The relationship between mortgage defaults and mortgage-backed securities : the role of the financial crisis

(1)

UNIVERSITY OF AMSTERDAM

AMSTERDAM BUSINESS SCHOOL

Bachelor in Finance & Organization

Bachelor Thesis

THE RELATIONSHIP BETWEEN MORTGAGE

DEFAULTS AND MORTGAGE-BACKED SECURITIES:

THE ROLE OF THE FINANCIAL CRISIS

Submitted by: Hidde Schipper

Supervised by: Dr. Martijn Dröes

(2)

Abstract

Mortgage-backed securities and mortgage defaults had a large part in the financial crisis of 2008. Previous research mainly focused on what happened during and before the financial crisis concerning the relationship between mortgage defaults and the value of mortgage-backed securities. This study investigates the relationship between those mortgage defaults and the value of mortgage-backed securities in the time period after the financial crisis, which is from the 1st of January 2010 till the 1st of March 2018. During and before the financial crisis mortgage defaults had first no effect on the value of mortgage-backed securities, but were eventually incorporated in the value of mortgage-backed securities which led to disastrous situations. The value of mortgage-backed securities had a positive effect on mortgage defaults, meaning the high demand for mortgage-backed securities resulted in more mortgage defaults. This study expects this relationship to change, due to the reform of the financial system with new regulation. Mortgage defaults should now be incorporated in the value of mortgage-backed securities, meaning that this study expects a negative relationship between mortgage defaults and the value of mortgage-backed securities. Whereas mortgage-backed securities should no longer have a positive effect on mortgage defaults, thus this study expects no effect of the value of mortgage-backed securities on mortgage defaults. Because of the simultaneous causality between mortgage defaults and mortgage-backed securities, this study uses two regressions with both the variables as the dependent as well as the main explanatory variable. As a preferred specification, also a VAR-model is used. In order to be able to investigate the longer-term effects, orthogonalized impulse response functions and a forecast error variance decomposition are calculated. This study did not find any significant statistical evidence that mortgage defaults have an effect on the value of backed securities. This could mean that mortgage-backed securities are still not valued accurately, which was also the case before the financial crisis, and led to disastrous situations. Neither did this study find any significant statistical evidence that the value of mortgage-backed securities has an effect on mortgage defaults, which is in line what this study expected.

Statement of originality

This document is written by Hidde Schipper who declares to take full responsibility for the content of this document. I hereby state 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 of the University of Amsterdam is not responsible for the content of the work, merely for guidance.

(3)

1. Introduction

Mortgage-backed securitization has played an influential role in the financial world since the 1900s. Since the beginning of the 21st century the market of mortgage-backed securities kept rising until the almost total collapse of this market around 2008. There have been numerous papers written about what were the causes of the financial crisis, this will be discussed in the literature section. The restricting conditions of obtaining a mortgage loan for consumers, pre-2008, were basically non-existent. Almost every consumer applying for a mortgage loan received one, even when their credit was questionable, and that is an understatement (Keys et al, 2010). These loans were pooled together and transformed into mortgage-backed securities, and accredited an AAA or an AA rating, which are the highest ratings available (Ashcraft et al., 2011). These mortgage-backed securities were repackaged again, into the so-called “collaterized debt obligations”. This demand for securitization caused for an increase of approximately 20% of mortgage defaults, due to reduced incentives of lenders to control for the risk that borrowers had (Peicuti, 2013). This, among other trends, caused a major bubble, because these securities and derivatives were classified as top-notch, but were in fact not and consisted of bad mortgages. When the bubble burst, it led to sudden huge write-offs for companies due to a large decrease in the value of mortgage-backed securities when people found out what the mortgage-backed securities really consisted of (Bhat, Frankel & Martin, 2011). A part of what caused this large decline of value can be attributed to mortgage defaults. Even though mortgage defaults were rising there was not really an effect on the rating of these securities and there was no decrease in value which makes no economic sense (Fender & Scheicher, 2008). Eventually mortgage defaults, among other things, were incorporated in the value of mortgage-backed securities and caused such a sharp decline that led to the bankruptcy of many companies.

Today’s economy has recovered and the faith of the public is more or less restored. However, Bloomberg issued press releases recently in which they warned that banks are, once again, looking into riskier mortgages to pool together into a mortgage-backed security (McNeely & Quinn, 2018). That is not all, regulators are also signing off on banks selling off some of the risk they have on their balance sheets using derivatives. Bloomberg states that these transactions are not as dangerous as in 2008, but they are still risky and a cause for concern (Abramowicz, 2017). Taking all these events in the past and now together, the main research question of this study will be:

(5)

“What is the relationship between mortgage defaults and the value of mortgage-backed securities in the period after the financial crisis?”

This study expects a relationship between mortgage defaults and the value of backed securities, because during and before the financial crisis, the demand for mortgage-backed securities resulted in higher mortgage defaults, whereas these mortgage defaults caused at the beginning of the financial crisis little to no impact on the value of mortgage-backed securities (Fender & Scheicher, 2008; Peicuti, 2013). When following economic intuition, if mortgage defaults rise, it is riskier to invest in mortgage-backed securities and therefore the value decreases. During and before the financial crisis markets were not functioning optimally and therefore this reasoning did not apply (McSweeney, 2009). However, this study expects that the relationship has changed after the financial crisis, due to the reform of the financial system after the crisis, because new regulation was implemented (Duffie, 2016). The relationship between mortgage defaults and the value of mortgage-backed securities should now follow economic intuition and have a negative relation with the value of mortgage-backed securities, and the value of mortgage-backed securities should no longer have an effect on mortgage defaults. This is further specified in the hypotheses paragraph. This study contributes to existing literature, because most research focused on only mortgage-backed securities or mortgage defaults. If the relationship between these two variables was taken into account, the time frame of the research was before the financial crisis, whereas this study focuses on the period after the financial crisis.

Several sources are used to obtain the data that are necessary to conduct this study. The first explanatory variable, which is the value of mortgage-backed securities, is extracted from the Economic Research section from the Federal Reserve Bank of St. Louis. It contains the values of mortgage-backed securities from all commercial banks in the United States. It consists of monthly data, which starts in July 2009. In this study, the time period will be from January 2010 to March 2018. The second explanatory variable, mortgage defaults, is obtained from the S&P/Experian First Mortgage Default Index. This dataset consists of monthly percentages of consumers credit accounts going into default at the first of the month. The other variables that are included in the model are control variables, these are used to reduce the noise in the error term and to obtain an unbiased estimate. The control variables that are included in the model are average housing prices, inflation, mortgage interest rates and the value of other asset-backed securities. All these variables are described in monthly data (United States Census Bureau,

(6)

2018; Freddie Mac, 2018; Board of Governors of the Federal Reserve System, 2018; U.S. Bureau of Labor Statistics, 2018).

The main research question will be answered through an analysis of a multivariate regression model on time-series data. At the start of this study only a one-directional relationship between mortgage defaults and the value of mortgage-backed securities was expected, namely an influence of mortgage defaults on the value of mortgage-backed securities. However, after further research, see literature section, these two variables both seem to have an impact on each other, therefore two regressions will be run with both variables as the dependent variable and main explanatory variable. The first explanatory variable is the value of mortgage-backed securities, whereas the other explanatory variable is mortgage defaults. The regressions are all computed with robust standard errors, to correct for heteroscedasticity. As a preferred specification, a VAR-model will also be used to investigate the relationship between mortgage defaults and mortgage-backed securities. A VAR-model takes the simultaneous causality of these two endogenous variables into account, a joint estimation. Since a VAR-model is the better model to test the relationship between these two endogenous variables, this study will use results from the VAR-model.

The unconditional relationship between the value of mortgage-backed securities and mortgage defaults is measured by the correlation between these two variables, which is -0.8658. This is logical when the value of mortgage-backed securities is the dependent variable and mortgage defaults the explanatory variable, because it follows the aforementioned economic intuition. However, it is less logical when mortgage defaults is the dependent variable and the value of mortgage-backed securities the explanatory variable, because that would suggest that an increase in the value of mortgage-backed securities results in lower mortgage defaults, whereas in the financial crisis it was a positive relation. When investigating the conditional relationship, the results of the first part of the VAR-model show that mortgage defaults lagged with one month have an effect of -0.0087388 on the value of mortgage-backed securities. This effect is not significant at any level. This means that a 1%-change in mortgage defaults lagged with one month result in a -0.0087388% change in the value of mortgage backed-securities. The results of the second part of the VAR-model show that the value of mortgage-backed securities lagged with one month have an effect of -1.49069 on mortgage defaults. This effect is also not significant at any level. This means that a 1%-change in the value of mortgage-backed securities lagged with one month result in a -1.49069% change in mortgage defaults. The implications of the results of the first part of the VAR-model are that it seems that mortgage default have zero effect on the value of mortgage-backed securities. This was also the case in

(7)

years before the financial crisis, which is mentioned above, where mortgage defaults were firstly not incorporated in the value of mortgage-backed securities and this led to disastrous circumstances. This means that investors and regulators should be aware of this and look into this, because it could mean that mortgage-backed securities are, still, not valued accurately. The implications of the results of the second part of the VAR-model are that the new regulation seems to have had the desired effect, because the value of mortgage-backed securities have no longer a positive effect on mortgage defaults. It is now a negative effect, which is insignificant, but if it were significant it implicates that if the value of mortgage-backed securities goes down, mortgage defaults go up. Based on the results of the first part of the VAR-model mortgage defaults seem to have no effect on the value of mortgage-backed securities, but mortgage defaults should have an effect when one follows economic intuition. This could lead, again, to the aforementioned circumstances in the financial crisis where the value of mortgage-backed securities suddenly had a large decrease in value because mortgage defaults where incorporated in the value after some time. In other words, a decrease in the value of mortgage-backed securities could set off a chain of reactions which suddenly leads to a rapid decline in value of mortgage-backed securities. Investors and regulators should be very aware of this.

The remainder of this thesis is organized as follows. Chapter 2 continues with a literature review that is included to describe important components of this study. In chapter 3, the data will be discussed. Following that, the econometric analysis is elucidated, which is in chapter 4. Regressions, hypotheses and variables will be specified. In chapter 5, the results will be disclosed and dissertated. The last chapter includes the conclusion where the main research question will be answered in full and limitations of this study will be discussed.

(8)

2. Literature Review

In this section, the two most important elements of this thesis, mortgage-backed securities and mortgage defaults will be described. Their features will be elaborated, and also the relationship between these variables will be discussed.

2.1 Description of mortgage-backed securities

The mortgage-backed security is a piece of the multibillion dollar market that consist of asset-backed securities. These securities can be asset-backed by various assets, such as cars, credit cards and also mortgages. Securitization is the process of forming assets that were characteristically hard to sell to a more liquid security. In this study, mortgage-backed securities will be discussed, a type of security that is backed by a mortgage or a group of mortgages pooled together. The U.S. secondary mortgage market is the largest fixed-income market in the world, because almost 60% of mortgage debt in the U.S. is repackaged into mortgage-backed securities (Chomsisengphet & Pennington-Cross, 2006). Essentially, when an investor is investing in a type of mortgage-backed security you are investing in people paying off their mortgages. A mortgage-backed security is created when a financial institution purchases multiple residential home mortgages, usually these mortgages have the same characteristics such as maturity and interest rates. The financial institution then pools these loans together and a bond is issued against this pool. The mortgage-backed security will now be sold to investors. Every month that the mortgages are paid, this will be passed through to the financial institution which created the mortgage-backed security and then the financial institution passes it through to the bondholders according to the proportionate share they invested in the mortgage-backed security (Fabozzi, 2016).

2.2 Risks of investing in mortgage-backed securities

These mortgage-backed securities are accompanied with a rating, which are determined by a certified credit rating agency. The rating gives insight on how much risk an investor in mortgage-backed securities is exposed to. When investing in mortgage-backed securities you have to take two types of risks into account. The first one is the risk that a mortgagor defaults, also known as credit risk. Some financial institutions have a government guarantee of payments, that even if the mortgagor defaults, your payments are protected. If these payments are not protected, an investor faces credit risk. To mitigate this risk, these loans are sometimes bundled into tranches, with tranches for more risk-averse investors and riskier investors (Lucas, Goodman & Fabozzi, 2006). In this study, we will be looking at the kind of mortgage-backed

(9)

securities that are not insured by the government. The second kind of risk is prepayment risk, which basically means that if a mortgagor refinances, the investor is paid back a part or the whole principal and then loses out on future interest payments. Prepayment by the mortgagor can happen for different kind of reasons, economic reasons such as a lower interest rate or a significant rise in their house which gives them reasons to sell, or non-economic reasons such as problems with the family or the environment (Kariya & Kobayashi, 2000).

2.3 Valuation of mortgage-backed securities

Over the years multiple models have been developed to value mortgage-backed securities. Fabozzi, Kalotay and Yang (2004) mention that there are two ways to come up with the value of a security that has a fixed income but the cash flow is interest-rate sensitive. The first one is to construct an econometric model based on historic data to predict future cash flows. This approach is rarely used nowadays because there have been limited accurate predictions of the future cash flows. The other model, more commonly used, is an arbitrage-free model which has been fitted to the market price of other, similar liquid instruments that have a credit that is very comparable to the security that is being priced. The last model is being used for almost all securities that are interest-rate sensitive, except for the mortgage-backed security that mostly still is valued using econometric models. This last approach uses statistical analysis of historical prepayment data which is used as input for this econometric model. The problem with this approach is that it does not account for microeconomic elements that are essential for the mortgage origination market. Even though that analysts persistently try to improve their prepayment models, they structurally will come up short in the microeconomic components of the mortgage market. There have been presented many option-theoretic models in valuing mortgage-backed securities, however most of them have not been successful in constructing the correct model, and virtually none have been implemented in the mortgage market (Fabozzi, Kalotay & Yang, 2004). This shows that in the years prior to the financial crisis there were still major issues with the valuations of mortgage-backed securities.

2.4 Mortgage defaults

There are different interpretations of mortgage defaults, this study will be discussing the type of mortgage default in which a consumer is said to be in default when a consumer is 90 days or more delinquent or the lender deems the remainder of the loan to be uncollectible, and takes repossession of the collateral that the loan is based on, in this study that are residential homes (S&P/First Experian Mortgage Index, 2018). Brueggeman & Fisher (2015) state that normally

(10)

lenders view foreclosing as a last option in the case that a consumer goes into default and they prefer to retrieve a part of the principal through other ways rather than foreclosing. However, when a foreclosure is inevitable, or even when lenders choose to foreclose, they have a strong incentive to foreclose quickly (Ling & Archer, 2017). Foreclosures or defaults are preferably avoided by lenders, particularly when the foreclosure is being contested, for multiple reasons. The first reason being that residential homes decrease significantly in value if the consumer is in default. For example, Campbell, Giglio and Pathak (2009) found that a property of a consumer in default, has a decrease in value of almost 28%. On top of that, in a foreclosure sale the home usually is being sold at a discount to speed up the process of liquidating the asset. There are better ways to solve the problem of a mortgage default than the aforementioned means. The first way that is a better solution for both parties is the short sale, which means that the owner of the home finds a new buyer for the property and this purchase price is usually higher than the price on an auction. Also, the owner of the home could also agree to something that is called a “friendly disclosure”. A friendly disclosure basically means that the owner of the home complies with the foreclosure, and agrees to not contest. The benefit for the lender is that it is less time consuming. The lender also normally agrees to waive their right to a deficiency judgment, which implies that if the sale of the home is insufficient, the lender has no right to claim the rest of the funds (Ghent & Kudlyak, 2010). Ghent & Kudlyak (2010) also state that this right to a deficiency judgment is an important factor for borrowers in deciding whether to default and if so how to default. Some borrowers choose to go into default if exercising this option means that their wealth will increase, even though these borrowers do not encounter any changes in their income or their mortgage payments. This kind of financial cleverness also has negative sides, for example that after going into default borrowers do not have the option to re-finance anymore. Moreover, lenders try to discourage this cleverness by allowing a recourse to other assets than the mortgaged home (Deng, Quigley & Van Order, 2000).

2.5 Relationship between mortgage-backed securities and mortgage defaults

As stated before, the relationship between mortgage-backed securities and mortgage defaults have played an important role in the recent financial crisis. Keys et al. (2010) mentions that the sudden rise in defaults during the financial crisis period are highly associated with the looser lending standards that are inherent to the “originate-to-distribute” mortgage loan model, because this this model the lender has less incentive to screen and monitor borrowers, since the loans are almost sold immediately again to securitizers. Arentsen et al. (2014) also argues that

(11)

there is not only a one-directional relationship between mortgage-backed securities and mortgage defaults, but that because there was such a strong demand for mortgage-backed securities, that resulted in more and more loans to borrowers that were by definition riskier, which ultimately led to a higher default rate of mortgagors. Petersen and Rajan (2002) advocate this as well, by stating that due to securitization a distance is created between the originator of the loan and the holder of the loan, who carries all the risk. Therefore, the lender has less incentive to screen and monitor borrowers thoroughly. However, supporters of securitization state that regulations or fear of a damaged reputation can obviate moral hazard by lenders (Keys et al 2010). During the period before the financial crisis, 80 to 90 percent of mortgage-backed securities that were non-prime, were accredited with the highest possible rating, a triple-A rating. In spite of that, many of these securities were in danger of default or had already done so. Many people suggested that if these subprime mortgage-backed securities had been accredited with a lower rating before the financial crisis took place, the effect of this could have been mitigated (Ashcraft et al., 2011). Besides the weak screening of borrowers on the part of lenders due to securitization, securitization also increased the complexity of financial products, which made it for borrows more difficult to assess risks that were incorporated in these products (Peicuti, 2013).

(12)

3. Data

In this section, data that is used in this study will be discussed and defined. Particularly where the data is acquired, what the data consists of, an elaboration about the descriptive statistics and why these data are used. The explanatory variables will be discussed first and then the control variables.

3.1 Explanatory variables

The first explanatory variable of this study is the value of mortgage-backed securities. This paragraph will go into detail about this variable. The values of mortgage-backed securities are collected from the database of Economic Research Federal Reserve Bank of St. Louis and the source is Board of Governors of the Federal Reserve System. The dataset consists of monthly data containing the value of all mortgage-backed securities held by commercial banks. The data are seasonally adjusted (Board of Governors of the Federal Reserve System, 2018). The reason that only mortgage-backed securities held by commercial banks are included and not also the mortgage-backed securities held by the Federal Reserve is because those held by commercial banks are not guaranteed by a government institution, such as Fannie Mae, Freddie Mac or Ginnie Mae. Those guaranteed mortgage-backed securities need to be excluded from the regression, considering the different and lower type of risk these securities consist of. The first data point of this database is at July 2009, and is updated weekly. To use this dataset, the weekly data needed to be computed in monthly data due to the lack of availability of weekly data of other variables. The aggregation method that is used is averaging. As one can see in Table 1, the mean is 1399 in billions of U.S. dollars and the standard deviation is 225, also in billions of U.S. dollars. Figure 1 shows a steady rise of the value of mortgage-backed securities.

Figure 1: Value of MBS held by commercial banks (in billions of U.S. dollars)

0,0000 200,0000 400,0000 600,0000 800,0000 1000,0000 1200,0000 1400,0000 1600,0000 1800,0000 2000,0000 2010 -01 -01 2010 -05 -01 2010 -09 -01 2011 -01 -01 2011 -05 -01 2011 -09 -01 2012 -01 -01 2012 -05 -01 2012 -09 -01 2013 -01 -01 2013 -05 -01 2013 -09 -01 2014 -01 -01 2014 -05 -01 2014 -09 -01 2015 -01 -01 2015 -05 -01 2015 -09 -01 2016 -01 -01 2016 -05 -01 2016 -09 -01 2017 -01 -01 2017 -05 -01 2017 -09 -01 2018 -01 -01

(13)

The second explanatory variable of this study is mortgage defaults which will now be further explained. The data of mortgage defaults are derived from the database of S&P Dow Jones Indices. In this index, mortgage defaults are defined as consumers whom are delinquent for 90 days or more, or when the lender takes repossession of the collateral, in this study, residential homes. This dataset contains the percentages of consumers going into mortgage default at the first of every month. Consequently, this dataset consists of monthly observations. The data goes as far back as to March 2008 (S&P/Experian First Mortgage Default Index, 2018). The reason for using this index is that it provides a clear overview of how many consumers go into mortgage default every month, which is what is needed in this study. Figure 2 shows a sharp decline of mortgage defaults between January 2010 and July 2012, and a steady decline thereafter.

Figure 2: Proportion of consumers that go into mortgage default

Table 1: Descriptive statistics

Variable #Observations Mean Std. Dev. Min Max Value of MBS (1) 98 19753.63 874.338 18263 21244 Mortgage Defaults (3) 98 1.457475 28.72281 999.8824 1816.892 Inflation Rate (4) 98 233.5445 8.605682 217.199 249.619 Average Housing Prices (2) 98 322.4323 42.57149 250.0 402.9 Asset-backed CP (1) 98 292.6941 67.20281 213.1042 489.3452 Interest Rate (3) 98 4.062061 0.430219 3.345 5.098 Note: (1) are in billions of U.S. Dollars, (2) is in thousands of U.S. Dollars, (3) is in percentages, (4) is an index 0,00% 0,50% 1,00% 1,50% 2,00% 2,50% 3,00% 3,50% 4,00% 4,50% 5,00% Ja n-2010 Ap r-2010 Ju l-2010 Oc t-2010 Ja n-2011 Ap r-2011 Ju l-2011 Oc t-2011 Ja n-2012 Ap r-2012 Ju l-2012 Oc t-2012 Ja n-2013 Ap r-2013 Ju l-2013 Oc t-2013 Ja n-2014 Ap r-2014 Ju l-2014 Oc t-2014 Ja n-2015 Ap r-2015 Ju l-2015 Oc t-2015 Ja n-2016 Ap r-2016 Ju l-2016 Oc t-2016 Ja n-2017 Ap r-2017 Ju l-2017 Oc t-2017 Ja n-2018

(14)

3.2 Control variables

In this paragraph, there will be a further elaboration on control variables. Firstly, the control variable housing prices will be discussed. Housing prices are derived from the United States Census Bureau. It contains monthly data of the average price of homes sold in the United States (United States Census Bureau, 2018). The reason for including housing prices in the multivariate regression model is that when housing prices are rising, the amount that consumers can mortgage increases too, not only when purchasing new homes, consumers could also possibly refinance. Therefore, the whole mortgage market increases in value, and consequently also the value of mortgage-backed securities. From Table 2 can also be retrieved that there is a strong effect of housing prices on the value of mortgage-backed securities, as the correlation is 0.8875, which indicates that there is indeed a strong correlation between the two.

The second control variable that will be discussed is the interest rate of mortgages. This dataset is also obtained from the database of Economic Research of the Federal Reserve Bank of St. Louis. The source is Freddie Mac. The dataset consists of weekly 30-year fixed average mortgage rates, and are not seasonally adjusted (Freddie Mac, 2018). To use this dataset, the weekly data needed to be computed in monthly data due to the lack of availability of weekly data of other variables. The aggregation method that is used is averaging. The mortgage rate has been fluctuating between 5.10% and 3.45%, which are the maximum and minimum variable as can be seen in Table 1. The reason for including mortgage interest rates as a control variable in the regression model is because when mortgage interest rates are rising, the demand for mortgages will decrease and therefore there are less mortgages to be repackaged into mortgage-backed securities which could impact the value of mortgage-mortgage-backed securities. According to Table 2, interest rates do have a negative effect on the value of mortgage-backed securities, shown by the coefficient of -.5440, which indicates a mild correlation.

The third control variable that will be discussed are the value of asset-backed commercial papers. This dataset is also obtained from the database of Economic Research of the Federal Reserve Bank of St. Louis. The source is the Board of Governors of the Federal Reserve System. The dataset consists of the value commercial paper that is outstanding and asset-backed. The data are seasonally adjusted (Board of Governors of the Federal Reserve System, 2018). The value of asset-backed commercial paper is included in the regression model as a control variable because there might be a trend in asset-backed financial products that have no relation to mortgage defaults whatsoever, which could cause for a bias in results if this variable is not included in the model. However, asset-backed commercial paper has had a decline in value from January 2010 to around January 2015, and then stabilized, whereas the

(15)

value of mortgage-backed securities keeps rising. Also, according to Table 2, asset-backed commercial papers have a negative effect on the value of mortgage-backed securities, indicated by a coefficient of -0.8082, which shows that there might not be a positive relation between these two variables.

The fourth and last control variable that will be discussed is the inflation rate. This dataset is, once again, obtained from the database of Economic Research of the Federal Reserve Bank of St. Louis. The source is the U.S. Bureau of Labor Statistics. The dataset consists of monthly consumer price indices for all items. The index of 100 is between 1982 and 1984. The data are seasonally adjusted (U.S. Bureau of Labor Statistics, 2018). Economic Research of Federal Reserve Bank of St. Louis defines the consumer price index as follows:“a measure of the average monthly change in the price for goods and services paid by urban consumers between any two-time periods.” The inflation rate has been constantly rising since January 2010 up until now. The implication of this is that due to this constant rise, it is wiser for consumers to purchase a product today rather than later. Therefore, the demand for houses will rise and, consequently, also the demand for mortgages. This could affect the value of mortgage-backed securities and hence why this control variable is included in the regression model. This reasoning is supported with the coefficient of 0.9670 from Table 2, which indicates that there is a positive effect of the inflation rate on the value of mortgage-backed securities. It also shows a very strong correlation between these two variables.

From Table 2 can be derived that some independent variables show a very strong correlation, which could affect the regression results due to multicollinearity. Grewal, Cote and Baumgartner (2004) argue that it is better in order to obtain accurate estimation results high correlations between independent variables should be avoided. However, Grewel, Cote and Baumgartner (2004) also state that due to practical considerations researchers may ignore multicollinearity. Therefore, these variables can be implemented in the regression model.

Table 2: Correlation matrix

MBS MD ITR AHP ABCP IFR

MBS 1.0000 MD -0.8658 1.0000 ITR -0.5540 0.6427 1.0000 AHP 0.8875 -0.7991 -0.3915 1.0000 ABCP -0.8082 0.9215 0.5332 -0.8237 1.0000 IFR 0.9670 -0.9183 -0.5036 0.9070 -0.8863 1.0000

(16)

4. Methodology

In this section, the methods of this study will be discussed and the arguments for using these methods. At first the arguments for using a multivariate regression model will be provided, then an explanation for why a VAR-model and the corresponding impulse response functions and forecast error variance decomposition were added. Finally, hypotheses of this study are stated.

4.1 Multivariate regression model

To estimate the effect of mortgage defaults on the value of mortgage-backed securities and vice versa, a regression analysis is needed. A regression will be run on both mortgage defaults and the value of mortgage-backed securities as the dependent variable as well as the main explanatory variable. These two variables are assumed to be endogenous, whereas the control variables included in the regression are assumed to be exogenous. Because of the multi-causality assumption, consider the following models:

∆𝑙𝑛 𝑀𝐵𝑆' = 𝛽+∆(𝑀𝐷)'+ 𝛽0∆(𝐼𝑇𝑅)'+ 𝛽4∆(𝐼𝐹𝑅)'+ 𝛽6∆𝑙𝑛(𝐴𝐵𝐶𝑃)'+ 𝛽:∆𝑙𝑛(𝐴𝐻𝑃)' + 𝜀_' (1) ∆ 𝑀𝐷_' = 𝛽₊∆𝑙𝑛(𝑀𝐵𝑆)_'+ 𝛽₀∆(𝐼𝑇𝑅)_'+ 𝛽₄∆(𝐼𝐹𝑅)_'+ 𝛽₆∆𝑙𝑛(𝐴𝐵𝐶𝑃)_'+ 𝛽_:∆𝑙𝑛(𝐴𝐻𝑃)_' + 𝜀_' (2)

These regression models will be used to test the hypotheses. The mathematical hypotheses are: H0: 𝛽₊∆(𝑀𝐷)_'=+ & 𝛽₊∆𝑙𝑛(𝑀𝐵𝑆)_'=+ = 0 and H1: 𝛽₊∆(𝑀𝐷)_'=+ & 𝛽₊𝑙𝑛∆(𝑀𝐵𝑆)_'=+ ≠ 0.

Table 3: List of variables

MBS_D MBS_D=+

MD_D

Value of mortgage-backed securities at time t

Value of mortgage-backed securities lagged with one month Mortgage default rate at time t

MD_D=+ Mortgage default rate lagged with one month ITR_D

IFR_D ABCP_D

Interest rate at time t Inflation rate at time t

Value of asset-backed commercial paper at time t

AHPD Average prices of houses at time t

Due to the non-stationary characteristic of several variables, the data needed to be tested whether it is unit root or stationary. A test that can be used for this, is the Dickey Fuller test. When executing a Dickey Fuller test, the null hypothesis is always that the variable is a unit root (Hamilton, 1994). If these tests are not executed, it could lead to spurious regressions results. That is because if variables that are not-stationary are regressed on each other it leads

(17)

to inaccurate results. When looking at Table 4 (1), one can see that the null hypothesis of this test can only be rejected of the variables mortgage defaults and asset-backed commercial paper. This means that all the other variables, which do not have significant values, are variables with a unit root and therefore unable to implement in the regression model to prevent spurious regression results. To eliminate unit root in the variables, the first differences transformation method is needed (Becketti, 2013). According to Table 4 (2), one can see that the null hypothesis of this test can be rejected with all variables. This means that all the variables have significant values, and therefore are now stationary variables. Due to these adjustments to the data, these variables can now be accurately regressed on each other. One could argue that when it comes to the variables mortgage defaults and asset-backed commercial paper due to their significance in both tests, there are no preferred data. The data that contained absolute values, the value of mortgage-backed securities, average housing prices and asset-backed commercial paper, were first differenced by taking the logarithmic scale of the differences. This is computed according to the following equation:

∆ ln 𝑀𝐵𝑆' = ln 𝑀𝐵𝑆' − ln 𝑀𝐵𝑆'=+

The data containing percentages, mortgage default rate, interest rate and the inflation rate, were first differenced by just subtracting. This is computed according to the following equation:

∆ 𝑀𝐷_' = 𝑀𝐷_' − (𝑀𝐷_'=+)

Due to these adjustments to the data, all the variables can now be used in the model to obtain accurate regression results.

Table 4: Unit root tests (Dickey-Fuller)

(1) (2)

Level data First-differenced data

Value of MBS -0.298 -7.267*** (0.9259) (0.0000) Mortgage Defaults -7.238*** -5.387*** (0.0000) (0.0000) Interest Rates -2,339 -7.045*** (0.1596) (0.0000) Housing Prices -1.718 -15.420*** (0.4420) (0.0000) Inflation Rates -0.334 -6.492*** Asset-backed CP (0.9205) -3.027** (0.0000) -7.506*** (0.0000) (0.0000)

(18)

The argument for two regressions on both the explanatory variables instead of just one regression is due to simultaneity. Simultaneous causality means that two variables within the regression model are assumed to be endogenous, in this case the value of mortgage-backed securities and mortgage defaults (Theil & Zellner, 1962). Simultaneous causality leads to a biased & inconsistent OLS estimate. According to Stock & Watson (2015) this can be avoided by running two regressions instead of only one. The reasons for lagging the effects of mortgage-backed securities and mortgage defaults is that if these are not lagged, one is more looking into correlation rather than into causality, which is what this study is aiming for. This study uses the method of Ordinary Least Squares. A multivariate regression model is used, because the main explanatory variables, the value of mortgage-backed securities and mortgage defaults, are also influenced by other macroeconomic factors. If these factors are not included in the model, this could cause omitted variable bias. This means that the main explanatory variable is correlated with a variable that is not included in the model, yet has an effect on the dependent variable. This would produce biased results. To observe how much of the variance of the dependent variable is explained by the model, there will be looked at the 𝑅0_{of the model and also the}

adjusted-𝑅0_{of the model. The reason for also looking at the adjusted-𝑅}0_{, is that when adding}

variables to a regression model 𝑅0_{automatically increases, even if these variables are not}

related to the dependent variable whatsoever. The adjusted-𝑅0_{corrects this increase. Lastly, to}

test the significance of the regression model, the t-test will be used to test individual variables, whereas the F-test will be used to test the entire model.

4.2 Vector-autoregressive model

To investigate the relationship between two endogenous variables, it is of vital importance to use a VAR-model. That is because a VAR-model takes the simultaneous causality of both variables into consideration. When using a VAR-model, the lagged value of the endogenous variables will be regressed on each other in a simultaneous equation, with both variables as the dependent variable (Sims, 1980). In this study, these dependent variables are the value of mortgage-backed securities and mortgage defaults. There are conditions restricted to a VAR-model that need to be fulfilled before accurate regression results can be retrieved. The first requirement is that the data needs to be stationary and the second requirement is that the lag should have a logical length. As mentioned before, the non-stationary characteristics of the data were removed by first-differencing the data. The second requirement can be fulfilled by performing a selection-order criteria test, which indicates the sufficient number of lags. Pegkas & Tsamadias (2014) state the Akaike information criterion (AIC) is the most commonly used

(19)

test to identify the number of lags. Results of Table 5 show that the AIC identifies one lag as an optimal lag length. Therefore, one lag will be implemented in the VAR-model.

Table 5: selection-order criteria

lag FPE AIC HQIC SBIC

0 0.00000041 -9.02009 -8.99823 -8.96597

1 0.00000030* -9.35953* -9.29396* -9.19719*

2 0.00000031 -9.31942 -9.21013 -9.04885

3 0.00000030 -9.35566 -9.20266 -8.97687

4 0.00000031 -9.30233 -9.10552 -8.81522

Notes: asterisk (*) identifies optimal number of lags

To investigate the long-term relationship between mortgage-backed securities and mortgage defaults, impulse response functions will be illustrated. Besides that, also a forecast error variance composition will be computed. Before the long-term relationship can be tested, it is recommended to perform an Eigenvalue test. This test measures the stability of the VAR-model. The moduli of the eigenvalues of the dynamic matrix must lie within the unit circle. From Table 6 can be retrieved that the eigenvalues lie inside the unit circle, therefore the VAR-model satisfies the stability condition. Thus, the long-term relationship between mortgage-backed securities and mortgage defaults can be accurately tested with impulse response functions and a forecast error variance decomposition.

Table 6: Eigenvalue stability condition

Eigenvalue Modulus

0.6448015 0.644802

0.2560133 0.256013

In this study, the variance decomposition is constructed from the VAR-model. The forecast error variance decomposition computes the portion of the forecast error variance of an endogenous variable that can be contributed to orthogonalized shocks to itself or to a different endogenous variable (Masih and Masih, 1997). In other words, it shows the contribution an endogenous variable has on the forecasting dependent variable. Many papers argue that a variance decomposition is one of the most relevant methods to identify a causal relationship between two variables established on economic theory (Hye, 2012).

(20)

4.3 Hypotheses

After extensive literature research on the relationship between mortgage defaults and mortgage-backed securities it has become very clear that there is indeed a relationship between these two variables. During and before the financial crisis, the demand for mortgage-backed securities resulted in higher mortgage defaults, whereas these mortgage defaults caused at the beginning of the financial crisis little to no impact on the value of mortgage-backed securities. These variables are not only affected by each other, they are also affected by variables such as interest rates and housing prices. Earlier research mainly focused on the period during and before the financial crisis, whereas this thesis will be focusing on the period after the financial crisis. That is why this study contribute to existing literature. Due to the aftermath of the financial crisis, the government imposed many regulations on financial institutions. Therefore, the first hypothesis is that ‘there is a negative effect of mortgage defaults on the value of backed securities’ and the second hypothesis is that ‘there is no effect of the value of mortgage-backed securities on mortgage defaults’.

(21)

5. Results

In this section, the results of the multivariate regression model and VAR-model will be elaborated. First, the two outcomes of the multivariate regression models will be explained. Then, an interpretation of the results of the VAR-model with the corresponding impulse response functions and the forecast error variance decomposition will be shown. At last implications of these results will be discussed.

5.1 Analysis of the multivariate regression model and the vector-autoregressive model

To check for robustness, a multivariate regression model was used with the value of mortgage-backed securities and mortgage defaults and vice versa. These results can be retrieved from Table 7.

Table 7: Regression analysis

(1) Δln(Value of MBS(t)) (2) ΔMortgage defaults (t) ΔMortgage defaults (t-1) -0.0087388 0.6077676*** Δln(Value of MBS(t-1)) (0.007749) 0.2930472*** (0.0981879) -1.49069 (0.1048064) (1.17139) Δlnterest Rate (t) -0.0194245*** -0.1218576* (0.0054062) (0.0622395)

Δln (Average housing prices (t)) 0.0506056*** 0.1751529 Δln (Asset-backed CP (t)) (0.0161615) -0.0123781 (0.2014669) -0.01180787 (0.127363) (1.364688) ΔInflation rate (t) 0.0042434 -0.000262 (0.003505) (0.00541) Constant 0.0029207*** -0.0117868 (0.0009945) (0.0083935) Number of observations 98 98 R-squared 0.2883 0.3777 Adjusted R-squared 0.2414 0.3367 F-statistic 6.95 7.49

Notes: robust standard errors in parentheses; *p<0.10, **p<0.05, ***p<0.01

Then, a VAR-model is used. These results can be retrieved from Table 8. As these results are almost identical, and a VAR-model is the preferred specification, in this section only results from the VAR-model are discussed. Except for the F-statistic, which is obtained from the

(22)

multivariate regression model. First, the model will be analyzed as a whole, where the value of mortgage-backed securities is the dependent variable. The results are shown in Table 8 (1).

Table 8: Vector-autoregressive analysis

Notes: standard errors in parentheses; *p<0.10, **p<0.05, ***p<0.01

The R0 is 0.2883 and the adjusted R0 is 0.2414. As mentioned before, the adjusted R0 will be

used. This number of 0.2883 means that 28.83% of the variance in the value of mortgage-backed securities is explained by the variables in the model. This implicates that there might still be other variables not included in the model relevant to the value of mortgage-backed securities. According to Table 7, the F-statistic (6, 91) = 6.95. This number is significant at a 1%-significance level, and thus the null hypothesis can be rejected. Therefore, at least one of the variables included in the model has a significant effect on the value of mortgage-backed securities.

Further, the individual variables will be analyzed. According to Table 8 (1), The effect of the momentum, which is measured by the effect of the values of mortgage-backed securities lagged with one month on the current value of mortgage-backed securities, has a coefficient of 0.2930. This is significant at a 1%-significance level. This means that a 1%-change in the lagged values of backed securities result in a 0.2930% change in current values of mortgage-backed securities. Thus, showing that lagged values of mortgage-mortgage-backed securities have an influence on the current value of mortgage-backed securities. Table 8 (1) shows that mortgage Δln(Value of MBS(t)) (1) ΔMortgage defaults (t) (2)

Δln(Value of MBS(t-1)) 0.2930*** Δln(Value of MBS (t-1)) -1.49069

(0.0867094) (0.942025)

ΔMortgage defaults (t-1) -0.0087388 ΔMortgage defaults (t-1) 0.6078***

(0.0075546) (0.082074)

Δlnterest Rate (t) -0.0194*** Δlnterest Rate (t) -0.1219*

(0.0051296) (0.055729)

Δln (Average housing prices (t)) 0.0506*** Δln (Average housing prices (t)) 0.1751529 Δln (Asset-backed CP (t)) (0.0147299) -0.012378 Δln (Asset-backed CP (t)) (0.160028) -0.11808 (0.15758) (0.171197)

ΔInflation rate (t) 0.00042434 ΔInflation rate (t) 0.0340172

(0.0034707) (0.037706)

Constant 0.00292*** Constant -0.011787

(0.0010165) (0.011044)

Number of observations 98 Number of observations 98

R-squared Adjusted R-squared 0.2883 0.2414 R-squared Adjusted R-squared 0.3777 0.3367

(23)

defaults lagged with one month have an effect of -0.0087388 on the current value of mortgage-backed securities. Which means that a 1%-change in mortgage defaults lagged with one month leads to a -0.0087388% change in the value of mortgage backed-securities. This follows economic intuition, because one would say that if mortgage defaults are rising, investing in mortgage-backed securities is riskier and therefore the value decreases. This effect is, however, not significant and also very low. This insignificant effect could be due to the number of low observations or omitted variable bias. From Table 8 (1) can be retrieved that interest rates have a significant negative effect of 0.0194 on the value of mortgage-backed securities at a 1%-significance level. Meaning that a 1%-change in interest rates results in a -0.0194% change in the value of mortgage-backed securities. This is perfectly explainable. When interest rates rise, consumers have to pay more on their mortgage and therefore the credit risk increases. Investors want to be compensated for that risk. Hence, the lower value of mortgage-backed securities when interest rates rise. When looking at Table 8 (1), average housing prices have a small but significant positive effect of 0.0506 on the value of mortgage-backed securities. The effect is significant at a 1%-significance level. This means that a 1%-change in average housing prices will lead to a 0.0506% change in the value of mortgage-backed securities. This effect falls within the range of what could be expected. One would say that if average housing prices rise, the amount of people can mortgage rises too. Therefore, the whole mortgage market would increase. Table 8 (1) shows that asset-backed commercial paper has a small negative effect of 0.012378 on the value of mortgage-backed securities. This means that a 1%-change in asset-backed commercial paper results in a -0.012378% change in the value of mortgage-asset-backed securities. This effect is not significant. As mentioned before, there seems to be no particular trend in the whole asset-backed securities market. From Table 8 (1) follows that the inflation rate has a very small effect of 0.00042434 on the value of mortgage-backed securities. Meaning that if inflation rates change with 1%, it leads to a 0.00042434% change in the value of mortgage-backed securities. This seems logical. When inflation rate rises, it is wiser for consumers to purchase securities now rather than later. This would increase the demand for mortgage-backed securities and therefore also the value of mortgage-backed securities, on the condition that supply does not change. On the other hand, due to the very low effect one cannot really conclude much. The constant has a positive coefficient of 0.00292, significant at a 1%-significance level.

Now the second part of the regression will be explained, where the dependent variable, value of mortgage-backed securities, and explanatory variable, mortgage defaults, from the first regression are switched. The results can be retrieved from Table 8 (2). Again, the model will

(24)

first be analyzed as a whole. The R₂ is 0.3777 and the adjusted R₂ is 0.3367, meaning that 33.67% of the variance in mortgage defaults is explained by variables in the model. According to Table 7, the F-statistic (6, 91) = 7.49, which is significant at the 1%-significance level. Therefore, the null hypothesis is rejected and that implicates that at least one of the variables in the model have a significant effect on mortgage defaults.

Next, the individual variables will be analyzed. From Table 8 (2) follows that the effect of the momentum, which is measured by the effect of the lagged value of mortgage defaults on the current rate of mortgage defaults, has a coefficient of 0.6078. This means that a 1%-change in mortgage defaults lagged with one month result in a 0.6078% change in current levels of mortgage defaults. It can be stated that both the values of mortgage-backed securities and mortgage defaults are directly affected by changes in the past of those variables. This shows that there is indeed a lagging effect in both variables. According to Table 8 (2) the effect of the lagged value of mortgage-backed securities on mortgage defaults, is -1.4069. This means that if the lagged value of mortgage-backed securities change with 1%, this results in a -1.4069% change in mortgage defaults. This effect was different during the financial crisis, because then mortgage-backed securities had a positive effect on mortgage defaults. An explanation could be that an increase of the value in mortgage-backed securities is due to economic welfare, and therefore there are less mortgage defaults. However, this effect is not significant and thus definitive conclusions cannot be drawn. The other variable that has a significant effect on mortgage defaults is the interest rate. As can be seen in Table 8 (2), the coefficient of interest rate is -0.1219, which means that a 1%-change in interest rates result in a -0.1219% change in mortgage defaults. This effect is significant at a 10%-significance level. It seems illogical, because when interest rates rise, it is more expensive for consumers and therefore one would say mortgage defaults increase. Omitted variable bias could be an explanation for this. Multiple variables do not have a significant effect on the value of mortgage-backed securities, mortgage defaults, or both. The insignificant effects could be due to the low number of observations. The coefficients of asset-backed commercial paper and inflation rate, are in both parts of the regression very low. So, even if these values were significant, they seem to have a very little effect on both the value of mortgage-backed securities and mortgage defaults. In neither the first or second part of the regression, did the explanatory variable have a significant value. Only some control variables and the lagged effects of the same variable were significant in the regression. The insignificant values of the explanatory variables could be due to multicollinearity, since from Table 2 follows that mortgage defaults and asset-backed

(25)

commercial paper have a correlation of 0.9215, and the value of mortgage-backed securities and inflation rate have a correlation of 0.9670.

5.2 Analysis of the impulse response functions (IRF) and the forecast error variance decompositions (FEVD)

IRF’s are calculated in order to measure not only the short-run marginal effects but also the longer-term effects of the endogenous variables mortgage defaults and the value of mortgage-backed securities. In particular, orthogonalized IRF’s. To obtain these graphs, one must determine the most endogenous variable. When observing the regression results, mortgage defaults is determined as the most endogenous variable because it has a higher adjusted R₀, a higher F-statistic and mortgage defaults are more affected by the value of mortgage-backed securities then the other way around. The reason for using orthogonalized IRF’s is that it measures the isolated effect from a shock in one variable with respect to another, which is what this study is interested in.

The graphs are illustrated in Figure 3. The first graph, Figure 3 (1), shows a shock of one unit standard deviation in mortgage defaults and the effect of that shock on the value of mortgage-backed securities. The effect, which is positive, lasts for about a month, then it turns into a small negative effect before returning to zero after around four months. The effect is not significant. The second graph, is illustrated in Figure 3 (2). This shows a shock of one unit standard deviation in the value of mortgage-backed securities and the effect of that shock on mortgage defaults. This effect, which is negative, also lasts for approximately one month, before returning to zero. This is a longer lasting effect, since it returns at zero after around eight months. This effect is also not significant. The third graph, in Figure 3 (3), shows a shock of one unit standard deviation in in the value of mortgage-backed securities and the effect of that shock on the value of mortgage-backed securities. It is a positive effect which lasts about two months where it returns to zero. This effect is significant for about two months, since zero is not within the 95%-confidence interval at that time. The fourth and last graph, in Figure 3 (4), shows a shock of one unit standard deviation in mortgage defaults and the effect of that shock on mortgage defaults. This is a positive effect which lasts about six months, where the effect turns to zero. This is a significant effect for approximately four months, because zero is not within the 95%-confidence interval at that time.

(26)

Figure 3: Orthogonalized impulse response functions

Lastly, the FEVD’s will be analyzed and interpreted. In Table 8 (1) the forecast error variance decomposition of mortgage defaults is shown. The results reveal that the value of mortgage-backed securities does not have a large contribution in explaining mortgage defaults. The value of mortgage-backed securities only explains mortgage defaults with 1.33% in the short term and increases up to 2.89% percent in the longer term. This shows that the value of mortgage-backed securities does have a longer-term effect in explaining mortgage defaults, however this effect is very low. In Table 8 (2) the forecast error variance decomposition of the value of mortgage-backed securities is shown. These results reveal that mortgage defaults, also, do not have a large contribution in explaining mortgage defaults. Mortgage defaults only explain 4.11% in the short term of the value of mortgage-backed securities. This percentage fluctuates and rises up to 4.70%. This shows that mortgage defaults also have a longer-term effect in explaining the value of mortgage-backed securities, but again, the percentage is very low. Both mortgage defaults and the value of mortgage-backed securities are predominantly explained by itself.

(27)

Table 9 (1): Forecast error variance decomposition of MD(t) Periods MD (t) Standard error MBS (t) 1 1 0* 0 2 0.986671 0.016916 0.013329 3 0.978151 0.027326 0.021849 4 0.974126 0.03231 0.025874 5 0.972377 0.034579 0.027623 6 0.971638 0.0356 0.028362 7 0.971328 0.036066 0.028672 8 0.9712 0.036259 0.0288

Notes: asterisk (*) means approximately zero

Table 9 (2): Forecast error variance decomposition of MBS(t)

Periods MD (t) Standard error MBS (t) 1 0.04114 0.039292 0.95886 2 0.039129 0.035037 0.960871 3 0.042547 0.035589 0.957453 4 0.044937 0.036988 0.955063 5 0.046108 0.037869 0.953892 6 0.046626 0.038306 0.953374 7 0.046846 0.038507 0.953154 8 0.046938 0.038596 0.953062

5.3 Discussion of the implications of the results

On the basis of the VAR-model and the corresponding orthogonalized IRF’s and the FEVD’s, there can be concluded for the first hypothesis that there is not enough statistical evidence to state that mortgage defaults have a negative effect on the value of mortgage-backed securities. Thus, the first null hypothesis cannot be rejected. For the second hypothesis, however, it can be stated that there is not enough statistical evidence to conclude that the value of mortgage-backed securities has an effect on mortgage default, therefore the second null hypothesis can be rejected.

The results of the first part of the VAR-model implicate that that since mortgage defaults have no effect on the value of mortgage-backed securities. From economic intuition follows that if the underlying asset of the security becomes riskier, this should have a negative effect on the value of the security. Whereas these results imply that mortgage defaults have zero influence on the value of mortgage-backed securities. Therefore, investors and regulators should take notice of this, because that could mean that mortgage-backed securities are, still, not valued accurately. This could, again, lead to circumstances that are similar to the

(28)

circumstances before the financial crisis, where mortgage defaults also were not incorporated in the value of mortgage-backed securities.

The results of the second part of the VAR-model implicate that the reform of the financial system seems to have had the desired effect because the value of mortgage-backed securities no longer has a positive effect on mortgage defaults. There seems to be no effect at all, because it is insignificant. However, because this could be due to multicollinearity or low observations, the negative coefficient should still be taken into account. The negative relationship between the value of mortgage-backed securities and mortgage defaults implies that if the value of mortgage-backed securities decreases, mortgage defaults increase. As mentioned before, mortgage defaults seem to have no effect on the value of mortgage-backed securities. However, they should have an effect when one follows economic intuition. This could cause, again, situations that were similar to the ones in the financial crisis where the value of mortgage-backed securities suddenly had a sharp decline in value because mortgage defaults were incorporated in the value after some time. In other words, a decline in the value of mortgage-backed securities could set off a chain of reactions that causes an even greater decline in the value of mortgage-backed securities. Investors and regulators should be very aware of this.

(29)

6. Conclusion and discussion

This last section of this study will draw conclusions on the results that were found. Then, limitations of this study will be discussed and what can be improved in future research.

6.1 Conclusion and discussion

This study investigated the relationship between mortgage defaults and the value of mortgage-backed securities, with evidence from the United States and a time frame from the 1st of January 2010 till the 1st of March 2018. The main research question was: “What is the relationship between mortgage defaults and the value of mortgage-backed securities in the period after the financial crisis?”. Firstly, this study included a literature review to provide a description about mortgage defaults and mortgage-backed securities. The relationship between mortgage defaults and mortgage-backed securities was also investigated. Empirical studies showed that there was not only a relationship between mortgage defaults and the value of mortgage-backed securities, but also the other way around. This study expected a different relationship than the relationship that were found in and before the financial crisis, due to the reform of the financial system. This resulted in two hypotheses, with the first one being: ‘there is a negative effect of mortgage defaults on the value of mortgage-backed securities’ and the second one is that: ‘there is no effect of the value of mortgage-backed securities on mortgage defaults’. In order to answer these hypotheses, this study used a multivariate model and a VAR-model to analyze the relationship between mortgage defaults and mortgage-backed securities.

The results show that neither mortgage defaults or the value of mortgage-backed securities have a significant effect on each other. This could be due to the relative low number of observations, or multicollinearity. The results did show that mortgage defaults have a very small negative effect on the value of mortgage-backed securities. So even if it was a significant effect, mortgage defaults show little effect on the value of mortgage-backed securities. The value of mortgage-backed securities had a larger negative effect on mortgage defaults, which is different from the effect in the financial crisis, where it was positive. To further gauge the relationship between mortgage defaults and the value of mortgage-backed securities, orthogonalized impulse response functions (IRF) were calculated, as well as a forecast error variance composition (FEVD). The orthogonalized IRF’s also showed that mortgage defaults and the value of mortgage-backed securities had both an effect on each other, but neither were significant, which corroborates the results from the VAR-model. The FEVD’s showed that both mortgage defaults and the value of mortgage-backed securities explain very little of each other. This also supports the results of the VAR-model and the results from the orthogonalized IRF’s.

(30)

There can be concluded that for both variables not enough statistical evidence was found that mortgage defaults have a significant effect on mortgage-backed securities and vice versa.

Due to these results, the first null hypothesis cannot be rejected at any significance level. The second null hypothesis, however, can be rejected because no significant effect was found between the value of mortgage-backed securities and mortgage defaults. The implications of these results are that because mortgage defaults show no effect on the value mortgage-backed securities, investors and regulators should be aware of this because it could mean that mortgage-backed securities are, still, not valued accurately. The value of mortgage-mortgage-backed securities also has no significant effect on mortgage defaults, which implies that the new regulation after the financial crisis seem to have had the desired effect. However, because the insignificance could be due to multicollinearity or low observations, the negative effect of the value of mortgage-backed securities on mortgage defaults was still taken into account. This implies that if the value of mortgage-backed securities decreases, mortgage defaults increase. Following the results, mortgage defaults have no effect on the value of mortgage-backed securities, but if one follows economic intuition they should. This means that a decline in the value of mortgage-backed securities could set off a chain of reactions, and cause situations that also arose in the financial crisis, where mortgage defaults eventually were incorporated in the value of mortgage-backed securities and caused for a sudden large decline in value. Investors and regulators should be very aware of this.

There are also some limitations and suggestions for future research. One of them is the relative low number of observations. Due to the lack of weekly data, monthly data had to be used. Because there is a limited number of observations from the time of the financial crisis until now, only 98 observations could be used. This could be helped by finding and retrieving weekly data and obtain more accurate results. Another limitation is that all the values of mortgage-backed securities are taken together, when in fact mortgage-backed securities have different credit ratings. Further research could differentiate these different credit ratings of mortgage-backed securities and then regress it on mortgage defaults to observe whether different results are found. The results showed mortgage defaults and the value of mortgage-backed securities seem to have little effect on each other. This could be due to omitted variable bias, or due to multicollinearity and thus further research could include more or other control variables.

(31)

7. References

Abramowicz, L. (2017). A Synthetic CDO by Any Other Name Is Still Risky. Retrieved from:

https://www.bloomberg.com/gadfly/articles/2017-02-03/a-synthetic-cdo-by-any-other-name-is-still-risky

Arentsen, E., Mauer, D. C., Rosenlund, B., Zhang, H. H., Zhao, F. (2014). Subprime Mortgage Defaults and Credit Default Swaps. The Journal of Finance 70 (2).

Archer, W., R. Ling, D., C. (2017). Real Estate Principles: A Value Approach. McGraw-Hill Education (5th).

Ashcraft, A., Goldsmith-Pinkham, P., Hull, P., Vickery, J. (2011). Credit Ratings and Security Prices in the Subprime MBS Market. American Economic Review 101 (3).

Baumgartner, H., Cote, J. A., Grewal, R. (2004). Multicollinearity and Measurement Error in Structural Equation Models: Implication for Theory Testing. Marketing Science 23 (4). Becketti, S. (2013). Introduction to Time Series Using Stata. Stata Press.

Bhat, G., Frankel, R., Martin, X. (2011). Panacea, Pandora’s box, or placebo: Feedback in bank mortgage-backed security holdings and fair value accounting. Journal of Accounting and Economics 52.

Board of Governors of the Federal Reserve System. (2018). Asset-backed Commercial Paper Outstanding. Retrieved May 1, 2018 from: https://fred.stlouisfed.org/series/ABCOMP

Board of Governors of the Federal Reserve System. (2018). Treasury and Agency Securities: Mortgage-Backed Securities (MBS), All Commercial Banks. Retrieved May 1, 2018 from: https://fred.stlouisfed.org/series/TMBACBW027SBOG

Brueggeman, W., B., Fisher, J. D. (2015). Real Estate Finance and Investments. McGraw-Hill Education (15th).

Campbell, J. Y., Giglio, S., Pathak, P. (2011). Forced Sales and House Prices. American Economic Review 101 (5).

Chomsisengphet, S., Pennington-Cross, A. (2006). The Evolution of the Subprime Mortgage Market. Federal Reserve Bank of St Louis Review 88 (1).

Deng, Y., Van Order, R., Quigley, J. M. (2003). Mortgage Terminations, Heterogeneity and the Excercise of Mortgage Options. Econometrica 68 (2).

Duffie, D. (2017). Financial Regulatory Reform After the Crisis: An Assessment. Stanford University Conference version: 2016 ECB Forum on Central Banking.

Fabozzi F. J., Kalotay, A., Yang D. (2004). AN OPTION-THEORETIC PREPAYMENT MODEL FOR MORTGAGES AND MORTGAGE-BACKED SECURITIES. International Journal of Theoretical and Applied Finance 7 (8).