transaction-level accounting

This work addresses the performance degradation in smart contract vulnerability detection caused by label noise introduced through reliance on unreliable open-source annotation tools. To mitigate this issue, the authors propose CGBC, a novel approach that integrates granular-ball computing between the encoder and classifier to generate coarse-grained representations via clustering, thereby correcting noisy labels. The method further enhances robustness by combining unsupervised contrastive pretraining, semantics-preserving data augmentation, and symmetric cross-entropy loss. Notably, this is the first study to synergize granular-ball computing with contrastive learning for vulnerability detection, introducing tailored intra- and inter-granular-ball loss functions. Extensive experiments demonstrate that CGBC significantly outperforms state-of-the-art methods across multiple benchmarks, effectively alleviating the adverse impact of label noise and substantially improving both detection accuracy and model robustness.

Existing systemic risk models struggle to integrate multi-source, heterogeneous, fine-grained banking data and often rely on oversimplified or single-channel network representations that fail to capture the complexity of multi-channel risk contagion mechanisms. This study proposes a novel multilayer network model of systemically important banks in the euro area, constructed from real regulatory and statistical data, where each layer corresponds to a distinct risk transmission channel—such as interbank lending, securities holdings, and short-term funding—and features consistent exposure matrices across layers over a unified set of nodes. Through multilayer network modeling, entity alignment, and microsimulation, the analysis reveals significant heterogeneity across layers in connectivity and centrality, demonstrating that aggregated networks can misidentify systemically important institutions. This work establishes a new paradigm for layered, data-driven systemic risk assessment.

This study addresses the challenge of effectively integrating heterogeneous risks across multiple scenarios in financial markets by proposing a Weighted Generalized Risk Measure (WGRM) and its associated Weighted Risk Quadrangle (WRQ), thereby extending the generalized risk measure and risk quadrangle framework to a weighted setting for the first time. Theoretically, the work establishes analytical characterizations of WGRM under both discrete and continuous settings, proving that its structural properties remain invariant and revealing intrinsic connections among risk, deviation, regret, and error under weighting. Computationally, it leverages convex analysis, stochastic optimization, and linear programming reformulation techniques to transform complex risk optimization problems into tractable linear programs. Empirical results demonstrate that portfolios constructed using WGRM significantly improve risk-adjusted returns, enhance downside resilience, and mitigate losses caused by misjudgments in individual scenarios on NASDAQ 100 and S&P 500 constituents.

This paper addresses the robustness of information aggregation among partially informed traders in dynamic trading, extending Ostrovsky (2012) by introducing a periodic paid-signal acquisition mechanism to analyze how declining information acquisition costs affect market aggregation efficiency. Methodologically, it integrates a dynamic rational expectations model, Market Scoring Rule incentives, a Shannon entropy-based cost function, and Bayesian equilibrium analysis. The paper introduces the novel concept of “k-separability” and proves it is both necessary and sufficient for full information aggregation. It shows that as signal costs approach zero, almost all securities satisfy k-separability, enabling full equilibrium aggregation even before costs vanish. Furthermore, it establishes that “unique-state-payoff” securities achieve efficient aggregation with probability one and yield strictly higher precision than equally informative opinion polls—demonstrating a nonlinear precision leap. These results deepen understanding of cost-driven aggregation thresholds and structural conditions for efficient market learning.

Financial regulation faces challenges in modeling high-dimensional, heterogeneous institutional data with missing values, where conventional methods fail to achieve robust and interpretable structured compression. This paper proposes the first Lloyd-type clustering framework tailored for probability distributions, introducing a novel generalized Wasserstein centroid and establishing a metric system in distribution space based on the generalized Wasserstein distance. This approach effectively addresses data incompleteness and distributional heterogeneity. By mapping financial institutions onto geometrically interpretable risk clusters, the method significantly enhances clustering robustness and interpretability on real-world regulatory datasets. It enables more precise and efficient risk identification and regulatory response, thereby supporting actionable supervision.

This study addresses the “granularity paradox” in time series forecasting, wherein fine-grained modeling improves in-sample fit but suffers from error accumulation due to recursive structures, degrading out-of-sample performance, while coarse-grained approaches incur information loss. Leveraging 13 years of public procurement data, the authors systematically evaluate ten model classes—spanning statistical, machine learning, and deep learning methods—across six temporal granularities using multidimensional metrics including TPFE, R², and RMSE. Their analysis reveals that recursive feedback topology, rather than model complexity, is the primary driver of error propagation. The work introduces a “consensus–discrepancy diagnostic” framework and advocates incorporating cumulative error metrics to overcome limitations of conventional point-wise error measures. Empirical results demonstrate strong model-dependent granularity effects—for instance, LSTM achieves a TPFE of 4.35% at daily granularity, whereas Holt-Winters fails catastrophically with R² = −151.

This study addresses the limitations of traditional financial sentiment analysis, which predominantly relies on news texts and uses stock returns as the sole predictive label, thereby overlooking the predictive power of regulatory filings—particularly the Item 1A risk factors section of 10-K reports—for market volatility. The authors propose a supervised dictionary learning approach to construct sentiment indicators across three aggregation levels—firm, portfolio, and industry—using 1,383 10-K filings from 94 Nasdaq-100 technology companies, with both returns and volatility as target variables. Results reveal that full-text sentiment performs better at higher aggregation levels (industry and portfolio), whereas Item 1A excels at the individual firm level. In contrast, the conventional Loughran-McDonald dictionary exhibits significant negative correlations. These findings underscore the necessity of supervised learning for regulatory text and highlight the interaction between textual scope and aggregation level in financial sentiment modeling.

This study addresses the challenge of priority reasoning among multidimensional rules—such as material, form, and function—in mapping product descriptions to Harmonized System (HS) codes. The authors propose a deterministic agent workflow that employs a fixed control flow to invoke large language models in distinct stages, embedding structured knowledge from China’s HS tariff nomenclature to produce interpretable and institutionally coherent classification decisions. The approach integrates a six-stage online inference pipeline with offline knowledge engineering, leveraging constrained calls and localized validation using the Qwen3.6 model series. Evaluated on the HSCodeComp dataset, the method achieves 77.4% top-1 accuracy at the six-digit level (using Qwen3.6-27B-FP8) and 84.2% at the four-digit level. Manual auditing reveals that certain ground-truth annotations deviate from the General Rules for the Interpretation of the HS.

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.

This study addresses the challenges of excessive discounting, inefficient manual review, and inconsistent pricing decisions in enterprise software contract negotiations, which often stem from a lack of data-driven benchmarks. To overcome these issues, this work proposes an adaptive nearest-neighbor-based peer comparison scoring system that embeds dynamic peer benchmarking directly into the contract design workflow. By leveraging tree-based ensemble models to learn similarity from historical contracts and defining data-driven proximity through shared leaf nodes, the system generates real-time letter-grade ratings and product-line-level insights for new contracts. The approach enables auditable, real-time pricing decisions and centralized oversight, significantly enhancing discount discipline post-deployment and driving commercially meaningful revenue uplift across scored contract portfolios.