Securitization in the auto market : a risky business

(1)

(2)

2 Statement of Originality

This document is written by Jim Pel who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

Securitization is often used as an example to illustrate how todays financial system can create an environment where financially sophisticated brokers take advantage of private investors. The global financial crisis has been argued to be a direct effect of securitization in the market for mortgages. In this thesis, the effect of securitization on delinquency rates of auto loans will be investigated. The main point is to research whether the often negative impact of this financial tool, as described in prior research, is also observed in the car market. This is done in two ways. First, prior research is investigated and some arguments about possible effects are found. Secondly, an empirical research, using data of the auto market, is done to research a statistical effect of both securitization and subprime loans on delinquency rates. This thesis does not find such a relation in this market, although theory suggests a positive effect. It is argued that moral hazard might not be present in the car market due to the depreciating value of cars. Further research can investigate how loan contracts are set up in the market, possibly explaining differences in the behavior of lenders.

(4)

4

1. Introduction

The world is recovering from what is widely regarded as the worst global financial recession since the Great Depression of 1930. Researchers agree that securitization caused the market for mortgages to collapse, initiating a worldwide financial disaster. This financial tool has therefore been the subject of much research in the past years. Meanwhile, a fair portion of securities outstanding consisted of car loans. This reached its peak in the year 2006, when the net worth of auto securities outstanding was 196 billion dollars in the USA (sifma). This was 12 percent of all non-mortgage related securities outstanding. Due to worldwide negative attention,

securitization decreased in the years following the recession. However, it is quickly regaining popularity. In fact, in the first quarter of 2015, auto securities outstanding amount to 185 billion dollars, which is almost back to pre-crisis levels and nearly a 25 percent increase since

2013. Furthermore, this is now 13,7 percent of all non-mortgage loans. Securitized car loans thus seem to be growing relative fast. It is therefore important to investigate whether the quick rise in auto securities poses a threat to default rates of car loans. In order to shed some light on the events preceding the crisis, it is important to identify how the banking system changed and how securitization led to a subprime lending crisis.

In the traditional banking system, banks would write out loans and hold them on their balance sheets, receiving monthly payments from their customers. This was called “orginate and hold”. According to Brunnermeier (2009), all loans were no longer held on banks’ balance sheets. Instead, they operated under a new form of banking, the so called “originate and distribute” model. After the loan was written out, it was sold to private investment companies. Those companies now received the monthly payments instead of the banks. This was done with the financial tool of securitization. Next will the process of securitization be explained.

First of all, a bank collects payments it receives on different types of loans, including mortgages credit card receivables and car loans and passes them on to investment companies. This so called pool of loans is then sliced up into tranches with different risk profiles and sold to individual investors. The risk profile is decided by the order in which payments out of the pool are made. The safest tranche will receive a relatively low return, but is paid out before all other tranches. On the contrary, the least safe tranche will receive their payments after all other payments are done. Investing in this security will thus expose an investor to the highest risk, but grants the highest return to compensate for the risk taken. It thus allows banks to pass on risk to private investors, making profits in the process of selling these securities to investment

companies.

Eggert (2009) states that securitization lays the foundation for a system of abusive lending practices, where subprime lenders can take advantage of both borrowers and investors. Securitization causes a rise in subprime lending, because banks no longer hold risk of default

(6)

6 and can give out mortgages to less than prime lenders more often than before. Investors

are given a sense of security, for they are protected against the liability and receive higher returns for their risk taken. But when default rates rise higher than expected, investors see their AAA-rated securities downgraded again and again. The demand for securities falls and they become almost worthless. This causes major investment companies to fall, dragging banks down with them. Questions must be asked if it is the use or rather abuse of securitization that leads to this financial bankruptcy.

As discussed, practices involving securitization have been the cause of worldwide financial distress in the past. This is derived from the argument that it leads to an increase in subprime lending, which in turn causes default rates to rise. This thesis will further investigate securitization and its implications on default rates of auto loans. The main question is whether securitization of these car loans lead to a higher default rate. It will try to answer this question, using quarterly data from the auto market. The data used will be of loans 90+ days delinquent. This is because delinquency rates give an indication of the inability to repay the loan. If

this thesis finds a relation between securitization of auto loans and delinquency rates, it will add to the evidence that these two are correlated. If this research does not find a correlation,

then it can be an indication that the effects of securitization are less severe in the auto market. Other research could then examine this further.

This thesis consists of a literature review and statistical analysis. First, existing literature on securitization will be reviewed in section 2. Much research on this topic has been done, especially in the context of subprime mortgages. Firstly it can explain why the demand for securities rose so much in the years preceding the financial crisis. Secondly, and more

importantly, the literature review will try to identify how the process of securitization can lead to more subprime lending, accompanied with higher delinquency rates. Further research on other factors that drive delinquency rates in the auto loan market will also be investigated. Then, section 3 will consist of a description of the dataset and all the variables

used. The empirical research in section 4 will then try to statistically investigate whether a relationship exists between securitization and delinquency rates. Multiple regressions, each time adding or subtracting variables will be done, in order to create a regression that tries to explain the dependent variable. Based on this model, some statements will be made about the effects of the explanatory variables on the dependent variable. Finally, section 5 will be the conclusion of this paper. Based on the literature review in section 2 and the statistical analysis in section 4, some conclusions will be made about the effect that securitization has in the market for auto loans.

(7)

7

2. Literature Review

This section will describe earlier research done on the topic of securitization. Most of this research has been about securitization of mortgages. But according to Heitfeld (2004), automobile loans are similar to residential mortgages. First of all, they are both held by

households. Secondly, they have a tangible asset as collateral, that can be repossessed by lenders in the case of default. Thirdly, both types of loans usually have a fixed payment scheme to repay the loans. Therefore Heitfeld argues that borrowers of both loans will act similar. Research on subprime mortgages can thus give insight in the subprime auto market. Researchers have found a relationship between securitization and subprime lending and default rates. For

example, Nadauld and Sherlund (2013) find that securitization generates an expansion of credit demanded. Furthermore they state that an increase of 75% in credit, results in 2.5-6.5% more subprime loans. Keys et al (2008) investigate the subprime mortgage market and conclude that a doubling of securities outstanding result in 10-25% more default. This section will try to analyze through which channels securitization can lead to more subprime lending and more default. To start of it will explain how banks could profit from the use of this financial tool. Next, the section will describe how the demand for subprime borrowers rises due

to securitization and how this leads to an increase in default rates of loans. Finally, literature on the drivers of default on auto loans will be reviewed. This literature combined, will construct a full picture of variables that can be used to statistically analyze the market for car loans.

2.1 Subprime loans

Popularity of securitization rose quickly because of multiple reasons. Firstly, risk can be more precisely distributed to investors. Because banks no longer hold the risk, they can charge lower interest- and mortgage rates, leading to a higher flow of capital. Second of all, regulations force banks to hold at least 8 percent of outstanding loans in capital. But because pooling of loans is an off balance sheet operation, this rule does not apply. Thus less capital is needed for giving out loans. Thirdly, the loans are collected in a diversified portfolio. This means that higher ratings are given to securities than they would receive individually. A loan that would normally receive a BBB-rating, could now receive an AAA-rating, because its risk can be diversified in the portfolio of loans. Mutual funds that are by regulations only allowed to invest in AAA-rated investments, are now able to invest in these assets. Again, leading to a higher flow of capital for banks (Brunnermeier, 2009)

Securitization can be a profitable business for banks and investment companies. By selling their securities to private investors, they collect profit without taking immediate default risk. As it becomes more popular, the demand for loans to securitize rises. So then why would

(8)

8 the percentage of subprime loans increase? Theoretically, subprime borrowers that are too risky for banks before, can now be diversified and be made into higher rating securities. And even if subprime borrowers entail more risk, investors that are willing to bear that risk are given higher returns as compensation. So subprime borrowers that would have been refused before, can now be accepted for these reasons. This fact does not necessarily pose a problem, if the amount of risk is properly recognized and distributed. Mishkin (2009) however, states that the market has been too optimistic and has not recognized the true risk associated with these tranches.

Tranches deemed safe, turned out to be risky as default rates increase, and resulted in losses on the securities of around $500 billion dollars in 2008.

2.2 Screening and regulations

So why did the market not recognize the true risk involved with these loans? Keys et al (2008) research the effect of securitization on screening and regulations surrounding

loan contracts. They do this by looking at the market for mortgage and find that because banks can pass on risk through securitization, screening and regulation standards decrease. In his research, it is stated that lenders deal with two types of information, when deciding on the terms of the loan contract. Hard information is the credit score of the borrower. This is based on factual quantitative data such as default history, accounting reports and financial foreclosure. This is used by rating agencies to decide what rating the loan will get in the securitized pool. It is therefore important information in the securitization process.

However, soft information is more subjective and less quantifiable. It takes into account personal factors, such as future prospects and the long-term relationship between bank and borrower. He finds that the degree of screening for soft information depends on the amount of risk the lender bears. This is where a problem arises with securitization. Because the lender, in this case the bank, can pass on the risk of default on written loans, they have less incentive to perform a proper screening on soft information. Borrowers that would be rejected before due to low “soft scores”, were now allowed to take a loan because the risk of their default was shifted from banks to individual investors. This is a form of moral hazard. Banks do not fulfill their duty to properly screen borrowers and create a safe environment for investors, because they can pass on the risk anyhow.

Eggert (2009) provides further arguments, linked to screening and regulations. He argues that even though hard information is still properly screened, the market is pushing the limits of what is responsible. Because there are many different investors involved in the securitization process, the quantity rather than quality of loans is of interest. Investment companies are asking for more loans so that they can sell them to private investors. In this process, the quality of loans do not matter as much, as long as they are sellable. In other words, a

(9)

9 lack of screening can also be in the interest of investment companies. Banks therefore create the largest, riskiest loans that are still legally allowed to be sold to private investors, with as little credit enhancement possible needed to get the highest ratings. The pushing of boundaries in combination with laxer screening and regulations means that these securities are often more risky than anticipated by the market. Moreover, Eggert states that defaults in a pool of loans has a cascading effect. Once certain loans get written off, the whole pool can get downgraded. So even the highest rated loans in that pool will then be downgraded.

2.3 Predatory Lending

Lack of proper screening is not the only problem that arises from securitizing loans. According to Reiss (2010) , subprime loan contracts are less regulated and standardized than prime loans. Therefore it is more prone to predatory lending. Although it does not have an official

definition, it consists of practices by lenders that are disadvantageous to borrowers. Because there are less regulations in the subprime market, lenders can set up loan contracts in ways that it becomes beneficial for themselves more than for the borrowers. Quercia (2010) adds that subprime borrowers are more vulnerable to predatory practices. This is because the percentage of minorities, low-income and elderly people is higher among subprime borrowers. It is exactly this group that is more prone to “push-marketing” and are generally less experienced

with lending. Moreover, they tend to have a higher need for credit. All of these

factors combined put subprime borrowers in a weak bargaining position, where they are easily taken advantage of by financially sophisticated brokers.

So how can predatory lending be put in to numbers? In his paper, Stein (2001) estimates that 9.1 billion dollars is lost every year to these practices. He describes three forms of predatory lending. Firstly, equity stripping means that subprime loans are accompanied by exorbitant fees, either paid upfront or in the form of prepayment penalties or financial credit insurance.

Often packaged in such a way that it is difficult for inexperienced borrowers to estimate the cost beforehand. These fees do not have to be paid by prime borrowers. Next he states that interest rates on loans are too high for the amount of credit risk the borrowers represent. It is estimated that subprime borrowers pay 1% interest more than is justified by their credit risk. Finally, lenders write out loans without investigating whether or not the borrowers will be able to repay them. He finds that this leads to excessive default rates. It is expected that subprime borrowers will default on their loans more often than prime borrowers, but controlled for this fact,

subprime borrowers have to default on their loans even more due to predatory lending practices.

(10)

10 The existing literature thus shows how securitization became a popular financial tool used by banks and investment companies. A better distribution of risk, less capital needed and diversification of risk are advantages of securities. However, this section has also revealed the darker side, involving lax screening and predatory lending practices, highly disadvantageous for subprime borrowers. This is exactly the reason it has been the subject of worldwide discussion and research. As the consequences of improper use of securitization have been felt globally. In short, a possible positive effect of securitization on delinquency rates, through the channel of subprime lending might be expected.

2.4 Liquidity constraints

Besides possible effects of securitization on delinquency rates as described before, liquidity constraints are an important factor driving defaults in the auto market. Liquidity

constraints mean that households have a limit to the amount of money they can

borrow. It creates imperfect intertemporal consumption in the sense that consumers can’t perfectly use future income by borrowing money today. In a world without these constraints, unexpected short-term shocks in income would make no difference on the ability to repay a loan, as borrowers could use future income to repay their loans of today. Only long-term income would have an effect, because consumers would be perfectly able to borrow and lend money. However, because of the existence of liquidity constrains, short-term income shocks do have an impact on default rates of loans (Heitfeld, 2004). It is therefore expected that default rates on auto loans rise after a negative income shock.

Adams et al (2007) research the effect of liquidity constraints on moral hazard and adverse selection in the market for auto loans. Adverse selection arises because borrowers with lower income and thus a higher risk of default, are exactly the ones looking to get a loan in economically bad times. Adverse selection can partly be avoided by credit scores and other statistics that decide the creditworthiness of a borrower. But as been argued before, screening can worsen due to securitization. So adverse selection becomes a bigger problem when banks fail to screen their borrowers. Furthermore, he finds that the adverse selection problem also induces moral hazard. This is rooted in the fact that adverse selection brings more subprime borrowers, who in turn tend to have higher loans due to a lower down payment. Why this is, is described in detail in their paper but the general idea is as follows. The higher the monthly payments, the higher the chance that defaulting on the loan is in the interest of the borrower, even if he could technically pay. In fact, they find that for any given borrower, a $1.000 dollar increase in loan size, increases the probability of default by 16 percent. So there is not only moral hazard on the lender side, in the sense that banks would not thoroughly screen

(11)

11 new loans due to the lack of risk, also borrowers show signs of moral hazard, possibly leading to higher delinquency rates under subprime borrowers.

A final possible impact on default rates on loans is the LTV-ratio, or the loan-to-value ratio of a loan. Research suggests that the chance of default on a certain type of loan, holding all else constant, decreases when this LTV-ratio decreases. That is, if the value of the loan in terms of the collateral decreases, the chance of default falls. For example, mortgages have a decreasing LTV-ratio. While prices of houses tend to remain the same or increase, the value of the loan decreases as it gets paid off. Therefore, ceteris paribus, the chance of default on mortgages decreases over time. A car, however is a different type of collateral in the sense that its value decreases over time, along with the decreasing loan value. The LTV-ratio thus either stays constant or increases. An increase could occur if the value of the car falls quicker than the value of the loan. If this is the case, an increasing LTV-ratio could mean that there will be a certain point in a car loans lifetime where this ratio is higher than at the start and therefore, the

probability of default is higher than it was before. This might be during the first year, as the price of a car decreases non-linearly. That is, a drop in its value will be greater in year one than in year two. According to Carsdirect, a car depreciates 20 percent in the first year. Then, a yearly

depreciation of 15 percent is observed from year two until five. If the payments on a loan contract are constant, then the first year will have the highest LTV-ratio and thus the a positive effect on delinquency rates.

2.5 Chapter Summary

Now that the literature on this subject has been reviewed, some conclusions can be made on the expected effects of securitization on delinquency rates. First of all, some predictions can be made about the popularity of securitization through the last 15 years. A yearly increase is expected from 2000 until 2006, when the financial crisis began, because of advantages for banks such as distribution or risk, diversification and rules regarding required capital After this a decrease is expected during the financial crisis and all the negative publication on securitization. However, because of the potential profits and advantages banks get, there might be an increase of its popularity in the last few years.

Furthermore, earlier research can explain why certain variables are important in the analysis of securitization and delinquency rates. Firstly, securitization can lead to moral hazard by banks, in the form of lax screening. As argued before, mainly the screening for soft

information is affected. So it can lead to higher default rates, even if the percentage of subprime loans does not increase. Subprime lenders are defined by a credit score lower than 620, but perhaps banks were hesitant to accept borrowers with a credit score slightly above that. So

(12)

12 without securitization, perhaps these borrowers would not be accepted based on low soft

scores, or would at least be thoroughly screened. Without this incentive, default rates on borrowers just above subprime might increase as well.

Secondly, it can have a positive effect because demand for subprime loans may rise following an increase in securitization. Because risk is redistributed and banks have less skin in the game, they have less of an incentive to refuse certain borrowers from taking a loan.

Moreover, they have less incentive for proper screening and regulating these loans. The combination of less screening and more subprime loans can result in higher delinquency rates. Next, predatory lending practices have shown that subprime borrowers are subject to rather disadvantageous loan terms. Again, this means that because of the bad loan terms, subprime borrowers are expected to have higher than expected default rates. Concluding, there might be a positive relationship between securitization and delinquency rates, caused by a less screening for soft information and an increase in the demand for subprime loans by banks.

Next, research on what drives delinquency in the auto market has been investigated. Some arguments arose that are important for the statistical analysis. Due to liquidity constraints, shocks in liquidity are proven to have an effect on delinquency rates. Next, these constraints have proven to induce moral hazard and adverse selection problems. This could possibly mean that subprime lending is also correlated with liquidity shocks, although evidence for this claim is weaker.

Finally, LTV-ratios could have an impact on delinquency rates. A way to measure this is through changes in total auto sales in a certain period. Namely, if car sales increase, then a positive effect on delinquency rates can be seen in the first year, when the LTV-ratio is highest. Therefore, the variable of auto sales is included in the empirical research, to investigate whether this claim has a statistical argument.

(13)

13

3. Methodology

This section consists of an empirical research, using relevant variables described in section 2. First, a description is made of all the data uses. Then, a regression is done, describing the results and deriving some implications from this result. In the previous section, various arguments have been positive effect of securitization on delinquency rates of auto loans. This section will

research the effect empirically, using historical data. It will do this using Linear Regression (OLS). First it will describe the regressions and which dependent and independent variables are used. Then the dataset will be described and shown graphically. Finally, the results of the regression are written down and some implications discussed

3.1 data description

. All the variables are measured over time, so it is a time-series analysis. In order to perform such an analysis, unit roots of the variables are tested. The Dicky-Fuller test for stationarity is used to see if a regression can be performed on variable levels or differences. The results are shown in below. Variable Test statistic 10% critical value delinq -0,762 -2,6 securit -1,98 -2,6 subprime -2,08 -2,6

The variables are non-stationary at the 10 percent level, therefore a regression on levels could be spurious. In order to create stationarity in variables, they are all first differenced, including the control variables. The model that is used is the following.

∆Y = 𝛽0 + 𝛽1∆𝑋1t-1 + 𝛽1∆𝑋1t-2 + 𝛽1∆𝑋1t-3 + 𝛽1∆𝑋1t-4 + 𝛽∆2𝑋2t-1 + 𝛽∆2𝑋2t-2 + 𝛽∆2𝑋2t-3 + 𝛽∆2𝑋2t-4 + 𝛽3∆𝑋3t-1+ 𝛽3∆𝑋3t-2+ 𝛽3∆𝑋3t-3+ 𝛽3∆𝑋3t-4 +𝛽4∆𝑋4t-4 + 𝛽4∆𝑋4t-8 + 𝛽4∆𝑋4t-12

+ 𝛽4∆𝑋4t-16 + 𝜀

Where y is denoted as delinq. The independent variables are denoted as follows: * 𝑥1 as lnsecurit

* 𝑥2 as subprime * 𝑥3 as unemploy * 𝑥4 as lnautosales * 𝑥5 as lngdp

(14)

14 * 𝑥𝑖 as a dummy variable for a seasonal effect

The variables for securities and auto sales are transformed to logarithmic values, because they are denoted in dollars, while the other variables are percentages. To match these variables, logarithmic differences are regressed. Furthermore, Xi is a dummy variable for seasonal effects. This can take on the values 1 until 4. The regression will leave out 1, while adding 2-4 to control for seasons.

Now the data used in the regression will be described. In order to analyze the effect of securitization on delinquency rates and subprime lending, multiple datasets have been used. Historical data on Asset Backed Securities outstanding has been provided by sifma, from the 1st quarter of 1999 until the 4th quarter of 2014. From 2006 onwards, the data has been provided quarterly. From 1999 until 2006, however, only yearly data is available. In order to create a dataset completely in quarters, a mutation has been performed. Yearly data from before 2006 has been interpolated into quarterly data points. This has been done as follows: denote the change in auto securities outstanding from year t-1 to year t as ∆x, then the change in securities outstanding from each quarter to the next in year t-1 is ∆x/4. This is done because both

delinquency rates and subprime rates have been found quarterly, starting 2000. In order to match their longevity while maintaining the quarterly variances over this period of time, interpolation of securities outstanding has been done.

Variable Description

delinq Auto loans 90+ days delinquent

lnsecurit Log Auto Securities outstanding (in billions)

subprime Percentage of subprime auto loans in the market

unemploy Unemployment rate

lnautosales Log Total sales of automobiles (in thousands)

lngdp Log GDP

Table 3.1: Variable description

Variable Obs Mean Std Dev Min Max Time Source Unit

delinq 64 0,0312 0,0106 0,0188 0,0527 15 years New York Fed Percentage

securit 64 156.761 30.556 93.749 218.523 15 years sifma US Dollars

subprime 64 0,02515 0,0418 0,175 0,239 15 years Ycharts Percentage

unemploy 64 0,0627 0,0181 0,039 0,0993 15 years B.L.S Percentage

autosales 64 1.847.579 290.867 1093.3 2433.3 15 years Motor Intelligence US Dollars

(15)

15

Table 3.2: Variables Summary

Now, the data will be shown graphically and discussed briefly. It can give us an image of how the variables have developed through time, and some remarks are made. The timelines are shown graphically on the next page in figure 3.1 until 3.6.

Firstly, quarterly information on delinquency rates and subprime lending has been obtained from Ycharts and the New York Fed respectively. The criteria used for delinquency rates is 90+ days and subprime lenders are defined by lenders with a credit score lower than 620. Both graphs are shown below. Figure 3.1 shows delinquency rates and figure 3.2 the percentage of subprime loans from 1999-2015. Note that percentages are written down on a scale of 0-1. The percentage rate of delinquencies remained relatively stable from the year 2000 through 2007. However from 2008 onwards there was a spike in delinquencies to 5% suggesting that the rate doubled. From the year 2000 to 2006 subprime loans remained relatively stable with gentle peaks and troughs throughout. Between the years 2008 – 2010 there was a noted drop in subprime loans which subsequently remained stable again albeit at a lower rate than previously shown years.

Figure 3.3 illustrates the total amount of auto securities outstanding through the years. This graph shows a sharp rise in auto securities from 2000 – 2004, ultimately peaking in 2007. Following this, a large drop was seen until 2010 after which it remained stable. We can then see a progressive increase from 2011 onwards The unemployment rate, in figure 3.4, rose from the year 2001 to 2002 albeit remaining within the values of 4 to 6%. We can then see a significant spike in 2008 which then dropped steadily by 2014.

Last, GDP and total auto sales outstanding are presented in figure 3.5 and 3.6. GDP grows steadily from 1999 onwards, except for a decrease in 2008, caused by the financial crisis.

Because GDP has the tendency to rise yearly, it might not be a good variable to measure income shocks. It will therefore not be included in the regression. Auto sales have peaked and troughed consistently until 2008. This is a clear example of a seasonal effect. Therefore a seasonal dummy has been added to the regression, as state before. There was then a drop which was followed by a gentle rise in auto sales through to 2014.

(16)

16

Figure 3.1: delinquency rates of auto loans, in percentages. Figure 3.2: Subprime loans in percentages

Figure 3.3: total auto sales outstanding in (bln) USD Figure 3.4: Unemployment rates in percentages

Figure 3.5: GDP, expressed in (bln) USD Figure 3.6: Total Auto sales in (thousands) USD 0 0,01 0,02 0,03 0,04 0,05 0,06 Delinquencies 0 0,1 0,2 0,3 0,4 Subprime 0 50 100 150 200 250 Auto Securities 0 0,02 0,04 0,06 0,08 0,1 0,12 Unemployment 0,0 5000,0 10000,0 15000,0 20000,0 GDP 0 500 1000 1500 2000 2500 3000

Auto sales

(17)

17

4. Regression Results

In this section, the variables are tested using Linear Regression. First, the regression is shown and some significant results discussed. Then, some practical implications are derived. In table 3.3 the results of multiple regressions are shown. Because delinquencies are defined as 90+ days, there can only be an effect after a lag of 1 period minimum. This is because periods are expressed in quarters, which is 90 days. So the effect of a change in one of the variables will only be able to turn into a delinquency after 1 quarter.

Model OLS regression: delinq

Variables (1) (2) (3) (4) (5) Δ Ln(securit) Δ Ln(securit) t-1 -0,016** -0,013** (-2,19) (-2,43) Δ Ln(securit) t-2 -0,011 -0,009** (-1,66) (-2,12) Δ Ln(securit) t-3 0,002 (0,24) Δ Ln(securit) t-4 -0,004 (0,75) Δ Subprime Δ Subprime t-1 -0,033*** -0,005 -0,009 (-2,86) (-0,28) (-0,55) Δ Subprime t-2 -0,014 0,031 0,025** 0,021 0,022** (-1,13) (1,65) (2,18) (1,06) (2,09) Δ Subprime t-3 -0,008 0,024 0,015 0,011 0,015 (-063) (1,30) (1,37) (0,54) (1,59) Δ Subprime t-4 0,011 0,007 0,013 0,013 0,015 (0,63) (0,48) (1,32) (0,91) (1,65) Δ Unemploy Δ Unemploy t-1 -0,001 0,001 (-0,18) (0,13) Δ Unemploy t-2 0,021** -0,023*** 0,013 0,013** (2,50) (5,77) (1,46) (2,65) Δ Unemploy t-3 0,009 -0,003 (1,01) (-0,33) Δ Unemploy t-4 -0,004 -0,001 (-0,50) (0,07)

(18)

18 Δ Lnautosales Δ Lnautosales t-4 -0,006* -0,006** -0,008** -0,006** (-1,92) (-2,33) (-2,16) (-2,37) Δ Lnautosales t-8 0,000 0,002 (0,16) (0,81) Δ Lnautosales t-12 0,006* 0,006** 0,008** 0,008*** (1,98) (2,15) (2,53) (3,14) Δ Lnautosales t-16 0,003 0,000 (0,75) (0,09) Q 2 -0,001* -0,002* -0,002** -0,002* -0,002** (-1,95) (-2,03) (-2,16) (-1,73) (-2,46) 3 0,002 -0,000 0,000 0,000 0,001 (0,74) (-0,18) (-0,18) (0,40) (0,81) 4 0,001 0,001 0,001 0,001 0,001 (1,21) (0,62) (0,75) (1,02) (1,62) Constant 0,000 0,001 0,000 0,000 0,000 (0,79) (1,38) (1,11) (0,59) (0,71) Obs 62 62 62 62 62 Adj R-Sq 0,3169 0,4915 0,5451 0,5412 0,6333 R-squared 0,4464 0,6573 0,6269 0,7307 0,7140

Note: t-statistics in parentheses

Column 1 shows the regression of subprime rates on delinquencies, because subprime loans are expected to be an important driver of the effect caused by securitization. The independent variable is lagged four times. These lags represent each quarter up to a year, because the risk of default is the highest in the first year (Heitfield, 2003). Therefore the assumption is made that if there is an impact on the dependent variable, this can be seen in the first year. Only the first lag coefficient of subprime loans has a t-statistic that is significant at the 5 percent level. However, this coefficient is negative, while theory clearly suggests a positive relation. This might be an indication that macro-economic variables such as unemployment and auto sales should be included in the regression, in order to create a better prediction. Finally, a variable dummy for seasonal effects has been included. Season two is significant at the 10 percent level, but this might be due to seasonal effects in auto sales, as was observed in the data description.

The second column shows the results when only the variable for subprime rates is used in the regression, along with the macro-economic control variables total auto sales and

(19)

19 might have a correlation with subprime lending. In order to observe this possible correlation, securitization is left out at first, but added in column 4 and 5. If the coefficients of subprime rates change, this might be an indication of multicollinearity, because it indicates that some of the effect of securitization is taken on by subprime rates. All the independent variables are lagged 4 times. Note that for auto sales, yearly lags are used. As previously stated, different years might represent different LTV-values and probabilities of default. The first lag of subprime rates has a negative and non-significant sign, indicating that an (positive) effect of subprime lending might not yet be observed after the first quarter. The second until the fourth lag all have positive effects, although none of the coefficients have a t-statistic that is significant at the five percent level. For the control variables, any order of lag with a t-statistic of at least one, might indicate that it has some effect on the dependent variable.

In column 3, a similar regression has been carried out but with the most relevant lags. The effect of subprime loans on delinquency rates as described in literature is quite

unambiguous, because subprime borrowers represent people with lower credit scores, lower income and less financial knowledge. So they are expected to have a higher probability of default on their loan. The impact could even be strengthened by predatory lending practices by banks, which may or may not be present. The order of lags used for each variable is derived from column 2, using lags with relatively high t-statistics for the control variables and all the lags except for the first one for subprime lending. Subprime lending has a significant result in period 2, while period 3and 4 do not show significant t-statistics at the 10 percent level. The macro-economic variables for unemployment and total auto sales both have positive and significant coefficients. Unemployment rate has a positive effect at the 1 percent level, while auto sales seem to have a negative and positive effect in the first year and the third year respectively.

Column 4 and 5 include securitization as an independent variable, as well as subprime rates. Again, all variables are lagged four times at first in column 4, to acquire some information on the relevant lags. A first glance shows that securitization might have an impact after the first and second lag, although it is a negative. This is different than what was argued in the literature review. Subprime lending and unemployment rates both have positive values with a t-statistic of more than 1, indicating that they can have a positive effect on delinquency rates. Auto sales has a negative, significant coefficient after the fourth, but a positive significant coefficient after the twelfth lag. The relevant lags observed in column 2 thus still seem to apply for these variables. Again, seasonal variances have been controlled for. Column 5 is similar to the regression in column 3, but this time the first and second order lag of securitization are included. Both

generate significant t-values at the five percent level, but negative coefficients. This negative sign is hard to justify, but the statement can be made that this regression does not show that

(20)

20 securitization increases delinquency rates. Moreover, the coefficients of subprime rates are similar to the ones obtained in regression 3, indicating that subprime lending and securitization might not be highly correlated. Column 3 and 5 are used to deduce some arguments for the effect of subprime loans and securitization on delinquency rates.

4.1 Implications

So what do these results illustrate? First of all, securitization is expected, according to prior research, to lead to higher delinquency rates. However, this regression does not find a positive relation. The coefficient even turns out negative, although this effect seems unlikely. It is hard to find an argument as to why securitization would lead to a decrease in delinquency rates. It can however suggest that in the car industry, the problem of moral hazard by banks is not present as much as in the market for mortgages. It is argued that moral hazard leads to a decrease in soft screening, possibly leading to higher delinquency rates for both subprime and prime borrowers. The fact that this is not found in this thesis, can indicate that screening standards might be higher in the auto market. It is possible that this is because of the

characteristics of car prices. Because cars serve as collateral on the loans, these characteristics can be an important factor in explaining this finding. Prices of cars tend to decrease rapidly in the first year. If a loan defaults, the collateral has lost a big part of its value already. In other cases such as mortgages, it was believed that the collateral, namely houses, had ever rising prices. If the borrower would default, the new owner of the collateral could always sell the house for a profit. This is impossible with cars, as the value has depreciates.

Next, the results of subprime rates are discussed. The third column shows subprime rates in separation of securitization, while the fifth column adds the 2 variables together. Both times, subprime lending has a positive and significant coefficient in period 2, implying a positive effect on delinquency rates after roughly 6 months. This is expected from the literature review. Subprime borrowers have a higher probability of default, so an increase of these borrowers will then lead to an increase in delinquency rates. This result is hardly relevant in relation to

securitization. The main point in this thesis is to find a relationship between securitization and delinquency rates, also caused by an increase in the demand for subprime loans. However, looking at the differences in coefficients in column 3 and 5 might give be an indication of a low correlation between subprime lending and securitization. Because the coefficients of subprime lending do not differ much in both columns, it does not seem to pick up much of the effect of securitization. In other words, the correlation seems low. A test shows that the correlation between the two variables is 0,18. This adds to the argument that securitization of auto loans might not be related to a higher demand for subprime loans. This might also be the reason why the coefficient for securitization is not positive. Because a large part of the impact of securities is

(21)

21 caused by an increase in subprime lending, along with laxer screening. If subprime lending and securitization are not found to be positively correlated, then this argument no longer holds.

Unemployment rate seems to have a significant, positive effect on delinquency rates. The arguments about liquidity constraints thus seem to be of importance in the industry for cars. An income shock such as a rise in unemployment will cause people to default on their loans more often. This is intuitive, a decrease in income means that more people will have problems making monthly payments, resulting in a rise in delinquency rates. This is in agreement with the

research of Heitfield, that suggests that defaults in the auto market are also driven by liquidity constraints (2001).

Finally, auto sales have been added to the regression to find some results regarding LTV-ratios and default rates of auto loans. A negative effect on delinquency rates is shown after one year, while a positive effect is found after the third year, both with significant t-statistics. This is an indication that there might not be a relationship between LTV-ratios and delinquency rates as described earlier. The logic behind it is that if sales do have an impact on delinquency rates, while holding other variables equal, then it might say something about after how much time people are more likely to default. It was argued that the value of a car decreases the most in year one. This could mean that the LTV-ratio would be highest in year 1 so that the chance of default is higher in this year. This seems to not be the case in this regression, where sales have a positive impact after one year. This could mean that loan contracts are constructed in such a way as to avoid the problem of first year default for banks. They also know that the price of a car decreases most in the first year. Perhaps they anticipate this and construct the loan contract accordingly. The positive coefficient after year three might be an indication that this year has the highest probability of default. But in order to make a proper statement about this, the loan terms need to be further investigated to see how LTV-ratios develop during the terms of loan contracts.

(22)

22

5. Conclusion

This thesis has investigated the relationship between securitization and delinquency rates (90 days) in the market for auto loans. The importance of investigating this is that securitization of mortgages is regarded to be the cause for the global financial crisis. It is thus important to keep an eye out for these practices and keep investigating whether or not it is used in a proper way. And although many researches have investigated the effects of Mortgage-backed securities, research on the implications in other markets such as the auto industry seems scarce.

Economic literature has pointed towards arguments for a positive relationship between securitization and delinquency rates. The first driver behind this effect is that it might lead to a higher demand for subprime loans, because of the way risk is distributed under private

investors. Furthermore, the distribution of risk might induce moral hazard by banks. Screening and regulations become weaker as the lender bears less risk, possibly causing higher

delinquency rates in both prime and subprime loans. Furthermore, subprime borrowers might be the subject of predatory lending practices, involving high fees, down payments or interest rates. All these factors combined lead to believe there might be an effect of securitization on delinquency rates in the auto market.

Next, this has been investigated using a linear regression. In order to isolate the possible effects of subprime lending and securitization, control variables have also been derived from the literary review. Liquidity constraints are considered to have an impact in the ability to repay loans. Therefore, negative income shocks can be expected to cause more people to become delinquent on their loans. Finally, data on total auto sales have been included to investigate a possible effect of LTV-ratios. It was argued that the loan-to-value ratio might be the highest in year one, resulting in higher delinquency rates in this year.

The empirical results however only partially coincide with the theory. In this thesis, no direct positive impact of securitization on auto delinquencies have been found. The significant coefficient was negative, although this does not seem to have an economic explanation. A positive, significant effect was found on delinquency rates, which was highly expected based on existing literature. This is because subprime borrowers have a higher probability of default. Furthermore, unemployment rates had a positive effect on defaults, as was predicted by the liquidity constraint theory. Finally, auto sales are found to negatively impact delinquency rates after one year, while a positive effect is found after the third year. The former result might be of more importance in this thesis, because it can be an indication that loan contracts are set up differently in different markets.

(23)

23 So what are the implications of this thesis? The key result is that securitization does not seem to have a positive effect on delinquency rates in the auto market, even though this was argued by previous literature. Most of this literature has been in the context of mortgages. A market in which it had a great negative impact. Since then, the practice of securitizing has been the subject of much critique, maybe rightfully so. This research however, might show that it does not necessarily has these negative effects on borrowers. A great deal of the effect is caused by a higher demand for subprime loans, which is not found here. Also, moral hazard by lenders due to risk distribution does not seem to be present in auto loan market. If more research adds to these findings, it can indicate that different markets react differently to it. The important step is then to investigate what exactly causes these differences. This can help to create a responsible, relatively safe financial environment for investors in asset-backed securities of all loan types.

In order to further investigate these statements, characteristics of cars and car loans must be researched thoroughly. This research has not taken into account the way loan contract terms are set up. It is thus limited in explaining exactly why securities do not seem to have an effect on delinquency rates. Furthermore, limited data on securities outstanding has been available, which was solved by interpolating the yearly data points from 1999 until 2006. This might have led to some distortion of the regression. In conclusion, based on this research, securitization does not seem to result in more delinquency rates in the market for auto loans.

(24)

24

6. Reference List

Adams, W., Einav, L., & Levin, J. (2007). Liquidity constraints and imperfect information in subprime

lending (No. w13067). National Bureau of Economic Research.

Eggert, K. (2008). Great Collapse: How Securitization Caused the Subprime Meltdown, The. Conn. L.

Rev., 41, 1257.

Heitfield, E., & Sabarwal, T. (2004). What drives default and prepayment on subprime auto loans?. The Journal of real estate finance and economics , 29(4), 457-477.

Keys, B. J., Mukherjee, T., Seru, A., & Vig, V. (2009). Financial regulation and securitization: Evi dence from subprime loans. Journal of Monetary Economics,56(5), 700-720.

Mishkin, F. S. (2010). Over the cliff: From the subprime to the global financial crisis (No. w16609). National Bureau of Economic Research.

Nadauld, T., & Sherlund, S. M. (2009). The role of the securitization process in the expansion of subprime credit. Available at SSRN 1410264.

Quercia, R. G., Stegman, M. A., & Davis, W. R. (2007). The impact of predatory loan terms on subprime foreclosures: The special case of prepayment penalties and balloon payments. Housing

Policy Debate, 18(2), 311-346.

Reiss, D. J. (2006). Subprime Standardization: How Rating Agencies Allow Predatory Lending to Flourish in the Secondary Mortgage Market. Florida State University Law Review, 33

Stein, E. (2001). Quantifying the economic cost of predatory lending. Center for Responsible Lending,