Machine Learning Forecasting of EUR/USD Trends with Qualitative Cross-Validation from Institutional Reports - paper

Aug 4, 2026

Upload by Eden

This study proposes a comparative framework to forecast weekly EUR/USD direction using daily financial data, combining a baseline Classification and Regression Tree (CART) with an enhanced Random Forest model. Models are trained to classify weekly movements into two categories (+1 for expected increase, -1 for expected decline) and generate predictions every 5 days for the following 5-day window under strict causality. Statistical performance and robustness are evaluated through systematic backtesting, while interpretability is strengthened by an LLM-based qualitative sentiment layer extracted from institutional FX research reports to validate model outputs against expert directional views.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Machine Learning Approaches: CART and Random Forest

Data and Preparation

Results of the CART Model

Interpretation of the Decision Tree

Machine Learning Forecasting of EUR/USD Trends with Qualitative Cross-Validation from Institutional Reports Mohamed Adil Khalifa * E-mail of the corresponding author: [adilkhalifa.ak@gmail.com](adilkhalifa.ak@gmail.com) Abstract This study presents a comparative methodology for forecasting weekly trends in the EUR/USD exchange rate using daily financial data. A first model is calibrated using a Classification and Regression Tree (CART), followed by an enhanced version based on the Random Forest algorithm. The analysis evaluates statistical performance, model robustness, interpretability, and decision-making quality through systematic backtesting. The machine learning models are trained to classify weekly currency movements into two categories: +1 for an expected rise and-1 for an expected decline. A prediction is made every 5 days to anticipate the market direction for the subsequent 5-day period, using only information available at the time of the prediction (ensuring strict causality) . To complement and validate the quantitative results, a qualitative layer is introduced using a large language model (LLM) to extract directional sentiment from institutional FX research reports. This dual approach enhances the interpretability and contextual relevance of the forecasting framework. Keywords: EUR/USD, machine learning, Random Forest, CART, forecasting, large language models, sentiment analysis, time series DOI: 10.7176/EJBM/17-4-06 Publication date: May 30th 2025 1. Introduction Supervised learning models (James et al. 2013) for predicting financial market trends are attracting increasing attention. Currency markets are particularly complex, influenced by a mix of technical, fundamental, and behavioral factors, which complicates the modeling of short-term price dynamics (Fama 1970) . Financial time series such as exchange rates typically display high noise, regime shifts, and persistent volatility. (Tsay 2010) In this context, the present study aims to evaluate the effectiveness of a decision tree model (CART) as a baseline for forecasting weekly EUR/USD movements, and to compare it to a more robust ensemble method: the Random Forest algorithm (Hastie et al. 2009) . These models are trained to generate directional trading signals based on technical indicators derived from daily data. To enrich the predictive framework and enhance model interpretability, the study also incorporates a qualitative validation layer using a large language model (LLM) . Institutional FX research reports are analyzed semantically to extract directional sentiment, allowing for a cross-validation of model outputs with expert market views. The core objective is to determine whether these relatively simple machine learning techniques, when complemented by LLM-based sentiment analysis, can generate signals that are not only statistically sound but also economically actionable within a systematic trading context. 2. Machine Learning Approaches: CART and Random Forest 2.1 Data and Preparation The dataset spans the period from 1990 to 2025, with observations recorded at daily frequency. The target variable is the weekly directional movement of the EUR/USD exchange rate, derived from daily returns. The explanatory variables include technical indicators such as the Relative Strength Index (RSI), Moving Average Convergence Divergence (MACD), Bollinger Bands, moving averages, and a volatility measure. These indicators were computed over short-, medium-, and long-term horizons to capture multiple dimensions of market behavior. Such indicators are widely used in the modeling of financial time series, especially due to their ability to reflect trend strength and momentum (Tsay, 2010) . External variables like the DXY index, VIX, and Brent crude prices were initially considered but excluded due to low correlation and statistically insignificant p-values (see Figures 1 and 2) . Figure 1: Correlation heatmap between EUR/USD daily

FAQs

What prediction target and labeling scheme are used for weekly EUR/USD forecasting?

Weekly directional movement is labeled into two classes: +1 for an expected rise and -1 for an expected decline, derived from daily returns.

How is strict causality enforced when making predictions?

Explanatory variables are lagged by five days, and each prediction is made every 5 days using only information available at that time to prevent data leakage.

How does the qualitative layer work and what is its purpose?

A large language model extracts directional sentiment from institutional FX research reports, providing qualitative validation to complement and interpret the quantitative model outputs.

Aug 4, 2026

Upload by Eden