Written by Jeff Keltner, SVP Business Development
When talking to lenders about alternative data in lending and leveraging enhanced data sets to assess credit risk, one of the questions that regularly comes up is: Are that many variables really necessary?
The short answer is yes.
Increased variables improve accuracy and allow users to leverage more sophisticated predictive intelligence. In fact, Upstart's model uses more than 1,500 variables as part of its underwriting analysis, and many of these data points are highly correlated.
For example, many of the multiple debt-to-income variables (DTI ratios) Upstart analyzes look similar, but the ratios are different depending on which types of debt are included. Each one tells a different story and can help uncover a different pattern.
While each of these variables is similar, the differences between them often have important implications for overall predictive power.
Increased variables improve accuracy and allow users to leverage more sophisticated predictive intelligence. Upstart's model uses more than 1,600 variables as part of its underwriting analysis.
The power of variables in credit assessment
Individual variables on their own may not be that helpful. Users might be able to take them out of the more sophisticated models and it would not change things much, but combining them becomes very powerful because each variable tells its own side of the story. The little power of each additional variable, across a wide number of variables, provides additional data and allows for more sophisticated models with greater accuracy.
While most traditional lenders continue to use simple FICO-based models to analyze credit risk and assign rates, such scorecard methods only provide snapshots and are limited in quantifying risk. For example, consumers that don’t already have an account at their financial institution and have a limited credit profile may not provide the data needed for an accurate assessment.
A third of Millennials say they’ve been rejected when applying for credit because of their credit score, yet many millennials don’t have credit cards or established credit. Many of them use debit cards and pay all of their bills on time and may be great candidates for loans. In situations like these, FICO scores might not be useful.
There are still hundreds or thousands of data points that may be available across other sources. By going beyond FICO scores and using non-conventional variables, Upstart leverages artificial intelligence, machine learning, and predictive analytics to assess creditworthiness at scale.
Provide greater credit access without increased risk
Using a large number of variables allows lenders to provide more access to credit for consumers without increasing their risk profile. In one study, the Upstart methodology approved 173 percent more loans at the exact same loss rate as compared to traditional bank models for approving credit. At the same approval rates, the study showed 75 percent fewer defaults.
Intelligent decisioning based on extensive variables leads to better overall outcomes. Because of this heightened accuracy, financial institutions can also provide nearly instant approval of loans and fully automate the process for most borrowers and lenders.
So, the more variables that are analyzed, the more accurate the credit profile is and the better credit decisions financial institutions can make. Upstart allows users to process loans using thousands of variables quickly and at scale.