Debt To Income
Debt to Income (DTI) is a metric that many active investors in Marketplace Lending are familiar with and potentially use in their investment strategies. It represents the total amount of debt payments the borrower owes each month excluding mortgage (e.g. credit cards, student loans, car loans) divided by the stated monthly income. The more debt a borrower has relative to the stated income, the higher the DTI ratio. Both Prosper and LendingClub use DTI to underwrite. LendingClub only allows borrowers on their marketplace if DTI is < 35%, and may have more restrictive cutoffs for borrowers based on other credit attributes. Prosper has a DTI cutoff of 50% and similarly may also have additional restrictions. The DTI that Prosper displays on its website and in the data download includes the loan being considered in the “monthly debt” amount of the calculation, however the pending loan is not included in the 50% cutoff used to consider the loan for the marketplace.
Given that both marketplaces are already cutting out borrowers with high DTI, does it make sense to further segment or cut out borrowers based on this variable? Does including the pending loan amount provide a more predictive variable based on past performance?
Below is the distribution of DTI used in Prosper’s underwriting decision (monthly debt / stated monthly income) for loans originated in 2014. We found that the 50% cutoff does appear to be enforced.
In the table below, we show the the average FICO and interest rate of these loans by DTI band. While the FICO score is stable, the interest rate is much higher for higher DTI. Therefore, not only is Prosper using DTI to decline certain applicants, it is also likely used to set the interest rate on loans that are approved on the marketplace.
DTI Band Avg FICO Avg Interest Rate
0-10% 694 13.50%
10-20% 696 13.50%
20-30% 697 15.00%
30-40% 697 17.70%
40-50% 695 19.80%
We reviewed the LendingClub declines from 2013 and 2014 and found that 23% of the declined loans had a DTI > 35%. The only other variable we have to assess of the rejects is FICO, and we see 60% of those failing the DTI would have passed (assuming a 660 cutoff). Assuming that LendingClub and Prosper have similar applicant pools, it appears that about 20% of declines are due to high DTI, but a portion of these would have been declined for FICO or possibly other reasons.
In order to understand if the increased interest rates outweigh any increase in default for higher DTI bands, we looked at the interest rate and default rate (defined here as missing at least one payment or worse) of Prosper loans from 2012. What we see is that both rates increase as the DTI increases. However, it appears that the default rate increases at a faster rate in the higher DTI bands (30%+).
In order to assess this, we looked at the difference between the interest rate and default rate (as a proxy for profit). In the higher DTI bands, the margin between interest rate and loss rate tightens. Based on this, it would suggest that the applicants with higher DTI have historically had a lower profit margin for investors.
Next we looked at whether borrowers with a stated purpose of debt consolidation (78% of current loan volume) have similar performance by DTI as all other borrowers. The default rate for the debt consolidation borrowers is lower than the others in each DTI band – but it appears that DTI still ranks within both populations.
Based on what we’ve seen in this analysis, it appears that DTI would be a useful variable to consider when designing an investment strategy. In higher DTI bands, we’ve seen higher default rates that are not necessarily offset by the higher interest rates. Both LendingClub and Prosper provide many credit and application variables to assess on backtested data. At Orchard we’ve built models on both populations using this data, in both cases DTI has not proven to be a significant enough input to be included when assessed against other potential variables. We wondered if DTI would prove significant as an overlay on top of our model on Prosper’s 2012 population. What we found is that when we used our standard model and a sample strategy on top of that (this strategy qualifies about 60% of population), the interest rate and default rate differences are more subtle across DTI bands.
The profitability proxy we looked at shows that the highest DTI band (40-50%) is slightly worse than all other bands.
When looking at the same metric, but in 5% DTI increment bands as opposed to 10%, we see that the 45-50% DTI band is overall negative.
Based on this information, DTI is clearly an important factor to be assessed. If using a filter strategy, it would make sense to assess the higher DTI levels based on profitability. If using a statistical model that does not include DTI (such as Orchard’s) it may make sense to put in a limit of 45% or so on DTI, although DTI has less of an impact in this case.