Print

[2potatoes]

This morning I took my first look at the loans from Lending Club http://lendingtuber.blogspot.com/2011/08/needs-series-comparing-to-lending-club.html to see if the trend carries over. Looking at the 2007 and 2008 loans, it looks like there was a 7.5% difference in payback between loans with the word "need" and those without.

		Total Loans		Percent Charged Off or Defaulted
All 2007 & 2008 Loans		2996		21.2%
"Need" in title or description		814		26.8%
"Need" not in title nor description		2182		19.2%

The discrepancy for "Payday" was much larger, but on a tiny number of loans.

[2potatoes]

Lending Club D, E, F and G loans from inception to about two months ago.

It looks like 'need', 'payday', 'help', and 'behind' are words that may be indicative of loans that pay less often. 'Chance', contrary to my initial guesses, doesn't seem have the same affect.

Description	Total Loans	Percent Good	Percent Bad	Fully Paid	Current	Charged Off	Default
D,E,F,G All	8840	81.1%	9.4%	15.5%	65.6%	7.5%	.1%
D,E,F,G With 'need'	1455	77.9%	16.2%	21.6%	56.3%	13.8%	.1%
D,E,F,G With 'help'	1422	80.3%	11.3%	18.9%	61.4%	9.3%	.4%
D,E,F,G With 'chance'	84	83.3%	7.1%	22.6%	60.7%	6%	0%
D,E,F,G With 'behind'	65	78.5%	16.9%	18.5%	60%	15.4%	0%
D,E,F,G With 'payday'	12	41.7%	58.3%	25%	16.7%	58.3%	0%

[JGuide]

So what words have a better than average pay history?

[2potatoes]

So what words have a better than average pay history?

I haven't investigated these nearly as much, but a first-look is available at http://lendingtuber.blogspot.com/2011/07/words-of-win.html

[havastat]

So what words have a better than average pay history?

I haven't investigated these nearly as much, but a first-look is available at http://lendingtuber.blogspot.com/2011/07/words-of-win.html

A brief word of caution. The list gives the ten words most likely to appear in loans that paid without looking at the loans that didn't pay. For all we know, these loans might have appeared with the same frequency in loans that didn't pay. Their use in loans that paid may be due to their use in all loans. It might have nothing to do with whether or not the loan pays. We'd have to look at the loans that didn't pay to see.  

[2potatoes]

A brief word of caution. The list gives the ten words most likely to appear in loans that paid with looking at the loans that didn't pay. For all we know, these loans might have appeared with the same frequency in loans that didn't pay. Their use in loans that paid may be due to their use in all loans. It might have nothing to do with whether or not the loan pays. We'd have to look at the loans that didn't pay to see.   

A good word of caution, but I think you have misunderstood (and looking at my blog I see that I didn't explain anywhere) the method I used to gather the words. Instead of looking at the frequency of words in all Paid loans, as I believe you are thinking, for this study I looked at how likely, when a word was used in a loan, that loan was likely to be paid back.

Allow me to elaborate:

For every loan created successfully on Prosper before the end of 2007 I created a list of all of the words used in the Title and Description of the loan. For every instance of a word in a loan that was Paid I added 1 to a running total of PaidInstances. For every instance of a word in a loan that had any other status I added 1 to a running total of UnpaidInstances.

I then calculated the percentage for the word with the formula: PaidInstances / (PaidInstances + UnpaidInstances)
(Which is to say: PaidInstances / TotalWordUsage)

I reduced the list to words which had been used at least 1000 in the loan set and sorted it from words that were most often in Paid loans to words that were least often in Paid loans and compared that list to the overall likelihood of any loan to be paid back.

I found the word 'lender' at the top, with loans containing the word having been Paid 68.96% of the time. I found the word 'payday' at the bottom, with loans containing the word having been Paid only 38.89% of the time. (This compared to an average Paid percentage, across all loans, of about 61% for this time period.)

Now I think that there is an argument to be made that it would have been better to count each word a maximum of once for each listing -- what I did measures use of the word itself, more than it measures the use of the word in the listing ("help, help, help, help!" in one listing counts 4 times, instead of just once), but I think that the best choice really depends on what you're trying to do with the information.

[havastat]

Thanks for the clarification.

The phrase "10 most likely words to have been used in a loan that repaid" may not quite covey what you mean, it suggests that given that the loan was repaid, how likely was it that word was used. You're saying that instead what you did was given the word was used, how likely was it the loan was repaid.

I agree it might have been more helpful to count the word only once per loan, since that would better simulate choosing loans by looking for words. One would have to weight the amount one invested by the number of times the word was used to get performance analogous to what you found, and aren't likely to do that (and there's no reason they should). If one weighted money invested by the loan rather than by the word, the probability would better predict the outcome if it was weighted the same way.

Finally, I'll point out that if words are distributed in the applications randomly (if all the applications were produced by monkeys typing on typewriters or randomly selecting words from a dictionary), than by chance there will be a word associated with the most and a word associated with the least repayment, and the chance difference would look substantial. Just as there'll be a best and a worst five year performance for mutual funds and the difference between the two will look impressively large even if all the performance numbers were the result of chance. One quick way to see if the result is due to chance is to divide the set of loans into two randomly, run the model on them, and see if the results look similar.  If it's all chance, they'll look quite different.

[2potatoes]

Finally, I'll point out that if words are distributed in the applications randomly (if all the applications were produced by monkeys typing on typewriters or randomly selecting words from a dictionary), than by chance there will be a word associated with the most and a word associated with the least repayment, and the chance difference would look substantial. It's much like there'll a best and a worst five year performance for a mutuals fund and the difference between the two will look impressively large even if all the performance numbers were the result of chance. One quick way to see if the result is due to chance is to divide the set of loans into two randomly, run the model on them, and see if the results look similar.  If it's all chance, they'll look quite different.

A good point, and I think I ran into that problem in this post http://lendingtuber.blogspot.com/2011/08/look-at-prospers-2008-loans.html where I found that, when taken as a whole, the loans with words that went bad before 2008 actually performed better than the average when they were used in loans made during 2008. This is why, for my Lending Club analysis, I looked at just a hand-picked subset of the words: http://lendingtuber.blogspot.com/2011/08/initial-investment-in-lending-club.html

I'll try running a randomly split search on the loans, as you suggest, this weekend. Maybe I'll try doing something similar for Lending Club, too.

[cubbiesnextyr]

Nominate for the lobby.

The thread analyzes general data in a unique way.

We wouldn't want the lobby to appear dead...

[ira01]

Nominate for the lobby.

The thread analyzes general data in a unique way.

We wouldn't want the lobby to appear dead...

+1.

[JGuide]

+1

[Risk_Reward]

I don't want lobby.  Why give borrowers these tips.

[God-Father]

Lobby ---- Lobby  -----   Lobby

[cubbiesnextyr]

Lobby ---- Lobby  -----   Lobby

It's already in the lobby...

[God-Father]

Well good job then!