Paper: arXiv 2409.04471

Authors: Kevin Cedric Guyard, Michel Deriaz

Abstract

The Foreign Exchange market is a significant market for speculators, characterized by substantial transaction volumes and high volatility. Accurately predicting the directional movement of currency pairs is essential for formulating a sound financial investment strategy. This paper conducts a comparative analysis of various machine learning models for predicting the daily directional movement of the EUR/USD currency pair in the Foreign Exchange market. The analysis includes both decorrelated and non-decorrelated feature sets using Principal Component Analysis. Additionally, this study explores meta-estimators, which involve stacking multiple estimators as input for another estimator, aiming to achieve improved predictive performance. Ultimately, our approach yielded a prediction accuracy of 58.52% for one-day ahead forecasts, coupled with an annual return of 32.48% for the year 2022.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.0)

Why this score: The paper uses standard machine learning algorithms and PCA with minimal advanced mathematical derivations, but includes comprehensive backtesting over nearly a decade with specific metrics like accuracy and annual returns, plus detailed data preparation steps.

Research Flowchart

  flowchart TD
  Start["<b>Research Goal</b><br>Predict EUR/USD Daily Direction"]
  
  Data["<b>Data Inputs</b><br>EUR/USD Historical Data"]
  
  Feat["<b>Feature Engineering</b><br>Decorrelated vs Non-decorrelated<br>using Principal Component Analysis"]
  
  Models["<b>ML Models</b><br>Individual Estimators +<br>Stacked Meta-Estimators"]
  
  Eval["<b>Evaluation</b><br>Directional Accuracy &<br>Backtested Returns"]
  
  Result["<b>Key Outcomes</b><br>Accuracy: 58.52%<br>Annual Return: 32.48% (2022)"]
  
  Start --> Data
  Data --> Feat
  Feat --> Models
  Models --> Eval
  Eval --> Result

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 comparative analysis of 21 machine learning algorithms with and without PCA and evaluates meta‑estimators that stack multiple models.

Source sentence
The contributions of this paper are significant and include: • A comparative analysis of 21 machine learning algorithms, both with and without applying Principal Component Analysis (PCA) to decorrelate the input features. • An evaluation of these 21 mac…

Assumes

No explicit assumptions found in the text.

Data

Backtest sample: EUR/USD · fx · daily

  • Dataset 1 market-prices unknown
    Source sentence
    Dataset 1 was composed exclusively of daily data.
  • Dataset 2 market-prices unknown
    Source sentence
    Dataset 2 combined daily data with the preceding 90 days’ data.
  • Dataset 3 macro unknown
    Source sentence
    Dataset 3 incorporated daily data along with technical indicators from time series of market indices and Forex pairs.

Costs

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

Finds 5

Reported performance (Meta estimators for EUR/USD direction prediction): annual return 32.5%

  • Histogram gradient boosting achieved the highest prediction accuracy of 58.52% for one‑day ahead forecasts. predictive-accuracy
    Source sentence
    In this study, histogram gradient boosting achieved the highest prediction accuracy, scoring 58.52%.
  • Logistic regression yielded the best profit among basic models, delivering an annual return of 29.54% in 2022. performance
    Source sentence
    Logistic regression achieved the best profit, yielding an annual return of 29.54% (Table 2).
  • Using PCA prior to model input reduced both accuracy (max 54.98%) and profit (max 24.05%). negative
    Source sentence
    Using PCA prior to model input resulted in lower accuracy and profit. Indeed, when using PCA, the highest achieved accuracy was 54.98%, and the greatest profit was 24.05% (Table 3).
  • Meta‑estimators achieved an accuracy of 57.23% and a higher annual return of 32.48% in 2022. performance
    Source sentence
    By utilizing data and predictions from basic models, meta estimators achieved an accuracy of 57.23%, slightly lower than that of the basic estimators
  • Feature engineering with domain knowledge (Dataset 3) produced the best accuracy across all model types. statistical
    Source sentence
    Whether for models with PCA, without PCA or meta estimators, dataset 3 reached the best accuracy. This highlights the fact that feature engineering with domain knowledge enhances the models’ understanding of the problem.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 28 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.