This is the second post of a two part series on Lending Club loan descriptions by guest writer Sam Kramer. Sam spent the last 15 years working in the finance industry, where he was exposed to financial analysis and consumer credit. Sam is married, has two children, and includes investing in Lending Club amongst his hobbies. He can be contacted on Twitter @P2P_CT.
This post references charts included in the first part of this series. Please read Part 1 first – it will make much more sense that way.
Very short loan descriptions – 1-10 characters
The very short loan description default rate is notably high. Obviously these loans should be avoided, but is there something other than description length that can be used to identify problem loans? I don’t think I’d be comfortable investing with a strategy which allows 11 character loan descriptions (which show a lower default rate) but not 10 characters.
An examination of the very short loan descriptions showed a few different types:
- Blank descriptions – these descriptions only consist of 1 space, and are primarily responsible for the increased incidence of 1-10 character loan descriptions in Q4 2009 and Q1 2010. These are very rare outside of these two quarters;
- Descriptive – these are single or double word descriptions. “Car loan” and “pool” are quite common within this category;
- Thanks – a number of borrowers simply wrote “thanks” or “thank you”;
- Other – this category includes borrowers who inserted only a dollar amount, a single word (such as “help,” “hello” or “personal”). Gibberish was also included in this category, but is rare;
- None – a number of borrowers went to the trouble of writing “none,” “nothing” or “n/a” in their loan descriptions.
Regardless of my categorization, we can see all types of very short loan descriptions have relatively high default rates:
Based on this analysis, it would appear that loan descriptions which would be more appropriate as a loan title should be avoided. It would also seem logical that a description which only says “debt consolidation” (which I recall seeing often) should also show a high default rate, even though it is 19 characters. An examination of a sample of loans shows that this description also has a high default rate.
Other loan descriptions – 11 and more characters
Revisiting Charts 1 and 3, the short (11-350 character) description loans exhibit a lower default rate, while longer descriptions start showing a higher default rate.
This observation might be justified by assuming long-winded borrowers are really pleading their case, which is a good indication of financial distress, and therefore a poorer credit risk. This sounds logical, and a couple of studies which caught my eye shed further light on this topic:
- Tell Me a Good Story and I May Lend You Money: The Role of Narratives in Peer-to-Peer Lending Decisions, by Michal Herzenstein, Scott Sonenshein, Utpal M. Dholakia.
- Is Silence Golden? – How Non-Verifiable Information Influences Funding Outcomes On Peer-to-Peer Lending Platforms, by Fabio Caldieraro, Marcus Cunha Jr., Jeffrey D. Shulman, Jonathan Zhang.
Tell Me a Good Story focuses on Prosper loans. The study included the categorization of loan descriptions by the researchers, which required that each loan description be read and the contents categorized. While this study was not performed on the LC platform, I believe its findings are still relevant and recommend that anyone who uses the loan descriptions in their credit vetting process read this study.
I noted two particularly interesting findings from this study:
- Borrowers who claimed to be trustworthy or moral are more likely to pay on time, while borrowers claiming hardship tend to have higher late and default rates; other identities (including successful, hardworking and religious) are less useful as indicators of loan performance; and
- Borrowers who made many claims (for example, “I am a good credit risk because I consistently spend more time at the office than my colleagues and have been rewarded for this behavior with large bonuses and frequent salary increases. God bless you all and thank you for your support” – this description displays hardworking, successful and religious traits – 3 in total) tended to be poorer credit risks. The study also found that the default rate increases as the number of claims made increases.
Is Silence Golden was performed on the LC platform and specifically looked at the no description loans. This study concluded that the most credit worthy borrowers (and only the borrower itself knows this piece of information) do not feel the need to provide a loan description, or non-verifiable information, as these borrowers feel confident that the data provided by LC, which is verified, will be sufficient for investors. By not providing any information, these borrowers are “counter-signaling” their credit worthiness.
Wikipedia defines counter-signaling as individuals, “with the highest level of a given property invest less into proving it than individuals with a medium level of the same property.”
Both of these studies might be applicable to the short description loans; that is, the shorter description might be a form of counter-signaling, in that these borrowers spend less time trying to demonstrate their credit-worthiness than the borrowers entering long descriptions. Further, borrowers who make fewer claims would do so in fewer characters than borrowers who make multiple claims, and therefore the findings in Tell Me a Good Story appear to be consistent with the observations in Charts 1 and 3. The converse also appears to be true for both studies’ conclusions.
As the 2010 and later vintages repay their loans, the true default rates of no description loans will start to emerge. This will start happening over 2013 for the 2010 36-month loans. Perhaps I will need to revisit the no description loan analysis in a year.
For the time being, I plan on reading the loan descriptions more closely and will try to count the number of traits exhibited by borrowers, with the preference being a single claim of trustworthiness. I will also try to identify borrowers who are counter-signaling in their loan descriptions, keeping in mind that very short descriptions should be avoided entirely. I have also made a point to go through my portfolio and sell all loans with loan descriptions which would be better suited as titles.
I haven’t decided whether I will avoid the no description loans. It might be difficult ignoring such a large proportion of loans. If LC keeps attracting borrowers at their current pace this shouldn’t be an issue.