This project aims to analyze loan data from Prosper Marketplace, a company based in San Francisco, California, that specializes in providing peer-to-peer loans at low interest rates to borrowers. The goal is to identify the different borrower motivations when applying for loans, and several factors that may influence loan favorability (measured in terms of annual percentage rates).
The prosper loan dataset contains information on 113,937 loans described with 81 variables. The variables included loan amount, borrower rate (or interest rate), borrower employment status, borrower income, and many others. The dataset can be found on Udacity servers, and a detailed explanation of all its features can be obtained from this spreadsheet.
Of the 81 features in the dataset, 14 were selected for exploration. Prosper rating and Borrower APR were identified as the key variables to measure loan favorability. During cleaning, duplicate records were removed, data types were properly converted, and null records for the key variables were dropped. State longitude and latitude information was gathered from data uploaded by GitHub user, Rashida048, to aid the visualization of loan patterns across states. At the end of the wrangling process, 83,982 loan records and 17 variables remained for analysis.
The exploration process yielded a mix of both intuitive and rather unexpected results.
The largest population of people took loans to finance previously accrued debt, rather than for business or asset purchase. Debt consolidation also accounted for higher loan amounts than other reasons. Asides business purposes, borrowers depended on huge loans to finance weddings, child adoptions, boat acquisitions, and the purchase of engagement rings.
The annual percentage rate on a loan (Borrower APR) was negatively correlated with the loan size, loan term, and Prosper rating. Prosper rating appeared to be a key factor on its own, influenced positively by high and verifiable incomes, stable means of employment, homeownership, and low debt-to-income ratio.
Further exploration reveals a dichotomy between borrower APR and Prosper rating. Both variables correlated negatively, up to the B rating. However, the pattern reverses after that. The possible involvement of lurking variables was discovered. High-rated borrowers may have a propensity to borrow huge sums when long-term loans are involved. Hence, increasing APR could disincentivize 'overborrowing'. Alternatively, lowering APR could build capacity for low-rated borrowers.
Explanations start with the most common reasons why people borrow through prosper, and the average amounts they borrow for those reasons. The information was conveyed with two barplots.
Furthermore, the influence of Prosper rating on loan interest rate was explained. Prosper rating was chosen as the key variable to compare with Borrower APR, since the company already factors inputs from other variables in its estimation. Box plots, scatterplots, and point plots were used to show the disparities in borrowing power across Prosper ratings. The interesting influence of lurking variables, like loan term, was also explained.