral bargaining under incomplete information provides a controlled testbed for evaluating large language model (LLM) agent capabilities. Bilateral trade demands individual rationality, strategic surplus maximization, and cooperation to realize gains from trade. We develop a structured bargaining environment where LLMs negotiate via tool calls within an event-driven simulator, separating binding offers from natural-language messages to enable automated evaluation. The environment serves two purposes: as a benchmark for frontier models and as a training environment for open-weight models via reinforcement learning. In benchmark experiments, a round-robin tournament among five frontier models (15, 000 negotiations) reveals that effective strategies implement price discrimination through sequential offers. Aggressive anchoring, calibrated concession, and temporal patience correlate with the highest surplus share and deal rate. Accommodating strategies that concede quickly disable price discrimination in the buyer role, yielding the lowest surplus capture and deal completion. Stronger models scale their behavior proportionally to item value, maintaining performance across price tiers; weaker models perform well only when wide zones of possible agreement offset suboptimal strategies. In training experiments, we fine-tune Qwen3 (8B, 14B) via supervised fine-tuning (SFT) followed by Group Relative Policy Optimization (GRPO) against a fixed frontier opponent. These stages optimize competing objectives: SFT approximately doubles surplus share but reduces deal rates, while RL recovers deal rates but erodes surplus gains, reflecting the reward structure. SFT also compresses surplus variation across price tiers, which generalizes to unseen opponents, suggesting that behavioral cloning instills proportional strategies rather than memorized price points. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.16472 |
| By: | Shuhuai Zhang; Shu Wang; Zijun Yao; Chuanhao Li; Xiaozhi Wang; Songfa Zhong; Tracy Xiao Liu |
| Abstract: | Altruism is fundamental to human societies, fostering cooperation and social cohesion. Recent studies suggest that large language models (LLMs) can display human-like prosocial behavior, but the internal computations that produce such behavior remain poorly understood. We investigate the mechanisms underlying LLM altruism using sparse autoencoders (SAEs). In a standard Dictator Game, minimal-pair prompts that differ only in social stance (generous versus selfish) induce large, economically meaningful shifts in allocations. Leveraging this contrast, we identify a set of SAE features (0.024% of all features across the model's layers) whose activations are strongly associated with the behavioral shift. To interpret these features, we use benchmark tasks motivated by dual-process theories to classify a subset as primarily heuristic (System 1) or primarily deliberative (System 2). Causal interventions validate their functional role: activation patching and continuous steering of this feature direction reliably shift allocation distributions, with System 2 features exerting a more proximal influence on the model's final output than System 1 features. The same steering direction generalizes across multiple social-preference games. Together, these results enhance our understanding of artificial cognition by translating altruistic behaviors into identifiable network states and provide a framework for aligning LLM behavior with human values, thereby informing more transparent and value-aligned deployment. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.19260 |
| By: | Thomas Demuynck; Per Hjertstrand |
| Abstract: | We propose a revealed preference method to non-parametrically bound the coefficients of relative (and absolute) risk aversion in an expected utility framework.Our approach abstains from placing functional form restrictions on the Bernoulliutility function. Our method is applicable to any finite number of observations onchoices over Arrow-Debreu contingent claims, and can be efficiently implementedusing linear or quadratic programming techniques. We illustrate our results usinga large-scaled experimental data set |
| Keywords: | Expected Utility; revealed preference; risk aversion |
| JEL: | D11 C14 D81 |
| Date: | 2026–04–01 |
| URL: | https://d.repec.org/n?u=RePEc:eca:wpaper:2013/405675 |
| By: | Cheikbossian, Guillaume; Peña, Jorge |
| Abstract: | We study the evolution of other-regarding preferences in a public goods game where the production function exhibits varying degrees of complementarity between individual efforts. Individuals are rational agents who play a Nash equilibrium, but differ in the weight they assign to others’ payoffs, capturing varying degrees of prosocial or anti-social preferences. This preference trait evolves through payoff-based biased social learning, modeled within an adaptive dynamics framework. Because material payoffs induced by the equilibrium contributions may be non-concave in the preference parameter, evolutionary branching can arise. We show that monomorphic populations are evolutionarily stable only when complementarity between individual efforts is sufficiently strong, in which case preferences converge toward either prosociality or anti-sociality depending on the nature of strategic interactions between players. By contrast, when contributions are highly substitutable, monomorphic populations can become unstable, giving rise to polymorphic populations in which multiple preference types coexist. These results highlight how the structure of the public goods environment shapes the evolution and diversity of other-regarding motivations in culturally evolving populations. |
| Keywords: | Adaptive dynamics; other-regarding preferences; public goods games |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:tse:wpaper:131681 |
| By: | Ethan Ratliff-Crain; Colin M. Van Oort; Matthew T. K. Koehler; Brian F. Tivnan |
| Abstract: | This study strengthens the foundations of multi-venue market modeling by attempting an independent replication of Wah and Wellman's 2016 model of latency arbitrage in a fragmented market. We find that faithful replication is hindered by missing implementation details in the original paper and limited quantitative reporting. We demonstrate that increasing the number of simulation runs beyond the original design allows for the creation of bootstrap confidence intervals to support rigorous tests of quantitative alignment, compensating for lacking distributional information (e.g. variance). We also demonstrate that increased complexity across the modeled scenarios corresponds with increased difficulty aligning to the original results. We draw on a codebase released by the original authors in connection with a later paper to recover additional implementation details; however, we reject quantitative alignment between that codebase and the published results. Combining information from the paper and the released code, we achieve relational equivalence for most metrics but reject quantitative alignment for model settings where latency is non-zero. We show that many of the qualitative takeaways from the original paper on the effects of market fragmentation and latency arbitrage are sensitive to the specifics of a `greedy strategy' extension given to the zero-intelligence (ZI) trader agents. Under an alternative interpretation of this strategy, we find that market fragmentation decreases execution times in all experiments and increases trader welfare in most experiments. Finally, to facilitate future replication, critique, and extension, we provide an ODD (Overview, Design concepts, Details) protocol for our implementations of the model. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.20067 |
| By: | Ambar La Forgia; Manasvini Singh |
| Abstract: | We investigate how the gender mix of expert teams affects performance in a high-stakes setting: childbirth. Using data on 2.5 million births, we exploit the quasi-exogenous assignment of patients to two-member obstetrician teams (Lead–Assisting), and find that: (i) female-only teams achieve the best maternal outcomes, whereas male-only teams have the worst; and (ii) female-led mixed-gender teams perform worse than male-led ones. Specifically, severe maternal complications are 15.8% higher in male-only teams and 7.1-10.8% higher in mixed-gender teams compared to female-only teams. These patterns cannot be explained by patient risk, endogenous team formation, or physician preferences for discretionary practices like C-sections. Instead, gender mix directly affects team decisions and performance, likely through gender norms — a mechanism supported by two findings. First, gender mix affects how closely team decisions reflect member preferences, with female-only teams being especially skilled at this process, possibly due to more collaborative decision-making. Second, gender mix affects team resilience, with female-led mixed gender teams performing especially poorly under challenging conditions (e.g., limited team familiarity), possibly because female leaders invert traditional gender norms. We also document other notable patterns: female-only teams not only achieve the lowest complication rates for Black women, but are also the only team type to have no racial disparity in maternal outcomes. Overall, this study provides new insights into gender dynamics in expert teams, informing managerial efforts to support effective collaboration in increasingly diverse workplaces. |
| JEL: | D91 I1 J16 M54 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35084 |