Skip to content
Review Open access

Unsupervised Learning for Financial Market Regime Detection Using Volatility Features

Sep 2026 · Theoretical and Natural Science · 0 citations

Abstract

Financial markets fluctuate under several relatively persistent conditions, and returns, volatility, correlation, and risk patterns vary depending on the conditions. Because market regimes and their boundaries are not directly observable and widely recognized labels are unavailable, unsupervised learning provides a suitable framework for identifying hidden states from financial time series. This review examines the transformation of volatility-related features into model inputs for clustering and latent state detection. In order to illustrate the challenges posed by non-stationarity, outliers, and unequal feature scales, it first explains core variables including log returns, rolling volatility, drawdown, trading volume, implied volatility, and dynamic correlation. Then, it compares K-means, hierarchical clustering, Gaussian mixture models, spectral clustering, and hidden Markov models in terms of assumptions, temporal coherence, probabilistic output, interpretability, and evaluation. According to recent studies, market regime identification approaches are evolving from single-asset, hard-label classifications to multidimensional, probabilistic, rolling, and information-enriched systems. Nevertheless, the results remain sensitive to feature design, window length, state number, initialization, sample period, and validation criteria. Future research should focus on online updating, uncertainty quantification, cross-market robustness, and economically meaningful interpretation.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.