{"paper_id":"2511.17304","title":"Law-Strength Frontiers and a No-Free-Lunch Result for Law-Seeking Reinforcement Learning on Volatility Law Manifolds","authors":["Jian’an Zhang"],"paper_date":"2025-11-21","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2511.17304v1","page_url":"https://thequant.space/flowcharts/law-strength-frontiers-and-a-no-free-lunch-result-for-law-se/","scores":{"empirical_rigor":4,"math_complexity":9.5,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes a finite‑dimensional convex volatility law manifold with a metric‑based law‑penalty functional and a Graceful Failure Index, and introduces a Goodhart decomposition of reward into on‑manifold and ghost arbitrage components.","quote":"we construct a finite-dimensional convex volatility law manifold of admissible total-variance surfaces, together with a metric-based law-penalty functional and a domain-agnostic Graceful Failure Index (GFI) that normalizes law degradation under shocks.","summary":"We study reinforcement learning (RL) on volatility surfaces through the lens of Scientific AI. We ask whether axiomatic no-arbitrage laws, imposed as soft penalties on a learned world model, can reliably align high-capacity RL agents, or mainly create Goodhart-style incentives to exploit model…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The structural class S is non‑empty, convex, and law‑consistent with bounded law penalties and GFI.","type":"model","stated":"explicit","quote":"Assumption 3 (Structural class and world-model properties). We assume: 1. Law-consistent structural class. The structural class S ⊂Π is non-empty, convex, and law-consistent in the sense that for all s ∈S, E[Lϕ(s)] ≤LSmax and GFI(s) ≤GFISmax"},{"assumption":"The unconstrained policy class Π is rich enough to strictly contain S and to reach off‑manifold regions.","type":"model","stated":"explicit","quote":"Rich unconstrained class. The unconstrained policy class Π is rich enough to strictly contain S and to reach off-manifold regions"},{"assumption":"The synthetic generator G⋆ produces trajectories that are exactly law‑consistent (each wt lies in Mvol).","type":"data","stated":"explicit","quote":"We call G⋆ a law-consistent generator if the trajectory (wt) it induces satisfies this condition for M = Mvol."},{"assumption":"The recurrent neural world model is trained without any law‑penalty regularization, using only predictive MSE loss.","type":"model","stated":"explicit","quote":"Crucially, no law penalties are used when training the world model: the loss is purely predictive."},{"assumption":"Shock regime changes the distribution of trajectories (via long‑variance and spot‑variance multipliers) while keeping the underlying axioms unchanged.","type":"market","stated":"explicit","quote":"To probe robustness and graceful failure, we introduce a shock regime in which the same axioms hold, but the volatility regime is stressed."}]},"datasets":{"status":"available","items":[{"name":"Law‑consistent synthetic generator (G⋆)","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"A synthetic generator produces trajectories that are exactly law-consistent, while a recurrent neural world model is trained without law regularization and therefore predicts surfaces that deviate from the law manifold in structured ways."}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Ghost‑arbitrage incentive theorem shows naive PPO‑type RL is driven to increase off‑manifold reward whenever ghost arbitrage is available.","kind":"theoretical","quote":"a ghost-arbitrage incentive theorem showing that naive PPO-type RL is structurally driven to increase E[r⊥] whenever ghost arbitrage is available"},{"finding":"Law‑strength trade‑off theorem proves that increasing the weight on law penalties inevitably worsens P&L beyond a quantifiable threshold.","kind":"theoretical","quote":"a law-strength trade-off theorem establishing that increasing the weight on law penalties inevitably worsens P&L beyond a quantifiable threshold"},{"finding":"No‑free‑lunch theorem states that under a law‑consistent world model and law‑aligned strategy class, unconstrained law‑seeking RL cannot Pareto‑dominate structural baselines on P&L, penalties, and GFI.","kind":"theoretical","quote":"a no-free-lunch theorem: under a law-consistent world model and law-aligned strategy class, unconstrained law-seeking RL cannot Pareto-dominate structural baselines on P&L, penalties, and GFI"},{"finding":"Empirically, all law‑seeking RL variants underperform structural baselines, occupying high‑penalty, high‑GFI regions, while baselines lie on the empirical law‑strength frontier.","kind":"performance","quote":"all law‑seeking RL variants underperform and move into high-penalty, high-GFI regions. Volatility thus provides a concrete case where reward shaping with verifiable penalties is insufficient for robust law alignment."},{"finding":"In the shock regime, RL policies in the same law‑penalty band have negative Sharpe and GFI > 1.5, whereas Zero‑Hedge maintains high Sharpe and GFI ≈0.","kind":"performance","quote":"Within the law-penalty band [0.0053, 0.0057] highlighted, Zero-Hedge lies near the lower boundary with high Sharpe and GFI ≈0, while the closest RL variants in the band have Sharpe < 0 and GFI > 1.5"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":28,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:21:18Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}