Print

[zcommodore]

I've just posted in my blog about how loans "age" and the rate at which loans go late seems to change over time.  I figured this may merit some discussion so I thought starting a thread here would be a good place to post comments.

Read my post, then let me know what you think, either there or here.

[j9359]

Good Stuff!
Confirms my suspicion that a lot of loans go bad very early.  I wonder if many of the early bad loans are planned?  In other words borrower goes into the loan knowing that they aren't going to make it.  After this early period things settle down so that only the normal things like job loss, sickness, etc.. kick in
john.

[mjerryfirst]

Nothing here that is useful to a prudent lender

[ira01]

Nothing here that is useful to a prudent lender

http://www.prospers.org/forum/what_members_think_of_mjerryfirst-t7380.0.html

[zcommodore]

Nothing here that is useful to a prudent lender

Pride goeth before a fall....

There has been more than one lender who thought he had it all figured out.  One of the things I observed from my study is that better loan picking picks better loans at first.  Unfortunately, over time, there is a regression to the mean.  Before three years are out, you will have defaults as well.  I don't know how many but they will happen.

[yankeefan]

Thanks for another thoughtful blog topic.

One dimension of the late loan/time relationship that isn't obvious is the changing average credit level of the remaining loans.  That is, the HR loans tend to go late quickly, and so there are relatively fewer to go late in the out months, while the AA loans (the ones that don't repay early, anyway) stay around.  Some, perhaps a large, portion of the overall improvement could be from the (postulated) imrpovement in average credit quality.

This effect may also explain some of the changes seen in your other graphs.  The extended credit data and lender guidance skewed the funded loans more and more away from the poor credits, so that the credit levels are better to begin with, and don't have much room to improve with the default of teh HR.

[mjerryfirst]

Nothing here that is useful to a prudent lender

Pride goeth before a fall....

There has been more than one lender who thought he had it all figured out.  One of the things I observed from my study is that better loan picking picks better loans at first.  Unfortunately, over time, there is a regression to the mean.  Before three years are out, you will have defaults as well.  I don't know how many but they will happen.

Death, natural disaster, illness and loss of employment should result in a default rate of 3% or so.

Please explain your concept of "better loan picking."

[mjerryfirst]

Good Stuff!
Confirms my suspicion that a lot of loans go bad very early.  I wonder if many of the early bad loans are planned?  In other words borrower goes into the loan knowing that they aren't going to make it.  After this early period things settle down so that only the normal things like job loss, sickness, etc.. kick in
john.

A lender who bids with no idea that or with reservations that the borrower can make it,  maybe shouldn't be bidding

[mjerryfirst]

Nothing here that is useful to a prudent lender

http://www.prospers.org/forum/what_members_think_of_mjerryfirst-t7380.0.html

Ninety percent of the posts on this forum belong on the "OFF Topic" board.  Less than ten percent of the posts help persons become better Prosper lenders and borrowers

[zcommodore]

Thanks for another thoughtful blog topic.

One dimension of the late loan/time relationship that isn't obvious is the changing average credit level of the remaining loans.  That is, the HR loans tend to go late quickly, and so there are relatively fewer to go late in the out months, while the AA loans (the ones that don't repay early, anyway) stay around.  Some, perhaps a large, portion of the overall improvement could be from the (postulated) imrpovement in average credit quality.

This effect may also explain some of the changes seen in your other graphs.  The extended credit data and lender guidance skewed the funded loans more and more away from the poor credits, so that the credit levels are better to begin with, and don't have much room to improve with the default of teh HR.

You bring up some very interesting points.  I wanted to create graphs based on credit grade but realized it would take a lot more time to do it so I just went with all loans for now.  I may go back and separate everything out by credit grade but don't hold your breath waiting.

I haven't memorized all the data I've collected but I noticed that the higher grades, particularly AA grade borrowers tend to pay off their loans earlier so to some extent those borrowers are removed from the data about as fast as the HR's that default.  I don't have a good feel for how the average credit grade changes over time though.

[Risk_Reward]

I think there is a problem with your analysis, unless I am missing something.  You reduce the denominator by paid of loans, but it should also be reduced by late and defaulted loans.  A loan that is already late cannot go late in a given month.  This is causing at least part of the downward bias over time.  Later months are rewarded for high late rates in earlier months.  Prosper gives out lots of these rewards.   

You should have the following formula (late loans at end of month /  current loans at beginning of month).  current loans = all loans - paid off loans - late loans - defaulted loans - repurchased loans. 

[zcommodore]

Nothing here that is useful to a prudent lender

Pride goeth before a fall....

There has been more than one lender who thought he had it all figured out.  One of the things I observed from my study is that better loan picking picks better loans at first.  Unfortunately, over time, there is a regression to the mean.  Before three years are out, you will have defaults as well.  I don't know how many but they will happen.

Death, natural disaster, illness and loss of employment should result in a default rate of 3% or so.

Please explain your concept of "better loan picking."

I'm very curious how you came up with the 3% number.  Do you have something concrete that you've based it on?  Remember borrowers are affected by those around them, not just themselves so just because a borrower is healthy doesn't mean a family member won't get sick or injured and need expensive medical care.

My comment regarding "better loan picking" was based on the graphs in my post.  If you look, you will see that as improvements came on the site and lenders learned to pick higher grade borrowers, the rate of defaults out of the gate went down.  My point is that loan picking skills affect the loans chosen in the near term.  However, as time goes on, the chances of bad things happening goes up and the better loan picking skills have less effect on the overall performance of the loan.  My personal opinion is that your 3% default rate is very optimistic.

[zcommodore]

I think there is a problem with your analysis, unless I am missing something.  You reduce the denominator by paid of loans, but it should also be reduced by late and defaulted loans.  A loan that is already late cannot go late in a given month.  This is causing at least part of the downward bias over time.  Later months are rewarded for high late rates in earlier months.  Prosper gives out lots of these rewards.   

You should have the following formula (late loans at end of month /  current loans at beginning of month).  current loans = all loans - paid off loans - late loans - defaulted loans - repurchased loans. 

I was about to post "you missed something" when I went back and looked at my calculations.  You were right that I made a mistake.  I've subsequently gone in and corrected the calculations and updated the graphs to the right numbers.  Now the calculation is new lates divided by current loans.  The intent all along was to use the number of current loans as the denominator but I got mixed up when I was actually doing the numbers.  I was originally thinking that I was subtracting the lates from the previous month from the calculations so it didn't matter but really I was only subtracting them from the numerator which was obviously wrong.

Interestingly enough the graphs don't look a whole lot different now than they did before.  The line looks a little flatter over time but it still appears to go down as time goes on.

[Fred93]

Good stuff.

This agrees with my (less detailed) observations.  The default rate (or go-bad rate or whatever you want to call it) appears to decrease over time.  I noticed that when I tried to fit exponential (with a negative exponent) curves to those charts of late rates I publish from time to time they didn't fit well.  When I added a 2nd parameter to the model, then I could get a curve that fit better.  Now adding parameters (ie degrees of freedom) always results in a fit curve that fits better, so this doesn't imply the cause. 

There's one sort of intrinsic cause.  The whole-prosper portfolio is a mix of very different credit grades.  The HR loans of course go bad at a much higher rate than the better loans.  The rate for the really low grade loans, HR and E especially is so high that the mix of credit grades changes as the loans age.  So one reason that older loans have a lower default rate is simply that they are a different mix of credit graes.  I wasn't the guy who realized this.  Someone pointed it out to me.  Don't remember who.

So that made me realize that instead of an exponential with a time-verying exponent (ie uniform portfolio with a decreasing default rate vs time), I should have modeled this as a sum of exponentials each with a different exponent.  (ie different default rates for different credit grades, or some such catagorization).  At that point I ran out of energy.

So if you get back to this and have the energy to plot by credit grade it would be interesting to see if this changes the result.