Paper: arXiv 2509.16707

Authors: Sid Ghatak, Arman Khaledian, Navid Parvini, Nariman Khaledian

Abstract

There are inefficiencies in financial markets, with unexploited patterns in price, volume, and cross-sectional relationships. While many approaches use large-scale transformers, we take a domain-focused path: feed-forward and recurrent networks with curated features to capture subtle regularities in noisy financial data. This smaller-footprint design is computationally lean and reliable under low signal-to-noise, crucial for daily production at scale. At Increase Alpha, we built a deep-learning framework that maps over 800 U.S. equities into daily directional signals with minimal computational overhead. The purpose of this paper is twofold. First, we outline the general overview of the predictive model without disclosing its core underlying concepts. Second, we evaluate its real-time performance through transparent, industry standard metrics. Forecast accuracy is benchmarked against both naive baselines and macro indicators. The performance outcomes are summarized via cumulative returns, annualized Sharpe ratio, and maximum drawdown. The best portfolio combination using our signals provides a low-risk, continuous stream of returns with a Sharpe ratio of more than 2.5, maximum drawdown of around 3%, and a near-zero correlation with the S&P 500 market benchmark. We also compare the model’s performance through different market regimes, such as the recent volatile movements of the US equity market in the beginning of 2025. Our analysis showcases the robustness of the model and significantly stable performance during these volatile periods. Collectively, these findings show that market inefficiencies can be systematically harvested with modest computational overhead if the right variables are considered. This report will emphasize the potential of traditional deep learning frameworks for generating an AI-driven edge in the financial market.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.5)

Why this score: The paper uses established deep learning architectures (feed-forward, recurrent networks) without heavy mathematical derivations, focusing on practical implementation rather than novel theory. It demonstrates high empirical rigor with detailed backtesting, risk metrics (Sharpe >2.5, max drawdown ~3%), regime analysis, and performance evaluation across 814 equities, though it omits full disclosure of proprietary model details.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Uncover & exploit subtle patterns<br>in financial markets"] --> B["Methodology<br>Domain-focused DNN: FFNN & RNNs<br>Curated Features"]
  B --> C["Data & Inputs<br>Price, Volume, Cross-sectional<br>800 US Equities Daily"]
  C --> D["Computation<br>Minimal overhead, robust to low SNR"]
  D --> E["Backtesting & Metrics<br>Sharpe Ratio, Max Drawdown, Correlation"]
  E --> F{"Outcomes"}
  F --> F1["Sharpe > 2.5"]
  F --> F2["Max Drawdown ~3%"]
  F --> F3["Near-zero correlation with S&P 500"]
  F --> F4["Stable in volatile markets"]

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 deep‑learning framework that generates daily directional signals for over 800 U.S. equities using feed‑forward and recurrent networks with curated features.

Source sentence
we built a deep-learning framework that maps over 800 U.S. equities into daily directional signals

Assumes 2

  • The signal generated at the end of day t is intended for trading on day t+1. model
    Source sentence
    Since positions are opened one business day after the signal is generated, we assume the signal pertains to day t + 1.
  • The tracking process provides a true ex‑ante prediction record free of look‑ahead bias and information leakage. execution
    Source sentence
    This rigorous tracking ensures a true ex-ante prediction record, immune to post-hoc adjustment or backtest contamination, and serves as a strong guardrail against look-ahead bias and information leakage.

Data

Backtest sample: 814 US equities · equities · daily

  • EODHD price data · www.eodhd.com market-prices commercial
    28 June 2021 to 30 June 2025 · 814 U.S. equities
    Source sentence
    We collected daily OHLC prices for all 814 U.S. equities from 28 June 2021 to 30 June 2025.

Costs

Not modeled A trading or execution result is reported without costs.

Finds 2

Reported performance (MDDEW portfolio): Sharpe 2.54 · total return 26.4% · max drawdown 3.0%

  • The best portfolio combination using the signals achieves a Sharpe ratio above 2.5, maximum drawdown around 3%, and near‑zero correlation with the S&P 500. performance
    Source sentence
    The best portfolio combination using our signals provides a low-risk, continuous stream of returns with a Sharpe ratio of more than 2.5, maximum drawdown of around 3%, and a near-zero correlation with the S&P 500 market benchmark.
  • Applying 2× leverage (with 4% annual cost of capital) raises the MDDEW portfolio’s cumulative return to approximately +47% versus +26.38% without leverage. performance
    Source sentence
    With a 2x leverage factor, the MDDEW portfolio’s cumulative return at the end of the sample is approximately +47%, with a peak gain near +49%. This represents an increase in absolute nominal returns compared to the non-leveraged MDDEW portfolio’s cumulative return of +26.38%.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 20 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.

Cited References (13)

  1. High resolution subject-specific mitral valve imaging and modeling: Experimental and computational methods (2016) — link
  2. Algorithmic trading and financial forecasting using advanced artificial intelligence methodologies (2022)
  3. Automated trading systems statistical and machine learning methods and hardware implementation: a survey (2019)
  4. High frequency trading strategies, market fragility and price spikes: an agent based model perspective (2019)
  5. Grid trading system robot (GTSbot): A novel mathematical algorithm for trading FX market (2019)
  6. Intelligent forecasting with machine learning trading systems in chaotic intraday Bitcoin market (2020)
  7. Lookahead bias in pretrained language models (2024)
  8. R e (Visiting) Large Lan guage Models in Finance (2024)
and 5 more
  1. Can chatgpt forecast stock price movements? return predictability and large language models — → 2304.07619
  2. A review of reinforcement learning in financial applications — → 2411.12746
  3. LSTM-Based Deep Learning Model for Financial Market Stock Price Prediction (2024)
  4. Improving the Performance of Stock Price Prediction: A Comparative Study of Random Forest, XGBoost, and Stacked Generalization Approaches (2024)
  5. Fin-gan: Forecasting and classifying financial time series via generative adversarial networks (2024)

2 of 13 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.