Mad Rush at Lending Club Loan Release Time: Part I
Mad Rush at Lending Club Loan Release Time: Part III - Delinquency Rate and FICO Score Change
Mad Rush at Loan Release Time: Part II - Loan Performance with Time to Fund
In the previous post Mad Rush at Lending Club Loan Release Time: Part I, we analyzed listing volume, issue volume, and time to issue with Credit Grade and Time to Fund. The key takeaway from the last post was that half of listed loans are funded within the hour of being listed, i.e. early bird gets the worm.
In this post we will review the performance of loans with time to fund, i.e. the quality of the worm. As most of the loans in this study have only been issued for 7 to 10 months, we expect to find fewer loans that may have been charged off.
Charged Off and Default Loan Status
Figure 1 below shows the percentage of loans that have been charged off or defaulted with Time to Fund. The solid red horizontal line at 1.081% shows the charged off rate for all loans in this dataset. The gray band shows the 95% confidence interval (0.648 - 1.238%). The highest charged off rate is for the loans funded within a minute (1.875%), whereas the lowest charged off rate is for loans funded between 18 to 32 hours (0.395%). The charged off rate for loans funded within 10 minutes appears to be much higher than that for all loans.
A caution is in order here. As the previous post mentions almost 50% of listed loans that were funded within 10 minutes have credit grade between D and G that carry higher interest rate and have higher risk. It is possible that charge off rate for loans funded within 10 minutes may be higher due to predominantly higher risk of D through G grade loans. To verify this assumption, we decided to review charged off rate by credit grade with Time to Fund. The Credit Grades were divided into three groups:
- Loans with Credit Grade A and B (Loan Count: 10,764)
- Loans with Credit Grade C, D, and E (Loan Count: 11,700)
- Loans with Credit Grade F and G (Loan Count: 1,209)
Figure 2 below shows the percentage of Grade A and B loans that have been charged off with Time to Fund. There is no discernible pattern with A and B grade loan charge offs with Time to Fund. The charged off rate for A and B grade loans funded between 4 and 10 minutes after being listed (1.06%) is highest and more than double of all A and B grade loans (0.48%). The charged off rate for A and B loans that were funded within 10 minutes (0.65%) is about 35% higher than for all A and B loans (0.48%). The charged off rate for A and B loans that took more than 5 hours to fund (0.35%) is about 25% lower than for all A and B loans.
Figure 3 below shows the percentage of Grade C, D and E loans that have been charged off with Time to Fund. The charged off rate for all C, D, and E grade loans (1.46%) is almost three times higher than that for all A and B grade loans (0.48%). There is a steady decline of charged off rate for C, D, and E loans with Time to Fund for loans funded within first 60 minutes after being listed. The charged off rate for C, D, and E loans that are funded within a minute (2.02%) is almost twice of that for C, D, and E loans funded between 11 and 60 minutes (1.02%). The C, D, and E grade loans that take 12 to 15 hours to fully fund seem to have lowest charged off rate (0.43%).
Figure 4 below shows the percentage of Grade F and G loans that have been charged off with Time to Fund. You may notice that there were no charge offs for loans that were funded between 11 to 60 minutes and 15 to 18 hours. The charged off rate for all F and G grade loans (2.73%) is almost double that for all C, D, and E grade loans (1.46%). The charged off rate for F and G grade loans that were funded within a minute (5.65%) is almost twice of that for all F and G grade loans. Also, it appears the Grade F and G loans that stay on the platform for 5 - 15 hours tend to have higher charged off rate.
Overall, the lending strategies that depend on being early to lend to new loans as soon as being released may not be as good as some may believe from default perspective.
Bad to Good Loan Ratio
We would like to evaluate effectiveness of any strategy by comparing Bad to Good Loan Ratio. In peer to peer lending scenario, we assume the bad loans are the ones that are either charged off or defaulted and the good loans are the ones that are fully paid. The Bad:Good Loan ratio is the underlying concept used in Bad Loan Experience (BLE) Risk Index as used by PeerCube and covered in PeerCube March 2013 Newsletter. The strategies that have low Bad:Good Loan Ratio or high Good:Bad Loan Ratio are considered better.
Figure 5 below shows the Bad:Good Loan Ratio (in percentage) with Time to Fund for all loans. The solid red horizontal line at 0.129 shows the Bad:Good Loan Ratio for all loans in aggregate. The gray band shows the 95% confidence interval (0.094 - 0.139). The pattern in this chart is very similar to that of Figure 1. The loans that are funded within 1 minute have the highest Bad:Good Loan Ratio. The loans that are funded between 12 and 32 hours appear to have lowest Bad:Good Loan Ratio.
Figures 6, 7, and 8 show the Bad:Good Loan Ratio for Credit Grades A and B; C, D and E; and F and G respectively with Time to Fund. Following observations can be made from these charts:
- Grade A and B loans appear to benefit from early investing. The Bad:Good Loan Ratio appears to be low for Grade A and B loans that are funded within first 3 minutes of being released or take between 12 and 15 hours to fully fund.
- Grade C, D, and E loans appear not to benefit from very-early investing, within a minute. The Bad:Good Loan Ratio appears to be low for C, D, and E loans that are funded between 2 and 60 minutes and between 12 and 32 hours after being released.
- There were no Charged Off Grade F and G loans that were funded between 11 and 60 minutes and 15 and 18 hours thus creating a gap in Figure 8 for Bad:Good Loan Ratio (Orange line). Once again very-early investing doesn’t seem to pay-off for Grade F and G loans also.
- Lenders using automated systems that depend on speed and trying to invest in loans within a minute of being released may be chasing fool’s good. This is particularly true for lenders who are investing in low quality loans, i.e. loans with Grade C through G. There appears to be some benefit from very-early investing for lenders who invest in Grade A and B loans.
- It appears that the periods of low loan listing volumes available on Lending Club may have created the impression that all “good” loans are snapped up quickly. The data doesn’t seem to support such view.
- Overall, the above analysis points out that the early worms may be sour. It seems the patience might be the virtue that pays off in the end.
In the next post, we will review a few other measurements and predictors of loan performance with time to fund.
- Bryce Mason | Monday June 16, 2014, 2:12 pm
- Speed is a necessary condition to performance. It is not sufficient. One must also have a valid model to know what to buy. Your analysis suggests the existence of uninformed investors who happen to act quickly, attenuating (or even reversing) the true effect of investing intelligently.
- Anil Gupta | Monday June 16, 2014, 5:42 pm
Thanks for the comment. Are you suggesting that uninformed investors are skewing the performance results for loans that are funded quickly? I will expect the distribution of uninformed investors to be random or possibly reverse of what you are suggesting. Uninformed investor is not going to setup automated investing based on random loan selection instead of doing some work in picking loan selection criteria.
- Victor | Tuesday June 17, 2014, 9:04 am
why is your default rate so low? i thought default rates at LC in general is around 3-4%
also, how is this actionable? if I see a loan outstanding at 12 hours, is it going to be a 12-15h loan or will it be an 18h+ loan.
- Anil Gupta | Tuesday June 17, 2014, 10:05 am
- @Victor, The default rate is low because these loans have only been issued for 7-10 months. You are right, making the findings actionable will be difficult. Possibly filtering loans based on how long they have been on the platform plus the percent funded. But your point is good. I will review and analyze the data to see if it is possible to predict when a loan may be fully funded based on funding progress.
- Carl | Tuesday June 17, 2014, 10:38 am
I really like your Key Takeaways. Those who use automated systems that depend on speed to invest in low quality/high risk loans are indeed chasing fools gold. You can use an automated system to flag and save for later analysis potential low quality loans but it is essential that a human analyze each loan before the investor "pulls the trigger" and actually invests in the loan. There are simply too many variables for a computer to muster.
Here's a variable I hadn't mentioned: After analyzing the data of a potential 19% loan I find its not worth the risk: but if the potential loan had an interest rate of 24% I might commit my funds.
I think many folks like the idea of an automated system producing big returns. You can get good returns from Lending Clubs' automated programs. To achieve better returns requires knowledge and time. Many folks don't have the knowledge and don't have the time.
- Peter Renton | Tuesday June 17, 2014, 12:41 pm
Great analysis Anil, thanks for sharing. I have a few comments/questions:
1. Because these loans are so new the numbers are very small as far as charge-offs go. I am wondering how statistically significant when the actual number of charge-offs is often in the single digits or even zero in some instances.
2. Did you see any difference between 36 and 60 month loans? That would be interesting to note.
3. You mention nothing of ROI. I am totally fine having a ton of defaults as long as my ROI is high. Maybe the loans in the 0-1 minute bucket have a higher interest rate than the average more than making up for the higher default rate.
I would love to see you revisit this analysis every six months for the next 2+ years so we can follow this analysis. Personally, I feel it is premature to say investing quickly is not beneficial. While the data certainly looks that way, I would like a more thorough and seasoned analysis before making a definitive call.
- Anil Gupta | Tuesday June 17, 2014, 5:19 pm
@Peter, It is difficult to include everything in one post. I plan to cover different measures of performance for this dataset in future posts. Same for loan terms 36 months vs 60 months and other loan attributes.
The issue of higher interest rate loans in 0-1 minute bucket is addressed by using credit grade buckets. Previous post also showed the breakdown of loan listings by credit grade. The higher interest rate loans in 0-1 minute is a mute point considering the same grade bucket loans are being compared with other Time to Fund.
I plan to keep track of performance of these loans in the future. I have performed similar analysis on smaller datasets in the past in response to PeerCube users' request for automated lending, data has never supported the general perception of "Good" loans are picked up quickly put forward by a few vocal lenders and/or benefits of automated lending. Automated decision has helped Lending Club take out the cost from lending decision workflow for middlemen (loan brokers) like itself, but the cost of decision making hasn't changed for actual lenders who are lending their money with automation. You are trading savings in time with reduction in performance.
- Megan | Saturday June 21, 2014, 8:28 am
Great analysis. I wonder how having loans in your shopping cart vs having the loan truly fully funded would impact this data. If I added it to my cart within sec, but then took a couple minutes to review and decide how much would that impact the "fully funded" time?
does this include both whole and incremental loans?
- Anil Gupta | Saturday June 21, 2014, 9:25 am
- @Megan, the shopping cart lockup will not impact the Time to Fund numbers. Time to Fund is the difference between the time when the loans was first listed and the last time it was detected as available on the LC platform by PeerCube. The dataset only includes fractional loans.
- Taavi Pertman | Sunday June 22, 2014, 1:08 am
I'm pretty certain that time to funding doesn't affect loan quality in any way. You have discovered a correlation that is likely caused by something else and just mediated by time to funding.
You can get the same result if you look at the rate of drowning cases and the amount of ice cream eaten on a day. Although you'd get a similar graph suggesting that eating ice cream causes more drowning, it actually is likely just the fact that people just swim and eat more ice cream when the weather is hot and ice cream has nothing to do with drowning.
Following the same logic here I would then assume that the fastest funded loans have a higher interest rate or some other criteria that makes them attractive for investors who then rush to fund them. If they weren't attractive they'd get funded more slowly, but still default in the same fashion.
Even though you looked at the credit grades to "eliminate" the effect of interest rates, you merged F and G categories. Looking at the rates table on lendingclub homepage, the difference can be around 3% for the rate: https://www.lendingclub.com/public/rates-and-fees.action
I would lend any time with a 3% higher interest rate even if that meant 0.5% higher default rate. It would just be more profitable in most cases.
The conclusions could easily be made actionable, if these were the results according to the actual ROI, not just initial default rates:
1. If you are investing into best quality loans, then you can follow the majority and try to snatch up the best loans fast, as it seems to be giving good results (currently at least from default perspective) and the main criteria for picking good loans seems to be correct.
2. If you have investing into lower quality higher interest rate loans, you should perhaps reconsider your strategy and invest into the loans that actually provide higher returns. If you can't figure out what the difference is within the group, then perhaps just use a slower strategy where you'll miss the fastest funded loans.
However, I somehow think that the current strategy is working and the saying that the best loans are snatched up fast, is correct when you take the interest rate into consideration.
- Filip | Tuesday August 5, 2014, 6:02 pm
- HI Anil, thanks for putting together the analysis. Was an interesting read. I think peter highlighted this in his comment before but I think your data set groups all C-E notes and F-G notes together and looks at funding times. There is a huge yield spread even within Letter Grades (E1's yield 18.99% and E5's yields 22.99%). You're analysis at a summary level isn't picking up the right comparison, apples with apples. Even with a 2% default on the 0-1min funding notes you don't really know the effective yield the investor is walking away with. I'd like to hear more why you think this is a mute point.
- Anil Gupta | Tuesday August 5, 2014, 10:04 pm
Thanks for your comment. See if Part IV provides some answers to your questions.