turnover analysis

Reinforcement learning (RL) faces critical challenges in financial decision-making, including poor interpretability, limited robustness, deployment difficulties, regulatory compliance hurdles, weak adaptability to non-stationary market environments, and the absence of standardized benchmarks. This study conducts a systematic review of RL applications in market making, portfolio optimization, and algorithmic trading from 2017–2025, performing a meta-analysis of 167 peer-reviewed works. Integrating insights from financial microstructure theory, regulatory constraints, and risk management—augmented by synthetic-data-driven empirical validation—the work proposes a novel unified framework that explicitly embeds domain knowledge into architectural design, prioritizing implementation quality and model interpretability over algorithmic complexity. Results demonstrate that RL significantly outperforms conventional approaches in market-making tasks; however, regulatory-compliant, interpretable models are urgently needed, alongside cross-task, reproducible, standardized evaluation protocols to bridge the gap between research and real-world deployment.

This study addresses the governance divergence between total portfolio approaches and strategic asset allocation in institutional investing, revealing that their fundamental difference lies in the specification of tracking error constraints. Using U.S. equity and bond data from 2004 to 2026, historical backtesting, portfolio simulations, and statistical tests demonstrate that while Sharpe ratios exhibit no significant variation across different static tracking error constraints, realized tracking error volatility can differ by up to twelvefold, and constraint-related costs surge dramatically during crises. To reconcile these frameworks, the paper proposes a dynamic tracking error mechanism as a novel governance paradigm, which substantially enhances decision-making efficiency and allocation flexibility during periods of market stress.

This study addresses the long-standing lack of systematic measurement of “implementation risk” in quantitative investment backtesting—the performance discrepancies arising from differences in backtesting engine implementations. The work formally defines this risk for the first time and proposes four metrological metrics alongside a taxonomy of five failure modes. These are derived from parallel execution of 15 benchmark strategies across five open-source backtesting engines, incorporating transaction cost modeling, non-overlapping stratified asset buckets, and source code defect analysis. Experiments reveal that while engine outputs converge under zero-cost assumptions, performance divergence can reach up to 3.71% when transaction costs are introduced. Crucially, however, the relative ranking of strategy efficacy remains unchanged across engines (conclusion stability index = 1), indicating that implementation risk affects performance attribution but does not alter investment decisions.

Current LLM-driven trading research lacks standardized execution assumptions and reproducibility criteria, hindering cross-study comparisons and economic interpretability. This work systematically reviews 30 related studies and introduces the first evidence matrix encompassing execution semantics, turnover handling, and temporal control to evaluate transparency across dimensions such as data recency, backtest partitioning, and transaction cost modeling. Through bibliometric coding, methodological analysis of backtesting practices, and friction sensitivity experiments on ten stocks, the study quantifies how execution details compress strategy returns. It reveals that most papers inadequately disclose execution assumptions and proposes a standardized reporting framework emphasizing execution transparency as critical for result credibility, advocating for stricter community-wide standards of realism and reproducibility.

Financial foundation models are widely used in time-series forecasting, yet the risk composition of their driven trading strategies remains poorly understood, hindering real-world deployment. Method: We propose the first risk-decoupling framework tailored to foundation-model-based trading strategies: it disentangles systematic risk (epistemic uncertainty) from idiosyncratic risk (aleatoric uncertainty), extends the Capital Asset Pricing Model (CAPM) to incorporate uncertainty dimensions, and introduces the “aleatoric collapse hypothesis” for computationally tractable estimation—implemented via Monte Carlo Dropout and other uncertainty quantification techniques. Contribution/Results: Empirical evaluation demonstrates that our framework precisely identifies performance bottlenecks, early signs of model degradation, and actionable optimization pathways. It significantly enhances risk transparency and decision interpretability, thereby providing both theoretical foundations and practical tools for the trustworthy application of pre-trained models in high-frequency financial decision-making.

Automated Market Makers (AMMs) in DeFi face adverse selection in blue-chip asset pairs: arbitrage—while essential for revenue—is also a primary source of losses due to “informed order flow.” Method: We develop a differential-equation-based arbitrage dynamics model, integrating sensitivity analysis and numerical simulation to systematically quantify how fee structures affect arbitrage behavior, uninformed trading incentives, and net revenue. Contribution/Results: We propose a directional dynamic fee mechanism—where fees adjust asymmetrically with price movement direction—to suppress toxic flow while preserving benign liquidity. Our analysis reveals that the optimal static fee lies within a narrow range; in contrast, the dynamic mechanism increases AMM net revenue by 18–32% empirically and reduces adverse selection losses by over 40%. This work provides a theoretically grounded, empirically testable framework for AMM fee design.

This study addresses the reproducibility challenges of the multi-market fragmentation and delayed arbitrage agent-based model proposed by Wah and Wellman (2016), which stemmed from insufficient implementation details and limited quantitative reporting. Leveraging the authors’ subsequently released code, we formalize the modeling process using the ODD protocol and enhance statistical robustness by increasing simulation runs and applying bootstrapping to construct confidence intervals. Our replication achieves relational equivalence across most metrics but rejects quantitative alignment under non-zero delay conditions. Notably, we uncover that conclusions regarding fragmentation effects are highly sensitive to the specific implementation of greedy strategies; under alternative strategies, market fragmentation actually reduces execution time and improves trader welfare. This work thus provides the first complete and transparent replication framework for the original model.

Traditional risk metrics struggle to capture the structural fragility of systematic investment strategies under shifts in market regimes. This study proposes a “Minimum Regime Performance” (MRP) framework that, for the first time, quantifies strategy decay risk as a measurable indicator by evaluating the lowest risk-adjusted return a strategy achieves across distinct historical market regimes. By delineating historical regimes and conducting cross-regime performance analysis, the research validates the efficacy of MRP across a broad sample of factor-based strategies. The findings reveal a trade-off between long-term Sharpe ratios and strategy resilience, offering investors a practical tool to identify and manage the risk of performance deterioration stemming from regime transitions.

This study addresses the common oversight in existing reinforcement learning trading environments, which often neglect or oversimplify transaction costs, leading to strategies that fail in real-world deployment. Building upon the Almgren-Chriss framework and the square-root market impact law, this work proposes three open-source, Gymnasium-compatible trading environments that, for the first time, systematically incorporate empirically validated nonlinear market impact models. These environments support modular cost structures, exponentially decaying permanent impact, and fine-grained logging. Integrated with FinRL-Meta extensions and Optuna-based hyperparameter optimization, five state-of-the-art deep reinforcement learning algorithms are evaluated on NASDAQ-100 data. Results demonstrate that adopting the proposed model reduces average daily trading costs from $200,000 to $8,000 and turnover from 19% to 1%; hyperparameter optimization further cuts costs by up to 82%, with algorithm performance shown to be highly sensitive to the fidelity of cost modeling.

This study investigates behavioral alignment and representational dynamics of large language model (LLM) trading agents in financial markets, with a focus on early warning signals preceding failure. Using TradeArena—a novel auditable testing platform—the authors analyze LLM reasoning, position-taking, and intervention behaviors within risk reporting, execution simulation, and memory replay environments. They identify, for the first time, robust pre-failure signatures including stable representational drift, decoupling between planning and risk assessment, and contraction of the effective rank of latent manifolds. Structured risk feedback is shown to serve as an external alignment mechanism without requiring fine-tuning. Experiments across 80 rolling failure anchor points and 8 LLM trajectories confirm the robustness of these signatures. While real-world audit feedback improves calibration or returns for some models, it often exhibits myopic reward optimization and misalignment with diagnostic objectives, revealing rational blind spots in LLMs under coupled asset exposure.

Current large language model (LLM)-based trading systems lack rigorous evaluation of whether their intelligence translates into net profitability, making it difficult to ascertain if the costs of reasoning and decision-making are offset by incremental gains. This work proposes TradeLens, a novel toolkit that introduces the first “intelligence self-compensation” evaluation paradigm. By reconstructing trading trajectories, attributing costs and profits to interpretable evidence, and conducting multidimensional ablation studies across models, capital scales, trading frequencies, and architectures, TradeLens shifts the focus from mere performance ranking to diagnosing how intelligence converts into profit. The study reveals that intelligence self-compensation hinges on decision quality rather than system scale, with different models exhibiting distinct failure modes—such as asset selection bias or timing errors—while architectural factors influence profitability only indirectly through their impact on timing value.