Prosper Lending Resources

Notes on Prosper Lending: Interest Rate & Bidding Guidance

2007-12-31T07:00:00.000-08:00

Prosper has recently done something *wonderful* for current and future Prosper Lenders. But I bet you don't know what it is - or, if you do, why it's so unbelievably important.

Since it's garnered very little attention in the P2P Lending Community thus far, I wanted to shine a little bit of light on it here, because, quite frankly, it's a monumental step - both for Prosper Marketplace *and* for the evolution of the entire Peer to Peer Lending space. Ok, alright already... enough pre-amble.

Let's hear it:

Since Prosper's launch in February 2006, lenders were presented with Experian default projections as baseline guidance for predicting default rates for each credit grade. Judging by the bidding behavior of many lenders, these default projections were trusted, to a large degree, as being both accurate and applicable. I won't go into why they turned out to be so grossly inapplicable to Prosper loans (or online Peer to Peer loans in general) - since I covered it a long time ago, and since I'm so happy to see them go! - but if you're interested in the history behind the story, you can see one of my old posts on Default Projections.)

Regardless, they *are* history! Gone. Finito. Outta here. And none to soon.

On October 29th, 2007, Prosper released a site update which leverages historical Prosper Marketplace performance data to present lenders with *enormously* improved bidding guidance. And not only that, but they chose to display it directly on the bid entry screen. My heart goes pitter patter for all the lenders who will be saved time, money, and a considerable amount of frustration by getting a much more accurate picture of what rates they *ought (more) rationally* to be bidding.

Here is the (understated) announcement of that release:

Improved bidding guidance for lenders

As the Prosper portfolio has grown and matured, we finally have a volume of loan payment activity that has allowed us to replace the Experian historical default data with Prosper’s own estimated default data.

We have also performed some analysis on the differences between borrowers within the same credit grade, and segmented the borrower population into 54 unique segments (one segment, for example, is C-grade borrowers with no automatic funding, 0 now delinquent accounts, and 2 or more inquiries). Based on which segment the borrower belongs to, we will display the estimated loss (due to default), rate adjustment (uncollected interest and fees), and annual servicing fee, and come up with an estimated return for loans to borrowers of that type.

Here’s an example of the new bid input for the example borrower:

What the above translates to in practical language is that each time a Prosper Lender places a bid, they will now see a detailed breakdown of the past performance of Prosper loans with similar credit characteristics. Not only that, but Prosper also breaks down the estimated default loss percentage, adjusts properly for interest loss and service fees, and provides a properly calculated Estimated Return figure - all based on the past performance of *actual loans made in the Prosper Marketplace*.

Read: REAL GUIDANCE.

It took Prosper a year and a half to implement it, but, it's important to understand that you can't roll out a feature like this in the absence of data to base it upon. It simply took a while for the first Prosper loans to age enough for data to be available - and getting this feature out, this soon, and available to every lender, is a MAJOR coup.

At nearly the same time, on the borrower side, Prosper began providing initial interest rate guidance to borrowers, also based on historical marketplace performance data.

So, let's look at a couple of charts.

First, loans by credit grade per month (chart courtesy of Prospers.org Statistics Wiki):

Are those the early signs of a sharp decrease in low quality loans being funded? And, if so, could there be a related increase in interest rates for loans that *do* fund?

Let's take a look at the rates of funded loans by grade (courtesy: EricsCC.com):

The sharp up-trends in interest rates for funded loans, especially those of lower grades, tell the story quite well.

It might be too little or too late to save some of the early-adopters from lending pain, but you really do have to give Prosper the credit they deserve for this one.

From now on, we might still be flying partially blind - but at least now we've got a map!

(We now return you to your regularly scheduled Prosper-drama, already in progress.)

Notes on Prosper Lending: Reinvestment Risk

2007-03-29T00:05:00.000-07:00

One of the most often misunderstood issues I see popping up on the official forums with some frequency is that of reinvestment risk. To be absolutely clear, the kind of "reinvestment risk" that I'm referring to here is that which is taken on when good borrowers make their payments or pay off their Prosper loans (remember that there are no prepayment penalties) - and lenders are, naturally, put into the position of having to attempt to re-deploy that money into a new loan (or new loans).

As Prosper offers no interest on deposited funds at this time, having funds repaid puts the lender in the situation of earning zero return on those dollars (and having to find new loans to fund). This causes a (usually small) period of dead money, while lenders wait to find a new listing that they feel comfortable bidding on (or wait for their SO to do the same).

In addition to the selection time involved in finding a new listing, one must also wait some number of days for that new listing to close. So, now you're looking at 2 (business) days of undeployed funds while your borrowers payment (or payoff) clears, plus X amount of time for selection of a new listing to invest in, plus X number of days for that listing to close, plus X number of times you have to repeat this process in the event that you are outbid from your selection(s) or in the event that your selections do not fully fund.

Add to this another 1-7 days (usually somewhere around 5 days, on average, currently) for Prosper to verify the borrower's details (post-auction) and actually originate the loan. You don't make any money during this downtime either - and there's no guarantee that the borrower will pass verification. In the event that they don't, you're back to step 1 again.

In my opinion, the above issues (aside from the issue of Prosper not providing some sort of investment vehicle which offers interest on unlent funds - the desirability of which they ARE painfully aware) are marginal, in the grand scheme of things, and not really worth worrying about given the likely small amounts of both money and time involved. However... there are some follow-on issues which I do believe are worth giving some serious consideration:

Defaults occur unevenly in time. That is to say: Defaults occur at different points during the lifespan of any particular loan. If a loan defaults in month 30, you will suffer less of a loss (maybe even still managing to make a small profit, depending on rate) than if a loan defaults in month 3.

Since defaults occur unevenly in time, it is instructively necessary to plot aggregate loan performance data in order to get some rough idea of the default behavior we should expect over the life of the average 36-month Prosper loan.

Plotting this data yields us a "default curve". The default curve is simply a visual representation of the behavior (over time) of Prosper loans. We look at the totality of the loans made in the marketplace (though you can certainly be more specific, if you'd like) and say "x loans defaulted in month one, y loans defaulted in month two, etc", all the way up to 36 months, and then present this information in "bar graph" format. We then fit a single line to connect each peak on the graph in order to form a curve. Basic stuff.

You may find it helpful to think of this curve as something like a "roller-coaster". Every new dollar (and every reinvested dollar) that you invest in a Prosper loan can be thought to follow this curve, from origination at month 0 to payoff at month 36.

Keep in mind, though, that there's no "safety restraint" holding individuals borrowers to their seats: borrowers may pay early or pay off their loan entirely, at any point and with no penalty (ie, they may jump out of their "car" at any point during the ride). They may also default at any point in time.

When reinvestment is necessary, either because of borrowers making their normal payments or paying off early, the dollars that you have invested with them (that they have just repaid) are sent to get back in "line" and to be reinvested into a new loan ("to go for another ride on the Default Curve Express").

Why does that matter? It matters because, just as with a roller coaster, some parts of the default curve are more dangerous (or "exciting") than others. We only have approximately six months worth of viable data at this point, but there is already a clear initial incline (just as many roller coasters start by pulling you up to some height) - an incline which, it is hoped, will be followed by a distinct drop and levelling off (of defaults).

Let me say that in another way: If, for example (and I'm not saying this is the case, it's still too early to tell), the first six months of a loan are when it is most at risk for defaulting, then, each time you have to reinvest repayments into new loans, you are forcing those dollars not only back "into line" (and into the delays mentioned above), but also, once the new loan originates, those dollars will be travelling the riskier initial parts of the default curve. (ie, your dollars were tooling along in the virtual flats, post loop-de-loop, of the latter parts of the ride, coming safely back into the station - but, a few bucks decided to get off early, and now they get to start over and do the "fun" parts again).

When you take this default curve behavior into account, you are able to approximate the risk differential (or expected return) between a loan aged, say, 3 months (still travelling the dangerous first parts of the ride - with plenty of track left to cover), and a loan aged, say, 30 months (that has already successfully passed most of the risky parts of the curve - and therefore presents less of a risk).

Quantifying this risk is difficult, due to the lack of historical data. But, it is a safe bet that there will be parts of the curve that are more risky than others - perhaps substantially so.

Though we can't extrapolate out very far from the data that we do currently have, there are, at least, some hypothetical assumptions (probable psychological effects) that may come into play and give us some idea of what to expect (such as a possible slope in default behavior toward the latter part of the life of a loan, perhaps driven by something like a "no sense ruining your credit now that you've only got six payments left to make" or an "almost-there" effect). That kind of thing is exceptionally hand-wavy and uncertain, but it does underline the uncharted-ness of the territory, and the lengths one has to go to in order to "guess".

The indisputable, and non-hand-wavy, fact, though, is that subjecting repaid funds to another go on the 36-month default curve roller coaster incurs an increased level of risk for those funds. We might not know exactly what it is yet, but it's certainly there...

The fear that follows from this uncertainty is thus: That good borrowers will make their payments on time or pay off early, and that, since the "good" borrowers return funds to us that must be reinvested (and since the bad borrowers do not return some percentage of funds to us), that there may be an amplification of negative effects. Since our money is so often made to travel the riskier parts of the curve - and, since it is therefore, much more often than we would like, exposed to new borrowers in new loans - it is therefore to exposed to the higher risk of default/loss. Couple that with the fact that at the beginning of a loan we have the most outstanding principal, "good money", at risk (exposed to the entirety of the "risk track") and you begin to see why this may become a non-negligible issue in the long-term (and one that should certainly be watched in the short term).

In particular, if loans that will default will also tend to default early in the life of the loan, then we may find ourselves open to losing larger (initial) amounts of principal at an increased frequency (increased, at least, from the frequency that one might think if one considered only "default rate" in their risk calculations).

Prosper lender "pninen" graciously maintains (and updates, usually bi-weekly) a blog at the unofficial Prosper forums with charts of the current aggregate default curve.

Notes on Prosper Lending: ScoreX Plus

2007-03-27T13:24:00.000-07:00

Prosper.com uses a proprietary & predictive credit scoring model called Experian ScoreX PLUS(sm) to calculate borrower credit scores. Prosper then assigns a letter grade based upon specific score ranges.

While the score ranges may look similar to FICO(tm) scores, they are not FICO(tm) scores.

As ScoreX is a predictive model, the credit grade that borrowers are assigned by Prosper (based upon the numeric ScoreX score) may differ considerably (sometimes better than you might think appropriate - sometimes worse than you might think appropriate) from the credit grade that would be assigned if a more traditional FICO(tm) or FICO-like scoring product were employed.

ScoreX is built for assessing risk for new accounts on different types of existing credit products. Prosper loans are a new credit product, and one which has little in common with several of the types of data included in the sample that the ScoreX model uses for prediction.

Of the types of credit products that the ScoreX sample consists of (auto finance loans, mortgages, home equity lines, and credit cards), many are credit products that are secured by real property - whereas Prosper loans are unsecured.

The same caveats about "credit qualification" that apply to the Experian default projections also apply to the ScoreX model: The sample consists of the performance of borrowers who were approved for new accounts on traditional banking products. Many Prosper borrowers in the mid to lower credit grades would not qualify for traditional bank or credit card products, and so correlation (and therefore predictive ability) seems to decline sharply as credit quality falls.

After close observation of ScoreX over the last year, it is at least clear that the majority of the same factors that would lower your traditional credit score also affect your ScoreX score - though perhaps not to the same degree. For example, DTI and utilization of available credit seem to be more heavily weighted in the ScoreX model, as does length of credit history.

Notes on Prosper Lending: Default Projections

2007-03-27T00:12:00.000-07:00

Prosper loans are an entirely new asset class.

There is, quite simply, no (truly) analogous (or available) existing historical performance data to look to for guidance - even when it comes to lending purely "by the numbers".

History is being made each day at Prosper.com - and the performance you see in your portfolio today will set the "true" (well, relatively!) baseline for this marketplace for the coming years. Until then, we're not exactly flying blind - but we are certainly flying in uncharted territory. I think that's more exciting than frightening - but not because I've shut my eyes to the realities of the marketplace:

Prosper's Experian default projections should probably be used as a very rough baseline reference only.

Keep in mind that Prosper presents projected default rates in an annualized format. That is, the default rates that are presented are for one year (12 months) - while Prosper loans are for a 3-year (36 month) term. (This is a huge point, and I hope you're reaching for a calculator or a spreadsheet if you didn't already realize this.)
Here's some mathiness to get you started:

[defaultrate] = projected default rate as a decimal (ie, the 6.2% default rate for D's would be expressed .062)

(1 - (1-[defaultrate])*(1-[defaultrate])(1-[defaultrate]))
= the 3-year default rate (expressed as a decimal)

If you've been doing: [lenderrate]-[defaultrate]=return,
don't raise your hand, but... don't keep doing it.

This will only get you so far, though, as defaults occur unevenly in time. While x percent of loans may be projected to default at some point during the 3 year term, there's a significant difference to your actual losses depending on at what point the default actually occurs (since you'll have received payments until that point).

Modelling that behavior is beyond the scope of this post, but Prosper lender "pninen" actively maintains (and updates, every 2 weeks or so) a blog on the unofficial forums (at prospers.org) with charts of the default curve that's shaping up.

The projected default rates presented have both an "average" and a "range". For example, while D grade borrowers (with less than 20% DTIs) may have a projected yearly default rate of 6.20%, this 6.20% average represents a range of 4.5-8.2%. As we're all unlikely to match Experian's sample size of 251,000 accounts, it's probably a good idea to expect some further anomalies here (especially due to the relative smallness of our individual loan portfolios/samples).

The Experian default projections do not seem to be very tightly correlated with Prosper loans (as an asset class). This is likely because the Experian projections are derived from a sample that included existing credit products that have characteristics (and credit qualifications) quite dissimilar from Prosper loans.

The Experian sample is based upon data from bank card (credit card) products. As such, the sample includes only the performance data of borrowers who qualified (based upon their credit histories) for these traditional credit products.

As you delve into the mid to low credit grades of Prosper loans, keep in mind that some (not marginal) subset of those borrowers would not qualify for the credit card products used to make the default projections (very, very roughly: the worse the borrower's credit, the less weight to give to the Experian projections).

Also of note that is that Experian itself warns that the projections go out the window for borrowers with Debt To Income ratios > 20% (though, in the actual Prosper marketplace data to date, DTI is rather loosely correlated with performance, on the whole).

This is by no means everything you need to consider when becoming a lender, but it's a damned good place to start!

More to come, as I find the time...