EXFormer: A Multi-Scale Trend-Aware Transformer with Dynamic Variable Selection for Foreign Exchange Returns Prediction

ArXiv ID: 2512.12727 “View on arXiv”

Authors: Dinggao Liu, Robert Ślepaczuk, Zhenpeng Tang

Abstract

Accurately forecasting daily exchange rate returns represents a longstanding challenge in international finance, as the exchange rate returns are driven by a multitude of correlated market factors and exhibit high-frequency fluctuations. This paper proposes EXFormer, a novel Transformer-based architecture specifically designed for forecasting the daily exchange rate returns. We introduce a multi-scale trend-aware self-attention mechanism that employs parallel convolutional branches with differing receptive fields to align observations on the basis of local slopes, preserving long-range dependencies while remaining sensitive to regime shifts. A dynamic variable selector assigns time-varying importance weights to 28 exogenous covariates related to exchange rate returns, providing pre-hoc interpretability. An embedded squeeze-and-excitation block recalibrates channel responses to emphasize informative features and depress noise in the forecasting. Using the daily data for EUR/USD, USD/JPY, and GBP/USD, we conduct out-of-sample evaluations across five different sliding windows. EXFormer consistently outperforms the random walk and other baselines, improving directional accuracy by a statistically significant margin of up to 8.5–22.8%. In nearly one year of trading backtests, the model converts these gains into cumulative returns of 18%, 25%, and 18% for the three pairs, with Sharpe ratios exceeding 1.8. When conservative transaction costs and slippage are accounted for, EXFormer retains cumulative returns of 7%, 19%, and 9%, while other baselines achieve negative. The robustness checks further confirm the model’s superiority under high-volatility and bear-market regimes. EXFormer furnishes both economically valuable forecasts and transparent, time-varying insights into the drivers of exchange rate dynamics for international investors, corporations, and central bank practitioners.

Keywords: Foreign Exchange, Transformer, Time Series Forecasting, EUR/USD, Multi-scale Attention

Complexity vs Empirical Score

• Math Complexity: 8.0/10
• Empirical Rigor: 8.5/10
• Quadrant: Holy Grail
• Why: The paper employs advanced deep learning concepts like Transformers, multi-scale trend-aware attention, and squeeze-and-excitation blocks, requiring dense mathematical formulation, while providing rigorous empirical validation with multiple currency pairs, sliding-window backtests, transaction cost adjustments, and statistical significance testing.

  flowchart TD
    Start["Research Goal<br>Daily FX Returns Forecasting<br>EUR/USD, USD/JPY, GBP/USD"] --> Inputs["Data & Inputs<br>28 Exogenous Covariates<br>Daily Historical Prices"]
    Inputs --> Method["Core Methodology<br>EXFormer Architecture"]
    Method --> A["Multi-Scale Trend-Aware<br>Self-Attention (Parallel Convs)"]
    Method --> B["Dynamic Variable Selector<br>Time-Varying Weights"]
    Method --> C["Squeeze-and-Excitation Block<br>Feature Recalibration"]
    A & B & C --> Process["Learning & Out-of-Sample Forecasting<br>Sliding Windows & Backtesting"]
    Process --> Results["Key Findings & Outcomes"]
    Results --> R1["Outperforms Baselines & Random Walk<br>Dir. Acc. Gain: 8.5–22.8%"]
    Results --> R2["Backtest Returns (No Costs):<br>18%, 25%, 18%"]
    Results --> R3["Returns w/ Costs (>0):<br>7%, 19%, 9% vs Baseline Negative"]
    Results --> R4["Robust in High Volatility<br>Transparent Driver Insights"]