Blog

One of the most persistent challenges in quantitative finance is that markets don’t behave consistently over time. A strategy that performs well during a calm trending period can blow up during a crisis. The underlying issue is that markets switch between distinct regimes — periods of low volatility interspersed with bursts of high volatility — and most models ignore this structure entirely.

Hidden Markov Models (HMMs) offer a principled way to handle this. The core idea is simple: assume there is a small number of hidden states (e.g., “calm” and “turbulent”), and that the observed returns are drawn from different distributions depending on the current state. The states transition between each other with some probability, and we never directly observe which state we’re in — we infer it from the data.