Filtering Loans – Part 2: Avoiding Defaults on Lending Club and Prosper

Avoiding-Defaults-Lending-Club-Filters

Part 1 – Introduction: Risk
Part 2 – Avoiding Defaults
Part 3 – Low-Grade Loans
Part 4 – Multiple Filters
Part 5 – From Tools to Platforms

Unlike a mortgage, peer to peer loans are unsecured. When a borrower fails to pay their loan back (called a default), the lender loses their entire investment. Almost every lender will experience a default eventually, but there are two important ways to soften the blow. First (and most important), we can be diversified in at least 200 loans. Second, we can choose to invest in the highest quality loans available, the loans that have the lowest chance of defaulting.

So which loans are least likely to default?

In part 2 of this filtering series, we will do our best to answer this important question. The reality is, just a little filtering goes a long way in our attempts to avoid defaulting borrowers.

Dating Truthful People

In part 1 we looked at how statistics can help our filtering process. Let’s look at another example:

Imagine you are a single and looking to begin dating (maybe you don’t have to imagine). When you meet potential partners, you always take time to examine their character. You begin with some basic questions to assess compatibility, like “What is your favorite music?” First dates are often awkward because so much of it is just trading raw information that people have little time to enjoy each other. But this process is important. The fact is, people who have a certain characteristics will often be worse to date. For instance, people who routinely lie are generally not good boyfriends or girlfriends.

Imagine all the habitual liars were put in one (very large) room, and all the habitual truth-tellers were put in another. By only dating people in the truthful room, you would probably find your dates to be more fun and authentic.

The world of peer to peer lending isn’t quite like dating, but we can similarly ask some basic questions to improve our experience. We can approach our huge pool of historical loans, discover the riskier factors in them (like how habitual liars are bad boyfriends), and filter out the loans that have these qualities.

Platform Underwriting Rocks

Helpfully, Lending Club and Prosper have already removed the very worst borrowers through their underwriting, the process of examining each applicant and approving or denying them for a possible loan.

For example, here is an actual applicant for a Lending Club loan in 2008:

girlofmydreams

Lending Club rightly denied this one. While all of us would love to help people meet the girl of their ‘dream’, this particular person is probably irresponsible with their finances, seen in their poor FICO score (credit score) of 489. Both the platforms have put a lot of effort into solid underwriting, and with great effect. For Lending Club alone, over 900,000 people have been denied for a loan in comparison to the 120,000 who were funded.

p2p platform underwriting

Filtering the platform simply takes the above graphic one step further; it is like underwriting your own account. Filtering simply takes all the available loans and selects a smaller higher quality group of them that are less likely to default.

p2p filtering graphic

Finding Responsible Borrowers

So how do we find these loans? The platforms have a ton of different filters we can use. For instance, we can filter their loans by income, employment, or credit score. We can filter loans by how many times they have been late on their bills or whether they have ever had a bankruptcy. There is a long list to choose from, and this list can be quite confusing, especially at first.

LendingClub filters

(click to zoom)

We can reach the filtering section of the Lending Club website on the left hand side of the screen in their Browse Notes area.

prosper filter screen

(click to zoom)

Similarly, we can find the Prosper filtering section in the Advanced Search screen of their Invest section.

The best way to begin is to start with what makes sense. For instance, people who have been employed a long time have more consistent income, and as a result are more likely to have money to keep their payments up. Conversely, many people (like our “girl of my dream” applicant above) have poor employment, and these people will be less likely to pay you back over time.

Four qualities of good borrowers:

  • Longer employment
  • Higher incomes
  • Few bankruptcies (if any)
  • Fewer recent credit inquiries

If you can filter the platform for these kind of borrowers, then you are already off to a good start at lowering your future defaults.

This is not just my opinion. There is mathematical evidence that these four borrower qualities (and many others) have a statistically significant ability to lower defaults. Michael over at NickelSteamroller has done the lending community a huge favor with his Return Forecaster tool. We can use this tool to run some simple statistics to create excellent filters.

Lending Club Filtering Exercise

Let’s try an exercise for Lending Club using the p2p website NickelSteamroller.com. We will insert some filters and watch the default rate drop and the return increase.

Step 1. Go to NickelSteamroller.com and click on the link that says “Return Forecaster”. The page that comes up lists 120,000+ loans that have been issued on the Lending Club platform. If you look further down you can see how the return on investment (ROI) for all these loans combined is about 7.72%. Not bad! This is way better than the savings rate at your local bank which is probably mired somewhere around 0.5%.

NSR April 2013

We want to make this ROI go up! This is done by filtering these thousands of loans.

Step 2. Insert Two Simple Filters: In the “Return Forecaster” on NickelSteamroller you will see a list of filters on the left hand side of the page. Scroll down until you see the following filters and change them to zero:

  • Inquiries in Last 6 Months = 0: I only invest in Lending Club loans with zero inquires on their credit report. Filtering a person’s inquiries is perhaps the best filter we can use to avoid risk. These inquiries are the number of times the borrower has applied for credit within the past six months. If a borrower has multiple inquiries, they are more likely to default, because they may have been rightly denied by somebody else and are shopping around for a loan.
  • Public Records on File = 0: If you scroll down to the bottom of the page, you will see this great filter. A public record shows up on a person’s credit report if they have had a bankruptcy, lien, or judgment against them. Avoid these borrower as well.

Step 3. Create Your Filter: Now scroll to the top and click on the button that says ‘Filter’. What happens when we trim down those thousands of historical Lending Club loans with these two simple filters? Our (historical) ROI goes up to 8.5%! Additionally, the default rate dropped from 3.1% to 2.1%. Not bad for such little effort. As you can see, the 120,000 loans have also been trimmed in half, meaning that half the issued loans in Lending Club’s history are borrowers with a past inquiry or public record. This is a good reminder that some loans are better than others.

NSR April 2013b

Adding Employment & Income

employment length filterLet’s add the next two, filtering the historical loans for those with longer employment history and higher incomes. Go to the Employment Length filter and check the six boxes for 5+ years of employment.

monthly income filterThen go lower on the page and make the borrower income at least $5000 per month.

Hit Filter again. Just like that, the historical ROI climbs to 9.54% and our default rate drops to around 1.6%.

NSR April 2013c

Adding Additional Filters

You can continue to make the historical ROI go up and defaults go down by adding even more filters. For instance, many lenders do the following:

  • Exclude California & Florida – This filter is a whole post in itself, but many lenders (including myself) have found that borrowers in California and Florida are statistically less likely to pay back their loans.
  • Increase Total Credit Lines – A borrower having had more historical lines of credit often means their history has more reliability.
  • Debt-to-Income Ratio – This is the percentage of a person’s monthly gross income that is currently going towards paying their debts

There are many others as well! You should take time to get used to the different filters, their terminology and attributes. Filtering the platforms with these tools is a skill that you can get better at with time.

Trust the Numbers

One of the first things I learned when developing a good filter was to second guess my instincts. Sometimes we assume that a specific filter is good, when it may not be. For instance, if we take our current filter and remove all borrowers who have had a late payment (called a delinquency or DQ) in the past two years (set Delinquencies Last 2 Years to zero), you would assume that the ROI would go up. Instead, it actually goes down. Interesting, right? Sometimes we can guess why this happens. Perhaps people who have had a late payment or two also have a more authentic credit history than those whose accounts are relatively untouched (meaning, some people have not yet had a chance to reveal themselves as the poor borrowers they actually are). Whatever the case, trust the data. Trust the historical statistics. If you are using these tools correctly, you will begin to realize that the peer to peer lending historical data is brilliant if only we give it space to be so.

Exception: When Not to Trust the Numbers

tennessee filterYou should not use a filter that does not return enough historical loans (this is why I kept circling it above). For instance, if we removed the Delinquencies setting from the last paragraph and check only the box for Tennessee in the State section, the ROI jumps up to 12.97%. However, only 39 historical loans match this filter, a number that is not statistically valid, meaning this filter does not have enough history and is not trustworthy.

TNROI

As a general rule, a good filter should contain at least 2000 historical loans, the more – the better. If you can get a filter to return at least 2000 historical loans, you can be more certain that future loans will act in a similar way.

Filtering Prosper

prosper-stats.comThis post spent the majority of its time focusing on developing a filter for Lending Club. However, all this information can also be used on Prosper.com. Prosper is just as good an avenue of investing, often offering even higher interest rates than those at Lending Club. Currently, my Prosper account is earning a higher return than my Lending Club account, as seen in my first portfolio update for 2013.

For Prosper, you can use statistical tools like Rocco’s site Prosper-Stats.com to discover a great filter before investing with this filter on Prosper. You can use the filter tool to get a good handle on what kinds of borrowers are more historically likely to pay you back on time, and use this info to improve your Prosper returns.

A Final Disclaimer: Inflated ROIs

Third party peer to peer lending tools like NickelSteamroller go a long way in helping us find better loans on the Lending Club platform. But the ROI on these sites is almost certainly inflated. We will explore this reality in a future post, but for now let’s take a moment to remember that our historical loan data is very young. Out of 120,000 issued loans on the NickelSteamroller tool, only about 10,000 have actually had the full three years to be fully paid off (or default). This means that 91% of the loan data is for loans still actively being repaid, and as a result should not be equally trusted.

The reason this artificially inflates the ROI is because the earliest months in a loan’s history are when the loan gives us the most interest. Peer to peer loans are amortized, meaning they give off less and less interest each year, so a historic pool of very young loans will give a prematurely high ROI. As the loan data continues to mature, I think we will see the overall 7.7% return drop, perhaps to 5-6%. So take these ROI numbers with a grain of salt; keep them in perspective. Peer to peer lending is an emerging asset class. It is risky because it is so new and unknown.

That said, I continue to have great confidence in peer to peer lending (see here). I think it will eventually prove itself trustworthy to the world, providing filter-folks like us with solid returns for decades to come.

Continue to part 3 – increasing our return by adding some low-grade loans.

Questions or comments? If you enjoyed this please Like or Tweet it below.

[image credit: TaxCredits.net "Money" CC-BY 2.0]

Comments

  1. Jonathan Ching says

    Thanks, Simon – this article is very helpful in understanding the principles underlying P2P lending platforms. My main question after reading is how (or whether) the P2P lending platforms verify a borrower’s employment history and income. Do lenders simply “trust” that the borrower is self-reporting truthfully or is there a verification process?

    • says

      Hi Jonathan, the platforms largely trust the borrower’s statements. That said, some borrowers are asked to verify their income using W2s or paystubs (as seen in loans like this), but many are not. There is debate about whether investing in those with verified income is actually even profitable, since it can be required only when a borrower’s application is questionable. Is verified income a negative mark? Debatable. I personally don’t care about whether my borrower’s income or employment is 100% accurate, since I invest statistically. Borrowers who say they have five years of employment historically have lower defaults than those who say they have two years. So I invest in the five years. Hope this helps, Simon

  2. Jill says

    Hello. Thanks for info! I can’t seem to get anything on Nickel Steamroller ressembling the ‘Return Forecaster’. Nor can I do it on Lending Stats either. Can you help? Have they taken away this feature?

  3. Tabby Cat says

    What the %#€?! have they done to Nickel Steamroller?
    Looks like they’ve turned the most valuable P2P research site into a giant paperweight. What happened to the zillions of past loans that used to be accessible for research? The new Returns Forecaster’s abominable interface (no defaults, unclear) never seems to return more than a few thousand loans. I can’t do anything with numbers like that.

    Have you been there lately?

Leave A Reply