TLDR Tech / Crypto
Market Making for Prediction Markets
Classical Avellaneda-Stoikov market-making fails for prediction markets because prices are bounded probabilities with non-constant volatility and binary settlement, and thus requires a rebuilt framework using probability-space AS with VPIN-based flow toxicity detection, proportional inventory clamping, and Lagrange multiplier coordination across multi-outcome markets. Backtesting across four model iterations validated the approach, moving from a $1,114 loss in the logit-space v1 build to a $453 profit in v4, with profitability across all tested spread widths.
Read the full story