A variable that is not as widely used and/or discussed is the Number of Open Accounts. Lending Club labels it “open_acc” with the following definition: “The number of open credit lines in the borrower’s credit file.”
As we know, LendingClub only pulls one bureau (TransUnion), and the number of accounts can range from bureau to bureau, so the number being analyzed is technically the number of credit lines currently open and being reported by lenders to that particular bureau.
Also, the number of open accounts impacts a borrower’s credit score. The relationship between these two is not specifically defined or even linear. A borrower having very few lines of credit would tend to have a lower score given the limited information. As the number of credit lines increases, there is a point when it negatively impacts the score, as it is perceived as “credit seeking behavior” or simply risky.
How do we pick which variable to use?
Determining which “number of accounts” variable to use is tricky, because there are many variations, and it is not always clear exactly what they mean or which is the best to use. Currently, LendingClub’s data extract includes other similar variables:
- total_acc – total accounts (this will include closed accounts as well as open ones)
- num_rev_accts – number of revolving accounts
- acc_open_past_24mths – total number of trades opened in the past 24 months
In addition to these, there are also variables related to balances and open credit amount (literally how much credit is available.) In this post, the focus is number of accounts; future posts will be devoted to the “amount” variables.
To figure out which variable is the best to use, we start with some basic diagnostic analyses of the related variables to determine what differences exist. This includes comparing to see if any are likely inclusive of each other:
- total_acc is always larger than or equal to the other variables, confirming the description
- open_acc is usually smaller than num_rev_accts (82% of the time), confirming that num_rev_accts probably includes closed accounts
We’ve found that the open_acc variable is optimal as they give the most accurate picture of the borrower’s current situation. Looking at how many total (open + closed) accounts is not as useful, since some/all of those accounts may not be available to the borrower, and the numbers can be very different (the average difference between total and open accounts is 13 accounts!!)
How is it distributed?
From the below distribution, we see that the bulk of funded loans have < 12 accounts open at a time. However, there is quite a tail on this distribution – with a decent number of loans being funded to borrowers with quite a few open lines.
The underwriting strategy used by LendingClub makes a difference in the distribution, and we know that they require at least 2 open trades, which is confirmed by the data.
How does this predict risk of default?
To answer this question, we banded the number of open accounts into groups of 10. This way, we can determine if behavior changes as the number increases. What we see is that the presence of more accounts is correlated to higher returns. Some of this is due to higher interest rates (these rates are likely related to the impact that number of accounts has on the borrower’s credit score).
Given the volume in each of these bands, we will only look at the < 19 group.
What we see here is that while the interest rates are essentially constant for all groups, the default rates are much higher for those with 4 or fewer accounts, dropping by almost 20% when you go from 4 to 5 open accounts.
As with any analysis, it is important to understand the impact of implementing a finding. Based on this analysis, the conclusion appears to be:
- Definitely do not lend to a borrower with only 2 open accounts
- Potentially do not lend to a borrower with 3-4 open accounts
Of course, whether you utilize this variable really depends on your investment strategy and also the volume it eliminates from the fundable population.
Unfortunately, this variable is not available on LendingClub’s browse notes filter interface at this time, so the only ways to use it would be either manually or in a model that was deployed via the API.