Safe Portfolio Learning from Untrusted Predictions: Signed Trust, Abstention, and Cross-Universe Polarity Failure

Paper: arXiv 2611.04826

Authors: Amirhossein MohammadpourAzari, Mohammad Zarei, MohammadAmin Fazli

Abstract

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.

Complexity vs Empirical Score

• Math Complexity: 7.0/10
• Empirical Rigor: 7.0/10
• Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper presents a novel framework for safe portfolio learning from untrusted predictions, incorporating advanced online learning theory with practical safety mechanisms. The empirical evaluation, while not demonstrating alpha, rigorously tests the safety and adaptability of the proposed architecture across different market conditions. The dual certificate and emphasis on polarity instability are significant contributions.

Research Flowchart

  flowchart TD
    A[Research Goal: Safe Adaptive Portfolio Learning from Untrusted Predictions] --> B{Key Methodology Steps};
    B --> C[1. Controller Chooses: Safe, Follow, Reverse Actions];
    C --> D[2. Share-Adaptive Master Updates Action Probabilities];
    D --> E[3. Evidence Gate: Abstain if Unsupported Directional Probability];
    E --> F[4. Delayed Collateral Budget & Protected Split Account];
    F --> G[Data/Inputs: 24-stock Development Universe, 82-stock Frozen Holdout];
    G --> H[Computational Process: Portfolio Simulation with Adaptive Controller];
    H --> I[Key Findings/Outcomes: Conservative Prediction-Assisted Overlay near Equal-Weight];
    I --> J[Post-Holdout: Protected Wealth remains within 15.9 bps of Safe Benchmark];
    J --> K[Contribution: Safe Adaptive Use of Untrusted Predictions, not Demonstrated Alpha];