Author Topic: Model Questions  (Read 4374 times)

SBryantMS

  • Jr. Member
  • **
  • Posts: 95
    • View Profile
Model Questions
« on: February 16, 2013, 07:21:32 PM »
I unilaterally will not invest in some loan categories i.e. Business, Medical, Moving, Vacation or Other. 

I have a bias toward Credit Card and Debt Consolidation.

What effect does this external action have on your model, assuming that I regularly pick loans selected by your model from the other categories?

Peter

  • Administrator
  • Hero Member
  • *****
  • Posts: 759
    • View Profile
    • Lend Academy
    • Email
Re: Model Questions
« Reply #1 on: February 18, 2013, 08:16:20 AM »
Thanks for starting some threads here. I know Bryce was off the grid this past weekend but I am sure he will get back with you early this week.
Publisher of the Lend Academy blog

See my returns here: http://www.lendacademy.com/returns

brycemason

  • Hero Member
  • *****
  • Posts: 801
    • View Profile
    • P2P-Picks.com
    • Email
Re: Model Questions
« Reply #2 on: February 18, 2013, 09:38:54 PM »
I suggest that it's not a big deal, because loan purposes that are statistically significantly related to default are absolutely included in the model. Admittedly, certain categories under- and over-perform. However, if many other characteristics about the loan suggests repayment, then sometimes even a small business loan can meet the expected return criteria (it's definitely more rare than just picking randomly off the platform).

This is an example of a fundamental difference between a multi-dimensional, simultaneously estimated model and a filter. In a filter, it's unilateral. The other (my) kind of of model allows us to use shades of gray. For example, if a small business loan was at 1000% interest, it might be worth the risk. There are a lot of moving parts.

Thanks for the question. I hope the answer was helpful.

Keltset

  • Full Member
  • ***
  • Posts: 176
    • View Profile
Re: Model Questions
« Reply #3 on: February 18, 2013, 11:23:34 PM »
I think there is something we are all kind of wanting but just doesn't matter. We (myself included) tend to all like to run our own stats and use our own methodology while that's not exactly what this is. This is already the cream of the crop based on Bryce's model which presumably includes all these factors such as state, purpose, income etc... I suppose it's difficult to not want to over-analyze it and tear it apart to ensure it fits into our own individual models and buying patterns.

brycemason

  • Hero Member
  • *****
  • Posts: 801
    • View Profile
    • P2P-Picks.com
    • Email
Re: Model Questions
« Reply #4 on: February 18, 2013, 11:58:43 PM »
It's natural for the hobbyists here to want to understand and incorporate into their way of working. I'm fine with that. I get it because I'm like you. And I'll further the conversation here as well as I can while maintaining the proprietary nature of the work.

At the end of the day, though, for every one person who is a hobbyist here, there are probably 100 people out there who are interested in LC and want it to be simple. They can think about the things that are specific to them (such as maturity (36/60 eventually), general risk (A-F), and perhaps other things) without worrying about the risk/reward so much.

rawraw

  • Hero Member
  • *****
  • Posts: 2795
    • View Profile
Re: Model Questions
« Reply #5 on: February 19, 2013, 06:28:50 AM »
It's natural for the hobbyists here to want to understand and incorporate into their way of working. I'm fine with that. I get it because I'm like you. And I'll further the conversation here as well as I can while maintaining the proprietary nature of the work.

At the end of the day, though, for every one person who is a hobbyist here, there are probably 100 people out there who are interested in LC and want it to be simple. They can think about the things that are specific to them (such as maturity (36/60 eventually), general risk (A-F), and perhaps other things) without worrying about the risk/reward so much.
Does the model produce an expected return?  I'm not nearly familiar enough with your model to know what it is capable of, but it seems people would want more customization than the current two options.  I'm going to try out your model and make a separate portfolio for it and see what happens.  I need to see if it generates a spread greater than the commission.