Machine Learning and the Random Walk Puzzle: Forecasting the CAD/USD Exchange Rate with Expanding Window Evaluation and SHAP Interpretability

In three linesStudy comparing 5 ML models (linear regression, random forest, gradient boosting, XGBoost, AdaBoost) to forecast monthly CAD/USD rate (2017-2026, 113 observations). Only linear regression statistically outperforms random walk (DM=3.06, p=0.0071). Random Forest achieves MAPE=1.17%. SHAP shows short lags (lag1-2) and rolling means dominate predictions.

Summary generated by Claude — human-verified