Loan-level vs. customer-level credit risk measurement

Let’s consider a simple example, typical for many consumer loans portfolios:

  • 10.5% of loans granted to new customers are “bad” (90+ days past due) after one year from origination (black line in the plot below).
  • 7.1% of loans granted to existing customers are “bad” one year past origination (green line).

Credit risk measurement - vintage default rates agreggated by loans

So what percent of customers are in “bad” status one year after taking the first loan?

  • Somewhere between 10.5% and 7.1%? No chance.
  • Exactly 10.5%? Rarely the case.

It is 14.3% (blue line in the plot below). At least in case of the dataset used for this article. The difference can be even greater in some portfolios. Of course 14.3% default rate after one year is already very high risk. The difference between 10.5% (loan-level) and 14.3% (customer-level) is very material. It can easily be a difference between profitable and unprofitable lending.

Credit risk measurement - vintage default rates

As illustrated above, there are three important risk ratios measured at 12 months on books:

  • 10.5% bad rate on new loans,
  • 7.1% bad rate on loans granted to existing customers,
  • 14,3% bad rate at customer-level,

The third number, customer-level, is probably the best one to characterize overall portfolio performance. It can be interpreted as the share of customers, whose risk status is “bad” one year after they received the first loan. The “bad” status is bad and has the same consequences for the lender as well as for the customer regardless if it is on the first loan or on another loan that was taken during this first year in the portfolio.

Calculations are based on actual portfolio of a European lender, one of our Clients. Their management board kindly expressed explicit consent in the consultancy contract to use their depersonalized data for this purpose.


Why and when customer-level bad rate is higher and more meaningful than loan-level

Of course, there would be no difference if one customer always had only one loan. No new loans would be granted to a customer up to one year from origination, even if the first loan is fully repaid. In reality, this is the case mainly in banks specialized in consumer credit cards. More often it is actually a strategy of consumer finance players to offer a new loan as soon as the first loan is repaid, either on schedule or by an early repayment of the whole balance. It can be a successful strategy but it should not be excersised without proper risk measurement system on customer-level.

The strategy, which makes it perfectly possible for a customer to have two, three or even more loans in a sequence during one year, is currently very common. Traditionally the domain of non-banking lenders, it is also in place in some universal banks. Moreover, it is often combined with the possibility of one customer having multiple loans with one lender in parallel and a high-risk “product” called “top-up”, “internal consolidation”, “extension”. This procedure, known under different names in different credit pollicies, allows repayment of an existing loan with a new one. When this kind of “product” is available, the need for customer-level risk measurement becomes very urgent. Loan-level measurement is not only insufficient in this case. It is also affected by the fact that some loans are artificially closed (by new loans) and marked as “good” while the customer can quickly appear to be “bad” just demonstrating non-repayment under a new loan number.


The importance of customer-level in credit risk measurement

Surprisingly, retail credit risk is still too often measured mainly with account-level ratios. Customer-level aggregation of bad rates over time should be used alongside loan-level in:

  • stress-testing for adverse economic scenarios (e.g. in case of limited funding there will be fewer new loans to existing customers, consequently vintage curves at loan-level will deteroriate, likely at least to the current shape of customer-level),
  • evaluating portfolio quality to make a decision about an investment, understanding overall health of the portfolio or quality and business perspective of a customer profile,
  • making decisions about credit policy changes – especially related to: product tenor, ticket size, limit strategy, so called top-ups and loan increases (top-ups are very risky if not managed with extreme care from risk perspective)
  • selecting dependent variables for risk scoring models (a new customer who repaid the first loan on time but quickly took another loan and then defaulted should not be classified as “good”, we don’t want the model to “learn” that this is the kind of customer to approve or even get attactive risk-based pricing).


Comparison of multiple close-ends vs. credit cards

The situation when new customers often repay their loans rather quickly and then take new loans from the same lender, often with greater amounts, is similar to credit cards with limit increases. In case of credit cards, customer risk status is tracked over time under one loan account number. This is why credit risk measurement system for card portfolios is considered to be slightly more complex than the one for close-ends. The truth is, in case of close-ends, these complexities still exist but they are hidden when only loan-level aggregation is used.



Looking only (or mainly) at account-level (loan-level) risk performance is not enough. Customer-level aggregation is essential. Many risk managers look at dozens of loan-level vintage curves every month while customer-level vintages are still missing. Sometimes a proxy of customer-level vintage is drawn based on some management assumptions and loan-level results, without proper data aggregated at customer-level. It is a risky idea, which can lead to incorrect conclusions. Constructing a customer-level vintage credit risk MIS is usually relatively simple and strongly recommended. It is very important for every lender with short term loans and extremely important for lenders with very short term loans or allowing loan increases (top-ups). It would also be very useful and informative to analyse customer-level vintage performance for the whole sector, as they are reported by the Credit Bureau (currently – aggregated at loan-level).