Skip to content

Can innovative thinking on credit risk modelling better predict defaults?

Joe Rothermich
Joe Rothermich
Head of StarMine Quantitative Research, Refinitiv

Credit risk modelling is foundational to understanding the probability of default or bankruptcy among public companies. With COVID-19’s continued economic impact, innovative solutions that meet the rising challenges of managing and moderating credit risk are required. Our recent report reviews the performance of Refinitiv’s StarMine credit risk models.


  1. The COVID-19 pandemic has exacerbated the challenges of managing and moderating credit risk.
  2. In such uncertain times, a new approach is needed in credit risk modelling to more accurately predict the probability of default among public companies.
  3. Refinitiv’s StarMine credit risk models have been proven to consistently outperform its benchmark widely used as a measure of default risk.

On 19 May 2021, as part of its coordinated coronavirus credit risk strategy, the European Central Bank (ECB) published a supervision newsletter entitled COVID- 19: Gaps in credit risk management identified.

It stated: “A strong data infrastructure is vital as it underpins a bank’s ability to understand the risks it is facing. Data should be readily available and easily aggregated.”

It also raised concerns about the ability of some banks to aggregate required data, causing weakness in credit risk assessments, and concluded: “It is clear that how banks use data is becoming even more important.”

Download the StarMine Credit Risk Model performance report to discover more

An innovative approach to credit risk modelling

Refinitiv’s StarMine credit risk models address the concerns raised by the ECB and outperform the Altman Z-Score (the benchmark) published in 1968 and widely used as a measure of default risk.

Specifically, the Refinitiv paper notes that over three time periods, the performance of the StarMine models remained essentially unchanged and significantly outperformed the benchmark in predicting the probability of default or bankruptcy within one year for over 47,000 public companies globally.

The StarMine suite includes four credit risk models:

  • StarMine Structural Credit Risk Model (SCR) that evaluates credit risk from the equity market’s view using a proprietary extension of the Merton structural default prediction framework that models a company’s equity as a call option on its assets.
  • The StarMine SmartRatios Credit Risk Model (SRCR) that uses financial ratio analysis for credit risk assessment and incorporates both reported information and forward-looking estimates using StarMine SmartEstimate.
  • The StarMine Text Mining Credit Risk Model (TMCR) that mines the language in textual data from multiple sources (Reuters news, StreetEvents conference call transcripts, corporate filings, and select broker research reports) to evaluate companies’ potential financial distress.
  • StarMine Combined Credit Risk Model (CCR) that combines the power of StarMine SCR, StarMine SRCR and StarMine TMCR to generate a single, final estimate of public company credit risk.

The coverage universe of StarMine’s suite of credit risk models is all public companies globally, including those in the financial sector.

The primary performance metric for all models is the area under the receiver operating characteristic curve (AUC).

Another way to measure a default prediction model’s performance is to look at the fraction of default events captured within a specified ‘danger zone’. In this case, the bottom quintile (20 percent) is the ‘danger zone’.

Case study: performance through the pandemic

The performance of the StarMine suite of credit risk models before and through the turbulent and uncertain times of the pandemic is demonstrated in a case study included in the paper, which was written in mid-2021. The study covered the period of January 2019 to December 2020.

The analysis finds that using StarMine CCR at the end of February 2020, the beginning of the 2020 stock market crash, it was possible to steer clear of over 90 percent of defaults in the following six months. Using the Altman Z-Score, fewer than 70 percent of defaults would have been avoided.

Fraction of F6M defaults captured in the bottom quintile between January 2019 and December 2020

Fraction of F6M defaults captured in the bottom quintile between January 2019 and December 2020. How can credit risk modelling help to predict default?
Fraction of defaults captured (in the following six months) in the bottom quintile of StarMine CCR, versus the benchmark between January 2019 and December 2020. Source: Refinitiv StarMine Combined Credit Risk Model.

A model approach to outperformance

Considering the StarMine Credit Risk models individually, they also provide a positive outcome in terms of outperformance, adding value to the risk management process.

The paper details the performance of each model, breaking down the AUC of each model by region, year and sector across the three time periods.

StarMine CCR is still the best performer. It achieved AUC of 0.91 and captured 86.4 percent of default events in the bottom quintile between January 1998 and June 2020, compared with 0.78 and 63.7 percent respectively for the Altman Z-Score.

StarMine credit risk models are a valuable tool

Without doubt, Refinitiv’s StarMine credit risk models help identify companies most at risk, making them a valuable tool for financial professionals tasked with the assessments of credit and counterparty risk, fixed income security selection and valuation, and equity selection and risk management.

Download the StarMine Credit Risk Model performance report to discover more


Faqs

What's the best way to predict credit defaults?

New analytical models have detected credit defaults more accurately than traditional methods.