Machine learning models struggle to beat random walk in USD/CAD exchange rate forecasting

A new study published on arXiv explores the effectiveness of various machine learning models in forecasting the USD/CAD exchange rate against the random walk benchmark. Researchers found that while most machine learning models showed only marginal improvements, linear regression was the only model to statistically outperform the naive random walk. The study utilized daily data from the Bank of Canada, resampled into monthly observations, and employed an expanding-window framework for evaluation. SHAP analysis was used to interpret the best-performing model, revealing that short-term lags and recent rolling means were the dominant predictors, aligning with the near-random-walk nature of exchange rates. AI

RANK_REASON The cluster contains a research paper published on arXiv detailing an academic study on machine learning models for financial forecasting. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• AdaBoost
• Bank of Canada
• Diebold-Mariano (DM) test
• gradient boosting
• Holt-Winters Forecasting for Brazilian Natural Gas Production
• linear regression
• random forest
• Shap
• Shapley Additive Explanations
• USD/CAD
• XGBoost

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →