Saturday, March 31, 2007

Proper Listing Extraction

I've finally discovered the magic to extract all the listings from the XML data. My previous problems finding enough listings was due to user error. The CR-LF codes in the listing descriptions were messing up my parsing program. Quite maddening!

Now that I've got the extraction down, it's time to rerun the matching algorithm.

Monday, March 12, 2007

Prosper Around The Web

A few web clippings from the last few days

Iinnovate has a podcast with Chris Larson, founder and CEO of

In this podcast, Chris chats with Julio and Min Li about Prosper's innovative vision, its unique challenges as an eBay platform for money (check out Prosper's cool tools for academics and researchers for performing case studies on its model), and the fine balance between the countervailing forces of transparency and privacy in the world of Web 2.0.

Catallarchy looks at default rates over time

Default rate is almost zero for the first 6 months, then rises in a roughly linear fashion, more linear for the higher credit grades (graph). This means that for portfolios with many loans younger than 6 months, projecting current lateness rate forward is erroneous.

My Money Blog has a Prosper review is a person-to-person lending service where you can lend out money to complete strangers. My first and only post about Prosper was back on February 13, 2006, when it was first released to the public. Since then, I haven’t written a peep about them. An online service that offers high interest rates for my cash? Why haven’t I written about them? The simplest answer is that I’ve been waiting for more information to review.

Determining Funded Listings

To get a reasonable loan to listing link-up, I had thought it'd simplify things to remove all the unfunded listings from the XML data to get down to the essentials. The two primary metrics I focused on were in the PercentFunded and AmountRemaining tags. I had assumed that if the PercentFunded was 1.0 or the AmountRemaining was 0.0, the listing had been funded. A few oddities popped up:

  • One loan (38580) reports 99.990% funded, but 0.0 remaining. Which is right?
  • One loan (59) reports the amount remaining at -98.00. It's been removed from the website, but is still in the XML data.
  • In this dataset, there are 8386 loans listed and only 7950 listings that meet my criteria for funding.

Tuesday, March 6, 2007

More On Historical Interest Rate Table

Following up on my previous post, it looks like there's been some resolution on the HistoricalInterestRateTable XML tag field. From the Prosper Forums, Dzogchen noted that the data looked suspiciously like the 30-day average rate for lenders.

Armed with this suggestion, I went back into the XML data and managed to calculate the historical interest rate table using loan listings less than 30 days old. This provides good information on how to interpret the rate table provided in the XML and a validation of Prosper's data dump. Good stuff!

Saturday, March 3, 2007

How Is Historical Interest Rate Data Calculated?

RateLadder had posted a note that Prosper has exceeded $40 Million in loans (go Prosper) when they did posted the Marketplace data. This lead to an interesting question: does the historical interest rate information reflect borrower rates or lender rates? This matters because it tells whether groups are lowering the borrower's interest rate (as Prosper intends) or merely redistributing the interest.

I started chewing on the published loan data, and I can't seem to figure out how Prosper got their historical interest rate information. I took a look at the NC credit group since it's a smaller group. Just looking at data for NC loans with groups, I took the average loan interest rate and a dollar-weighted interest rate and compared it to the Prosper published info.

  • Prosper Group'ed NC: 17.676%
  • My Group'ed NC, Loan Average: 21.46% (Borrower), 20.28% (Lender
  • My Group'ed NC, Dollar-Weighted Average: 22.44% (Borrower), 21.24% (Lender)
This is a wildly different result, not just a rounding issue. I also noticed that there are NC loans in the $5k - $10k region even though Prosper does not publish historical information on these loans.

So, how is Prosper calculating their historical interest rates?

Prosper Around The Web

I stumbled on a few bloggers writing about their Prosper thoughts.

The first is a follow-up post, So How Is Prosper Doing?, to earlier posts Why Will Fail and Why Will Succeed.

I have nothing against Prosper and feel a little bad that my negative take on it shows up so high on Google while my thoughts on the potential positives do not (although I linked to the positive post on the negative post).

So, 7 or so months later, I thought I'd look to see how Prosper is doing.

According to this NPR interview with Prosper CEO Chris Larsen, Prosper has over 140,000 members and has funded over $27 million in loans. My thoughts obviously haven't dampened others' enthusiasm.

The other, Lending on – Successes, Issues and Suggestions, is about an individual's success making loans with Prosper.

It's now been almost a quarter since I began using Overall the Prosper experience has been lucrative. My portfolio currently has a risk-adjusted return of over 10%. That certainly beats what you can get currently with both money market accounts and CDs both of which currently will return you around 5%

First let me say that I now have over a hundred active loans. Most performances been good to give you a flavor of what I've experienced so far I've had four loans go late. Two of these loans have already been remedied and brought back to being current. The other loans are currently less than 15 days late. We'll have to wait and see how those go.

Friday, March 2, 2007

Matching Listings To Loans

Prosper is a statistics junkie's paradise since they make a LOT of their statistics available in XML form. From this, statistics like defaults versus credit rating can be calculated. Unfortunately, they do not provide a clean way to connect the listing information to the loan information. Without this information, all the juicy stats like default rates versus state or default rates versus home ownership cannot be determined. Fortunately, the Prosper Forum wizards are on it:

For what it's worth, I align based on amount, interest rate, credit grade, group (or lack thereof), and DTI (looking for cases where DTI remains constant). When I do that, 97% of loans have 1 listing candidate, 2.5% have multiple candidates, and 0.5% have zero candidates. (The 2.5%-0.5% split is from memory, but it's something like that.) I've manually verified at least 100 samples of the 97% and I'm confident in their accuracy.

Also interesting was an explanation on why Prosper doesn't directly link listings and loans

We don't do this for regulatory reasons. Basically, if we exposed a person's payment history (which is implied in the loan's daily status), even though that person's identity is hidden behind a screen name, Prosper would effectively become a credit reporting agency, and therefore become subject to a pile of credit reporting regulations and requirements which we are not prepared at this time to take on (and may never be - we'll leave that game to Experian, Equifax, etc.). So that's why we don't include a link between listings and loans.