Updated: Oct 29, 2020
Much of the credit risk stream of the conference was forward looking and posed the questions with which we are all wrestling – what will be the impacts of COVID and what tools do we have to better second guess the impact to profits and balance sheets whilst continuing to support the economy and write business?
The economic picture
The current position of the health crisis, as viewed through the lens of financial services, is much like the pregnant pause associated with an impending tsunami. The earthquake has happened way out at sea, the water has rushed off the beach and people have run for higher ground. Now they sit and wait to see the damage the sea will do. In an online vote during the conference, the majority of attendees believe that peak losses will not be until Q2-21 or beyond, once government subsidies fall away.
However, forecasting the damage from the crisis is made all the more tricky
There has been no comparable crises on such a global scale - you have to go back 100 years to the Spanish Flu)
The current crisis is different to Great Financial Crash (GFC) in 2008. It is not initially financial driven, severity of the impacts come from lockdowns and the logistical consequences of social distancing, many of the impacts are largely unseen as yet and the magnitude is way beyond GFC.
The governments’ fiscal, monetary and macroeconomic policy responses have been unprecedented and so looking back to past crises will not be particularly useful as templates for what will occur in 2021 and beyond.
This is a truly global crisis, world trade (both internal and external) has taken a knock, there are sector specific hits with transport, leisure and tourism particularly bad.
There is huge uncertainty in exit back to “normal”. There is uncertainty in the outcome of a potential cure, the transmission of economic stress from one industry to another and wide ranging debate on the recovery shape (U, V, W, K). Where will banks end up?
So, where did delegates expect to place their bets? Most people expect a further downturn is coming. Leisure, tourism and transport will continue to struggle. Leisure that managed to survive the summer by providing outdoors options (e.g. restaurants), will find trading much more challenging in winter. There is early evidence is that there will be lower demand for office space (particularly in city centres) and residential property in densely populated areas and so there will be increase pressure on the infrastructure (shops and services) that supports city office workers.
Conversely, products and services that support working from home (tech firms, residential property market outside cities like London) are likely to do well, as are pharmaceuticals.
Regulatory stress testing
The overall policy responses of the regulators and government were, and continue to be, to cushion the shock and avoid amplification of the crisis – especially for banks’ solvency and ensuring sufficient credit supply to those that needed it.
The stress testing regimes set up following the GFC have been leveraged by the different regulators (Fed, BoE, ECB etc.) with a focus on high severity outcomes, specific to risks faced at appropriate granularity for those risk. The tests have been top down rather than bottom up, enabling multiple scenarios, with severity increased and a baseline closer to stress environment already. Further tests will continue into 2021.
The current conclusions are that there has been sufficient resilience to date, with external shocks having limited impact on banking. Financial services firms were in a better position going into this health crisis than they were going into the GFC. There was a short-lived hit via financial markets with policy responses making a big difference.
However, we are clearly not out of the woods yet. There is growing expectation of a 2nd wave and real side shocks accompanying it. Expiration of policy effects and policies also have unknown consequences, as do lagged effects on P&L and capital.
Going into 2021, the objectives are to understand the impacts of policy (IFRS 9, TLTRO etc), publishing only at the aggregate level with no bank level info – closer to macro-prudential approach. The tests will implement a new extreme and plausible benchmark to replace GFC and make use of reverse stress testing.
Measurement, Data and Modelling
As technological capabilities continue to develop at a rapid pace, there is an expectation that models built on big data and machine learning become increasingly used within credit risk management. Historically, the slow response from regulators has been seen as something of a blocker to progress as has some of the transparency of the machine learning models themselves.
There was much discussion on the need for banks to catch up in this space. Internet firms such as Google and Facebook are light-years ahead in the exploitation of the technology and, given their existing access to information from news flows and social media, they present a real threat to the distribution channels of financial services firms. It is not inconceivable that the banks become disintermediated as a result.
This shouldn’t be the outcome if they begin to more aggressively use the abundance of data they do have from customer activity and the additional data they can now access through open banking in addition to news flows. With the current Covid crisis, the ability to see trends that only machine learning uncovers is even more vital.
To do this, banks need to be more data driven, even mirroring the strategies of the big IT companies. This may seem like something of an extreme statement but banks’ reliance on IT is huge: for risk management alone stress testing, risk appetite, pricing, reporting and risk models are all tightly interwoven and impact each other and demand vast quantities of data and processing.
This is not to say that the banks face significant headwinds in adopting machine learning. Bank databases are of varied data quality, fragmented and are a struggle to use and the older and larger the bank, the more so. Additionally, regulators are not yet ready for capital, impairment and stress testing models to be built through machine learning algorithms (albeit, they can provide a useful benchmark to existing models) and so both the incentives to spend on development and the resources to perform the work are held back for the existing model suite.
This shouldn’t stop firms moving forward. Understanding the ecology of data systems and building use cases through small scale trials of the technologies available, whilst actively engaging the regulator to move faster, will help to shape a viable strategy for the future.