Print

[SLN]

I was catching up on some reading in Google Reader today and noticed this blog post from a couple of days ago on Prosper's blog. Looks like it has been deleted already because there is no sign of it any more. Seems like Prosper is looking to go on the offensive against Lending Club. Comments anyone?

This is the second in a series of posts about risk management. See the previous installment here.

In part two of our series on risk management, we take a deeper look at the two biggest players in the peer-to-peer lending industry. Both companies employ models reliant on internal risk teams to fairly balance interest rates with strong returns for lenders.

Over the last couple of years, Prosper has assembled the best risk team, models, and capabilities in the business, supported by the fact that we accurately matched our expectations of risk with actual performance, as noted in our last blog post.

In contrast, Lending Club’s losses are running as much as four times higher than their original expectations - seriously dampening their actual lender returns. See below for an example of a D-rated loan on Lending Club’s platform. Lending Club estimates that this loan and others like it will have an average annualized loss rate of 2.8% (see Footnote #1 below to see how to find these numbers):



Because the actual losses on this type of loan are really closer to 10% (per Lending Club’s own numbers), actual annual returns are likely to be under 5%, or not quite half of what was predicted. (For more information about the relationship between cumulative and annualized loss rates, see Footnote #2 below.)

Compare that with our ‘D’ rated loans where we forecasted an 11.9% loss rate that actually came in at 10.5%:



Also of note is how close Lending Club’s actual losses on D-rated loans are to our actual losses on similar loans. The issue isn’t that these loans shouldn’t be issued - these are reasonable loss rates in consumer credit. The real issue is that it underscores the importance of accurate predictive risk models. At the end of the day, our losses were about 12% under expectation, while Lending Club’s were off by more than 300% – and in the wrong direction.

We feel that customers from both companies need to have access to all the information they can in order to make informed decisions. That’s why we have always made our data widely available. We also want to help you understand those numbers so that you can make informed decisions about your money.

Notes:

1.   To see the Expected Default Rate for a particular sub-grade at Lending Club, you’ll need to create an Order for a single Note of that sub-grade. Once you have the order created, proceed to the View Order page:



We’ve circled the Default Rate in red. It’s a bit onerous to aggregate these one at a time, so we’ve done it for you in the Excel attachment to our previous post. You can find the figures on the “LC Expectation Calcs” tab.

As an aside, we also thought it was interesting that Lending Club’s Expected Default Rate “is an annualized rate based on industry-wide performance data collected by TransUnion”. We found that out by clicking on the 3% link from our order:



Given the unique nature of peer-to-peer lending and its imperfect correlation with industry-wide measures, we suspect this is one of the primary reasons that their loss expectations have not been very accurate to date.

2.   The relationship between cumulative principal losses and annualized default rates can be a bit confusing. In the attachment to our previous post, we provided a breakdown of how to visualize the cumulative loss curves forecast for a given annualized loss rate, but we wanted to provide some more context on that.

For a given set of loans the only way to have a perfect annualized rate is to wait until all the loans have completed – either by paying in full or defaulting. If we waited that long, though, we would be ignoring a wealth of useful information that the loans provide as they age. One of the most effective tools that we have is the cumulative loss chart: on the x-axis we have the age of the loans (i.e., the number of payments or “cycles” that have come due so far); on the y-axis we have the total principal losses as a percentage of what was originally lent. Using actual losses as of a certain cycle, we can extrapolate an estimated annualized default rate.

Because of the varying effect of individual defaults, we would expect Lending Club’s “D”-notes, with a 2.82% weighted average expected default rate, to experience 3.50% principal losses in aggregate over their 36-month life – and 1.62% of that would occur within the first 12 months! So when we see that actual losses within the first 12 months are closer to 6%, we can estimate that the annualized loss rate – when all is said and done and the loans are all paid or defaulted – will be closer to 10% than 3%.

Because the loss curve for a particular vintage may be somewhat skewed in either direction, it’s important to remember that this conversion is only an estimate. The actual annualized loss rate actually may be closer to 9% or 11%. That said, with losses over expectation by 300% or more after a year of data, it’s safe to say that there is a problem.

Again, if you’re interested in the math, we encourage you to check out the data appendix on our previous blog post – there’s lots of detailed commentary there as well as the math that we’re doing to evaluate these relationships. And if you have questions, we encourage you to submit them!

[Cushie]

Thank you for posting this.  I'd say more but I'm laughing too hard at Prospers' attempts at making themselves look good.

[ira01]

Over the last couple of years, Prosper has assembled the best risk team, models, and capabilities in the business, supported by the fact that we accurately matched our expectations of risk with actual performance,

Bwahahaha -- maybe someone ought to remind Prosper of their own Experian default expectations and the guidance based on that they gave lenders regarding appropriate interest rates to bid (using an incorrect mathematical formula to boot).

In contrast, Lending Club’s losses are running as much as four times higher than their original expectations

Well, let's see -- IIRC, weren't the Experian expected default numbers somewhere around 5% on average across the credit grades (did anyone save those numbers)?  Yet actual Prosper performance was around 40% of Prosper 1.0 loans defaulting.  So Prosper was off what, about eight times?

[Cushie]

Now I'm not real good with stuff like this (as a caveat) but is Prosper boasting that they correctly guessed that they'd have bigger losses than Lendingclub and are, in fact, pleased that they guessed correctly?  Am I reading this wrong?

Also, during 13 or so days of the period they've chosen to highlight Prosper was in the quiet period and not producing any loans.  The quiet period ended July 13, 09 - coincidentally, when Prosper.com decided to pull from for stats. 

http://blog.prosper.com/2009/07/13/welcome-back-lenders/

[SLN]

Now I'm not real good with stuff like this (as a caveat) but is Prosper boasting that they correctly guessed that they'd have bigger losses than Lendingclub and are, in fact, pleased that they guessed correctly?  Am I reading this wrong?

That is how I read it, too. The thing they seem most proud of is the fact that they predicted their higher expected loss rate more accurately. When looking at "expected lender return" they claim theirs will be triple Lending Club, but it is interesting they didn't emphasize that number.

My guess is that someone at Lending Club had a quiet word with Prosper and informed them of some mistakes in their rationale - hence the post has been pulled. Just a guess....

[Nora_Lenderbee]

Over the last couple of years, Prosper has assembled the best risk team, models, and capabilities in the business, supported by the fact that we accurately matched our expectations of risk with actual performance, as noted in our last blog post.

This is a complete and total lie. Their risk predictions were wildly off. 40% of 1.0 loans defaulted. Most lenders' returns were negative.

In contrast, Lending Club’s losses are running as much as four times higher than their original expectations - seriously dampening their actual lender returns.

Dunno if this is true--but even if it is, four times is still WAY better than Prosper did.

I couldn't read any further because I was laughing too hard.

[Nora_Lenderbee]

Lobby?

[Urbi_et_Orbi]

Lobby?

Second.

[dlv1945]

Those images may vanish soon. Anybody saving them?

[Urbi_et_Orbi]

Never a good idea to simply link to an image hosted by Prosper.  Download, place them on a free image host and re-post.

[Xenon481]

Here are the images hosted at a site other than Prosper.

[Xenon481]

My guess is that someone at Lending Club had a quiet word with Prosper and informed them of some mistakes in their rationale - hence the post has been pulled. Just a guess....

I suspect that Prosper's blog post got pulled because they put up a screenshot of Lending Club's website without approval from Lending Club.

[TotoMMB]

My guess is that someone at Lending Club had a quiet word with Prosper and informed them of some mistakes in their rationale - hence the post has been pulled. Just a guess....

I suspect that Prosper's blog post got pulled because they put up a screenshot of Lending Club's website without approval from Lending Club.

I'm guessing someone got super giddy with their findings and couldn't wait to post...legal ramification be damned.

[SLN]

Never a good idea to simply link to an image hosted by Prosper.  Download, place them on a free image host and re-post.

Urbi et al, I have updated my original post to link to the image URL's posted by Xenon.

[Cushie]

Thanks SLN!