When I started considering my investment into Lending Club loans, I wanted to choose 36 month loans only. Because these are unsecured personal loans, sixty months seemed to transfer too much risk to me as the lender. That also matches with my personal philosophy to borrow short-term and as cheaply as possible.
I gathered all of the loans from 2012 and 2013, which Lending Club freely makes available, to analyze how 36-month term loans were performing. There are a surprisingly large number still making payments, but it was the Charged Off portion I was most interested in. My goal in this endeavor is to limit my risk, specifically by choosing loans from borrowers least likely to default. For the total pool of loans, 16% were Charged Off.
There are a few data items provided with each loan that give a little bit of information on the purpose of the loan and the background of the borrower. By looking into each item, there might be some useful information to narrow down the list of loans. A data scientist would likely perform a Principal Component Analysis, a look at each independent variable to create a coefficient for a linear equation to determine a predicted value for the dependent variable. Instead, I wanted to do this for a specific subset of variables and I think minimizing risk is a different set of identifiers. We will get into those.
The first variable I wanted to look at was the purpose of the loan. I’ll call this the Category of the loan. There are seven principle categories. In looking at other’s analyses, Credit Card and Debt Consolidation were categories that were possible problems. I also found commentary that people would take personal loans to pay down credit card debt, but would build the card debt back up while paying down the personal loan. As a lender, I have no issue with that. It is a behavior problem that I do not mind as long as they remain current on the loan I own.
In my analysis, four of the categories showed much lower charge off rates. One of those was Credit Card. For these four Categories, the charge off rates were lower for the first six of the seven grades. For the other three categories, the lower charge off rates were only valid for the first three grades.
The result of this took the 16% default rate down to about 14%.
The next variable I looked at was employment length. I don’t really care what the job title is since that can be somewhat misleading. The key to me was an employment length of N/A. This could be because of unemployment, retirement, or privacy wanted by the borrower. In any event, when I looked at charge off rates, the N/A selection was significantly higher than a number in this field. Adding this into the filter brought the charge off rate down to about 13%.
A third variable was Home Ownership. There are five selections here — rent, own, mortgage, none, and N/A. The None and N/A bothered me and the analysis showed these two were clearly different from the other three selections. Including this variable to the filter brought the charge off rate down to 11%.
The final variable I examined was one I created. By taking the annual installment amount and dividing it by the annual salary, I created a payment ratio. I used a simple histogram to show the charge off rate by payment ratio and the was a jump condition if the payment ratio exceeded 7%. Adding this variable to the filter brought the charge off rate down to 9%.
There were a few variables that provided no differentiation — salary and amount borrowed to salary. Both showed some randomness to the output indicating they are likely not a factor towards creating a charge off situation.
Adjusting the charge off rate from 16% to 9% by filtering on a few variables feels like success to me. I had written this analysis up using R and looking at loans for sale on the secondary market. There was quite a bit of manual activity to get through the list. I spent a few minutes converting the program over to accepting a CSV file of the loans for sale and then running the filter. The final step was to sort the loans based on their discount value.
People sell loans for all kinds of reasons but for those who have an urgent need to sell, I am happy to take their loan at a good discount. This is step one. Step two is to look at two characteristics of the loan to see if it fits with my portfolio. That will be the next post.