Unveiling Latent Market Dynamics: A Hidden Markov Framework for Informed Trading

Abstract

Abstract The Probability of Informed Trading (PIN) treats informational events as independent across trading days, leaving it silent on how long informational regimes last. We propose the Hidden Probability of Informed Trading (HPIN), which lets the informational state evolve as a hidden Markov chain while keeping the sequential trade model intact. Existing hidden Markov treatments of daily order flow replace those regimes with generic latent states, whose economic content must be recovered by clustering. The HPIN needs no such step: its emission intensities are tied to the PIN parameters inside the likelihood, so the latent states remain the three informational regimes. We derive the tied Baum–Welch updates in closed form and show that the classical PIN is the degenerate case in which all transition rows are identical. To measure what that costs, we simulate six environments holding the unconditional PIN fixed at 0.1163 while varying the persistence of the informational process. Under correct specification the HPIN recovers the spectral gap with mean absolute error 0.037 and correlation 0.999; under overdispersion the estimates compress but preserve the ranking. The static model reports exactly 1.000 everywhere—an arithmetic consequence of its degenerate transition matrix, not an error any estimator could repair. On the level of informed trading the two are equivalent: both overstate the true PIN comparably (mean absolute bias 0.016 and 0.019), with indistinguishable one-step-ahead out-of-sample forecasts. The case for modeling informational dynamics therefore rests on what the static specification cannot represent, not on sharper measurement of what it can.

The authors' abstract, as published at the source. Annals of Data Science, 2026 · DOI ↗