A Review of Lending Club Defaults

[Update: Graphs have been updated with data through Q1 of 2016.]

In Peer Lending, similar to bonds and other types of loans, a substantial portion of investment performance depends on the probability of default for the investments.

As such it is useful to monitor default rates, not only to elucidate past performance but to form reasonable expectations of future returns. We’ll examine Lending Club and their defaulting loans briefly to see how default rates have changed over time and if there are substantial deviations from historical default averages.

First, we check our hazard rate and historical probabilities of default per grade by iterating month after month and checking how many loans we expected to default in each month compared to how many loans actually defaulted in each month. An adjustment was made to the most recent months, where there are loans of late status that are likely to default but could not default since they are not past 120 days old. Examining numbers of loans soon becomes hard to track as the number of loans issued and loans defaulting grows exponentially, so we also provide a plot of the log(default loan count).

Screen Shot 2016-05-20 at 10.36.14 AM

Note how the log(number of newly issued loans) appears parallel to the log(default loan count), which suggests that the rate at which number of defaulting loans is increasing is proportional to the rate at which newly issued loans is increasing. If the slope of log(default loan count) were more positive than the slope of log(number of newly issued loans), that would show that the rate at which defaults were increasing was outstripping the rate at which new loans were being issued, which could indicate increasing probabilities of default across the platform. This does not appear to be the case. It is worth mentioning that the log(actual default loan count) starts in 2008. Our data shows that the first loan defaulted in January 2008, so since the actual amounts of defaulting loans was 0 prior, the logarithm in those periods is undefined.

Next, we look at the percent error \((\frac{Actual – Theoretical}{Theoretical} * 100\%) \) of our number of defaulting loans predictions. The dashed line is the simple average of the monthly percent errors.

Screen Shot 2016-05-20 at 10.55.17 AM

We point out that percent error before the start of 2008 is 100%, but fairly meaningless. The fact that there were no actual loans defaulting, as well as there being only fractions of a loan being predicted to default, combined with the small number of loans in existence make the percent error uninformative during those times. If the percent error is below 0 or above 0 that means we over-predicted or under-predicted the number of defaults, respectively. The years 2013 and 2014 appear to show below average numbers of defaulting loans, which suggests that performance on Lending Club was relatively good during those times since defaults were lower than expected. The movement of percent error into positive territory in the most recent times suggest that there are more defaults occurring than we expected. One possible explanation for this is that loans that were issued in 2013 and 2014 that were going to default did default, but later than we predicted, which would explain the period of over-predicting followed by the period of under-predicting defaults.

Finally, we examine historical and/or projected default rates for grades annually on a moving quarter basis. Examined loans have a term of 36 months and dashed lines are simple averages of quarterly values until one year ago. Quarterly values more recent than one year ago are excluded from averages to prevent the inclusion of default rates that are heavily based on prediction.

Screen Shot 2016-05-20 at 10.36.38 AM

To generate this graph, we started in Q3 2007, looked at all loans issued from Q3 2007 to Q3 2008, split them into their respective grades, and looked at what proportion defaulted. We then moved to the next quarter (Q4 2007) and repeated the process until Q1 2016. For quarters where loans are still ongoing (e.g. 2013 – 2016), we counted the loans that had actually defaulted and added to them the number of loans that we predict will still default in the future. Again, on this graph we see a period from 2013 through 2014 that default rates appear lower than average, followed by some increase in default rates towards the end of 2014 and extending through 2015. This trend is similar to the one noted in the percent error graph above.

While this analysis is fairly simple, there are a few points that an investor can glean. First, there does not appear to be any long term differences in the rates at which loan issuances and number of defaulting loans are growing, which suggests that there has not been anything that greatly alters the proportions of loans failing at any given time. Second, that from around 2012 to 2014, there appeared to be a period of above-average loan performance in the form of less defaults. Third, that this over-performance appears to be in the process of correcting itself, but not to a degree that deviates significantly from average past performance. Finally, and perhaps most importantly, that for Peer Lending (defaults) it appears to be business as usual.

 

 

2016 in 6 Numbers

Leave a Reply