Author Topic: Portfolio (Loan Picking) Strategy  (Read 4022 times)

dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Portfolio (Loan Picking) Strategy
« on: April 27, 2016, 05:38:45 PM »
I've had discussions touching on how I pick loans for my portfolio.  I've never laid it out fully. 

I would like to
a) get thoughts and feedback on mine (follow up post)
b) hear how others build their portfolios of loans

Obviously people have their secret sauce. I'm not asking for specifics, just an overview.

dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #1 on: April 27, 2016, 05:55:47 PM »
My strategy/framework is broken into 2 components, Expected Return and Allocation.

Expected Return:
I start by scoring each loan in a model.  This produces a "score" that is highly correlated with return. 

Next I apply a simple filter, all scores below the threshold that correlates with 0 expected return are removed.

I sort loans by grade.

Each grade has a prior distribution of scores conditional on the filter above.  I find the percentile rank of each loan in that distribution.

I then filter only loans greater than a cutoff value.  This cutoff depends on how picky I want to be.  Remember, these are all loans I expect to have positive returns.

Allocation:
I am not swinging for the fences here, so I want to control the risk in my portfolio. 

Here I want to include macro economic variables.  I am still working on the model, but the point is to correlate and stress defaults.

I run a simulation based on the model above.

Using something similar to an efficient frontier, I then create an allocation that fits my risk tolerance.

Loan Buying:
I look at my current portfolio, the allocation, and decide how many loans I need in each grade.  I pick that many loans from each grade in the available loans (surviving the process above) starting with the highest scored loan.

Happy to hear thoughts and suggestions?

rawraw

  • Hero Member
  • *****
  • Posts: 2790
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #2 on: April 28, 2016, 09:45:14 PM »
I buy loans with good metrics in larger proportions than those with weak metrics.  And I monitor my portfolios for their aggregate characteristics to determine what sorts of loans need to be in the portfolio.  While your method sounds good, it's more work than I want to do for already priced consumer credit.  "expected return" is the key in your method and you don't tell us how that is calculated

dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #3 on: April 28, 2016, 09:59:27 PM »
Currently I use a logistic regression.  I've played with decision trees, neural nets, and a few others.  I haven't found anything that greatly outperforms the logistic regression out of sample.  The simplicity of the logistic regression is a big plus -- it's easy to understand.

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #4 on: April 29, 2016, 03:15:04 AM »
How long have you run this strategy?

If you see your portfolio here https://www.lendingclub.com/info/statistics-performance.action, and see yourself in the 90% percentile, I'd say don't ask and don't tell.  8)

dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #5 on: April 29, 2016, 08:34:59 AM »
How long have you run this strategy?

If you see your portfolio here https://www.lendingclub.com/info/statistics-performance.action, and see yourself in the 90% percentile, I'd say don't ask and don't tell.  8)
Since Feb 1 of this year.  So not long enough.

rawraw

  • Hero Member
  • *****
  • Posts: 2790
    • View Profile
Portfolio (Loan Picking) Strategy
« Reply #6 on: April 29, 2016, 08:41:25 AM »
A regression over what time period? What is your out of sample period? Sounds like you are doing fun stuff. And maybe I haven't worked on the street as long as Fred, but I'm in the top of that chart but don't think I do anything difficult. It's not like I'm running neural networks

Sent from my SAMSUNG-SM-G935A using Tapatalk


AnilG

  • Hero Member
  • *****
  • Posts: 1109
    • View Profile
    • PeerCube
Re: Portfolio (Loan Picking) Strategy
« Reply #7 on: April 29, 2016, 01:58:18 PM »
As rawraw hinted before, you most probably running into multicollinearity if you are using sub grades along with other credit attributes. Logistic regression is the basis of most credit scoring models. Have you done any readings on credit scoring and credit risk assessment? It may be of help in improving your model specially 3 C's of credit scoring and decision trees.

Currently I use a logistic regression.  I've played with decision trees, neural nets, and a few others.  I haven't found anything that greatly outperforms the logistic regression out of sample.  The simplicity of the logistic regression is a big plus -- it's easy to understand.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

Rob L

  • Hero Member
  • *****
  • Posts: 2108
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #8 on: April 29, 2016, 06:13:38 PM »
Currently I use a logistic regression.  I've played with decision trees, neural nets, and a few others.  I haven't found anything that greatly outperforms the logistic regression out of sample.  The simplicity of the logistic regression is a big plus -- it's easy to understand.
You may have set yourself for more questions than answers, but smart experienced guys like Fred, rawraw and Anil may lend a hand. So .. if it's not "secret sauce" what's your definition of good/bad (won't ask about ugly)?

Rob L

  • Hero Member
  • *****
  • Posts: 2108
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #9 on: May 01, 2016, 05:45:47 PM »
In the interest of promoting a dialogue like "regression over what time period", or "what is your definition of good/bad":


dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #10 on: May 01, 2016, 08:11:34 PM »
First, thank you to all for the feedback.  All good so far.

I've been away this weekend, but I will be looking at the multicollinearity issue early this week.  I will report back what I see.

Good/bad is full payment/default over the life of the loan. 

I separately look at the loss given default.  My simulation in the allocation step takes into account the range of losses possible.  That distribution is the observed empirical losses calculated out of the big payment history file.

Rob L

  • Hero Member
  • *****
  • Posts: 2108
    • View Profile
Re: Portfolio (Loan Picking) Strategy
« Reply #11 on: May 09, 2016, 06:32:57 PM »
Gee, I had my day whole planned out; living the life of a semi-retired engineer.
Write some new code to do a better job of cleaning, prepping and validating the LoanStats files prior to modeling.
Then listen to the LC webcast at 5pm and find out how the past quarter went. I was optimistic.
The plan went horribly awry upon learning of RL departure et. al. in the AM.
Started to code a bit and felt like I was re-arranging the chairs on the Titanic. Bummer.
Then I set back and watched LC's stock and felt like I was on the Titanic.

AnilG

  • Hero Member
  • *****
  • Posts: 1109
    • View Profile
    • PeerCube
Re: Portfolio (Loan Picking) Strategy
« Reply #12 on: May 09, 2016, 07:47:35 PM »
Well said, sounds like my day except you didn't have to deal with reporters calling and sending emails.

Gee, I had my day whole planned out; living the life of a semi-retired engineer.
Write some new code to do a better job of cleaning, prepping and validating the LoanStats files prior to modeling.
Then listen to the LC webcast at 5pm and find out how the past quarter went. I was optimistic.
The plan went horribly awry upon learning of RL departure et. al. in the AM.
Started to code a bit and felt like I was re-arranging the chairs on the Titanic. Bummer.
Then I set back and watched LC's stock and felt like I was on the Titanic.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

jheizer

  • Sr. Member
  • ****
  • Posts: 480
    • View Profile
    • LC Tools
Re: Portfolio (Loan Picking) Strategy
« Reply #13 on: May 09, 2016, 08:33:30 PM »
I'd be happy with just the first sentence being semi-retired.
Replacement to P2P Quant's Percentile Tool http://lc.geekminute.com