Credit inquiries

[Smkj79]

While trying to educate myself on P2P leading, there is one bit of advice or information I don't quite understand. A few websites suggest avoiding loan applicants that have credit inquiries. Even if it's just 2. Why is this such a bad thing? Or is it just a red flag but not necessarily a bad thing? Thank.

[rawraw]

Credit inquiries imply that their current access to liabilities is changing.  So if they have 3 inquiries for loans, their financial condition may drastically change once those liabilities become obligations of the borrower. 

[LoanShark]

Data is more important than intuition.  There doesn't seem to be any reason why a credit inquiry or two should make a borrower more risky, but we have seen that borrowers with credit inquiries on file default at a significantly higher rate than borrowers without them.

For example, borrowers who approached LC for a 36 month loan between 01/09 and 06/11 with a B3 LC grade caused losses of 4.4%, giving a rate of return of 5.7%.  The loss rate for the same subset of borrowers without credit injuries on file during that period was only 2.6%, giving an roi of 7.4%.

When the data tells me something like that, I don't really care what the intuitive reason or cause is.

With that being said, you could think of previous inquiry borrowers approaching LC as a "lender of last resort."  They might be desperate because they were turned down by other lenders they would have preferred.  Maybe one of those other lenders is the local loan officer, who knows that Borrower is a bad dad who had drug problems 8 years ago.  Who knows?  Who cares?  Follow the data!

[TravelingPennies]

Data without intuition is equally as dangerous.

[standby]

It seems my biggest mistakes early on were taking people with inquiries.  Go for zero because most likely, they're looking for some sucker to clean out before they disappear or file for bankruptcy.

[Thatguybil]

https://forum.lendacademy.com/index.php?topic=2926.msg26663#msg88888888Quote"> from: rawraw on January 03, 2015, 05:18:18 PM

[TravelingPennies]

Im sorry if I in any way implied that one should never take his intuition into account at all whatsoever.  Of course, at least a little intuition is necessary. 

The broader point here is that there are serious holes in LC's underwriting system and that you can exploit them.  Within each sub-grade, there are very different loans.

[Fred93]

The original poster used the word "why".  This can mean different things to different people.  As in "Why is there air?"  Nobody knows the fundamental reason, and the discussion becomes theological.  I'll try to answer the original question without falling into that rathole.

So lets break it down by level.

Question 1: "Why do people filter on inquiries?"

Answer 1: Because the statistics show that loans with low inquiries have in the past performed better than loans with high inquiries.  This result is robust.  You can see it at all loan grades.  You can see it at all income levels.  You can see it in all years.  You can see it at both Prosper and Lending Club.  By the way, this is apparently a well known thing in more traditional consumer loan underwriting too.

Lending Club used to publish their credit grade formula.  You can still find this info in some of their old prospectuses from 2012 or so.  In those writeups, you see that there is a term in the formula to reduce the grade (thus increase the interest rate) for folks with more inquiries.   We don't know the details of LC's formula any more, but it is likely that they still use inquiries.

Question 2: Why do loans with more inquiries default more often?

Answer 2:  I don't know.  Lots of people speculate.  Here's a speculation that makes sense to me.  Your mileage may vary. 

Lets consider a guy who places a bunch of loan applications all at once with different lenders.   Some of the people who do this will be desperate folks.  We probably want to filter out the desperate folks, because something may be going wrong in their life, for which a loan may not be the answer.  How would we find these people?  DTI won't help you, because those other applications haven't turned into loans yet.  However, those other lenders may have made pulled credit reports, and these events increment the inquiries number.  Thus inquires is an indicator which detects desperation.  It also ensnares non-desperate folks who are casually looking around before deciding where to borrow and whatnot.   It is not perfect.  However, we benefit by filtering away the high inquiry guys, probably because we eliminate some desperate folks who are desperate for reasons which have not yet otherwise shown up on their credit reports.

I want to be clear here.  I don't use inquiries because of this speculation about borrower intent.  Its just my story for what may be going on.  I use it because the data shows me clearly that it works.  When I was starting in P2P lending, I found this very non-intuitive.  I had no background in consumer lending, so my intuition was at the very least not well tuned to consumer lending realities.

Question 3: If LC already takes inquiries into account, why is there information left in that number for us to use?

Answer 3:  Hard question.  The smart people at LC have already used inquiries in their algoirthm for picking an interest rate for a loan application.  If this perfectly compensated for the extra risk (on the average, ie statstically speaking) for a person with more inquiries, then the information contained in the "inquiries" number would be all used up, and there would be no value in us trying to use it further.  And yet, the statistics show that, at least in the past, there was additional information, and loans with more inquiries provided lower returns.   My conclusion is that while LC uses inquiries, all the information value contained therein is not perfectly represented in the interest rate LC chooses.

Question 4: Why is there information left over after LC uses not only inquiries but all the numbers on the credit report?

Answer 4: Dunno.

This is a deep subject.  if LC used all the information on the credit report perfectly (from your point of view, that is optimizing what you want to optimize) then there would be no value in any kind of filtering at all. 

[AnilG]

Attached image may provide supporting data to what Fred93 and rawraw mentioned. Credit inquiries indicate potential change in financial situation of a borrower In such cases, it is better for such transitions to occur/stabilize before making any decision. Personally, 0 or 1 credit inquiries in last six months are fine but more than that needs more exploration before deciding to lend.

[TravelingPennies]

"Question 4: Why is there information left over after LC uses not only inquiries but all the numbers on the credit report?

Answer 4: Dunno.

This is a deep subject.  if LC used all the information on the credit report perfectly (from your point of view, that is optimizing what you want to optimize) then there would be no value in any kind of filtering at all.  "

Good insight Fred!

I would argue that is what makes a market. Asymmetrical information between parties and different values / weighted placed on the information that is symmetrically between the two.

[sociallender]

Here is a graph showing the percent of charged off notes (complete) by inquiries.  I fitted a linear regression line to show the increase in defaults as inquiries increase.

https://dl.dropboxusercontent.com/u/415842/PLS/inq.png" alt="" class="bbc_img" />

However, you should consider other criteria as well.  Having a rigid rule of selecting only notes with 0 inquiries may preclude  potentially "good" notes. 

[edward]

Do you view the better than expected results (i.e., below your regression line) for 4 & 5 inquiries as anything special, or just random variation/noise? I'm not advocating that as a filter, simply noticed it.

[TravelingPennies]

I don't think that there is a perfect linear relationship between inquiries and defaults.   The upward trend just represents a strong relationship that can be used to formulate a dynamic filter.  For example, I would definitely consider adding this covariate into a learning model.  Back to your question, it may be noise but it also could be explained by other variables as well (such is the case with univariate analysis).