- 1 Abstract
- 2 Motivation
- 3 Introduction
- 4 Data and Sources
- 5 Sallie Mae Website Data
- 6 Bloomberg Data
- 7 Models and Analysis
- 8 Visualizations
- 9 Further Research
In June 2014, The Economist, reported that the total student loan debt in the market is over $1.2 trillion. These loans are either categorized as federal or private loans. Federal loans are given out by the United States government and in 2014 totaled $98.1 billion of all student loans. Private loans on the other hand are given out by financial institutions, and tend to charge a higher rate of interest, thereby increasing the rate of default. Defaulting on student loans can have a lasting impact on the economy. Our paper will center around the macroeconomic impact on the economy and markets of student loan default and aim to give a clearer picture of the influences of the potential systemic risks around it. Using any available Federal data on student loans as well as comparable research, our paper will additionally try to classify any factors attributable to student loan default. The motivation behind our research thesis is to introduce a more quantitative aspect to looking at Student Loans and their macroeconomic effects, when compared to more qualitative papers already published.
The 2008 mortgage crisis was a wakeup call and effectively pushed regulators and policy makers to pay closer attention to the kind of loan practices conducted on Wall Street. The relevance of Student Loans lies in the unanimous belief that the next default crisis will originate from what was initially an idea to help students achieve their life goals. With national Student Loan debt levels, both private and public, surpassing the trillion mark, it is expected that the current Presidential administration would invest millions of dollars into understanding the problem better. Thus the key problem lies in understanding the correlation and causation of potential defaults on the various parts of the economy. Relevant indices include Case-Shiller for the Housing Sector, Prime Rate/Libor Rate, S&P Index as well as Consumer Price Index etc. The motivation behind our report lies in visualizing and understanding the causation of ABS prices on marquee economic indicators. Different models and correlation tests will allow us to statistically define areas of the economy with ranging levels of impact. Another important factor behind our research is a call for policy makers to review loan practices under a magnifying glass. Reports over the years exposed the world to the term Robo-Signing and delved into the criminal practices financial institutions were involved in. Robo-Signing is refereed to the loose practices conducted by financial institutions in awarding loans to almost anyone who applied. The Justice Department was able to track back and pin the root cause on temporary workers as well as middle management who never paid close attention to the paperwork. Such practices led to families and individuals with shaky credit to take on massive amounts of debt and eventually default. Reports over the past year point to Robo-Signing activities being linked to Student Loans which could in turn leave the economy exposed, similarly to the 2008 crisis. Thus, there is a need to better understand the relation between the loans and the economy. Currently, large quantities of loans awarded across the country have been packaged into Asset Backed Securities that date back to 1998. Every year, Sallie Mae, the government backed Student Loan Provider, issues multiple ABSs, with underlying loans originating na-tionwide. Thus, increasing default risk with the underlying loans would consequently impact the Asset Backed Securities which, in turn, would affect the different areas of the economy highlighted within the report, in later sections. The following sections will delve into the depths of the report and will highlight the various types of data that were used, the models and tests as well as the methodology of our research.
Data and Sources
Our entire research project is driven around the Student Loan and ABS data available both through Sallie Mae as well as Bloomberg. The data can be further divided into:
- ABS Loan Breakdown Data
- External Market Factors
- CUSIPS and Fields through Bloomberg
Initially, all CSV files of the CUSIPS as well as ABS Loan Breakdown was retrieved through the Navient website. It provided us with any and all data on Student Loan Asset Backed Securities for years ranging from 1998 to 2014. Each year had multiple different ABS issuances with respective student loans nationwide as the underlying.
Sallie Mae Website Data
ABS Loan Breakdown Data
Our team was able to convert a pdf file comprising a large subset of Sallie Mae ABS issuances from 2001 to 2012 into a worksheet. The file was further broken down into respective 12 other CSV files. The following is a description of the different important Loan Breakdown Data retrieved per issuance over the spanning years:
Interest Rate Breakdown
The following data breaks down each ABS issuance by range of interest rates the loans are originated by. Ranges span from 3-3.50% to greater than 8.50%. This allows us to look at the concentration of interest rate loans per ABS issuance. It is evident from further research that issuances with more than 60% of loans with rates 6% and higher have a higher default risk, especially for lower tranches within the ABS. The data is broken up by loans and outstanding principal balance. Thus for some issuances, there might be higher number of loans in the lower interest rate ranges but larger outstanding principal balance in higher interest rate ranges. The following pie charts give a visual perspective on the data.
College Type Breakdown
The following data breaks down each issuance by the proportion of loans from 4 year College, 2 Year College, Proprietary/Vocational and Unidentified. Similar to the above data, it is broken down further by loans and by outstanding principal balance. While the data doesn’t delve into more specifics about the type of colleges, it allows us to gain a high level perspective on the proportion per issuance and look into issuances with larger than average weighting in Proprietary/Vocational schools that might have a higher default risk.
Original Pool Breakdown
The following data comprises different fields and educates us into the outstanding balance, proportion of Treasury Bill to Commercial Paper, Number of Loans, Weighted Averaged Coupon etc. The following attributes listed above are the most important when analyzing the different ABS. It was expected that tranches, highlighted later, that constituted more risk stemmed from ABS issuances with higher Weighted Average Coupon. The following bar plot below represents the distribution of WAC over the issuances.
Static Loan Pool Breakdown
The following data represents one of the more important during our analysis. It breaks down each ABS Issuance Underlying by In-School, Grace, Deferment, Repayment, Forbearance and Claims in Process Loans. It helped us to gauge the change in proportion of type of loans over the range of Collection Periods listed. Similar to previous datasets, the following data is also broken down by loans and outstanding principal balance. It is observed for a majority of issuances of a larger proportion of Repayment Loans. The red flag within the dataset is an increasing proportion of forbearance loans within the ABS issuance. The following represents the change in the proportion of loans and outstanding principal balance for a select ABS issuance.
Loan Type Characteristics
The following data set gives a brief breakdown by ABS issuance of the proportion of Unsub Stafford, Sub Stafford, PLUS and SLS Loans. It further breaks it down by loans and out- standing principal balance. Not as important as other datasets, it was still able to give more information on key ABS issuances highlighted as comprising more risk.
Prepayment Rate Breakdown
The following dataset represents one of the more important pieces of information we used to analyze the ABS issuance. The data, broken down by ABS issuance, lists the Since Issuance Conditional Prepayment Rate (CPR), Quarterly CPR-1, Quarterly CPR-2 over a range of Collection Periods. It allows us to isolate an ABS issuance and track the change in Conditional Prepayment Rates over the given time. It is expected that the CPR rate will increase over time as Students, with steady jobs after a couple years, would want to decrease the principal balance on which they are charged their effective interest rate. The following plot represents the change in the three rates over the given Collection Period range.
Delinquency Loan Breakdown
The following data set represents one of the more important pieces of information during our analysis. The data breaks down each ABS Issuance by the range of days in delinquency per loan. Delinquency is a state of a loan where the borrower is late on his/her repayments. The delinquency ranges from 0-30 days all the way to greater than 360 days where a delinquency greater than 270 days is considered to be in default. This helps us identify ABS issuances with higher than average proportion of delinquencies and defaults ( ¿270 days). The data is also further broken down by per loan as well as outstanding principal balance. The following plot represents the mix of delinquency ranges above 240 days or potential defaults for a select ABS Issuance over the Collection Period. File:SallieMaeFigure6.png
The data set is also broken down into geological distribution, by state. For each state, there is information on the number of loans outstanding, the total principal balance outstanding for that state, and the percentage of the total balance held. These values change over time, but are generally concentrated in certain areas of the United States where more colleges and universities exist.
Using the Navient database online, we retrieved all 182 CUSIPS related to every ABS Is- suance that allowed us to pull any available attribute through Bloomberg to get a clearer picture of the different tranches of the Asset Backed Securities. Some key attributes include:
- Last Prices
- Moody Credit Rating
- S&P Credit Rating
- Fitch Credit Rating
- Tranche Number
- Yield Curve Bid/Ask/Mid etc.
The following attributes along with a select others helped narrow down the kind of tranches and Issuances we should be paying more attention to. The core idea would be to highlight tranches with higher default risk using the Underlying Loan Breakdown data along with any respective fields from Bloomberg and find any causation/correlation to our other dataset, External Market Factors. It was observed using a collection of fields that more than half of our CUSIPs or tranches had been Paid-in-Full. This considerably helped narrow down our dataset and helped us focus on the remaining to highlight any key risky tranches.
External Market Factors
In order to understand the relationship between student loans and the larger economy, we explored a list of external market factors which serve as benchmarks for the overall health of the US economy. The external market factors are listed below with an explanation:
- The Case Shiller Index provides an index for housing prices. Many papers on the topic of student loans mentioned that the debt burden of students prevents many to invest into housing.
US CPI College Tuition & Fees
- The Consumer Price Index for College Tuition & Tees provides us with an aggregate college tuition index. The rising college tuition rates have been a factor which causes many students to borrow more money.
ICE LIBOR USD 3 Month
- The ICE 3 Month LIBOR rate provides us with the floating rate for interbank lending. The ABS loans are all quoted in LIBOR + X%, therefore, ABS prices and yields are dependent.
- The Prime Rate is the rate at which a bank lends to its best customers with the best credit. The prime rate is used for mortgage lending, which may have an impact on individuals with student loans.
US Unemployment Rate (%)
- The US Unemployment Rate is the measurement of the percentage of workers receiving unemployment. A high unemployment rate is a strong indicator of a systemic issue.
10 Year Yield
- The 10 Year Yield for US Government Bonds. The inclusion of this index provides a longer term structure which may be significant.
US Real GDP (Annual YoY %)
- Real GDP growth percent shows whether the economy is expanding or shrinking. A growing economy typically has more room for jobs, and a shrinking economy has less room - with possibility for layoffs.
Federal Reserve Money Supply
- Federal Reserve Money Supply (M2) is a measurement of available money. It includes cash and checking deposits, savings deposits, money market mutual funds, and other highly liquid accounts.
Consumer Price Index
- The Consumer Price Index (CPI) is a weighted basket of consumer goods and services such as healthcare, transportation, and food. A rising CPI is indicative of an expanding economy.
S&P 500 Index
- The Standard and Poors 500 Index is the benchmark index for large cap US equities. A rising S&P 500 Index is a very strong indicator of a good market and overall health in US companies.
Producer Price Index
- The Producer Price Index measures the average price change in domestic goods and services sold, from the seller’s perspective. Our goal is to understand how these external market factors are correlated, and how they relate to the time series of ABS yields.
Models and Analysis
In order to model the systemic importance of student loans, we used at different models to measure correlation and causality. After doing some research, we discovered granger causality. The granger model assumes that one time series has a unique relation to another, and impacts in the first time series are reflected in the second. The null hypothesis in this statement is that there is no granger causality; and as such, a low p-value indicates that the first time series granger causes the second. The model also takes into account a significant lag, which may be indicated by Akaike’s Information Criterion or Schwarz information criterion.
This model will have one of three outputs: the one variable granger causes the other, neither variables granger cause each other, or both variables granger cause each other. The model formula is listed below: yit = α0 + α1 ∗ yt−1 + α2 ∗ yt−2 + ... + αm ∗ yt−m + βp ∗ xt−p + ... + βq ∗ xt−q + residualt
Where: X and Y are stationary time series Null hypothesis: X does not granger cause Y
p and q are lags, short and long respectively
Correlation and Causality
To understand the correlation between the different external market factors, we used R’s cluster package to build dendrograms of the euclidean distance using the correlation matrix. The code also corrected for missing data (NAs) with the na.spline function,which creates a smooth transition between available data points. The following image illustrates the distance between Risky Sallie Mae Loans using a dendrogram.
This Dendrogram shows that the first tranche of the 2008 issuance has a farther distance than the other tranches issuing during that year. The subset was shown because there are a very large number of loans to be visualized. Below, we have presented a distance dendrogram of the Navient Student Loans, which are yet another subset of the total Sallie Mae student loan portfolio.
Methodology and Findings
The methodology we adopted was to locate tranches within our database and measure the correlation and causation towards the different External market Factors such as Unemploy- ment, Money Supply, Consumer Price Index etc. We used the long term rating assigned by Moody’s and Fitch to isolate several tranches that comprised the most credit risk as judged by the respective credit agencies. To further narrow down the list, we used Delinquency Loan ratios as well as the distribution of loans in default. The respective tranches are displayed below:
- SLMA 2008-1 B
- SLMA 2008-2 B
- SLMA 2008-4 B
- SLMA 2008-6 B
One observation to take note of is the concentration of credit risky tranches in the year of 2008. This can be explained due to the higher concentration of interest rates in the 6% and higher range when corroborating with our ABS issuance breakdown data. The underlying loan borrowers are thus paying a higher interest rate during the start of the 2008 crisis. The following attributes would have to be a major factor of the credit risky tranches. We were able to pull the last price data and create a mass spreadsheet of External Market Factors and last day prices for the credit risky tranches. The next move was to run Granger Causality tests on the new dataset to highlight what key Market Factors cause the fluctuation in the tranches and vice versa. Figure 1 represents the results of this test.
The outlying Market Factor is 3M Libor that unanimously causes fluctuations in the Tranche pricing. Though, this is expected since the tranches have coupon payments with the Libor rate. The Money Supply Index is the next significant cause that affects the Tranche Pricing. This is also intuitive since an increase/decrease in Money Supply directly affects the disposable income that is used to pay the underlying loans. An important note to keep in mind is the illiquidity of the Tranches as well as the frequency of economic indicators. In the end, the dataset to run models on isn’t as expansive. The more important causality is the kind of effect Tranches have on Market Factors. Figure 2 represents the results from the opposite test.
Figure 2: Granger Causality Results: Do Tranches Cause Change in External Market Factors.
Figure 3: represents the individual correlations with the Tranches and the respective External Market Factors.
Figure 4: Correlation Matrix.
One alarming observation is the significant cause of Tranches on the Case-Shiller Index. If such credit tranches were to default, they might affect the housing market, that initially led to the 2008 collapse. The high correlations between the Case-Shiller Index and the 4 Tranches further paints a picture of the effect on the housing market. The 3M Libor is another significant cause of the Tranches that is intuitive since their coupons are based on Libor. This fact can be further corroborated by high inverse correlations that describes the relation between interest rates and the pricing of such Asset Backed Securities. Thus any fluctuations in the Tranches would affect the Libor rates as well. There seems to be evidence of tranches significantly causing changes in the Consumer Price Index, the correlations are not too high to warrant as much of an effect. One feature of our analysis and data to take into consideration is the frequency of External Market Factor data and the illiquidity of the ABS Tranches. Often times, the size of present data for both series may not be large enough for the Causality and Correlation models. This is not to take away from the quality of the results but to inform of the right quality of our data. As part of our attempt to understand our data, we proceeded to visualize the geographical distribution of Student Loans within each ABS Issuance. We thus created a heat map with darker colors representing the higher concentration of loans within the particular state. Since all 4 credit risky tranches are from 2008 ABS Issuances, we looked at the heatmap for the year of 2008 where the loans from all of the respective issuances are included. As can be seen from the heatmap from 2008 in the Geographical breakdown section, states such as California, Florida comprise the most concentration of loans and are at risk of collapsing state economies if following students were to default.
Thus we were able to find and map conclusive areas of the economy where such ABS defaults might affect that include the housing sector, Libor rates etc and looked at the loans from a geographical point of view as well.
With a data set this large, on an ever-increasing market, there are many more avenues for research, both building on the work we have done, and diverging into different paths. It is our belief that the following topics would allow this research to grow to a new level. The possibility of looking into private loans instead of the public Sallie Mae loans is one topic to be explored. Unfortunately, because of the very nature of private loans, there is not an easy way to access data on them to continue this line of research. This study focuses heavily on one issuance, but there are different issuances that can be studied, and in a more quantitative approach, determining which issuances are riskier for investors based on different factors and attributes. We hope to be able to focus on this as one of our main courses of action moving forward. Another aspect that we want to work on is evaluating how this asset-backed security is priced, and develop new pricing methods based on the influx of quantitative data and analysis that is now possible. While there is only so much that is feasible for a researcher to accomplish, we hope this initial study has fueled a market for continuing research into the student loan market and its securitization.