Welcome to a Post-FICO World!

Transcription

1 Welcome to a Post-FICO World!

2 Consumer credit modeling relies on data and analytics that haven t changed in decades

3 A smarter prime lender could approve almost twice as many borrowers and yet have fewer defaults 100% 80% 60% Percent in US with loans but have never defaulted** 40% 20% Average lender approval rates* 0% Defaults Traditional Underwriting Modern Data Science * Source: Prosper, Lending Club **Source: Upstart data study with TransUnion

4 So why doesn t everyone do it? Real data science is hard Regulatory risk is daunting

5 So you want to add a new variable? Some helpful attributes Broadly available Decade+ of training data Easily verifiable Unbiased and legal Hint: Facebook is not the answer!

6 We ve assembled a collection of variables that are more predictive than the entire credit bureau file 20 3-Year Student Loan Default Rate (%) School ranking

7 And by layering all of these variables together, we can make smarter credit decisions instantly 15 Default rate of best 40% from sample population 12 Default Rate (%) Random Financial variables Financial variables Obtained a degree Financial variables Obtained a degree School ranking Major Financial variables Obtained a degree School ranking Major SAT/GPA Data from NCES National Education Longitudinal Study

8 Data that is predictive in a recession is even more valuable Unemployment rate by level of education

9 A disruptive credit model requires unique predictive data, better math, and faster learning Traditional Upstart Variables Credit file Income Credit file Income Occupation Employer Work Experience Degrees Schools GPA Test Scores Job Offers Cost of Living etc. Methods Black/white decision logic, simple regression Continuous decision logic, cross-validated logistic regression, higher-order variables, random forest, monte carlo methods, ensemble learning Learning Speed Lenders 2-3x per year, FICO 2-3x per decade Automated training, daily updates

10 When you re building a disruptive credit model, verification of inputs is essential Upstart Borrower income verified 100% Borrower education verified 100% Borrower savings verified 100% Verification phone call 100%

11 Proof in the pudding: steadily increasing approval rates and consistent investor returns Approval Rate of Control Group IRR by Origination Month 30% 25% 20% 15% 10% 5% 0% MAY 2014 JUN 2014 JUL 2014 AUG 2014 SEP 2014 OCT 2014 NOV 2014 DEC 2014 JAN 2015 FEB 2015 MAR 2015 APR 2015 MAY 2015 JUN 2015 JUL 2015 AUG 2015 SEP 2015 OCT 2015 NOV 2015 DEC 2015 JAN 2016

12 Our model has learned quickly, with each cohort performing better than the prior Cohort # Originated % DQ121+ Q % Q % Q % Q % Q % Q %

13 Our delinquencies by loan grade also provide evidence that we re accurately pricing our loans Loan Grade # Originated Average Age (Months) % DQ121+ Modeled % DQ121+ AAA % 0.02% AA % 0.15% A % 0.46% B % 1.31% C % 2.22% D % 3.70% E % 0.94%

14 Sounds great, but my lawyers say no! - You

15 Q: So you give loans to wealthy grads from elite schools? A: No. Less than 2% of Upstart borrowers come from elite schools. And wealthy people don t need our loans.

16 Q: Your average borrower is 28 years old - are you biased against older borrowers? A: No. In fact, all else being equal, an applicant with longer credit history will get a lower rate on Upstart.

17 Q: Does your system discriminate against people based on race, gender, or other protected classes? A: No. Using a tool provided by the CFPB, we were able to demonstrate that our model demonstrates no statistical bias with respect to race or gender.

18 ( All successful credit models are based on the same tried & true concepts = f ( Financial Capacity to Repay X Propensity to Repay Income Earning potential Unemployment potential Personal Characteristics Credit history Personal responsibility Awareness of credit score Expenses Debt obligations Living expenses Support Network Network connectedness Backstop financial support Spending habits Assets Available to service debt but modern data science can make these concepts better

19 Success in our case means reducing the price of credit to 65M underserved borrowers Upstart Lending Club Percent of borrowers Borrower age

20 Thank you!!