AIQF Daily digest — 36 new papers

Quant Paper Radar · 14 of 36

1. 1

   Interpretable Allocation and Model-Predictive Rebalancing of Equity Portfolios, Tested on Real and Synthetic Markets

   Jeffrey D. Varner · ARXIV · Trend/Beta Allocation and Event-Triggered Rebalancing

   we backtested it — Sharpe 0.89 · quant relevance 1.00

   The system combines trend/beta stock scores, constrained portfolio allocation, and forecast-band or drawdown-triggered rebalancing, reporting lower drawdowns in most tested combinations but no consistent wealth improvement from early replanning. Its daily-price mechanism is buildable, although an unbiased yearly top-500 universe differs from the study's survivor-filtered current-member sample. Replication must also disclose dividend-adjustment differences, substitution for the unavailable DTB3 series, and that the reported September 2026 endpoint exceeds the platform's stated historical backtest window.

   original paper

2. 2

   Penalize the Common Trade: A Collision-Aware Layer for Learned Portfolios

   Eric Benhamou, Jean-Jacques Ohana, Beatrice Guez et al. · ARXIV · Momentum with common-factor exposure control

   we backtested it — Sharpe 0.74 · quant relevance 1.00

   The study tests a weekly ETF momentum portfolio that penalizes assets with large absolute loadings on the leading recent return principal component. It reports lower drawdown and volatility but nearly unchanged Sharpe; the bootstrap interval for drawdown improvement includes zero. The return-based strategy is reproducible here, but its results do not establish that the common component represents AI crowding.

   original paper

3. 3

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

   Amirhossein MohammadpourAzari, Mohammad Zarei, MohammadAmin Fazli · ARXIV · Online Portfolio Allocation and Benchmark-Relative Risk Control

   we backtested it — Sharpe 0.13 · quant relevance 1.00

   The controller learns whether to follow, reverse, or abstain from a prediction, using delayed rewards, evidence gating, collateral reservations, and a protected benchmark-following account. Reported equity backtests do not establish alpha; the principal empirical result is limited deviation from the Safe benchmark, which is not protection against absolute market losses.

   original paper

The rest of the batch

• Mitigating Return Drag and Regime Whipsaws in Sector Rotation: An Asymmetric Downside-Aware Nested Clustered Optimization Framework

  Supriadi · ARXIV · relevance 1.00

  The framework allocates monthly among 11 US sector ETFs using downside-risk clustered optimization and a persistent two-state market-regime model.

• CosmosAlpha: Multi-Relational Graph Search with Pareto-Guided Exploration for LLM-Driven Alpha Mining

  Zhan-He Wang, Zi-Yang Lang, Yi-Yang Li et al. · ARXIV · relevance 1.00

  CosmosAlpha mines interpretable stock-return signals using LLM-generated formulas, a multi-relational factor graph, and Pareto-guided search.

• Beyond Anonymous Order Flow: Wallet Identity as a Feature for Short-Horizon Return Prediction

  Jingqiu Ding, Daojing Zhai · ARXIV · relevance 0.99

  Persistent wallet identities improve short-horizon return forecasts for Hyperliquid perpetual futures beyond anonymous order-book features.

• PanelGP: A Probabilistic Factor Model for Asset-Return Panels

  Andrea Ruglioni, Damir Filipovic · ARXIV · relevance 0.99

  PanelGP jointly forecasts stock expected returns and covariance using characteristic-dependent Gaussian-process factor loadings and macro-dependent expected factors.

• Thinking Fast, Trading Late: System One Decision Models vs the Market

  Amir Dehkordi, Alireza Kargarzadeh, Nariman Khaledian et al. · ARXIV · relevance 0.99

  This paper directly evaluates whether AI readings of company news generate executable stock-return forecasts under a blinded, pre-registered protocol.

• TensorLOB: Structure-Preserving Tensor Learning for Limit Order Book Forecasting

  Yuxuan Gu, Danilo Mandic · ARXIV · relevance 0.97

  TensorLOB preserves time, bid/ask side, price level, and order attribute as distinct tensor dimensions to forecast short-horizon mid-price movements.

• Optimal portfolio and consumption under no-shorting constraints: The exploratory formulation and reinforcement learning

  Xun Li, Xiang Yu, Fengyi Yuan et al. · ARXIV · relevance 0.97

  The paper develops entropy-regularized reinforcement learning for portfolio allocation and consumption, with nonnegative portfolio-policy means and deterministic modal deployment.

• Can We Stop the Backtest Yet? Anytime-Valid Promotion Tests for Adaptive Financial Research Agents

  Eric Benhamou, Jean-Jacques Ohana, Beatrice Guez et al. · ARXIV · relevance 0.96

  PROMOTE is an implementable sequential gate for promoting trading strategies while controlling false promotions across adaptive candidate creation and repeated monitoring.

• Specification LASSO and a Flexible Characteristics-Based Asset Pricing Model

  Chaohua Dong, Shuyi Ge, Shaoran Li et al. · ARXIV · relevance 0.94

  S-LASSO selects linear, nonlinear, and firm-size interaction effects for characteristics-based cross-sectional return prediction, providing a concrete systematic stock-selection application.

• Herding or Racing? Decomposing Cross-Account Order-Flow Persistence in On-Chain Markets

  Yichong Bai, Dingyue Kan · ARXIV · relevance 0.94

  The study separates cross-wallet order-flow persistence from genuine common-signal trading and links CEX-driven arbitrage flow to liquidity-provider losses.

• The Costs and Benefits of Leveraged ETFs

  Chris Murray, Marco Sammon · ARXIV · relevance 0.94

  The study directly informs leveraged-ETF selection and holding decisions by quantifying volatility drag, financing costs, and historical investor outcomes.

and 22 more · see the full radar