Per lending stats

395 lenders qualify for the bucket of
>20 loans
>6 months loan age
>$5000 invested

Only 11 of those 395 have estimated ROI >12% - thats 3% of the lender basket
Only about 30 of those 395 have estaimted ROI >10% - about 8%

So borrowing on Prosper at 8% to invest, again - thats a toughie. YOu have to be in the elite 3% for it to even begin to make any sense.

At least one graph is at Prospers.org - Data

Then add in taxes. It doesn't make sense, really. Only if you can get $$ cheap or free elsewhere to invest, but getting it on Prosper isn't the way to go.

-arebelspy

bump

Stats are not always what they appear to be. If you take a look at my portfolio, you will notice that I am averaging 9.69%. This is not because of poor investing on Prosper or poor choices, it is because I used this site to make a large and very low interest rate loan to my brother which lowers my overall average.
However, this site allows me to not worry about asking my brother for a loan payment and works great for that reason. I have to consider my ROI higher than the 9.69% when you deduct out a family member loan.

I wonder if there is a way to take out specific loans and see exactly what the ROI would be?

Another point to consider is that if the lenders have been lending on Prosper for over a year now, that was before the additional credit information was made available. Those loans have typically faired much worse because of poor information being given to the lenders at that earlier time table.

Here is the bimonthly update. New data points added for 4/1/07.
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosperlate-2007-04-01.gif)
This month it looks like Prosper must have bought back some more loans. The number of originated loans in Oct'06 went down from 743 to 742. The last time that happened it was caused by Prosper buying back loans, so I assume it had the same cause this time. Similarly Nov'06 went from 701 to 700, Dec'06 from 969 to 966, Jan'07 from 1020 to 1018. Didn't buy back yours yet? Ah well. Patience.

There was a nice article about Prosper in the Baltimore Sun. http://www.baltimoresun.com/business/yourm...home-columnists

This article quotes Chris Larsen... "Larsen says the default rate - payments 120 days past due - is about 0.50 percent. But that will likely climb as the three-year loans mature, he says."

Well, that is misleading. You can see from the chart above that about 20% of the Mar'06 loans are now >1 month delinquent. Although Mar'06 seems like a full year ago, remember that we are observing with a 2 month lag, because the first payment is due 1 month after origination, and you can't be 1 month late until another month has passed. So the stats for Mar'06 are actually 1 month shy of a year. Anyway, it looks like the loans that have gone bad (with "1 month late" as the definition of "bad") in the first year will be about 20% for the Mar'06 loans. Now we don't know exactly how many of those will go on to default, but the Prosper/Penncro collections performance stats give us a hint. It shows about 12% of loans that have gone 1 month late have recovered. If that means 88% of loans that go 1 month late are gonna default, then we can multiply the 20% at the tip of that Mar'06 curve by 88%, and estimate that about 17% of those Mar'06 loans will default. Now 17% is way different than 0.5%, which is why I find the quote misleading. (Of course I hope the collections guys will do much better than the historical statistics.)

Newspapers almost never get quotes right. (I know from personal experience.) Therefore, I don't blame Chris. However I do find the statement as printed misleading. It is correct to say that about 0.5% of all loans originated so far have already defaulted, so that might be what Chris actually said. However, I just don't think that's a meaningful number, as it mixes loans of all ages, including a large number of loans so young that they couldn't possibly have defaulted yet. Makes no sense to compute a ratio of defaults to things that could not have possibly defaulted. I urge clarity on this point.

Lendingstats recently removed loans so young that they could not possibly be late from the denominator when computing late rates, which helped improve his estimated ROI calculation. That's the same sort of thing.

Interesting how each month appears to be roughly linear. I would have expected them to begin to flatten slightly by now. If it isn't a trick of the eye, then that means the pool of non-late loans is going late at a faster rate each month s.t. the absolute number of loans going late in each cohort is roughly equal across units of time.

Wow. Will those paths ever deviate from a straight line??

I thought we should expect about a 5% default rate (based only on the cumulative data for all loans), but...wow.

Can you make a similar chart for 3+ month lates plus defaults?

~

PS: Prosper did repurchase 9 loans, and included their new status in the XML download. They seem to be the loans previously reported in the forums as possible identity theft.

http://forums.prosper.com/index.php?showtopic=19469


ETA: Wish you'd join us at prospers.org!

Yeh only Aug 06 and Apr 06 show any signs of flattening. That sucks.

Can you do these same charts by Grade?

i.e. have an April 2006 chart, with the 7 grades each as their own line

And so on and so forth each month?

Lots of work I know, just askin' :)

Was curious how these loans performed after 3 months of "life", in terms of how many were 1 month late and if over time the performance has gotten better
i.e. more data should make for better results

So if this is bimonthly than the 6th data point should be the 3 month mark

Mar 06: approx 9% of loans 1+ month late
Apr 06: 10%
May 06: 6%
Jun 06: 7.5%
Jul 06: 6.5%ish
Aug 06: 7.5%
Sep 06: 6.75%ish
Oct 06: 7%
Nov 06: 6.5%
Dec 06: not enough data

So not much improvement, basically after a bad start for the March and Apr 06 loans (which I think coincided with more credit reporting starting in Apr 06), May 06 through Nov 06 stats pretty similarly pattern a 6.5-7%ish 1+ month late payment expectation within 3 months.

December 06 could be the first to break the pattern, with 5 data sets its right around 3% so unless a big spike in next 2 weeks could show the first sign of improvement... now is than an anamoly or trend.

I wonder how this chart would look if rather than % of loans late, it used % loans by original loan size late. I'd assume all of the small, crappy, HRs are mucking up the stats. I need to get off my ass and download the Prosper data. :P

Instead of loan size loan grade would be better. BTW you would not have to download data. You could draw the chart using Prosper's Performance data by hand. It may take awhile to do though. For instance Aug 06 loans on Apr 2, 2007 would be as follows:

Grade:% Late - AA:0 A:12.82 B:4.41 C:9.08 D:12.33 E:23.3 HR:32.74

From performance search:
https://www.prosper.com/lend/performance.as...0&sn=&tg=0&vb=0

Man that credit Grade A was a killer that month. :o

Thanks for the great work pninen!

The one thing that is comforting is to note how consistent it all is. Without any major changes, it seems that we are safe to assume that the Apr 07 loans will preform just like the Apr 06 ones.

The other thing that pninen's data shows is that there isn't a high percentage of loans that don't pay the first month. That might indicate that fraud is not much of a problem (despite all we hear on the forums about people not making the first payment) - of course, it also means there is no reason to pat ourselves on the back when the first payment comes in ok...

Nice work pninen. Hope the charts continue to get updated on a regular basis. I've been lending for a few months and tho i only have a single late loan at this point I am concerned. My work and experience have exposed me to the "other side" culture wise. Unfortunately, there are those driven by purely selfish motives, not altruism. They care not of credit ratings or truthful disclosure. Others are naive in their personal finances, focused only on the short term - a loan gets them over the current crisis, but doesn't solve the underlying problem. Its absolutely wrong to assume that delinquencies will decline after a few payments are made.

Where a loan goes late after only a few months there is a real potential for fraud. In this area we should encourage Prosper to analyze these transactions and look to generate additional safeguards. If lenders are not successful ...... (I don't have to spell it out). Also, as with all things internet, there is the risk of organized attempts.

Finally, the current discussion in the news about defaults on subprime mortgages and rising adjustable rates is directly relevant here. We are entering a "challenging" part of the economic cycle. If I followed my own advice I'd start doing more "due diligence" and possibly reducing my exposure until Prosper proves it is mature and viable. Still, I love the concept.

delinquent
delinquencies
late loans
default
risk
statistics

The 4/15/2007 update...
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosperlate-2007-04-15.gif)
Read from the beginning of the thread for info on methodology, etc.

The ants go marching 12 by 12, hurrah...hurrah... <_<

The consistency is scary. Just move the line for each new month over one to the left, and it appears almost identical.

Great.......

I wonder what this will look like in each consecutive month for the earlier loans....

Maybe I should just open an Apple Bank account for this money.....

Its too soon, I suppose, for a Mar 07 line?

Pretty soon you'll have to break that graph to have 1 for each year...

The scary part is that even the older lines don't appear to be flattening out much if any. They're pretty much a straight line of ugliness.

The scary part is that pninen keeps moving the top of the chart up. Welcome to the chart, 22%! How long before we see 24%?

Thanks, btw, for your work.

Jan 07 had such a great start too!

And then it just fell into line with the rest

Agree - the consistency is...amazing

Sep 06 went down?

miracle.

pninen can you show this same exact thing for 3+ mo late?

Some of these 1+ months 'recover' and get on time... but once something hits 3+ mo late I'd toss it directly into the lost cause column so I am curious how that would look?? (please) :)

Just take ~10% off of it and you have a good approximation for how many of those would become 3+ late.

It takes a lot of faith in SOMETHING - be it yourself or a higher power - to keep lending on prosper when you see information like this...(shiver, chill, groan!)

All it takes is the realization that there are a lot of stupid lenders out there that will fund anything and that you aren't one of them.

So what's with those May '06 lenders? Some of them must have not gotten the "payment optional" memo.

You mean borrowers?

Must mean June 06 borrowers in yellow.

Notice that the June 06 borrowers are worse than the rest while the May 06 are better then the rest. If you average the two together then the result would fall in to the same pattern as the rest. :(

Now, that's the rule of dumb luck!

Without sticking the lines on top of each other to see how they trend over the same time perspective, how do you guys really tell to what extent things are better, worse or same?

Are you just going by visual observation? Is there some underlying data I am missing about the slope of these lines?

I think this is the most up-to-date version of sliding everything over.

I was making a visual observation on the slope (or the angle) of the line. Geometry was my favorite subject in school. :)

It's quite easy to compare the Xth dot on a line to the Xth dot of another line. That is how I do my comparison work.
Another quick visualization thing is to see that the last dot on the June 06 line is just over 13%, about the same spot as the August and September dots... And the Aug and Sep borrowers have had 2 & 3 less months to default.

Would it be possible (or has it been done) to break out this graph into multiple graphs, one for each credit grade. Prosper is heavy in low credit grade loans, and I would expect that drives the curves to where they are. Some of us are making it our policy to avoid E-HR listings. I'd be interested in seeing the data without the influence of these loans.

LR

LoanRanger0714,

Here is my earlier comment on an earlier chart per grade request. I do not know of anyone who has invested the time to create such a chart.

pjz

Thanks pjz! I was hoping either someone already did this or that one of the data miner types could do it easier than I could.

Yes.
(IMG:http://img.villagephotos.com/p/2006-6/1187065/prosper-late-loan-stats-vs-time-2_apr06bygrade_2007-04-15.gif)
Be careful when interpreting this chart. Once you break out by origination month and then by grade, you get buckets that don't have very many loans in them. Therefore when a single loan goes late or recovers it makes significant jumps in the curves. For example, you can't really interpret that magnitude of the A late rate, because there's only 1 late loan in the bucket of 38, and 1 isn't enough to get a reasonable estimate of a rate. You can see something about the general shape of things.

Thanks, pninen, for all your work so far. Just when you think it's leveling off... WHAM! It's like some kind of snuff film in slow motion...

I sure am going to miss my $300. :( Oh well, at least I'll still pay off my loan. I promise! :P

The charts are useful, but don't describe the whole story.

# of lates isn't valid for determining performance... Since the larger number of loans go to C and below, but the larger size loans go to B and above.

$10,000 A loan counts as 1 late, whereas 10x$1000 HR loans count as 10. Even though the same amount of money is late.

Gog,

Don't you think they are more relevant for those of us who pretty much bid the same amount on loans regardless of the loan size?

LoanRanger,


Since most people seem to bid the $50 minimum on their loans, the larger loans will have more lenders on them also. Therefore, the size of the loan at the time of default will affect the # of lenders with defaulted loans and skew the projections regarding lender performance.

That's true, but I suppose the question is what are you using the data for. My first concern is my own interests. Once we get enough default data to make sense out of, my first goal is going to be to determine a lending strategy that I'm comfortable with. I'll make a run at the data myself and explain better where I'm coming from.

LR

This is what I was after...

AA-D Loans - Performance

E-NC Loans - Performance

In my relatively short time here, it has become apparent to me that E-HR loans are not profitable unless you happen to be one of the very few who can pick the diamonds out of the rough (and have the time to do it). Thus, I was primarily interested in the default performance to date on AA-D loans. These graphs I'm presenting here don't tell me all I'd like to know (really not enough data on individual credit grades), but they do give me a general idea about the feasibility of making a return and the trend of the default performance on the loan population I'm interested in. I did produce these same graphs for each individual credit grade, but I agree with Pninen that there is a lack of data (default % is too sensitive because of lack of sufficient loans in each grade). Of course, eventually, I'd like more reliable default numbers so I can determine the APR I need for each credit grade to achieve an ROI.

Given that I would probably stick to $50 bids, the default rate as a function of # of loans rather than $'s is more important to me.

I followed a similar methodology to Pninen to generate these, if I correctly understand how Pninen generated his graphs and I didn't hose up my own data handling.

LR

Thank you, LoanRanger! I had been wanting to see things broken down that way for a long time. It is doubly useful since the definitions of E & HR have changed.

Thank you Pninen and LoanRanger for your insightful data and analysis. As a lender who was just starting to get educated, it sure makes a difference in my strategy to rely on the actual historical Prosper late rates vs. the published Experian default rates.

Certainly not a venture to be jumped into without a bit more study and ongoing analysis.

Thanks, LR. I will now even more carefully consider the E and HR loans I bid on.

ETA: I hope that with the new tighter criteria for HRs and the new data, we see these numbers dramatically improve.

pninen

If you could (please) do this exact same chart but with 2 changes

1) A version excluding only HRs
2) A version excluding Es and HRs

I'd love to see the exact same graph as you have it now, with no format changes, only changing the data set to the 2 versions above.

I am hoping for a great improvement just by excluding HRs alone :)

ETA: obviously exclude NC as well, keep forgetting about that data set since its no longer around.