Financial Covid-19 Measures: the impact on default modelling

November 18, 2021 General

Mees blog

In 2020, governments around the world introduced several supporting measures for enterprises and consumers in order to overcome the consequences of the coronacrisis. Also banks, which are usually in the spotlight during crises, were now part of the solution. Besides government guaranteed credits, they provided the possibility to postpone repayments for a certain period of time, known as payment holidays. According to the Corona Monitor by the Nederlandse Vereniging voor Banken (NVB), 129.000 commercial clients, 26.000 mortgage holders and 12.000 consumer credit holders received a payment holiday. Those numbers are indirectly related as commercials defaults fuel unemployment, which leads to mortgage arrears and defaults. The payment holidays were meant to provide financial stability for credit holders and overcome their financial distress. Most of these payment holidays ended in the last quarter of 2020, with a possible increase of payment arrears or even defaults; the debt needs to be repaid in the end. But did this actually happen? This blog first discusses the consequences in modelling caused by the payment behavior data, draws up the balance and finally extrapolates future implications.

The missing information about the number of customers in arrears results in Probability of Default (PD) models that are not functioning properly anymore, which is especially true for point-in-time (PIT) models. These models typically rely on variables like the client’s payment behavior and macroeconomic factors, both of which are seriously distorted during the crisis. More on PD models in this blog: A solution to the problem of missing arrear information could be to simulate this data using historical information of previous crises. This method then provides proxies for missing risk drivers in the PD-model, which could to a better estimation than when using no arrear information. This solution is applicable for similar situations that may occur in the future, when payment holidays are provided by the bank and the underlying payment status will be unknown. In the blue box below, this method is explained.

Currently, we are in Q3 of 2021 and we can take a look to the bigger picture and draw up the balance. The number of households benefited from the payment holiday is low, with a total postponed amount of 88 million, which is marginal compared to the total outstanding debt. Obviously, this is fueled by the relative low number of commercial defaults and related low unemployment rate. The financial supporting measures by the government (especially the NOW) and the payment holidays have done their work, although it should be mentioned that the most financially distressed enterprises are still able to use the latest version of NOW. Therefore, it could be that after the termination of the last NOW, a spike of commercial credits defaults will appear, but since the economy is booming and the demand for employees is already at its pre-pandemic level, we do not expect that the default rate for mortgages will heavenly increase.

It even could be argued that the payment holidays were not necessarily required for the mortgage and consumer finance market, as the support for enterprises seems to be sufficient. The financial measures and the payment holidays even led to a default rate lower than before the crisis. This has resulted in an unprecedented relationship between the default rates and the GDP growth, for instance Q4 2020, where the GDP has fallen with 2,9 % (CBS) and the number defaults decreased with approximately 30 % (CBS). We could state that a lot of defaults that would have happened without financial support measures are missing in the historical observations. To summarize, the decline in GDP was counteracted by the government support measures and the payment holidays.

But what does this mean for modelling credit risk in the feature? Macro-economic risk drivers like GDP growth and unemployment rate in combination with low default rates could lead to modelling difficulties. Historically, the correlations between these factors and the default rates in different years show similar values, but the observed values in 2020 disturbs this trend and can even show inverted correlations. These could lead to problems when estimating models in the future. To incorporate this fact in modelling in the future, there are several solutions. A simple solution is to leave the observations out of the period used to estimate the correlation between macro-economic factors and default rates. A more sophisticated approach could be to create a Covid-19 dummy variable, to account for the extraordinary situation including the financial support measures.

We are aware that providing a general solution to this problem is almost impossible, with extremely adverse outcomes per company sector and unpredicted situations that appear again and again. Each portfolio requires a different approach and a tailor-made solution. We at RiskQuest use our outstanding credit risk knowledge to provide the best solution in order to help our clients in this interesting period.