Modern technology that re-engineers credit risk scoring, and decisions systems lead to better outcomes for lenders, says SAS Director of Financial Services, Paul Franks.
A confluence of factors is driving changes in the way that banks extend credit. Chief among them is the International Financial Reporting Standard 9 (IFRS 9) which requires banks to radically reassess how they calculate and recognise credit losses. Before the arrival of IFRS 9 banks were only required to allocate capital for losses on receiving objective evidence of impairment, disregarding any future losses. IFRS 9 introduces expected credit loss models that account for possible future losses, a shift that is likely to increase banks’ loan provisioning requirements and constrict their available economic capital. Proposed revisions to the Basel regulatory framework will further limit banks’ capacity to use capital. Digital banking, with its demand for decision speed and higher levels of customer experience, is placing pressure on credit risk decision approaches and risk modelling.
Today’s analytics pair a desire for granularity with the need to process ever larger datasets they must process more data, with greater depth and accuracy.
Time for a change in credit risk modelling
The shifting nature of credit has led to advances in analytics with widespread effects on internal systems and processes. The days when traditional credit default risk models provided sufficient analytics are diminishing. Today’s analytics pair a desire for granularity with the need to process ever larger datasets – they must process more data, with greater depth and accuracy. To accommodate and exploit the information generated by new analytical techniques and models, banks’ data processing systems have become vastly more complex. Rather than simple unidirectional or bidirectional information flows, data now moves across densely interconnected networks.
Organisations that develop their own credit risk models see long lead times to get them built and deployed. Business users determine a need for a new model, which triggers weeks or months of data collection and model development effort. By the time the new model is deployed, market conditions and customer needs have changed, so the process starts over. Restricted by handoffs and manual interventions, the entire analytics and modelling process can be slow and inefficient.
Legacy credit risk tools generally cannot access the novel, alternative data that would generate a more accurate risk score – or get the data fast enough to react to market changes. Tools based on disparate products that don’t integrate well can also increase regu¬latory risk. Generating customer-level versus account-level credit scores is practically impossible and could triple the data volume.
Credit risk scoring is only as good as the data that feeds the process. It’s essential to have strong capabilities to access, transform, standardise and cleanse all requisite data. This includes third-party bureau, transactional, application, billing payment and collec¬tions data. Getting from a credit score to a customer-facing decision can be sluggish. Even with online loan sites that appear to front a sophisticated decision environment, there is often a lot of manual processing behind the scenes.
It’s time to re-evaluate the cost efficiency and sustainability of credit risk scoring models and processes.
There is new urgency to resolving these deficiencies in the credit scoring and decision process. It’s time to re-evaluate the cost efficiency and sustainability of credit risk scoring models and processes. Banks and non-bank lenders alike are starting to rethink how they assess credit risk and make decisions, from approving a credit card applicant to identifying at-risk accounts, to balancing portfolios and mitigating enter¬prise-wide risk.
Resilient model risk governance process
When you layer multiple analytics methods, you can more accurately assess credit-worthiness of an individual account or at the portfolio level. Credit models must reach production while they are still accurate and reflect what’s happening today. This calls for estab¬lishing a cohesive model risk management process – an automated environment to quickly develop, deploy, track and assess models.
To gain a better understanding of the risks presented by a given model, stakeholders must become more involved in the model development and deployment lifecycle. Robust procedures for model oversight will become even more important as models become more complex.
Given the persistent low level of regulatory tolerance for “black box” analytic models and processes, everything from data creation to analytics, deployment and reporting should be transparent. Not only is this information valuable corporate intellectual property, it will be needed if regulators question your credit decision methodology.
Applying machine learning
Machine learning can uncover correlations among different attributes that traditional linear models would miss. Banks and non-traditional financial entities are turning to machine learning to make more precise decisions about risk-based pricing, loan approval, and debt collection. Machine learning is proving especially valuable when using alternative data, where attributes may be indirectly correlated. Other examples include using machine learning to identify which variables have good predictive quality and then pouring that insight into traditional models, or, alternatively, building a primary model with traditional modelling techniques or rules, and a parallel model with machine learning. Where the machine learning model can be difficult to interpret, the companion traditional model can explain the result.
Lenders that capitalise on modern technology to re-engineer their credit risk scoring and decision systems can improve the quality of leads and make better recommendations, while reducing manual activities, maintenance costs and losses. How is your credit risk modelling system placed to meet the requirements of digital banking?