Forecasting Intraday USD/CAD Exchange Rate with News-Derived Monetary-Policy Signals

Battle Count: 0

Highly relevant for intraday FX strategies. The paper demonstrates that while level prediction is difficult (random walk dominance), directional prediction and uncertainty quantification benefit significantly from specific communication signals (timing and activity). The attribution framework helps traders identify which news features actually add alpha versus noise.

Moderate to High. Requires integrating LLM APIs for sentiment extraction, complex temporal feature engineering (impulse, decay, rolling windows), and implementing a rigorous rolling-window ablation study with FDR correction. Standard ML libraries are used, but the pipeline orchestration is non-trivial.

The paper provides detailed experimental design, fixed random seeds, and specific library versions (pandas, NumPy, scikit-learn, LightGBM, XGBoost). It uses public APIs (OpenAI, TheNewsAPI, Yahoo Finance). However, the specific news dataset (927 articles) and exact prompt engineering details for the LLM might require reconstruction if not explicitly shared in a repository.