Loan Prediction with Deep Learning / A.I.

[cameron]

Hi all, I have been investing with Lending Club for a few months now using a deep learning model, and I thought I'd share the work I did if it's of interest to anyone else. All the code and analysis is at http://scaubrey.github.io" class="bbc_link" target="_blank">http://scaubrey.github.io. Cheers.

[SLCPaladin]

I read your work and found it very impressive, thanks for sharing! The extent of my programming is an intro to C++ and Java in my freshman year of university, so I wasn't able to follow everything in its entirety. But I want to ask you, how have your returns been doing and how many notes have you invested in? I am curious as to the answer to this because your model seems to suggest E and F notes plugged into your model have a higher predicted return rate. It seems that with the latest increase in defaults over the past year or so that the lower grade notes have been the worst performers. Obviously, the model can only go off historical data, and some of the recent deterioration and uptick in delinquencies may not have been apparent in the model data you trained your algorithm with. I'd be curious to hear if a more recent application of the data is showing that A/B or A/B/C notes are having a higher return?

[panther02912]

Thanks for sharing. Very interesting. Are thinking about developing this work more, like more features or a more granular output performance?

[Skeptical]

Nice work. Like you, I am new to Lending Club. In fact, I bought my first note this morning. I intend to be methodical and patient when buying my notes.

What strikes me about investing in the stock market and investing in Lending Club are the similarities and differences. It is the differences that are most pronounced for me. I have some passive stock market investment strategies and I am comfortable with them. But I have one account that I actually trade. I would never let a stock go to zero. But I am not sure I could sell a note, so there is a possibility the note could go to zero. This is a big difference. Your returns at Lending Club are capped. Whatever the highest interest rate is, that is all you can make. In the stock market a stock can gain 100%, 200% or a 1000%. This really helps the performance of the portfolio and erases a lot of poor performers.

It seems to me that one will want to stay away from D, E, F and G notes because of the greater risk of the borrower defaulting and one having to take a charge off.

Also, not appreciated is, when you buy a lower grade note is the possibility of going on a losing streak. When I trade stocks in my active account, I am always aware (even with the greatest research) that I will pick losers and I may pick them in a row. Out of 30 trades there is almost a 100% certainty that 6 trades in a row will be unprofitable. And a 67% chance that 8 trades in a row will be unprofitable. When buying D, E, F and G notes, it is easy to go on a losing streak. Most investors don't factor the losing streaks into their investing equation. That is why most investors throw in the towel. To be successful, you have to have the mentality of a trader and factor in the losing streaks that will come and then how to survive them.

I remember reading an article in the Wall Street Journal in the late summer of 07. Matthew Rothman ran a quant fund (no emotions, just numbers). And this sentence stood out for me, "Events that models only predicted would happen once every 10,000 years happened every day for three days." Models are useful but models can never account for what people will do.

I'm looking for a 5% to 7% return from Lending Club. I think that is reasonable, so I don't have any illusions I can pick the profitable D, E, F and G notes with any regularity.

Good luck with Lending Club. 

 

[TravelingPennies]

@SLCPaladin @Skeptical

I am currently invested in only about 150 notes (LC is a small portion of my investment portfolio) and Lending Club today has my ANAR at 13.65%. I have a young portfolio though, just 6 months old. The historical data I was working with showed that the lower grade notes outperformed, but I don't know how that has changed now. I would question the nature of the change if one has occurred though. It may be a systematic population change (e.g. economy effecting all people), a borrower change (e.g. higher number of unreliable people are applying for LC loans), a LC change (e.g. new algorithm for assessing risk), or something else. The modeling I've done can't handle the 1st type of change without retraining on recent data, but potentially can handle the 2nd and 3rd types of changes.

https://forum.lendacademy.com/index.php?topic=4419.msg40664#msg88888888Quote"> from: Skeptical on April 28, 2017, 02:49:48 PM

[TravelingPennies]

@Cameron

Thanks for the response. I hope you'll add your data to lascott's ongoing spreadsheet to track returns of users across the platform. I'd be curious to see how your performance stacks up against the other forum participants. My portfolio's age is about 26 months and I've dipped from about 12% ANAR to about 7% ANAR now. The data seems to indicate that notes below A and B have been particularly bad in the past couple of years, mostly due to increased defaults. Unfortunately, I had a lot of those lower grade notes, so I rotated out into higher quality notes with my reinvestments over the past 18 months. I think that was a good decision. For now, my reinvestment is on pause. I have quite a few more notes than you (7000). LC is a small part of my investment portfolio.