2020 was a year that no-one predicted and not many will forget. The banking industry showed great resilience and adaptability and played a key role in propping up many of the individuals and businesses that were impacted. Hopefully, 2021 will be the year we emerge from the pandemic although the ramifications will be felt for some time to come. Nevan McBride looks at some of the key trends in risk management and makes some predictions for hot topics for the rest of 2021:
Most lenders have embraced digitalisation to take advantage of advancements in data and technology. While some providers were already racing to keep up with the fintechs or established competitors who had adopted more quickly, others took the view that the cost and disruption of the journey called for a more leisurely pace.
The pandemic, however, forced the hand of those firms that were behind the curve as, almost overnight, customers, switched to phone services or digital-based platforms to fulfil their banking needs. While this may have started out of necessity rather than choice, digitalisation will continue to be key in 2021 as preferences and behaviours learned during the pandemic will inevitably carry on afterwards. Customers who may have been previously content with using a branch network for transactional activities now increasingly have expectations of quick lending decisions, enhanced online transactional capability and frictionless digital journeys. A new raft of bank branch closures will make this even more essential, whilst presenting opportunities for some brands who retain a more local presence.
Getting the prioritisation right within a digitisation programme is important; increasingly, our clients are requesting help planning out the best way to embark on this journey. Throughout the transition to digitalisation, lenders should be acutely aware of where human interaction must be retained to handle situations that are better addressed by keeping the human in the loop. But in general, the trend is towards fast decisions, greater accuracy and harnessing tech to deliver great customer experience.
Managing Covid losses
Ever since the recovery from the Global Financial Crisis the industry has been experiencing relatively low and stable credit losses through tightened lending criteria driven by more stringent regulation. Covid-19 however has caused widespread disruption to the economy with government interventions prolonging the emergence of losses.
Lenders must now prepare their collections and recoveries teams for the deluge of delinquency as unemployment increases and business loan deferment periods come to an end. Policies and strategies must be adapted to fit as the profile and circumstances of Covid debtors differ from typical debtors. There are clear benefits of capturing additional data such as debt reason and industry type to support agile repayment plans. Increasing collections automation will be key, with digital collections tools handling the increase in volumes and Open Banking data supporting Income and Expenditure assessments to determine repayment plans (see next prediction).
Artificial intelligence can enable more powerful models to be produced, which will enhance collection-based decisions and enable strategies to keep up with the ever-changing environment. All of these measures, of course, must be considered carefully in light of bereavements and customer vulnerability, with regular reviews of customer sentiment in place and policies altered where necessary.
The PSD2 directive came into force in January 2018 and unsurprisingly, given the potential to transform banking, it earned its spot on our risk predictions list in 2019. It’s fair to say that the adoption of open banking falls short of forecasts made 2 years ago, with only 2 million users to date in the UK - although, given our clients’ plans for this year, that looks set to change.
Many firms are now looking seriously at Open Banking solutions to enable the implementation of automated income and expenditure calculations for affordability assessments and tailored debt payment plans. APIs provide direct and rapid access to customers’ bank accounts (upon granting of consent from the customer) meaning quicker and more accurate assessments can be made. Open Banking data provides far from perfect assessments and this is likely to have put the brake on adoption across the industry.
The key to measuring affordability is the ability to differentiate between discretionary and mandatory payments, which relies on the accurate categorisation of all transactions. Moreover, customers could potentially hide poor behaviour on undeclared accounts preventing a true and complete picture from being drawn up. Firms are currently managing these risks with simplistic and potentially conservative implementations, including benchmarking against industry data from the likes of ONS. With higher customer consent rates, and improving transaction categorisation - coupled with the above-mentioned necessity for lenders to digitalise - 2021 is set to be the year for a ramp-up in adoption of Open Banking.
Fairness in AI and Credit Scoring
The topic of fairness in the context of AI Models has gained a fair amount of media attention. There is a perception by some that AI models are implicitly fair as they are trained on underlying data. However, there have been high profile cases where predictive models have manifested high levels of unintended bias (eg. Applecard).
Indeed, the concept of ‘Fairness’ is quite complex and difficult to define. Does this mean that for protected groups (eg. gender, race) the accept rates should be equal across all groups? For this to be achieved positive discrimination would be required to offset underlying differences in estimated credit quality, which may require data that lenders simply do not have at their disposal.
The underlying issue here is that of proxy data. By way of example, typical inputs to a credit scorecard might include the length of residence or age of the oldest credit account, and these features will be heavily correlated to the age of the applicant. So, whilst a model should not include age as a direct predictor, lower accept rates for younger people will still occur based on proxy relationships.
Within Archetype, Jaywing’s proprietary AI software, it is possible to control the marginal impact on the prediction made by the model. This ensures the model is explainable and at least aligned to risk-based expectations.
A further step rarely performed within the industry would be to check that the model-based predictions and underlying bad rates are well aligned across all Protected Groups.
As most firms are starting to make the switch from traditional models to AI-based techniques, we think a renewed focus on model fairness will be centre stage in 2021 – alongside the standard industry requirements for risk models to be explainable and to always behave in predictable ways.