Oliveira et al. 2025
“Tactical Asset Allocation with Macroeconomic Regime Detection” (arXiv:2503.11499, v1 14 March 2025; SSRN 5183762) is by Daniel Cunha Oliveira (University of São Paulo), Dylan Sandfelder and Xiaowen Dong (Oxford / Oxford-Man Institute), André Fujita (São Paulo / Kyushu) and Mihai Cucuringu (UCLA / Oxford Statistics). It classifies monthly market regimes using a modified k-means clustering — a fuzzy-c-means-style wrapper that emits regime probabilities from centroid distances rather than hard labels — over more than 100 monthly FRED-MD US macroeconomic series. The paper is explicit that it is “the first to apply a regime detection model from a large macroeconomic dataset to tactical asset allocation,” extracting regimes from stable macro data rather than noisy asset returns.
The pipeline has three stages: detect the current regime and forecast next-month regime probabilities; convert those into expected returns/volatilities via four forecasting schemes (naive Sharpe-conditioning, ridge regression, mean-variance optimisation, Black-Litterman); and map the forecasts into ten-ETF portfolios under long-only, long-short, long-or-short and mixed sizing. The test is a fixed 48-month rolling window, walk-forward, one-month-ahead over 746 monthly observations spanning 2000-2022. The headline numbers — all volatility-scaled to 10% — are genuinely strong on paper: the best long-only ridge configuration reaches a Sharpe of ~1.505 with a maximum drawdown of just ~-4.4%, against SPY at ~0.818 / ~-33.5%. The authors note explicitly that “regime information primarily enhances return generation capabilities rather than downside risk management,” since maximum-drawdown improvements are statistically insignificant.
It appears in this vault as a recent, methodologically careful pro-regime data point for Regime Classification and Tactical Asset Allocation. Note that, despite the existing draft’s label, the core regime engine is k-means clustering, not a Hamilton-style Markov-switching model — though the paper does build a regime transition probability matrix (a Markov chain over the discovered states) and cites Markov-switching TAA (Ang & Bekaert, Ang & Timmermann) as antecedents. It is graded weak for profitability for three concrete reasons. First, it models no transaction costs and no slippage at all — a material omission for monthly ETF rebalancing across long/short books. Second, its headline statistical test compares structured regimes against random-shuffled-label controls; this demonstrates that the regime labels carry information, not that the strategy produces net-of-cost economic alpha over a passive benchmark. Third, robustness is uneven: long-short variants frequently produce negative Sharpe ratios (mvo_lns_2 at -0.516), the result is concentrated in long-only low-dimension ridge configurations, and replication code is not stated as released. The clustering regime detector is a real contribution; the profitability claim is not substantiated after costs.
Connections
- Regime Classification — reports_profitability, source: https://arxiv.org/html/2503.11499v1
- Tactical Asset Allocation — tests_strategy, source: https://arxiv.org/html/2503.11499v1
- Regime-Based Asset Allocation — tests_strategy, source: https://arxiv.org/html/2503.11499v1
- K-Means Regime Clustering — proposes_model, source: https://arxiv.org/html/2503.11499v1
- Transaction Costs and Slippage — excludes_costs, source: https://arxiv.org/html/2503.11499v1
- Out-of-Sample Backtesting — relates, source: https://arxiv.org/html/2503.11499v1
Ontology
Oliveira et al. 2025 supports Regime Classification Oliveira et al. 2025 tests_strategy Regime-Based Asset Allocation Oliveira et al. 2025 contradicts Transaction Costs and Slippage
Sources
- Oliveira, D. C., Sandfelder, D., Fujita, A., Dong, X. & Cucuringu, M. (2025). “Tactical Asset Allocation with Macroeconomic Regime Detection.” arXiv:2503.11499. https://arxiv.org/html/2503.11499v1
- SSRN preprint (abstract id 5183762, 18 March 2025). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5183762