Idle ramblings of a wandering mind...

I have just came back from AMCIS 2007 - Americas Conference on Information Systems. The conference was held at the beautiful Keystone, Colorado. I presented my preliminary paper on Prosper.com titled - "Bank of One: Empirical Analysis of Peer to Peer Financial Marketplace". The presentation was received very well. I even received an invitation to submit it to a very good journal - which is quite nice. Several attendees asked me for a copy of my presentation - so here is a PDF copy of my AMCIS 2007 presentation on Prosper.com. The paper itself is available through AMCIS Proceedings or you can leave a comment on this post with your email and I would be happy to send you a copy of the paper.

Update:
I received several appreciate emails about the presentation file posted above. The post was linked in several blogs including Tom's excellent blog: Prosper Lending Review - Thanks Tom.

Labels: Conference Presentations, Prosper, Research

   

Recently one of the readers of this blog - Akash, posed this question about Prosper.com:

Great blog. I am trying to understand the p2p lending biz models out there, such as prosper and zopa. What I am struggling with how these guys make money:

- looking at the public data, these sites take on average 1% origination fee from borrower, and 0.5% from lenders (on outstanding loan), then it would appear that they are not making a lot of money even on a loan volumes of $1 Billion ($15M in revenue)
- when will these companies make money? their expenses even running an internet outfit will be high
- they have raised over $20M in VC money;

Questions:

- what are investors seeing that I am not able to grasp? (assuming they will get traction)
- will they be able to monetize their loans in the secondary market somehow

Here are my thoughts on the subject:

Current Revenue and Cost Streams for Prosper
I agree that on the current scale of operation, Prosper is not making too much money. However - look at the other side - the business model is risk free - once the loan has originated, the future revenue stream (apart from defaults) is guaranteed. The cost elements of the business are very low - running the site costs only little, they do not have too many employees (double figures only), Prosper is not doing any paid marketing and once the platform is established, they can outsource most of the routine work. With sustained growth for a few years - they will have enough scale to make decent money with very little downside.

My sense is that their current cash burnout rate is quite low - so their $50 million venture funding should go a long way. Even if they need extra cash, they have a positive cash flow in future with all the loan repayments - so they shouldn't have problem raising either secureed debt or the next round of venture financing.

Future Revenue Streams
I imagine that once they have established their current business model, they will start cross-selling other financial products. Lender's insurance, database mining, prosper security backed credit cards, syncing your prosper account with your paypal account... secondary trading in Prosper loans... the possibilities are endless. Oh yes - Prosper Loan insurance - insurance than lenders can buy that will protect them against loan default - this can be a pretty good business - I will buy it, especially after suffering through the pain of a couple of defaults!!

Secondary Trading of Prosper Loans
As far as secondary trading in Prosper loans is concerned - I hope Prosper introduces something similar soon. Currently there is no "exit" for Prosper lenders - you got to wait 36 months for your payments to slowly trickle in. What if you need some cash out? Right now the only good way is for you to actually borrow in Prosper for a similar amount that you have lent out. I am not sure whether anybody has really tried it as an exit strategy - although there are several lenders which try to use it as an arbitrage strategy. Creating a secondary market is not difficult at all - in fact it will be remarkably similar to the current platform. A current lender can put a portfolio of loans up for auction and other lenders can bid for it. Once the auction clears - the documentation can get shifted to the new owners of the loan. Easy! This secondary market will bring liquidity to Prosper loans and encourage more lenders to participate as they will always have an opportunity to exit the market.

Labels: Prosper

   

My last post on the group reputation system at Prosper.com focussed on the effectiveness of the system. Since then I have continued my research on the same and will likely deploy a survey to group leaders in next couple of months to test my hypotheses.

There has been a lot of research recently on online reputation systems like the ones used at eBay and Amazon. There are some important differences between the "individual" reputation systems at eBay etc and the group reputation system at Prosper.

• Seller Only Reputation: First of all, Prosper's reputation system is essentially a seller reputation system. Buyers do not need to prove their reputation as they bear all the risk (since they pay money upfront). This is a key difference compared to eBay where the reputation system includes both buyer as well as seller transactions.
• Credit Scores: While the reputation system is the ONLY window to a seller's reliability in eBay; in Propser, buyers have access to the credit history of sellers - which gives them all sorts of information to judge seller's reliability. So the reputation system here is of only incremental importance over and above the information available in the credit history. The group reputation system is essentially as secondary reputation system with the credit score being the primary reputation system.
• No Repeat Transactions: Unlike eBay and Amazon, sellers at Prosper rarely conduct repeat transactions (second, third loan). Since the sellers are unlikely to come back to the market, their incentive for keeping a great reputation in the market is quite low. Hence, there needs to be factors beyond the marketplace to force them into complying with buyer's expectations.
• Group's Core Connection: Prosper expects that if sellers are part of a tightly knit group then there would be peer pressure from the group for the seller to behave nicely. However, this would only be true if the group has a strong core connection (whether offline or online) which has some value to the seller. Currently most groups in Prosper do not have a core connection - they are just an online group of people with similar needs. This is unlikely to lead to good behavior by sellers and we have seen the evidence of same in the repayment data.
• Filtering by Groups: Prosper also expects the group to be a filtering mechanism where group leaders check the listings and only let trustworthy borrowers come to the market. However, many groups do not even enforce the mandatory review of the listing by the group leaders - so even this expectation is futile.
• Conclusions: Based on the above - we come to the following conclusion:
• Groups that share a strong core connection (offline or online) would perform better
• Groups that have a strong process for filtering listings would perform better

We have refined the conclusions above to generate research hypotheses and are testing these hypotheses using forthcoming survey data and data collected from Prosper. I will post the results once the analysis is done in next couple of months.

Links

1. Thanks are due to Prosperous PF Strategies for leaving a thoughtful comment. Link to the comment.
2. Rate Ladder is an excellent resource for Prosper participants. It is a little cluttered with ads - but has great information.

Labels: Prosper, Research

   

Today is turning into a pretty nice day. First, I checked the current copy of the SmartMoney, the magazine of the Wall Street Journal, and saw that it referred to my research on peer to peer financial marketplaces in one of the articles. Second, I heard back from the Academy of Management Conference and one of my papers have been accepted for presentation there.

The current issues of SmartMoney has an article on Prosper.com titled "The Banker Next Door" by Ann Kadet. She quotes my research in the article. I am quoting the article below:

As a whole, Prosper's lender community is still more subject to irrational whims and emotional behavior than professional credit analysts. Sanjeev Kumar, a doctoral student at the University of Michigan's Ross School of Business, analyzed six months of Prosper activity and found that quantitative data such as credit scores and borrower income explain just a third of bidding behavior. "Everything else is subjective, like whether there is a photo of the kids", he says.

Here is the link to the original article: Banker Next Door.

Now the other good news. Academy of Management is the largest association of management researchers and professionals. The Annual Meeting of Academy is one of the largest gathering of management researchers. It would be a great experience to present in the Academy 2007 conference. The paper accepted is titled: Embedded Trust in Open Source Software Development Communities. Following is the abstract of the paper:

Structure of naturally evolving collaborative relationships in open source software (OSS) development communities has been identified as a critical factor in success of OSS projects. This study attempts to extend the literature by examining impacts of collaborative relationship structures on trust, an important yet under researched construct in the OSS context. Based on social network and organizational theories, this study proposes that collaboration network structures have significant effects on individual OSS developer’s trust in a project team. Subsequently, we expect that trust has positive effects on effectiveness of an OSS development community. Empirical analysis of project and survey data using partial least square (PLS) verifies that collaboration network structures significantly affect trust, which subsequently enhances OSS team effectiveness in terms of perceived software quality, team cohesiveness and satisfaction with the teamwork. Results of this study shed light on the antecedents and relevance of embedded trust in OSS communities with significant implications for both researchers and practitioners.

This year has started well - I attended the first conference of the year and my paper got accepted at the first paper submission of the year. I hope the luck is continue for the rest of the year as this is my job market year. Talking of jobs - I finally finished writing my academic Vita.

I also got the full student feedback for the course I taught in Spring 2006. Here is the feedback image from the evaluation website:

Quite Nice, Huh?

Labels: Conference Presentations, Prosper, Research, Teaching

   

Continuing my data analysis pieces on Prosper.com (click here for all my posts regarding Prosper), I am looking at the group rating system in this post (using the more academic term - Group Reputation Systems).

Prosper introduced the rating system a few months back. Borrower's have the option of being part of a group and groups are given a star rating (1 star to 5 star) based on the repaying history of past loans originated from the group. More details are available on Propser's page. Now, the first thing which is clear to anyone with any experience lending on Prosper is that this group reputation system is broken - it does not work! In this post I will first provide "empirical evidence" (fancy research term for a statistically tested argument) that the group reputation system is not working and then I will discuss reasons why this is so and how can this be fixed.

Effectiveness of Prosper's Group Reputation System

The essential idea is the groups will keep tabs on their member borrowers resulting in lower default rates for good groups. This will lead to lenders charging lower interest rates (lower risk premiums) for borrowers belonging to these better groups, providing incentives for borrowers to be part of these groups. A nice virtuous cycle. Now, lets look at some numbers and see if this cycle is in fact working:

The image below is a screenshot from Prosper of interest rates for last 30 days. Look at differences between interest rates for borrowers belonging to a group and those not.

As we expect, for most categories interest rates for group members are below those for non group members. This is good - the system is mostly working for the borrowers. However, there are many categories where the system breaks down and non members in fact get better rates (for example: AA for loans between 5K and 10K).

Now lets see if the system is working for lenders. I can't divulge all the details here as its part of a research paper that is under review - but this is essentially what I did. I collected six months of listing information and the loan repayment information for all resulting loans. Then I analyzed this data to discover the factors that are antecedents of loan default - factors that drive loan default. The analysis was done using something called a duration model (also known as survival analysis). This statistical technique is widely used to model failure events where time to failure is known. Failure in our case is of course a loan default. Duration models output hazard ratio, which is an estimate of the relative risk of failure resulting from the factors included in the model. A hazard ratio greater than 1 would mean that the risk of failure increases with the factor. In my preliminary models the hazard ratio for group membership came out well above 1. This means that the risk of default is actually higher for borrowers that belong to a group! So as far as lenders are concerned - the group system completely does not work.

The duration model of loan defaults also uncovers some interesting results.

• The Debt to Income ratios of borrowers seem to have no impact on the risk of loan default
• Amount of information provided in listing description (on an overall description length basis) has no impact on the risk of loan default
• Being a homeowner also does nothing to the risk of loan default
• Listings with endorsements from group leaders seem to have a lower risk of loan default
  

I will provide more details of this analysis as the paper passes through the review process and I finalize my dissertation.

Whats the Problem?

This is a topic for future post. There has been extensive academic research on reputation systems in past years - I will borrow from those and analyze Prosper's reputation system in that light.

GMAT Preparation?

One of my long time friendshas started this wonderful online service to help students preparing for GMAT. Look up his site WinGMAT.com. I have looked through the contents of the site and they are good, quite good!

Labels: Prosper, Research, Statistics

   

I have not posted anything here for a while as I have been busy trying to give shape to my dissertation. As part of my research I have been working with Lending, Borrowing and Loan Repayment Data at Prosper.com and there are some pretty neat results. I finished writing an article about my results and submitted it to one of the major Information Systems conferences AMCIS (Americas Conference on Information Systems) for review. I of course can not detail all the results here as it is under review, here is the abstract of the paper I submitted:

Peer to peer financial marketplaces provide a platform for individual lenders and borrowers to interact and transact. These marketplaces dis-intermediate the traditional financial services business models. In this exploratory paper we study the operation and effectiveness of one such marketplace: Prosper.com. We analyze six months of lender, borrower and loan repayment data to answer preliminary research questions about lender behavior, market effectiveness and antecedents of loan default. We show that lenders mostly behave rationally and charge appropriate risk premiums for antecedents of loan default. We also show that there are mismatches between risk premiums charged and relative importance of factors that drive loan default. We then explore the dynamic process of lenders adjusting their lending strategies to reduce these mismatches. Interestingly, our results indicate that the group reputation used in marketplace is not effective and needs to be enhanced. Our analysis provides a base for future research in this exciting and evolving context. Our results provide directions for practice applications as well as future research in design of financial marketplaces, investing and risk mitigation strategies and improving effectiveness of financial marketplaces.

I am excited about my research on Prosper.com (or Peer to Peer Financial Marketplaces - the name I have given to the broader phenomenon). I am continuing to enhance my analysis and looking for improving my paper and submitting to a top journal. If you would like to read a copy of my paper then pls drop a comment with your email and I will send one to you.

It seems my research will be quoted in the SmartMoney Magazine. Ann Kadet, who is a columnist and Senior Editor with the magazine got in touch with me about my research on Prosper.com. The coming issue of SmartMoney will have an article on Prosper.com and according to Ann, there will be a mention of my work. Cool...

Portfolio Update

Like everyone else, my stock portfolio also took a hit in last two weeks with the sharp decline in US markets. Although, I continue to believe that in the medium term stocks are worth keeping - so I used the drop to add to my holdings and I have now mostly recovered. My current major holdings are (in order of size of holding): STX, AMD, TTM, TNE, UL, FTE, SCSS, CX, CAT, IBN, TTMI, HMC and TKC. As you can see - I have diversified considerably - I now have large technology MNCs, Indian Auto, Brazilian Telecom, Mexican Cement, Consumer Goods, Global Auto, Turkish Telecom and some good old US companies in my portfolio... Before the Feb end decline, I had booked profits to the tune of some 16% return - not bad for a two month period.

Research Update

Since I wrote last one of my papers was nominated for the best paper award at HICSS 2007 (Hawaii International Conference on Systems Sciences). The paper is available here. I have further improved this paper and it is now going to be a part of my dissertation.

Labels: Conference Presentations, Investing, Prosper, Research

   

I have not written for a while as I am attending the International Conference on Information Systems (ICIS) 2006 in Milawaukee, Wisconsin. Its a delightful downtown full of holiday decorations and lots of beer places. Yesterday I presented a research paper in the OASIS Workshop - Empirical Analysis of Determinants of Perceived Customer Satisfaction with Open Source Software (extended abstract).

This post is essentially an update of previous posts. First - my stock portfolio, then an update on my Prosper adventure and lastly, research.

Stock Portfolio

My post on a 60%+ stock market return in 9 months generated a lot of interest. Well - my portfolio is continuing to perform and I am on on track for ~80% by the end of the year. I will be paying capital gains tax on this end of the year - but thankfully my marginal tax rate is low so its not going to hurt too much. Here is my current stock portfolio:

• Seagate (STX): It has been the largest stock in my portfolio for a while. Now I have booked profits to the extent of about half of my initial holding but I still have quite a bit left.
• LCA-Vision (LCAV): I bought some when it dropped after bad results as I am positive on its medium term growth. It has been reasonably volatile since then and that had given me the opportunity to book profits and re-enter.
• Cemex (CX): It is the classic business school case - and its growing. There have been some unpleasant acquisition news going around - that gave me an opportunity to enter. This is my first time holding on Cemex.
• TurkCell (TKC): I put in a lot of money on this a few months back - was down a significant percentage one time then it recovered sharply and I was able to book profit on my entire holding. Then it dropped and I have re-entered.
• Tata Motors (TTM): I have a smaller holding in my one time employer than I would like to have. I am looking for entry points but it has been consistent and strong.
  
• Other small holding are - Select Comfort (SCSS), Christopher & Banks (CBK), PepsiAmericas (PAS) and HiMax Tech (HIMX). I picked initial stakes in all of these after their sharp down move recently.
  

You can see that I am quite heavy on emerging markets. This also insulates me from the coming decline of the dollar and the slowing down on the US economy. Here is a screenshot of how my portfolio has grown in last 6 months. Not all of this is stock appreciation - I also infused cash along the way. I have cut out the values on the y-axis for privacy reasons.


Prosper Adventure

I am continuing to put money in Prosper - although the careful lending strategy that I use necessitates slow scaling up. Now I have ~3.5% of my total investments in Prosper. All the data analysis pieces I posted here has helped me make better sense of the dynamics in Prosper. All my loans are current and I have average interest rate of 20%+. Lets see how long the honeymoon continues. This is my current Prosper portfolio distribution. As I had recommended in my previous posts - D credit group is the way to go in Prosper.


Research

I am in the middle of all the big names in IS research here in this conference. I have been talking to several researchers and here is the big thing everyone is talking about - impact of IS on Innovation. The troublesome thing is that I am also looking at the same thing - although I am confident that the niche I am following is quite unique and I will be able to look at it from a different angle than other researchers.

PS> LFCFan has inspired me to put some ads on my blog and "monetize" the traffic. I am not completely comfortable about it - but its exciting to see that all this that I write has a hard cash value. If you find the ads in bad taste, not appropriate, unpleasant... then please leave a comment and I will take them out.

PPS> Its that time of the year when NASDAQ 100 index is rebalanced. This year Infosys (INFY) is being added to the index and its a great great story for the Indian businesses. Although I am very positive about future growth of Infosys - I am quite uncomfortable with current valuations (PE 45, PEG 1.29) - so I will stay out of it till valuations come down to more reasonable levels.

Labels: Investing, Prosper, Research

   

This is the third article in a series of data analysis pieces I have written on Prosper.com. Links to my previous two articles are there at the end of this article, in case you want to dig deeper. Okay - so what am I saying here? As would be clear to anyone who has read my previous posts, I like data and I like visual representation of data. So here are three charts and what we can infer from these charts. All these charts are drawn on a monthly sampling basis for 10 months from Feb 2006 to Nov 2006. I am starting my analysis from Feb as the Prosper's size before Feb was quite minuscule and not really comparable. All data is from Prosper's Lending Performance page. It takes a while to retrieve all this data but its all publicly available.

Chart 1: Growth of Prosper

This is a chart of dollar value of loan's originated every month from Feb 2006 through Nov 2006.
Two things are obvious from the chart. First - Prosper's growth has been phenomenon. It has grown from half a million dollars of loans per month to more than three million dollars of loans per month in less than a year. Second - the growth has levelled out in last three months - in fact it the loan volume declined in Sep and Nov. Whats going on? My read of the situation is that in Aug Prosper received some good media coverage and it lead to an influx of borrowers and lenders leading to a huge spike in Aug. However, this spike also meant a decline in quality of loans - more loans of lower quality got funded. Prosper then clamped down on its verification process and started rejecting some loans (even when they got funded). Further, some lenders who got burnt after lending indiscriminately in June-July-Aug tightened their belts. Both these combined to reduce volume of loans in following months. However, the long term uptrend is very much intact and we should see continuing growth in Prosper's loan volume.

This loan volume also gives us a fair idea of how much money Prosper might be making. Lets assume an monthly loan volume of 4 million dollars per month and all repayments being reinvested. Prosper charges 1% of loan amount from borrowers - so that makes $0.48 million a year. Further, Prosper charges 0.5% of outstanding loan principal from lenders - so that makes $0.24 million. So a total of $0.72 million. Prosper will also make some money from interest (it does not pay any interest to lenders with free cash in their Prosper account), some from fees etc - so may be we are looking at $0.75 million per year. Lets say that in three years (I guess venture caps will stand by you for that long before pulling the plug) Prosper grows to be ~$40 million per month - not too outlandish considering past growth rates. Then Prosper will be making $7.5 million per year. Now - its not too big an amount but remember that running costs for Prosper is quite low - it does very little advertising, it employs few (maybe 20-25) people and it doesn't cost much to have servers and the site running. Combine this with potential for rapid scaling up and perhaps cross selling and other revenue sources and you are looking at a sustainable, profitable business here.

Chart 2: Distribution of Prosper Loans across Credit Grades

The chart below shows distribution of Prosper loans across credit grades for loans originated in every month from Feb through Nov 2006. The chart is a little busy - so look at the attached table to make sense of the graph.

You can see that the composition of Prosper loans has changed significantly since Feb. Initially AA and A credit category used to be quite big - more than 20% each of total loan amounts in Mar. However, over time, their proportion came down and by Nov - they were barely 20% of total loans combined. This declined has happened because of the increase of share of E and HR category loans. They combined for ~14% of loans in Feb but by Nov, E category alone claimed higher than 14% of total loan amount. B, C and D credit grades have mostly held their ground over the period. So what can we infer from this? Perhaps lenders have become more open to funding low credit loans after previous good experience with them. This is unlikely because except D, all other low credit category has performed miserably. Perhaps word has gotten out that Prosper is a good and affordable source of credit for low credit borrowers and many people with payday loans and such are not coming to Prosper in droves and hence the increase in E and HR category borrowers on Prosper - this is quite possible. There may be other explanations as well. Let me know if you have a different opinion.

Chart 3: Reduction in Credit Quality at Prosper

This is a followup of the previous chart. I thought - well - loan composition is changing - so lets see how the average loan profile is changing over these months. So I created a simple index of credit quality. Under this index - a loan with credit AA has an index of 0, A has an index of 1 and so on till HR has an index of 6. Then I take a weighted average of all loans for the month to calculate a credit quality index for the month. By definition, this index will be between 0 to 6 and higher number will mean lower average credit quality. So how has this average credit quality index changed in past months - here is the chart:

Its clear from the chart that the average quality of credit at Prosper is declining - rather sharply. In Feb and Mar, average credit quality was less than 2.5 - that is - a credit rating better than middle of B and C. It has now changed to higher than 3 - that is - a credit rating worse than C on average. It seems to have stabilized after July to a 3-3.25 level. It will be interesting to see how this develops in coming months.

I must post a disclaimer that this is still early days for Prosper and its numbers have not really stabilized - so reading too much into overall trends may not be accurate. So lets continue to watch Prosper and see what direction all these numbers are taking. I will be updating the above analysis every month or so and post updates on this blog.

In case you are interested in my previous two data analysis pieces on Prosper - here are the links:

• How to select a borrower on Prosper - Part II
• How to select a borrower on Prosper - Part I
  

Labels: Investing, Prosper

   

In my last post on borrower selection at Prosper.com, I emphasized drilling into the delinquencies data that Prosper.com provides. It was in response to StingyJoe's question in his excellent blog - StingyFinance. StingyJoe was kind enough to put a link to my post on his blog post. I am continuing the chain of thought and detailing various kinds of data exploration you can do and helpful insights that you can gather.

If you are a member if Prosper, then you can access all the data on the Lending -> Performance page. You can change performance criteria and see the impact on default rates. Here are the results of my dabbling with this page (all these