The Prosper Credit Risk Rating System: Does it Improve Market Decision Making Efficacy?

Objectives:

Anonymous online interaction presents new challenge in peer-to-peer consumer credit markets: effectively predicting and screening risk. Prosper has implemented new policies to minimize its disadvantage in information access.  These are improved information transparency and the introduction of the “Prosper Credit Rating System”. Therefore, the objective of this research endeavor is to examine the credit risk rating system newly introduced into Prosper’s peer-to-peer loan marketplace to see if it leads to improved market decision making efficacy. The proprietary credit rating system was specifically designed to evaluate and rank borrower credit risk associated with their loan request listings so as to support lenders in their pricing decision. It is untested, only introduced in July 2009.

Data/Methods:

The Prosper Rating system consists of seven grades: AA, A, B, C, D, E and HR corresponding to lowest to highest credit risk. The focus of this research is to determine whether the Prosper credit rating system does or does not differ with respect to its efficacy on interest rate.  The null hypothesis is that the Prosper credit ratings will have no impact on individual loan interest rate determination.

Ho: τ_{ProsperRatingAA} = τ_{ProsperRatingA} = τ_{ProsperRatingB}  =  τ_{ProsperRatingC} =  τ_{ProsperRatingD} = τ_{ProsperRatingE} =  τ_{ProsperRatingHR} = 0.

Ha: at least one (τ_{ProsperRatingAA}, τ_{ProsperRatingA}, τ_{ProsperRatingB}, τ_{ProsperRatingC,} τ_{ProsperRatingD}, τ_{ProsperRatingE}, τ_{ProsperRatingHR)} ≠ 0. 

The sample covers 2,525 funded loan listings. The data employed came directly from Prosper itself.

 As the proprietary system was introduced in July 2009, the dataset spans July 2009 through January 2010. An analysis of covariance was employed to test whether the newly created Prosper credit risk rating system, τ_{ProsperRating}, has an effect on the outcome of Y_interestrate after removing the variance for which the other four variables denoted as X_loanamount, X_{debt-to-income}, α_homeowner and γ_FICOrange account.

The dependent variable being measured in this framework is individual loan interest rate on loans receiving funding, Y_{interestrate.} Explanatory variables include dollar amount of the loan request (X_loanamount), Fair Issac credit score range (γ_FICOrange), borrower’s home ownership status (α_homeowner), borrower’s debt-to-income ratio (X_{debt-to-income}) and Prosper credit risk rating (τ_{ProsperRating}).

 

Results/Expected Results:

All variables in the model under consideration are statistically significant.

Results show that should the borrower possess the best Prosper credit rating, AA, the impact to interest rate is -23.6%. Given the intercept of 23.5% and upward influences of the numerical variable measuring for loan amount (+6.91% to +10.13%) and an average debt-to-income influence of +0.5%, the context of the -23.6% does indeed make sense. The inverse impact on interest rate lessens in steps, -21.6% for “A”, -17.8% for “B”, -11.6% for “C” and -5.9% for “D”. The interest rate actually ticks upward +0.9% for a low Prosper rating of “E” and no impact is experienced at the “NR” Prosper rating.           

Discussion

The writer must reject the null hypothesis that the Prosper credit ratings will have no impact on individual loan interest rate determination. Research findings clearly indicate that six out of the seven Prosper credit ratings have a statistically significant impact on the dependent variable, individual loan interest rates.