Even large firms such as Walmart, Apple, and Coca-Cola face persistent fluctuations in costs, demand, and raw material availability. These are not \textit{rare events} and cannot be evaluated using traditional disruption models focused on infrequent events. Instead, sustained volatility induces opportunistic behavior, as firms repeatedly reconfigure partners in absence of long-term contracts, often due to trust deficits. The resulting web of transient relationships forms opportunistic supply chains (OSCs). To capture OSC evolution, we develop an integrated mathematical framework combining a Geometric Brownian Motion (GBM) model to represent stochastic price volatility, a Bayesian learning model to describe adaptive belief updates regarding partner reliability, and a Latent Order Logistic (LOLOG) network model for endogenous changes in network structure. This framework is implemented in an agent-based simulation to examine how volatility, trust, and network structure jointly shape SC resilience. Our modeling approach identifies critical volatility threshold; a tipping point beyond which the network shifts from a stable, link-preserving regime to a fragmented regime marked by rapid relationship dissolution. We analytically establish monotonic effects of volatility on profitability, trust, and link activation; derive formal stability conditions and volatility-driven phase transitions, and show how these mechanisms shape node importance and procurement behavior. These theoretical mechanisms are illustrated through computational experiments reflecting industry behaviors in fast fashion, electronics, and perishables. Overall, our contribution is to develop an integrated GBM-Bayesian-LOLOG framework to analyze OSC stability and our model can be extended to other OSCs including humanitarian, pharmaceutical, and poultry networks.

This paper re-examines the empirical Phillips curve (PC) model and its usefulness in the context of medium-term inflation forecasting. A latent variable Phillips curve hypothesis is formulated and tested using 3,968 randomly generated factor combinations. Evidence from US core PCE inflation between Q1 1983 and Q1 2025 suggests that latent variable PC models reliably outperform traditional PC models six to eight quarters ahead and stand a greater chance of outperforming a univariate benchmark. Incorporating an MA(1) residual process improves the accuracy of empirical PC models across the board, although the gains relative to univariate models remain small. The findings presented in this paper have two important implications: First, they corroborate a new conceptual view on the Phillips curve theory; second, they offer a novel path towards improving the competitiveness of Phillips curve forecasts in future empirical work.

Building on the finding that Market Cap Normalization ($\SMC$) isolates the ``pure'' directional signal of informed trading \citep{kang2025}, this paper investigates the physics of how that signal is transmitted -- and how it breaks down. We employ \textbf{Tikhonov-regularized deconvolution} to recover the impulse response kernels of investor flows, revealing a dual-channel market structure: Foreign and Institutional investors act as ``architects'' of price discovery (positive permanent impact), while Individual investors act as liquidity providers (negative total impact). However, using \textbf{Multivariate Hawkes Processes}, we demonstrate that this structure is fragile. We find that individual investor order flow exhibits near-critical self-excitation (Branching Ratio $\approx$ 0.998). During periods of high retail herding, the market undergoes a \textbf{phase transition} into a ``critical state.'' In this regime, the signal-to-noise ratio collapses, causing the price impact of sophisticated investors to reverse from positive to negative. These findings suggest that retail contagion acts as a physical barrier that temporarily disables efficient price discovery.

Can fully agentic AI nowcast stock returns? We deploy a state-of-the-art Large Language Model to evaluate the attractiveness of each Russell 1000 stock daily, starting from April 2025 when AI web interfaces enabled real-time search. Our data contribution is unique along three dimensions. First, the nowcasting framework is completely out-of-sample and free of look-ahead bias by construction: predictions are collected at the current edge of time, ensuring the AI has no knowledge of future outcomes. Second, this temporal design is irreproducible -- once the information environment passes, it can never be recreated. Third, our framework is 100% agentic: we do not feed the model news, disclosures, or curated text; it autonomously searches the web, filters sources, and synthesises information into quantitative predictions. We find that AI possesses genuine stock selection ability, but only for identifying top winners. Longing the 20 highest-ranked stocks generates a daily Fama-French five-factor plus momentum alpha of 18.4 basis points and an annualised Sharpe ratio of 2.43. Critically, these returns derive from an implementable strategy trading highly liquid Russell 1000 constituents, with transaction costs representing less than 10\% of gross alpha. However, this predictability is highly concentrated: expanding beyond the top tier rapidly dilutes alpha, and bottom-ranked stocks exhibit returns statistically indistinguishable from the market. We hypothesise that this asymmetry reflects online information structure: genuinely positive news generates coherent signals, while negative news is contaminated by strategic corporate obfuscation and social media noise.

Global inequality has shifted inward, with rising dispersion increasingly occurring within countries rather than between them. Using 8,790 newly harmonised Functional Urban Areas (FUAs), micro-founded labour-market regions encompassing 3.9 billion people and representing approximately 80% of global GDP, we show that national aggregates systematically, and increasingly, misrepresent the dynamics of growth, convergence, and structural change. Drawing on high-resolution global GDP data and country-level capability measures, we find that the middle-income trampoline that previously drove global convergence is flattening. This divergence in the lower-income regime does not reflect poverty traps: low-income FUAs exhibit positive expected growth, and the transition curve displays no stable low-income equilibrium. Instead, productive capabilities, proxied by the Economic Complexity Index, define distinct growth regimes. FUAs converge within capability strata but diverge across them, and capability upgrading follows a predictable J-curve marked by short-run disruption and medium-run acceleration. These findings suggest that national convergence policies may be systematically misaligned with the geographic scale at which capability accumulation operates.

We analyze daily lead-time distributions for two Airbnb demand metrics, Nights Booked (volume) and Gross Booking Value (revenue), treating each day's allocation across 0-365 days as a compositional vector. The data span 2,557 days from January 2019 through December 2025 in a large North American region. Three findings emerge. First, GBV concentrates more heavily in mid-range horizons: beyond 90 days, GBV tail mass typically exceeds Nights by 20-50%, with ratios reaching 75% at the 180-day threshold during peak seasons. Second, Gamma and Weibull distributions fit comparably well under interval-censored cross-entropy. Gamma wins on 61% of days for Nights and 52% for GBV, with Weibull close behind at 38% and 45%. Lognormal rarely wins (<3%). Nonparametric GAMs achieve 18-80x lower CRPS but sacrifice interpretability. Third, generalized Pareto fits suggest bounded tails for both metrics at thresholds below 150 days, though this may partly reflect right-truncation at 365 days; above 150 days, estimates destabilize. Bai-Perron tests with HAC standard errors identify five structural breaks in the Wasserstein distance series, with early breaks coinciding with COVID-19 disruptions. The results show that volume and revenue lead-time shapes diverge systematically, that simple two-parameter distributions capture daily pmfs adequately, and that tail inference requires care near truncation boundaries.

This paper develops a micro-founded economic theory of the AI industry by modeling large language models as a distinct asset class-Digital Intelligence Capital-characterized by data-compute complementarities, increasing returns to scale, and relative (rather than absolute) valuation. We show that these features fundamentally reshape industry dynamics along three dimensions. First, because downstream demand depends on relative capability, innovation by one firm endogenously depreciates the economic value of rivals' existing capital, generating a persistent innovation pressure we term the Red Queen Effect. Second, falling inference prices induce downstream firms to adopt more compute-intensive agent architectures, rendering aggregate demand for compute super-elastic and producing a structural Jevons paradox. Third, learning from user feedback creates a data flywheel that can destabilize symmetric competition: when data accumulation outpaces data decay, the market bifurcates endogenously toward a winner-takes-all equilibrium. We further characterize conditions under which expanding upstream capabilities erode downstream application value (the Wrapper Trap). A calibrated agent-based model confirms these mechanisms and their quantitative implications. Together, the results provide a unified framework linking intelligence production upstream with agentic demand downstream, offering new insights into competition, scalability, and regulation in the AI economy.

This study investigates the economic complexity of Indian states by constructing a state-industry bipartite network using firm-level data on registered companies and their paid-up capital. We compute the Economic Complexity Index and apply the fitness-complexity algorithm to quantify the diversity and sophistication of productive capabilities across the Indian states and two union territories. The results reveal substantial heterogeneity in regional capability structures, with states such as Maharashtra, Karnataka, and Delhi exhibiting consistently high complexity, while others remain concentrated in ubiquitous, low-value industries. The analysis also shows a strong positive relationship between complexity metrics and per-capita Gross State Domestic Product, underscoring the role of capability accumulation in shaping economic performance. Additionally, the number of active firms in India demonstrates a persistent exponential growth at an annual rate of 11.2%, reflecting ongoing formalization and industrial expansion. The ordered binary matrix displays the characteristic triangular structure observed in complexity studies, validating the applicability of complexity frameworks at the sub-national level. This work highlights the usefulness of firm-based data for assessing regional productive structures and emphasizes the importance of capability-oriented strategies for fostering balanced and sustainable development across Indian states. By demonstrating the usefulness of firm registry data in data constrained environments, this study advances the empirical application of economic complexity methods and provides a quantitative foundation for capability-oriented industrial and regional policy in India.

In this paper, we study the order between the standard deviation (SD) and the Gini mean difference (GMD) and derive sharp, interpretable sufficient conditions under which one exceeds the other. By expressing both the SD and the GMD in terms of pairwise differences and linking their comparison to the mean excess function of the absolute difference of two i.i.d.\ copies, we reduce the problem to structural properties of the underlying distribution. Using tools from reliability and survival analysis, we show that SD dominance arises under heavy-tailed regimes, characterized by decreasing hazard rates or increasing reverse hazard rates. Conversely, when both tails are light -- equivalently, when the hazard rate is increasing and the reverse hazard rate is decreasing -- the GMD dominates the SD. We further demonstrate that these dominance relations are preserved under affine transformations, mixtures, convolutions, and tail truncation, and we extend the analysis to discrete distributions. Numerous examples illustrate the sharpness of the results and highlight the distinct roles played by tail behavior and distributional regularity. Our findings provide a unified framework for understanding dispersion ordering and offer clear guidance for the choice of variability measures in risk-sensitive applications.

The diffusion of Generative AI (GenAI) constitutes a supply shock of a fundamentally different nature: while marginal production costs approach zero, content generation creates congestion externalities through information pollution. We develop a three-layer general equilibrium framework to study how this non-convex technology reshapes market structure, transition dynamics, and social welfare. In a static vertical differentiation model, we show that the GenAI cost shock induces a kinked production frontier that bifurcates the market into exit, AI, and human segments, generating a ``middle-class hollow'' in the quality distribution. To analyze adjustment paths, we embed this structure in a mean-field evolutionary system and a calibrated agent-based model with bounded rationality. The transition to the AI-integrated equilibrium is non-monotonic: rather than smooth diffusion, the economy experiences a temporary ecological collapse driven by search frictions and delayed skill adaptation, followed by selective recovery. Survival depends on asymmetric skill reconfiguration, whereby humans retreat from technical execution toward semantic creativity. Finally, we show that the welfare impact of AI adoption is highly sensitive to pollution intensity: low congestion yields monotonic welfare gains, whereas high pollution produces an inverted-U relationship in which further AI expansion reduces total welfare. These results imply that laissez-faire adoption can be inefficient and that optimal governance must shift from input regulation toward output-side congestion management.

No-arbitrage asset pricing characterizes valuation through the existence of equivalent martingale measures relative to a filtration and a class of admissible trading strategies. In practice, pricing is performed across multiple asset classes driven by economic variables that are only partially spanned by traded instruments, raising a structural question: does there exist a single admissible information structure under which all traded assets can be jointly priced as martingales?. We treat the filtration as an endogenous object constrained by admissibility and time-ordering, rather than as an exogenous primitive. For any finite collection of assets, whenever martingale pricing is feasible under some admissible filtration, it is already feasible under a canonical minimal filtration generated by the asset prices themselves; these pricing-sufficient filtrations are unique up to null sets and stable under restriction and aggregation when a common pricing measure exists. Our main result shows that this local compatibility does not extend globally: with three independent unspanned finite-variation drivers, there need not exist any admissible filtration and equivalent measure under which all assets are jointly martingales. The obstruction is sharp (absent with one driver and compatible pairwise with two) and equivalent to failure of admissible dynamic completeness. We complement the theory with numerical diagnostics based on discrete-time Doob--Meyer decompositions, illustrating how admissible information structures suppress predictable components, while inadmissible filtrations generate systematic predictability.

This paper develops a dynamic insurance market model comprising two competing insurance companies and a continuum of insureds, and examines the interaction between strategic underreporting by the insureds and competitive pricing between the insurance companies under a Bonus-Malus System (BMS) framework. For the first time in an oligopolistic setting, we establish the existence and uniqueness of the insureds' optimal reporting barrier, as well as its continuous dependence on the BMS premiums. For the 2-class BMS case, we prove the existence of Nash equilibrium premium strategies and conduct an extensive sensitivity analysis on the impact of the model parameters on the equilibrium premiums.

Liability sharing and staffing jointly determine service quality in AI-assisted online medical consultation, yet their interaction is rarely examined in an integrated framework linking contracts to congestion via physician responses. This paper develops a Stackelberg queueing model where the platform selects a liability share and a staffing level while physicians choose between AI-assisted and independent diagnostic modes. Physician mode choice exhibits a threshold structure, with the critical liability share decreasing in loss severity and increasing in the effort cost of independent diagnosis. Optimal platform policy sets liability below this threshold to trade off risk transfer against compliance costs, revealing that liability sharing and staffing function as substitute safety mechanisms. Higher congestion or staffing costs tilt optimal policy toward AI-assisted operation, whereas elevated loss severity shifts the preferred regime toward independent diagnosis. The welfare gap between platform and social optima widens with loss severity, suggesting greater scope for incentive alignment in high-stakes settings. By endogenizing physician mode choice within a congested service system, this study clarifies how liability design propagates through queueing dynamics, offering guidance for calibrating contracts and capacity in AI-assisted medical consultation.

Generative models for financial time series often create data that look realistic and even reproduce stylized facts such as fat tails or volatility clustering. However, these apparent successes break down under trading backtests: models like GANs or WGAN-GP frequently collapse, yielding extreme and unrealistic results that make the synthetic data unusable in practice. We identify the root cause in the neglect of financial asymmetry and rare tail events, which strongly affect market risk but are often overlooked by objectives focusing on distribution matching. To address this, we introduce the Stylized Facts Alignment GAN (SFAG), which converts key stylized facts into differentiable structural constraints and jointly optimizes them with adversarial loss. This multi-constraint design ensures that generated series remain aligned with market dynamics not only in plots but also in backtesting. Experiments on the Shanghai Composite Index (2004--2024) show that while baseline GANs produce unstable and implausible trading outcomes, SFAG generates synthetic data that preserve stylized facts and support robust momentum strategy performance. Our results highlight that structure-preserving objectives are essential to bridge the gap between superficial realism and practical usability in financial generative modeling.

The growing adoption of artificial intelligence (AI) technologies has heightened interest in the labour market value of AI-related skills, yet causal evidence on their role in hiring decisions remains scarce. This study examines whether AI skills serve as a positive hiring signal and whether they can offset conventional disadvantages such as older age or lower formal education. We conduct an experimental survey with 1,700 recruiters from the United Kingdom and the United States. Using a paired conjoint design, recruiters evaluated hypothetical candidates represented by synthetically designed resumes. Across three occupations - graphic designer, office assistant, and software engineer - AI skills significantly increase interview invitation probabilities by approximately 8 to 15 percentage points. AI skills also partially or fully offset disadvantages related to age and lower education, with effects strongest for office assistants, where formal AI certification plays an additional compensatory role. Effects are weaker for graphic designers, consistent with more skeptical recruiter attitudes toward AI in creative work. Finally, recruiters' own background and AI usage significantly moderate these effects. Overall, the findings demonstrate that AI skills function as a powerful hiring signal and can mitigate traditional labour market disadvantages, with implications for workers' skill acquisition strategies and firms' recruitment practices.

We study how radiologists use AI to diagnose pulmonary embolism (PE), tracking over 100,000 scans interpreted by nearly 400 radiologists during the staggered rollout of a real-world FDA-approved diagnostic platform in a hospital system. When AI flags PE, radiologists agree 84% of the time; when AI predicts no PE, they agree 97%. Disagreement evolves substantially: radiologists initially reject AI-positive PEs in 30% of cases, dropping to 12% by year two. Despite a 16% increase in scan volume, diagnostic speed remains stable while per-radiologist monthly volumes nearly double, with no change in patient mortality -- suggesting AI improves workflow without compromising outcomes. We document significant heterogeneity in AI collaboration: some radiologists reject AI-flagged PEs half the time while others accept nearly always; female radiologists are 6 percentage points less likely to override AI than male radiologists. Moderate AI engagement is associated with the highest agreement, whereas both low and high engagement show more disagreement. Follow-up imaging reveals that when radiologists override AI to diagnose PE, 54% of subsequent scans show both agreeing on no PE within 30 days.

Dealers in foreign exchange markets provide bid and ask prices to their clients at which they are happy to buy and sell, respectively. To manage risk, dealers can skew their quotes and hedge in the interbank market. Hedging offers certainty but comes with transaction costs and market impact. Optimal market making with execution has previously been addressed within the Almgren-Chriss market impact model, which includes instantaneous and permanent components. However, there is overwhelming empirical evidence of the transient nature of market impact, with instantaneous and permanent impacts arising as the two limiting cases. In this note, we consider an intermediate scenario and study the interplay between risk management and impact resilience.

We estimate the national social cost of carbon using a recent meta-analysis of the total impact of climate change and a standard integrated assessment model. The average social cost of carbon closely follows per capita income, the national social cost of carbon the size of the population. The national social cost of carbon measures self-harm. Net liability is defined as the harm done by a country's emissions on other countries minus the harm done to a country by other countries' emissions. Net liability is positive in middle-income, carbon-intensive countries. Poor and rich countries would be compensated because their current emissions are relatively low, poor countries additionally because they are vulnerable.

We develop a mathematical framework to optimize leveraged staking ("loopy") strategies in Decentralized Finance (DeFi), in which a staked asset is supplied as collateral, the underlying is borrowed and re-staked, and the loop can be repeated across multiple lending markets. Exploiting the fact that DeFi borrow rates are deterministic functions of pool utilization, we reduce the multi-market problem to a convex allocation over market exposures and obtain closed-form solutions under three interest-rate models: linear, kinked, and adaptive (Morpho's AdaptiveCurveIRM). The framework incorporates market-specific leverage limits, utilization-dependent borrowing costs, and transaction fees. Backtests on the Ethereum and Base blockchains using the largest Morpho wstETH/WETH markets (from January 1 to April 1, 2025) show that rebalanced leveraged positions can reach up to 6.2% APY versus 3.1% for unleveraged staking, with strong dependence on position size and rebalancing frequency. Our results provide a mathematical basis for transparent, automated DeFi portfolio optimization.

Healthcare sector indices consolidate the economic health of pharmaceutical, biotechnology, and healthcare service firms. The short-term movements in these indices are closely intertwined with capital allocation decisions affecting research and development investment, drug availability, and long-term health outcomes. This research investigates whether historical open-high-low-close (OHLC) index data contain sufficient information for predicting the directional movement of the opening index on the subsequent trading day. The problem is formulated as a supervised classification task involving a one-step-ahead rolling window. A diverse feature set is constructed, comprising original prices, volatility-based technical indicators, and a novel class of nowcasting features derived from mutual OHLC ratios. The framework is evaluated on data from healthcare indices in the U.S. and Indian markets over a five-year period spanning multiple economic phases, including the COVID-19 pandemic. The results demonstrate robust predictive performance, with accuracy exceeding 0.8 and Matthews correlation coefficients above 0.6. Notably, the proposed nowcasting features have emerged as a key determinant of the market movement. We have employed the Shapley-based explainability paradigm to further elucidate the contribution of the features: outcomes reveal the dominant role of the nowcasting features, followed by a more moderate contribution of original prices. This research offers a societal utility: the proposed features and model for short-term forecasting of healthcare indices can reduce information asymmetry and support a more stable and equitable health economy.

We study whether Fintech disrupts the banking sector by intensifying competition for scarce deposits funds and raising deposit rates. Using difference-in-difference estimation around the exogenous removal of marketplace platform investing restrictions by US states, we show the cost of deposits increase by approximately 11.5% within small financial institutions. However, these price changes are effective in preventing a drain of liquidity. Size and geographical diversification through branch networks can mitigate the effects of Fintech competition by sourcing deposits from less competitive markets. The findings highlight the unintended consequences of the growing Fintech sector on banks and offer policy insights for regulators and managers into the ongoing development and impact of technology on the banking sector.

Using data on school shooting incidents in U.S. K--12 schools from 1981 to 2022, we estimate the causal effects of high temperatures on school shootings and assess the implications of climate change. We find that days with maximum temperatures exceeding 90$^\circ$F lead to a 80\% increase in school shootings relative to days below 70$^\circ$F. Consistent with theories linking heat exposure to aggression, high temperatures increase homicidal and threat-related shootings but have no effect on accidental or suicidal shootings. Heat-induced shootings occur disproportionately during periods of greater student mobility and reduced supervision, including before and after school hours and lunch periods. Higher temperatures increase shootings involving both student and non-student perpetrators. We project that climate change will increase homicidal and threat-related school shootings in the U.S. by 8\% under SSP2--4.5 (moderate emissions) and by 14\% under SSP5--8.5 (high emissions) by 2091--2100, corresponding to approximately 23 and 39 additional shootings per decade, respectively. The present discounted value of the resulting social costs is \$343 million and \$592 million (2025 dollars), respectively.

How do authoritarian regimes strengthen global support for nondemocratic political systems? Roughly half of the users of the social media platform TikTok report getting news from social media influencers. Against this backdrop, authoritarian regimes have increasingly outsourced content creation to these influencers. To gain understanding of the extent of this phenomenon and the persuasive capabilities of these influencers, we collect comprehensive data on pro-China influencers on TikTok. We show that pro-China influencers have more engagement than state media. We then create a realistic clone of the TikTok app, and conduct a randomized experiment in which over 8,500 Americans are recruited to use this app and view a random sample of actual TikTok content. We show that pro-China foreign influencers are strikingly effective at increasing favorability toward China, while traditional Chinese state media causes backlash. The findings highlight the importance of influencers in shaping global public opinion.

In an incomplete financial market with general continuous semimartingale dynamics; we model an investor with log-utility preferences who, in addition to an initial capital, receives units of a non-traded endowment process. Using duality techniques, we derive the fourth-order expansion of the primal value function with respect to the units $\epsilon$, held in the non-traded endowment. In turn, this lays the foundation for expanding the optimal wealth process, in this context, up to second order w.r.t. $\epsilon$. The key processes underpinning the aforementioned results are given in terms of Kunita-Watanabe projections, mirroring the case of lower order expansions of similar nature. Both the case of finite and infinite horizons are treated in a unified manner.

I study how firms adjust to temporary disruptions in international trade relationships organized through relational contracts. I exploit an extreme, plausibly exogenous weather shock during the 2010-11 La Niña season that restricted Colombian flower exporters' access to cargo terminals. Using transaction-level data from the Colombian-U.S. flower trade, I show that importers with less-exposed supplier portfolios are less likely to terminate disrupted relationships, instead tolerating shipment delays. In contrast, firms facing greater exposure experience higher partner turnover and are more likely to exit the market, with exit accounting for a substantial share of relationship separations. These findings demonstrate that idiosyncratic shocks to buyer-seller relationships can propagate into persistent changes in firms' trading portfolios.

Electricity price forecasting approaches generally fall into two categories: data-driven models, which learn from historical patterns, or fundamental models, which simulate market mechanisms. We propose a novel and highly efficient data-driven merit order model that integrates both paradigms. The model embeds the classical expert-based merit order as a nested special case, allowing all key parameters, such as plant efficiencies, bidding behavior, and available capacities, to be estimated directly from historical data, rather than assumed. We further enhance the model with critical embedded extensions such as hydro power, cross-border flows and corrections for underreported capacities, which considerably improve forecasting accuracy. Applied to the German day-ahead market, our model outperforms both classic fundamental and state-of-the-art machine learning models. It retains the interpretability of fundamental models, offering insights into marginal technologies, fuel switches, and dispatch patterns, elements which are typically inaccessible to black-box machine learning approaches. This transparency and high computational efficiency make it a promising new direction for electricity price modeling.

Incarceration-diversion treatment programs aim to improve societal reintegration and reduce recidivism, but limited capacity forces policymakers to make prioritization decisions that often rely on risk assessment tools. While predictive, these tools typically treat risk as a static, individual attribute, which overlooks how risk evolves over time and how treatment decisions shape outcomes through social interactions. In this paper, we develop a new framework that models reoffending risk as a human-system interaction, linking individual behavior with system-level dynamics and endogenous community feedback. Using an agent-based simulation calibrated to U.S. probation data, we evaluate treatment allocation policies under different capacity constraints and incarceration settings. Our results show that no single prioritization policy dominates. Instead, policy effectiveness depends on temporal windows and system parameters: prioritizing low-risk individuals performs better when long-term trajectories matter, while prioritizing high-risk individuals becomes more effective in the short term or when incarceration leads to shorter monitoring periods. These findings highlight the need to evaluate risk-based decision systems as sociotechnical systems with long-term accountability, rather than as isolated predictive tools.

Detecting anomalous trajectories in decentralized crypto networks is fundamentally challenged by extreme label scarcity and the adaptive evasion strategies of illicit actors. While Graph Neural Networks (GNNs) effectively capture local structural patterns, they struggle to internalize multi hop, logic driven motifs such as fund dispersal and layering that characterize sophisticated money laundering, limiting their forensic accountability under regulations like the FATF Travel Rule. To address this limitation, we propose Relational Domain Logic Integration (RDLI), a framework that embeds expert derived heuristics as differentiable, logic aware latent signals within representation learning. Unlike static rule based approaches, RDLI enables the detection of complex transactional flows that evade standard message passing. To further account for market volatility, we incorporate a Retrieval Grounded Context (RGC) module that conditions anomaly scoring on regulatory and macroeconomic context, mitigating false positives caused by benign regime shifts. Under extreme label scarcity (0.01%), RDLI outperforms state of the art GNN baselines by 28.9% in F1 score. A micro expert user study further confirms that RDLI path level explanations significantly improve trustworthiness, perceived usefulness, and clarity compared to existing methods, highlighting the importance of integrating domain logic with contextual grounding for both accuracy and explainability.

We introduce a novel model for time-varying, asymmetric, tail-dependent copulas in high dimensions that incorporates both spectral dynamics and regularization. The dynamics of the dependence matrix' eigenvalues are modeled in a score-driven way, while biases in the unconditional eigenvalue spectrum are resolved by non-linear shrinkage. The dynamic parameterization of the copula dependence matrix ensures that it satisfies the appropriate restrictions at all times and for any dimension. The model is parsimonious, computationally efficient, easily scalable to high dimensions, and performs well for both simulated and empirical data. In an empirical application to financial market dynamics using 100 stocks from 10 different countries and 10 different industry sectors, we find that our copula model captures both geographic and industry related co-movements and outperforms recent computationally more intensive clustering-based factor copula alternatives. Both the spectral dynamics and the regularization contribute to the new model's performance. During periods of market stress, we find that the spectral dynamics reveal strong increases in international stock market dependence, which causes reductions in diversification potential and increases in systemic risk.

Diseases originating from wildlife pose a significant threat to global health, causing human and economic losses each year. The transmission of disease from animals to humans occurs at the interface between humans, livestock, and wildlife reservoirs, influenced by abiotic factors and ecological mechanisms. Although evidence suggests that intact ecosystems can reduce transmission, disease prevention has largely been neglected in conservation efforts and remains underfunded compared to mitigation. A major constraint is the lack of reliable, spatially explicit information to guide efforts effectively. Given the increasing rate of new disease emergence, accelerated by climate change and biodiversity loss, identifying priority areas for mitigating the risk of disease transmission is more crucial than ever. We present new high-resolution (1 km) maps of priority areas for targeted ecological countermeasures aimed at reducing the likelihood of zoonotic spillover, along with a methodology adaptable to local contexts. Our study compiles data on well-documented risk factors, protection status, forest restoration potential, and opportunity cost of the land to map areas with high potential for cost-effective interventions. We identify low-cost priority areas across 50 countries, including 277,000 km2 where environmental restoration could mitigate the risk of zoonotic spillover and 198,000 km2 where preventing deforestation could do the same, 95% of which are not currently under protection. The resulting layers, covering tropical regions globally, are freely available alongside an interactive no-code platform that allows users to adjust parameters and identify priority areas at multiple scales. Ecological countermeasures can be a cost-effective strategy for reducing the emergence of new pathogens; however, our study highlights the extent to which current conservation efforts fall short of this goal.

Platforms matching spatially distributed supply to demand face a fundamental design choice: given a fixed total budget of service range, how should it be allocated across supply nodes ex ante, i.e. before supply and demand locations are realized, to maximize fulfilled demand? We model this problem using bipartite random geometric graphs where $n$ supply and $m$ demand nodes are uniformly distributed on $[0,1]^k$ ($k \ge 1$), and edges form when demand falls within a supply node's service region, the volume of which is determined by its service range. Since each supply node serves at most one demand, platform performance is determined by the expected size of a maximum matching. We establish a uniformity principle: whenever one service range allocation is more uniform than the other, the more uniform allocation yields a larger expected matching. This principle emerges from diminishing marginal returns to range expanding service range, and limited interference between supply nodes due to bounded ranges naturally fragmenting the graph. For $k=1$, we further characterize the expected matching size through a Markov chain embedding and derive closed-form expressions for special cases. Our results provide theoretical guidance for optimizing service range allocation and designing incentive structures in ride-hailing, on-demand labor markets, and drone delivery networks.

Recent advances, such as RegretNet, ALGnet, RegretFormer and CITransNet, use deep learning to approximate optimal multi item auctions by relaxing incentive compatibility (IC) and measuring its violation via ex post regret. However, the true accuracy of these regret estimates remains unclear. Computing exact regret is computationally intractable, and current models rely on gradient based optimizers whose outcomes depend heavily on hyperparameter choices. Through extensive experiments, we reveal that existing methods systematically underestimate actual regret (In some models, the true regret is several hundred times larger than the reported regret), leading to overstated claims of IC and revenue. To address this issue, we derive a lower bound on regret and introduce an efficient item wise regret approximation. Building on this, we propose a guided refinement procedure that substantially improves regret estimation accuracy while reducing computational cost. Our method provides a more reliable foundation for evaluating incentive compatibility in deep learning based auction mechanisms and highlights the need to reassess prior performance claims in this area.

We extend the results of (Liu and Firoozi, 2025), which develops the theory of linear-quadratic (LQ) mean field games in Hilbert spaces, by incorporating a common noise. This common noise is an infinite-dimensional Wiener process affecting 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 forward-backward deterministic evolution equations. We establish the existence and uniqueness of solutions to the coupled linear FBSEEs associated with the LQ MFG setting for small time horizons and prove the $\epsilon$-Nash property of the resulting equilibrium strategy. Furthermore, for the first time in the literature, we develop an analysis that establishes the well-posedness of these coupled linear FBSEEs in Hilbert spaces, for which only mild solutions exist, over arbitrary finite time horizons.

We introduce Look-Ahead-Bench, a standardized benchmark measuring look-ahead bias in Point-in-Time (PiT) Large Language Models (LLMs) within realistic and practical financial workflows. Unlike most existing approaches that primarily test inner lookahead knowledge via Q\\&A, our benchmark evaluates model behavior in practical scenarios. To distinguish genuine predictive capability from memorization-based performance, we analyze performance decay across temporally distinct market regimes, incorporating several quantitative baselines to establish performance thresholds. We evaluate prominent open-source LLMs -- Llama 3.1 (8B and 70B) and DeepSeek 3.2 -- against a family of Point-in-Time LLMs (Pitinf-Small, Pitinf-Medium, and frontier-level model Pitinf-Large) from PiT-Inference. Results reveal significant lookahead bias in standard LLMs, as measured with alpha decay, unlike Pitinf models, which demonstrate improved generalization and reasoning abilities as they scale in size. This work establishes a foundation for the standardized evaluation of temporal bias in financial LLMs and provides a practical framework for identifying models suitable for real-world deployment. Code is available on GitHub: this https URL

We introduce BallotRank, a ranked preference aggregation method derived from a modified PageRank algorithm. It is a Condorcet-consistent method without damping, and empirical examination of nearly 2,000 ranked choice elections and over 20,000 internet polls confirms that BallotRank always identifies the Condorcet winner at conventional values of the damping parameter. We also prove that the method satisfies many of the same social choice criteria as other well-known Condorcet completion methods, but it has the advantage of being a natural social welfare function that provides a full ranking of the candidates.

Systemic risk measures (SRMs) are crucial for the stability of financial markets, yet classical formulations fail to capture the complexity of market volatility. We propose a new framework for systemic risk measurement on the variable-exponent Bochner-Lebesgue space (VEBLS) $L^{p(\cdot)}$, where the exponent $p(\cdot)$ is modeled as a random variable, thereby incorporating latent volatility effects. By introducing appropriate deterministic functions (DFs) and single-firm risk measures (SFRMs), we decompose systemic risk assessment in $L^{p(\cdot)}$ into two sequential components and establish corresponding dual representations. Several examples are provided to illustrate the theoretical results.

Many countries have adopted negative interest rate policies with tiering remuneration, which allows for exemption from negative rates. This practice has led to higher interbank trading volumes, with market rates ranging between zero and the negative remuneration rates. This study proposes a basic model of an interbank market with tiering remuneration that can be tested with actual market data because of its simplicity and can indicate the level of the market rate created by the different exemption levels. By generalizing the model, we found that a tiering system is also suitable for maintaining a higher trading activity, regardless of the level of the remuneration rate.

The random values and volumes of consecutive trades made at the exchange with shares of security determine its mean, variance, and higher statistical moments. The volume weighted average price (VWAP) is the simplest example of such a dependence. We derive the dependence of the market-based variance and 3rd statistical moment of prices on the means, variances, covariances, and 3rd moments of the values and volumes of market trades. The usual frequency-based assessments of statistical moments of prices are the limited case of market-based statistical moments if we assume that all volumes of consecutive trades with security are constant during the averaging interval. To forecast market-based variance of price, one should predict the first two statistical moments and the correlation of values and volumes of consecutive trades at the same horizon. We explain how that limits the number of predicted statistical moments of prices by the first two and the accuracy of the forecasts of the price probability by the Gaussian distribution. This limitation also reduces the reliability of Value-at-Risk by Gaussian approximation. The accounting for the randomness of trade volumes and the use of VWAP results in zero price-volume correlations. To study the price-volume empirical statistical dependence, one should calculate correlations of prices and squares of trade volumes or correlations of squares of prices and volumes. To improve the accuracy and reliability of large macroeconomic and market models like those developed by BlackRock's Aladdin, JP Morgan, and the U.S. Fed., the developers should explicitly account for the impact of random trade volumes and use market-based statistical moments of asset prices.

We evaluate two interventions facilitating technology-sector transitions for women in Poland: Mentoring, focused on expanding professional networks, and Challenges, focused on building credible skill signals. Randomizing oversubscribed admissions, we find both programs substantially increase technology employment at twelve months - by 15 percentage points for Mentoring and 11 p.p. for Challenges. The distinct mechanisms through which the programs operate translate to heterogeneous treatment effects across geography, career stage, and baseline credentials. These differential effects create scope for improved allocation: algorithmic targeting across programs outperforms random assignment by 86% and experts' selection into Mentoring by 11%.

A new version of the database for the meta-analysis of estimates of the social cost of carbon is presented. New records were added, and new fields on gender and stochasticity.

The primary challenge of market making in spot precious metals is navigating the liquidity that is mainly provided by futures contracts. The Exchange for Physical (EFP) spread, which is the price difference between futures and spot, plays a pivotal role and exhibits multiple modes of relaxation corresponding to the diverse trading horizons of market participants. In this paper, we model the EFP spread using a nested Ornstein-Uhlenbeck process, in the spirit of the two-factor Hull-White model for interest rates. We demonstrate the suitability of the framework for maximizing the expected P\&L of a market maker while minimizing inventory risk across both spot and futures. Using a computationally efficient technique to approximate the solution of the Hamilton-Jacobi-Bellman equation associated with the corresponding stochastic optimal control problem, our methodology facilitates strategy optimization on demand in near real-time, paving the way for advanced algorithmic market making that capitalizes on the co-integration properties intrinsic to the precious metals sector.

Current AI systems are better than humans in some knowledge dimensions but weaker in others. Guided by the long-standing vision of machine intelligence inspired by the Turing Test, AI developers increasingly seek to eliminate this "jagged" nature by pursuing Artificial General Intelligence (AGI) that surpasses human knowledge across domains. This pursuit has sparked an important debate, with leading economists arguing that AGI risks eroding the value of human capital. We contribute to this debate by showing how AI capabilities in different dimensions shape labor income in a multidimensional knowledge economy. AI improvements in dimensions where it is stronger than humans always increase labor income, but the effects of AI progress in dimensions where it is weaker than humans depend on the nature of human-AI communication. When communication allows the integration of partial solutions, improvements in AI's weak dimensions reduce the marginal product of labor, and labor income is maximized by a deliberately jagged form of AI. In contrast, when communication is limited to sharing full solutions, improvements in AI's weak dimensions can raise the marginal product of labor, and labor income can be maximized when AI achieves high performance across all dimensions. These results point to the importance of empirically assessing the additivity properties of human-AI communication for understanding the labor-market consequences of progress toward AGI.

This paper develops distribution-based measures that extract policy-relevant information from subjective probability distributions beyond point forecasts. We introduce two complementary indicators that operationalize the second and third moments of beliefs. First, a Normalized Uncertainty measure applies a variance-stabilizing transformation that removes mechanical level effects around policy-relevant anchors. Empirically, uncertainty behaves as a state variable: it amplifies perceived de-anchoring following monetary-policy shocks and weakens and delays pass-through to credit conditions, particularly across loan maturities. Second, an Asymmetry Coherence indicator combines the median and skewness of subjective distributions to identify coherent directional tail risks. Directional asymmetry is largely orthogonal to uncertainty and is primarily reflected in monetary-policy responses rather than real activity. Overall, the results show that properly measured uncertainty governs state-dependent transmission, while distributional asymmetries convey distinct information about macroeconomic risks.

We study strategic interactions in a broker-mediated market in which agents learn and exploit each other's private information. A broker provides liquidity to an informed trader and to noise traders while managing inventory in a lit market. The informed trader infers the broker's trading activity in the lit market, while the broker estimates the trader's private signal. Information leakage in the client's trading flow generates economic value for the broker that is comparable in magnitude to transaction costs: the broker can speculate profitably and manage risk more effectively, which in turn adversely affects the informed trader's performance. Brokers therefore hold a strategic advantage over traders who rely solely on prices to filter information. When the broker only relies on prices rather than client trading flow to infer information, their trading performance becomes indistinguishable from the performance of a naive strategy that internalises noise flow, externalises informed flow, and offloads inventory at a constant rate.

The long term estimation of the Marxist average rate of profit does not adhere to a theoretically grounded standard regarding which economic activities should or should not be included for such purposes, which is relevant because methodological non uniformity can be a significant source of overestimation or underestimation, generating a less accurate reflection of the capital accumulation dynamics. This research aims to provide a standard Marxist decision criterion regarding the inclusion and exclusion of economic activities for the calculation of the Marxist average profit rate for the case of United States economic sectors from 1960 to 2020, based on the Marxist definition of productive labor, its location in the circuit of capital, and its relationship with the production of surplus value. Using wavelet transformed Daubechies filters with increased symmetry, empirical mode decomposition, Hodrick Prescott filter embedded in unobserved components model, and a wide variety of unit root tests the internal theoretical consistency of the presented criteria is evaluated. Also, the objective consistency of the theory is evaluated by a dynamic factor autoregressive model, Principal Component Analysis via Singular Value Decomposition, and regularized Horseshoe regression. The results are consistent both theoretically and econometrically with the logic of Classical Marxist political economy.

What happens to pollution when developing countries open their borders to trade? Theoretical predictions are ambiguous, and empirical evidence remains limited. We study the effects of the 1991 Indian trade liberalization reform on water pollution. The reform abruptly and unexpectedly lowered import tariffs, increasing exposure to trade. Larger tariff reductions are associated with relative increases in water pollution. The estimated effects imply a 0.11 standard deviation increase in water pollution for the median district exposed to the tariff reform.

While electric vehicle (EV) adoption has been widely studied, most research focuses on the average effects of predictors on purchase intent, overlooking variation across the distribution of EV purchase intent. This paper makes a threefold contribution by analyzing four unique explanatory variables, leveraging large-scale US survey data from 2021 to 2023, and employing Bayesian ordinal probit and Bayesian ordinal quantile modeling to evaluate the effects of these variables-while controlling for other commonly used covariates-on EV purchase intent, both on average and across its full distribution. By modeling purchase intent as an ordered outcome-from "not at all likely" to "very likely"-we reveal how covariate effects differ across levels of interest. This is the first application of ordinal quantile modeling in the EV adoption literature, uncovering heterogeneity in how potential buyers respond to key factors. For instance, confidence in development of charging infrastructure and belief in environmental benefits are linked not only to higher interest among likely adopters but also to reduced resistance among more skeptical respondents. Notably, we identify a gap between the prevalence and influence of key predictors: although few respondents report strong infrastructure confidence or frequent EV information exposure, both factors are strongly associated with increased intent across the spectrum. These findings suggest clear opportunities for targeted communication and outreach, alongside infrastructure investment, to support widespread EV adoption.

We introduce the resilience rate as a measure of financial resilience. It captures the expected rate at which a dynamic risk measure recovers, i.e., bounces back, when the risk-acceptance set is breached. We develop the corresponding stochastic calculus by establishing representation theorems for expected time-derivatives of solutions to backward stochastic differential equations (BSDEs) with jumps, evaluated at stopping times. These results reveal that the resilience rate can be represented as a suitable expectation of the generator of a BSDE. We analyze the main properties of the resilience rate and the formal connection of these properties to the BSDE generator. We also introduce resilience-acceptance sets and study their properties in relation to both the resilience rate and the dynamic risk measure. We illustrate our results in several canonical financial examples and highlight their implications via the notion of resilience neutrality.

The growth of large-scale AI systems is increasingly constrained by infrastructure limits: power availability, thermal and water constraints, interconnect scaling, memory pressure, data-pipeline throughput, and rapidly escalating lifecycle cost. Across hyperscale clusters, these constraints interact, yet the main metrics remain fragmented. Existing metrics, ranging from facility measures (PUE) and rack power density to network metrics (all-reduce latency), data-pipeline measures, and financial metrics (TCO series), each capture only their own domain and provide no integrated view of how physical, computational, and economic constraints interact. This fragmentation obscures the structural relationships among energy, computation, and cost, preventing a coherent optimization across sector and how bottlenecks emerge, propagate, and jointly determine the efficiency frontier of AI infrastructure. This paper develops an integrated framework that unifies these disparate metrics through a three-domain semantic classification and a six-layer architectural decomposition, producing a 6x3 taxonomy that maps how various sectors propagate across the AI infrastructure stack. The taxonomy is grounded in a systematic review and meta-analysis of all metrics with economic and financial relevance, identifying the most widely used measures, their research intensity, and their cross-domain interdependencies. Building on this evidence base, the Metric Propagation Graph (MPG) formalizes cross-layer dependencies, enabling systemwide interpretation, composite-metric construction, and multi-objective optimization of energy, carbon, and cost. The framework offers a coherent foundation for benchmarking, cluster design, capacity planning, and lifecycle economic analysis by linking physical operations, computational efficiency, and cost outcomes within a unified analytic structure.

Accurately forecasting daily exchange rate returns represents a longstanding challenge in international finance, as the exchange rate returns are driven by a multitude of correlated market factors and exhibit high-frequency fluctuations. This paper proposes EXFormer, a novel Transformer-based architecture specifically designed for forecasting the daily exchange rate returns. We introduce a multi-scale trend-aware self-attention mechanism that employs parallel convolutional branches with differing receptive fields to align observations on the basis of local slopes, preserving long-range dependencies while remaining sensitive to regime shifts. A dynamic variable selector assigns time-varying importance weights to 28 exogenous covariates related to exchange rate returns, providing pre-hoc interpretability. An embedded squeeze-and-excitation block recalibrates channel responses to emphasize informative features and depress noise in the forecasting. Using the daily data for EUR/USD, USD/JPY, and GBP/USD, we conduct out-of-sample evaluations across five different sliding windows. EXFormer consistently outperforms the random walk and other baselines, improving directional accuracy by a statistically significant margin of up to 8.5--22.8%. In nearly one year of trading backtests, the model converts these gains into cumulative returns of 18%, 25%, and 18% for the three pairs, with Sharpe ratios exceeding 1.8. When conservative transaction costs and slippage are accounted for, EXFormer retains cumulative returns of 7%, 19%, and 9%, while other baselines achieve negative. The robustness checks further confirm the model's superiority under high-volatility and bear-market regimes. EXFormer furnishes both economically valuable forecasts and transparent, time-varying insights into the drivers of exchange rate dynamics for international investors, corporations, and central bank practitioners.

We study choice among lotteries in which the decision maker chooses from a small library of decision rules. At each menu, the applied rule must make the realized choice a strict improvement under a dominance benchmark on perceived lotteries. We characterize the maximal Herfindahl-Hirschman concentration of rule shares over all locally admissible assignments, and diagnostics that distinguish rules that unify behavior across many menus from rules that mainly act as substitutes. We provide a MIQP formulation, a scalable heuristic, and a finite-sample permutation test of excess concentration relative to a menu-independent random-choice benchmark. Applied to the CPC18 dataset (N=686 subjects, each making 500-700 repeated binary lottery choices), the mean MRCI is 0.545, and 64.1% of subjects reject random choice at the 1% level. Concentration gains are primarily driven by modal-payoff focusing, salience-thinking, and regret-based comparisons.

We identify volatility spillovers across commodities, equities, and treasuries using a hybrid HAR-ElasticNet framework on daily realized volatility for six futures markets over 2002--2025. Our two step procedure estimates own-volatility dynamics via OLS to preserve persistence, then applies ElasticNet regularization to cross-market spillovers. The sparse network structure that emerges shows equity markets (ES, NQ) act as the primary volatility transmitters, while crude oil (CL) ends up being the largest receiver of cross-market shocks. Agricultural commodities stay isolated from the larger network. A simple univariate HAR model achieves equally performing point forecasts as our model, but our approach reveals network structure that univariate models cannot. Joint Impulse Response Functions trace how shocks propagate through the network. Our contribution is to demonstrate that hybrid estimation methods can identify meaningful spillover pathways while preserving forecast performance.

We study how the general public perceives the link between monetary policy and housing markets. Using a large-scale, cross-country survey experiment in Austria, Germany, Italy, Sweden, and the United Kingdom, we examine households' understanding of monetary policy, their beliefs about its impact on house prices, and how these beliefs respond to expert information. We find that while most respondents grasp the basic mechanisms of conventional monetary policy and recognize the connection between interest rates and house prices, literacy regarding unconventional monetary policy is very low. Beliefs about the monetary policy-housing nexus are malleable and respond to information, particularly when it is provided by academic economists rather than central bankers. Monetary policy literacy is strongly related to education, gender, age, and experience in housing and mortgage markets. Our results highlight the central role of housing in how households interpret monetary policy and point to the importance of credible and inclusive communication strategies for effective policy transmission.

We present a simple, numerically efficient but highly flexible non-parametric method to construct representations of option price surfaces which are both smooth and strictly arbitrage-free across time and strike. The method can be viewed as a smooth generalization of the widely-known linear interpolation scheme, and retains the simplicity and transparency of that baseline. Calibration of the model to observed market quotes is formulated as a linear program, allowing bid-ask spreads to be incorporated directly via linear penalties or inequalities, and delivering materially lower computational cost than most of the currently available implied-volatility surface fitting routines. As a further contribution, we derive an equivalent parameterization of the proposed surface in terms of strictly positive "discrete local volatility" variables. This yields, to our knowledge, the first construction of smooth, strictly arbitrage-free option price surfaces while requiring only trivial parameter constraints (positivity). We illustrate the approach using S&P 500 index options

We develop a framework for composite likelihood estimation of parametric continuous-time stationary Gaussian processes. We derive the asymptotic theory of the associated maximum composite likelihood estimator. We implement our approach on a pair of models that have been proposed to describe the random log-spot variance of financial asset returns. A simulation study shows that it delivers good performance in these settings and improves upon a method-of-moments estimation. In an empirical investigation, we inspect the dynamic of an intraday measure of the spot log-realized variance computed with high-frequency data from the cryptocurrency market. The evidence supports a mechanism, where the short- and long-term correlation structure of stochastic volatility are decoupled in order to capture its properties at different time scales. This is further backed by an analysis of the associated spot log-trading volume.

Large language models (LLMs) are increasingly used to support the analysis of complex financial disclosures, yet their reliability, behavioral consistency, and transparency remain insufficiently understood in high-stakes settings. This paper presents a controlled evaluation of five transformer-based LLMs applied to question answering over the Business sections of U.S. 10-K filings. To capture complementary aspects of model behavior, we combine human evaluation, automated similarity metrics, and behavioral diagnostics under standardized and context-controlled prompting conditions. Human assessments indicate that models differ in their average performance across qualitative dimensions such as relevance, completeness, clarity, conciseness, and factual accuracy, though inter-rater agreement is modest, reflecting the subjective nature of these criteria. Automated metrics reveal systematic differences in lexical overlap and semantic similarity across models, while behavioral diagnostics highlight variation in response stability and cross-prompt alignment. Importantly, no single model consistently dominates across all evaluation perspectives. Together, these findings suggest that apparent performance differences should be interpreted as relative tendencies under the tested conditions rather than definitive indicators of general reliability. The results underscore the need for evaluation frameworks that account for human disagreement, behavioral variability, and interpretability when deploying LLMs in financially consequential applications.