Papers

This paper introduces and studies the Multi-Objective Enterprise Green Supply Chain Network Design problem (MOEDGSCND) an extension of the well known Multi-Objective Supply Chain Network Design (MO-SCND), in this case a three-objective combinatorial optimisation problem that simultaneously minimises total cost f1, carbon dioxide emissions f2, and customer dissatisfaction f3 over binary decisions on which distribution centres to open and how to assign customer zones to open facilities. The problem generalises capacitated facility location and is known to be NP-hard. Furthermore this paper lays out a formal specification of decision variables, objective functions, all constraints and then investigates and compares three solution approaches using a single numerical instance: (A) exact weighted-sum integer linear programming solved via PuLP, (B) three Fully Polynomial-Time Approximation Schemes (FPTAS) providing (.94 + $\epsilon$)-guarantee approximations for $\epsilon$ := 0.15, and (C) two baseline heuristics for comparison. On an eight-customer, five-candidate-DC instance the exact solver certifies the optimal Pareto front across eight weight combinations, and all three FPTAS variants achieve their approximation guarantees, finishing within milliseconds compared to centiseconds for the exact ILP using the PuLP solver on this scale.

Posted prices for AI inference have fallen steadily since 2024, yet the measured speed of that fall depends almost entirely on the method of measurement. This paper constructs quality-adjusted price indices for the AI inference market from public data. The panel assembles 21,024 posted-price observations across 3,208 models and 86 providers and joins them to 4,605 benchmark scores through a latent quality index estimated from benchmark response patterns, so the quality ladder of the hedonic tradition is built here from evaluations in place of product characteristics. Measured by the matched-model methods that statistical agencies apply to software, inference prices fell at 0.10 log points a year. The quality-adjusted index fell at 0.73, so 87% of the decline is invisible to current methods, with direct consequences for measured competition, concentration and productivity in this market. Counted per completed task, moreover, the buyer's price stopped falling. Reasoning models raised token consumption faster than token prices fell, and the seller's and buyer's prices accordingly diverged. A pre-registered validity audit disciplines the quality measure and yields the sharpest result. Excluding contamination-flagged benchmarks leaves model rankings intact at 0.998 yet moves the index by 0.49 log points a year, so the leaderboard-stability arguments standard in AI evaluation offer no defence of economic statistics built on benchmarks. Prices, quality and the audit are fully reproducible from public sources at zero cost.

Renewables now dominate Great Britain's generation mix but rarely occupy the marginal price-setting position, which raises the question of which flexible technologies translate a renewable-rich system into real-time price formation. This study reconstructs the price-ranked edge of the eligible bid or offer stack in the GB Balancing Mechanism across 50,684 half-hourly Settlement Periods from 2023 to 2025 and attributes it to individual Balancing Mechanism Units, separating long-system bid-active from short-system offer-active conditions and retaining co-marginal ties. Batteries rose from 0.8% to 36.2% of bid-active and from 3.2% to 26.9% of offer-active marginality, displacing combined-cycle gas on the bid side and pumped storage on the offer side. Capacity-normalised marginal capture rose on both sides, and unit-quarter fixed effects show that an additional 100 MW of Balancing Mechanism active capacity was associated with a 0.55 and 0.48 percentage-point increase in quarterly bid- and offer-side marginal share. By 2025, battery actions were around {\pounds}9 to 10/MWh more favourable than volume-matched non-battery alternatives drawn from the same ladder, yet batteries were under-represented by 12.9 percentage points in the highest-priced 5% of short-system periods. Individual batteries remain consistent with price-taking, while the fleet has become endogenous to routine balancing price formation.

This paper develops a latent-state framework for recovering borrower-level posterior beliefs in credit-risk analysis. Creditworthiness and financial fragility are represented as latent dimensions, while observed borrower scores follow a finite Gaussian mixture model and default depends on the latent profile. Borrower-specific probabilities of default are obtained by averaging profile-specific default probabilities over the posterior distribution of latent states. A joint Expectation–Maximization procedure is used to estimate the mixture structure and the profile-specific default probabilities from observed score–default pairs. After estimation, predictive posterior beliefs are computed using the observed scores alone, thereby preserving the information available before default realization. A controlled simulation experiment evaluates the recovery of structural parameters, posterior beliefs, and borrower-level probabilities of default. Posterior distributions are interpreted as points on the probability simplex, and their recovery is assessed using both conventional error measures and information-geometric divergences. The results provide a controlled benchmark for studying the interaction between latent economic structure, posterior uncertainty, and credit-risk prediction.

Portfolio risk assessment ordinarily relies on reliable estimates of cross-asset return covariances, which are difficult to obtain in short, high-dimensional panels. We show that firm-level distribution-valued characteristics can instead provide one-sided certificates of portfolio risk. Under maintained links from characteristics to systematic exposures and from exposures to returns, multi-firm Wasserstein-2 dispersion yields a sharp upper bound on systematic portfolio variance and a corresponding bound for standardized returns. A weighted pairwise relaxation produces an objective that is convex under a checkable condition and requires marginal volatility scales but no cross-asset return covariances. With zero firm-specific slack, the common-map scale changes the certified variance reduction but not the normalized allocation, which depends only on observed information geometry. In a 52-firm panel from 2018-2022, an allocation constructed from Qwen3-Embedding-8B news representations lies between the 0.69th and 1.33rd in-sample variance percentiles across four prespecified capped portfolio populations; equal risk weighting lies between the 21.1st and 28.6th percentiles. The lower in-sample variance ranking relative to equal risk also appears across the reported frozen language-model representations. The framework therefore distribution-valued firm information into a coherent risk bound and an implementable allocation rule constructed without cross-asset return covariances.

Spatial asset-pricing models take the structure of inter-firm interaction as given. We infer that structure from firms' information environments using language-model representations. Each firm is represented as a distribution of news-article embeddings, and a target-anchored Wasserstein barycentric reconstruction selects, for every firm, the weighted combination of other firms whose information footprints jointly reconstruct its own. The resulting directed peer field enters a quadratic exposure-adjustment model in which the spatial coefficient indexes alignment with information peers relative to stand-alone exposure. Using fields built from 2018-2022 news and frozen before 2023-2026 returns, we find that the constructed field organizes cross-sectional return dependence beyond the Fama-French five factors and momentum and raises the held-out mean Gaussian quasi-log score relative to a matched factor-only model. Because factor betas are unchanged, the gain lies in residual covariance. The field outperforms pairwise distance weighting and equal weighting of the same peers, and remains incrementally informative beside persistent news co-mentions under the primary factor-conditioned specification. Linear and quadratic transport generate nearly identical peer-return signals and equivalent held-out predictive performance. The barycentric-proximity ordering persists across alternative embedding models, and a pre-period encoder preserves the held-out advantage under the primary specification. Language-model representations thus serve as a measurement instrument for latent inter-firm information structure in capital markets.

This paper is devoted to developing a framework for stochastic growth models with environmental risk, in which rare but catastrophic shocks interact with capital accumulation and pollution. Building on the Poisson point process formulation studied in arXiv:2511.13568, we extend the model to disasters driven by a marked Hawkes process, allowing past disasters to temporarily increase the likelihood of subsequent shocks. Our work focuses on a subcritical Markovian Hawkes specification, in which the state space is augmented by the self-excitation component of disaster risk. We establish the well-posedness and nonexplosion of the resulting controlled Hawkes dynamics. Using the Hamilton-Jacobi-Bellman characterization of the corresponding Poisson control problem, we obtain quantitative estimates for the Hawkes excitation process and prove, under a small-excitation scaling, that the Hawkes value function converges to its Poisson counterpart as the magnitude of self-excitation vanishes. This provides a rigorous Poisson approximation of the stochastic control problem and quantifies the effect of self-excitation on optimal growth under environmental disaster risk.

We study the quadratic tracking problem of a general stochastic target process with absolutely continuous controls, with and without terminal constraint. We derive explicit, non-asymptotic upper bounds in terms of a Besov-type modulus of the target. These bounds yield sharp explicit rates that specialize to the square-root order for semimartingale targets. We then apply these results to a generalized Obizhaeva–Wang execution model with random terminal inventory. We first develop a Hilbert-space approach to characterize its optimal strategy, which includes jumps. To avoid such trading spikes, one regularizes the problem by a quadratic trading-rate penalty with coefficient $\varepsilon$. We then show that the regularized optimal execution cost—and therefore the excess price impact cost of the regularized optimal strategy—converges at the sharp rate $O(\sqrt{\varepsilon})$. Since the regularized optimal strategy is not available in closed form, we further construct a nearly optimal strategy which is readily implementable and shares the same approximation rate.

The factor HJM stochastic volatility model introduced by Sepp and Rakhmonov (2025) obtains tractable swaption pricing by freezing the nonlinear swap-rate loading along a deterministic expected-state path. This removes the dependence of conditional swap-rate variance on the current yield-curve state. We introduce a first-order Taylor correction to the loading that adds no calibration parameters. Conditional on retaining the frozen annuity-measure drift, we show that the first variation of the swap-rate transform is affine in the centered rate states and reduces to one-dimensional equations in volatility. For quadratic-drift lognormal stochastic volatility, these equations yield a finite-dimensional ODE representation and a direct log-volatility formulation. Calibrations to independently generated nonlinear-model prices show substantially lower stochastic volatility parameter bias and held-out pricing error, with only modest changes in local calibration identifiability.

Classical option-hedging methods like Black-Scholes delta assume constant, free rebalancing, which real markets don't allow. Deep hedging trains a neural network to handle these frictions directly, and prior work reports strong results. But those comparisons usually pit deep hedging against a frictionless classical baseline on simulated price data. That's not a fair fight, and it leaves open whether the advantage is real. We test this using five years of actual BTC options data from Deribit (2020-2024), comparing Black-Scholes delta, Leland's cost-adjusted hedge, and the Whalley-Wilmott no-trade band against three deep hedging setups: an LSTM and a feedforward network, each trained with a CVaR loss and, in some runs, a penalty for trading too often. All six strategies face the same 5 basis point transaction cost. On 11,546 test episodes from September 2023 to December 2024, Whalley-Wilmott cuts transaction costs significantly versus hourly rebalancing, saving $1.79 per episode against plain BS delta (95% CI [-2.21, -1.39], p < 0.0001) by trading about eight times less often. Its P&L and tail-risk numbers are better too, though not quite significant at this sample size. None of the three deep hedging models beat any classical benchmark on any metric, and all kept trading almost every hour regardless of penalty weight, a twenty-fold range barely moved the needle. A calmer validation period shows Whalley-Wilmott's P&L edge shrinks or disappears there while its cost edge holds, so the result depends on market regime. We think the likely causes are a small training set and the lack of any built-in mechanism for sitting still in the architectures tested. Not a flashy result, but an honest one: a real-data check on a claim mostly supported by simulations so far.

In financial markets, a sequential policy that reacts systematically to price movements may become predictable to other market participants. This paper studies whether large language model (LLM) agents exhibit such directional structure through RetailAgent, an experimental framework in which an LLM observes anonymized intraday equity price histories and permitted state, then repeatedly chooses long (hold the stock) or flat (stay out) before the subsequent interval return is revealed. We compare returns during long and flat intervals along the same stock's intraday path after removing the overall fraction of long decisions. This exposure-matched measure reveals persistent negative timing across modality, horizon, state, and model family. Shuffling saved action sequences substantially attenuates the effect, showing that alignment between actions and subsequent returns drives the negative score. Feeding self-authored memories into decisions further increases policy persistence, while timing becomes more negative among stock-days on which the agent uses both actions. These results reveal stable, recoverable directional structure in sequential LLM financial decisions and a behavioral signal for studying how another participant could respond to a predictable policy.

We develop a market-informed valuation framework for guaranteed minimum maturity benefit (GMMB) riders with rational surrender under the Heston stochastic-local volatility (SLV) model. The guarantee is written on the fee-deducted account value and is considered both in its terminal-only form and in the presence of early surrender rights. The Heston SLV specification combines stochastic volatility with a leverage function calibrated to a prescribed local-volatility surface. The leverage surface is obtained through a forward Markovian-projection equation so that, at the model level, the SLV dynamics are constrained to the same one-dimensional marginals as the corresponding local-volatility (LV) model. The latter is used only as a one-factor benchmark, allowing us to isolate the effect of stochastic volatility on continuation values and surrender decisions while preserving the same option-calibrated local-volatility target. We derive the associated backward pricing equations and propose a hybrid tree/finite-difference algorithm for the SLV model with a calibrated leverage function. Synthetic experiments and a market-informed case study show that SLV and LV valuations are numerically close for terminal-only guarantees, as expected from the common marginal target, whereas materially larger differences can arise once surrender is allowed. These differences are reflected in guarantee values, fair insurance fees and volatility-dependent surrender regions. The results indicate that matching one-date marginals implied by vanilla-option prices does not eliminate model risk for insurance liabilities whose value depends on conditional continuation dynamics and endogenous surrender decisions.

South Korea has the world's lowest fertility and an intense education race. Families devote nine percent of lifetime income to education, mostly to private tutoring with near-zero measured returns. I show that competition for coveted careers generates an assignment externality: when all families spend more, the admission bar rises and children become costly. In a quantitative model calibrated to this mechanism, replacing score-based assignment with a capacity-preserving lottery raises completed fertility by 0.24 children per couple. Pronatal transfers deliver one eighth of that increase. Education taxes deliver almost none. Preference shifts, not a fiercer race, explain the decline across cohorts.

To achieve carbon neutrality, electrifying vehicles in a cleaner electricity grid has become a key pathway for global cities. While extensive research has examined the environmental benefits of electric vehicles (EVs), their financial impacts remain underexplored. Using Singapore-a smart city-state - as a social laboratory, we combine high-resolution data on the geographic distribution of EV chargers with nearly 1 million individual-level monthly expenditure records from a major commercial bank and about 65,000 housing transaction records, spanning January 2019 to March 2021. Our results show that proximity to charging stations during the initial phase of EV infrastructure rollout significantly increases household consumption by 1-3%. This effect is driven primarily by public housing residents, who are generally more financially constrained than their private housing counterparts. We identify three channels underlying this increase in consumption: appreciation in housing wealth, catalysts for commercial activity and discretionary spending around EV infrastructure, and budget reallocation from fuel expenses to other consumption categories. Notably, hybrid car owners exhibit the strongest trade-offs-reducing fuel spending while increasing expenditure on entertainment or necessity goods among public housing residents. Overall, our findings demonstrate that green infrastructure can shape economic behavior and enhance consumer welfare, extending its impact beyond environmental objectives.

In this chapter, we provide a synthesis of empirical work on the intergenerational mobility in China, examining how much a child's success depends on their parents' success. We start with an overview of data sources and measure of intergenerational mobility in China's context, followed by discussion on the empirical findings in five categories of intergenerational mobility: income, education, social class, wealth, and health. Major channels of transmission of socioeconomic status across generations are investigated, including human and social capital, fertility and migration decision, housing price and trade liberalization in the market transition, and belief. Finally, we provide policy implications to promote upward mobility for children born to less advantaged parents in China's context.

Using the October 2021 announcement of Hong Kong's Northern Metropolis Plan as a quasi-natural experiment, we examine its impacts on households, firms, and demographics. Leveraging 100,576 housing transactions, 4.7 million consumption records, and district-level demographic data, difference-in-differences estimates show housing prices in the treated region rose by 3.9% within one year, alongside a 4.1% increase in consumption. Firm entry expanded, with inflows of higher-income, higher-educated households. We find a mild, lagged spillover to Shenzhen's housing market but no significant firm response. While the policy narrows cross-region inequality, it increases within-region inequality and raises affordability concerns for lower-SES households.

Large language models (LLMs) are increasingly used to discover trading strategies, and much of the resulting literature shares a methodological weakness: many candidate strategies are generated, the best is reported, and neither look-ahead bias nor the intensity of the search behind the reported result is corrected for. We present a strategy-discovery system that makes both corrections structural rather than procedural. First, the agent can only act through registry-validated tools whose feature space excludes look-ahead by construction; we show that this guardrail is not redundant with statistical correction: a deliberately leaky oracle posting a Sharpe ratio of 35 survives Deflated Sharpe and probability-of-backtest-overfitting testing completely. Second, the system records every strategy evaluation its search performs and deflates all reported performance by that trial count, tracing how the best in-sample Sharpe ratio climbs with each trial while the deflation threshold, driven by the agent's own search, climbs faster. Across a 453-stock point-in-time US equity universe and a 39-ETF multi-asset universe with realistic transaction, impact, and borrow costs, honest evaluation certifies passive benchmarks (out-of-sample confidence intervals excluding zero), rejects every LLM-discovered strategy (across two frontier models, search budgets up to one hundred candidates, and five repeated runs), catching selection luck, predicted rank degradation, and out-of-sample collapse through complementary instruments, and evaluates a human trader's production rule system under identical instruments. The framework formalizes why pre-registered hypotheses earn lower evidential bars than brute search, and quantifies the sample sizes that credible certification of moderate edges actually requires.

This study examines whether customer disclosures affect how supplier firms allocate capital across business segments. Customer disclosures can shape supplier investment decisions through two competing channels. They can improve suppliers' information about downstream demand, helping suppliers align capital with growth opportunities ("information channel"), or erode incumbent suppliers' private information advantage, inducing costly investments to defend customer relationships ("competitive-threat channel"). I use the adoption of SFAS 131 as a customer-level disclosure shock. Suppliers exposed to expanded customer disclosures experience increased product-market competition and reallocate capital toward segments with relatively weak growth-opportunity signals. Suppliers that deviate from allocations predicted by growth signals are more likely to preserve market share and expand their customer base in subsequent years. Using a novel approach to link supplier segments to customer segments, I show that this reallocation is driven by investing in capacity in affected segments, rather than by a correction of prior under-investment. Consistent with the competitive-threat channel, the investment adjustment is stronger for segments linked to larger customers and for segments operating in more concentrated industries. Segments making these investments subsequently experience lower ROA, consistent with suppliers accepting lower profitability to defend customer relationships. Overall, the findings show that disclosures can shape how economically linked firms allocate capital internally.

Bitcoin inverse options, traded on the Deribit exchange and settled in the underlying cryptocurrency rather than in fiat currency, combine extreme and genuinely rough volatility dynamics with a non-linear, currency-dependent payoff structure. This paper develops and empirically validates a pricing and calibration framework for these instruments based on the rough Bergomi (rBergomi) model of Bayer, Friz and Gatheral (2016). We adapt the rBergomi dynamics to the inverse payoff max(S_T - K, 0)/S_T, and implement and compare three computational pipelines that differ in the simulation scheme for the driving fractional Brownian motion (coarse-grid Cholesky vs. the Hybrid Scheme of Bennedsen et al., 2017) and in the Monte Carlo pricing estimator (plain log-Euler vs. the Mixed Estimator of McCrickerd and Pakkanen, 2018). The model is calibrated to thirty implied volatility surfaces extracted from Deribit trade data between May 2022 and March 2025, spanning seven major market-stress events and nine baseline regimes stratified by volatility level. The Hybrid and Mixed pipeline is simultaneously the most accurate (mean unweighted RMSE 22.83 percentage points, versus 41.76 pp for the Cholesky and Euler benchmark) and the fastest (17 seconds per snapshot, a 20-fold speed-up). The calibrated Hurst exponent is consistently close to the lower bound of the search space (H approximately equal to 0.01–0.06 in most regimes), confirming that Bitcoin's volatility is genuinely rough, and calibration error scales approximately linearly with the level of at-the-money implied volatility (Pearson r = 0.89).

Quiet quitting reflects a form of worker disaffection that operates along the intensive margin of labor supply rather than through job exit. We study how alternative workplace narratives shape workers' behavioral responses using a randomized survey experiment on a representative sample of employees in Italy and France. Respondents are exposed to empirically grounded moral framings of work emphasizing either social justice and collective rights or work organization and employment practices. We find that moral framings reallocate behavior across margins: justice-oriented narratives increase detachment while reducing passive disengagement, whereas organization-centered framing generates no systematic effects.

We consider the distribution-constrained optimal stopping problem introduced by Bayraktar and Miller (Mathematical Finance, 2019) and Beiglbock et al. (PTRF, 2018). Motivated by the multi-marginal Skorokhod embedding problem (SEP) and applications in financial mathematics, we generalize the Root-type solution to the multi-marginal case. The key difficulty is that the associated stopping barriers need not be ordered, so the problem in general cannot be reduced to a sequence of one-marginal problems. First, we give a probabilistic characterization in terms of sequential optimal stopping for an auxiliary time-reversed process, in the same spirit as Cox et al. (PTRF, 2019). Then, we prove the optimality of this construction by martingale arguments for a general class of reward functions, including multi-marginal versions of all examples in Beiglbock et al. (PTRF, 2018) as special cases.

We study how sophistication in generative AI (genAI) use varies among the back-office workforce of a large firm. Using proprietary data, we observe 713,564 employee prompts and their corresponding large language model responses from nearly 4,000 back-office employees across 15 functional areas over eight months in 2025. We document three main findings. First, senior employees exhibit more sophisticated genAI use, consistent with domain expertise complementing genAI capabilities. Second, sophistication varies considerably across functions and is highest in Strategy, Digital Innovation, and Project Management, three groups that share a focus on firmwide strategic initiatives and organizational change. Third, we observe neither improvements in sophistication over time nor lasting improvements following formal AI training, suggesting that sophisticated use can be difficult to change. Together, our study provides measures of and insights into sophisticated genAI use that managers can use to improve outcomes and that researchers can use in future research.

This paper develops a unified framework for assessing systemic risk and identifying contagion channels in the global banking system using a Temporal Heterogeneous Multiplex Graph Neural Network. We construct a harmonised quarterly panel combining bank fundamentals, CDS spreads, and macroeconomic indicators, and represent these data as dynamic multiplex networks linking banks through financial similarity and liquidity co-movement, augmented with country-level macroeconomic relationships. The model integrates graph convolutional layers with recurrent GRU dynamics and incorporates a learnable fusion gate to capture time-varying reliance on alternative contagion channels. Empirical results show that the framework outperforms conventional econometric, machine learning, and graph-based benchmarks for short-term changes in CDS spreads. Beyond forecasting, we provide an interpretable framework to quantify bank-level systemic importance via stress testing, assess country-level spillovers under macroeconomic shocks, and uncover transmission pathways through edge perturbation analysis. Robustness tests confirm the stability of both predictive accuracy and systemic risk rankings.

Bayesian inference in compound loss models must often be repeated across policies, market scenarios, and prior specifications. Outside conjugate cases, this may require repeated numerical integration or Markov chain Monte Carlo (MCMC). We formulate this problem as amortized posterior approximation and construct a conditional Wasserstein generative adversarial network conditioned on sufficient statistics, prior mean and coefficient of variation, and mixture weights of prior families. Notably, a single shared generator is able to approximate the posterior laws of both the Poisson intensity and the Pareto shape parameter under mixtures of Gamma, inverse-Gaussian, and lognormal priors. We assess the approximation by simulation-based calibration and by comparisons with analytical posteriors, deterministic quadrature, and extensive MCMC simulations. In an application to data on extreme natural catastrophe losses, we produce rolling one-year posterior predictive distributions, and examine the effects of heavy-tailed severity and prior-family uncertainty on aggregate tail risk.

We model equity markets using geometric Brownian particles entering and exiting at rank-dependent intensities. In the many-firm limit, the capital distribution converges to the solution of a reaction-diffusion equation with reaction term built from the intensities. Calibrated on CRSP data, the reaction term is bistable, and the long-run distribution is a traveling wave: we prove existence, uniqueness, and, for constant coefficients, exponential relaxation. Turnover, not drift, stabilizes the calibrated market. With measured volatility, the wave tracks the empirical capital distribution in every decade, determines the capitalization growth of diversity-weighted portfolios, and places the market just inside the boundary of the diverse phase. Turnover reclaims most of what rebalancing gains.

Algorithmic trading now represents a market exceeding $20 billion, where even marginal gains in signal robustness can translate into economically significant returns. Existing evaluations of equity prediction models do not explicitly target regime robustness during hyperparameter selection. Five model classes are trained on daily observations from approximately 300 large-cap US equities over eleven years, with Bayesian optimisation configured to target trading performance across three statistically different market regimes. Regime-robust hyperparameter selection is associated with out-of-sample generalisation, as signal precision remains above the random baseline across all four quarters of the test period, and portfolio performance slowly degrades under simulated input noise before collapsing beyond a defined threshold. No individual tabular deep learning architecture outperforms gradient-boosted trees, but combining XGBoost and TabNet using rank aggregation produces a Hybrid ensemble with an annualised return of 51.26%, a Sharpe ratio of 2.44, and a statistically significant CAPM alpha of 0.423 (p = 0.011). A near-zero beta indicates this outperformance is driven by stock selection, not market exposure. Alternative data plays a secondary role once technical and fundamental features are accounted for, as well as contributing more strongly on the short side than the long, and varies by model class. An interactive application makes these results explorable in real time, with live data integration the remaining step toward practical deployment.

How do we define an occupation? By its job title? An accountant at a small trading company keeps the books; at a listed firm the same title demands a certified-accountant licence, and the week goes to the reports that regulators and the board read. Same title, different bar, different work. What defines an occupation is who it lets in and what it asks them to do. In a rapidly changing labor market, tracking those requirements and tasks is how to take the market's pulse. Yet no instrument reads both at the speed they change. Official occupational directories like O*NET report one national average per occupation, updated every few years. Job postings are timely but unstructured. Research built on them works from job titles plus proprietary skill keywords, which blur what is asked of a candidate into what a candidate is asked to do. The blur matters, because rising requirements and changing tasks are different events with different causes. We separate them. From 752.6 million job ads posted on China's five leading recruitment platforms between 2022 and 2026, we extract the phrases employers write, unify those that name the same thing, and validate the mapping from text back to entry. By doing so we construct two catalogs, 20,721 requirements a candidate must meet and 44,479 tasks the hire will do. With the entries standardized, we annotate them further. Each task, for example, carries a score for how far a language model could absorb it. Matched back onto every ad, the catalogs read the market month by month. Two examples show what the layer beneath the job title buys. First, the occupational registry records one accountant where the ads record a staircase, the junior certificate at the bottom of the wage range and the intermediate one at the top. Second, counting occupations says the work most exposed to language models is disappearing, and counting tasks says far less of it is.

We extend a residual-learning hybrid credit scoring framework (logistic regression scorecard plus a gradient-boosting correction on its residuals, decomposed at each prediction into an interpretability ratio $\rho(x)$ that measures the share attributable to the linear branch) along three axes: an East African empirical instantiation on the Zindi Financial Inclusion in Africa data (Kenya, Rwanda, Tanzania, Uganda); a fairness audit at the granularity of the framework's three interpretability regions; and a thin-file segmentation analysis. On the Taiwan Credit Default benchmark retained for continuity, the calibrated hybrid attains AUC $= 0.776$ ($\Delta\mathrm{AUC} = +0.057$ vs.\ standalone logistic regression, $+0.001$ vs.\ standalone XGBoost), reduces Brier Score by 23%, and concentrates the highest-default-rate borrowers (69.5%) in the fully interpretable region. On Zindi, the calibrated hybrid attains AUC $= 0.869$ ($\Delta\mathrm{AUC} = +0.015$ vs.\ LR, $p < 0.001$; $-0.004$ vs.\ XGBoost), cuts Brier from $0.158$ to $0.085$ (a 46% reduction), and replicates the regional routing pattern. The fairness audit detects severe routing into the opaque ML-driven region along socioeconomic axes: rural respondents by 18 percentage points relative to urban, primary-or-less-educated by 32 points relative to secondary-and-above, and Ugandan respondents by 22 points relative to Kenyan, while gender shows essentially no routing disparity. The audit pipeline surfaces subgroup-routing violations that aggregate fairness metrics miss, in a form directly usable by African central-bank supervisors of digital credit.

Evaluating customer creditworthiness is crucial for retail banking operations, as it impacts marketing strategies, customer relationship management, and credit risk control. Traditional methods often struggle to capture complex temporal dependencies and extract pertinent information from customer data, crucial for accurate risk assessment. Specifically, they fail to differentiate between temporal patterns indicative of credit risk and those reflecting general customer behavior or preferences, leading to suboptimal risk predictions. In this study, we introduce the Disentangled Temporal Dependencies Variational Autoencoder (DTD-VAE), an advancement over conventional VAE, designed to disentangle temporal dependencies and distinguish credit risk-related features from past customer preferences. The feature inference module of the DTD-VAE incorporates an autoregressive temporal dependency learning mechanism that adeptly captures the temporal dependencies among latent variables, enriching the model's comprehension of the inherent data structure. Furthermore, the feature generative module utilizes an element-wise gating mechanism that assigns independent weights to each dimension of the expert models, enabling a finer-grained disentanglement of latent variables, particularly those relevant to credit risk prediction. Extensive experiments on six real-world datasets demonstrate that the proposed framework consistently outperforms existing methods, achieving performance gains of 3.2%-4.86% in ROC-AUC and 6.41%-9.71% in Accuracy Ratio.

Employers are increasingly using large language models (LLMs) to automate their hiring process. This paper investigates the risk of monocultural biases, in which the widespread deployment of large language models homogenizes biases across the labor market, leading to greater systemic exclusion for certain demographic groups. For ten LLMs, we measure hiring biases across their base and post-trained versions to identify which stage, pre-training or post-training, lead to monocultural biases. We find that, compared to their base models, post-trained models are 3.6% less likely to callback older applicants. This negative shift occurs in eight of the ten models that we evaluate. Post-trained models have much more correlated decisions than base models which is likely driven by human capital traits like skills or college major. However, greater consensus among models increases global systemic exclusion rates from 5.6% to 17.3% and exacerbates demographic inequalities, with intersectional systemic exclusion rates ranging from 12.2% to 21.7% for post-trained models. We find that this inequality is primarily driven by age-based discrimination that is exacerbated in post-training. These results indicate that while post-training techniques may improve models' abilities to select the best applicants, they may raise systemic inequality risks for those at the margin by uniformly introducing new biases.