Commentary Incentives and Asset Bubbles: Evidence from Learning-to-Forecast Experiments

Abstract

This paper examines how alternative incentive structures for public online commentary affect asset-price dynamics in a learning-to-forecast environment. We compare four settings: a baseline without commentary, non-incentivized commentary, commentary with peer endorsement, and commentary with financial incentives. We find that allowing commentary is associated with lower mispricing than the baseline. Adding a like mechanism reverses much of that improvement, leading to larger bubbles and slower convergence. When a financial incentive is added to the like-based system, mispricing and volatility are significantly lower than in the like-only treatment. Contextual analysis (employing 100 coders and an LLM) offers suggestive evidence that financial rewards encourage high-quality comments that stabilize subsequent market expectations, especially those offering fundamental insights. However, in a like-only environment, the desire for social recognition often associates with negative emotional venting, resulting in a failure to break speculative price momentum.