Can a Hidden Markov Model Find Trading Regimes?
I put a Hidden Markov Model on order-flow data expecting it to discover sticky, tradeable market regimes. On one feature it became a fancy way to smooth the sign of order flow — 50% accuracy. On many features it latched onto volatility instead of direction and actively destroyed the small signal that was there. Here’s how HMMs work, how to wire one up without secretly leaking the future, and why the machinery didn’t earn its keep.