Market Regime Aware Portfolio Analytics for Retail Investors Using AI Techniques

Abstract

Retail participation in Indian equity markets has grown sharply over the past decade, yet regulatory data indicate that a large majority of self-directed retail traders lose money, in part because static buy-and-hold and rule-of-thumb strategies do not adapt to abrupt shifts in market behaviour. This paper develops and evaluates a market-regime-aware portfolio analytics framework aimed at retail investors, built on an unsupervised Gaussian Hidden Markov Model (HMM) applied to the NIFTY 50 index. Using daily log returns and their squared values as observable features, we fit a four-state HMM that identifies latent Bull, Bear, High-Volatility, and Crash regimes over the period January 2014 to June 2024. A regime transition matrix is estimated and combined with regime-conditional return and variance estimates to construct a forward-looking, risk-adjusted trading signal that governs equity exposure. We first evaluate the strategy in-sample, where it outperforms a passive buy-and-hold benchmark (cumulative return 4.17× versus 3.35×; Sharpe ratio 1.09 versus 0.80; maximum drawdown −35.4% versus −40.0%). We then subject the framework to a rigorous walk-forward, out-of-sample validation with periodic retraining. Out-of-sample, the regime strategy reduces volatility (10.9% versus 17.3% annualised) and drawdown (−20.1% versus −40.0%) relative to buy-and-hold, and achieves a higher Sharpe ratio (1.02 versus 0.79). We further test a hybrid extension incorporating a rule-based re-entry trigger following price recovery from a rolling trough (Sharpe 1.08 versus 0.82).