The effect of the Euro on the housing price boom

1 University of Groningen

Faculty of Economics and Business

Master Thesis International Economics and Business

The effect of the Euro on the housing price boom

Name Student: Judith Reijnders Student ID number: S2043467

Student email: j.m.j.reijnders@student.rug.nl

Date Paper: July 5, 2012

Name Supervisor: Dr. D.J. Bezemer

Co-assessor: Dr. G. Lanjouw

2

Abstract

While research in the field of the Euro is a booming topic, its effect on the housing price boom has not been researched yet. This paper will empirically research this within a sample of the Euro countries and the other OECD countries using quarterly panel data from 1991 until 2012. The direct effect of the Euro on the housing price boom has been proven to be positive and significant, combined with three other determinants: The percentage change of GDP per capita, the percentage change of population and the percentage change of the production of the construction sector. A second model has been used to explain the effect of the Euro on the housing price boom via the percentage change of mortgage loans. This model however is not proven to be significant.

Key words: Euro, housing price boom, mortgage loans

3

Table of Contents

1. Introduction ... 4

2. Literature Review ... 6

2.1 Determinants of the housing price boom ... 7

2.2 Euro effect on the housing price boom ... 9

2.2.1 Control variables mortgage loans ... 10

3. Methodology ... 11

3.1 Sample ... 12

3.2 Measures & Data collection ... 12

3.2.1 Housing price boom ... 12

3.2.2 Population growth ... 14

3.2.3 Income per capita ... 14

3.2.4 Number of households ... 14

3.2.5 Production of construction sector ... 14

3.2.6 Euro ... 15

3.2.7 Mortgage loans ... 16

3.2.8 Control variables mortgage loans ... 16

3.3 Limitations dataset ... 16

3.4 Methods ... 17

3.4.1 Methods with the dependent variable housing price booms ... 17

3.4.2 Methods with the dependent variable housing prices ... 18

3.4.3 Method of this paper ... 19

4. Empirical results ... 21

4.1 Assumptions ... 21

4.1.1 Assumption of multicollinearity absence ... 21

4.1.2 Assumption of homoskedasticity ... 21

4.1.3 Assumption of autocorrelation absence ... 22

4.2 Descriptive statistics ... 22

4.3 Direct model results ... 24

4.4 Mechanism model results ... 27

4.4.1 Regression results ... 27

4.4.2 Logit results ... 29

5. Robustness check ... 31

6. Conclusion ... 34

6.1 Limitations & future research ... 35

6.2 Contribution to the literature ... 36

7. References ... 37

8. Appendix ... 40

4

1. Introduction

“Monetary union fosters integration by raising price transparency and reducing transactions costs. But it also implies the loss of the sovereign interest rate and exchange rate instruments.” (Hoeller & Rae 2006, p.1)

2012 is an important year in the history of the Euro, as it will celebrate its 10th physical anniversary. In 2002 the Euro replaced the domestic currency of 14 countries (Appendix 1). The anniversary is also a good evaluation point for making up the balance; with the subsequent question whether it has been a success or a failure. The Euro brought along many benefits such as the rise in the price transparency and the reduction of the transaction costs, it also caused disadvantages i.e. losing the ability of using country specific monetary policy to adjust the interest rates and the exchange rates to restore the equilibrium settings.

The ECB has the power to adjust the interest rates and exchange rates for the entire Euro area.

The interest rates set by the ECB are based on the average of inflation of the entire Euro area. Faust et al. (2001) compared the actual interest rates set by the ECB to a prediction of the interest rates the Bundesbank would have set. They showed that the interest rates of the ECB are well below the interest rates they predicted for the Bundesbank. One explanation being that the inflation (on which the interest rate prediction is based) across the Euro zone has large differentials ranging from 1.8 percent in France to 5.3 percent in Ireland in the year 2002 (Faust et al. 2001). In setting monetary policy the ECB might base their interest rates more on large countries such as Germany and France with low inflation rates. Mainly the smaller and less developed countries within the EMU benefitted from these low interest rates which occurred with the implementation of the Euro, as their initial interest rates were higher (Conefrey & Gerald, 2010).

Another advantage of the Euro was the fact that a large pool of savings became available, without the well-known exchange rate risk within the member countries. This advantage would appear for all the member countries, not just the smaller ones. The combined effect of lower cost of capital (lower interest rates) and the higher availability of capital would lead to an increased access to credit for the EMU member states (Conefrey & Gerald, 2010).

Borio et al. (1994) showed that if households and firms have an increase in access to credit this would contribute to asset price booms in the 80’s for several developed countries. These asset price booms were reflected in the massively increasing prices of equity and real estate.

Normally monetary policy would provide stable macro-economic conditions with an interest

rate that slows down major inflation booms, as the interest rates would be increased whenever

5 there is an asset price boom (inflation) and decreased during the asset price busts (deflation) (Bernanke & Gertler, 2000). However as mentioned before the differentials of inflation were large in the EMU, leading to interest rates which would target inflation in certain countries, yet kept rising inflation in others. The housing price boom is a part of this larger asset price boom phenomenon and therefore it can be expected that the increase of credit also influenced the housing price boom. Up to 2007 there has been a massive increase in the prices of houses, which was larger than any increase in housing prices shown before. This increase was larger than the economic growth at the same time, suggesting that it has been a housing price boom (Girouard et al. 2006).

In recent years the EMU has experienced an unusually strong growth of credit which is accompanied by strong housing price increases. Not only the EMU countries but also other industrialized countries experienced this growth (Goodhart & Hofmann, 2008). The attractiveness of the “new Europe” caused a rapid credit growth in the EMU countries, the start of the credit boom-bust cycle. This cycle ended with the start of the global crises (Bakker

& Gulde, 2010). Furthermore the Euro caused that the exchange rates were fixed, meaning that the member countries could not adjust the exchange rate whenever large capital inflows occurred. Alongside the capital inflows, money supply increased, resulting in a downward pressure on the real interest rates which in turn leads to a further increase of credit (Bakker &

Gulde, 2010).

Consequently the research questions will be:

What are the determinants of the housing price boom?

What is the effect of the Euro on the housing price boom?

This paper will measure the housing price boom as the increase of the housing prices, 10 percent above the trend of the percentage change of the housing prices, for at least a period of 3 quarters in a row.

The sample which will be used consists out of the EMU countries and the benchmark

countries which are the remaining OECD countries. These countries are chosen as these are

the most developed countries next to the euro zone countries. To show other effects, which

determine the housing price boom next to the euro, these countries are chosen.

6

2. Literature Review

The existing literature contains confusion on the concept of the housing price boom and what the requirements are to be able to call an increase in housing prices a housing price boom.

Housing price booms are a part of a larger phenomenon: the asset price boom. The asset price booms are the rapid increases of the prices of particularly equity and real estate (Bernanke &

Gertler, 2000). This paper is the first in researching determinants of the housing price boom;

however some research has been done on the concept of the asset price boom. The housing price boom is part of the asset price boom, for this reason literature on the asset price boom can be used for explaining the concept of the housing price boom.

The housing price boom can be measured as growth rates or as cumulative imbalances (Bruggeman, 2007). Cumulative imbalances are also known as gap measures, and are used by several authors as a measure of asset price booms i.e. Bruggeman (2007), Borio and Lowe (2002; 2004). Bruggeman (2007) defines the gap measure as a deviation of the value of a certain variable from its own trend.

The rapid increase of housing prices can be narrowed down to a period of nonstop rise in housing prices more than 10 percent higher than its trend (Detken & Smets, 2004; Adalid &

Detken, 2007). Alessi and Detken (2009) suggested that this period of rapid housing price increases should be at least 3 quarters in a row above the trend. Van Schijndel (Board member Rabobank Netherlands) suggested in an interview with the Dutch newspaper Het Financieele Dagblad (2012) that the housing price bubble is shown by the difference between the actual prices of houses and the inflation corrected trend of the housing prices.

Equity prices fluctuate too easily due to for example speculation; therefore it is difficult to find a pattern that might be explained by variables such as the Euro. Real estate prices on the other hand move more smoothly and do not respond instantly to any sudden external shocks (Gros, 2006). It is also proven that the changes in the prices of houses affect the economy to a larger extend than the changes of the stock prices do (IMF, 2003; OECD, 2004; Debelle, 2004).

The duration and the size of the housing price increase up to 2007 have been

unprecedented. The housing prices have increased way beyond the actual business cycle

(GDP), indicating that it has been a housing price boom (Girouard et al. 2006). Appendix 2a

depicts the housing prices across the world; it clearly shows the large increase since 1998.

7 Therefore from now on the concept of the housing price boom will be narrowed down to the increase of the housing prices, above the trend of the housing prices, for at least a period of 3 quarters in a row.

2.1 Determinants of the housing price boom

In the section above the differences in the definition of asset and housing price booms have been discussed briefly. In this section the choice of the determinants of the housing price boom will be explained. The previous section highlighted that this paper is the first researching the determinants of the housing price boom. The definition which has been mentioned above indicates that the housing price boom is based on the increase in housing prices. Therefore most of the literature concerning the determinants has been based on the growth of housing prices.

An important feature proven to determine the housing prices is the demand level for real estate. The demand will be influenced by the real income per capita and demographic factors (i.e. population growth and number of households).

Over the long term the increase in housing prices is influenced by the change of the income per capita (GDP per capita growth). Whenever countries grow faster an increase in the income level is expected as well. Whenever people have more to spend, the demand for real estate will grow (based on the assumption that more income leads to the demand of a bigger house or even a second house), leading to an increase in housing prices (Case & Schiller, 1990; Jud & Winkler, 2002 and O’Neill et al., 2007). O’Neill et al. (2007) showed the direct correlation of these two variables, and thereby proved that a 1 percent increase in real income per capita is correlated to a 0.28 percent increase in housing prices. This research is based on data from 1970-2006 for the following countries: United States, Germany, Great Britain, Spain, Japan, France and Australia. Appendix 2b shows this effect graphically; a clear correlation can be seen. The percentage change in real income per capita explains 28 percent of the variance of the percentage change in real housing prices in their research. Jud &

Winkler (2002) also researched this effect, concluding more modest results: a 1 percent increase in real income per capita is associated with a 0.17 percent increase in housing prices.

Their research equation has been tested with a pooled time-series cross-sectional study of 130 metropolitan markets in the US with annual data from 1984 to 1998.

Therefore the first hypothesis will be:

The change in income per capita will have a positive effect on the housing price boom.

8 A second variable that influences the demand and therefore the housing price boom is the change of population (Jud & Winkler, 2002). O’Neill et al. (2007) showed that a positive change in population (population growth) has a large correlation with the change in housing prices. The increase of the population ratio of 1 is correlated to a 0.014 percent increase in housing prices. This means that a one percent increase in population correlates to a 1.4 percent increase in housing prices. The change of population explains 51 percent of the variance of the percentage change in housing prices. Appendix 2c shows this relationship graphically, a clear pattern can be observed. Jud and Winkler (2002) on the other hand proved that population growth rate has a major effect on the change in housing prices, as a 1 percent increase in the population leads to a 1.09 percent increase in housing prices.

The increase in population, leads to an increase of the demand for housing, which in its turn results in the increase of housing prices. Consistently the second hypothesis will be:

The change in population will have a positive effect on the housing price boom.

The second demographic factor that influences the housing price boom is the change of the number of households. The number of households is also referred to as headship rate.

FitzGerald (2005, p. 13) defined the headship rate as: “the proportion of individuals that list themselves as head of household“.

The latest trend that couples tend to have a higher divorce rate combined with a lower marriage rate (OECD, 2011), may cause a rise in the number of households. Another trend which raises the number of households in Europe has been the increase of aging. Reim (2009) stated that the aging effect causes an increase in the number of one person households, which has a positive effect on the demand of real estate. This rise leads to an increase in demand for real estate, which in its turn is a determinant of the housing price boom (Conefrey & Gerald, 2010).

One note on this effect is the endogeneity between these two variables. Garcia and Rodriguez Hernandez (2008) proved that the housing prices also influence the number of households. The higher the housing prices, the longer young people remain living at their parents’ house, thereby decreasing the number of households. However, for time reasons this paper will not focus on the effect of the housing prices on the number of households. It will be recommended for further research to deal with the potential issue of the causality between these two variables, as it might influence the level of significance.

The growth of the number of households increases the demand for real estate and thereby leads to the increase in housing prices (Poterba et al. 1991; FitzGerald, 2005;

Conefrey & Gerald, 2010). The third hypothesis which will be tested is:

9 The change in the number of households will have a positive effect on the housing price boom.

Most papers tend to explain the housing price rise only by discussing the demand side, however if demand equals supply there should not be a rise in housing prices (Glaeser et al.

2005). Therefore supply factors should also be taken into consideration, as housing prices only increase whenever there is a limit on supply. The change of the production of the construction sector is important in determining whether demand outweighs supply. If there is an increase in the supply (production), the prices of houses are likely to decrease in the next 5 years and vice versa (O’Neill, 2007). The fourth hypothesis will be:

The change in the production of the construction sector will have a negative effect on the housing price boom.

2.2 Euro effect on the housing price boom

In the previous section the determinants of the housing price boom have been defined. In this section the effect of the euro on the housing price boom will be discussed. This effect has not been researched before, and will therefore be the new contribution of this paper.

In the introduction a short history of the Euro has been reviewed, and the possible effects it could have for the EMU member countries. In the years before the global crisis, there was a large increase in housing prices leading to a housing price boom. In this period the Euro zone has seen a larger deviation between the output (GDP) and the trend of output, compared to other OECD countries (Hoeller & Rae, 2006). As stated in the former, housing prices are partly explained by the growth of GDP per capita of a country.

One of the effects that might have influenced the large deviation between the housing prices and the trend since the start of the Euro is the change of the interest rate (Conefrey &

Gerald, 2010). Faust et al. (2001) compared the actual interest rates of the ECB with the

original formula of the Bundesbank in Germany to estimate whether the interest rates set for

the Eurozone would be comparable to the ones Germany should have had. They showed that

the interest rates set by the ECB are well below the ones the Bundesbank should have had,

through applying the Taylor rule. This means that the cost of capital has been reduced; people

can receive a loans (i.e. mortgage loans) at lower cost (lower interest rates) and thereby are

able to increase the demand for housing which in turn may lead to a housing price boom. The

interest rates have decreased significantly after the start of the Euro for the long-run; therefore

a permanent reduction in the cost of receiving a mortgage for buying real estate occurred,

leading to a housing price boom (Conefrey & Gerald, 2010).

10 Not only the price of the (mortgage) loans influences the housing prices but also the change of the availability of credit. With the start of the Euro there was a wider pool of savings available for the member countries, without the previous exchange rate risk (Conefrey

& Gerald, 2010). The start of the monetary union can be seen as a financial reform which increased the access to credit from the firm, household and government point of view (Borio et al. 1994). Borio et al. (1994) proved that the increased access to credit by these parties contributed significantly to the housing price boom in the 80’s in several OECD countries.

Reim (2009) states the importance of the availability of credit and mentioned it as one of the leading determinants of the housing price boom.

In the latest housing price boom the importance of the increase in the availability of cross-border credit has been outlined by Borio et al. (2011). Not only the Euro-member countries, but other developed countries also experienced the financial reforms, which led to a wider pool of savings being available (Conefrey & Gerald, 2010). Conefrey and Gerald (2010) showed that due to the no longer existing exchange rate risk at the start of the EMU, Ireland and Spain’s banking systems raised massive resources for investments in new buildings, which led to a housing price boom in these economies.

The combination of the lower price of credit and the increased availability of credit can be tested by the concept of the total amount of mortgages being issued.

Therefore the following two hypotheses will be tested:

The Euro will have a positive effect on the change of mortgage loans.

The change in mortgage loans will have a positive effect on the housing price boom.

2.2.1 Control variables mortgage loans

In the previous section the mechanism of the Euro effect on the housing price boom has been outlined. The first hypothesis in section 2.2 concerns the effect of the Euro on mortgage loans.

In order to test this effect several control variables on mortgages should be tested as well.

The first variable which influences the mortgage loans is the change in average income per person (Petrides et al. 2009; Cunha et al. 2009). The height of the mortgage loan is usually determined by a multiple of the income from the person that is asking for a loan. This means that the higher the income per capita, the higher the mortgage supplied (Cunha et al. 2009).

Another variable which influences the mortgage change is the change in

unemployment (Petrides et al. 2009). The larger the unemployment growth, the lower the

mortgage change (or even negative change) as banks will be more careful with the extension

of capital for housing if the insecurity employment grows.

11 Besides these variables, the growth of the population also affects the mortgage loans (Japelli et al. 2008). Mortgage loans are based on the future income expectations for young households. Therefore Japelli et al. (2008) suggest that the higher the population growth, which automatically causes an increase in the proportion of young households, the higher the increase in the growth of mortgage loans.

3. Methodology

The previous section has outlined the theory which has been written on the topic of the housing price boom. This section will describe the economic model which will be used in this paper based on the hypotheses formulated in the former. With this model at the basis the sample, the measures and the data collection will be presented, followed by the methods which have been used in similar papers. The final part of this section will describe the methods which are used for this paper to estimate the economic model.

The following economic model is derived from the theory & hypothesis section and will be tested:

Depending on the type of data and the method which will be used, this economic model can be used as a starting point.

This model will determine the effect of the Euro directly on the housing price boom. Theory predicts that will be positive, whereas will be negative.

If is significant this means that the Euro has an effect on the housing price boom.

Besides the direct model, an indirect model which depicts the mechanism will be

tested as well. It is interesting to also test whether the effect of the Euro on the housing price

boom can indeed be explained by the change in the cost of capital and the change in

availability of capital. The total mortgage loans comprised are used as a proxy for both of

these effects. As mentioned above, the Euro caused an increase in credit being available and a

decrease in interest rates; both of these effects are well reflected in the change of the total

amount of mortgage loans comprised. This leads to the following models:

12 The theory predicts that in the mortgages model will be positive, whereas on the other hand is most likely to be negative. In the second model depicted above the theory predicts that will be positive, whereas will be negative. The addition to the direct model will be in the shape of the change in mortgage loans which is predicted to be positive. If both in the mortgage loans and in the last model are significant (under the condition that both estimators have the same sign), the indirect effect of the Euro on the housing price boom is shown empirically.

3.1 Sample

The sample which will be used consists out of the EMU countries and the benchmark countries which are the remaining OECD countries (appendix 3) within the 1991-2012 time frame on a quarterly basis. These countries are chosen as these are the most developed countries next to the EMU countries. To show global effects or other effects, which determine the housing price boom next to the Euro; these countries are the most suitable benchmark countries. These years are chosen as they represent the starting year of most of the data up to the most recent data.

3.2 Measures & Data collection

This section will explain the measures of all the variables used in the economic models described above. It will also explain the data sources from which the data will be collected.

Appendix 4a depicts a summary table of the information described in this section.

3.2.1 Housing price boom

The housing price boom is the dependent variable of this paper. It can be measured as growth rates or as cumulative imbalances (Bruggeman, 2007). Cumulative imbalances are also known as gap measures, and are used by several authors as a measure of equity, asset or housing price booms i.e. Borio and Lowe (2002; 2004), Bruggeman (2007). Bruggeman (2007) defines the gap measure as a deviation of the value of a certain variable from its own trend.

The rapid increase of housing prices can be narrowed down to a period of nonstop rise in

housing prices more than 10 percent higher than its trend (Detken & Smets, 2004; Adalid &

13 Detken, 2007). The 10 percent is a threshold value taken to show that the growth of the housing prices exceeds the trend of the growth of the housing prices. Alessi and Detken (2009) suggested that this period of rapid housing price increase should be at least 3 quarters in a row above the trend. Therefore the housing price boom will be measured as a dummy value representing the increase of the housing prices, 10 percent above the trend, for at least a period of 3 quarters in a row.

The data which will be used for the housing price boom will be the change in house prices data from Oxford Economics

. It is derived as the change of the housing prices per quarter compared to the same quarter a year before. This is done to overcome seasonal variation in the growth of housing prices. This dataset contains data from 1991-2012. Chile, Mexico, Slovenia and Turkey are not included in this dataset, of these countries only Slovenia is a Euro country (as of 2008). New Zealand, Israel and Iceland only have data on an annual basis and will therefore be interpolated to match the rest of the data.

The next step is to create a new variable: the trend, which will be done with the help of the Hodrick-Prescott filter (the HP filter)

. The HP filter decomposes the trend (housing price changes) from the business cycle component. In the HP filter the parameter λ determines the tendency. If λ=o then the trend is the same as the actual values, whereas if λ increases the smoothness of the trend increases as well. Whenever λ approaches infinity a linear trend will be reached. In this paper the approach of Alessi & Detken (2009) will be used. They chose a slowly adjusting HP filter (100000) to make a boom possible with the difference between the actual value and the trend value.

A second new variable will be created; this variable will measure the difference between the actual housing price change in percentages and its trend*1.10 (the trend plus 10%

above the trend).

If the newly created value is larger than zero for at least 3 quarters in a row, it will receive the dummy value of 1 and if not the dummy value of zero

. This dummy variable will define whether a housing price change is part of a housing price boom.

1

Collected via the data stream computers, the data is not freely available at their website.

2

See: Mohr (2005) for a detailed explanation of the HP filter including the formulas.

3

If the HP trend is negative, the value will be manually corrected so that the 10% will be added towards 0.

4

If one quarter does not match the criteria of the housing price boom, and at the same time is in between a longer

period consistent with the criteria, it will receive the dummy variable 1 after all. This is due to the fact that even

within a housing price boom there might be one quarter which does not have the needed threshold value to be

named a boom, however the periods before and after do and therefore it can be expected that it still is a boom.

14 3.2.2 Population growth

Population growth in this research will be measured as the growth in population on a quarterly basis compared to the same quarter a year before. The data for population growth will be used from Oxford Economics

, which contains data from 1991-2012.

3.2.3 Income per capita

In most of the research that has been done, income per capita is measured as GDP per capita.

The data for GDP will be collected from the OECD

, which has a dataset on the entire OECD sample of real GDP growth year-on-year on a quarterly basis. This means the percentage change of GDP compared to the same quarter a year before. The data from this database is collected per country, from the statistical offices. The GDP is most of the time collected while using the expenditure approach.

The dataset contains variations in the starting point of the countries, some starting in 1980 and others starting only in 2001.

Change in income per capita will be computed from the data on GDP growth and population growth using the following formula: [(1+ change GDP in percentages/100) - (1+ change population in percentages/100)] / (1+ change population in percentages/100)*100.

3.2.4 Number of households

The data of the number of households cannot be found. An alternative would to divide the population with the average number of households. However the data that is available only starts in 2004 for many countries and is a full number without any decimals on an annual basis. The conclusion in here is that there is no data available concerning the number of households on a quarterly basis, therefore this variable will be dropped, and be held as a recommendation for future research.

3.2.5 Production of construction sector

The production of the construction sector can be measured as a volume index or at current prices (OECD). Obviously the production of construction at current prices will be determined by the housing prices, and therefore the volume index will be chosen as a measure of production of construction sector.

5

See footnote 4

6

www.oecd.com

7

New Zealand is the exception as the OECD was advised to use the GDP data based on the production approach.

15 The data for the volume index of the construction sector will be collected from the OECD

; Main Economic Indicators, production in the construction. This dataset contains data from 1991-2011, and is derived at quarterly basis with the base year of 2005=100. The volume index is per country collected, most of the country data is compiled based on physical output or value added of the construction sector.

The growth rate on a quarterly basis compared to the same quarter a year before will be compiled manually. The data has the same limitations as the GDP growth per capita, namely that there is a large variation between the countries concerning the starting point varying from 1991 to 1999. Besides this limitation, the dataset is also not complete. It is missing the following countries: Chile, Estonia, Israel, Japan, South Korea, Norway, Slovenia and Turkey. Of these countries Estonia (2011) and Slovenia (2007) are both in the Eurozone. This can be solved by comparing one model with the production of the construction sector included to one without this variable, to see whether there are large differences in the significance levels and the coefficients.

Another limitation of this dataset is that the data is compiled from all the countries statistical offices, which used different methods of collecting the data and units of measure varying from type of construction to enterprises.

A final limitation is that the housing prices might also influence the volume index of the production of the construction sector. When housing prices rise, the construction sector might want to produce more so that it will have a higher revenue. This endogeneity issue will be checked for with comparing lags of the change in the production of the construction sector to the housing price boom. This will check whether past performances of the production of the construction sector may influence the housing price boom.

3.2.6 Euro

A dummy variable will be created for the years in which certain countries have been using the Euro. For example the Netherlands have been using the Euro since 2002 and therefore will receive a dummy variable with the number one from quarter 1 2002 onwards. Another example is Estonia which started using the euro in 2011 and therefore will only receive the dummy value of 1 for Euro from 2011 onwards.

8

www.oecd.com

9

Information collect via personal contact with the OECD, with the help of David Brackfield.

16 3.2.7 Mortgage loans

In the indirect mechanism models the percentage change of mortgage loans on a quarterly basis compared to the same quarter a year before, are used as a measure. Mortgage loans are measured as the total of mortgages or housing loans, both including the same content and will be transformed into growth rates per quarter compared to the same quarter a year before. The data for mortgage loans is collected by Dr. Bezemer, Nicholas Morgan, Alex Kochlashvili and I from all the central banks of the countries included in the dataset

. This contains loans comprised by monetary financial institutes to domestic residents & companies for house purchase/investment. The countries missing are: France, Japan, South Korea, Norway and Slovakia. Of these countries: Japan, South Korea and Norway also experienced missing data for the production of the construction sector. France and Slovakia on the other hand are new countries missing in the dataset, and are both Euro members. France started using the Euro from 2002 onwards whereas Slovakia started using the Euro from 2009 onwards. These missing countries therefore are an important limitation for the generalizability of the research.

3.2.8 Control variables mortgage loans

Section 1.3.3 explained the control variables for the mortgage loans. The population change in percentages and the percentage change of income per capita have already been specified. The third control variable is the percentage change of unemployment. This will be measured as the number of unemployed persons as a percentage of the labour force. The OECD harmonized unemployment quarterly data will be used from 1991 to 2012. The data will be transformed into growth rates compared to the same quarter a year before. The data is quite complete, only Estonia and Israel start in respectively 1997 and 1995. This is not a problem as these two countries already have missing data for the production of the construction sector variable.

3.3 Limitations dataset

There are limitations which can be identified from the dataset. For starters, there are several countries missing which has an impact on the generalizability of the outcomes from this research. The main part of the data is based on only the OECD, which means that some member states from the EMU are not included. These states are: Monaco, San Marino, The

10

In the case that the data was not available on-line or not available completely, an email has been sent to the central banks. Therfore I would like to thank: Manfred Ackerler, Viviane De Pré, Éva Kaponya, Philip

Anderson, Andreas Kuchler, Grzegorz Pietrzak, G.19 Staff, Essi Tamminen, Amy Birdee, Linda Groulx, Lydia

Kleebinder, Statistical Information Banco de España, Tina Hansson, fjola agnarsdottir

17 Vatican, Cyprus and Malta. These are only small member states, and therefore the consequences of their data not being included will not be of crucial importance.

Within the datasets of the OECD some countries are also missing for some variables, these countries are: Chile, Estonia, France, Israel, Japan, South Korea, Norway, Slovenia, Turkey and Mexico. As has been mentioned in the latter, Estonia and Slovenia are Euro- member states. To start with Japan and Norway are important countries within the OECD, as they are part of the most developed countries. These countries not being included means that some important benchmark countries will not be included. Slovenia entered the Euro in 2007 and therefore the fact that their data is missing is a limitation. Estonia on the other hand has only entered the Euro in 2011 and the possibility of it meeting the definition of the housing price boom with the usage of the Euro is quite limited, therefore the consequences of the data non availability is not of crucial importance.

For the second mechanism model there are also data restrictions. The data concerning the variable mortgage loans is missing data for several countries. Some of these countries have already been mentioned in the latter as missing data, however there are some extra countries: France and Slovakia. Both of these countries are Euro members since respectively 2002 and 2009. The absence of these two countries in the dataset therefore has implications for the generalizability of the outcome of the second model.

Other limitations are that the data for the construction of the production sector is being conducted based on different data per country. However as the data is based on growth percentages instead of levels, it can be used as valid data. Also the missing data for the number of households is a large limitation.

3.4 Methods

This section will describe the methodology used in this paper. First several methods used in other papers will be discussed and evaluated; second the methods used in this paper will be discussed.

3.4.1 Methods with the dependent variable housing price booms

Most of the papers examined asset price booms, which is a combination of the housing price

boom and the equity price boom. The only difference in determining the asset- compared to

the housing price boom is the fact that different data has been used. Therefore most of the

methodologies described in here will concern asset price booms; however the methodologies

are equally applicable for housing price booms.

18 Alessi and Detken (2009) researched the early warning indicators for asset price boom in the OECD countries from 1970 to 2007. They have used panel data to conduct a signaling approach, which has a binary dependent variable. “In the signaling approach a warning signal is issued when an indicator exceeds a threshold, here defined by a particular percentile of an indicator's own distribution. This approach assumes an extreme non-linear relationship between the indicator and the event to be predicted” (Alessi & Detken, 2009 p. 6). To determine the asset price boom, they manually created a dummy variable under the conditions that the asset price growth exceeds 1.75 times the standard deviation of the model above the trend, for at least 3 quarter in a row.

This method uses a model for signaling when an asset/housing price boom may occur according to certain pre-stated threshold levels of the determinants. In this paper there is no pre-stated threshold level necessary as the idea is not to signal if an housing price will occur, but instead to determine whether the Euro has an effect on the already occurred housing price boom.

Detken and Smets (2004) researched the stylized facts for real, financial and monetary policy changes during the asset price boom. Their sample consists of OECD countries from 1970 onwards. They have defined 2 types of asset price booms; low and high ones, depending on the relative post-boom growth performance. The booms have the same requirements as the paper described above, except for the fact that deviation from the trend does not have to be 1.75 times the standard deviation but instead 10% above the trend. They compare the median of many real and financial variables between the 2 types of booms in a Wilcoxon-Mann- Whitney test.

The disadvantage of this method is that it does not empirically show how much the financial and real variables predict the 2 different types of booms. They only show that there are some differences in the levels of these variables concerning the 2 types of booms. This method would not be an option for this paper as the median of a dummy variable (Euro) will not state anything.

3.4.2 Methods with the dependent variable housing prices

Conefrey and Gerald (2010) wrote a paper on managing housing bubbles in regional

economies under the EMU. They used time series data for Ireland and Spain, on which they

have used 2 separate models to determine housing prices. They have conducted a regression

model in which the dependent variable is housing prices (new housing prices deflated by the

personal consumption deflator); this is determined (for Ireland) by income per capita, housing

19 stock per capita, percentage of the population aged between 25 and 34 years old, the real cost of capital for housing (mortgage interest rate minus the change in housing prices) and a dummy variable for financial liberalization (after 2003). For Spain, the same model is being used, with the addition of another dummy from 1987 which captures the joining of Spain to the EU and subsequently the changes in their financial system.

The disadvantage of this model is that the real cost of capital for housing will be determined by the change in housing prices, which is constructed from the dependent variable of their model (housing prices).

This model would not be completely applicable for this paper as it cannot use the definition of the housing price boom: the increase of the housing prices, 10 percent above the trend, for at least a period of 3 quarters in a row. Besides this, they used 2 models in which financial liberalization dummy is dependent per country. By testing a different model for every country it becomes too time consuming for this research.

3.4.3 Method of this paper

As the sample already suggests, a panel analysis will be used to test the hypotheses. As has been defined in the former the housing price boom will be measured as the increase of the housing prices, 10 percent above the trend, for at least a period of 3 quarters in a row.

The model that fits the research question and the definition of the housing price boom the best is a binary model. Alessi and Detken (2009) also used a type of binary model, however, the signaling approach they had used is not the most suitable model for this research. Another type of binary model is a probability model such as a probit or a logit analysis. A logit will be used for the direct model, in which the probability of a housing price boom to occur is between 0 and 1. Under this condition the dependent variable is the log of the odds ratio: ln(p/1-p). The independent variables will be regressed against the logit and not the dependent variable. Suggesting the following model:

For the mechanism model a multiple regression will be used to determine the effect of the

Euro on the mortgage loans, and a probability model to determine the effect of the mortgage

loans on the housing price boom.

20 For both models country fixed effects will be used, as there are location specific fixed effects that influence the growth of the housing prices and thereby the housing price booms.

Some examples are government policies and the lack land available (Jud & Winkler, 2002).

Appendix 4b shows the Hausman test for fixed and random effects. The null hypothesis is that both measurement methods are appropriate to use as they give the same coefficients. However as the outcome shows; the test is significant, suggesting that the null hypothesis can be rejected. This means that the coefficients are not the same and therefore the fixed effects are better to use, due to a specific random effects assumption which is not met: that the random errors are statistically unrelated to the independent variables.

If the Euro dummy will have a significant effect in the first model, the research question can be answered. If it has a positive significant effect then the following conclusion can be drawn: The euro does have a positive effect on the housing price boom, and made the housing price boom larger in percentage change relative to the non-Euro countries/before the Euro was introduced.

If the Euro dummy has a negative effect it would mean that the euro caused a negative effect on the housing price boom compared to the non-Euro countries and the Euro countries before the Euro was introduced.

If on the other hand the Euro does not have a significant effect at all, this would suggest that the Euro did not have an effect on the housing price boom in general. If the Euro dummy in the indirect mechanism model is not significant either, then the outcome is clear as well: The euro does not have an influence on the housing price boom.

Another possibility is that the Euro dummy in the first model is significant but in the

mechanism model not, future research has to determine what the mechanism might be of the

influence of the Euro on the housing price boom.

21

4. Empirical results

This section will describe the empirical results of the previously determined methods. First the assumptions will be checked, and then the descriptive statistics of the data will be shown.

If the assumptions are met, the probability models and the regression will be tested to prove the hypothesis and to be able to answer the research question.

4.1 Assumptions

Both the logit probability model and the multiple regression (fixed effects) model require certain assumptions which have to be met. The probability models are different from the multiple regression (fixed effects) models in a way that certain assumptions do not have to be met. For example the autoregression assumption is not possible to be met as the only values being there will be 0 and 1. It does however require the assumption of multicollinearity absence.

4.1.1 Assumption of multicollinearity absence

One of the assumptions of the probability model is that the independent variables are not allowed to be perfectly correlated. This assumption can be tested with the Pearson correlation test. Appendix 5a depicts the table of the Pearson correlation test. It shows that there exists no multicollinearity as the highest value of correlation is 0.57 between the percentage change in GDP per capita and the percentage change in unemployment.

4.1.2 Assumption of homoskedasticity

The homoscedasticity test is applicable to both the regression model and the logit model.

Appendix 5b-5d depict the residual plots of the independent variables. The first two residual

plots show some sort of heteroskedasticity (the opposite of homoskedasticity), as a pattern in

the variance of the errors can be observed. In the first plot, the errors seem to increase in

variance, the lower the percentage change in the population. For the percentage change in

GDP per capita the opposite effect is happening, the larger the percentage change in GDP per

capita, the larger the variance of the error term. The third residual plot does occur to be

homoskedastic. Appendix 5e shows the result of the modified Wald test for groupwise

heteroskedasticity. The outcome shows that there is a significant difference at the 99 percent

confidence level, that the variance of the errors is dependent on different values of the

independent variables. Thereby the assumption of Homoskedasticity is not met for the

regression model.

22 For the logit model the assumption of homoscedasticity cannot be tested with a panel analysis combined with fixed effects. Therefore the residual plots compared to the predicted logit value will be taken into consideration; these are plotted in appendix 5f. A diagonal line is expected, and for the data to be homoscedastic the errors should be at equal distance from this line at all values of the independent variables. The plots show that in the plots of the percentage change in population and the percentage change in the production of the construction sector the errors are positioned around this line at the same variance across different values of the independent variables. However the change in GDP per capita plot does show that between 0 and 5 percent change in GDP per capita the variance is slightly larger. However appendix 5g depicts the histogram of the GDP per capita, it shows that most of the values are in between the 0 and 5 percent change in GDP per capita and thereby it might explain the slightly larger deviation of the errors in this area. Therefore the assumption of homoskedasticity is met for the logit model.

4.1.3 Assumption of autocorrelation absence

The third test is only applicable for the regression model. Autocorrelation is only an issue when time series or panel data is being used. It means that the values of one variable are being correlated with its own past or future values. The autocorrelation can be tested with the Wooldridge test for autocorrelation in panel data. Appendix 5h depicts the results of this test.

There are two methods which can account for both the heteroskedasticity and the autocorrelation of the regression model, namely the Generalized Least Square method (from now on GLS) and the robust estimator of covariance (from now on VCE). The advantage of the GLS is that it is used the best when there are many time periods and few clusters, in which the VCE is the opposite. One problem with the GLS is that first of all it cannot use fixed effects, which could be solved with creating time dummies. Another problem with GLS is that it does not work when there are holes in the data, which is the case with outliers in the data which is being used. Therefore the VCE method will be used, which uses robust estimates of variance, and is also referred to as heteroskedasticity robust estimator of covariance.

4.2 Descriptive statistics

The table in appendix 6 depicts the descriptive statistics of the variables used in this paper.

The dummy variables of the Housing price boom and the Euro are not very informative in this

table. The other 5 variables are in percentage change compared to the same quarter a year

before, to overcome seasonal effects. The table shows that the standard deviation of three of

23 the variables is more than twice the mean, and thereby it is assumable that there are several outliers influencing the mean and the errors of these variables. Stata has the option to identify the outliers using the Hadi method (1992, 1994). The outliers are the observations that are not in line with the model’s assumptions. Many outliers have been identified, after careful looking at the variables to check whether they are truly outliers or just some slightly higher variables, appendix 4a (the last column) presents the outliers which have been taken out of the data for the analysis.

The table below depicts the descriptive statistics of the data without the outliers. The standard deviation of the change of the production of the construction sector is still quite large. The standard deviation of the percentage change in unemployment almost 1000 times as large as the mean, this however has more to do with the low mean, considering that the range is only 11.1 percent.

Table 1: The descriptive statistics Descriptive Statistics

Number of observatio ns (N)

Range Minim

um

Maximu m

Mean Standard Deviatio n Housing price

boom

2463 1 0 1 0.378 0.485

Euro 2463 1 0 1 0.195 0.396

Δ GDP per capita

2090 24.490 -10.429 14.061 1.880 2.997

Δ Population growth rate

2436 4.920 -1.865 3.055 0.582 0.584

Δ Production construction

1612 99.252 -48.144 51.008 1.685 10.608

Δ Mortgage loans 1460 123.369 -26.266 97.103 13.161 14.684

Δ Unemployment 2219 11.1 -4.7 6.4 -0.009 1.119

24 As has been mentioned, the descriptive statistics of the dummy variables do not explain much, therefore a frequency table has been added below for the housing price boom and the Euro.

The frequency table shows that a little less than one third of all the quarters in the data is in a period of a housing price boom; one third of the increases in housing prices on a quarterly basis is 10% above the trend for at least a period of 3 quarters in a row. Only one fifth of the entire dataset is filled with data from a country having the Euro as their currency. This is due to the fact that only from 2002 onwards some of the countries included in the dataset started using the Euro s their currency.

Table 2: Frequency table for the dummy variables: housing price boom and Euro Frequency

Number of observations (N)

Value Frequency Percentage Cumulative percentage

Housing price boom

2463 1 1531 62.16% 62.16%

Housing price boom

2463 0 932 37.84% 100%

Euro 2463 1 1981 80.43% 80.43%

Euro 2463 0 482 19.57% 100%

4.3 Direct model results

This section will show the results of the direct model, for which a logit model will be used.

Fixed effects are applied to include country specific effects on the housing price boom.

The table below shows the logit results of the effect of the determinants and the Euro on the

Housing price boom. The first model includes the determinants of the Housing price boom,

whereas the second model also adds the effect of the Euro and the third model excludes the

change in the production of the construction sector. All the models’ coefficients have also

been transformed into odds ratios to make the interpretations more clearly.

25 Table 3: The logit results from the effect of the Euro on the housing price boom.

Variable Expected sign

Model 1:

Control variables

Model 1:

Odds ratio

Model 2:

Complete model

Model 2:

Odds ratio

Model 3:

Complete model without production construction

Model 3:

Odds ratio

Euro + 0.848***

(0.205)

2.336***

(0.480)

0.711***

(0.162)

2.036***

(0.331) Δ GDP per

capita

+ 0.225***

(0.032)

1.252***

(0.040)

0.242***

(0.032)

1.274***

(0.041)

0.255***

(0.022)

1.290***

(0.028)

Δ Population + 2.234***

(0.291)

9.339***

(2.718)

2.093***

(0.291)

8.116***

(2.370)

1.971***

(0.224)

7.180***

(1.615) Δ

Production construction

- 0.043***

(0.008)

1.044***

(0.008)

0.045 (0.008)

1.046***

(0.008)

Observations 1506 1506 1506 1506 2090 2090

Z-statistic 254.33*** 254.33*** 271.68*** 271.68*** 245.53*** 245.53***

* significant at 90 percent confidence level

** significant at 95 percent confidence level

*** significant at 99 percent confidence level

The first model is significant at the 99 percent confidence level. Showing that without the effect of the Euro the entire model is significant as well. All variables separately are also significant at the 99% level of confidence and have the right expected sign, except for the production of the construction sector.

The second model shows the complete model, which in terms of coefficients and significant levels is pretty similar to the control variables model except for the added Euro effect which has a positive and significant influence at the 99% level of confidence.

The third model has been used to expand the number of observations as the variable

the change in production of the construction sector limits the total amount of observations. It

26 proves that even with less data the models are consistent as the signs and the significance levels remain the same, only the coefficients are slightly lower.

The main variable of this paper, the Euro, has a positive and significant effect on the housing price boom. This means that the main hypothesis is proven: The Euro has as positive effect on the housing price boom. The Euro increases the log odds of a housing price boom to occur by 0.84 in the complete model. This interpretation of the coefficient does not explain much; therefore the odds ratio has also been conducted. The Euro causes that the odds of a housing price boom to occur to get multiplied by 2.33, meaning that with the Euro the odds of a housing price boom is more than twice as big then without the Euro.

The second variable which in theory has been shown to determine a housing price boom is the change in GDP per capita. This variable is estimated to have a significant positive influence on the housing price boom, which proves the first hypothesis of this paper. A one percent increase in GDP per capita increases the log odds of a housing price boom by 0.24.

Transforming the log odds into odds ratios this means that a one percent increase in GDP per capita leads to a multiplication of the odds of a housing price boom to occur by 1.27.

The third independent variable in the second model is the change in population. As well as the two variables presented above, the change in population also has a significant positive influence on the housing price boom at the 99 percent confidence level, which proves the second hypothesis of this paper. A one percent increase in population increases the log odds of a housing price boom by 2.09. In terms of odds ratio this means that a one percent increase in population multiplies the odds of a housing price boom by 8.11.

As has been mentioned in 3.2.4, the change in the number of households could not be supported by appropriate data, therefore the third hypothesis of this paper cannot be proven.

The last independent variable of the second model is the change in the production of the construction sector. The sign does not match the expected sign based on the theory.

Instead of the influence being negative, the change in production of the construction sector

has a positive and significant effect on the housing price boom. However the effect does not

have a large influence, as a one percent increase in the percentage change of the production of

the construction sector increases the log odds of a housing price boom by 0.04. In terms of

odds ratio this means that a one percent increase in the percentage of the production of the

construction sector multiplies the odds of a housing price boom by 1.046. Thereby the fourth

hypothesis is proven to be wrong. An explanation might be that the demand side may had

outweighed the supply side (even though it increased) and thereby still have led to the

increase of the housing prices and eventually the housing price boom. Another explanation for

27 the positive sign might be that at the time the housing price boom started, the production of the construction sector increased positively as well due to the higher prices and thereby higher revenues the construction sector could achieve. This would suggest that there is an endogeneity issue regarding the housing price boom and the production of the construction sector. Jud & Winkler (2002) explained that in the past large increases in the production of the construction sector preceded drops in housing prices, and thereby explain that the negative effect of the change in the production of the construction sector on the housing price boom might need a lag. Appendix 7 shows the logit results with 3 lags of the change in production of the construction sector. Even though the significance does decrease from the 99% to the 95% level of confidence, it is shown that even past percentages of change of the production of the construction sector do influence the housing price boom. This proves that the effect indeed goes from the percentage change of the production of the construction sector towards the housing price boom.

4.4 Mechanism model results

In this section the multiple regression (fixed effects) and logit results of the mechanism model will be discussed briefly. In the assumptions section heteroskedasticity has been detected, and therefore the robust estimator of covariance will be used for the multiple regression.

4.4.1 Regression results

The first part of the mechanism model is the empirical testing of the effect of the Euro on the Mortgage loans, this will be tested with a multiple regression (fixed effects). The table below shows the results of several models, the control variables model and the complete model.

28 Table 4: The regression results from the effect of the Euro on the percentage change of the mortgage loans

Variable Expected sign

Model 4:

control variables

Model 5:

Complete model

Constant 8.184***

(1.906)

8.905***

(1.804)

Euro + -2.719

(2.755) Δ GDP per

capita

+ 1.265***

(0.305)

1.213***

(0.292)

Δ Population + 4.537

(2.995)

4.752 (2.946) Δ

Unemployment

- 0.171

(0.905)

0.188 (0.893)

Observations 1335 1335

Within R² 0.119 0.125

Z-statistic (F) 9.22*** (F) 7.94***

First of all there are no large differences between the control variables model and the complete model. However the complete model’s variables are explaining 12.5% of the variance of the percentage change in mortgage growth, whereas the control variables model’s variables only explain 11.9% of this variance.

The table above is showing some striking results, as the Euro dummy has a negative sign but is not significant. This is at odds with the fifth hypothesis and the theory of the mechanism in which the Euro influences the housing price boom. Besides the Euro variable the percentage change in unemployment also has an unexpected positive sign; however it is not significant either.

The change of GDP per capita is the only significant variable in this model at the 99 percent confidence level. It has a positive effect on the change in mortgage loans. The other two variables: the change in population and the change in unemployment do not have a significant effect on the percentage change of mortgage loans.

29 4.4.2 Logit results

The previous section has shown some striking and contradicting results; however to complete the mechanism model the second test should be done as well. The previous test has shown the effect of the Euro on the percentage change of mortgage loans, whereas this second model will depict the effect of the percentage change of the mortgage loans in its turn on the housing price boom.

Table 5: The logit results from the effect of the percentage change of mortgage loans on the housing price boom

Variable Expected sign

Model 6:

Control variables

Model 6:

Odds ratio

Model 7:

Complete model

Model 7:

Odds ratio

Δ Mortgage loans

+ 0.021**

(0.008)

1.022**

(0.008) Δ GDP per

capita

+ 0.225***

(0.032)

1.252***

(0.040)

0.226***

(0.038)

1.254***

(0.047)

Δ Population + 2.234***

(0.291)

9.339***

(2.718)

2.181***

(0.320)

8.857***

(2.834) Δ

Production construction

- 0.043***

(0.008)

1.044***

(0.008)

0.041***

(0.009)

1.042***

(0.009)

Observations 1506 1506 1189 1189

Z-statistic 254.33 254.33 238.56 238.56

The results do not differ much from the results of the first logit model, which showed the

effect of the Euro on the housing price boom. The only variable which is interesting in these

results is the change of the mortgage loans variable. The effect on the housing price boom is