Research projects where I implement a paper or a model and stress-test it on real market data.
The Price Impact of Order Book Events
Implemented Cont, Kukanov & Stoikov's 2010 order-flow paper from scratch on four years of SOXL tick data — 23 million one-second bars — and reproduced its headline result: R² ≈ 0.67, positive on all 1,000 trading days. Then spent weeks proving the signal never survives the spread — six strategies, taker and maker, every one dead. A measured study of the difference between a signal and an edge.
Hidden Markov Models for Trading Regimes
I put a Hidden Markov Model on order-flow data to find tradeable market regimes. Instead it learned volatility, ignored direction, and destroyed the small signal that was there. A clear example of when a more complex model doesn't help — and how to build an HMM on time series without leaking the future into your backtest.