One last shot at salvaging my Lending Club account

My last post discussed a way of salvaging my Lending Club account by buying notes from the secondary market. I’m buying notes that have had at least two years of regular payments and are in their last year. Hopefully, this will reduce the number of charge offs.

The only other way of adding notes is to fund new loans as they become available. The question then becomes, how to identify the better opportunities? As I write this Lending Club has 645 loans available for funding. A few hours a go there were 691 potential loans. In between, Lending Club added some new loans (they do this 4 times a day) and some have dropped off. That’s too much activity for me to want to handle manually.

Fortunately, they do have a download mechanism to allow you to get a data file. Like to other data files, it contains lots of information. There are 121 potential data points for each loan. This leads me to want to develop an automated method (computer program) to process these loans and select the better ones. To do this you need to identify the fields that are most likely to indicate desirable loans. This can be done by using previous loan data and back testing scenarios.

I wrote another program to accomplish this back testing by simulating what would have happened if I had implemented this strategy beginning in 2015. I picked two potential data fields that I felt in combination might provide a selection method of identifying notes with potentially better results. I passed all loans prior to 2015 building a table of counts. I used a technique like the one described a few posts back. I put the two fields together in a match code and built a table from the pre-2015 notes. I used only the completed loans. The table had the match code, the number of notes with this match code, the amount of the loans and the amount paid back. From these I computed the return on investment.

I only used 36-month notes since those are the ones I want to invest in. This initial table had 17,707 entries (number of unique combinations of the two fields). I then passed all the 1^st quarter 2015 notes, building the match code and getting the appropriate values from the table. I only selected notes that had at least 100 table entries and a return of $1.05 or more. Finally, I added the 1^st quarter note values to the table and then moved on to the 2^nd quarter notes. I repeated this process for all 8 quarters in 2015 and 2016. Following is a table of the results.

36 Months	All Notes			Selected Notes
Quarter Issued	Number	% Charged Off	Return/$	Number	% Charged Off	Return/$
15_Q1	56,569	11.5%	$0.953	4,130	9.2%	$0.956
15_Q2	64,222	10.6%	$0.883	4,191	8.5%	$0.888
15_Q3	73,572	8.6%	$0.810	4,724	7.1%	$0.811
15-Q4	88,810	6.9%	$0.717	5,215	5.6%	$0.720
16_Q1	96,120	5.6%	$0.608	4,049	3.9%	$0.614
16_Q2	74,537	4.3%	$0.518	2,106	2.6%	$0.514
16_Q3	73,898	2.4%	$0.425	2,005	1.2%	$0.407
16_Q4	78,940	0.7%	$0.304	1,768	0.4%	$0.297

These are 36-month notes so some are notes not yet fully paid or charged off and are still being paid monthly by the borrower. The selected notes are better relative to charged off rate. The amount returned thus far is better for the older notes and not quite as good for the newer one. However, the newer ones have a higher percentage of currently active loans.

The results are not encouraging. There is an improvement in return rate, but less than 1%. This is not a combination that I will be using in the future. I can and might generalize the program to look at other fields and combinations. But at the moment, I’m inclined to move on to other projects for a while.

If I find something of interest, I’ll post it. Follow me on Twitter @billlanke and I’ll let you know when that happens.