Global Economic Cycles Unveiled: A Hybrid TCN-HMM Approach for Regime Dynamics Across Eight Nations

This study developed a hybrid Temporal Convolutional Network and Hidden Markov Model (TCN-HMM) framework to identify economic regimes across eight nations: Brazil, China, Egypt, Indonesia, Mexico, South Africa, United Kingdom, and United States, using GDP, inflation, policy rates, equity indices, exchange rates, and yield curve factors. The method integrated MSSA-smoothed data with a TCN autoencoder to extract multi-scale temporal features, optimized through sensitivity analysis, followed by an HMM to classify three regimes: Natural State, Substantial Growth, and Restrictive Policy. The TCN-HMM outperformed standalone HMMs in capturing non-linear dynamics and temporal dependencies. Key findings highlighted several insights. Twelve extracted features were optimal, outperforming eight features (the number of original variables), implying that the TCN enhanced regime differentiation through a richer feature representation. Furthermore, the framework demonstrated generalizability across the panel, as developing markets like Brazil and Egypt exhibited volatile inflation, while developed economies showed stable policy transmission. Finally, China maintained very high growth rates across all regimes, and currencies across the panel showed a high correlation with real interest rates rather than with overall economic conditions, reflecting their sensitivity to capital-flow dynamics.