Just in case the region chart was misunderstood - the split between population >10 <=10 (i.e. blue/orange) was based on the population in the data group, not the population of the region.
Printable View
More:
Attachment 9986
Plot of 'loan payment' / 'income' ratio - in general a rise to the right, as expected - not at all clear cut though. Might need to include more info e.g. loan size, grade?
Attachment 9987
Does not include groups with less than 10 to try to 'lift' the detail, but it needs to be interpreted - in general the 'more gaps' to the right suggests the longer the better (outliers hide the detail).
Attachment 9988
Added two more to this one:
- payment protect doesn't have the big jump that some thought it might (they are higher though)
- re-writes stand out (for me at least)
Seems you were onto something, leesal. Have been observing for this since you hinted at the possible connection. Big lumps are being filled and note availability fall fast in many loans, as I'm being served this error screen in many loans too - and repeatedly as I retry for fills
A huge pile of loans on.
Further to the red screens - am getting them consistently when placing. Could it be system overload?
In those loans, a D5 with no less then 5 defaults recorded. Has to be a typo surely! Also a rewrite with 16 successful payments.
Attachment 9992
Yes I did catch that, but forgot that you excluded the regions without claims, most of which would have been rural locations
A lot of the data is confirming what most would expect. Good way of mining insights, won't ever tell the full story, but helps paint an overall picture. Payment protect having a higher default % was something I hadn't considered - wonder whether their is a difference in grading mix between those on PP and those without.
Numbers start to be a bit dodgy at those levels - pretty much across the board for those with significant detail (an obvious bump up in E). Partial PP looks like a waste of time and the values are too small to have much meaning.
Attachment 9993
Note: This could be influenced by other criteria that I use to select loans e.g. loan/income ratio?
Well I've sent harmoney a large list of loans that have extremely obvious reporting errors
Broken down into the following main areas
Showing as Paid off but still owe money = 67
Showing as Current but $0 owing = 54
Negitative outstanding principal = 17
They have said "The team has been made aware of the loan IDs and problems with each as detailed in your email and look to have this corrected soon. I don't have a timeframe at this stage as the list of loans with errors is, as stated, large."
Only time will tell how long they will take to fix these
Trying to come up with some charts that give a good overview of a portfolio. These two I think give a lot of condensed detail:
Attachment 9998
This show the age of all loans (top of yellow), current loans (height of yellow) and paid off loans (pink).
Also shows the 'hazard curve' ~ where loans are defaulting, as well as where loans are going into arrears. (hazard curve similar to Harmoney's but needs a bit more time to develop)
Interesting for me is that arrears are appearing more frequently in older loans, but that is still where the bulk of my 'current' loans are.
The 'Paid Loans' plot shows major loan 'churn' between months 5 and 10 - it will tail off to the right as time goes on.
The yellow 'horn/wave' on the right is my 'startup' loans moving through.
Attachment 9995
Overview of the status of all loans per grade. The issue will be that the 'Paid Off' area will push down everything over time - I did try weighting newer loans over older loans which looked good but values are then meaningless.
Any thoughts/comments on improvement? You do have to spend a bit of time figuring out the detail they are showing.
Rejigged that first chart and fixed some more data errors - it's now more focused on lender loans rather than loan characteristics:
- you can see where you current loans are - in time
- see where/when you'll get most defaults
- still see the churn (though not as well)
Not sure if arrears fits - it will change all the time and perhaps isn't that meaningful?
Attachment 9999
I can't quite work this out, perhaps I've been looking at it for too long... It appears that the default rate for newer loans has significantly dropped since the start of this year i.e. I've not had any defaults from loans that started this year!
I've gone through and checked that the flagging of loans as 'Charged Off' is correct (it is based, on date of last payment).
Has the debt collection process 'scared' borrowers into paying or paying off loans, or is this something else? Or have I got this wrong?
Attachment 10002
Blue: rate of loans that default (i.e. defaults/total loans per month as %) against loan start date
Dotted: number of loans purchased
Red: 9 months default free
Purple: arrears go back 6 months only (there are no major errors in the allocation of arrears)
So, it appears that since the start of this year, loan defaults have dropped on new loans - suggesting that changes to borrower selection or debt collection has made a difference?
Excellent if it has, but I just can't quite convince myself that I might not have this wrong.
The only thing that could have influenced this was that I tightened up on loan selection in the lead up to Christmas, but I don't believe that would explain it...
I think I've figured it out...
Loans don't (typically) start defaulting until at least 2 months into the loan (borrowers can pay the first couple of instalments), with the arrears process then taking 180+ days (6 months), plus a month for the processing gets you to 9 months. However, it does appear that default rates may be on the decline as typically some defaults come through well before the 180 days...
So trying to get some meaningful information on defaults is going to have to be based on at least 9 month old data :( [unless you use arrears info]
The reason I'm interested in this is because it allows you to create stats like the table below for current loans - the missing detail in being able to determining this is the default rate, so it needs to be 'calculated' based on historic info (or something else), which can be problematic...
Loans Principal Interest Fee Default Tax Income MRAR 1251 $121344.46 $26996.53 ($4049.48) ($3002.13) ($2834.64) $17110.29 14.10%
Grade Loans Principal Interest Fee Default Tax Income MRAR A 10 $373.55 $47.46 ($7.12) ($18.97) ($4.98) $16.38 4.39% B 143 $12631.25 $2071.81 ($310.77) ($134.26) ($217.54) $1409.24 11.16% C 493 $46936.72 $9339.55 ($1400.93) ($1238.18) ($980.65) $5719.79 12.19% D 436 $46098.19 $11152.73 ($1672.91) ($602.01) ($1171.04) $7706.77 16.72% E 167 $15235.64 $4357.85 ($653.68) ($1008.70) ($457.57) $2237.89 14.69% F 2 $69.11 $27.14 ($4.07) ($0.00) ($2.85) $20.22 29.25%
The above is calculated on my 'current' loans using 'default rates' from loans older than 10 months, going back a further 12 months to get a weighted average. Ignoring A's and F's due to low numbers (note I'm using a tax rate of 10.5%), it actually looks pretty close to what I'm getting (compared to XIRR calculation on total loans).
I've been looking around to see what others (academics) have done to try to determine default rates - some are using a 'derived' weighting on arrears so it looks at more recent data.
Has anyone else looked into this and come up with any good options?
Note: My B grade loans are weighted on the high side (i.e. mostly B3, B4, B5, so B grade is showing higher than what an average spread would be).
Nice observation. Going to really watch that. I've been keen to grab pp loans, due to the extra premiums available at the same grading (and few reports of envocations). Your data suggests a near doubling of risk at the B to D grades. Am also noticing that that 3 or my 4 defaults, and 3 of 4 60-90 days are pp loans.
You are correcty Myles about the 9 months give or take, I'm benchmarking my cohorts against HM's static loss data - H1 2018 is currently at 0% 3 months through - which is the same as all the other cohorts except 2016 H1 at a tiny 0.03%. Looks like the 7th month (approx 13 months through), is when defaults from a cohort start approaching a meaningful level as the 120+ days start converting to charge-offs.
Attachment 10003
I'm not sure about the value of estimating default rates to give estimated returns, always seemed to me to be somewhat manufactured. Instead I tend to run my actuals alongside benchmarks derived from HM data (Weighted Ave of HM assigned expected default rates ), and run comparisons on the monthly cohorts. Each line in the table has a separate sheet where forecasts are derived from plugged in data (the expected interest rate, and expected default and some others), and the running and projected defaults/RAR etc are returned. EG the first capture represents all loans, 2nd is only "D" grade.
Attachment 10008
Attachment 10009
I'd rather get my average (not estimated) rates from my own data which is likely to be a better match than using Harmoney 'all in' averages?