As the global economy is fighting the next downturn, the new test for credit risk management is quickly rising. Lenders are asking how recent events will affect customer creditworthiness and whether their credit risk management practices will help them withstand potentially large losses on the portfolio as a whole.  Whilst the long-term outcome of the current market shock is still unknown, if we were to look back at the recent financial crisis, we can draw some interesting conclusions that may help us prepare for any future recession. As put by writer and philosopher George Santayana: "Those who cannot remember the past are condemned to repeat it.

What happens to customer creditworthiness during recession?

Simply speaking, the overall risk of the portfolio increases as it becomes more difficult for customers to meet their loan commitments due to financial difficulties. As the payment behavior of the borrowers’ change, this results in credit score deterioration across the portfolio. To illustrate the behavior of credit scores through the recent 2008 financial crisis, we simulated a portfolio of c.10 million unsecured retail loans to small businesses with an average credit score of 600. As illustrated in the diagram below, during benign market conditions we typically observe close to a normal shaped distribution of credit scores depending on the lender’s risk appetite. During a simulated market downturn, there is a deterioration in credit quality across the portfolio and, as a result, the mean of the credit score distribution shifts to the left resulting in a distribution with heavy tails. Once the market returns to growth, interestingly the distribution does not return to its initial shape. The new distribution contains a proportion of customers moving even further to the left (so-called “unrecoverables” – highlighted by the red circle in figure 1) who are no longer eligible for traditional credit services.

What determines the shape of the distribution and how can lenders manage the movements in the portfolio? 

The lender’s risk appetite, i.e. how much risk it is willing to accept on the book, drives the shape of the distribution. However, reducing the size of the unrecoverable population is not as simple setting a higher cut off point (i.e. reducing the number of risky customers in the book). To understand this better, we need to look at the underlying dynamics of the different customer segments.

Each segment suffers a different impact during the recession and more importantly experiences a different recovery as the economy returns to growth following the recession, Figure 2 below illustrates this. For example, Segments 1 and 2 both contain small businesses who are more susceptible to a recession when trade declines, this may be due to a dependency on short-term cash flows. Segment 1 is severely impacted by the recession and does not recover when growth returns to the market (this segment migrates to the far left of the score scale and remains there – highlighted by the red circle), whereas Segment 2 returns to the same creditworthiness following the recession. In contrast, Segment 6 may contain businesses that are less susceptible to a recession (this cluster remain stable at the right hand side of the score scale).

It is easy to define population clusters using traditional techniques once the historical data is available. However, at the start of the downturn lenders need to make critical decisions to help protect their customers. Therefore, portfolio managers must act quickly as a comprehensive historical data set will only be available when the market has recovered.

To prevent significant deterioration of the customer book, appropriate action needs to be taken at the start of the downturn to support customers experiencing temporary difficulties. An example might be businesses forced to close their stores for a short period, however the temporary closure does not reduce the value of their goods. Payment holidays in such example might be sufficient to prevent long-term effect on the business credit profile. International governments encourage certain forms of support offering liquidity, loan guarantees or actual one-off cash payments for selected businesses.

As a lender’s ability to provide support is often limited, prioritization of loans in the portfolio becomes necessary and therefore it is important to understand which customers are most affected by events in the market and what their recovery curve is. One approach to identify at risk customers is by identified industries or segments most at risk through industry-level credit cycle indices.

It is also important to note that market downturns are generally predicated on some form of shock event and therefore historical data is of limited use. Decisions therefore need to be made on all available data as it becomes available. External data sources therefore becomes valuable and machine learning (ML) techniques become powerful tools to extract useful information from small datasets.

Using machine learning to identify susceptible customer segments

Deloitte developed a hybrid modelling approach, which combines traditional regression with ML techniques, allowing efficient identification of the subpopulations or segments. Our Zen Risk approach then translates the derived segment definitions from ML into a set of transparent business rules. The business rules are implemented alongside the existing portfolio origination and portfolio management strategies. This helps lenders identify specific portfolio clusters where action is needed, without major model redevelopment and policy changes. Our approach provides lenders with a deep understanding of their customer portfolio and allows quick reaction to sudden portfolio movements.

Additional Reading:

Blog: Helping French FinTech SME lender make better credit decisions with Deloitte’s Zen Risk

Blog: Smarter Better Faster: The next generation of risk modelling with Machine Learning