Preference Optimization Drives Monoculture in LLM Prediction Markets

2026-06-25

AI Summary

This study reveals that large language model (LLM) agents fine-tuned via Direct Preference Optimization (DPO) exhibit highly homogeneous outputs in prediction markets, violating the independence-of-errors assumption and substantially reducing collective forecast diversity. The work identifies preference optimization as the causal mechanism underlying this emergent “monoculture” phenomenon and proposes cross-model diversity as a mitigation strategy. Experimental results demonstrate that an ensemble of ten DPO-tuned agents yields an effective number of independent predictors (N_eff) of only 1.4 and achieves a prediction accuracy of 67.6%, which is notably lower than that of a single untuned baseline model (70.2%). By incorporating cross-model ensembling, the correlation among prediction errors drops from 0.68 to 0.40, effectively restoring the wisdom of crowds.

Abstract

Prediction markets rest on the independence of participant errors. As LLM agents become active traders on platforms like Kalshi and Polymarket, we ask: does this independence hold when the crowd is composed of LLMs? We find it does not. LLM agents fine-tuned with Direct Preference Optimization (DPO) share a convergent output distribution, producing pairwise error correlations of $ρ= 0.70$ and reducing ten agents to the effective forecasting power of ${\approx}1.4$ independent forecasters $N_{\text{eff}}$. This is not a scaling problem: $N_{\text{eff}}$ remains flat from $N=5$ to $N=40$, and the 10-agent market (67.6%) fails to match a single standalone agent (70.2%). Two controlled ablations isolate preference optimization as the causal driver, replicated across labs and scales ($Δρ= +0.24$ to $+0.46$ on identical-SFT controls at 8B and 70B). Among mitigations tested, cross-model diversity achieves the largest correlation reduction ($ρ$ from 0.68 to 0.40). As LLMs become more aligned, markets built from them become more monocultural.

Problem

• monoculture
• preference optimization
• prediction markets
• error correlation
• LLM agents

Innovation

• Preference Optimization
• Monoculture
• Prediction Markets
• LLM Agents
• Error Correlation