Paper: arXiv 2410.03913

Authors: John Phan, Hung-Fu Chang

Abstract

This paper investigates the application of machine learning models, Long Short-Term Memory (LSTM), one-dimensional Convolutional Neural Networks (1D CNN), and Logistic Regression (LR), for predicting stock trends based on fundamental analysis. Unlike most existing studies that predominantly utilize technical or sentiment analysis, we emphasize the use of a company’s financial statements and intrinsic value for trend forecasting. Using a dataset of 269 data points from publicly traded companies across various sectors from 2019 to 2023, we employ key financial ratios and the Discounted Cash Flow (DCF) model to formulate two prediction tasks: Annual Stock Price Difference (ASPD) and Difference between Current Stock Price and Intrinsic Value (DCSPIV). These tasks assess the likelihood of annual profit and current profitability, respectively. Our results demonstrate that LR models outperform CNN and LSTM models, achieving an average test accuracy of 74.66% for ASPD and 72.85% for DCSPIV. This study contributes to the limited literature on integrating fundamental analysis into machine learning for stock prediction, offering valuable insights for both academic research and practical investment strategies. By leveraging fundamental data, our approach highlights the potential for long-term stock trend prediction, supporting portfolio managers in their decision-making processes.

Complexity vs Empirical Score

  • Math Complexity: 6.0/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper employs advanced machine learning models (CNN, LSTM) and foundational quantitative finance concepts like DCF, but uses a very small dataset (269 data points) with no mention of rigorous backtesting or transaction costs, placing it in a controlled experimental setting rather than a production-ready environment.

Research Flowchart

  flowchart TD
  A["Research Goal"] --> B["Data & Feature Engineering"]
  B --> C["Model Training"]
  C --> D["Model Evaluation & Comparison"]
  D --> E["Key Findings"]

  subgraph B ["Data & Inputs"]
      B1["Dataset: 269 data points<br>2019-2023"] --> B2["Features: Financial Ratios &<br>Discounted Cash Flow DCF"]
      B2 --> B3["Targets: Annual Stock Price<br>Difference & Current vs Intrinsic Value"]
  end

  subgraph C ["Computational Processes"]
      C1["LSTM"] --> C2["1D CNN"]
      C2 --> C3["Logistic Regression LR"]
  end

  subgraph E ["Outcomes"]
      E1["LR Outperforms LSTM & CNN"]
      E2["Accuracy: ASPD 74.66%<br>DCSPIV 72.85%"]
      E3["Validated Fundamental Analysis<br>for Long-Term Prediction"]
  end

  style A fill:#e1f5e1,stroke:#333,stroke-width:2px
  style E fill:#ffebee,stroke:#333,stroke-width:2px

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 integrating fundamental financial statement data and intrinsic value calculations with machine learning models (LSTM, 1D CNN, Logistic Regression) to predict stock trends.

Source sentence
Unlike most existing studies that predominantly utilize technical or sentiment analysis, we emphasize the use of a company's financial statements and intrinsic value for trend forecasting.

Assumes 4

  • A universally applicable intrinsic value formula can be created using DCF with EV/EBITDA multiples and growth rates. model
    Source sentence
    Although various methods exist to determine a company's intrinsic value across sectors, we aimed to create a formula that is both universally applicable and aligned with investment industry standards.
  • Labels are assigned based on whether historical financial data positively or negatively impacted the stock price over the year. data
    Source sentence
    In the first method, if historical financial data positively impacted the company's stock price over the year, it was labeled as 1; if it had a negative effect, it was labeled as 0.
  • Companies from stable sectors exhibit less volatility, making them suitable for studying the link between fundamentals and price movement. market
    Source sentence
    companies within the “stable” or consumer goods sectors are less volatile than high growth expectation technology companies due to their needs for our daily consumption.
  • Five years of annual fundamental data (2019‑2023) are sufficient to train and test the models. data
    Source sentence
    The fundamental data was retrieved from Yahoo Finance in the range of 5 years from the first day of 2019 to the last day of 2023.

Data

  • Fundamental data from Yahoo Finance · Yahoo Finance fundamentals public
    2019-2023 · stable sector stocks (Industrial, Utility, Consumer Staples, Consumer Discretionary, Dow J
    Source sentence
    The fundamental data was retrieved from Yahoo Finance in the range of 5 years from the first day of 2019 to the last day of 2023.

Recognised sources: Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Logistic Regression achieved the highest average test accuracy: 74.66% for ASPD and 72.85% for DCSPIV. predictive-accuracy
    Source sentence
    Logistics Regression achieved the best average testing accuracy out of three models for both datasets at 74.66% accuracy in
  • The CNN model performed the worst, attaining only 55.36% test accuracy on the ASPD task. performance
    Source sentence
    the CNN model performed the worst among all of them, achieving just 55.36% accuracy on ASPD
  • LSTM achieved a high recall of 92.66% on the DCSPIV task, indicating strong true‑positive detection. performance
    Source sentence
    Among all the LSTM values in both ASPD and DCSPIV, LSTM’s average recall value reaches 92.66% in DCSPIV.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 18 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.