Paper: arXiv 2512.18648
Authors: Sungwoo Kang
Abstract
We demonstrate that the choice of normalization for order flow intensity is fundamental to signal extraction in finance, not merely a technical detail. Through theoretical modeling, Monte Carlo simulation, and empirical validation using Korean market data, we prove that market capitalization normalization acts as a ``matched filter’’ for informed trading signals, achieving 1.32–1.97\(\times\) higher correlation with future returns compared to traditional trading value normalization. The key insight is that informed traders scale positions by firm value (market capitalization), while noise traders respond to daily liquidity (trading volume), creating heteroskedastic corruption when normalizing by trading volume. By reframing the normalization problem using signal processing theory, we show that dividing order flow by market capitalization preserves the information signal while traditional volume normalization multiplies the signal by inverse turnover – a highly volatile quantity. Our theoretical predictions are robust across parameter specifications and validated by empirical evidence showing 482% improvement in explanatory power. These findings have immediate implications for high-frequency trading algorithms, risk factor construction, and information-based trading strategies.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; reported Sharpe ratio of 5.48 is implausibly high for daily or lower-frequency data; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)
Why this score: The paper employs advanced signal processing theory (matched filter) and theoretical modeling with rigorous Monte Carlo simulations and validation on large-scale Korean market data (2.1 million observations), demonstrating both high mathematical density and strong empirical implementation.
Research Flowchart
flowchart TD
A["Research Goal: Optimal Normalization<br>for Order Flow Signal Extraction"] --> B["Key Methodology<br>Monte Carlo Simulation &<br>Empirical Validation (Korean Market)"]
B --> C["Input: Raw Order Flow Data<br>+ Market Microstructure Variables"]
C --> D{"Computational Process:<br>Compare Normalization Methods"}
D --> E["Method 1: Trading Volume<br>Traditional Approach"]
D --> F["Method 2: Market Cap<br>Matched Filter Approach"]
E --> G["Outcome: Signal Corruption<br>Heteroskedastic noise<br>Low correlation"]
F --> H["Outcome: Clean Signal<br>Aligned with informed trade scale<br>High correlation"]
G --> I["Key Findings:<br>Market Cap normalization yields<br>1.32–1.97x higher correlation<br>482% improvement in explanatory power"]
H --> I
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 general matched filter principle for order flow normalization, matching the scaling behavior of the signal-generating process.
Source sentence
We establish a general matched filter principle for order flow normalization: optimal normalization must match the scaling behaviour of the signal-generating process.
Assumes 6
- Daily turnover is heteroskedastic, providing the heteroskedasticity central to the theoretical model. market
Source sentence
Daily turnover exhibits substantial cross-sectional variation (mean 2.6%, std 10.9%), providing the heteroskedasticity central to our theoretical model.
- Order flow scaling differs across investor types (capacity‑scaled vs volume‑scaled). model
Source sentence
the direction of normalization dominance should reverse across investor types whose scaling behaviours differ.
- The KRX daily disclosures accurately capture net order flow by investor category. data
Source sentence
All data sourced from Korea Exchange (KRX) daily disclosures and CREON market data.
- Signal extraction can be framed as SNR maximization using Jensen’s inequality. other
Source sentence
deriving the general matched filter principle from Jensen’s inequality and SNR maximization
Show 2 more
- Monte Carlo simulations with 500 stocks each approximate real‑world heterogeneity. other
Source sentence
We validate these predictions through 1,000 Monte Carlo simulations with 500 stocks each
- Fama‑MacBeth standard errors provide valid inference for the regressions. statistical
Source sentence
t‑statistics in parentheses from Fama‑MacBeth time‑series standard errors
Data
2020 – 2024 · Korean equity market · equities · daily
- Korean market order flow data order-book public
Source sentence
2.7 million stock‑day observations from the Korean market (2020–2024).
- CREON market data market-prices commercial
Source sentence
All data sourced from Korea Exchange (KRX) daily disclosures and CREON market data.
Costs
Not modeled
Source sentence
These are gross, pre‑transaction‑cost Sharpe ratios assuming frictionless daily rebalancing; actual implementable Sharpe ratios would be substantially lower after accounting for bid‑ask spreads, market impact, and short‑selling costs.
Finds 5
Sharpe 5.48 · annual return 39.8% · max drawdown 8.2%
- Monte Carlo simulations show matched‑filter normalizations achieve up to 1.99× higher signal correlation. theoretical
Source sentence
Monte Carlo simulations confirm this principle works bidirectionally, with matched filters achieving up to 1.99×
- Domestic institutional net order flow normalized by market cap predicts next‑day returns (t = 9.65, p < 0.001). predictive-accuracy
Source sentence
domestic institutional flows predict next‑day returns significantly under SMC (t = 9.65)
- Foreign institutional net order flow normalized by trading value predicts next‑day returns (t = 16.35, p < 0.001). predictive-accuracy
Source sentence
foreign flows exhibit stronger predictability under STV (t = 16.35)
- Market‑cap normalization yields R² = 0.29% versus 0.14% for trading‑value normalization (≈2.07× higher). performance
Source sentence
The empirical R2 ratio (0.29%/0.14% = 2.07×) is broadly consistent with, and somewhat exceeds, this prediction
- Long‑short portfolio based on SMC achieves annualised Sharpe 5.48 and mean daily return 15.84 bps (t = 12.11), outperforming STV (Sharpe 3.29, 8.41 bps). performance
Source sentence
The SMC portfolio achieves an annualised Sharpe ratio of 5.48 with mean daily return of 15.84 Basis Points (bps) (t = 12.11), substantially outperforming the STV portfolio (Sharpe 3.29, 8.41 bps daily).
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 27 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.