-
13-10-2018, 03:53 PM
#3861
Originally Posted by myles
Basically showing what Harmoney say the default is estimated to be vs what it actually is (grouped by risk grade).............
I see it now, thanks.
Originally Posted by myles
Summary at top of report may not be completely correct - will be when I finalise the data set.
Okay, thanks.
So, is it safe to assume that (barring the jump over 30% due to debt sold) the default line on page 20 is the up-to-date hazard curve of our data pool?
-
13-10-2018, 05:35 PM
#3862
yeah, nah
Originally Posted by beacon
So, is it safe to assume that (barring the jump over 30% due to debt sold) the default line on page 20 is the up-to-date hazard curve of our data pool?
Nope, it's broken. If you go back to that last time-lapse graph you'll see that when those loans were sold off it affected loans across time, not at a specific time i.e. the loans all actually defaulted at different times, but where sold off together. When they were sold off the date used to determine when they defaulted was set to when they were sold - most appeared in the 25 - 35 'ish month period (100-150 week on time-lapse). Just looking at it now, I don't think it's annualised anyway. I'll remove it to avoid confusion.
-
14-10-2018, 03:47 AM
#3863
-
14-10-2018, 10:43 AM
#3864
Member
Originally Posted by myles
Have updated the summary.pdf document with quite a few characteristic by grade chart sets where they make sense. Some of these appear to offer some hints of how the Harmoney grading process may be working.
At this stage I'll take a break from making new charts unless someone asks for something specific or I think of something useful - if I've missed anything, let me know.
I'll put up the final summary document and unique.csv and raw.csv sometime late tonight. If you find any errors or have problems with these let me know.
Time to digest some of this info
fantastic work myles. Lots of info to make sense of. Reinforces that previous default is something to avoid and to consider scaling down pp loans. Owning home seems to perform better then renting, and suprisingly 20-29 age group performs comparatively well.
The only other thing that could be done (but really getting quite pedantic), is determine the average age of loan in each category to ensure comparing apples with apples. But its probably safe to assume simplistically the loan age mix would be similar.
How are you calculating HM estimated default rate per group in the bubble graph? In "Half yearly paid off loans", does "paid off" exclude part paid?
Thanks again!
-
14-10-2018, 11:02 AM
#3865
yeah, nah
Originally Posted by leesal
How are you calculating HM estimated default rate per group in the bubble graph? In "Half yearly paid off loans", does "paid off" exclude part paid?
Attempted explanation a few posts back - calculated by grouping loans with same estimated default values (i.e risk grade) - it is not time based. Different Scorecard risk grades had different estimated default values so make up different circles.
In case it's not clear, each loan has the Estimated Default Rate recorded with it - so calculation is similar to all other bar charts, but annualised as per Harmoney's estimated rate.
Last edited by myles; 14-10-2018 at 11:14 AM.
Reason: Clarity
-
14-10-2018, 12:08 PM
#3866
Member
Excellent document myles. Just a query about co-borrowers defaults. The chart on p6 shows loans with a co-borrower default at half the rate of single borrower loans yet on p25 co-borrower defaults seem to be almost always at a higher rate than single borrower loans. It maybe the way I'm reading it but I would appreciate your view on this. Cheers and thanks for what is obviously a huge amount of time consuming work on your part.
-
14-10-2018, 01:53 PM
#3867
Member
co borrowers
Originally Posted by joker
Excellent document myles. Just a query about co-borrowers defaults. The chart on p6 shows loans with a co-borrower default at half the rate of single borrower loans yet on p25 co-borrower defaults seem to be almost always at a higher rate than single borrower loans. It maybe the way I'm reading it but I would appreciate your view on this. Cheers and thanks for what is obviously a huge amount of time consuming work on your part.
Myles, I will try to answer this.
Good spotting Joker.
The difference is in the scale of things.
In page 25, for grades where the co-borrower's defaults are higher, the numbers (as in population) are not significant - for example F, only 2 defaults in co-borrowers out of 8 gives it a 25% default rates. OR the absolute difference in percentages is not significant as in A where the single is 0.44% (14/3111) and coborrowers 0.97% (5/517) - an absolute difference of just 0.52%. Both Population size and Absolute percentage difference are important when you combine the grades.
Whereas, where the singles are higher eg E, both the numbers and absolute percentage difference is significant. So E has single 291/2336 = 12.46% and coborrowers 7/72 = 9.72%, a difference of 2.74%.
As an example, if you add E and F, the numbers are single (291+250)/(2336+1203)=15.29% and coborrowers (7+2)/(72+8) =11.26%. So, singles are still way higher than coborrowers. But looking at the two chart (without noticing the numbers) one would have thought that they even out.
An analogy would be if you add a pot of boiling water to a pot of cold water, the temperature of the cold water would rise considerably. But if you add a spoonful of boiling water to a pot of cold water, it does not make much difference to the cold water. Or if you add a pot of slightly warmer water to the pot of cold water, there is not much difference to the cold water either.
Last edited by Cool Bear; 14-10-2018 at 02:00 PM.
Reason: expand analogy
-
14-10-2018, 01:57 PM
#3868
-
14-10-2018, 02:05 PM
#3869
yeah, nah
Originally Posted by joker
The chart on p6 shows loans with a co-borrower default at half the rate of single borrower loans yet on p25 co-borrower defaults seem to be almost always at a higher rate than single borrower loans. It maybe the way I'm reading it but I would appreciate your view on this.
If you add up all the individual numbers (population and defaults) you see that they all add up, so no smoke and mirrors
You need to consider the weighting/proportion of each grade bar to the whole - for example the E grade 'No' bar (appears a little higher) represents 2458 loans at ~12% default rate (305 defaults), compared to say the A Grade 'No' bar (appears much lower) that represents 3158 loans loans at ~0.4% default rate (14 defaults). The E swamps the A on defaults so at the overall level drives up the number of 'No' defaults.
The left and right bars don't represent the same base number of loans, it is the ratio to the base number of loans.
Harder to explain than it should be...?
Last edited by myles; 14-10-2018 at 02:11 PM.
-
14-10-2018, 03:54 PM
#3870
Member
Originally Posted by myles
If you add up all the individual numbers (population and defaults) you see that they all add up, so no smoke and mirrors
You need to consider the weighting/proportion of each grade bar to the whole - for example the E grade 'No' bar (appears a little higher) represents 2458 loans at ~12% default rate ( 305 defaults), compared to say the A Grade 'No' bar (appears much lower) that represents 3158 loans loans at ~0.4% default rate ( 14 defaults). The E swamps the A on defaults so at the overall level drives up the number of 'No' defaults.
The left and right bars don't represent the same base number of loans, it is the ratio to the base number of loans.
Harder to explain than it should be...?
Yep hard to explain - All to do with Mix. Coolbears teaspoon of boiling water was a good analogy.
Co-Borrower is a good example of adverse selection. Two average borrowers (D+D grade) given a B grade loan, does not equal one good borrower (B grade)
Tags for this Thread
Posting Permissions
- You may not post new threads
- You may not post replies
- You may not post attachments
- You may not edit your posts
-
Forum Rules
|
|
Bookmarks