HOME > OUR THINKING > Credit Analysis > BLOG

When Enough Is Enough: Assessing Credit Risk Of Companies With Incomplete Financials

Company financials are the backbone of fundamental credit risk analysis performed by risk managers at corporations, and are usually sourced from the income statement, cash flow statement and balance sheet.

This process becomes quickly cumbersome when an analyst needs to look at many public and private counterparties spread across the globe that follow different accounting standards.

At S&P Global Market Intelligence, we offer a large database of company financials, updated on a timely basis, covering both public and private companies globally,[1] standardizing reported values within a unified chart of accounts, and applying hundreds of pre- and post-collection quality assurance checks.[2]

In addition, analysts with access to our Credit Analytics suite can choose among a variety of statistical models to assess credit risk in a simple, quick, and scalable way, leveraging our pre-scored database or using their own financials, perform ‘what-if’ analysis by changing/stressing  input factors, generate reports with a comprehensive summary analysis directly from the S&P Capital IQ Platform or Excel templates, and dynamically link the analysis to PowerPoint via PresCenter to efficiently replicate credit memos or senior management presentations.

Risk analysts are often confronted with incomplete financial information when dealing with private corporations, and thus may not be able to fill all inputs required by a fundamental-based statistical model to output a credit assessment. When this happens, some analysts may approximate missing financial values with industry averages, or forego the analysis altogether.

In instances like this, we recommend a dual approach that takes into account both the quantity and materiality of the exposures:

  • When there is a large number of small exposures, it is reasonable to fill gaps by “looking at the overall picture”, leveraging company credit score benchmarks, such as our CreditModel™ and PD Model Fundamentals score benchmarks (available by country and/or industry), or looking at the Country Risk Scores and Industry Risk Scores, that capture the risk of doing business in a country/industry, and thus give a “broad brush” picture of two of the main drivers of the overall credit risk environment in which companies operate.[3]
  • When there are large exposures, the accuracy of the credit risk estimate becomes as important as the coverage itself. We therefore employ advanced statistical techniques, tailored to the company type, industry and region, to estimate missing financial ratios needed in our fundamentals-based models, and thus generate a credit risk score.[4]

To define the most appropriate statistical technique, we benchmarked different approaches on the same testing dataset, that included a large number of private companies with a complete set of financials and credit score, and from this dataset we randomly “deleted” a fraction of financials or, alternatively, a fixed number of randomly chosen financial ratios, to simulate a “real-life” situation.

Once applied to this simulated case, the selected statistical technique must estimate missing financial ratios that best approximate the actual ones, and thus lead to generation of credit scores that are as close as possible to the actual ones, when the complete set of financials is considered. As it turns out, for private companies that are largely heterogeneous due to differences in size and operating profile, the “nearest neighbor approach” is the best performing. For each company with limited financials, it identifies a subset of companies with complete financials, that best resemble that “focus” company, and from that subset it estimates the missing financial ratios with appropriate regression techniques.[5]

Table 1 below shows the discriminatory power (Receiver Operating Characteristic, ROC)[6] of PD Model Fundamentals – Private Corporates, when actual financials are used, or by using the advanced imputation mechanism to estimate an increasing number of missing financial ratios (randomly selected). Overall, the model starts with a high level of discriminatory power, considering that private companies are pretty hard to model from a credit risk standpoint mainly due to inherent heterogeneity, and the limited blend of quantity/quality of financials available.

More importantly, its performance remains pretty good under our advanced imputation technique, even when seven out of eight (!) model financial ratios get estimated. This is also achieved thanks to the inclusion within PD Model Fundamentals of additional non-financial factors, such as Country Risk Scores and Industry Risk Scores, that do not need to be imputed, and are also important drivers of credit risk.

The attentive reader may have noticed the “accelerated” reduction in model performance when extending the imputation framework from 5 to 7 financial ratios: looking at the right-most column in Table 1, a decrease of 516 basis points (bps) is visible, compared to 139 bps when moving from three to five variables. Why do we have this massive jump?

In a previous blog, we identified three main drivers of credit risk in PD Model Fundamentals Private: Total Revenues, Net Income/Total Revenue, and Short-Term Liabilities/Net Worth. In PD Model Fundamentals, the imputation mechanism requires at least knowledge of Total Revenue. It is easy to show that the marked fall in model discriminatory power from imputing five to imputing seven variables is due to the case when Net Income/Total Revenue and Short-Term Liabilities/Net Worth.

How can we be sure of this, if in our simulations we remove financial ratios at random? Simple: when we impute five out of eight financial ratios at random, the chance that we always impute Net Income/Total Revenue and Short-term Liabilities/Net Worth is smaller than when we impute seven out of eight financial inputs – when those ratios are imputed all the time. So, we can conclude that these are the culprits of the big fall in model performance!

This is also consistent with tests we performed during model development, where Net Income/Revenue and Short-Term Liabilities/Net Worth appeared as the most predictive variables when assessing the predictive/discriminatory power of each variables used in the model, on a standalone basis and in combination. Despite this, even when these important variables are missing, the technique that we have implemented for imputation allows the model to generate outputs with good discriminatory power.

In summary, our advanced imputation technique empowers analysts during their credit risk analysis of large exposures toward multiple private counterparties by bridging the gap of limited company financials via an automated, scalable, yet accurate, approach. There is more: all of S&P Global Market Intelligence’s fundamentals-based credit risk models come equipped with tailored imputation frameworks, to cover both private and public corporations!

Table 1- Model Receiver Operating Characteristic-ROC

Learn more about S&P Global Market Intelligence’s Credit Analytics models.

[1] S&P Global Market Intelligence collects financials for more than 42,000 public companies and 1,500,000 private companies across the globe (source: Capital IQ Platform, as of April 21, 2017).

[2] S&P Global Market Intelligence will award a $50 gift card for each error found in the public company data in the division’s latest desktop and enterprise products (subject to terms and conditions).  A similar program already covers public and private company data in the legacy SNL platform.

[3] See an example in S&P Global Market Intelligence’s blog: (July 4, 2016).

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

[5] Further details for each family of statistical models can be found in S&P Global Market Intelligence’s whitepaper: “Bridging the Gap of Missing Company Financials to Estimate Credit Risk” (August 2016).

[6] This is a measure of discriminatory power, i.e. the ability of the model to estimate high probability of default (low credit score) to companies that will default, and low probability of default (high credit score) to non-defaulters. Typically, for a model covering private companies, values of ROC between 70% and 80% are considered good sign of discriminatory power, between 80% and 90% are considered very good, and above 90% are considered excellent.

Subscribe to Insights
Jan 18, 2017
Credit Markets
Oct 07, 2016