Shu Yu and Mulvey 2024
A study by Yizhan Shu, Chenyu Yu and John M. Mulvey (Princeton University, ORFE), published in the Journal of Asset Management in 2024 (preprint arXiv:2402.05272; the preprint was retitled from “Regime-Aware Asset Allocation” to “Downside Risk Reduction Using Regime-Switching Signals” on the way to publication). It compares regime-identification models for a “0/1” stock-versus-cash equity-timing strategy on the S&P 500, DAX and Nikkei 225 over 1990-2023, with a conservative 10bp one-way transaction cost and trading-delay robustness checks of 1, 5 and 10 days. It benchmarks a two-state Gaussian hidden Markov model against the Statistical Jump Model.
The headline result is that the JM-guided strategy beat the HMM-guided strategy on every index — S&P 500 return 11.2% vs 8.5% (Sharpe 0.68 vs 0.54), DAX 8.6% vs 6.4%, Nikkei 4.7% vs 2.5% — and cut maximum drawdown below both the HMM and buy-and-hold (e.g. S&P 500 -26.6% vs the index’s -55.2%). The mechanism is turnover: the JM’s persistent regimes produced turnover of just 44-72% against the HMM’s 141-290%, and that churn is what eroded the HMM’s net edge. Read honestly, this makes the JM a superior risk filter, not a large alpha source: the JM’s return advantage over buy-and-hold is modest on the S&P 500 and comes mainly from drawdown control on the DAX and Nikkei. The paper appears in this vault as the primary trading-relevant evidence that all regime models suffer detection latency — a ~25-day median lag in detecting turning points; during the COVID-19 crash the online signal lagged the onset and end by roughly half a month, missing the rebound but avoiding a ~20% drawdown.
Two caveats temper the result. The jump penalty was tuned by time-series cross-validation to maximise the very 0/1 strategy being evaluated, so the JM-vs-HMM margin is partly a design choice rather than clean out-of-sample alpha. And the JM-beats-HMM trading result rests on this single research group; no fully independent replication with costs exists. The strategy does, however, clear genuine out-of-sample testing with realistic costs, a benchmark and drawdown metrics — hence a moderate profitability grade.
Shu Yu and Mulvey 2024 [reports_profitability] Statistical Jump Model Shu Yu and Mulvey 2024 [reports_underperformance] Hidden Markov Model Regime Detection Shu Yu and Mulvey 2024 [includes_costs] Transaction Costs and Slippage
Connections
- Statistical Jump Model — reports_profitability, 1990-2023, source: https://doi.org/10.1057/s41260-024-00376-x
- Hidden Markov Model Regime Detection — reports_underperformance, 1990-2023, source: https://doi.org/10.1057/s41260-024-00376-x
- Markov Regime-Switching Model — compares_benchmark, source: https://doi.org/10.1057/s41260-024-00376-x
- Regime-Based Asset Allocation — tests_strategy, 1990-2023, source: https://doi.org/10.1057/s41260-024-00376-x
- Transaction Costs and Slippage — includes_costs, source: https://doi.org/10.1057/s41260-024-00376-x
- Real-Time Regime Identification Lag — lacks_live_evidence, source: https://doi.org/10.1057/s41260-024-00376-x
- S&P 500 — trades_market, source: https://doi.org/10.1057/s41260-024-00376-x
- DAX — trades_market, source: https://doi.org/10.1057/s41260-024-00376-x
- Nikkei 225 — trades_market, source: https://doi.org/10.1057/s41260-024-00376-x
- Shu Yu and Mulvey 2024 Dynamic Allocation — precedes, source: https://arxiv.org/html/2406.09578v1
- Shu and Mulvey 2024 Dynamic Factor Allocation — relates, sibling jump-model paper extending the approach to factor allocation, source: https://arxiv.org/abs/2410.14841
- Steelman — Best Case for Markov Trading Models — reports_profitability, cited as supporting-tier pro-regime evidence, source: https://doi.org/10.1057/s41260-024-00376-x