As a lender, imagine an applicant came to you with no credit history or credit score. How would you determine whether to approve the loan? And, at what cost?
Such is the current state of much of the US lending industry. Though credit reports cover the majority of US adults, the sudden onset of this pandemic and a skyrocketing unemployment rate has left lenders driving blind. Given that no credit models have been trained on these pandemic-induced economic conditions... it is clear that making a credit determination by looking at credit history alone is helpful, but wholly insufficient.
This crisis reveals a central weakness of our credit system - we attempt to predict the future based on what has happened in the recent past. Notably, this brings up a few issues:
The future may simply look different than the past. The past is less valuable when “black swan” events - so called for their rarity - such as pandemics, financial crises, political crises, terror attacks, and climate change-related disasters now seem to occur with an unpredictable frequency.
The chicken-and-the-egg conundrum: you need a previous credit history to get credit, but you need credit to begin getting a history. Consequently, an estimated 45M Americans are considered “credit invisible.”
Overall, this crisis is likely to accelerate a key underwriting trend - the move away from backward looking models to one where lenders use more forward-focused credit models.
forward-looking > backward-looking credit models
Though all credit models attempt to predict the future, how they do so will shift: less reliant on backward-looking information like credit reports and more oriented toward forward-looking data like cash flow and future predicted income.
This shift will be enabled by changes in the underlying data and analytics. Here are 4 predictions:
#1: Cash flow data will become widespread in its use
Cash flow data helps lenders understand real-time measures of a person’s level of liquidity, savings, obligations, and any recent changes in income (e.g., a decrease due to a paycut, lost hours, or unemployment; or, an increase due to a bonus). The picture provided is more immediate and richer than a credit report alone.
Further, research on the use of cash flow data in underwriting has been found to boost financial inclusion. This data has only recently become readily accessible to lenders through fintechs like Finicity and Plaid and is expected to continue broader adoption.
#2: Employment & income data will be used to predict future income for underwriting
How lenders ask for employment and income data will change. In a forward-looking credit model, lenders will increasingly look at a person’s employment and income information to estimate future predicted income.
Future predicted income when used alongside current income helps measure income risk sensitivity to both adverse shocks and upside potential. Such a calculation will help lenders understand a person’s ability-to-repay both today and in the future.
Employment data such as sector, company, job category, and position will be used to determine how various scenarios affect future income. The same is also true of upside potential - a person in a growing field with fast wage growth may have a higher future predicted income.
Because employment and income data will become more crucial for underwriting, more lenders will seek to verify self-reported income and employment information.
#3: Machine learning (ML) will come to dominate credit risk analysis
Machine learning-based credit models will outperform more traditional credit models. This is true for two principle reasons:
First, ML allows for the inclusion of more data (like cash flow data, employment data, more macro-economic data, etc.) which increases the range of scenarios and factors the model is trained on.
Second, ML algorithms identify connections between sets and groupings of variables that traditional models don’t account for.
This will give lenders an edge in assessing future risk and hasten the transition towards more forward-looking approaches.
Though there are various flavors of machine learning used in underwriting, the industry will increasingly use the unconstrained, full-powered flavor of ensembled learning used by fintechs like Upstart, ZestAI, and Ensemblex as the explainability gap is addressed.
#4: Lenders will build the credit model infrastructure needed to enable rapid model enhancements
Credit models aren’t static. They’re meant to be refreshed and retrained as more data becomes available in a continuous feedback loop. In times of crisis, this becomes even more critical: feeding early data back into a credit model will immediately enhance its accuracy and staunch heavier bleeding.
As a way of future-proofing credit models, lenders will invest in rapid release capabilities. Rather than measuring the time from build to deploy in months (or even years), it could happen in days or weeks. The key advantage will be how fast lenders can turn around a refreshed model that incorporates new data.
What this all means
A world is coming in which credit is more forward-focused than backward-focused. For lenders, it has become a business imperative to insure downside risks. This shift will be enabled by new types of data and ML analytics.
Ultimately, this evolution should also be more beneficial to people: if more underwriting looks at future potential, more people without current histories would have access to affordable credit.