A regime-switching decision support system for dynamic portfolio optimization
Abstract
We present a regime-switching decision support system (DSS) for dynamic portfolio construction that links market data with investor-specific return goals through four transparent modules. In Module 1 (Data acquisition & features) , monthly market signals (e.g., SPX returns, VIX, 10-year yield, and relative volume) are standardized into a market-state feature vector. In Module 2 (Market-state estimation) , an unsupervised clustering model fitted on multi-decade history yields interpretable regimes, and a supervised classifier maps features to months-ahead regime probabilities. In Module 3 (User-specific layer) , a static risk profile and a goal-tracking rule combine the forecast with a user-defined annual return target to produce a discrete risk level $$u^{(t)}$$ . In Module 4 (Optimization policy) , conditional on $$u^{(t)}$$ , the DSS solves either a robust mean–variance optimization program or a CVaR minimization with asset-class constraints, and it rebalances whenever the risk level or regime forecast changes. We implement the DSS on a diversified universe of 256 liquid U.S.-listed assets, including equities, government and corporate bond ETFs, and commodity and alternative asset ETFs. In two rolling-horizon backtests on this universe—2008–2011 (bear) and 2013–2016 (bull)—the DSS outperforms a 70/30 benchmark and single-optimizer baselines, achieving higher performance metrics. These results demonstrate the practical viability and extensibility of regime-aware, goal-driven portfolio construction.