I wanted to get your thoughts on loans where there is at least one public record on file.
Normally I completely avoid those for obvious reasons.
What do you think about notes where the public record was around 10 years ago and everything else looks great (for whatever your personal selection criteria may be) ?
There is a certain argument to be made that someone who recovered from bankruptcy and now works hard to be financially responsible, may be as safe or a safer bet than an average borrower.
No, absolutely not for me.
Public records are almost (but not always, I admit) BK's.
If a person has filed for BK once, and is in debt consolidation mode again, to me, they just havent been able to learn from their past experience.
I absolutely filter them out without hesitation.
I did have a couple notes with public records that looked good otherwise, one defaulted 7 years after their last bankruptcy filing. . . I avoid these now.
All else being equal, I'd take the loan without the Public Record. But how often do you find 2 loans that are equal except for 1 thing?
In any case I've always assumed Public Records are taken into account in the FICO score.
Have I been mistaken all this time?!?
Public record is just one thing among many that affects the probability of a charge off. An attractive loan along many other dimensions may still be attractive with a public record. I buy them occasionally if the math works out.
If you look at the historical loan stats, the presence of a public record has a horrible effect. ROI is halved, default rate is doubled or tripled. No thanks. I am not so desperate for loans that I would buy one of those.
Perhaps if you look *solely* at public record, it has a tremendous effect. But if you look at dozens of factors simultaneously, you will come to realize they all share variance. The unique predictive power of public record is +0.045 to the probability of a charge off.
If you look at it that way, how could you ever come up with useful filters? Every one single criteria has a relatively small effect. But you need to pick *something* to filter on (if you're using filters, which I am). So why not filter out things with a strong negative effect like that?
I don't use filters. I calculate the probability of a charge off event for every loan, compare it to the interest rate, and then generate a list of all loans available for investment ordered by expected profit. I have a threshold value where I would be willing to invest in any loan above a certain expected profit. Back testing this strategy suggests returns in the top 10% of any strategy I have come across and has a huge upside compared to a filter strategy: on any given day I could find dozens of loans in which to invest.
Brycemason's method is actually very quick once you have the programs written and in place to go through the loans and it's much more effective then using a blanket filter through LC or something like that. It just takes some time to initially put it together then the system is pretty much set and forget with a quick glance over the note before you pull the buy trigger.
I spent the first part of 2012 making the programs and algorithms, and have refined and tested them (with $35k of my own money) this year also. I'm eager to launch a new website service where people will be able to take advantage of my work later in the month--at least a beta version.
It only takes about 2 minutes to make a beautiful Excel sheet with my picks, and then another handful of minutes to make the investment. Couldn't imagine doing it any other way.
When the API comes online I will make a server service that continuously runs my algorithms and automatically invests any idle balances. Or perhaps I will work with others who have developed such systems already. There's some good stuff brewing!
If you get enough loans in which to invest with a strict filter strategy, that's great. It's hard to scale up a filter strategy to sell, though, so that there's enough "product" for all your subscribers.
Many of those questions will be answered on my website within a couple weeks. However, I don't really want to give statistics lessons over the net. A few stats courses and you'd have enough knowledge to do what I'm doing.
By the top 10% thing, I mean that I programmed numerous strategies, including filters, and then compared the returns against each other. I used inception to Q2 2008 to train my modeling and tested all strategies on Q3 2008 to Q2 2009 loans. Their outcomes are known at this time.
Wow, thanks everyone. I wasn't checking this when I purchased loans on FolioFN, and I just noticed that 2 out of 3 of my >31 days notes had public records on file.
brycemason - looking forward to your site too.
I personally think that if you've gone through bankruptcy once and you are doing debt consolidation, you are more likely to do it again just because you are more familiar with it. At any rate, I avoid them generally. I think 2% of my notes have PRs
Sure, the method will be described in general on the site. It's going to be
www.p2p-picks.com, and we're just in the last steps of beta development. Maybe a handful of weeks? It's a side project for me, so it takes a long time to do anything.
The site will have extensive sharing of the results of the strategy, but maybe to the dismay of some people, it will not be couched in a percentage per year / APR standard. I wanted to reduce the number of assumptions in describing the performance of the picks, so I took a buy and hold (and not reinvest payments) approach to calculating everything. It's basically a percentage return over three years. If I had wanted to give a %/year like most people, I would have had to do all sorts of assumptions that have to do with reinvesting payments. As I only wanted to demonstrate that my picks had higher performance than other strategies, I didn't need to do an APR return. I just needed an easily understandable and calculable metric.
I look forward to feedback in a few weeks!
Three years of history and 6000+ completed notes during one of the darkest times in 80 years seemed like a fair training ground to me. People seem to think 6000 is some small number on which you can't base predictions. It is one on which I rest my investment choices, which evolve as more data comes in. Do what you like =). Plenty of room for all comers here.
Let's consider your two propositions:
1) Selection bias in who took out loans in 2008-9. Empirically we have seen principal loss curves improve for more recent vintages (see my guest blog post a few months back), meaning more recent vintages are better. If the best borrowers were selecting into LC in 2009 because they were the only game in town, they sure didn't behave very well. I'll take the more recent vintage performance any day.
2) Things are constantly changing. Some things don't really appear to change very much. I chose variables in my model whose coefficients have remained stable regardless of whatever time period I'm considering. I'd run my models six or seven times, lopping off old quarters each time and many things appeared to have substantially the same relationship to the outcome under study.
Best Wishes!
I never do.
Rawraw,
It's a good idea to compare the standard deviation of the asset class to the returns vs other asset classes, but I haven't made that chart. P2P lending right now is in rarified territory with having low variance and great returns.
Thanks for your interest in the site! I'm looking forward to launching shortly.
Bryce
Bryce,
Look forward to the site. Is there an email signup list?
Anyone interested can send me an email at
[email protected]. I'll add them to the distribution/announcement list.