1.87
Sharpe ratio, against an equal-weight baseline
Published · Grade 11
Regime-Aware Portfolio Optimization Using an AI–Markov Hybrid Model with Multi-Divergence Learning
A portfolio model that knows which market it is in before it decides what to hold.
International Journal for Multidisciplinary Research (IJFMR), Vol. 8, Issue 2, 2026. Mentor: Srijit Mukherjee, AI PhD student, Penn State; supported by Cheenta Academy.
Move to shift the regime
01 Why I did it
The question.
I wanted to know whether a model could admit it did not know which market it was in, and still make a decision. Most optimisers assume one world. M-ROLL assumes several and learns the odds of moving between them, which is also how I think about people's finances: the regime changes, and the plan should know that first.
02 Abstract
What the paper says.
Financial markets shift between regimes, and a portfolio built for one regime fails in the next. M-ROLL (Markov Return Optimization with Learning of Likelihoods) models asset returns as a discrete-time Markov chain over regimes, then uses metaheuristic optimisation to set portfolio weights that minimise the divergence between the empirical and a target return distribution. Ten divergence metrics were evaluated, including Kullback-Leibler, Jensen-Shannon, Hellinger and Wasserstein, to capture asymmetric and heavy-tailed returns. Chi-square tests validated stationarity and memorylessness.
03 Method and results
How, and what came out.
- Discrete-time Markov chains to model regime transitions in sector returns
- Particle swarm and grey wolf optimisation to learn weights that minimise distributional divergence
- Ten divergence metrics compared, Jensen-Shannon chosen
- Chi-square validation of the Markov property
- Tested across five US sectors: technology, finance, healthcare, energy, consumer
260%
improvement in Sharpe over equal weight
44.2% → 28.1%
maximum drawdown, cut by a third
10
divergence metrics evaluated