February has been slow month from blogging perspective for me. I spent majority of time making improvements to PeerCube. Three major enhancements released this month are ability to invest in multiple loans together from the list of loans, more robust BLE Risk Index that now includes 19 different loan and borrower attributes, and showing a list of loans with similar risk profiles as the one being viewed by a user.
One such method is to review the loans that are currently in default and when they were issued. This method assumes that similar pattern, in aggregate, will persist for loans in the future. One challenge with this method is that any variation in loan volume will skew the pattern. For example, if 1,000 loans defaulted out of 100,000 loans issued last year versus 400 loans out of 10,000 loans issued prior year, we may erroneously assume that loans default more within a year of being issued. In following analysis, I use the percentage of loans with default status to smooth out any effect of volume.
The chart below shows the percentage of loans with default status as a function of loan issued date. I only included 3 year term loans primarily because I have the historical data that covers the loans from issued date to maturity. Also, the 5 year term loans may have different default pattern. A peculiarity you may notice is the way I have chosen to plot X-axis (loan issued date). Instead of ascending issued year, I have reversed the axis and plotted percentage loan defaults with descending issued year. In fact, the chart below is a mirror image. As we are more interested in knowing when a loan may default, I believe flipping the X-axis better communicates visually the trend.
While reviewing this chart, think of that you are trying to find out how many 3 year term loans issued in January 2013 may default by the end of 2016. Think of right side of 2012 being end of 2013, right side of 2011 being end of 2014 and so on. The major assumption here is that future monthly default trend is exactly represented by the past monthly default trend.
For 100 loans issued in January 2013, 2.3 loans can be expected to default in first year, 3.9 loans default in second year, 4.3 loans default in third year, and another 1.9 loan default after third year.
When Default of Loans Start to Peak?
Recently, a participant on LendAcademy forum asked when defaults of loans start to peak. It is an interesting question so I decided to look further into historical data for Lending Club loans and see if I can find patterns for loan defaults. Unfortunately, Lending Club only provides point-in-time snapshot of historical loan data so I can't observe when a loan entered in default state. But, there are several different methods that can help provide insights from point-in-time snapshot.One such method is to review the loans that are currently in default and when they were issued. This method assumes that similar pattern, in aggregate, will persist for loans in the future. One challenge with this method is that any variation in loan volume will skew the pattern. For example, if 1,000 loans defaulted out of 100,000 loans issued last year versus 400 loans out of 10,000 loans issued prior year, we may erroneously assume that loans default more within a year of being issued. In following analysis, I use the percentage of loans with default status to smooth out any effect of volume.
The chart below shows the percentage of loans with default status as a function of loan issued date. I only included 3 year term loans primarily because I have the historical data that covers the loans from issued date to maturity. Also, the 5 year term loans may have different default pattern. A peculiarity you may notice is the way I have chosen to plot X-axis (loan issued date). Instead of ascending issued year, I have reversed the axis and plotted percentage loan defaults with descending issued year. In fact, the chart below is a mirror image. As we are more interested in knowing when a loan may default, I believe flipping the X-axis better communicates visually the trend.
While reviewing this chart, think of that you are trying to find out how many 3 year term loans issued in January 2013 may default by the end of 2016. Think of right side of 2012 being end of 2013, right side of 2011 being end of 2014 and so on. The major assumption here is that future monthly default trend is exactly represented by the past monthly default trend.
Observations
While reviewing the above chart, two observations right away stand out:- No loans are charged off or defaulted within first six months. This is understandable as majority of loans will go through stages of In Grace Period, Late (16 - 30 days), and Late (31 - 120 days) before being charged off. So, earliest most loans can be charged off is at least 120 days (4 months) after being issued.
- Th peaks appear during the year at regular interval. I am not very sure but I suspect this may have to do either with the time (end of December) when this historical loan file was downloaded or the Lending Club defaulting loans in batches at regular interval.
For 100 loans issued in January 2013, 2.3 loans can be expected to default in first year, 3.9 loans default in second year, 4.3 loans default in third year, and another 1.9 loan default after third year.
Key Takeaway
We may conclude from this analysis that the most default happen during the third year of 3 year term loans. But we need to be cognizant of the fact that the loans defaulting late in their maturity cycle have much lower impact on return as such loans have already paid back greater share of original principal.
In next post, I will further look in to default patterns of loans and also investigate other methods to analyze historical data for default patterns.
Interesting analysis Anil. The point you make in your key takeaway is worth emphasizing. Not all defaults are created equal. If someone defaults in month 35 of a 36 month loan it will have a microscopic impact on my bottom line. I think a more useful analysis would be a breakdown of total dollars lost from defaults. Because that is really what matters to most investors. And I think we would find in that analysis that year 1 has by far the most defaults.
ReplyDeletePeter, I agree with you about early defaults having higher impact on bottom line. I plan to cover lost amount due to defaults as part of this analysis in future post. There are multiple ways to look at defaults and I am hoping to cover as many angles as possible with this series of blog posts.
DeleteThank you for another great contribution. I always look forward to reading your new posts!
ReplyDeleteIs it too early to make any assumptions regarding how the new underwriting rules alter the slope of those lines? I wonder how the loss rate of the first x months of the newest issues compares to the first x months of the older vintages.
Andrew, thanks for your kind encouragement. Comparing the loss rate before and after underwriting rules will be an interesting exercise to see the effectiveness of rules change. Typically earliest the loans are charged off is 6 months after the issue. The rules were changed end of November. So the earliest we can do this comparison will be in May/June. I will make a note of your suggestion for future blog post in May/June.
DeleteVery interesting analysis.
ReplyDeleteI would expect a consumer loan portfolio of amortizing loans to incur the majority of its defaults when the loans are closer to issuance date. Do your findings contradict this? The statement "we can expect 2.3% loans in default by the end of first year, 6.2% loans in default by the end of second year and 10.5% loans in default by the end of third year" appears to say that more defaults will take place in years 2 and 3 than in year 1.
You mention there are different methods to determine when a loan defaulted - have you analyzed the Remaining Principal and Payments to Date fields to approximate a time of default?
Sam, there is slight distinction in this analysis from what you may be thinking. This analysis looked at what percentage of your portfolio will default at what stage of loan maturity cycle. It did't look at percentage of defaults that may happen in a time frame. Actually, the default analysis based on months of payments just got published this morning which is what I think you are interested in.
DeleteAnil, looks like I spoke too soon. I just read Part 2, and its results are in-line with other consumer loan portfolios' performance (at least the few that I have seen).
DeleteSam, yep I realized you made the original comment one minute before the part 2 post went live.
DeleteAnil,
ReplyDeleteThis is a terrific analysis. I wonder if the wintertime increase in default rates relates to an observation that some made over in the LendAcademy forum that LC was seeking out less creditworthy borrowers in order to increase volume?
Regards,
Grant
Grant,
DeleteVery unlikely that winter time increase in default rate was due to LC seeking out less creditworthy borrowers late in the year (most probably you are referring to underwriting changes late 2012). Such defaults will earliest materialize in May of 2013. LC will have to be seeking out low quality borrowers in summer of 2012 or earlier for default rate to increase in wintertime.
Thanks.
Anil