A few years ago, I opened a Lending Club account. I was intrigued by the idea of peer lending, sorting through the loans was kind of fun, and I actually did reasonably well investment-wise.
I wound up averaging something like 6-7%, but it ultimately became a bit of a time suck. Thus, I decided to wind down my portfolio.
A big part of the problem was that, though I had “developed” (i.e., picked out of thin air based on little more than a hunch) some filters for picking loans, I spent a lot of time sorting through notes, reading the backstory, etc.
While this was entertaining (to a point), I eventually lost interest. Honestly, I probably could’ve gotten similar returns with random note selection as opposed to using my “gut feeling” filters. I needed to make it more efficient.
Digging through data
During the past month or so, I started playing around with the historical data (available from Lending Club) looking for patterns. While it’s quite possible that I’ve been “Fooled by Randomness,” it looks like there’s some low hanging fruit.
Note: I’m an indexer at heart, and not usually the sort to backtest and look for strategies to give me an edge. But I’m also a stats geek, I enjoy poring over numbers, and I still love the idea of peer lending. So here we are.
The problem with analyzing highly multi-dimensional datasets is that there are an astronomical numbers of possible factor combinations, and things often interact in unexpected ways. Thus, it can be difficult to parse these data by hand.
I was thus fascinated when I ran across a guy who had written a genetic algorithm to explore the historical loan data in search of filtering criteria that would maximize returns while minimizing defaults.
Loan selection criteria
In short, this algorithm identified a relatively simple set of filtering criteria that, as it turns out, also make logical sense. While the analysis was run awhile ago, I’ve tested it on a couple of Lending Club stats sites and it seems to do quite well.
Here are the criteria:
- Credit grades: C, D, E
- Debt-to-income: ≤ 25%
- Inquiries, past 6 months: ≤ 1
- Home ownership: Mortgage, own
- Loan purpose: Refinancing credit card, consolidate debt, home improvement project, renewable energy financing, wedding expenses, paying for vacation, covering moving expenses, home down payment
- Months since last delinquency: ≥ 24 months
- Exclude loans with public records: Yes
- Exclude states: AZ, CA, FL, GA, IL, MD, NV
Based on the original backtesting, these criteria should provide somewhere around 12.5% annual returns with no more than about 2% defaults. Much better than average, but see above re: the possibility of being fooled by randomness.
It’s also worth noting that the code is available on GitHub, though I’ve had some trouble getting it to run on more recent data, primarily due to a change in data formatting. If I manage to get it working, I’ll run some analyses myself to see if/how things have changed — and I’ll be sure to let you know.
In the mean time, I’ve opened a new account and will be using these filters to see how well they really work. Of course, I’ll be detailing my experiences here.
A matter of scale
So how much money are we talking about here? That’s actually the primary challenge with Lending Club. Even if you’re comfortable deploying large amounts of cash via their platform, it will take some time to do so.
The reason for this is that, while they issue a significant number of loans on an ongoing basis, many of the notes are full subscribed in short order. This is likely due to the involvement of institutional investors with relatively deep pockets.
It’s also worth noting that Lending Club now holds back a random selection of ca. 20% of their loans for institutional investors that want to buy whole loans rather than making fractional investments.
Whatever the cause, this pattern of loans getting rapidly snapped up once they hit the market means that there might be slim pickings when you go to invest. This is especially true if you have fairly stringent filtering criteria.
I will note, however, that Lending Club adds new loans to their platform every day at 6AM, 10AM, 2PM and 6PM Pacific. Thus, if you time it right, the loan selection is far better than at other times of day.
For starters, I made an additional deposit of $1,500 and will be adding to this on a monthly basis. This will likely be on the order of $500-$1,000/month; I’ll fine tune the amount going forward based on how quickly I can deploy the money.
As for note size… The minimum is $25/note but I will be investing at $50/note. This strikes a balance between spreading my money across a relatively large number of notes and getting my money invested.
As for note selection… I’ve created a loan filter in the Lending Club interface using the criteria outlined above. Thus, I can simply log in, click the “Browse” button, click to filter, and buy my notes. The whole process takes about a minute.
I should also note that Lending Club offers both 36 and 60 month loans. I’m focusing solely on the former. The latter do have higher interest rates, but they also seem to have disproportionately higher failure rates, so I’m avoiding them for now.
Of course, this further narrows the field. If loan volume becomes a huge limitation, I’ll either have to revisit the note size or the selection criteria — or I’ll just curtail my lending activity entirely. We’ll see. I’ll update as we go along.
If you’re interested in trying it out, you can get started here.