Paper: arXiv 2504.13522

Authors: Yunhua Pei, John Cartlidge, Anandadeep Mandal, Daniel Gold, Enrique Marcilio, Riccardo Mazzon

Abstract

Accurate forecasting in financial markets requires integrating diverse data sources, from historical prices to macroeconomic indicators and financial news. However, existing models often fail to align these modalities effectively, limiting their practical use. In this paper, we introduce a transformer-based deep learning framework, Cross-Modal Temporal Fusion (CMTF), that fuses structured and unstructured financial data for improved market prediction. The model incorporates a tensor interpretation module for feature selection and an auto-training pipeline for efficient hyperparameter tuning. Experimental results using FTSE 100 stock data demonstrate that CMTF achieves superior performance in price direction classification compared to classical and deep learning baselines. These findings suggest that our framework is an effective and scalable solution for real-world cross-modal financial forecasting tasks.

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 employs advanced mathematics, including transformers, tensor representations, and Lasso regression for feature selection, indicating high complexity. It demonstrates strong empirical rigor by using real-world FTSE 100 data, reporting specific performance metrics (precision, recall, F1), and mentioning a GitHub repository and collaboration with an industry partner (Stratiphy) for prototyping.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Cross-Modal Financial Forecasting"] --> B["Data Collection:<br>FTSE 100, Macro Indicators, News"]
  B --> C["Methodology:<br>Cross-Modal Temporal Fusion<br>Transformer Framework"]
  C --> D["Computational Process:<br>Tensor Interpretation &<br>Auto-Training Pipeline"]
  D --> E["Outcome:<br>Superior Price Direction<br>Classification"]
  E --> F["Conclusion:<br>Scalable Cross-Modal<br>Solution"]

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 Cross-Modal Temporal Fusion (CMTF), a transformer‑based deep learning framework that fuses structured and unstructured financial data for market forecasting.

Source sentence
we introduce a transformer-based deep learning framework, Cross-Modal Temporal Fusion (CMTF), that fuses structured and unstructured financial data for improved market prediction.

Assumes 3

  • Binary classification assumes the closing price never stays the same on consecutive days. data
    Source sentence
    The data contains no instance where daily change in close price is exactly zero.
  • Labels are defined as 1 if the next‑day price increase is positive, otherwise 0. model
    Source sentence
    true_directioni t+1 = ( 1, if pi t+1 −pi t > 0 0, if pi t+1 −pi t < 0 )
  • The forecasting task is a binary direction prediction of the next‑day closing price. market
    Source sentence
    In this work, we tackle the classification task of financial market forecasting, in which the objective is to predict the direction of the movement of the stock price: whether the stock price will go up or down the next day.

Data

  • FTSE 100 stock data market-prices unknown
    FTSE 100 constituents
    Source sentence
    Experimental results using FTSE 100 stock data demonstrate that CMTF achieves superior performance in price direction classification compared to classical and deep learning baselines.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • CMTF achieves the highest recall of 84.88% and an F1‑score of 0.64, outperforming all baselines in the direction‑prediction task. performance
    Source sentence
    In classification, our proposed CMTF framework exhibits the highest recall of 84.88% and an F1-score of 0.64, outperforming all baselines.
  • CMTF improves precision by 1.52%, recall by 30.38%, and F1 by 0.17 compared with baseline methods on the FTSE 100 classification task. predictive-accuracy
    Source sentence
    The experimental results demonstrate that CMTF outperforms a suite of baselines in forecasting the next trading day close price, with average improvements of 1.52% in precision, 30.38% in recall, and 0.17 in F1 score for the classification task.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 46 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 by independent authors)

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.