AIQF Daily digest — 13 new papers

Quant Paper Radar · 5 of 13

1. 1

   Hard to Fool: A Bounded LLM Research Loop for Quantitative Finance, and What Two Dozen Campaigns Found

   Henry Zhang · ARXIV · LLM-Assisted Strategy Discovery and Backtest Validation

   we backtested it — Sharpe 0.64 · quant relevance 0.98

   The paper presents an auditable LLM-driven trading-research framework with frozen evaluation gates, chronological splits and a single-use holdout, and reports mostly unsuccessful campaigns. Its daily ETF research component is reproducible here; its intraday options, Kalshi, off-exchange-volume and proprietary alternative-data campaigns are not. A new ETF experiment would test the framework, not reproduce the reported option results or establish a profitable strategy.

   original paper

2. 2

   Optimizing Trading Strategies with Genetic Algorithms

   Diogo Da Silva, Nuno Mateus-Coelho · OPENALEX · Genetic Algorithm Technical-Rule Optimization

   quant relevance 1.00

   EvoTrade uses genetic algorithms to evolve daily technical-indicator trading rules for BTC, XRP, and ETH. The approach is implementable with available crypto price bars, but reported TradingView performance includes a negative Sharpe ratio despite a high win rate, and substantial discrepancies across simulators weaken the evidence of robustness.

   original paper

3. 3

   When the Market Changes Its Mind: Regime Fragility in Machine-Learning Forecasts of a Simulated NSE Index

   Evaan Behal · OPENALEX · Machine-Learning Direction Forecasting / Regime Robustness

   we backtested it — Sharpe 0.52 · quant relevance 0.94

   The study tests daily machine-learning direction forecasts and their trading value in a synthetic, NSE-like market where momentum reverses during turbulent regimes. Its reported accuracy and cost-adjusted performance are simulation results, not evidence from real NSE trading; the forecasting framework can be tested on available US ETF data.

   original paper

The rest of the batch

• Constrained defined-contribution pension management: A reparameterized neural network approach

  Yueman Feng, Wenyuan Li · ARXIV · relevance 0.94

  The study develops neural-network primal–dual methods for constrained pension allocation across cash, nominal bonds, inflation-linked bonds, and equities, jointly with term life insurance.

• Learning PDEs for Portfolio Optimization with Quantum Physics-Informed Neural Networks

  Letao Wang, Abdel Lisser, Sreejith Sreekumar et al. · OPENALEX · relevance 0.91

  The study develops quantum and quantum-inspired physics-informed neural networks to solve the Merton portfolio-allocation HJB equation.

and 8 more · see the full radar