Hidden Order in Trades Predicts the Size of Price Moves

Financial markets exhibit an apparent paradox: while directional price movements remain largely unpredictable–consistent with weak-form efficiency–the magnitude of price changes displays systematic structure. Here we demonstrate that real-time order-flow entropy, computed from a 15-state Markov transition matrix at second resolution, predicts the magnitude of intraday returns without providing directional information. Analysis of 38.5 million SPY trades over 36 trading days reveals that conditioning on entropy below the 5th percentile increases subsequent 5-minute absolute returns by a factor of 2.89 (t = 12.41, p < 0.0001), while directional accuracy remains at 45.0%–statistically indistinguishable from chance (p = 0.12). This decoupling arises from a fundamental symmetry: entropy is invariant under sign permutation, detecting the presence of informed trading without revealing its direction. Walk-forward validation across five non-overlapping test periods confirms out-of-sample predictability, and label-permutation placebo tests yield z = 14.4 against the null. These findings suggest that information-theoretic measures may serve as volatility state variables in market microstructure, though the limited sample (36 days, single instrument) requires extended validation.

💡 Research Summary

The paper investigates whether an information‑theoretic measure derived from high‑frequency order flow can forecast the magnitude of intraday price moves without providing any directional edge. The authors construct a 15‑state discrete Markov model at one‑second resolution by crossing price‑change sign (‑1, 0, +1) with volume quintile (1‑5). Using a rolling 120‑second window they estimate the transition matrix, compute its stationary distribution, and calculate a normalized entropy Hₜ that ranges from 0 (highly predictable) to 1 (completely random).

Two theoretical results are presented. Theorem 1 (Magnitude Predictability) states that, under standard microstructure assumptions, low entropy implies a higher expected absolute return over a future horizon Δ. Theorem 2 (Directional Unpredictability) shows that the same entropy is invariant under a swap of buy and sell labels, so the conditional expectation of the signed return is zero. These results formalize the intuition that informed traders generate structured order flow regardless of whether they are buying or selling.

Empirically the authors analyze 38.5 million SPY trades covering 36 trading days (Oct 1–Nov 19 2025), aggregated into 828,907 one‑second bars. Conditioning on entropy below the 5th percentile yields an average 5‑minute absolute return of 15.3 bps, a 2.89‑fold increase over the unconditional mean of 5.29 bps (t = 12.41, p < 0.0001). By contrast, directional accuracy under the same condition is 45.0 % (p = 0.12), indistinguishable from random guessing, confirming the predicted decoupling.

A walk‑forward validation with five non‑overlapping folds (10‑day training, 5‑day testing) reproduces the magnitude effect in every fold while the win‑rate remains below 50 %. A simple asymmetric payoff rule—enter when entropy is low, take a position in the direction of the trailing 5‑minute return, exit on a 5 bps stop, a 300‑second timeout, or a pre‑set profit target—generates 1,126 bps cumulative P&L over 240 trades. Decomposition shows that 87.8 % of the profit stems from timing (knowing when volatility spikes), 12.2 % from the asymmetric payoff structure, and 0 % from directional forecasting.

Robustness is examined through three placebo tests: (1) label permutation (1,000 trials) produces a magnitude ratio of 1.02 ± 0.08, yielding a z‑score of 14.4 for the observed ratio; (2) temporal scrambling (1,000 trials) gives z = 10.7; (3) random entry with identical payoff yields mean P&L 137 ± 201 bps (z = 4.9). Parameter sensitivity analysis (±50 % perturbations of four key thresholds) shows profitability persists in all 20 variations, with the worst‑case P&L reduction of 40 %.

The authors acknowledge several limitations: the study is confined to a single ETF over a short period, with a large share of profits concentrated on a single day; execution assumptions (instant fills, fixed 0.57 bps cost) may not hold at scale; and the market was in a low‑volatility regime (VIX 14‑22), leaving performance in stressed markets unknown.

From a theoretical perspective, the findings do not contradict weak‑form market efficiency because predicting the size of a move is distinct from predicting its direction. The entropy measure captures the presence of informed trading activity (structured order flow) but, due to its invariance under label permutation, cannot reveal whether the information is bullish or bearish. This aligns with Kyle and Glosten‑Milgrom models where informed traders affect prices through persistent order flow while the market cannot infer the sign of their private information.

The paper suggests that order‑flow entropy could serve as a state variable for intraday volatility timing, but broader validation—across assets, longer horizons, live execution, and high‑volatility environments—is required before practical deployment. The contribution lies in linking an information‑theoretic signal to a microstructure‑based prediction of return magnitude while rigorously demonstrating the absence of directional bias.