Author Topic: Have I found the Holy Grail of filters? (Nickelsteamroller)  (Read 12257 times)

trevor

  • Newbie
  • *
  • Posts: 39
    • View Profile
Just curious if anyone is able to beat this while maintaining a good amount of total loans (This filter has almost 3000)

In addition to the ones in this screenshot, the only other 2 filters I used were "exclude loans with public records" and "0 inquiries in past 6 months"


basically, I'm just investing into people with high income who have a spending problem


hoggy1

  • Sr. Member
  • ****
  • Posts: 401
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #1 on: July 21, 2014, 10:05:36 PM »
Yes, 3000 is not that many. I have seen 17+% in NS with that many loans passing.
Steve

trevor

  • Newbie
  • *
  • Posts: 39
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #2 on: July 21, 2014, 10:36:34 PM »
Yes, 3000 is not that many. I have seen 17+% in NS with that many loans passing.

So, does anyone have a link to it? What was the average loan age?

rawraw

  • Hero Member
  • *****
  • Posts: 2795
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #3 on: July 22, 2014, 07:10:23 AM »
Data mining can be a dangerous practice

hoggy1

  • Sr. Member
  • ****
  • Posts: 401
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #4 on: July 22, 2014, 07:27:17 AM »

I didn't save it, but just recreated the one below that is close and has over 5286 notes passing and just 0.11% lower OVERALL ROI than yours.


int_rate    >= 17
purpose    credit_card, debt_consolidation
pub_rec    0
inq_last_6mths    <= 1
num_actv_bc_tl    >= 3 and <= 8
num_actv_rev_tl    >= 7 and <= 12
num_accts_ever_120_pd    <= 0
annual_inc    >= 50000
Steve

hoggy1

  • Sr. Member
  • ****
  • Posts: 401
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #5 on: July 22, 2014, 07:29:28 AM »
Now that I look at yours again it is a similar approach.
Steve

trevor

  • Newbie
  • *
  • Posts: 39
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #6 on: July 22, 2014, 11:38:03 AM »
Data mining can be a dangerous practice

How so? Are you saying this isn't accurate? 3000 loans sounds like plenty to me

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #7 on: July 22, 2014, 12:48:26 PM »
I agree with rawraw. Blindingly using data mining is dangerous. You need to understand underlying data. Also, You need to be able to offer a reasonable theory on reasons for such an unusual return.

Have you looked at returns and loss rate by vintage (year of issue)? 60 month loans were only started to be issued in 2010 so no vintage has gone through a full maturity cycle yet.

How many of 3,000 loans identified in your backtesting are recent (2013-2014) and don't have enough history yet? I will bet more than half of your backtested loans were recent loans and average age less than 12 months.

Have you looked at loans with status Fully Paid, Default or Charged Off only? How has been the returns and loss rate when you exclude current and late loans?

Data mining can be a dangerous practice

How so? Are you saying this isn't accurate? 3000 loans sounds like plenty to me
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

trevor

  • Newbie
  • *
  • Posts: 39
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #8 on: July 22, 2014, 01:13:23 PM »
"Blindly" is a strong word to throw around. It's clear what kind of people I'm investing in: high income (100k+) earners who are seeking a debt refinance.

Average age is nearly 14 months. I agree, that's too recent, so I used the same filters but limited the average loan age to around 2 years. Loss rate went up by 1.4%



« Last Edit: July 22, 2014, 01:37:37 PM by trevor »

rawraw

  • Hero Member
  • *****
  • Posts: 2795
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #9 on: July 22, 2014, 03:45:25 PM »
Like Anil said, historical analysis of data is really just the first step.  Don't even think about the quantity of loans, think about the time period -- it doesn't even contain a full business cycle.  A great example of this data mining problem is people avoid certain states because "they default more."  Well. . . do they think the same geographies of the country are hit each recession?  No, it's just they didn't think through the data and the time period over which the data exists. 

hoggy1

  • Sr. Member
  • ****
  • Posts: 401
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #10 on: July 22, 2014, 04:15:32 PM »
Raw on rawraw,

I love CA loans. They are the only loans that score very highly (by my grading system) that I don't have to rush to get. They always take 2 or 3 days to fund no matter how good the borrower looks.
Steve

rawraw

  • Hero Member
  • *****
  • Posts: 2795
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #11 on: July 22, 2014, 05:51:07 PM »
"Blindly" is a strong word to throw around. It's clear what kind of people I'm investing in: high income (100k+) earners who are seeking a debt refinance.

Average age is nearly 14 months. I agree, that's too recent, so I used the same filters but limited the average loan age to around 2 years. Loss rate went up by 1.4%




BTW I'd be more concerned that the LC grading system under your filter doesn't appropriately rank order risk.

Booleans

  • Jr. Member
  • **
  • Posts: 76
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #12 on: July 26, 2014, 11:46:45 AM »
Like Anil said, historical analysis of data is really just the first step.  Don't even think about the quantity of loans, think about the time period -- it doesn't even contain a full business cycle.  A great example of this data mining problem is people avoid certain states because "they default more."  Well. . . do they think the same geographies of the country are hit each recession?  No, it's just they didn't think through the data and the time period over which the data exists.

But at the same time people from certain states could be defaulting more often for reasons not related to recession (like California's high cost of living, don't know if that's actually a factor but people mention it often).

I'm working towards getting my portfolio to be equally distributed between states so I don't have to worry as much about local recessions.

rawraw

  • Hero Member
  • *****
  • Posts: 2795
    • View Profile
Re: Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #13 on: July 26, 2014, 11:51:22 AM »
Like Anil said, historical analysis of data is really just the first step.  Don't even think about the quantity of loans, think about the time period -- it doesn't even contain a full business cycle.  A great example of this data mining problem is people avoid certain states because "they default more."  Well. . . do they think the same geographies of the country are hit each recession?  No, it's just they didn't think through the data and the time period over which the data exists.

But at the same time people from certain states could be defaulting more often for reasons not related to recession (like California's high cost of living, don't know if that's actually a factor but people mention it often).

I'm working towards getting my portfolio to be equally distributed between states so I don't have to worry as much about local recessions.
Of course. This is why it's a first step. And geographic diversity is very important imo. Texas looks great in filters, but when the next low in oil comes it may not

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #14 on: July 26, 2014, 08:39:20 PM »

But at the same time people from certain states could be defaulting more often for reasons not related to recession (like California's high cost of living, don't know if that's actually a factor but people mention it often).

I'm working towards getting my portfolio to be equally distributed between states so I don't have to worry as much about local recessions.

It will be very difficult to equally distribute your portfolio across states as except for CA, NY, TX, FL, IL, NJ, very few loans are issued to borrowers of other states. Best you can do is to keep state profile of your portfolio similar to all loans issued on Lending Club or use some other criteria such as population, income, economic factors to distribute your loans across states.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com