Portfolio system mostly follows rules with 4.8% rule violations on held days
A system learns to follow six rules that depend on market conditions, without post-hoc corrections.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Zeynep Yıldız, Satchit Chatterji, Ercan Atam, Erman Acar
- Universities and research institutions
- Boğaziçi University,Istanbul,Türkiye, Finesse Lab, University of Amsterdam,Amsterdam,the Netherlands, Boğaziçi University, Finesse Lab, University of Amsterdam
Institution metadata: Original paper ↗
What they did and found
Researchers trained a portfolio agent to follow six rules that change with market regimes. When evaluated on newer data, the agent violated the rules on about 4.8% of days, far fewer than learners with no constraints.
Why it matters
This shows a learning tool can follow set rules more often, helping manage risk without constant manual checks.
During training, a safety goal makes the policy stay within limits. It does not depend on live checks, keeping trading simple and fast.
In tough markets, strict rules can reduce gains but help limit losses and make behavior predictable across different regimes.
Original sources · 1
- Internalizing Investment Mandates: A Neurosymbolic RL Approach to Portfolio Allocation ↗arXiv · 2026-11-12
Check the original paper for its authors, methods, version and access terms.
What remains uncertain
Findings come from a specific market history and a fixed six-rule set; results may differ with other mandates or longer time frames.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.