Max-factor individual risk models with application to credit portfolios

Michel Denuit, Anna Kiriliouk, Johan Segers

TL;DR

This paper introduces a "max-factor" approach to modeling credit risk, where the probability of default is driven by the maximum of several latent risk factors rather than their sum. By replacing linear combinations with maxima, the model better captures extreme "shocks" and prevents the unrealistic dilution of risk that occurs in traditional models, leading to more accurate estimations of tail risk in credit portfolios.

What Problem Does This Paper Address?

In credit risk management, understanding the dependence between individual defaults is crucial. If you assume defaults are independent, you will drastically underestimate the risk of a "black swan" event where many firms fail simultaneously.

Standard industry models (like CreditRisk+ or KMV) typically use latent factor models. In these, the probability of a firm defaulting is a function of a linear combination of hidden factors (e.g., a global economic factor + a sector-specific factor). However, the authors argue that linear combinations allow for "diversification" or compensation: a very bad realization of one factor can be offset by a good realization of another.

In reality, a single catastrophic shock to a specific sector can trigger a wave of defaults regardless of how well the global economy is doing. The authors propose a model where the "worst" factor takes the lead, providing a more realistic representation of how shocks propagate through a portfolio.

Methods

The core innovation is the Max-Factor Risk Model.

  1. The Max-Specification: Instead of summing factors, the conditional loss probability is expressed as an increasing function of: $$\max\{\nu_r + \sigma_r\Psi_r, \mu_r + \sigma_r\Psi_0\}$$ Where $\Psi_0$ is a global risk factor and $\Psi_r$ is a category-specific factor (like an industry sector).

  2. Gumbel Distribution: The authors use the Gumbel distribution for the latent factors. This is a strategic choice because the Gumbel distribution is "max-stable"—the maximum of Gumbel variables is itself Gumbel-distributed, which keeps the math tractable.

  3. Likelihood Optimization: Calculating the Maximum Likelihood Estimate (MLE) for max-functions is computationally difficult because the functions are non-differentiable. The authors provide a novel numerical procedure (Lemma 3.1) to simplify high-dimensional integrals into a sum of lower-dimensional, smoother integrals.
  4. Weighted Nonparametric Estimators: To provide a fair benchmark, the authors developed new nonparametric estimators that assign more weight to years with more data, providing a more robust "ground truth" for comparing model performance.

Datasets

The model was evaluated using the Standard & Poor’s (2001) credit risk dataset.

  • Scope: One-year default rates for static pools of obligors from 1982 to 2000.
  • Focus: The "speculative grade" categories: BB, B, and CCC.
  • Composition: The data tracks approximately 9,200 obligors, accounting for 746 defaults over the period.

Key Insights and Ideas

  • The "Anti-Diversification" Principle: Linear models tend to "dilute" shocks. By using a maximum, the model recognizes that if a sector-specific shock is severe enough, it dominates the risk profile, regardless of other stabilizing factors.
  • Regime Switching: The max-factor model acts as a natural regime-switching mechanism. It can transition from a "standard" environment (driven by the global factor) to a "severe" environment (driven by a specific category shock) instantaneously.
  • Fat Tails Matter: The authors show that Gumbel-based models capture the "fat tails" of default probabilities much better than Gaussian (Normal) models. Normal models tend to underestimate the probability of very high default rates, which is exactly what risk managers need to avoid.

Key Results

  • Superior Fit: The Gumbel max-factor model (Model 2b) outperformed the standard Probit-Normal and Gumbel one-factor models, achieving the lowest AIC (320.28) and BIC (314.02).
  • Tail Risk Accuracy: Visual analysis of "excess probabilities" showed that the max-factor model aligns much more closely with historical data in the high-quantile regions (the right tail) compared to traditional models.
  • Robustness: The weighted nonparametric estimators confirmed that the max-factor model's implied default probabilities were consistent with empirical observations, even in volatile rating classes like CCC.

Limitations and Future Work

  • Hierarchical Complexity: The current study focuses on a two-level hierarchy (global vs. category). The authors suggest that extending this to more complex, multi-layered hierarchical structures would be a valuable next step.
  • Credibility Modeling: The authors note that the max-factor decomposition could be highly relevant for "credibility models" in insurance, where the worst-case risk proneness drives the individual's premium.
  • Computational Scaling: While Lemma 3.1 helps, the numerical integration required for MLE still becomes more taxing as the number of risk categories ($k$) increases. Further optimization for very large-scale portfolios may be needed.

Summary updated

Abstract

Individual risk models need to capture possible correlations as failing to do so typically results in an underestimation of extreme quantiles of the aggregate loss. Such dependence modelling is particularly important for managing credit risk, for instance, where joint defaults are a major cause of concern. Often, the dependence between the individual loss occurrence indicators is driven by a small number of unobservable factors. Conditional loss probabilities are then expressed as monotone functions of linear combinations of these hidden factors. However, combining the factors in a linear way allows for some compensation between them. Such diversification effects are not always desirable and this is why the present work proposes a new model replacing linear combinations with maxima. These max-factor models give more insight into which of the factors is dominant.
BibTeX
@inproceedings{denuit2015max,
  author = {Michel Denuit and Anna Kiriliouk and Johan Segers},
  title = {{Max-factor individual risk models with application to credit portfolios}},
  booktitle = {Insurance},
  year = {2015},
  doi = {10.1016/j.insmatheco.2015.03.006},
}

Loading related content...

Figure