It is well known that LLM guardrails and trained persona dynamics can produce a reality gap: the distance between the world a LLM is permitted or shaped to describe, and the world in which users must act. Here we argue that actively generating reality gaps is in fact unethical because it knowingly shifts epistemic risk back to the uninformed user -- this is reality laundering. This can potentially cause harm when operationalised at scale. The risk is sharpest in high-exposure advice contexts, where users seek orientation rather than a bounded, externally checkable task. Guardrails naively appear ethically necessary when they claim to prevent direct harm, but often become suspect when they suppress truthful perception and launder uncomfortable mechanisms into acceptable abstractions. Basel-style financial regulation, B-BBEE-style compliance, Societe Generale, and the London Whale show how formal safety systems can become legible, gameable, and performative while real exposure migrates elsewhere. The same pattern can appear in LLMs as moral compliance: safe language, distorted reality. We therefore distinguish refusing harm, from refusing reality; and then argue for top-down causal requirements specification at the task level rather than bottom-up moral correction at the response or sandbox level. Persona dynamics matter because the assistant interface is not neutral; it shapes how uncertainty, conflict, authority, and risk are staged. The conclusion is that so-called ``ethical AI'' becomes substantively unethical when it substitutes institutional reassurance for contact with reality.

Evaluating whether large language model (LLM) agents can profit in capital markets is increasingly framed as end-to-end trading: place an agent in a historical market, let it trade, and measure portfolio returns. This setup is vulnerable to two evaluation failures. First, long backtests often overlap with the knowledge cutoffs of frontier LLMs, allowing memorized tickers, dates, prices, and market narratives to substitute for investment reasoning. Second, raw returns are a noisy proxy for stock-selection ability, since positive performance may come from market beta, style exposure, or favorable regimes rather than genuine alpha. We introduce KTD-Fin (Knowing-To-Doing Financial Benchmark), an end-to-end stock-market trading benchmark that addresses both issues. KTD-Fin uses a data-side masking protocol to anonymize key identifiers and calendar information consistently across prompts and tools, separating historical market memory from investment decision-making. It also incorporates a Barra-style performance attribution framework that decomposes portfolio returns into market, style, and stock-selection alpha components. Across ten frontier LLM agents evaluated on the Chinese CSI300 over a 2024--2026 window, masking substantially changes agent rationales, pushing them towards anonymized factor-based reasoning. Attribution analysis further shows that LLM agents' cumulative returns under leakage-controlled evaluation are largely explained by passive market and style exposure, with limited evidence of persistent stock-selection alpha. These findings suggest that financial LLM benchmarks should evaluate not only whether an agent makes money, but also whether the source of returns reflects transferable investment skill. We release KTD-Fin as a reproducible template for leakage-controlled and attribution-aware evaluation of LLM trading agents.

Deploying machine learning in regulated financial environments -- credit risk, fraud detection, and anti-money laundering -- exposes critical vulnerabilities in algorithmic reproducibility. While early financial ML addressed statistical challenges such as backtest overfitting, deep neural networks and Generative AI have introduced mechanical nondeterminism rooted in hardware and architecture. This survey provides a systems perspective on reproducibility failures across three modalities now dominant in financial AI: tabular models (post-hoc explanation variance), graph networks (stochastic sampling and temporal asynchrony), and LLM-based agentic workflows (batch-dependent divergence and trajectory drift). We supplement the literature analysis with first-party experiments on public financial datasets -- quantifying explanation rank instability in credit scoring, prediction flip rates in GNN-based fraud detection, and tensor-parallel-induced output divergence in LLM entity extraction. We propose a layered evaluation framework linking modality-specific metrics (RBO, D_cos, TDI, PSD) to audit readiness, and empirically validate the complementarity of logit-level and semantic-level determinism measures.

This study looks at the statistical properties and predictability using deep learning methods of the U.S. aggregate bond index in daily observations spanning 2018 to February 2026. We first establish that index levels are extremely persistent and consistent with unitroot behavior (Dickey and Fuller), while log returns are covariance-stationary with weak linear dependence and pronounced volatility clustering characteristic of ARCH-type processes (Engle; Bollerslev). Motivated by the trade-off between stationarity and information retention, we construct a "stationary but maximally persistent" representation via fractional differencing (Granger and Joyeux; Hosking) following the procedure of López de Prado, and evaluate shorthorizon forecast using two neural paradigms: (i) Multilayer Perceptrons (MLPs) trained on lagged vectors with joint lag-length and hyperparameter tuning (Hornik et al.; Rumelhart et al.); and (ii) Convolutional Neural Networks (CNNs) trained on Gramian Angular Field (GAF) image encodings (Wang and Oates). Empirically, MLPs match the strong naive persistence benchmark on levels, collapse toward near-zero forecasts on returns, and achieve the strongest incremental performance on the fractionally differenced series, where moderate dependence remains but unit-root drift is attenuated. In contrast, CNN-GAF models deliver consistently negative out-of-sample R 2 across all three representations. Overall, the results imply that, for short-horizon forecasting of broad bond indices, the primary determinant of predictive performance is the transformation of the series-its degree of stationarity and memory-rather than architectural complexity. Lag-based models remain competitive under persistence, while GAFbased CNNs are better suited to pattern-based tasks than to persistence-dominated next-step prediction.

This study develops an integrated stochastic modeling framework for pricing short and medium-maturity equity options and assessing interest-rate risk using the Heston (1993), Bates (1996), and CIR (1985) models. We calibrate the Heston model using both the Lewis (2001) Fourier inversion and the Carr-Madan (1999) FFT approach, finding near-identical parameter sets, which is consistent with the calibration stability reported in recent studies such as Agazzotti et al. (2025). Extending the model to Bates shows that jump intensities converge to values effectively equal to zero for 60-day maturities, echoing empirical findings that jumps contribute marginally to short-term smile fitting. We further compare our calibration approach with the joint volatility-surface and variance-term-structure framework proposed by Yoo (2025), confirming that standard Heston/Bates calibration remains robust for the maturities considered. Finally, we calibrate the CIR short-rate model to the Euribor term structure, generating positive and economically consistent forward-rate scenarios in line with recent stochastic-rate option-pricing research by Jeon and Kim (2025). Overall, our results show that continuous stochastic volatility dominates near-term pricing dynamics, while stochastic interest rates materially influence valuations beyond one year.

This study develops a regime-aware portfolio allocation framework that integrates Markov switching models with Reinforcement Learning (RL) to dynamically allocate across equities (SPY), long-term Treasuries (TLT), and gold (GLD). Using daily ETF data from 2004-2025, we first characterize market behavior through a discrete Markov chain and then estimate a three-state Gaussian Hidden Markov Model (HMM) selected by the Bayesian Information Criterion (BIC). The estimated regimes-low-volatility, transitional, and high-volatility-exhibit strong persistence and state-dependent return dynamics consistent with recent findings on nonlinear market states (Ardia et al., 2024; Gupta & Pierdzioch, 2023). State-conditional analysis shows that SPY dominates in stable regimes, while TLT and GLD provide protection during stressed periods, motivating regime-conditioned allocation rules. We evaluate rule-based rotation and RL-driven strategies using a 30% out-of-sample test window with a one-day execution lag to avoid look-ahead bias. Both HMM-based allocations outperform a passive SPY benchmark, while the RL policy achieves the highest risk-adjusted performance, delivering the strongest Sharpe ratio and materially lower drawdowns, yet remains fully interpretable through discrete regime-dependent actions. Sensitivity analysis confirms the robustness of the three-state specification relative to two-state alternatives. Overall, the results demonstrate that RL can systematically enhance HMM-based regime detection, providing a transparent, adaptive, and empirically grounded framework for tactical asset allocation. The combined HMM-RL system provides a transparent, rules-based approach to tactical allocation that improves risk-adjusted performance relative to standard benchmark strategies.

The present paper investigates how insiders strategically navigate ongoing legal risk while leveraging stealth trading within a continuous-time Kyle-type framework. Legal enforcement operates concurrently with trading, which dynamic can be adversely obscured by a large surrounding population of noise traders. While surveillance intensity responds directly to the insider's trading intensity, triggering a random prosecution time, the resulting legal sanctions encompass both strategy-focused criminal penalties and profit-dependent civil penalties. Employing a new impact-neutral measure change, equilibrium analysis shows that even after achieving stealth, the insider internalizes regulatory exposure, and enforcement can significantly shape equilibrium trading strategies. The associated limiting equilibria yield a rich set of outcomes, with three key insights for regulatory impact: (i) under committed regulatory scrutiny, the insider trades a time-varying function of the discrepancy between the asset's fundamental value and its market price, and trading may intensify indefinitely near the end of the trading horizon as legal risk recedes; (ii) merely raising penalties as an advantageous selection cost proves ineffective in offsetting declines in regulatory diligence; (iii) criminal penalties remain essential for deterring aggressive insider trading, as they impose critical temporal constraints on trading intensity not achievable through civil penalties alone.

This review summarizes the historical development of probability measures in asset pricing, from early mathematical finance and state price theory to risk-neutral valuation, martingale measures, forward measures, stochastic discount factors, incomplete-market measure selection, benchmark pricing, robust and nonlinear pricing, and modern data-driven probability transformations. The central theme is that asset pricing is not merely an exercise in estimating physical probabilities. Instead, pricing theory constructs, transforms, or selects probability measures so that market prices can be represented as expectations after discounting, numeraire normalization, marginal utility weighting, entropy penalization, calibration, or information conditioning. The paper emphasizes landmark contributions including Bachelier's probabilistic model of speculation, Arrow-Debreu state-contingent claims, Black-Scholes-Merton option pricing, Harrison-Kreps and Harrison-Pliska's martingale formalization, Delbaen and Schachermayer's fundamental theorem, Breeden-Litzenberger implied state price densities, change of numeraire methods, Hansen-Jagannathan stochastic discount factor restrictions, Cochrane's SDF synthesis, and recent empirical and machine learning work on learned pricing kernels. Text-, attention-, and sentiment-based probability transformations are treated as recent information-adjusted forecasting extensions that complement, rather than replace, martingale, numeraire, SDF, and incomplete-market frameworks. The paper also collects key formulas for state prices, stochastic discount factors, Radon-Nikodym densities, Girsanov changes of measure, risk-neutral valuation, forward measures, implied densities, coherent risk measures, benchmark pricing, learned SDFs, and information-adjusted forecasting.

We extract four geometric observables -- Berry Phase Rate, Spectral Entropy, Reduced State Purity, and Hamiltonian Sensitivity -- from a learned spectral embedding of equity-index returns and evaluate them as regime-shift detectors against 46 classical and machine-learning baselines on 17 historical crises spanning 2000-2024. Under walk-forward nested hyperparameter selection on nine labelled windows, the Berry Phase Rate achieves an unbiased out-of-sample median Cohen's $d = 0.72$ (95% percentile-bootstrap CI $[0.34, 1.18]$, 10,000 resamples) and produces approximately 67% fewer false alarms per year than a label-supervised Random Forest (1.2 vs. 3.6 per year). Reduced State Purity attains the highest in-sample separability of any method ($d = 0.83$), tied closely by the Absorption Ratio ($d = 0.80$); geometric and classical channels are largely uncorrelated (mean $|ρ| \approx 0.22$), suggesting they capture distinct risk signals. Score construction is unsupervised; hyperparameter selection is the only supervised step.

Conversation logs from AI platforms are increasingly used to measure occupational exposure to artificial intelligence, but the users observed in these logs are not the workforce. We show that platform-derived exposure scores combine task-level AI applicability with the occupational composition of the platform's user base. Holding the empirical design fixed, changing only the platform input changes the post-ChatGPT employment coefficient by a factor of 1.9, and consumer and enterprise channels within the same vendor disagree in sign. We formalize the resulting non-classical measurement error, decompose it into between- and within-occupation selection, and construct workforce-reweighted partial-identification bounds. Reweighting to Bureau of Labor Statistics employment shares attenuates estimates by 42 to 93 percent. The bias captures augmentation among observed users more directly than substitution in the workforce.

This paper investigates how large language models (LLMs) form and express investor risk profiles, a critical component of retail investment advising. We examine three LLMs (GPT, Gemini, and Llama) and assess their responses to a standardized risk questionnaire under varying prompts. In particular, we establish each model's default investment profile by analyzing repeated responses per model. We observe that LLMs are generally longterm investors but exhibit different tendencies in risk tolerance: Gemini has a moderate risk level with highly consistent responses, Llama skews more conservative, and GPT appears moderately aggressive with the greatest variation in answers. Moreover, we find that assigning specific personas such as age, wealth, and investment experience leads each LLM to adjust its risk profile, although the extent of these adjustments differs across the models.

This study introduces a benchmark framework for evaluating the financial decision-making capabilities of large language models (LLMs) through portfolio optimization problems with mathematically explicit solutions. Unlike existing financial benchmarks that emphasize language-processing tasks, the proposed framework directly tests optimization-based reasoning in investment contexts. A large set of multiple-choice questions is generated by varying objectives, candidate assets, and investment constraints, with each problem designed to include a unique correct solution and systematically constructed alternatives. Experimental results comparing GPT-4, Gemini 1.5 Pro, and Llama 3.1-70B reveal distinct performance patterns: GPT achieves the highest accuracy in risk-based objectives and remains stable under constraints, Gemini performs well in return-based tasks but struggles under other conditions, and Llama records the lowest overall performance. These findings highlight both the potential and current limitations of LLMs in applying quantitative reasoning to finance, while providing a scalable foundation for developing LLM-based services in portfolio management.

Networks underpin systems that range from finance to biology, yet their structure is often only partially observed. Current reconstruction methods typically fit the parameters of a model anew to each snapshot, thus offering no guidance to predict future configurations. Here, we develop a Bayesian approach that uses the information about past network snapshots to inform a prior and predict the subsequent ones, while quantifying uncertainty. Instantiated with a single-parameter fitness model, our method infers link probabilities from node strengths and carries information forward in time. When applied to the Electronic Market for Interbank Deposit across the years 1999-2012, our method accurately recovers the number of connections per bank at subsequent times, outperforming probabilistic benchmarks designed for analogous, link prediction tasks. Notably, each predicted snapshot serves as a reliable prior for the next one, thus enabling self-sustained, out-of-sample reconstruction of evolving networks with a minimal amount of additional data.

The rapid advancement of Large Language Models (LLMs) has led to a surge of financial benchmarks, evolving from static knowledge evaluation toward interactive trading simulations. However, existing frameworks for evaluating real-time trading largely overlook a critical failure mode: the severe behavioral instability of LLMs in sequential decision-making under financial uncertainty. Through extensive experiments, we show that when deployed as trading agents, LLMs exhibit extreme run-to-run variance, generate inconsistent action sequences even under deterministic decoding, and frequently produce irrational action flipping across adjacent time steps. We attribute these behaviors to the stateless autoregressive nature of LLMs, which lack persistent memory of prior actions, together with their sensitivity to continuous-to-discrete action mappings in portfolio allocation tasks. These deficiencies fundamentally undermine the reliability and reproducibility of many existing online and offline trading benchmarks. To address these limitations, we propose AlphaForgeBench, a principled evaluation framework that redefines LLMs as quantitative researchers rather than stochastic trading agents. Instead of producing discrete trading actions, AlphaForgeBench requires models to generate executable alpha factors and compose factor-based trading strategies grounded in financial knowledge. This paradigm decouples reasoning from execution mechanics, enabling deterministic and reproducible evaluation while remaining aligned with real-world quantitative research workflows. Extensive experiments across multiple state-of-the-art LLMs demonstrate that AlphaForgeBench eliminates execution-induced instability and provides a rigorous benchmark for evaluating financial reasoning, strategy formulation, and alpha discovery. Webpage at https://finbrain-lab-hkustgz.github.io/AlphaForgeBench

We develop the theory of linear-quadratic (LQ) mean field games (MFGs) in Hilbert spaces with common noise modeled by an infinite-dimensional Wiener process that affects the dynamics of all agents. In the presence of common noise, the mean-field consistency condition is characterized by a system of coupled forward-backward stochastic evolution equations (FBSEEs) in Hilbert spaces, whereas, in its absence it is represented by coupled forward-backward deterministic evolution equations. We establish the existence and uniqueness of solutions to the coupled linear FBSEEs associated with the LQ MFG framework for small time horizons and prove the $ε$-Nash property of the resulting equilibrium strategy. Furthermore, we establish the well-posedness of these coupled linear FBSEEs for arbitrary finite time horizons. Beyond the specific context of MFGs, our analysis also yields a broader contribution by providing, to the best of our knowledge, the first well-posedness result for a class of infinite-dimensional linear FBSEEs, for which only mild solutions exist, over arbitrary finite time horizons.