History Rhymes: Macro-Contextual Retrieval for Robust Financial Forecasting

Financial markets are inherently non-stationary: structural breaks and macroeconomic regime shifts frequently cause forecasting models to fail when deployed out of distribution. Conventional multimodal approaches that fuse numerical indicators and textual sentiment often lack robustness to such shifts. We introduce macro-contextual retrieval, a retrieval-augmented forecasting framework that grounds each prediction in historically analogous macroeconomic regimes. The approach jointly embeds macroeconomic indicators and financial news sentiment into a shared similarity space, enabling retrieval of precedent periods during inference without retraining. Evaluated on seventeen years of S\&P 500 data and tested out-of-distribution on AAPL and XOM, the method consistently narrows the generalization gap and produces the only positive out-of-sample trading outcomes among tested baselines. Retrieved historical neighbors further provide interpretable evidence chains corresponding to recognizable macroeconomic contexts. These results demonstrate that macro-aware retrieval improves robustness and interpretability in financial forecasting under distributional change.