Paper: arXiv 2512.12924

Authors: Gagan Deep, Akash Deep, William Lamptey

Abstract

We develop a rigorous walk-forward validation framework for algorithmic trading designed to mitigate overfitting and lookahead bias. Our methodology combines interpretable hypothesis-driven signal generation with reinforcement learning and strict out-of-sample testing. The framework enforces strict information set discipline, employs rolling window validation across 34 independent test periods, maintains complete interpretability through natural language hypothesis explanations, and incorporates realistic transaction costs and position constraints. Validating five market microstructure patterns across 100 US equities from 2015 to 2024, the system yields modest annualized returns (0.55%, Sharpe ratio 0.33) with exceptional downside protection (maximum drawdown -2.76%) and market-neutral characteristics (beta = 0.058). Performance exhibits strong regime dependence, generating positive returns during high-volatility periods (0.60% quarterly, 2020-2024) while underperforming in stable markets (-0.16%, 2015-2019). We report statistically insignificant aggregate results (p-value 0.34) to demonstrate a reproducible, honest validation protocol that prioritizes interpretability and extends naturally to advanced hypothesis generators, including large language models. The key empirical finding reveals that daily OHLCV-based microstructure signals require elevated information arrival and trading activity to function effectively. The framework provides complete mathematical specifications and open-source implementation, establishing a template for rigorous trading system evaluation that addresses the reproducibility crisis in quantitative finance research. For researchers, practitioners, and regulators, this work demonstrates that interpretable algorithmic trading strategies can be rigorously validated without sacrificing transparency or regulatory compliance.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.0/10 (capped from 9.0: the paper states that its data carries survivorship bias; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 9.0)

Why this score: The paper employs advanced mathematical frameworks for walk-forward validation and reinforcement learning, but its core contribution is a rigorous, implementation-heavy empirical protocol with extensive backtesting data, strict information discipline, and open-source code.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Develop Rigorous Validation<br>for Microstructure Trading"] --> B["Methodology: Walk-Forward Framework<br>Signal Gen & Strict OOS Testing"]
  B --> C["Data Input:<br>5 MS Patterns, 100 US Equities<br>2015-2024"]
  C --> D["Computational Process:<br>34 Rolling Windows +<br>Transaction Costs +<br>RL Optimization"]
  D --> E["Outcome 1: Modest Returns<br>0.55% Annualized, Sharpe 0.33"]
  D --> F["Outcome 2: Regime Dependence<br>+0.60% High Vol vs -0.16% Stable"]
  D --> G["Outcome 3: Stat. Insignificant<br>P-value 0.34<br>Reproducible Template"]

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 rigorous walk‑forward validation framework for algorithmic trading that combines interpretable hypothesis‑driven signal generation with reinforcement learning and strict out‑of‑sample testing.

Source sentence
We develop a rigorous walk-forward validation framework for algorithmic trading designed to mitigate overfitting and lookahead bias.

Assumes 6

  • Features, signals, and execution decisions use only data available up to that point in time (strict information set discipline). data
    Source sentence
    First, it enforces strict information set discipline where features, signals, and execution decisions use only data available up to that point in time, preventing lookahead bias.
  • Walk‑forward validation is performed with rolling windows across 34 independent out‑of‑sample test periods. model
    Source sentence
    Second, it employs walk-forward validation with rolling windows, where the system must prove itself repeatedly across 34 independent out-of-sample test periods spanning multiple market regimes.
  • All trades must originate from a human‑interpretable hypothesis expressed in natural language. other
    Source sentence
    Third, it maintains complete interpretability by requiring every trade to originate from a human-interpretable hypothesis expressed in natural language.
  • Transaction costs consist of a $1 fixed commission and 5 basis‑points slippage per share. execution
    Source sentence
    where cfixed = $1 commission and cslippage = 0.0005 (5 basis points).
Show 2 more
  • Positions are limited to a maximum of 5 concurrent holdings, 20% allocation per position, and 50% per sector. execution
    Source sentence
    Position Constraints: Maximum 5 concurrent positions, maximum 20% allocation per position, maximum 50% allocation per sector.
  • The sample excludes delisted or acquired stocks, introducing survivorship bias. data
    Source sentence
    This process introduces survivorship bias (stocks delisted/acquired during 2015-2024 are excluded), which biases results upward.

Data

Backtest sample: 2015 – 2024 · US equities (100 US stocks) · equities · daily

  • Yahoo Finance daily price data · Yahoo Finance market-prices public
    2015-01-02 to 2024-10-31 · 100 US equities
    Source sentence
    All data obtained from Yahoo Finance via yfinance API with standard adjustments for splits and dividends.

Recognised sources: Yahoo Finance

Costs

Modeled Costs are part of the reported headline results.

Fixed $1 commission plus 5 basis‑points slippage per share; orders executed at next‑day open.

Source sentence
where cfixed = $1 commission and cslippage = 0.0005 (5 basis points). Orders placed at day t close execute at day t + 1 open with slippage.

Finds 5

Reported performance (hypothesis-driven microstructure signal framework): Sharpe 0.33 (benchmark 0.86) · annual return 0.6% · max drawdown 2.8%

  • The framework yields modest annualized returns of 0.55% with a Sharpe ratio of 0.33 and maximum drawdown of -2.76%, while beta to the market is only 0.058. performance
    Source sentence
    The system yields modest annualized returns (0.55%, Sharpe ratio 0.33) with exceptional downside protection (maximum drawdown -2.76%) and market‑neutral characteristics (beta = 0.058).
  • Aggregate results are not statistically significant (p‑value 0.34), indicating no reliable profitability across the 34 folds. statistical
    Source sentence
    The null hypothesis H0 : µ = 0 cannot be rejected at conventional significance levels: t‑statistic = 0.96, p‑value = 0.34 (two‑sided).
  • Performance is regime‑dependent: positive quarterly returns of 0.60% during high‑volatility 2020‑2024 period versus a negative -0.16% in stable 2015‑2019 markets. performance
    Source sentence
    The key empirical finding reveals that daily OHLCV‑based microstructure signals require elevated information arrival and trading activity to function effectively.
  • Win rate at the fold level is 41% (14 of 34 folds positive) and trade‑level win rate is 46.5% across 140 trades. performance
    Source sentence
    Trade-level win rate across all folds is 46.5%, with 140 total trades executed.
  • The framework is agnostic to hypothesis source and can incorporate rule‑based, genetic programming, neural network, or large‑language‑model generated hypotheses. theoretical
    Source sentence
    The framework accepts hypotheses from any generator G

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 34 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (2 by independent authors)

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (39)

  1. … and the cross-section of expected returns (2016)
  2. Why most published research findings are false (2005)
  3. Does academic research destroy stock return predictability? (2016)
  4. Replicating anomalies (2020)
  5. Is there a replication crisis in finance? (2023)
  6. Design, testing, and optimization of trading systems (1992)
  7. The evaluation and optimization of trading strategies (2008)
  8. Pseudo-mathematics and financial charlatanism: The effects of backtest overfitting on out-of-sample performance (2014)
and 31 more
  1. The deflated Sharpe ratio: correcting for selection bias, backtest overfitting, and non-normality (2014)
  2. The probability of backtest overfitting (2017)
  3. Backtest overfitting in the machine learning era: A comparison of out-of-sample testing methods in a synthetic controlled environment (2024)
  4. Empirical asset pricing via machine learning (2020)
  5. Deep learning with long short-term memory networks for financial market predictions (2018)
  6. A unified approach to interpreting model predictions (2017)
  7. Dissecting characteristics nonparametrically (2020)
  8. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead (2019)
  9. Interpretable machine learning for imbalanced credit scoring datasets (2021)
  10. Continuous auctions and insider trading (1985)
  11. Liquidity, information, and infrequently traded stocks (1996)
  12. Is information risk a determinant of asset returns? (2002)
  13. One security, many markets: Determining the contributions to price discovery (1995)
  14. Flow toxicity and liquidity in a high-frequency world (2012)
  15. BV-VPIN: Measuring the impact of order flow toxicity and liquidity on international equities markets (2016)
  16. Directional trading in option-to-stock volume and the predictability of future returns (2024)
  17. Learning to trade via direct reinforcement (2001)
  18. Deep direct reinforcement learning for financial signal representation and trading (2017)
  19. A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
  20. Data-snooping biases in tests of financial asset pricing models (1990)
  21. Data snooping, technical trading rule performance, and the bootstrap (1999)
  22. Regime-dependent performance evaluation and portfolio construction using machine learning (2022) — link
  23. " Why should I trust you?" Explaining the predictions of any classifier (2016)
  24. Interpretable machine learning (2020)
  25. Contemporary symbolic regression methods and their relative performance (2021)
  26. Hybrid AI: Combining symbolic and machine learning approaches (2022)
  27. Evaporating liquidity (2012)
  28. Algorithmic trading and information (2011)
  29. Machine learning and the futures markets: A systematic approach to hypothesis-driven artificial intelligence (2020)
  30. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI (2020)
  31. Advanced financial market forecasting: integrating Monte Carlo simulations with ensemble machine learning models (2024) — link

0 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.