Credit risk

Automate credit risk monitoring using statistical models

This article is written and published by S&P Global Market Intelligence, an independent division of S&P Global Ratings. The lowercase nomenclature is used to differentiate S&P Global Market Intelligence credit ratings from credit ratings issued by S&P Global Ratings.


S&P Global Ratings Issuer Credit Ratings (ICRs) assess a company’s willingness or ability to repay debt on time and in full. They are based on criteria that include both quantitative criteria[1] and qualitative considerations,[2] and are subject to committee review. S&P Global Ratings provides additional directional information (e.g. a credit outlook and credit monitoring) on ​​the potential future change in certain ICRs,[3] But any credit rating can be updated at any time, with critical implications for asset managers and credit risk managers:

  • Asset managers: The investment universe of a fund is often delimited by the indication of a minimum credit rating threshold. This allows investors to better understand the risk tolerance that a fund is prepared to assume. When building or rebalancing a portfolio, it is therefore important for asset managers to follow the overall credit risk profile of their portfolio in order to be able to stay in line with investors’ expectations, or capture possible arbitration opportunities.
  • Credit risk managers: transitions to or from the investment category can have a significant impact on the risk profile of companies included in a credit risk portfolio,[4] as well as in the calculation of the expected credit loss according to the International Financial Reporting Standard (IFRS) and the Current Expected Credit Loss (CECL) standard.

In this article, we show how a statistical model developed by S&P Global Market Intelligence can be used to generate early warning signals of a potential deterioration / improvement in credit risk that may trigger a rebalancing of the portfolio.

CreditModel â„¢ Business 3.0

At S&P Global Market Intelligence, we have developed a quantitative model that is trained on Autonomous Credit Profiles (SACP) of S&P Global Ratings companies:[5] CreditModel â„¢ Business 3.0 (CM3.0). CM3.0 generates a credit score[6] which aims to statistically match the SACP for rated companies, and can also be used for unrated companies above a certain income threshold. CM3.0 credit ratings are expressed on the same scale as S&P Global Ratings, but in lower case to distinguish them from actual S&P Global Ratings credit ratings.

Another overlay adjusts the CM3.0 score by including considerations of parental and government support, where applicable, allowing the model outputs to statistically match S&P Global Ratings ICRs 28% of the time, as shown in Table 1 for rated non-financial companies domiciled in North America.[7] As expected for a statistical model, the scoring agreement is not perfect (i.e. 100% of the time), because KPIs include qualitative aspects that cannot be exactly quantified. Either way, CM3.0 maintains excellent performance at one, two and three notches.

Table 1: Performance of CM3.0 North America, adjusted with parental and government support.

Source: S&P Global Market Intelligence as of December 31, 2018. For information only.

Interestingly, we notice a fraction of outliers whose adjusted CM3.0 score is three or more notches different from the actual ICR (around 12%). This group seems to defeat any technical attempt to improve the performance of the model. Two questions arise naturally:

  1. What if there was something else behind the existence of these outliers, aside from the limitations inherent in a quantitative model?
  2. What if these outliers could be used to play in a user’s favor by generating early warning signals of a potential change in credit risk?

Our discoveries

Figure 1 illustrates the historical frequency of changes in the ICR for companies where the adjusted CM3.0 score deviates by x notches from the ICR of S&P Global Ratings at time t. For example, the bar corresponding to +1 refers to companies whose CM3.0 score is one notch higher than the ICR score at time t, and which have been upgraded (green), downgraded (red) or remained unchanged. within one year. (left panel) or in a five-year time horizon (right panel) from time t. The dataset includes companies domiciled in the United States and Canada, rated between 2010 and 2021.

Figure 1: Percentage of downgrades, upgrades, no change.

Source: S&P Global Market Intelligence as of October 1, 2021. For information only

As the notch difference widens, the historical probabilities of an ICR change within the specified time horizon increase. As shown in Figure 1, outliers (companies with a difference of three notches and more) are particularly suitable for monitoring purposes, because the statistical probabilities of an upgrade (notch difference of + three and more) or a derating (notch difference of -three or less) are close to 80%.

Table 2 shows the percentage of ratings with an Outlook / CreditWatch before they changed, for a positive or negative notch difference in Figure 1.

Table 2: Outlook / CreditWatch for ratings that progressed to the CM3.0 score.

Source: S&P Global Market Intelligence as of October 1, 2021. For information only.

About half of all ratings that were upgraded (downgraded) within one year of recording a positive (negative) notch difference initially had an Outlook / CreditWatch from S&P Global Ratings.[8] Equally interesting, about 2% (7%) of ratings with a positive (negative) initial opinion actually got downgraded (upscaled).

Thus, statistical outliers could be used to automate the generation of directional signals of potential improvement or deterioration in creditworthiness, beyond existing credit monitoring and outlook.

These results also remain valid for a five-year time horizon, despite the volatility associated with a longer time horizon and the smaller number of observations that smooth the effect.


The stability or change in a credit rating is important for risk management and for investment purposes. In this paper, we show how a quantitative model designed to statistically match S&P Global Ratings ICRs can be used to automate risk monitoring or identify potential investment opportunities by appropriately exploiting the natural existence of mathematical outliers. generated by the model.

Our empirical analysis on historical data suggests that a significant divergence between a modeled score and the actual ICR is accompanied by a higher likelihood of a rating change, beyond the Outlook or CreditWatch indicator sometimes provided by S&P. Global Ratings. As is evident, these signals are statistical in nature and cannot predict rating movements deterministically.

In a follow-up analysis, we will explore how / if the same approach can be used for selecting equity portfolios and for finding excess returns relative to a certain benchmark, since rating moves are known for their impact. on stock prices.

About S&P Global Market Intelligence

At S&P Global Market Intelligence, we understand the importance of accurate, in-depth and relevant information. We integrate financial and industry data, research and news into tools that help track performance, generate alpha, identify investment ideas, perform valuations and assess credit risk. Investment professionals, government agencies, businesses and universities around the world use this essential intelligence to make business and financial decisions with conviction.

S&P Global Market Intelligence is a division of S&P Global (NYSE: SPGI), the world’s leading provider of credit ratings, benchmarks and analysis in global capital and commodities markets, offering ESG solutions, in-depth data and insight into critical business factors. S&P Global has been delivering critical intelligence for more than 160 years that unlocks opportunity, drives growth and accelerates progress. For more information, visit

[1] For example, a company’s financial ratios, macroeconomic scenario projections, etc.

[2] For example, country risk, sector risk, competitiveness, benchmarking with peers, quality of management, etc.

[3] Typically, there is a one in three chance that their rating will go down / up in the next six to 24 months, for companies with a negative / positive outlook. There is a one in two chance of a downward / upward revision for companies with negative / positive credit watch over the next three months. See, for example, “Guide to Credit Rating Essentials”, S&P Global Ratings (2019), Available here.

[4] The investment category includes any rating above BB +.

[5] A corporate SACP refers to an issuer’s credit rating before any consideration of parental or government support.

[6] S&P Global Ratings does not contribute or participate in the creation of the credit ratings generated by S&P Global Market Intelligence. The lowercase nomenclature is used to differentiate S&P Global Market Intelligence PD credit model scores from credit ratings issued by S&P Global Ratings.

[7] Evaluated on a training sample containing 2,801 unique companies, rated between 2003 and 2017.

[8] Initially, that is, the date on which the difference between the statistical score and the ICR was recorded.