Paper: arXiv 2512.15720
Authors: Mainak Singha
Abstract
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.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper uses advanced mathematical concepts like Markov chains, information theory, and stationary distributions, and includes formal theorems and derivations (math-heavy), while also presenting detailed backtest results, walk-forward validation, robustness checks, and specific trading rules (data/implementation-heavy).
Research Flowchart
flowchart TD A["Research Goal:<br>Predict Price Move Magnitude<br>without Direction"] --> B["Data & Inputs<br>38.5M SPY Trades<br>36 Trading Days<br>1-Second Resolution"] B --> C["Key Methodology<br>15-State Markov Transition Matrix<br>Compute Order-Flow Entropy"] C --> D["Computational Process<br>Real-time Entropy Calculation"] D --> E["Analysis<br>Condition on Low Entropy<br>5th Percentile Threshold"] E --> F["Key Findings / Outcomes"] F --> G["Predictability: 5-Min Abs Return<br>2.89x Increase (t=12.41, p<0.0001)"] F --> H["Non-Predictability: Direction<br>45.0% Accuracy (Chance Level)"] F --> I["Robustness<br>Walk-forward validation passed<br>Placebo test z=14.4"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a time‑dependent order‑flow entropy measure that predicts intraday return magnitude without providing directional information.
Source sentence
We formalize this insight by constructing a Time-Dependent Entropy measure from tick‑level market data.
Assumes 6
- Standard microstructure assumptions underpin the theoretical results. market
Source sentence
Theorem 1 (Magnitude Predictability). Under standard microstructure assumptions, E [|rt,t+∆| | Ht < H] > E [|rt,t+∆|] for sufficiently low H.
- Informed traders are equally likely to be buying or selling. model
Source sentence
Theorem 2 (Directional Unpredictability). Under the same assumptions, E [sgn(rt,t+∆) | Ht] = 0 when informed traders are equally likely to be buying or selling.
- Entropy is invariant under sign permutation of buy/sell labels. statistical
Source sentence
The proofs rely on entropy’s permutation invariance: swapping “buy” and “sell” labels transforms one transition matrix into another with identical entropy.
- Execution occurs with immediate fills and fixed transaction costs. execution
Source sentence
Third, execution assumptions (immediate fills, fixed costs) may not hold at scale.
Show 2 more
- Row i with no observed transitions is assigned a uniform probability of 1/15 for all j. model
Source sentence
When row i has no observed transitions, ˆpij = 1/15 for all j.
- Training and test windows are non‑overlapping and thresholds are frozen after training. other
Source sentence
All thresholds estimated on training data only; frozen for out‑of‑sample testing. Five folds total, covering October 15 to November 18.
Data
2025-10 – 2025-11 · SPY (S&P 500 ETF) · equities · intraday
- SPY tick‑level trades order-book commercial
Source sentence
We tested these predictions using 38,509,593 SPY trades from October 1 to November 19, 2025 (36 trading days), aggregated to second‑resolution bars (828,907 observations).
Costs
Modeled
0.57 bps round‑trip (half‑spread entry + half‑spread exit + 0.30 bps slippage + 0.10 bps exchange fees).
Source sentence
Transaction costs: 0.57 bps round‑trip (half‑spread entry + half‑spread exit + 0.30 bps slippage + 0.10 bps exchange fees, calibrated to SPY’s typical 0.17 bps quoted spread).
Finds 3
total return 11.3%
- Conditioning on entropy below the 5th percentile raises 5‑minute absolute returns by a factor of 2.89 (t = 12.41, p < 0.0001). predictive-accuracy
Source sentence
Conditioning on entropy below the 5th percentile yields 15.3 bps—a factor of 2.89 above the unconditional mean (t = 12.41, p < 10−4).
- Directional accuracy of the entropy‑based rule is 45.0%, statistically indistinguishable from random guessing (p = 0.12). statistical
Source sentence
Crucially, directional accuracy is 45.0% (108/240 out‑of‑sample trades), statistically indis‑tinguishable from 50% (z = −1.55, p = 0.12; Figure 2).
- Out‑of‑sample walk‑forward validation yields cumulative returns of +1,126 bps across 240 trades, with 87.8% of profit from timing and 0.0% from direction. performance
Source sentence
A simple asymmetric‑payoff rule—entering when entropy is low and using tight stop‑losses—yields cumulative out‑of‑sample returns of +1,126 bps across 240 trades (Figure 3). Profit attribution confirms: 87.8% from timing
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.