The Price of Having Been Right: Capital Lock-In in Prediction Markets
I study the behavior of agents in prediction markets with transaction costs. Agents use fractional-Kelly strategies based on competing probabilistic forecasts. In the frictionless benchmark, prices are wealth-weighted averages of forecasts, and wealth updates depend on likelihood ratios. Successful forecasts therefore come to dominate the price, making the market accurate but slow to adapt: after a change, repricing can take time proportional to the preceding regime’s length.
Shrinking betting fractions can preserve learning while removing this linear lock-in when trading is free. With a per-contract fee, the power-law schedules studied instead produce linear recovery times, and faster shrinkage makes recovery slower. I identify a sharp lower bound on the delay when betting fractions never increase. I then show how a policy combining shrinking baseline fractions with reversible increases in response to evidence of change restores sublinear recovery without new capital, while controlling false alarms.
I close with questions about how execution, fee incidence, and inventory constraints affect the speed at which prediction-market prices absorb new information.
Bio: Jan Vecer is a faculty member at Charles University in Prague and a Visiting Professor at the Frankfurt School of Finance & Management. He received his Ph.D. in mathematical finance from Carnegie Mellon University in 2000 and served on the faculty of Columbia University from 2001 to 2010. His research and publications span quantitative finance, portfolio choice, derivative pricing, and the connections between financial markets, statistics, and information theory. Alongside his academic work, he has developed market-making bots for prediction and energy markets. He is the author of the forthcoming book Principles of Portfolio Choice: An Information-Theoretic, Likelihood-Based Perspective (Chapman & Hall/CRC).