Seasonality in Loan Performance
Seasonality in loan performance is a well-established characteristic of mortgages and credit cards. Seasonality refers to any type of trend that is periodic in nature. For example, borrowers may be more likely to prepay or miss payments in certain months than in others. With several years of data now available for the marketplace lending industry, we thought this would be a good opportunity to investigate seasonal trends in loan performance. In this post, we will analyze seasonality in delinquency rates for consumer marketplace loans.
Seasonality can take on a variety of forms, but in consumer lending, there are two well-known trends. First, delinquency rates tend to rise in the 4th quarter, as consumers increase spending and credit card purchases during the holiday season. Second, delinquency rates tend to fall in the spring, as consumers receive tax returns that can be used to pay off loan obligations. As mentioned before, these trends are well established in the mortgage and credit card markets; now let’s investigate to see whether these same characteristics hold for the marketplace lending environment.
Methodology & Initial Analysis
To analyze the seasonal performance trends of a marketplace lending portfolio, we will focus on the measure of new entrants to delinquency. Specifically, we will track the proportion of beginning-of-month principal balance that enters the 30+ days past due delinquency bucket on a monthly basis, beginning in January 2013 and ending in December 2015.
Looking at the graph above, there are a few immediate observations to make. The first is that in 2013 the delinquency transition rates were both higher and more volatile than in more recent years. Looking a bit deeper, we can also see hints of a seasonal trend in the rates of new delinquencies. In general, we see lower transition rates in the spring months and higher rates later in the year, with peaks in October and November. To investigate seasonal patterns in more detail, we present this data stratified by year in the following graph.
Displaying annual data in this manner allows us to more easily identify any consistent intra-year trends over time. Here we do see a similar pattern presenting itself in each year: rates fall for the first few months of the year, reaching lows some time between March and April, then rise later in the year, peaking in October or November. In each of the years, the ratio of maximum to minimum transition rates is relatively consistent, with the yearly maximum rate approximately 1.3x the minimum.
Seasonal and Trend Decomposition Analysis
While the above discussion focused on the identification and investigation of seasonality in the data, we’re now going to explore a method one might use to quantify the seasonality and underlying trends in the data. For this, we will be employing Seasonal and Trend Decomposition using Loess (STL), a method developed by Cleveland et al. (1990). STL is a time-series method that allows us to decompose a series into trend, seasonal, and remainder components. Such an analysis could help us to answer questions about broader trends when seasonal components are present in the data. For example, if we see rising delinquency rates in Q4, STL could help us identify how much of those trends are due to seasonal factors and how much may be due to other structural changes in the underlying performance.
The above graph shows a visual representation of the STL analysis on the delinquency data. The 4 panels presented show:
- The raw data series
- The seasonal component of the data series
- A smoothed trend series, after removing the effects of the seasonal factor
- A remainder series (essentially normally distributed errors after removing the effects of the seasonal and trend factors)
There are a few interesting conclusions that we can draw from this analysis. First, we have identified a notable seasonal component in the data that corresponds to what we discussed previously: low values in the spring and higher values later in the year. From the lows of March and April to the peak in October, we would expect to see the transition rates in a given year increase by 12bps due to seasonal factors. Second, this allows us to identify broader trends over time. For example, we see that from 2013 through late 2014, the overall trend was toward lower transition rates, while through 2015 we saw a trend of rising rates, flattening slightly toward the end of the year.
In a future post we may perform a similar analysis to investigate seasonality in borrower prepayment behavior. As this industry continues to grow and mature, we expect many of the modeling techniques and analyses from the mortgage and consumer credit industries to become increasingly relevant and useful here. As the historical performance data for marketplace lending assets continues to grow, we look forward to building increasingly sophisticated analyses like the one presented here.