A Novel Budget-Based C+SVM Model for Credit Risk Prediction
Researchers introduced a budget‑aware C+SVM model that blends cost constraints with support vector machine classification to predict credit risk. The approach simultaneously selects optimal features and assigns weights, targeting the lowest total misjudgment cost while staying within a predefined budget. Empirical testing on a dataset of Chinese farmers showed the model delivers accuracy comparable to conventional methods but at reduced information‑gathering expense. The findings suggest banks can achieve reliable default forecasts without overspending on data collection.
Graph Neural Networks for Credit Default Prediction: Robustness and Model Evaluation
Graph neural networks (GNNs) improve credit default prediction by leveraging borrower similarity graphs, consistently beating strong tabular baselines such as logistic regression and gradient boosting. The study finds that well‑tuned GNNs occupy broad hyperparameter regions, offering stable and reproducible performance....
Abaxx: Meeting the Need for New Commodity Derivatives
Abaxx Exchange’s new white paper outlines a suite of physically deliverable commodity derivatives designed to tackle extreme price volatility, geopolitical uncertainty, and the shifting dynamics of the energy transition. Traditional benchmark contracts are increasingly misaligned with physical markets, creating heightened...
An Econometric Investigation on the Stability of Stablecoins: Are These Coins Stable or Is Their Stability Just a Flip of...
A new econometric study probes the volatility of U.S.-dollar‑backed stablecoins, using multilevel models such as GARCH and SVAR to assess reactions to macro‑financial shocks. The analysis finds USD Coin and TrueUSD highly sensitive to monetary policy changes and market turbulence, while...
Instant Payments as the New Normal: How Much More Money Do the Banks Need?
The shift from batch‑based to instant retail payments modestly raises banks' overall liquidity requirements, but the impact varies dramatically across individual institutions. Researchers modeled Finnish STEP2 transaction data to compare liquidity needs under net‑ting cycles versus real‑time gross settlement. Their...
The Merchant’s Hand in the Consumer’s Choice of Payment Instruments: An Agent-Based Model
The paper presents an agent‑based model that simulates the German retail‑payment ecosystem using detailed sociodemographic and transaction data from the Deutsche Bundesbank. By varying merchant acceptance rates—such as reducing cash outlets or expanding card terminals—the model shows that even modest...
Iran Confusion Makes the Case for Causal Modelling
Risk managers are grappling with the tangled uncertainties of the Iran‑US conflict, where traditional scenario testing—relying on past crises—fails to capture the unique dynamics of oil prices and prolonged infrastructure damage. Alexander Denev showed that Anthropic’s Claude LLM can construct...
Locked Out by Loyalty: Entry Deterrence Through Rebates in Payment Card Markets
A new academic paper shows that incumbent payment‑card networks use generous rebates to issuing banks as a strategic barrier against new entrants. The study finds that when transaction benefits for consumers and merchants rise, networks increase rebate levels, making entry...
Industrialising the Challenge Process: AI in Operational Risk Scenario Analysis
Banks are moving operational risk scenario analysis from a regulatory back‑stop to a core forward‑looking tool for capital planning and resilience. Patrick Naim and Nedim Baruh argue that structured modelling—exposure, occurrence, impact (XOI)—turns narrative scenarios into parameterised loss generators. Yet...
Fast Calculation of Cheapest-to-Deliver Curves
Multi‑currency collateral agreements create optionality that requires discounting with a cheapest‑to‑deliver (CTD) curve. While Monte‑Carlo simulation can price this exactly, it is computationally heavy. Researchers propose an analytic approximation that combines the Clark algorithm and Gauss‑Hermite quadrature, delivering near‑Monte‑Carlo accuracy...
Strong Order-One-Half Convergence of the Projected Euler–Maruyama Method for the Cox–Ingersoll–Ross Model
The paper proves that the projected Euler–Maruyama scheme attains Lp‑strong convergence of order one‑half for the Cox–Ingersoll–Ross (CIR) model. By integrating a projection step with the normalized error framework, the authors broaden the parameter space where the convergence guarantee holds....