Leaving the entire "old listings" debacle aside, a rather important
question:

I assume you threw some machine-learning and/or clustering stuff onto the dataset of loans/payments/defaults you've collected.
May I ask:
a.) what technique did you use?
b.) what validation did you do?

Both are rather critical to know, I'd say. Without having some confidence into the methodology you have used (which cannot be done without the answers to these questions), your "performance estimates" are rather meaningless.

- dvd

It is absolutely not a "step in the right direction[/b]. We have a great deal of evidence that higher rates default at a higher rate. You cannot push the rate beyond the rate people default at.

It makes no sense to me. On the one hand Prosper is warning lenders away from risky loans and on the other they are enabling riskier behavior by allowing (dubious) borrowers to list at higer rates.

Who here in his sane mind will take out a loan at 35% rates? Only stupid or desperate people will do so. Neither of those are good loan candidates.

While I see the rationale for it, it is regrettable that people will be further scared off from lower credit loans due to the result for the group. I fear it will make my approach harder because less loans that I bid on will be funded.

Prosper started with the concept that loans were an inefficent market (ie some people were better or worse than their credit score implied), but putting up the loss score for the segment as a whole may discourage lenders from doing their own thinking. Will interest rates move so that most loans end up with an equivalent ending ROI ? That will be interesting to watch.



I'm still unsure about the hidden ended listing thing. I haven't looked at it yet. Since it doesn't affect current listings, it might be work-able. I don't normally check ended listings of people I haven't bid on. I wonder how it's going to affect the stat sites, like Ericscc.

I didn't even consider that. What a disaster.

Edit: just saw Shen's reply. Whew!

Moth is barking up the wrong tree. If a borrower has a active listing all those old listings are still viewable to everyone.

Are you all insane??

Now lenders will be flying even more blind and honest borrowers will receive less constructive help too.

I was already out but now Prosper has proved that it is beyond redemption.

Read the immediately previous post:

If a borrower has a active listing all those old listings are still viewable to everyone.

Hi everyone,

Thanks for all of the great feedback, suggestions, and commentary. Let me try to address some of the bigger issues:

We can't see ended listings anymore. This sucks.

Yes, I understand. We don't expect that you're going to like this change. But like a lot of changes that we have made to the site, we made the choice to protect borrowers' privacy.

That said, I appreciate all of the commentary around how this was implemented (especially Pninen's), and I'll look into improving the experience of looking up an ended listing. We do still show the listing summary information on the borrower's member page (picture, title, amount, rate, etc.), so maybe we could just add this information to that "Listing not available" page, and it would alleviate some of your angst.


I don't understand why these portfolio plans are built the way they are. Why only bid on auto-funding listings?

Our credit team built the portfolio plans for a good mix of borrowers at different levels of risk/reward. They basically mixed and matched the credit segments found on this page into the various plans.


If a borrower was a member of a group when the first loan was taken out, can he change groups before posting a listing for the second loan?

If a borrower is a member of a group when his first loan was issued, then no, he cannot change groups before posting a second loan request. If this borrower is a member of a group that requires listing review, the group leader will not have the chance to review (and therefore reject) the borrower's listing. This is necessary because otherwise the GL could effectively block the borrower from requesting credit on Prosper, which would be illegal. For a first loan, the borrower can always leave the group if the GL rejects the listing.

Looks like we are to trust Prosper on their guidance since we are no longer able to do our own research. Bull----!

Not at all. The research that brought us to display this estimated loss and return data has been available since February 2007 on the marketplace performance page. What we're presenting on the bidding page now is simply a segmentation and re-formatting of the same data. In fact, if you think you are smarter than we are (although you are all so humble around here :) ), you should do your own research with the data and come up with your own bidding strategy to beat all the chumps that are using Prosper's analysis.

Those are the biggies - I'm sorry if I didn't answer your specific question, but I wanted to get a response out. Keep the questions/comments coming.

Best regards,
Andrew

Andrew,

on the assumption that you were replying to my question...

My point was rather the following: If you've used a clustering algorithm on the dataset that you have (Assuming you haven't magically conjured up a larger set), and end up with 54 clusters it is (in my book) a strong indication that you might be suffering from a bad case of overfitting.

Therefore, I'm wondering how you validated the thing, whether the clusters were stable, things like that.
Its just that 54 clusters is a quite large number, and the number of entries per cluster is dangerously low. (Especially if you've got the usual pattern of some large and many small clusters!)

- dvd

(TBH, I don't particular care about the guidance - Its just that, in my professional opinion, the thing looks a bit like a misused machien learning algorithm. Those beasts are damn tricky to use correctly!)

When, exactly, is Prosper going to start employing this balance that you speak of? Everything Prosper has done lately, with this being the most outrageous example to date, is to massively tilt this so-called balance towards borrowers and away from lenders. Let me put this as plainly as I can. This is complete bullshi*t. :angry: :angry:

The only way to learn what to bid on (and to teach the newbie lenders) is to be able to review the listings of the loans that go south and look for the red flags, and to review the portfolios of successful and unsuccessful lenders and see what they bid on. Now all of that is taken away from us. Obviously, Prosper's rationale is not protecting PII, but keeping as many lenders as possible as "mushrooms" (you know, keep them in the dark and feed them bullsh*t). Unless the third party sites can (and do) start archiving all listings, I'm done with Prosper. :angry:

I noticed that the servicing fee on this loan was still 0.5% even though it's an A. When do the new A / AA servicing fees really go into effect?

(IMG:http://i240.photobucket.com/albums/ff137/tashprosper/servicing_fee.png)

Eric has archived all listings.

I just can't wait for Prosper to try to force him to delete his archives.

Hmm, how about this guess to discuss this further:

In AF loans the interest rate, at which the loan closes, is predictable.

Since Prosper displays an estimated return the portfolio shall yield, it may help regarding predictability to take an larger than average amount of autofunds into the portfolio.

The assumption would be that this could increase predictability. So the portfolio might have a little less ROI then with less AFs but is more likely to meet predictions.

Could this be the strategy behind this?

I have no problem with borrowers being able to take out a second loan, but this aspect of it needs to be looked at more closely.

As a group leader, I am responsible for the reputation of my group, and therefore refuse to allow borrowers to list without a review. For this reason I have shut down my group until something changes.

Andrew, is there not some way to allow the borrower to still list as No Group, or to join a different group for a different listing? After all, there could be valid reasons. If a borrower lists first for debt consolidation they may wish to be in a debt consolidation group. Then later they decide to list a second loan for business expansion and may find more benefit in a group for small businesses. Now that there are no group leader rewards, is there any valid reason to restrict a borrower to a single group?

The more I think about it the more I agree with this statement.

How on earth will anyone be able to learn? New lenders are reduced to look at the performance stats prosper gives you or guess.

This would reduce Prosper to Lendingclub.

This is one edge I do not mind having over newer lenders... :D

Just imagine if the SEC was able to poof all the news stories for companies that were no longer being traded on an exchange. All that they kept was a record of their stock price and volume and perhaps their quarterly and yearly filings.

Isn't there a great deal that can be learned from reading the past and comparing it to how those stocks performed?

Even if ericscc does archive them (which he says he does) he doesn't have the credit data, which is needed to do a true analysis.

I beg you Prosper, put it back! Please let us keep our history!!

Thanks for the correction, LoanChimp and others; I wasn't logged in at the time and that was why I couldn't see past listings on relists. Sorry for the false alarm. :)

The same could be said for those who take out loans at 20% or 25%.

The performance of all C and D loans, originating 6/1/2006-6/1/2007, is 0.73% and 0.16%.

The average lender rate of these loans is 17.87% and 20.84%.

Add 500 basis points to each to compensate from Prosper's removal of City information, the elimination of vetted loans, and denial of PII to GL's.

If a lender is hoping for a 10% ROI, this would mean that an average C would need to start at 32.14% and a D at 35.68%. (E's would need 49.72%).

Prosper's change basically makes bidding on C and D listings viable again.

I heartily agree.

Also look at it from the borrower's perspective, as I've said already. If a listing doesn't fund, the borrower can't get any advice about what (s)he might do differently after the fact.

Consider the discontinuation of providing advice to people who are unable to put together a viable listing without extensive coaching.

What kind of red flags are you talking about ?

You can look at the default numbers of a group as a whole and look at the current and past listings of active borrowers. How is looking at an individual ended listing going to help you ? There's no way to establish a causation between an element in the listing and their default when looking at a single listing.
Aggregate data is still available in the performance page.

Advice on bad budgets, etc can be communicated without seeing actual listings.

I couldn't find any information there about the performance of business start-up loans vs business expansion loans, or business loans vs personal loans, or loans to pay off PDLs vs loans to buy Christmas gifts, or single parents vs married...

Unfortunately Eric did not archive the extended credit data. :(

First of all, that would not be illegal. Second of all, if it's liability that Prosper is worried about, I am a bit perplexed as to why you let scam artist GLs run free on here for months on end leading to millions of dollars in lost money for lenders. To be blunt, whatever firm/lawyer you are retaining for services is not doing a very good job.

By the way, I very much appreciate Prosper giving group leaders a heads up on this huge change in the group system. Just like when I advertised in Texas a day before you guys changed the rate cap in Texas.

The communication system between Prosper and it's members is severely lacking.

It's funny, I've tried to run a good group on here since I got here, and I think I've been moderately successful at it, yet Prosper seems intent on throwing constant screwballs in the direction of group leaders and lenders -- with no forewarning. Seems to me to be a poor way to run a business.

Just to chime in on one thing...

the data this morning still had the listings and the new authenticated API has access to the extended credit.

However eric or LS or ProP is going to have to monte carlo or something similar since the new tos API forbids row level republication of authenticated information.

Hey Rateladder,

I think you're a good guy and all, but when you chime in like this (and like you did at least one other place today about your positive inclination regarding a recent platform change), shouldn't you perhaps disclose the nature of your professional relationship with Prosper?

English please?

sorry let me try again... :)

I interpret the API TOS to say I can't show the extended credit of a specific listing on ProProsper or via my rateladder listing widget.

However I can build a simulation tool that takes a extended credit criteria and account funding patterns determines how a delinquency rate.

they same way all the older lenders have a delinquency rate.

When you run such a simulation with slightly different start dates you can get drastically different answers.

when you run a bunch of simulations it is called a monte calro and gives you both an average and a error rate.

Sorry if I slipped in and out of english again...

There are probably other examples of acceptable and not but this one makes sense to me.

Did you miss Urbi's question?

Rateladder, you say, "This rocks." I think you have the first two letters wrong.

Sorry to say, but if I were you, I would invest my money into a basic textbook that contains good chapters about statistical machine learning, especially classifiers and clustering algorithms.

My hint:
Re-read what a monte-carlo method is good for.
You're approaching the problem with the wrong tool.

- dvd

Thanks. The main thing I want from Erics or one of the other sites is to recreate what prosper just took away -- an archive of ALL listing data, including extended credit and Q&A, that we can use to get around Prosper's recent stupidity. If I don't get that, I'm through with Prosper (and I'm sure a lot of other people are too). I have no problem with restricting that information to signed in lenders, rather than making it available to the general public, but I want that data back. Period.

Rateladder, do you work for Prosper? If so, in what capacity?

Do you work with Prosper? An employee or independent contractor?

Do you have stock options?

Regarding the portfolio selections: I have one question I would like clarification on.

Let's say I choose a portfolio that is based on 4 underlying SOs in 25% each.

So I put in $400 @ $50 increments. Theoretically, if I understand this correctly, I should now get 8 loans (2 from each SO). But what happens if one SO is unable to find any listings to bid on? Or I get consistently outbid on one particular category of loan?

Will I end up with a skewed portfolio of say, $300 in 6 loans?


That portfolio may then not approach the return I'm hoping for.

Let's not drag this off-topic.

Rateladder works as a consultant/contractor for Prosper in the database area.

He's not getting stock options. At least not last time we talked about this.

He's enthusiastic about access to additional data through the Authenticated API - for him (and for me) that rocks.

A lot of what Prosper did don't rock (to put it mildly), but the authenticated API is a great stride for data miners.

Hollow Oak, sorry, but I do think it's ON topic to see the biases (if any) of those who are supporting these changes.

I would understand if a borrower supported these changes, but someone I've always -- mentally, at least -- associated with the lender community? I still welcome Rate Ladder's comments.

What will the authenticated API do for us? (As you know, I'm not computer literate.)

A couple of questions:

- How often will Prosper be updating the estimated loss data?
- Will Prosper be tweaking/optimizing/re-defining the portfolio plans at regularly scheduled intervals?
- Since this is starting to feel a bit like a "managed fund," will there be a designated "fund manager" with whom we can interact?
- Will Prosper be publishing something similar to a quarterly fund report?
- How will we be able to evaluate the performance of the portfolio plans releative to their projected performance?
- When will I be able to set up the UrbiFund and charge newbies a management fee for my subscription-based service?

Edit to add:
In the Portfolio plan definition:

Does this mean that it will allow ANY number in this box - or it must show at least 1 or greater current delinquency?

This post has been edited by Urbi_et_Orbi on Oct-30-2007 04:10 PM

EDIT(If the previous posts are accurate, and old listings will be available for borrowers with active listings, perhaps I can manage. Not a welcome change though)

Andrew,

Overall I am supportive of recent changes, I would ask you consider the following drawback on hiding old loan requests.

I have found on many occasions old loan requests for a borrower who dramatically changes the amount of money being requested, and the reasons behind it.

Start a new business, then assist a new bride in Nigeria, and after that debt consolidation? Avoiding these kind of fraudulent loans has been helpful in avoiding defaults. (Banana Books had 3 people listing the same loan request (not fraud, I contacted them, but worth knowing)

I am trying to imagine a happy medium behind privacy and fraud detection. and as I cannot see solution that would allow any data to be hidden, would suggest if we can't simply restore the old availability of data, perhaps putting a 6 month limit on old listing availability.

I hope you can consider the above, and that a working alternative can be found.

That allows third-party sites (like Erics and Lendingstats) to get the extended credit data. That is a welcome development. Too bad it's always one step forward, ten back with Prosper.

Just to clarify, When a borrower relists, ALL of the borrowers prior loan requests & past loans are fully viewable again. Yes. I tested it out. It's one of the first things I checked.

Andrew,
On the hidden listings issue, we understand Prosper's desire to keep borrower's information private. I also think you understand the lenders concern that hiding the prior listings will de-educate the lender pool. Is there some way to find a middle ground that largely satisfies both needs?

How about this? If a lender wants to have their portfolio reviewed to find out what he/she is doing wrong, allow that lender to flip a switch that temporarily makings all the listings in his/her loan portfolio visible to existing lenders for a week. This should be enough time for the lending community to search through the loans and identify the weaknesses. These peer lender review sessions are great, not just for the individual lender, but helps to educate all of the other lenders that join in the discussion.

If we don't come up with a peer-lender education solution, it will result in higher default rates, which will force lenders to compensate by raising their bid rates, which will raise the borrower's loan rates which will make Prosper far less competitive in the p2p market place.

Thanks for listening.

But, if what rateladder says is correct, they get the extended credit data subject to the condition that they not disclose it in any manner that can be connected to the individual borrowers. In other words, for statistical analysis only.

Ira and Trav, thanks.

What's the difference between the API available 2 days ago (before this change) and the API available now?

You can sign in. The difference is conceptually the same as the difference between what you can see on Prosper.com before and after you have signed on. With some llimitation still existing in the API, but ou should get the idea.

[edit]

I would like to make one suggestion for additional loans. When doing a test listing for a second loan I am presented with a bank information screen on step 1. This screen defaults to the primary account and has a link that says "Use a different bank account". When I click the link I am taken to fax sheet to send in a voided check. However I have 2 other verified bank accounts on file. The form should let me choose which bank account I want to use for the loan or have a explanation why I can't use one of the other accounts.

This would be particularly useful for borrowers (like myself) who might want to take both a business and a personal loan because it would allow you to have the business loan payments come out of a business bank account and the personal loan payments come out of a personal account.

The "hidden data" aspect is just another in a long series of bad decisions, in my opinion.

It is mostly just an annoyance to the lenders, but it is (another) kiss of death to any responsible Group Leader. How can he possibly make a reasonable decision with regard to group membership (or listing approval) without being allowed to know about prior listings that might have contradictory information from what he's being asked to approve now.

If anyone at P-----r had thought this through, it MIGHT have been a solution to only allow a GL to view this information after membership had been approved. Then, they could still stop bad listings from going out under their group's name by refusing to approve. The borrower could still go to the schlock GLs who don't give a rat's *** but at least he/she couldn't hurt the group or GLs reputation by turning out to be an easily-detected scammer.

Or, perhaps, I give too little credit to the P-----r thought processes. Since the net effect of so many recent decisions has been to drive more and more conscientious and helpful GLs away, it is entirely possible that this is just one more effort to continue that trend. If so, it's working admirably. The last one I was still following threw in the towel this morning based on this latest "enhancement", and my towel as a lender was right behind it.

Are you actually keeping stats on the default rates of those different types of loans ?
Have you been careful to eliminate other factors ?

That's a lot of manual work and I don't think wiseclerk or eric's or the other sites do it either.
How many people are actually doing that sort of stats keeping on a methodical basis ?

Then let's examine the biases of people who spend all day everyday trashing Prosper on the forums as well...
Why does support automatically equal suspiscion and yet people can trash all they want and no one may dare questions them with the implication that they are doing it for monetary gain ?

So I'm still waiting for Ira01 or someone else to answer
specifically, what have you lost from the hidden ended listings?
In a practical, specific sense what were you actually doing
(not what you could have done, or what a person in general did)
that you are no longer able to do ?

If you can't answer that, I'm going to consider it more hot wind.