AIQF Daily digest — 12 new papers

Quant Paper Radar · 8 of 12

1. 1

   Governed Self-Evolution for Autonomous Investing: A Multi-Agent LLM System with an Invariant Core

   Yebo Feng, Jingyang Dai, Liangrui Pan et al. · ARXIV · Multi-Agent Investing and Risk Management

   quant relevance 0.99

   Steward applies a multi-agent LLM architecture to US equity and ETF investing, with outcome-based memory, gated prompt updates and immutable risk controls. It reports forward paper-account trading results, recorded-input replays and historical backtests of its deterministic risk core; these are distinct evaluations, not evidence of independently established live-money alpha.

   original paper

2. 2

   Reinforcement learning-driven optimal capital allocation for intelligent finance

   Yan Luo, Qi Mingchun, Mingshan Cheng · OPENALEX · Reinforcement Learning Portfolio Allocation

   quant relevance 0.99

   The study proposes dynamic actor-critic reinforcement learning with clipped PPO updates for risk-aware portfolio allocation and reports a Sharpe ratio of 1.91. A US-equity implementation is feasible using available daily data, but the supplied excerpt does not establish the exact dataset universe, feature definitions, or evaluation protocol, so those reported results should not be treated as independently validated or directly comparable.

   original paper

3. 3

   Directionally Correct Views in Portfolio Optimization

   Ojas Kumar, Sreejit Dey, Sarvesh Ravichandran Iyer et al. · ARXIV · View-Adjusted Portfolio Optimization

   quant relevance 1.00

   The study incorporates directionally correct expected-return views into maximum-Sharpe and end-to-end portfolio optimization, reporting comparable performance on Indian equities. Its empirical views interpolate toward realized evaluation-period returns, so the results demonstrate the value of hypothetical foresight rather than an implementable forecasting signal.

   original paper

The rest of the batch

• When News Gets Old: Portfolio Allocation with Reinforcement Learning and Large Language Models

  Bolin Song, Surui Xiang · ARXIV · relevance 1.00

  The study compares stale-signal filtering with explicit news-age inputs for weekly US-stock portfolios selected by RL and LLM policies.

• Equilibria and Learning in Competitive Market Making

  Matteo Onger, Luigi Foscari · ARXIV · relevance 0.96

  The study develops a competitive market-making game, characterizes zero-profit equilibrium quotes, and analyzes convergence of no-internal-regret learners on discretized price grids.

• CROWD WISDOM VERSUS WALL STREET: A COMPARATIVE ANALYSIS OF REDDIT-FAVOURED AND INSTITUTIONALLY ACCUMULATED EQUITIES OVER A THREE-MONTH HORIZON

  Saidjon Murodov, Sakinabonu Aliyeva · OPENALEX · relevance 0.95

  The study compares three-month returns of five WallStreetBets attention leaders with five stocks ranked by institutional accumulation, reporting higher but statistically insignificant retail-cohort returns.

• Dynamics of the Liquidity Profile in Concentrated Liquidity Provision: An SPIDE Approach

  Cheuk Yin Lee, Jimmy Risk, Shen-Ning Tung et al. · ARXIV · relevance 0.94

  The study models and forecasts concentrated-liquidity profiles in Uniswap V3, with applications to liquidity-provider capital allocation, pool valuation, and loss-versus-rebalancing risk.

• Dispersion in Analysts’ Target Prices and Stock Returns in the Korean Stock Market

  Joon Chae, Ryumi Kim · OPENALEX · relevance 0.94

  Analyst target-price dispersion predicts higher next-month returns among Korean stocks, suggesting a cross-sectional stock-selection signal.

and 4 more · see the full radar