K-Means Regime Clustering

K-means regime clustering is an unsupervised machine-learning approach that partitions market or macroeconomic observations into a fixed number of clusters (“regimes”) by minimising within-cluster distance to centroids. Unlike the Markov Regime-Switching Model and Hidden Markov Model Regime Detection, it imposes no probabilistic transition structure and no likelihood model — regimes are defined purely by feature-space similarity. Fuzzy variants (fuzzy c-means and the “modified k-means” of Oliveira et al. 2025) emit regime membership probabilities from centroid distances, giving the soft, uncertainty-aware assignments that hard k-means lacks.

It appears in this vault as a representative non-Markov regime-detection method: the Statistical Jump Model is itself a clustering-based regime model (k-means plus a jump penalty), and Oliveira et al. 2025 uses a modified k-means over FRED-MD macro data for tactical asset allocation. Clustering methods can still be paired with a Markov layer after the fact — Oliveira et al., for instance, estimate a regime transition probability matrix from the discovered clusters. As with all regime detectors, the vault grades its usefulness as Regime Classification separately from any tradeable-edge claim.

Connections

Ontology

K-Means Regime Clustering supports Regime Classification K-Means Regime Clustering relates Statistical Jump Model K-Means Regime Clustering contradicts Markov Regime-Switching Model