financial backtesting

Traditional Monte Carlo methods for generating financial time series suffer from low fidelity, distorted tail statistics, and ill-conditioned covariance matrices. To address these issues, this paper pioneers the application of diffusion models to financial price dynamics modeling, proposing a high-fidelity synthetic data generation framework. Methodologically, the approach integrates numerical integration for accurate SDE solving, covariance matrix regularization to ensure numerical stability, and the Cramér–von Mises two-sample test to enforce statistical consistency across marginal distributions, extreme-value behavior, and multivariate dependence structures. Empirical results demonstrate strong alignment between synthetic and real market data in Q-Q plots and tail statistical tests; moreover, the condition number of the estimated covariance matrix is significantly reduced, enhancing numerical robustness. This framework establishes a novel generative paradigm for financial simulation, risk modeling, and algorithm training under data-scarce regimes—balancing statistical rigor with computational efficiency.

Historical simulation is often mistakenly regarded as assumption-free, yet it implicitly relies on strong and unspecified modeling assumptions. This study provides the first unified parametric framework for modeling asset returns that encompasses standard historical simulation, filtered historical simulation, and shifted historical simulation. By systematically reconstructing these methods through the extraction of realized innovations from historical data, the paper explicitly uncovers their underlying assumptions. The analysis demonstrates that these approaches impose far more stringent structural requirements on the model than commonly acknowledged, thereby significantly advancing the theoretical understanding of Value-at-Risk (VaR) estimation methodologies.

Existing market simulators struggle to simultaneously ensure controllability, plausibility, and cross-market/multi-frequency adaptability of synthetic financial data, hindering quantitative model development and robust evaluation. To address this, we propose the Retrieval-Augmented Financial Market Simulator (RA-FMS), the first framework integrating macro-level trend modeling and micro-level agent behavior via a retrieval-augmented diffusion architecture. RA-FMS enables causal “what-if” scenario generation and cross-market trend synthesis. We further design an automated model optimization framework grounded in simulated stress testing. Methodologically, RA-FMS unifies conditional diffusion modeling, cross-sectional information retrieval, and causal prompting for fine-grained control. Empirical results demonstrate that RA-FMS significantly enhances downstream quantitative models’ generalization under high-volatility regimes and improves stability of risk-adjusted returns. By providing interpretable, intervenable, and reproducible synthetic data, RA-FMS establishes a foundational infrastructure for trustworthy financial AI.

This paper frames asset pricing as a multiclass classification task, predicting whether individual stocks will outperform or underperform the market. Method: Supervised learning models generate out-of-sample trading portfolios, and a novel sequential binomial test rigorously evaluates over 3.34 million predictions for statistical significance—constituting the first systematic validation that historical information contains non-random, machine-learnable predictability. Contribution/Results: (1) Multiple classifiers pass stringent statistical tests, confirming short-term market inefficiency; (2) Model uncertainty—quantified by prediction probabilities—significantly affects portfolio performance, with high-confidence predictions yielding superior economic returns; (3) The constructed portfolios consistently outperform benchmarks out-of-sample. Collectively, this work establishes a reproducible, statistically verifiable methodological framework and empirical foundation for machine learning–driven quantitative investing.

This paper addresses the challenge of generating realistic tail-risk scenarios for high-dimensional, multi-asset dynamic portfolio optimization. We propose a novel generative adversarial network (GAN)-based scenario simulation method. Methodologically, we introduce an *elicitable GAN* loss function—first to leverage the joint elicitability of Value-at-Risk (VaR) and Expected Shortfall (ES)—to ensure accurate modeling of their joint tail distribution. To handle high dimensionality, we integrate principal component analysis (PCA) for dimensionality reduction and controlled expansion, overcoming limitations of univariate modeling. The framework uniformly supports risk assessment for both static and dynamic trading strategies. Empirical evaluation on synthetic and real financial market data demonstrates that our approach significantly outperforms existing data-driven scenario generation methods: it faithfully reproduces tail dependence structures, exhibits strong generalization across markets and time horizons, and scales effectively to high-dimensional asset universes.

This work addresses the limitations of existing financial agent evaluation frameworks, which often rely on static benchmarks or focus solely on final returns, thereby lacking traceability of decision-making processes and hindering fine-grained, fair performance assessment in dynamic markets. To overcome these challenges, the authors propose a unified performance tracking platform for financial agents that, for the first time, enables persistent logging of the complete decision trajectory—from market observation to trade execution. By integrating a time-consistent market data interface, a multi-agent collaborative architecture, and an end-to-end logging system, the platform supports interactive, cross-market and cross-model attribution analysis. Deployed across Hong Kong, U.S., and A-share markets, the system—augmented with a visual Trading Arena interface—significantly enhances the transparency, interpretability, and diagnostic capability of agent evaluation.

This work addresses three key limitations in existing financial reinforcement learning—asset lock-in, single-objective optimization, and static user modeling—by proposing a novel three-stage deep reinforcement learning framework. The approach introduces Chronos, a time-series foundation model, into portfolio management, leveraging self-supervised pretraining to learn asset-agnostic representations. It incorporates a goal-conditioned reward mechanism and an intent-routing strategy within a Mixture-of-Experts (MoE) architecture to jointly optimize six distinct investment objectives while mitigating gradient conflicts. Furthermore, personalized investment goals are dynamically inferred from real trading behavior through LoRA-based fine-tuning and natural language parsing, eliminating the need for user questionnaires. The method achieves zero-shot generalization across arbitrary publicly traded assets and demonstrates significant improvements in real-world investment performance.

This study investigates the efficient optimization and tail risk measurement of heterogeneous actively managed ETF portfolios. Leveraging daily data from 30 actively managed ETFs and one fixed-income mutual fund, it systematically evaluates static and dynamic strategies—including mean-variance optimization, CVaR minimization, tangency portfolios, and extreme value theory approaches (Hill estimator and Peaks-Over-Threshold with Generalized Pareto Distribution)—under varying constraints that incorporate dependence structures, dynamic allocation, transaction costs, and multidimensional tail risk metrics. The findings indicate that the tangency portfolio delivers superior cumulative returns and risk-adjusted performance, while a dynamic long-only CVaR-95 strategy proves robustly effective. Despite aggregation, portfolios exhibit pronounced downside tail risk. Innovatively treating actively managed ETFs as a joint opportunity set, this work elucidates how strategy heterogeneity collectively shapes overall portfolio performance.

Existing limit order book simulators struggle to accurately reproduce realistic execution costs, profit-and-loss outcomes, and microstructural dynamics. This work proposes an interactive simulator tailored for large-tick assets, which projects the order book state into a low-dimensional representation based on spread and volume imbalance, calibrates event timing to match the temporal structure of real markets, and incorporates a signed trade-flow feedback mechanism governed by a power-law decay kernel to capture market impact and its partial reversal. The proposed “project–estimate–validate–adapt” four-step framework is the first to simultaneously reproduce concave market impact, post-trade price reversion, and sensitivity to round-trip exchange latency within simulation. Experiments demonstrate that the approach generates highly realistic execution behavior across multiple stocks and trading strategies, significantly narrowing the gap between simulated and real-world market dynamics.

Traditional Gaussian hidden Markov models struggle to simultaneously capture the heavy-tailed nature of stock returns, their weak linear autocorrelation, and the slow decay of absolute return autocorrelations. This work proposes a continuous hidden Markov model that decouples temporal dynamics from marginal distribution modeling: temporal dependence is governed by a state transition chain, while each state employs heavy-tailed emission distributions—such as the Student-t or generalized error distribution—to accurately represent marginal characteristics. A unified EM framework is developed for parameter estimation. Theoretical analysis reveals that the inadequacy of conventional models stems from restrictive distributional assumptions rather than temporal structure, and that volatility clustering and high kurtosis can be reproduced with only a few latent states. Empirical results demonstrate that the proposed model significantly reduces fitting gaps across multiple U.S. equity datasets, generates paths that pass joint conditional coverage tests, and faithfully captures cross-asset dependencies, thereby supporting robust risk measurement and portfolio optimization.