Skip to main navigation Skip to search Skip to main content

Rational Roots of Irrational Behavior: Reinforcement Learning Agents and the Endowment Effect in Market Transactions

Research output: Conference Papers › RGC 33 - Other conference paper › peer-review

Abstract

Artificial Intelligence (AI) agents are increasingly deployed as autonomous decision-makers in economic environments. However, agents trained through environmental interaction may develop systematic behavioral patterns that deviate from classical rationality. We examine this possibility through the endowment effect, reflected in a persistent gap between willingness to accept (WTA) and willingness to pay (WTP). Using BDM elicitation, we train reinforcement learning (RL) agents in repeated market transactions. A 3 × 2 × 2 factorial design varies reference price, training depth, and training modality to test whether the endowment effect can emerge in agents without loss aversion or psychological ownership. Our findings show that RL agents consistently exhibit a significant WTA–WTP gap across all conditions. Reference prices moderate the magnitude of the effect, while training modality—whether agents are trained in fixed or alternating roles—substantially shapes its size. This study contributes a replicable RL-based experimental framework for studying emergent behavioral patterns in AI agents. It further provides evidence that biases traditionally attributed to human cognition can emerge from computational adaptation to environmental structure and yields practical implications for the design and governance of autonomous AI agents in digital markets.
Original languageEnglish
Publication statusPresented - 23 Jun 2026
Event22nd Symposium on Statistical Challenges in Electronic Commerce Research (SCECR 2026): The Generative Shift: Rethinking Work, Creativity, and Knowledge in the AI Age - Shangri-La Tokyo, Tokyo, Japan
Duration: 21 Jun 2026 → 24 Jun 2026
https://scecr.com/

Conference

Conference22nd Symposium on Statistical Challenges in Electronic Commerce Research (SCECR 2026)
Abbreviated title2026 SCECR
PlaceJapan
CityTokyo
Period21/06/26 → 24/06/26
Internet address

Bibliographical note

Research Unit(s) information for this publication is provided by the author(s) concerned.
Information for this record is provided by the author(s) concerned.

Research Keywords

  • reinforcement learning
  • endowment effect
  • reference price
  • artificial intelligence agents
  • machine behaviour

Fingerprint

Dive into the research topics of 'Rational Roots of Irrational Behavior: Reinforcement Learning Agents and the Endowment Effect in Market Transactions'. Together they form a unique fingerprint.

Cite this