Existing regulatory approaches for determining the capital backing of credit risks have a number of deficits: too unspecific for the credit institution, too complex, the procedure too unintelligible. It is time for a paradigm shift in credit risk management. Digitization and artificial intelligence are finding their way into credit risk management.
Current methods of quantifying credit risks do not always lead to satisfactory results. Regulatory approaches generally do not take a bank’s specific circumstances into account. Economically motivated models can either only be used to a very limited extent on retail portfolios or are so complex that the results cannot be reconstructed or made plausible.
While there is little to shake off the regulatory requirements, the question arises as to why existing problems are usually accepted without complaint from an economic perspective. With the help of artificial intelligence, the digital transformation could usher in a new age in credit risk management.
The model for managing operational risks
While credit risk management tries to calculate the required capital backing with more or less complex formulas and the information available at the account and customer level, institutions take a different approach when it comes to managing operational risks: on the one hand, they collect past claims in a loss database, on the other hand, they carry out a risk inventory for possible losses.
Subsequently, identified risks are evaluated and the options for controlling these risks are examined.
The risk inventory process, in particular, is significantly more complex than a purely analytical procedure, but the advantage lies in the fact that expert knowledge of the risk situation was used to determine and quantify the risks. This means that the acceptance of the results is considerably higher than when using an externally specified formula.
Development and use of a loss database in credit risk management
Building a loss database for loan portfolios shouldn’t be a problem for a bank. The amount of the losses is available to Controlling as additions to the individual value adjustments or as an impairment posting. Banks that perform a detailed counterparty risk result are even better.
In addition to the mere calculation of actual risk costs, it can also be used to differentiate according to creditworthiness, collateral and term effects.
For example, the VR Control application of Good Finance provides good results. Based on the changes in the value of the premium portfolio, it is easy to analyze in which periods the actual losses were greater than the expected losses covered by calculated risk premiums, even without artificial intelligence.
Limits of existing loan portfolio models
It is interesting to ask which data explain these deviations in particular upwards. And this is where the hour of artificial intelligence comes. In classic credit portfolio models, correlations observed for an economy are generally used to calculate a credit value at risk. When using CreditRisk +, the correlation of defaults in various industries is used as input.
These can be determined and easily calculated nationwide with the help of the Federal Statistical Office’s downtime series. If there are so-called cluster risks because the exposure shares of one or more sectors are dominant within a credit portfolio, this increases the value of the credit value at risk.
It becomes problematic for example with regionally operating banks. For an institute operating in the Wolfsburg area, spreading credit exposure across various sectors (retail, construction, vehicle parts) will have relatively little effect on the level of credit value at risk, since in this case loan defaults are much more closely related to the automotive industry than this nationwide is the case. And if a bank’s loan portfolio is made up primarily of private customers, there is generally no valid information available on the correlation of default probabilities.