In addition to posting the duds of the day, I would love it if you also post why you think each loan is the biggest stinker. This would be good for me to understand how other people evaluate bad/risky loans.

The short answer is that I don't know. A logistic regression analysis using independent variables such as FICO, inquires last 6 months, employment status, DTI and many others from historic LoanStats data was performed. No results of LC's model such as Grade, Interest Rate, EXPECTED_DEFAULT_RATE, or INSTALLMENT were included as independent variables in the logit analysis. Only borrower data was and should be included. The dependent variable of the analysis was the variable charged off. The analysis result was a set of weighting coefficients that when applied to the same variables in a new loan to be evaluated yields the predicted probability that loan will be charged off or not. Given this probability of being charged off (probability of default or "risk") and the interest rate ("reward") an overall score for the loan is computed. So, there are many many factors that contribute to a loan's score and I have no way to unwind that score into its individual components.

Logistic regression is the classical way consumer loans have been scored for decades. If there is a magic bullet dominant independent variable I am unaware of it. Of course some are more important (more heavily weighted) than others. And some independent variables are of little use as they closely follow others and add little or no value (covariate). For a good example see the following. It was an interesting and revealing post:

https://forum.lendacademy.com/index.php/topic,3570.msg31593.html#msg31593For each loan scored I am able to see the probability of default computed by my program and compare it with the EXPECTED_DEFAULT_RATE provided by LC. Since LC began providing a single EXPECTED_DEFAULT_RATE per grade rather than per sub-grade the comparison is less precise. However I am amazed by the magnitude of the differences and there are very few loans per drop where LC's rate is higher than that computed by my program. When LC's rate is higher it isn't ever by much. These differences have grown over time and it even seems as if LC's newer model expects consumer behavior to be improving. On the other hand I spend very little time with LC anymore and the model I'm using is pretty old. Since I'm no longer investing in new notes I'm more of a casual observer so take this with a grain of salt.