Thought leadership, research commentary, and insights on Market Risk, Quantitative Finance, Climate Risk, and the evolving landscape of financial risk management.
AI has quietly become one of the most influential writing assistants in history. But the real question is: what happens when we stop thinking and start outsourcing too much of our thinking to machines?
Few metrics are quoted as often and misunderstood as deeply as MAPE. This article breaks down what it is, how to calculate it, how to interpret it, and what goes wrong when it's misused.
An intuitive, actionable methodology for checking PD sensitivity to macroeconomic variables — covering transmission channels, linkage mechanisms, stress testing, and strategic interpretation for IFRS 9/CECL compliance and proactive portfolio management.
A practical guide to Cameron and Trivedi's test for overdispersion — what it is, how to compute and interpret it, and why ignoring it can lead to biased inference and flawed risk decisions in insurance, credit, and operational risk modeling.
A Python guide to building conditional PD term structures by integrating macroeconomic variables into logistic regression — covering feature lagging, coefficient interpretation, scenario-based PD forecasting, and real-world applications in stress testing, IFRS 9/CECL provisioning, and portfolio management.
A Python implementation of Gordy's Granularity Adjustment (GA) for credit portfolio concentration risk — correcting the Vasicek one-factor model's infinite granularity assumption with a portfolio-level capital add-on for name and sector concentration.
A Python implementation of the Pluto-Tasche method for estimating conservative PDs in zero-default and low-default portfolios using Clopper-Pearson confidence intervals — ideal for sovereign, corporate, and IFRS 9 overlays where traditional logistic models fail.
A comprehensive guide to the Vasicek one-factor model for PD estimation — covering theoretical foundations, Python implementation with scipy, interpretation of conditional PDs under macroeconomic stress, real-world IFRS 9 and low-default portfolio applications, and pitfalls to avoid.
A step-by-step guide to building robust internal credit ratings models for PD estimation — covering objective definition, data preparation, predictor selection, model fitting, score-to-PD mapping, and governance — plus the real-world consequences of getting it wrong.
A technical deep-dive showing how logistic regression naturally fits TTC/PIT one-year PD estimation and extends to lifetime PD via macroeconomic scenario projection — with full Python implementation including PD projection, cumulative lifetime PD aggregation, and backtesting.
A step-by-step Python workflow for estimating Probability of Default (PD) — from data preparation and EDA through logistic regression modeling, prediction, and validation with ROC/AUC, bridging theory and real-world credit risk practice.
A hands-on Python implementation guide for computing Loss Given Default (LGD) estimates — from data gathering and discounting recoveries to building and validating regression models with scikit-learn.
What happens when you don't have enough data to calculate LGD accurately? This article proposes a practical framework for approximating recovery rates in data-scarce environments using sectoral benchmarking, Bayesian inference, and structural models.
A systematic technical exploration of Loss Given Default (LGD) — its conceptual framework, step-by-step computational methodology, model interpretation, and the downstream consequences of methodological errors in credit risk modeling under Basel and IFRS 9.
A practical guide to Pearson Residuals — what they are, how to compute them, how to interpret them correctly, and what risks arise when they're misunderstood in finance, insurance, and applied statistics.
Why many of the most analytically trained professionals struggle to make their voices heard in the boardroom—and how to bridge the gap between model sophistication and strategic clarity.
Heteroskedasticity can undermine regression reliability with biased standard errors and misleading hypothesis tests. This guide covers the Breusch-Pagan Test—what it is, how to compute it in Python, how to interpret results, and when to use it versus the White test.
A practical guide to Cook's Distance — what it is, how to compute and interpret it, and why ignoring influential observations can lead to biased coefficients, unstable predictions, and regulatory risk in finance and insurance modeling.
Geopolitical risk is no longer about reacting to isolated events—it's about preparing for structural divergence in policy and regulation across major economies. This article explores how to shift from event-based monitoring to scenario-based strategic modeling of policy divergence.
Even the most well-structured Risk Committees can suffer from blind spots. Here are five areas where risk committees often fall short—from model risk to risk culture—and why these gaps matter for bank resilience.
CVA has evolved from a derivatives pricing concern into a strategic accounting requirement under IFRS 9. This article explores the conceptual convergence of CVA and ECL, governance and model risk, capital vs. accounting interpretations, and CVA's strategic relevance beyond valuation.
Market risk in Shariah-based finance poses unique challenges due to the prohibition of conventional interest, derivatives, and speculative instruments. This article explores VaR modification for Islamic portfolios, dual-layer Shariah governance, stress testing, and the future integration with ESG and digital finance.
While K-Fold Cross-Validation is widely known, advanced techniques like Stratified K-Fold, Repeated K-Fold, and Nested Cross-Validation are often overlooked or misapplied. This article explores what they are and how to use them correctly.
Building a model is easy—validating it is hard. K-Fold Cross-Validation is a robust resampling technique that provides a more reliable estimate of model performance. This guide covers what it is, why to use it, how to compute it in Python, and best practices.
The Ramsey RESET test is a powerful diagnostic tool that detects whether your linear model is correctly specified or missing critical functional forms. This article covers how to compute it, interpret results, and apply it in finance, economics, and insurance.
The Kolmogorov–Smirnov (KS) statistic is one of the most widely used tools for model validation in credit scoring—yet it's often misunderstood or applied mechanically. This practical guide demystifies what KS measures, how to compute it, and how to interpret the results.
AUC-ROC is among the most respected—and misunderstood—metrics in model evaluation. This guide explains what AUC and ROC curves represent, how to compute and interpret them, and how to avoid common pitfalls in real-world predictive modeling.
The Durbin-Watson test is an early warning system against autocorrelation in regression residuals—one of the most critical yet overlooked OLS assumptions. This guide covers what DW measures, how to compute it, and how to interpret results in time series and panel data models.
Multicollinearity is one of the most subtle yet destructive issues in linear regression—it inflates variances, destabilizes coefficients, and misleads inferences. The Variance Inflation Factor (VIF) is a simple but powerful diagnostic to expose it.
While the ADF test checks for non-stationarity, the KPSS test flips the question—its null hypothesis is that the series IS stationary. This guide explains how KPSS works, how to compute and interpret it, and why it's a critical complement to ADF in risk and forecasting models.
The assumption of normality lies behind much of statistical modeling—yet its violation can distort regression estimates, confidence intervals, and hypothesis tests. The Shapiro–Wilk test is a rigorous, powerful method for detecting non-normality, even in small samples.
Stationarity is the cornerstone of valid time series modeling—yet many skip the check. The Dickey-Fuller and Augmented Dickey-Fuller (ADF) tests are the gold standard for detecting unit roots. This practical guide covers both tests, their computation, and interpretation.