This article continues our occasional series of investigations into what the future might hold, for lifetime expected credit losses (LECL) in banks' loan portfolios. In a previous blog articles, we showed that:
- Lifetime credit losses at the end of March 2020 were likely to be a significant multiple of their year-end 2019 values; and
- The precise quantum is highly sensitive to assumptions about how the future might unfold.
In light of banks' Q1 2020 disclosures, we decided to revisit our assumptions and investigate the impact of adding a credit cycle index (CCI) data point for April month-end.
As of the end of April, the UK CCI has turned back towards normal conditions. Without downward momentum, our forecasting model predicts reversion to normal conditions, without the initial downward movement that was present in our March base case forecast. Across the base case, upside and downside forecasts, the quantum of CCI stress generally improves by +1. Compare the chart below with the range that we were forecasting a month ago:
Where the March base case forecast took 30 months to achieve Z=-1 , the April base case forecast achieves Z=-1 in 18 months. And where the March base case forecast had a distinct "U" or "J" shape, the April forecast has a pronounced "V" shape. This is considerably more-closely aligned with the UK Office of Budget Responsibility (OBR) reference scenario's 12 month return to normal output levels, and may even align if we believe that credit risk lags GDP - in some sectors, at least.
The impact on LECL estimates is profound. Across all maturities and scenarios investigated, LECL falls by 40% and 25% in secured and unsecured lending, respectively. At shorter maturities, the April base case LECL is reasonably close to the upside scenario's LECL in the March forecast.
Does this mean that the March forecast was wrong? Perhaps new information became available that enabled the market to update its view of credit risk (and hence our CCI)? Or perhaps the policy response and market participants' strategies were informed by knowledge of how severe the impact might be? If proven, this would have profound consequences for standard modelled assumptions. These generally pose the systematic (shared) risk factor as exogenous. Discussion of whether to jointly model the complex network effects that link exogenous to endogenous (or endogenous to exogenous) risk, is generally avoided.
These questions can only be answered conclusively via a back-test that traverses the entire quantum multiverse. Until then it must remain unresolved whether we, collectively, changed the result by looking at it?