When Is Inaction a Mistake? Continuation-Aware Auditing of PPO Trading Policies

Xingfei Zeng; Xin Zhong; Nanting Li; Ziyang Zhong; Lei Xiao; Guanghui Lu

Home/Archive/arXiv

arXiv · 2609.12536

When Is Inaction a Mistake? Continuation-Aware Auditing of PPO Trading Policies

Xingfei Zeng·Xin Zhong·Nanting Li·Ziyang Zhong·Lei Xiao·Guanghui Lu

View original source ↗Read source document ↗Export metadata ↓

Abstract

An optimal reference may recommend trading when a learned policy chooses inaction, but the recommendation depends on information and future decisions. We introduce a four-stage audit for frozen proximal policy optimization policies without retraining. It examines deployment occupancy, matches current information, tests isolated deviations under incumbent continuation, and evaluates repeated deployment of observation-based alternatives. In controlled linear-Gaussian simulations, information matching explains part of the disagreement, while continuation changes its interpretation. At unit observation noise, incumbent continuation reverses 99.3% of projected-hard missed-advantage mass; repeated projected-rule deployment improves all 50 policies. These comparisons distinguish isolated action changes from policy replacement. Historical Bitcoin/Tether (BTCUSDT) replay applies this deployment perspective to a hand-specified intervention selected using 2024 data and frozen for 2025. Daily net reward improves by 135.03 basis points, with gains in 46 of 50 policies, primarily through lower turnover costs. The audit clarifies what oracle-flagged inaction implies for deployed decision making.

Explore related subjects

cs.CE

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Xingfei Zeng, Xin Zhong, Nanting Li, Ziyang Zhong, Lei Xiao, Guanghui Lu. 2026-09-11. When Is Inaction a Mistake? Continuation-Aware Auditing of PPO Trading Policies. https://arxiv.org/abs/2609.12536

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Cosm: Collective switched motion for sparse Ising optimization

We introduce Collective Switched Motion (Cosm), a heuristic optimization framework based on switched collective dynamics. Cosm compiles the objective into subobjectives that are activated sequentially, temporally separating competing local influences. The interplay of switching and local interactions among variables gives rise to collective search behavior, while a new correlated perturbation mechanism encourages coordinated cluster motion. Tests on tuned-hardness spin-glass benchmarks suggest more favorable algorithmic scaling than that reported for leading dynamical solvers. Cosm heuristically attains the certified optima of three of the largest Gset instances (G72, G77, G81), exceeding the previously reported heuristic solutions. On the large random-graph instances G61 and G70, a CPU implementation reliably attains the best-known solutions, reaching cuts of 5799 and 9595 with 99%-confidence times-to-target of 15 s and 2.4 s, respectively. On the heterogeneous-degree G64 instance, Cosm establishes a new best-known cut of 8753. Broadly, the results suggest an alternative approach to heuristic design in which local dynamics and orchestration mechanisms are carefully designed so that effective search emerges.

cs.CE↗

arXiv2026

Decomposing Firm-Level Crisis Responses from Incomplete Market Signals: Evidence from China's IT Sector During COVID-19

Exogenous shocks generate heterogeneous behavioral responses across firms, yet event studies typically report only sector-level averages. This paper develops a multi-method approach combining causal identification (difference-in-differences with cluster-robust inference), unsupervised behavioral discovery (K-means trajectory clustering, Gaussian hidden Markov models), and cross-sectional resilience prediction (logistic regression with nested cross-validation) to decompose firm-level response heterogeneity from noisy market signals. We demonstrate the approach on 246 Chinese A-share IT firms (216 with complete data for all analyses) during the COVID-19 shock (January 2020), using 252 non-IT CSI 300 firms as controls. The return decline was market-wide, not IT-specific (DID p = 0.59); the IT-specific effect was elevated volatility (DID beta = 0.043, cluster-robust p < 0.001), with the effect surviving Benjamini-Hochberg correction in 13 of 30 alternative specifications. Unsupervised clustering produced three trajectory groups: fast recovery (36 companies, +29.7%), resilient/moderate (67 companies), and persistent drag (113 companies, -6.9%). Pre-crisis financial fundamentals showed only modest predictive power for resilience (nested CV AUC = 0.635, 95% CI: 0.558-0.711; permutation p = 0.016), consistent with the limited informativeness of publicly available signals for anticipating crisis outcomes. The combination of causal analysis, unsupervised learning, and prediction represents a reproducible framework which can be applied to crises in other market periods.

cs.CE↗

arXiv2026

An Insurance Broker for Every Small Business: The Economics of Exceptional Care at Scale

Small-business owners need expert guidance on their own terms, across schedules, languages, and channels, but low premiums make exceptional, continuous human service uneconomic for much of the market. Combining public evidence, Kinro operational data, and an illustrative five-year service model, we show why traditional brokerage economics leave 35 million U.S. small businesses underserved. An AI-native brokerage can change those economics by performing and coordinating routine work continuously, while licensed professionals govern consequential exceptions and the brokerage remains accountable.

cs.CE↗