Prosper Loan Geographical Analysis
One of the key features of marketplace lending is the availability of borrower data at the loan level that investors can use to develop models and make investment decisions. Along with the many borrower characteristics they provide in the data, most loan originators also provide state or zip-code level information on their borrowers. This loan-level geographic information allows investors to build additional macroeconomic factors from outside the loan tape into their investment criteria and can offer valuable insights in forming an investment strategy.
In this post we will be performing an analysis of Prosper loans by state. We’ll start by looking at the distribution of loan originations across states, and then we’ll build up an analysis of loan performance by state and investigate whether there are any noticeable trends between performance and state-level unemployment data.
Loan Origination Volumes
Let’s start with a simple map of Prosper loan originations by state from January 2013 through present:
As we can see from the map above, New York, California, Florida, and Texas all show large volumes of originations for Prosper loans. This is an interesting first level of analysis, but the problem with many maps like this one is that instead of offering real insight into trends across states, they are essentially population maps. After all, the four most populous states in the United States are currently California, Texas, Florida, and New York, so we’d expect these states to have the highest loan origination volumes. Let’s try to remove the effects of state population by instead reporting loan origination volumes on a per capita basis:
This chart offers much more interesting insights into state-level borrower behavior. The four largest volume states from before are now closer to middle of the pack, while states like Kansas, Nevada, Connecticut, and Rhode Island–all low origination states in the previous map–show the largest originations per capita. This may indicate increased awareness of Prosper in those states, or it may indicate lower levels of traditional bank financing for personal loans. In any case, it is interesting to note that the largest state for Prosper originations on a per capita basis (Kansas, with $5.58 per person), has around 3.5 times the origination volume of the lowest state (Utah, with $1.61 per person).
Now that we have an idea of the distribution of loan originations by state, let’s see what we can determine about state-level loan performance. For this analysis we’ll use the 2013 vintage loan population to allow for sufficient seasoning, and we will define delinquent loans as those that are either currently 30+ days past due or have already charged off. Rather than showing the raw performance numbers by state, here we’ll normalize the numbers to the national average, so that we can clearly see which states are performing better or worse than average. In the map below, the darker purple colors indicate better performance (lower delinquency rates than the national average), while the darker green colors indicate worse performance (higher delinquency rates than the national average).
We can see from this map that those states with the lowest levels of delinquency are Wyoming, Colorado, West Virginia, and Vermont. The states with the highest levels of delinquency include New Mexico, Missouri, and Mississippi.
Let’s compare the state-level loan performance map to a map that shows unemployment rates by state as of 2013, using data sourced from the Bureau of Labor Statistics. As before, we’ll normalize the numbers to the national average to quickly identify the relative unemployment rates of individual states. Similar to the map above, the darker purple colors indicate lower unemployment rates than the national average, while the darker green colors higher unemployment rates than the national average.
While this does not exactly match the map showing loan performance above, we do see some general trends here. The midwest and the south generally show higher levels of both delinquency and unemployment rates, while the great plains generally show lower levels of both delinquency and unemployment. To quantify the trend shown in these maps, I calculated the correlation between the state-level unemployment and performance measures to be .26. Although this value is not particularly high, it does indicate a weak positive relationship between the two variables, and suggests further analysis would be warranted.
As mentioned before, one of the key features of marketplace lending is the large volume of data that the loan originators make available to investors. The ability to combine this extensive data with broader market indicators should allow investors to generate additional insights when creating a marketplace lending investment strategy. As the industry grows and matures, we should expect to see increased use of macroeconomic variables such as unemployment, GDP, and interest rates playing a role in investor decision-making.