General

Reinforcement Learning in Financial Decision Making: A Systematic Review of Performance, Challenges, and Implementation Strategies ArXiv ID: 2512.10913 “View on arXiv” Authors: Mohammad Rezoanul Hoque, Md Meftahul Ferdaus, M. Kabir Hassan Abstract Reinforcement learning (RL) is an innovative approach to financial decision making, offering specialized solutions to complex investment problems where traditional methods fail. This review analyzes 167 articles from 2017–2025, focusing on market making, portfolio optimization, and algorithmic trading. It identifies key performance issues and challenges in RL for finance. Generally, RL offers advantages over traditional methods, particularly in market making. This study proposes a unified framework to address common concerns such as explainability, robustness, and deployment feasibility. Empirical evidence with synthetic data suggests that implementation quality and domain knowledge often outweigh algorithmic complexity. The study highlights the need for interpretable RL architectures for regulatory compliance, enhanced robustness in nonstationary environments, and standardized benchmarking protocols. Organizations should focus less on algorithm sophistication and more on market microstructure, regulatory constraints, and risk management in decision-making. ...

On the existence of personal equilibria ArXiv ID: 2512.08348 “View on arXiv” Authors: Laurence Carassus, Miklós Rásonyi Abstract We consider an investor who, while maximizing his/her expected utility, also compares the outcome to a reference entity. We recall the notion of personal equilibrium and show that, in a multistep, generically incomplete financial market model such an equilibrium indeed exists, under appropriate technical assumptions. Keywords: Personal Equilibrium, Utility Maximization, Incomplete Market, Reference Dependence, Game Theory, General/Asset Pricing ...

Stylized Facts and Their Microscopic Origins: Clustering, Persistence, and Stability in a 2D Ising Framework ArXiv ID: 2512.17925 “View on arXiv” Authors: Hernán Ezequiel Benítez, Claudio Oscar Dorso Abstract The analysis of financial markets using models inspired by statistical physics offers a fruitful approach to understand collective and extreme phenomena [“3, 14, 15”] In this paper, we present a study based on a 2D Ising network model where each spin represents an agent that interacts only with its immediate neighbors plus a term reated to the mean field [“1, 2”]. From this simple formulation, we analyze the formation of spin clusters, their temporal persistence, and the morphological evolution of the system as a function of temperature [“5, 19”]. Furthermore, we introduce the study of the quantity $1/2P\sum_{“i”}|S_{“i”}(t)+S_{“i”}(t+Δt)|$, which measures the absolute overlap between consecutive configurations and quantifies the degree of instantaneous correlation between system states. The results show that both the morphology and persistence of the clusters and the dynamics of the absolute sum can explain universal statistical properties observed in financial markets, known as stylized facts [“2, 12, 18”]: sharp peaks in returns, distributions with heavy tails, and zero autocorrelation. The critical structure of clusters and their reorganization over time thus provide a microscopic mechanism that gives rise to the intermittency and clustered volatility observed in prices [“2, 15”]. ...

Dynamic Inverse Optimization under Drift and Shocks: Theory, Regret Bounds, and Applications ArXiv ID: 2509.14080 “View on arXiv” Authors: JINHO CHA Abstract The growing prevalence of drift and shocks in modern decision environments exposes a gap between classical optimization theory and real-world practice. Standard models assume fixed objectives, yet organizations from hospitals to power grids routinely adapt to shifting priorities, noisy data, and abrupt disruptions. To address this gap, this study develops a dynamic inverse optimization framework that recovers hidden, time-varying preferences from observed allocation trajectories. The framework unifies identifiability analysis with regret guarantees conditions are established for existence and uniqueness of recovered parameters, and sharp static and dynamic regret bounds are derived to characterize responsiveness to gradual drift and sudden shocks. Methodologically, a drift-aware estimator grounded in convex analysis and online learning theory is introduced, with finite-sample guarantees on recovery accuracy. Computational experiments in healthcare, energy, logistics, and finance reveal heterogeneous recovery patterns, ranging from rapid resilience to persistent vulnerability. Overall, dynamic inverse optimization emerges as both a theoretical contribution and a broadly applicable diagnostic tool for benchmarking resilience, uncovering hidden behavioral shifts, and guiding policy interventions in complex stochastic systems. ...

Holdout cross-validation for large non-Gaussian covariance matrix estimation using Weingarten calculus ArXiv ID: 2509.13923 “View on arXiv” Authors: Lamia Lamrani, Benoît Collins, Jean-Philippe Bouchaud Abstract Cross-validation is one of the most widely used methods for model selection and evaluation; its efficiency for large covariance matrix estimation appears robust in practice, but little is known about the theoretical behavior of its error. In this paper, we derive the expected Frobenius error of the holdout method, a particular cross-validation procedure that involves a single train and test split, for a generic rotationally invariant multiplicative noise model, therefore extending previous results to non-Gaussian data distributions. Our approach involves using the Weingarten calculus and the Ledoit-Péché formula to derive the oracle eigenvalues in the high-dimensional limit. When the population covariance matrix follows an inverse Wishart distribution, we approximate the expected holdout error, first with a linear shrinkage, then with a quadratic shrinkage to approximate the oracle eigenvalues. Under the linear approximation, we find that the optimal train-test split ratio is proportional to the square root of the matrix dimension. Then we compute Monte Carlo simulations of the holdout error for different distributions of the norm of the noise, such as the Gaussian, Student, and Laplace distributions and observe that the quadratic approximation yields a substantial improvement, especially around the optimal train-test split ratio. We also observe that a higher fourth-order moment of the Euclidean norm of the noise vector sharpens the holdout error curve near the optimal split and lowers the ideal train-test ratio, making the choice of the train-test ratio more important when performing the holdout method. ...

Financial Data Analysis with Robust Federated Logistic Regression ArXiv ID: 2504.20250 “View on arXiv” Authors: Kun Yang, Nikhil Krishnan, Sanjeev R. Kulkarni Abstract In this study, we focus on the analysis of financial data in a federated setting, wherein data is distributed across multiple clients or locations, and the raw data never leaves the local devices. Our primary focus is not only on the development of efficient learning frameworks (for protecting user data privacy) in the field of federated learning but also on the importance of designing models that are easier to interpret. In addition, we care about the robustness of the framework to outliers. To achieve these goals, we propose a robust federated logistic regression-based framework that strives to strike a balance between these goals. To verify the feasibility of our proposed framework, we carefully evaluate its performance not only on independently identically distributed (IID) data but also on non-IID data, especially in scenarios involving outliers. Extensive numerical results collected from multiple public datasets demonstrate that our proposed method can achieve comparable performance to those of classical centralized algorithms, such as Logistical Regression, Decision Tree, and K-Nearest Neighbors, in both binary and multi-class classification tasks. ...

Unlocking NACE Classification Embeddings with OpenAI for Enhanced Analysis and Processing ArXiv ID: 2409.11524 “View on arXiv” Authors: Unknown Abstract The Statistical Classification of Economic Activities in the European Community (NACE) is the standard classification system for the categorization of economic and industrial activities within the European Union. This paper proposes a novel approach to transform the NACE classification into low-dimensional embeddings, using state-of-the-art models and dimensionality reduction techniques. The primary challenge is the preservation of the hierarchical structure inherent within the original NACE classification while reducing the number of dimensions. To address this issue, we introduce custom metrics designed to quantify the retention of hierarchical relationships throughout the embedding and reduction processes. The evaluation of these metrics demonstrates the effectiveness of the proposed methodology in retaining the structural information essential for insightful analysis. This approach not only facilitates the visual exploration of economic activity relationships, but also increases the efficacy of downstream tasks, including clustering, classification, integration with other classifications, and others. Through experimental validation, the utility of our proposed framework in preserving hierarchical structures within the NACE classification is showcased, thereby providing a valuable tool for researchers and policymakers to understand and leverage any hierarchical data. ...

Fitting an Equation to Data Impartially ArXiv ID: 2409.02573 “View on arXiv” Authors: Unknown Abstract We consider the problem of fitting a relationship (e.g. a potential scientific law) to data involving multiple variables. Ordinary (least squares) regression is not suitable for this because the estimated relationship will differ according to which variable is chosen as being dependent, and the dependent variable is unrealistically assumed to be the only variable which has any measurement error (noise). We present a very general method for estimating a linear functional relationship between multiple noisy variables, which are treated impartially, i.e. no distinction between dependent and independent variables. The data are not assumed to follow any distribution, but all variables are treated as being equally reliable. Our approach extends the geometric mean functional relationship to multiple dimensions. This is especially useful with variables measured in different units, as it is naturally scale-invariant, whereas orthogonal regression is not. This is because our approach is not based on minimizing distances, but on the symmetric concept of correlation. The estimated coefficients are easily obtained from the covariances or correlations, and correspond to geometric means of associated least squares coefficients. The ease of calculation will hopefully allow widespread application of impartial fitting to estimate relationships in a neutral way. ...

Large Language Model Agent in Financial Trading: A Survey ArXiv ID: 2408.06361 “View on arXiv” Authors: Unknown Abstract Trading is a highly competitive task that requires a combination of strategy, knowledge, and psychological fortitude. With the recent success of large language models(LLMs), it is appealing to apply the emerging intelligence of LLM agents in this competitive arena and understanding if they can outperform professional traders. In this survey, we provide a comprehensive review of the current research on using LLMs as agents in financial trading. We summarize the common architecture used in the agent, the data inputs, and the performance of LLM trading agents in backtesting as well as the challenges presented in these research. This survey aims to provide insights into the current state of LLM-based financial trading agents and outline future research directions in this field. ...

An Asymptotic CVaR Measure of Risk for Markov Chains ArXiv ID: 2405.13513 “View on arXiv” Authors: Unknown Abstract Risk sensitive decision making finds important applications in current day use cases. Existing risk measures consider a single or finite collection of random variables, which do not account for the asymptotic behaviour of underlying systems. Conditional Value at Risk (CVaR) is the most commonly used risk measure, and has been extensively utilized for modelling rare events in finite horizon scenarios. Naive extension of existing risk criteria to asymptotic regimes faces fundamental challenges, where basic assumptions of existing risk measures fail. We present a complete simulation based approach for sequentially computing Asymptotic CVaR (ACVaR), a risk measure we define on limiting empirical averages of markovian rewards. Large deviations theory, density estimation, and two-time scale stochastic approximation are utilized to define a ’tilted’ probability kernel on the underlying state space to facilitate ACVaR simulation. Our algorithm enjoys theoretical guarantees, and we numerically evaluate its performance over a variety of test cases. ...