arXane › Quantitative Finance › Quantitative Finance
Wednesday, 4 November 2026: 1 new paper, 1 cross-list and 1 replacement in Portfolio Management (q-fin.PM), in arXiv's order, with full abstracts. The same papers as arXiv's own listing, set as a newspaper; every title links to arXiv.
Hierarchical clustering and Nested Clustered Optimization (NCO) mitigate the mathematical ill-conditioning of classical Markowitz covariance matrices. However, two structural vulnerabilities persist in equity markets: (1) symmetric variance penalizes upside and downside return dispersion identically, inducing acute return drag against soaring momentum sectors; and (2) unconstrained Hidden Markov Models (HMM) exhibit hyper-active regime switching (whipsaw instability), incurring excessive turnover and transaction drag. We propose Asymmetric Downside-Risk Nested Clustered Optimization with Sticky Regime Dynamics (Asym-Sticky-NCO) for tactical sector rotation across all 11 S&P 500 GICS sectors. The framework substitutes symmetric covariance with the Bawa-Lindenberg downside semi-covariance matrix across both hierarchical tree clustering and two-stage allocation. Furthermore, we introduce a Sticky Gaussian HMM incorporating a diagonal self-transition persistence prior (\k{appa}) and confidence-adaptive boundaries ({\delta}) to enforce macroeconomic persistence. Strictly non-anticipative walk-forward backtesting across 65 out-of-sample months (August 2020 to December 2025) demonstrates that the architecture elevates CAGR from 10.65% (Plain NCO) to 13.41% (+276 bps/year), delivers a 97.74% total return versus 72.99%, suppresses maximum drawdown to 14.22% (versus 23.93% for the S&P 500), and achieves a Calmar ratio of 0.94 (+38.2% over S&P 500; +84.3% over Baseline Regime-NCO). Crucially, sticky dynamics cut regime whipsaws by 75.0% (from 16 to 4 switches) and reduce monthly turnover from 60.78% to 30.48%, addressing core limitations of machine learning asset allocation.
Portfolio use of machine predictions must often proceed while the predictions' economic direction, reliability, and timing remain untrusted, and while feedback from overlapping multi-period positions arrives with delay. A controller chooses among Safe, Follow, and Reverse portfolio actions, so signal orientation is learned rather than assumed. A share-adaptive master updates action probabilities from matured Safe-relative rewards, an evidence gate can abstain by transferring unsupported directional probability to Safe, a delayed collateral budget reserves worst-case loss capacity before exposure is funded, and a protected split account caps the capital adaptive decisions may touch. The architecture carries a dual certificate, tracking regret against switching action sequences under delay and one-sided intervention, together with a pathwise collateral safety bound that holds when predictions fail. On a 24-stock development universe a conservative prediction-assisted overlay stays near equal-weight performance at a fraction of the turnover. On a frozen 82-stock holdout, the pre-specified Reverse orientation fails to outperform Equal Weight; a separately labeled post-holdout diagnostic shows that permanently flipping the sign to Follow also does not recover a return edge. Under the post-holdout controller, protected wealth remains within 15.9 basis points of the Safe benchmark throughout. The contribution is safe adaptive use of untrusted predictions, not demonstrated alpha.