[Paper Review] Capturing Model Risk and Rating Momentum in the Estimation of Probabilities of Default and Credit Rating Migrations

Marius Pfeuffer, Gonçalo dos Reis+1 more|arXiv (Cornell University)|Sep 26, 2018

Credit Risk and Financial RegulationsEconomics, Econometrics and Finance36 references3 citations

TL;DR

This paper proposes two methodologies for estimating credit rating transition probabilities: a computationally efficient method for continuous-time Markov chains using discrete data, and a self-exciting marked point process model that captures non-Markovian rating momentum. The key contribution is that the non-Markov model better aligns with empirical default probabilities—predicting higher defaults in investment-grade firms and lower defaults in speculative grades—improving realism and reducing model risk in credit risk modeling.

ABSTRACT

We present two methodologies on the estimation of rating transition probabilities within Markov and non-Markov frameworks. We first estimate a continuous-time Markov chain using discrete (missing) data and derive a simpler expression for the Fisher information matrix, reducing the computational time needed for the Wald confidence interval by a factor of a half. We provide an efficient procedure for transferring such uncertainties from the generator matrix of the Markov chain to the corresponding rating migration probabilities and, crucially, default probabilities. For our second contribution, we assume access to the full (continuous) data set and propose a tractable and parsimonious self-exciting marked point processes model able to capture the non-Markovian effect of rating momentum. Compared to the Markov model, the non-Markov model yields higher probabilities of default in the investment grades, but also lower default probabilities in some speculative grades. Both findings agree with empirical observations and have clear practical implications. We illustrate all methods using data from Moody's proprietary corporate credit ratings data set. Implementations are available in the R package ctmcd.

Motivation & Objective

• To reduce computational time in estimating confidence intervals for continuous-time Markov chains from discrete, missing data.
• To transfer estimation uncertainty from the generator matrix to transition and default probabilities using the Delta method.
• To model non-Markovian rating momentum using a parsimonious self-exciting marked point process on full continuous-time data.
• To improve the accuracy of default probability estimates by capturing empirical patterns overlooked by standard Markov models.
• To provide a more realistic assessment of model risk in credit risk modeling, especially under IFRS 9 and Basel regulatory frameworks.

Proposed method

• Derives a simplified closed-form expression for the Fisher information matrix in discrete-time Markov chain estimation, cutting computational time for Wald confidence intervals by half.
• Applies the Delta method to propagate uncertainty from the generator matrix to transition probability matrices and default probabilities.
• Proposes a self-exciting marked point process model that captures rating momentum effects, where past downgrades increase future downgrade intensity.
• Uses full continuous-time rating transition data to estimate model parameters, avoiding the Markov assumption and enabling non-Markovian dynamics.
• Employs maximum likelihood estimation with numerical optimization for parameter inference in both Markov and non-Markov models.
• Implements parts of the methodology in the R package ctmcd for reproducibility and practical use.

Experimental results

Research questions

• RQ1 How can confidence intervals for transition probabilities in continuous-time Markov chains be computed more efficiently from discrete, missing data?
• RQ2 To what extent does uncertainty in the generator matrix propagate to default probability estimates, and how can it be quantified?
• RQ3 Can a non-Markovian model better capture empirical default patterns than a standard Markov model, particularly in investment-grade and speculative-grade ratings?
• RQ4 How does rating momentum—driven by past downgrades—affect the accuracy of default probability forecasts?
• RQ5 What are the practical implications of model risk in credit risk modeling, especially under IFRS 9 and Basel regulations?

Key findings

• The proposed closed-form Fisher information matrix reduces computation time for Wald confidence intervals by over 50% compared to existing numerical methods.
• The Delta method enables direct, interpretable uncertainty quantification for default probabilities derived from generator matrix estimates.
• The non-Markov model predicts higher one-year default probabilities for investment-grade firms (e.g., Baa), aligning better with empirical observations than the Markov model.
• For speculative grades like Ca and Caa, the non-Markov model yields lower default probabilities than the Markov model, bringing estimates closer to empirical data.
• The non-Markov model reduces the overestimation of risk in firms initially rated Ca or Caa, which are artificially penalized in the Markov framework due to data contamination from prior downgrades.
• The findings suggest that ignoring rating momentum leads to significant model risk, particularly in capital requirement calculations under IFRS 9 and Basel III.