Selected papers and technical notes on quantitative finance, machine learning, and market microstructure.
Current liquidity-mining pays for activity, not information. I developed a new incentive mechanism that rewards information discovery by measuring the Fisher-Rao distance the market probability travels. Wash trading earns nothing, while sustained discovery earns large rewards.
Estimating liquidity for rarely traded stocks has always been a guessing game. This paper introduces a method that uses highly liquid "reference stocks" to correct spread estimates, offering a more accurate view of market costs.
I built a production system that combines LLMs with formal verification for derivatives pricing. FVLM achieves 100% verification success, zero runtime failures, and 323x better numeric precision than standard approaches—processing over 127,000 requests in production testing.