Bank loans to corporations and private individuals are fundamental to a working economy. The credit risk of loans, the risk that borrowers default on their repayment obligation, is typically the largest risk category of a commercial bank. Regulators require banks to hold capital against the risk of unexpected losses. Bank failure due to individual problem borrowers are rare, nevertheless most banks fail due to bad lending decisions often due to misjudging and mispricing credit risk. As bank failures can send shock waves through the economy and society, the understanding and pricing of credit risk should be of foremost interest to credit investors, banks, regulators, central banks and governments.

Despite the importance of credit risk, investors have few tools available for quantifying credit risk, a hole that OSIS aims to fill. Long-term loan performance history is rare, even within banks, and much is held privately. Even when available, the quality and relevance of data from the distant past is questionable. Credit rating agencies could help investors understand credit risk, but arguably fail to do so. However, there are equally important but lesser known issues with traditional credit ratings that OSIS aims to address.

In the US, credit rating agencies are known as nationally recognized statistical rating organizations ‘NRSRO’ because ratings are based on statistical analysis. At the same time agencies emphasize that ratings are expert opinions and it is often unclear to what extent a rating was issued based on data or opinions. Even if the process was largely data driven, the information published to investors does not allow for independent validation and replication. This observation is important for a number of reasons. First, regulated investors are increasingly obliged to conduct their own risk assessment, but are forced to rely on ratings if those ratings reflect superior non-public information. Second, while good historical data is scarce, more and more data is becoming available to investors through better bank disclosure, central bank stress tests or transparency efforts (like the ECB sponsored loan level data initiative for ABS). A replicable, coherent way to capture such information seems desirable. Third, ratings are long-term forward-looking statements subject to significant uncertainty. More data reduces uncertainty and helps identify problems with rating models and their underlying assumptions. At OSIS™, we have devised rating robots that learn from data in a replicable, transparent way.

While data analysis is key, credit risk analysis depends on subjective choices and expert opinion. Whereas traditional rating processes largely keep the data analysis and prior assumptions hidden from investors and publish only the resulting risk assessment, OSIS™ advocates to transparently analyze the data and allow investors to use their own subjective opinions for the prior.

In summary, OSIS™ was set up in 2010 by 2 former bankers to offer an alternative way for banks and investors to analyze credit risk, by coherently combining a statistical analysis of data with subjective opinions about forward looking risk distributions. Such analysis captures parameter uncertainty and allows for designing more coherent stress testing scenarios.