Its time to update my analysis of Prosper's late rates.

I follow the rate at which loans become "late" in order to estimate the fraction of prosper loans that will default. Toward this end I follow the number of loans that have become "1+ month late". Loans that are late a shorter time often convert back to current, but Prosper loans that have become 1 month late seem to stick in the late mode, and mostly head toward default.

Loans go late over time, therefore it is essential in the early life of a portfolio to track late statistics over time. In the early months, the late rates vs time will look nearly like a straight line. The slope of that line is the rate at which lates occur. If we know the probability that a late will become a default, we can multiply the late rate by this probability and get an estimate for the default rate of the portfolio. You know from reading my other thread http://forums.prosper.com/index.php?showtopic=6523 that only about 5% of loans that become 1 month late are being cured by the collection agencies. Therefore, something like 95% of these lates will become defaults. If the collection agencies were doing a better job, the math would be more complicated, but as it is, almost everything that goes 1 month late defaults, so complicated math isn't necessary.

Many people have been looking at late rates for the total prosper portfolio. This is usually misleading. The prosper portfolio evolves over time. New loans are always being added, and of course brand new loans aren't late yet. This distorts how lates are evolving over time. To solve this problem, I have divided the prosper portfolio into separate portfolios for each month of loan origination. The contents of these portfolios is fixed, so the evolution of the late rate over time tells us simply how that fixed set of loans is doing.

The late rates are ratios. The denominator is the total number of loans in the portfolio (which prosper calls "loans originated"). The numerator is the number of loans in the portfolio which are 1+ months late.

Here's the evolution of late rates vs time. Each curve represents one month of prosper originated loans. The dots are data points observed on the 1st and 15th of each month. The data starts at 60 days past the beginning of the month, because that's the first time you could possibly have any "1+ months late". You can see that while the curves are similar, some months are clearly worse than others. April was a bad month. May was a good month.
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosperlate-2006-11-01-a.gif)
If the Prosper portfolio were behaving as the Experian data would suggest, then we would see about an 8% annualized late rate for the entire Prosper portfolio (weighted average of the Experian annualized default rates per grade).

But you can see that these portfolios are hitting 8% just a few months after the loans originate. This is the first clue that something is wrong.

To make this clearer, I have slid all these curves to the left so that they start at the same point. Here you can clearly see that April is the worst month and May is the best.
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosperlate-2006-11-01-b.gif)
Remember the slope of these curves is the rate at which lates are accumulating. I've added two grey dotted lines to show the range of slopes that contain most of this data. The slopes of those grey lines are 0.055%/day and 0.110%/day. I have also added an orange dotted line which shows the slope one would expect if the portfolio were to behave like the Experian data. The slope of the orange line is 0.23%/day.

Do you see a problem? All the data is way higher than the Experian prediction. The lower grey line is 2.5x the Experian numbers and the higher grey line is 5x the experian numbers!

I know that many of you haven't been thinking in terms of averages over the entire prosper portfolio. Lenders have been thinking about the individual credit grades in which they invest. So to help you see what this means to you, I have broken out the credit grades for a couple of months. I chose April and May. I didn't want to use very recent months, because there isn't yet much data on them. April and May happen to be the worst and best months so far, so you get to see the range.

This next chart is just like the prior charts, except that I have plotted April'06 loans only, and I have broken them out by credit grade.
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosperlate-2006-11-01-c.gif)
You can see that enough data has accumulated so that the credit grades are nicely spaced out in the expected order. AA is sitting down at the bottom at zero, and HR is up at the top. Ok, that makes sense. Now lets think about the numbers for a moment. In just 5 months, 36% of the HR loans have become >1 month late! This is huge. Remember, the collection agency statistics tell us that 95% or so of these will default. 95% is so close to 100% that its not worth multiplying by 0.95 all the time. You can pretty much just think of that late rate as a proxy for the default rate you will be seeing soon.

Now you don't have to be a mathematical genius to see that if you were expecting a 19.1% ANNUAL default rate on HR loans, and you were up to 36% after just five months that something is seriously amiss.

I said "5 months" had gone by. What I mean by that is that this data includes five points that have the possibility of being nonzero. Those are triggered by five payment events. Late events and default events are triggered by missing these payment events, so from the viewpoint of events that could trigger "1+ month late", this portfolio is 5 months old. I will extrapolate the 5 month statistics into annualized late rates below. That 35.90% late for the HRs becomes 65.6% after its annualized, which is a lot more than Experian's 19.1%. Egad.

Now lets look at the best month so far
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosperlate-2006-11-01-d.gif)
Here the data isn't quite as tidy. The AAs are doin' great, and the HRs are worst, but in between things are jumbled up. My guess is that they'll sort themselves out as time goes on. Once again, look at the HR late rate. It's 20.31% after only 4 months. Extrapolating that to annual gives us an annualized late rate of 49.4%. Lower than April, but still far in excess of Experian's 19.1% number.

Here's a table of annualized late rates for each credit grade for loans originating in these two months. Feel free to make comparisons to the Experian statistics.

Normally we would have to do some work to convert these late rates into default rates, but because the roll from "1 month late" to default seems to be 95% or so, you might as well consider these to be predicted default rates. On the other hand, if you think collections will improve, feel free to multiply by the number of your choice. For example, if you think collections will get twice as good, and prevent 10% from defaulting, then you could mulitiply the above numbers by 90%. Hardly changes things much.

Conclusion: Loans are going late at rates which extrapolate to defaults at 2.5x to 5x the Experian historical figures. All extrapolation is speculative, but I feel confident in saying at this point that Prosper's default rates will be more than 2x the Experian numbers. If this is correct, a great many lenders will lose money.

thank you for doing the work involved with your analysis

all I can say is, "gulp..." :o

Wouldn't one expect most of the bad loans to show immediately (in the first 6 months)? While the next 30 months be rather quiet? FWIW I'd feel lucky to break even at this point.

Great analysis work! Not the results that folks will WANT to see, but results that people NEED to see.

There's only one aspect of what you presented that I'd question, and that is when you extrapolate from the 5 month late rates to annualized late rates. Are you comfortable with your method for doing this?

As I look at your charts, it appears that at most credit grades the lines are starting to flatten out near the 5th month. Put another way, it appears that the percentage of loans that will go 1+ month late is higher in the earlier months, and lower as time goes by. That seems to run counter to the straight-line extrapolation that you used.

But I could be missing something (it wouldn't be the first time! :) ).

Thanks!

I agree, but I also think that January/February (holiday bills) and April will have more lates/defaults every year, regardless of how old the loan is.

Pninen-

thanks for developing this information. Pretty soon I'll have to give up my "it's got to get better next month" position.

Its possible. We really don't know.

I have found default curves in books, but they are always specific to the particular kind of loans that the author is talking about. One book discussing consumer loans had a peak at month 5. Another book (discussing mortgages) showed a peak in the 5th year. Clearly different loan portfolios behave differently. Every loan company sees different behavior, because they serve different communities, and implement different lending policies, etc. We don't know yet how prosper loans will evolve. Given all the things that are different about prosper loans, we're likely in uncharted territory.

Clearly during the first 5 months the late rate has been very steady. Those curves (my first chart above) look almost like straight lines! So we haven't seen any peak behavior yet. As we watch those curves for the next few months we will learn how quickly they bend toward horizontal. Just looking at the existing data I can't say that they are bending at all.

Pretty thorough, pninen. Thanks for the effort!

:o

This post has been edited by LoanChimp on Nov-2-2006 02:03 PM

If I remember correctly, April was the month that lenders were "social lending," rate-cap lending, giving HRs listing advice...

And May was the first month with loans coming late and expanded credit data available, and people freaked out.

My suggestion being that these results stink because lenders are being too generous with credit.

We have met the enemy, and he is us.

Prediciting is difficult, especially the future.

Given the lack of data on which to base a complex model, I chose the simplest possible model: flat (ie constant with time) default rates. I think doing anything else at this point would be just a way to get ones self into a bunch of muck.

Edited to add: The annualized numbers (extrapolations) do seem unbelievable and shocking. However, they are what results from an honest attempt to employ the existing data with a straightforwad model. So they have a foundation. They're not wild random guesses. They do give us a clue what is happening. While I'm not ready to tell you exactly what the default rates will be, I did predict that we're gonna end up >2x Experian.

I do see that, but I don't think there's enough data behind it to make it statstically significant yet. When you break down by month and then by credit grade you end up with pretty small sets. About 100 loans in some of those buckets, so it only takes 1 or 2 loans moving to inflect the curve a visible amount up or down. In other words the data is "noisy". I see inflections down but I also see inflections up. As more months go by, and there are more lates, the data will become less noisy, and if there is a slowing we'll see it, and I'll do something different.

In the meantime, feel free to extrapolate any way you like!

Which for me begs the question, which lenders won't lose money?

Anyway, super super interesting info here.

This one will probably lose the least :ph34r:

LOL - true!

But more seriously, these guys (referendum, LoanChimp) are doing SOMEthing right. :) Or they're just lucky. :D

With the statistics showing increased default rates - much higher than I expected - how is Prosper going to remian in business 2 years down the road?

Is anyone concerned about the business model?

it seems to be a great idea, but with defualts exceeding Experion anticipated rates Prosper must be getting concerned (as should the lenders).

Mark my prediction: Referendum is going to end up losing money. You have to consider the age of the loans, not just the number of lates. As his portfolio ages, he will have plenty of lates.

Sounds like you're saying all lenders will lose money...?

Hey! 97 cents may be chump change to you but if that loan defaults I'm screwed. :(

No. I think Referendum is going to lose money. I believe this is the guy who started recently bidding $50 each on a huge number of loans, accepting the market rate. If one dives into Prosper like this, having no personal standards, then you end up owning the prosper portfolio, and you are going to lose money. The fact that he doesn't have lates YET is irrelevant. Its easy for a relatively new portfolio to have no lates.

Personally, I still hope to come out positive. ;)

Just so long as your load doesn't default ..

I've got two kids, that's enough for me. I'd be fine if my load defaults. Or at least goes into delinquency.

Great work pninen,

However lets keep in mind that this data shows +1 month late loans, the experian data is for defaults. If you did the same graphs with +3 month lates then we can see where we are compared to experian. Currently its a comparison of Apples to Oranges...

Ummm, are you going to tell us why?
Or just leave us all in suspense. :D B)

They aren't the same, but they are related.

To map from "1+ month late" rates to default rates we need to know the probability (or "roll rate") for a "1+ month late" converting into a default.

Loans are sent to a collection agency when they go 1 month late, and prosper provides us with statistics on how the collection agencies are doing. We can therefore see what fraction of loans the agencies are curing. That's running about 5%. (see my thread on that subject) This means that about 95% won't be cured, and will default. 95% is close to 100%. I wish it were different, but those are the facts. We need better collection agency performance, but we're not getting it. Maybe its impossible. I don't know. What I do know is the present numbers.

You might argue that the loans in those collection agency statistics are still in the agency, and haven't defaulted yet, so I don't know that 95% of them are going to default. Look at the by-month data. No agency has ever cured a prosper loan in the 3rd month! Almost none in the 2nd month. Almost all the curing takes place in the 1st month. So we pretty much can see what's going to happen.

So, if 95% is the roll rate from "1+" to default, then all you need to do is multiply the "1+" rate by 95% to get the default rate, or conversely to convert a default rate to a "1+" rate, just divide by 95%. I did divide the Experian weighted average by 95% when I computed the slope of the yellow dotted line in the first graph. Therefore it is comparable to the "1+" late rates, to the extent that the collection agency data is truthful and predictive.

You are right--but you got to read between the lines a bit. Are those 1+month lates really going to cure? These unofficial defaults will be official in the coming months.

Pninen,
First off, thanks for all the work you did. Fascinating stuff. What I hope people take from this is that HR's need to really be looked at like a bomb that could explode at any time. Thus, I believe that over time the HR default slope will be looking uglier and uglier. I'm hopefull that the B,C,D borrowers will stabilize, and it looks like their slopes might be showing some signs of that. Concerning my portfolio, while I do have a ridiculous amount of small 50 dollar loans out, I don't randomly take all B,C,or D borrowers, though I concentrate my efforts on looking there. over time I will likely be putting 100 or more on a smaller sample. Still, the overall sample all must pass my basic criteria, which include a valid reason for getting the debt in the first place (usually to pay off higher interest rate credit cards) and indications of income stability. You'll notice a scarcity of business loans in my portfolio, and those few business loans that I've participated in were all to people who had other income sources. Also, good luck finding a "Help me, I'm desperate" story among my 221 loans. Also, about 40 loans are from June, another 50 or so from July, etc. etc. so the age of my loans aren't weighted toward just the last one or two months. Anyway, lets see how the default wave hits, I am expecting one and hope to still achieve my overall target of very high single digits real rate of return. Right now I'm at about 180 having made at least one payment (my other 41 are too new), and of them one is five days late, another is late but with money on the way (I love the sight of daggers in the morning) the rest current.

I like seeing all the numbers in one place:

If we don't see the 1+mo late curve flatten soon, as in very sharply and very soon, between the very high 1+mo late rates projected to default rates, and the over-simplified math Prosper is using to help borrowers estimate default-adjusted returns, it looks like no one would make money. I sure hope we are misinterpreting the numbers because, if we aren't, the "A" borrowers in May are doing worse than what Experian's annualized numbers predict for HR's, of all things, so this sure isn't good!

What I am hoping is that by using much tighter criteria than just the credit grade, we can get better results.

Don't forget that many of these loans probably would have never pass bank requirements (e.g., too high D/I, or no income, or for HR just too poor of a credit score) and thus are outside of Experian's sample from which they determined default rates for "bank card products".

So, pninen, it looks like I could very well have better returns form my postings than by my invested dollars. :o Maybe it is good I have more messages than dollars invested in Prosper. :(

Another thing to remember is that the sample set for this analysis is the ENTIRE Prosper portfolio.

There are a lot of what I consider to be "unwise" loans that originated in the Prosper portfolio that would never be in my portfolio.

What would actually be more interesting than the stats on the entire Prosper portfolio would be the same stats for loans that meet certain specific criteria, such as originated loans that had 1 or less current DQs.

Wow. Enlightening. Thanks for going to all the work to put that together pninen.

I too believe I'll lose money on my loans made from March until the summer (maybe middle of the summer). I also believe that my methods of picking loans has improved, and hope that I'll start making money.

-arebelspy

Many of us have the same hope for our personal portfolios.

Yes. Cellardoor has repeatedly made this point. The Experian data do not represent the WHOLE of folks who fall into a credit grade, but that subset which the banks found acceptable.

The banks are making a selection. On the other hand, the Prosper lenders are also making a selection. Thus the set of loans in the Prosper portfolio is also a selected subset. (Remember, only 2% of the HR listings are getting funded on Prosper, so there is quite a bit of selection going on!) So we're each selecting a subset. The prosper lenders on the whole just aren't doing it very well.

Or maybe the banks are using some other ingredient that we don't have access to. Remember, there's lots of stuff professional lenders have that we don't have. Perhaps we need more info. Perhaps Prosper needs to do better qualification (checking, vetting, review, whatever you call it). I don't claim to know how to fix it.

Recalling when I got my mortgage and later when I got it refinanced at a lower rate, I recall the following:

• Verification of job stability (they called the community college's HR department to find out how long I worked there),
• Verification of income (signed permission to check with employer, copies of most recent W-2 statement, last three pay stubs),
• Verification of value of collateral (yes, and they even made me pay for the appraisal),
• Verification of insurance coverage of the collateral (they checked with the condo association to see if the insurance was at an appropriate level)
• Verification of lack of liens on the collateral (and, again, they had me pay for that title search and title insurance)
• Verification of assets (signed permission to check credit union assets, fund family assets, but they decided that would be enough assets so they didn't want to also wait for verification of 403(b) assets.
• Verification of liquid assets (last two months of bank accounts, but they saw enough there so they didn't want to wait for verification of my Savings Bonds portfolio)
• Detail examination of my credit reports from all three CRAs
• Clear indication that any recent significant deposits in my bank accounts was not an undeclared loan. (In the mortgage where I purchased the condo, I had Mother write a "gift letter" because she decided to exercise "estate reduction" in the form of a check at an inconvenient time as far as timing goes to get the mortgage; in the refinance I showed where the money disappeared from one account and appeared in the other.)

I don't think there is any secret information involved there, but for something like a credit card the credit union did check all three credit reports from CRAs (not just the summary of numbers) and the FICO credit scores, and they did impose a fairly high standard.

In comparison, what we have available from Prosper.com is very limited: just some numbers from just one CRA. :(

1. prosper should have enough vc capital to fund operations for a couple of years.
2. if prosper becomes unprofitable for enough lenders, then the market rates should stabilize upwards
3. many of loans are defaulting because they should never have been made to begin with. The market will force lenders to become much more selective.

Even now, I see a lot of very high risk loans being made.


4. some lenders will adapt and will be able to profit from this.

amen to above. growing pains...too early to tell a wave from a rock.

Hey pninen, I believe you are incorrect regarding your stats.

So I propose a simple little game for you. (Assuming you will be in SF in mid Feb).

Post your est del numbers as of 1/31/07 in the next few days. If you are any where close to that number, I will be happy to buy you a beer.

If you are way off on your numbers, then you buy me a beer.

(No fancy beer, whatever they have on tap!). :lol:

I've known for two weeks that I'm scraping on a rock.

On my loans originating between 10 August and 2 October I have 7 lates (defaults) totalling $458 in loan value. Yet I have received only $230 interest. To break even I need another $198, and I cannot see where that is going to come from.

November 3 will be very interesting for me, and I can still hope that it won't be more disheartening. I have another 26 loans originating on 3 October from which I should see another $20 interest, but when I think about the new lates I have to shudder!

Incidentally, my lates average B-, while my portfolio averages C-. If I am going to be more selective, maybe I should avoid AAs, As, and Bs to upgrade my total portfolio?

Earlier in this thread there was discussion on how to estimate defaults from lates. A conversion of 95% was suggested by pninen for month+ rates to default. Prosper is using lower numbers in its group default estimation for non-HR loans. See near bottom of the group help ratings page: https://www.prosper.com/public/help/topics/...ps-ratings.aspx

Thanks for your reply zarp. I hope you're correct in trusting the VC patience.

I agree that lenders will become more selective before bidding on HR and NC loan requests. You would begin to think that at that point in time when fewer and fewer take the risk that those people who claim to be fixing their credit rating will be SOL once again.

It's sort of a self fulfilling prophecy. They are high risk for a reason.

We need to charge more interest to mitigate!

What a wonderful, thoughtful and carefully prepared analysis! Thank you.

I think Experian's numbers are correct, with the caveat that such numbers are based on traditional lending sources. Contrary to popular belief, some HR folks are financed (at outrageous, predatory rates) by subprime lending sources. For those able to be financed at such places, I'd venture to guess that only 19.1% would default on an annualized basis. Prosper, to me, is different in that some of the HR loans funded here would be laughed out of some subprime financing places. While there are a few (very few) HR folks with the ability and willingness to repay their loans here at Prosper, those folks are the exception and not the rule; and I'd venture to say that those HR folks paying their Prosper loans could have gotten funded at other venues too. Prosper, it seems, is really the place of last resort for those HRs who can't get funded ANY place. Quite frankly, when you scrape the bottom of the barrell you are bound to get more gunk.

As for the AA, A, B and C borrowers, who are quite aware that they could get funding elsewhere, I'd bet their numbers will be more in line with experiean's predications on an annualized basis.

Oddly enough, the performance statistics show that the delinquency rates for the higher grades (AA, A & B ) are actually higher than predicted.

Perhaps many of the early adopters are not able to get decent rates from traditional institutions (i.e. actually worse than their score) and / or trying to scam the system.

Middle grades have pretty good performance.

Perhaps it is the way the prosper has marketed itself to America.

In another recent saga, this potential borrower had a completely wrong impression of what prosper is.

From her blog...

Perhaps borrows have the impression that prosper is a cash cow, ripe for milking...

:angry:

Aren't you an angel?

Pninen,

Thanks for sharing your analysis. It surprises me that more people aren't hacking the numbers this way. Of course, if someone is doing the analysis, it would make sense for them to keep it secret.

In any case, thanks for the data. I've done my own analysis, especially on the April data, like you did. I think that assuming a 95% is a safe bet. My own numbers for April say that 95.1% of loans that go one month late will go three or more months late.

But as your data suggests, there is a certain amount of leveling off happening, at least for April. And earlier months seem to be leveling off too. This suggests that "bad" loans are being weeded out sooner rather than later. Of course, it's too early to see how far that goes.

I've also tried to determine annualized rates of defaults, using somewhat different methods from yours. I'd be happy to talk about how I do this, if anyone were interested. A big difference is that to get an overall default rate I'm looking at statistics for loans that are three or more months late, and considering only the history of loans that have been around at least long enough to be three months late, not just April or May loans. So this is what I get:

Annualized default rate, all loans, at least four months old:

All grades: 10.6%
AA: 1.1%
A: 2.7%
B: 4.3%
C: 6.6%
D: 5.7%
E: 11.9%
HR: 30.4%
NC: 27.0%

So these are worse than Experian rates, but still manageable, I think, from a risk/reward perspective.

But I also did a similar analysis just for loans with zero current delinquencies, and looked at loans that are or could be 1+ months late, so as to get a larger sample size. The annualized default rate:

All grades: 3.2%
AA: 1.7%
A: 2.0%
B: 3.3%
C: 3.9%
D: 1.7%
E: 2.3%
HR: 15.0%
NC: 0.0%

This has made the case to me that lending at D and E credit grades can be profitable for those loans with no current DQ's. Don't trust that NC number though! There have only been three NC's with zero current DQ's, so there's definitely not enough data to make good judgements.

These numbers are not recommendations and shouldn't be considered as such. Just my own data, subject to change.

Rob

unfortunately for those borrowers, they will find my teats have mostly shriveled up and dried. nothing but a few sour drops of left to squeeze out.....

This doesn't look right.

Could you explain yoru methods?

Kudos to Pninen for all this work. I do think that the prediction of default rates (and the huge differences from Experian rates) will be borne out - unfortunately for all us lenders.

What Pninen's data tells me is that borrowers are not taking Propser seriously. For some reason, it appears that the borrowers take lenders on Prosper a lot less seriously than insitutional lenders - dashing my hopes that the Prosper model of peer to peer lending would be better for lenders than traditional lending.

I also need to vent about group leaders. I have restricted my loans to borrowers in groups in the hopes that peer pressure and the involvement of group leaders would make a difference. However, the group leaders appear to have little or no contact with borrowers after a loan has funded. When I check in with group leaders about a late loan, I am invariably told that they have e-mailed the borrower but have had no response. Heck, I can do that and achieve the same result!

I admit that my initial lending was reckless - quite a few HRs, etc. However, I don't think the loans that I have made after instituting personal standard have fared materially better.

I had high hopes for Prosper as a source of consistent income. While I now realize that the mattress might be a better place for my money, the good news for me is that I only experimented and made a small number of small loans. Current status:

10 of 16 loans current (ouch!)

Late:

<15 days - 1
Late - 2
1 month - 2 (#350 zoeology (Doc Prosper) and #201 Falcons777 (Clean Slate))
3+ months - 1 (#347 souledout1 (Professors & University Grads))

bad visual, but funny nonetheless... :lol:

The Prosper rate cap must be raised for rational lenders to continue funding loans with the high risk levels pninen has outlined. Prosper please do it soon!

/right after you start paying interest on the standing funds I can't invest b/c listings and rates are so incongruous.

You better write your state legislator ( <_< Mark!). Those rate caps are set by each state to prevent usury. Payday loans have a way around it, but Prosper does not.

Two points:

1. Congressmen have nothing to do with state rate caps. Congress deals with federal laws and the like. The people to write to for state rate caps are members of the state legislature (whatever it is called in your state).

2. The 29% cap that Prosper puts on all other states is self-imposed by Prosper and thus is up to Prosper's discretion to raise.