Predicting Economic Quads through Asset Returns Using Ensemble Machine Learning

This study explores the predictive capacity of financial asset returns in forecasting macroeconomic regime shifts, specifically the transition among Inflation, Expansion, Stagflation, and Deflation (IESD) quadrants. While traditional economic indicators such as GDP, CPI, and interest rates are commonly used to define these regimes, their delayed release limits real-time applicability. Motivated by the forward-looking nature of markets, we develop two Machine Learning (ML) models: a Neural Network and an Ensemble Learning classifier combining Random Forest and Gradient Boosting. These models are trained on a ten-year dataset comprising 22 financial instruments, including equity sector Exchange-Traded Fund (ETFs), bonds, commodities, and volatility indices. To enhance predictive robustness, the study incorporates extensive feature engineering, such as logarithmic returns, rolling volatilities, and moving averages, alongside dimensionality reduction via Principal Component Analysis (PCA). Evaluation is conducted through K-Fold cross-validation and performance metrics including accuracy, F1-score, and ROC AUC. Results indicate that the Neural Network achieves an accuracy of 57% on first predictions, rising to 78% when considering the top-two predicted quads. The Ensemble model outperforms with a 96% accuracy, exhibiting consistent classification across all quadrants. The findings highlight the practical potential of ML models to identify economic cycles using asset return behavior. By treating asset returns as early-warning indicators, the study bridges the gap between market signals and delayed macroeconomic statistics, contributing a novel approach to economic state prediction.