The 1 month or worse curves that I published over the weekend have many potential uses in determining the interest rate required to make a certain return.
When the curves are complete we can simply solve the Markov Model since the vehicle is a fairly straight forward 3 year fully amortized loan. If we can assume the remainder of the Markov Model (we are missing months 11-37), we could also solve the Markov Model.
Solving the Markov Model is the holy grail of investing in Prosper, because having done so, one can simply fall back on the law of large numbers, by diversifying in many loans, and make a fairly consistent rate of return. Any large investors rigorously analytical that are going to come onto the platform will undoubtedly be solving the Markov Model.
As a first step to solving the Markov model I thought I would predict the 3 year cumulative default rate. At the moment there are 2 obvious options: linear and exponential decay. While the curves currently look linear it is not possible that they will remain so. In the HR category for example a linear trend would result in a cumulative 3 year default rate of a ridiculous 108%. The next obvious choice is exponential decay.
The thinking is: if 30% default in 10 months, 30% of the remaining good loans will default in the next 10 months, and 30% of the remaining good loans will default in the next 10 months. In other words, 70% of the HR loans survive 10 months; 70% of those (100*0.70*0.70) survive the following 10 months. Or, as a mathematical expression, pct_good_loans(p,r,t) = 100 * (1-r)^(t/p), where p is period, r is rate at period, t is time in months.
For HR = 100 * ( 1 – (1-0.30)^(36/10) ) = 72%.
Yikes! Clearly believable looking at the current curve; although it may turn out to be slightly high as most unsecured debt curves are actually S-shaped even though Prosper’s havn’t turned yet.
I averaged the exponential decay 3 year cumulative default rates for period 6-10. Here are the results:
| AA | A | B | C | D | E | HR |
|---|---|---|---|---|---|---|
| 3.9% | 12.7% | 20.8% | 21.2% | 32.8% | 53.8% | 73.7% |
- 1 Month Late or Worse Curves by Credit Grade I have pulled the public data on the performance tab by hand in order to produce a spreadsheet of loan performance. This spreadsheet was then pivoted...
- Prosper 1 month Late or Worse Curves by Monthly Cohort One of the tricks to determining your ROI is to be able to predict the default curve. Pninen of the Prosper Forums has been tracking...
- Prosper Days Day 2 Update The 2nd session was the Big Lender Panel. It was great. Worth the price of admission. The Lenders (with links to their portfolio) were pensioner, L5, and janeybooboo. ...
- Why Would a Lender Want to Join a Group? — Prosper Group Analysis Part 2 Previously I posed several questions on prosper groups. Here is the second installment of answers. Why would a lender join a group? I think that...
- Group RateLadder Lender Benefits I have sent members of my group the entire underlying spreadsheet to the upcoming 1 month late curves this include the Markov Chains that can...
If you liked this article, vote for it on del.icio.us and stumbleupon.
Categories:
Loan Review, Prosper.com, Statistics, Tools
Tags:
3 comments ↓
I think you’re making a faulty assumption that all credit grades will maintain an exponential decay. Or, more accurately, I think that you’re producing an overly pessimistic model for the good credit grades. Your 1 mo later or worse curves reflect this for AA – C grades.
I suspect that the exponential decay function is not proper for all loan grades. Your “1 Mo or Later” plots show different behaviors in the AA – B/C classes than the HR – D.
Mike
I suspect that you are reading my mind. This post is setting up the rebuttal. As I said, the exponential decay is “slightly high”. The Holy Grail will have to wait for more complete data. But it is fun to speculate.
Leave a Comment