HOUSE PRICES BOOMS IN 6 EUROPEAN COUNTRIES AND THEIR EFFECT ON CONSUMPTION GROWTH AND INCOME INEQUALITY

-Master thesis International Economics and Business-

HOUSE PRICES BOOMS IN 6 EUROPEAN

COUNTRIES AND THEIR EFFECT ON

CONSUMPTION GROWTH AND INCOME

INEQUALITY

Winnie van der Wal S1337165

Rijksuniversiteit Groningen Faculty of Economics and Business First supervisor: Dr. D.J. Bezemer Second supervisor: Dr. Akkermans

Abstract

Several countries have experienced a house price boom in the last decades. Often this was accompanied by consumption growth. In this thesis a panel data set consisting of 6 countries is used to analyse the effect of an increase on house prices on consumption. If there is a housing wealth effect, an increase in house prices leads to increased consumption. The strength of this wealth effect might depend on the degree of financial deregulation. Higher collateral values give better access to credit and so raise consumption. Deregulated financial markets facilitate this process. This thesis analyses how the effect of house prices on

consumption depends on the degree of financial deregulation.

House price booms can also lead to increases in income for the wealthier layer of the income distribution while decreasing disposable income for the lower layer. I will study the

Table of contents

1. Introduction and main research questions……….5

1.1 Introduction………...5

1.2 Main research questions………6

2. Review of theory and literature………...8

2.1 House prices and consumption………...8

2.2 House prices and income inequality………...………12

2.3 Hypotheses………..14

3.The house price booms in Ireland, Spain, the United Kingdom, Estonia, Latvia and Lithuania………...………16

3.1 The house price boom in Ireland……...………..16

3.2 The house price boom in Spain…………...………18

3.3 The house price boom in the United Kingdom………...………19

3.4 The house price boom in the Baltic states: Estonia, Latvia and Lithuania……...……..21

4. Empirical research………..24

4.1 Economic model. The effect of house prices on consumption………..………….24

4.2 Economic model. The effect of house prices on income inequality………..…….28

4.3 An overview of the methodologies used in research………..………31

4.4 Choosing a methodology…………..………..33

4.5 Testing the assumptions………….………35

5. Results……….40

5.1 Results of the consumption regression…………..……….………40

5.2 Results for the income inequality ……….……….41

6. Conclusion………...44

6.1 Conclusion………..44

6.2. Limitations and future research……….46

Appendix……….48

Graph 1. Global house price developments………..49

Table 1. Gini coefficients for Ireland, Spain and the United Kingdom 1995-2005…………49

Table 2. Gini coefficients for Estonia, Latvia and Lithuania 2000-2005………49

Table 3. Variable definition and their sources for the consumption regression………..50

Table 4. Variable definition and their sources for the income inequality regression…..……51

Table 6. Descriptive statistics. Gini regression……….52

Table 7. Unit root test for Δ log con..………...52

Table 8. Unit root test for Δ Gini………..………...53

Table 9. Correlation matrix for the variables in the consumption regression……….53

Table 10. Correlation matrix for the variables in the income inequality regression………….54

Table 11. Variance Inflation Factor values for variables in the consumption regression…….54

Table 12. Variance Inflation Factor values for variables in the Gini regression……..………54

Table 13. Regression output. The consumption regression………..55

Graph 2 House prices and income inequality in Ireland………56

Graph 3. House prices and income inequality in Spain………..56

Graph 4. House prices and income inequality in the United Kingdom………...57

Table 14. Regression output. Gini regression………..58

Table 15. Regression output. Gini regression with AR(1) model………59

References………60

1.Introduction and main research questions

1.1 Introduction

Since 1995, several countries in the world have experienced a sharp rise in house prices: a house price boom. Graph 1 shows global house price developments between 1995 and 2005. House prices in Ireland, Spain, Australia, the United Kingdom (UK) and the Netherlands have grown tremendously. Especially the UK and Ireland experienced an enormous increase in house prices between 1995 and 2005; the index rose from 100 to 270. Since 2000, numerous countries in Central and Eastern Europe also experienced double-digit annual growth rates of real housing prices. In the second quarter of 2006, Latvia ranked number one in terms of house price growth in Knight Frank’s global housing index1.

What is the effect of a house price boom on the economy of a country? House price booms result in an increase in wealth for those who own assets. This wealth increase can influence consumption through the wealth effect, implied by the permanent-income theories.

Consumers spread out their consumption over their lifetime, based on permanent income. When this permanentincome increases, for example by an increase in house prices,

and makes it easier to take on collateral-backed loans for households that already posses collateral. Therefore I will in this paper look at the influence of house prices on consumption and at how this effect depends on the degree of financial liberalisation in a country. I will focus on six European countries that experienced an asset boom: Spain, Ireland, the United Kingdom, Latvia, Estonia and Lithuania. The reason to focus on these countries is that Ireland, Spain and the UK have experienced the largest increases in house prices of all EU-countries between 1995 and 2005. In the Knight Frank’s global housing index Q1 2006- Q1 2007 the number 1,2 and 4 were the Baltic states Estonia, Latvia and Lithuania and I will also look a the house price boom in these countries.

Another possible effect of an increase in house prices is an increase in the income inequality. Not everyone in a country owns their house, so wealth inequality increases when house prices rise. When this wealth is transformed to income this could in turn lead to an increase in income inequality. In this paper I will use a regression analysis to find the effect of house prices on income inequality.

Furthermore, increasing housing costs might decrease the income available for spending after housing costs are paid, the disposable income. Rents usually move together with house prices. This brings down the disposable income of renters, while this on the other hand presumably increases for home-owners. This paper will look at the effect of house prices on income inequality in all the six countries and at the effect of house prices on disposable income of different income quartiles in Ireland, Spain and the UK.

1.2 Main research questions

but little research has been done based on cross-country data that takes into account the credit conditions in a country. I will add to the existent literature on the effect of house prices on consumption by including a variable for financial deregulation.

My first main research question is:

“What is the effect of an increase in house prices on consumption?”

Furthermore I will investigate the effect of house prices on income inequality. I will also look at the disposable income of different income groups and study how it moves with house prices to see whether an increase in house prices leads to increased disposable income inequality. My second main research question is:

“What is the effect of an increase in house prices on income inequality?”

2. Review of theory and literature

In this literature review first the literature on how house prices and consumption are related is discussed. Second, the literature on the influence of house prices on income inequality is described. At the end of this chapter I will outline my hypotheses, based on the literature.

2.1 House prices and consumption

Rising house prices can influence consumption through the wealth effect. According to the permanent income or life-cycle consumption theory a household’s consumption is spread out over its lifetime, dependent on its permanent income, defined as the annuity vale of household wealth. An unexpected wealth shock changes the permanent income of households and thus affects the life-cycle pattern of consumption and saving. A rise in house prices is likely to be seen as a permanent rise in income and therefore to increase consumption.

The wealth effect works through the credit market. The availability and cost of external finance by banks depend on their assessment of a household’s net worth. Therefore, higher house prices will make it easier for households to acquire loans and will lead to higher consumption.

Evidence on the wealth effect is mixed. Kenny (1998) argues that it is far from certain that increased housing wealth will unambiguously lead to higher consumption or lower savings. Increased consumption could also be the result of other factors, such as higher income, expectations thereof or looser credit constraints.

Case, Quigley and Shiller (2005) observe a panel of 14 countries for various periods during the past 25 years and a panel of US states quarterly during the 1980s and 1990s to study the link between increases in housing wealth, financial wealth and consumer spending.

They use the following equation:

Δlog ct = aΔlog ct-1 + β1 log yt + β2 log stockt + β3 Δ log houset + γ[log ct-1 - log yt-1] + β3 Δ log

stockt-1 + fixed effects + εt (1)

where c is consumption, y is income, stock is stock market wealth, and house is owner-occupied housing wealth, all in real per capita logarithmic forms. The results show strong evidence that variations in housing market wealth have important implications upon consumption.

Aron, Muellbauer and Murphy (2006) criticise the article of Case et al (2005) for leaving out important controls. Among the omitted controls are the interest rates, the unemployment rate, income growth expectations and a shift in credit conditions. They study the effect of

increasing house prices on consumption in the UK between 1972 to 2005 using the following equation:

Δ log ct = α[α0 – α1r1t – α2Ө2 + α3EtΔymt-k + γ1NLAt-1/Yt + γ2IFAt-1/Yt + γ3HAt-1/Yt +

(log yt – log ct-1) ] + β1Δlog yt – β2(DBt-1/ Yt ) Δ log nrt + εt (2)

Where c is consumption, r is the real interest rate, θ is an indicator of income uncertainty (proxied by the four quarter change in the unemployment rate), EtΔymt-k is a forecast of the

constraints. The γ’s measure the propensity to consume out of wealth for the different assets. They allow the parameters to shift with the credit conditions index (CCI).

Results show that the housing wealth effect largely works through the credit channel: high collateral values give better access to credit and so raise consumption. The marginal propensity to spend out of housing wealth is around 0.03 when the CCI is at the maximum and around 0.02 when the CCI is at the average.

According to Benito et al. (2006) there is evidence that income expectations have at times played and important role in the co-movement between house prices and consumer spending. The authors claim that the empirical association between house prices and spending has waned in recent years in the UK. They do acknowledge the fact that houses could serve as a collateral for increasing loans, and show that mortgage equity withdrawal and consumer spending started to move close together in the UK after financial liberalisation in the mid-80s. However, over the last decade, the association between mortgage equity withdrawal and consumer spending in the UK has been weaker.

Catte, Girouard, Price and André (2004) examine the link between housing wealth and consumption in 18 OECD countries. Their consumption equation is written as:

the contrary, the propensity varies between 0.01 and 0.02 in Japan, Italy and Spain and is statistically insignificant in France and Germany. The authors argue that the influence of the housing market on consumption depends on the extent to which homeowners are able to borrow against housing wealth. Housing equity withdrawal (HEW) is typically calculated by subtracting the household sector’s residential investment from the net increment in their mortgage debt. The marginal propensity to consume out of HEW ranges from 0.20 in the Netherlands and the US to 0.89 for the United Kingdom. By contrast, no effect of HEW is found for France, Spain, Germany, Italy and Japan.

The authors expect consumption responses to changes in housing wealth to be higher, ceteris paribus, in countries where:

• Financial markets are well-developed and provide easy access to mortgage financing and to financial products that facilitate equity withdrawal

• The rate of owner-occupation is high, which implies a wider distribution of housing wealth

• Housing transaction costs are low and housing wealth is exempted from capital gains taxes, which both stimulate home-owners to perceive housing assets as more liquid.

Smith (2006) analyses equity withdrawal by households in New Zealand. Household asset values increased in the previous years, while saving by households decreased. One of the ways in which households could be dissaving is via housing equity withdrawal. Results show that a 10 percentage point increase in the housing equity to income ratio boosts housing equity withdrawal by 0.7 per cent of income in the following quarter. A one percentage point

Overall, most research, especially when the credit conditions are taken into account, supports the view that there is a positive effect of house prices on GDP growth. The effect differs across countries and years, which could possibly be the result of changes in institutions and the credit conditions.

2.2 House prices and income inequality

In a review Hillyard (2007) deals with inequality in Northern Ireland. He argues that house prices play a large role in the wealth gap between rich and poor. He claims that the biggest increases in capital values have accumulated to those living in the largest houses. According to Hillyard, there has been plenty of evidence that many homeowners have released the increased equity of their house and invested in more property. This all leads to the rich getting richer and the poor finding it increasingly difficult to buy decent accommodation or enter owner occupation.

Piketty (2003) and Piketty and Emmanuel (2001) argue that ups and downs in the top incomes are mainly the result of capital income changes and that they therefore play an important role in income inequality. Both research uses time series data from tax returns from the 1910s to 2000 to study the long-term dynamics of the income share of the top tenth income percentile households in the U.S. and France. They conclude that the changes in income shares of the top tenth percentile households are predominantly a capital income event.

An increase in wealth inequality increases income inequality only if there is interest or dividend income and/or when at least a portion of the capital gain from the investment is realised. A household may choose no to liquidate assets or realise return for an extended period of time – for example, investment for retirement. Then there is no effect of an increase in wealth on income. However, on the aggregate level, there are always some households liquidating part of their wealth or realising their capital gain in each period, even more so in a booming market (Zhao, 2005b). Zhao (2005a) investigates income inequality between stockholders and non-stockholders between 1980 and 2001 based on the Michigan’s Panel Study of Income Dynamics. As a measure in income inequality Zhao uses the Gini index. The Gini coefficient measures income inequality in a country on a scale from 0 to 1, where 0 implies perfect equality and 1 perfect inequality. In Zhao’s research the Gini index is a function of the stock market participation and the income ratio of the two types of

stockholders. The author argues that movements in income inequality could also result from changes in skill premium or from different labour market behaviour and therefore also looks at unemployment and education of households to explain differences in income. The findings are that stock price appreciation has no long-run effect on income inequality but that there is an upward trend of the average income ratio between households with a college-educated head and households whose head is not college-educated.

Whiting (2004) scrutinises the optimism in the United States on the falling poverty rate. She looks at the housing affordability in the US between 1960 and 1999 and calculates the

gender. For all groups for which data are available, the data show an increase in housing affordability after 1960 that has reversed by 1975.

In general previous research has found that an increase in house prices leads to an increase in income inequality or a decrease in housing affordability.

2.3. Hypotheses

From the literature review it becomes clear that rising house prices can affect consumption through many channels. An increase in house prices raises households’ wealth, their ability to acquire loans and their confidence.

My first hypothesis is therefore:

Hypothesis 1: an increase in house prices has a positive effect on consumption.

According to Catte et al (2004) the housing wealth effect on consumption is higher when financial markets are well-developed. The easier it is to take on or increase collateral-based loans, the more home-owners can consume above their income and the higher the effect of an increase in house prices on GDP.

Hypothesis 1a: The positive effect of house prices on consumption increases when the degree of financial deregulation increases.

increasing wages. This is likely to lead to an increase in income of households at the lower end of the income distribution, decreasing income inequality. However, since in the countries discussed house prices growth has been much higher than wages , I expect the overall effect of asset booms on income equality to be positive.

Hypothesis 2: an increase in house prices increases income inequality.

When financial deregulation is high, it is easier for households to obtain a mortgage to buy their house. If more households own their house the effect of an increase in house prices benefits a larger part of the population.

Hypothesis 2a: the effect of house prices on income inequality is smaller when the degree of financial deregulation is higher.

However, the income distribution is not all there is to inequality. House prices also have an effect on disposable income, the income after housing costs are paid. When house prices rise, wealth of home-owners increases. When this wealth is liquidated income of home-owners rises, possibly increasing their disposable income. At the other end are the renters. Rents usually move together with house prices. If rents rise this will decrease the income left for spending for households. The income left for spending after house price costs are paid for is called disposable income. Since home-owners are generally at the higher end of the income distribution and renters on the lower end this could lead to increased disposable income inequality.

3. The house price booms in Ireland, Spain, the United Kingdom, Estonia, Latvia and Lithuania

To test the hypotheses for the countries discussed in this paper, we need information on house prices, consumption, financial deregulation, disposable income and inequality during the asset booms. This information will be given in this chapter. I will describe the house booms that took place in Ireland, Spain, the United Kingdom, Latvia, Lithuania and Estonia. Most data was found on the website of Eurostat. In table 3and 4in the appendix the sources are described in detail.

3.1 The house price boom in Ireland

In Ireland, house prices started to grow dramatically in 1995 as can be seen from graph 1 in the appendix. Between 1995 and 2001 Ireland faced the largest growth in house prices of all countries in this graph. Between 2001 and 2005 only the UK showed stronger growth. The Irish growth in housing prices can be explained by a number of factors:

• Economic growth. GDP in Ireland grew significantly during the asset boom. This growth was due to a combination of factors, such as successful attraction of FDI. • Unemployment rate and rising wages. Ireland has transformed from a country with a

high unemployment rate into an economy that has the lowest unemployment rate in the European Union. Ireland also experienced an increase in wages that was larger than growth in wages in the UK, Spain and the Euro Area. Lower unemployment and higher wages lifted the living standard of people previously unemployed, hence more people were able to buy a house.

rate. The fact that mortgage credit became much more accessible led to an increase in demand for houses.

• Population growth. The return of emigrants to their country when the economy took off from 1993 onwards and the reaching of maturity of the baby boom generation led to a strong increase in the young-adult population and an increased demand for housing (Power, 2007).

• Household size. The average household size in Ireland declined in the 1990s. One reason for this decline is the legalisation of divorce in the 1990s (Power, 2007). • Culture of home ownership and official policy favouring house ownership. Irish

culture traditionally has been in favour of home ownership, and more recently second or third homes for investment or holiday purposes have become very popular. Official policy supports this culture, through mortgage interest relief for owner occupied and investment homes (Power, 2007).

All these factors led to the enormous increase in house prices. The house price boom was accompanied by a large increase in consumption.

In 2007 the housing market started to slow down. House prices dropped by 7%, the highest percentage in Europe. Nowadays, property prices are crashing.

3.2 The house price boom in Spain

In the period 1995-2006 Spanish house prices sky-rocketed, as can be seen from graph 1.

House prices tripled in some parts of the country. The factors that contributed to this large increase in housing prices are described below.

• Economic growth. GDP growth took off 1995 and was higher than in the Euro area. • Unemployment rate and wising wages. Between 1995 and 2005 the Spanish

unemployment rate halved. Wages increased during the asset boom, although not as fast as in Ireland, the UK and the Euro area. Increasing employment and higher wages led to more people being able to afford a home, raising demand.

• Availability of mortgage credit and low interest rate. Since the early 1990s, the long interest rate has been falling considerably. The lower risk premia, in combination with the liberalisation of the credit market and a large demographic shock, considerably lifted Spain’s indebtedness.Total mortgage debt has grown twice as fast as house prices in the period 1995-2007, with total mortgage balances nearly six times the level of 10 years ago (Royal Bank of Scotland, 2007).

• Population growth. The period 1996-2005 is also characterised by strong population growth. The population increased by 5 million in this period. Immigration was a key factor (Royal Bank of Scotland, 2007). The Bank of Spain estimates that foreign residents now account for 10% of total population.

• Household size. The Spanish culture is changing. More people leave their parental home and household sizes decrease (Nogales, del Real and Portomarín, 2007). This led to an increase in demand for housing.

• Culture of home ownership and official policy favouring house ownership. The Spanish culture has always been in favour of house ownership. This has been

mortgage payments in income tax statements and legal obstacles to renting (Nogales, del Real and Portomarín, 2007).

• High number of second homes. In 2007, around a fifth of households owns a second home, and many households aspire to do so. Most second homes are located in the coastal areas or on the countryside (Royal Bank of Scotland, 2007).

All these factors combined led to a high demand for housing and higher house prices. During the house price boom consumption has risen significantly.

The slowdown in the Spanish housing market started in 2005 when growth decreased from 17.1% in the first quarter to 12.6% in the fourth (EGI Web News, March 5th 2007).

Between April 2007 and April 2008 house prices fell in real terms for the first time since late 1997. This is the result of the highest inflation rate in Spain in 13 years and a rapid slowdown in the housing market (El País. April 18th 2007).

As can be seen from table 1, Spain’s Gini coefficient increased slightly between 1995 and 1997 after which it declined. In 2005 there was a little increase again. Overall the Gini

coefficient declined, meaning that the Spanish income distribution became more equal during the asset boom.

3.3 The house price boom in the United Kingdom

Similar to Ireland and Spain, house prices in the UK started to rise after 1995, as graph 1 shows. House prices have more than doubled between 1995 and 2003. The house price boom can be explained by several factors:

• Economic growth. The UK experienced high GDP growth since 1995.

boom and more than they did in the Euro area. The declining unemployment and the rise in wages both increased household’s ability to buy a house and thus raised demand for housing.

• Availability of mortgage credit and low interest rate. The United Kingdom, similar to Spain and Ireland, faced a low and overall declining interest rate.This increased household’s capacity to borrow and therefore is likely to have had a positive effect on housing demand and housing prices. The annual increase in mortgage debt doubled between 1995 and 2005.

• Population growth. Since the late 1990s net international migration from abroad into the UK became an important factor in population growth (UK National Statistics). • Household size. Another factor for increased housing demand is the increase in

divorce rates and people living alone (Harris, 2003).

• Inelastic supply of housing. Supply of housing is very inelastic in the UK. Housing supply has been quite stable since 1990 and did not increase when demand increased around 1995, leading to a rise in prices. One reason is the shortage of land (Harris, 2003).

All these factors led to an increased demand for housing.

As in Spain and Ireland, consumption rose significantly during the house price boom. The boom is over now. Growth started to decline in 2005, when prices rose only by 1.3% between July 2004 and April 2005. This was due to an increase in interest and worrying development in the world markets, such as the rising oil prices (Cameron, 2005). Between September 2007 and September 2008, house prices fell by 12.7% (The Guardian).

3.4 The house price boom in the Baltic states: Estonia, Latvia and Lithuania

House prices in the Baltic countries took off about 5 years later than in the previous described countries: housing prices were stable during the 1990s but sky-rocketed in the period 2000-2006. The index of house prices rose from 100 to 386 in Lithuania, from 100 to 495 in Estonia and from 100 to 738 in Latvia. The house price booms in these countries were the result of several factors. Many factors are the same for the three countries and the countries will be described together.

• Economic growth. While the GDP growth rate of the Euro area was under 2%

between 2001 and 2005, it was between 7% and 10% in the Baltics in the same period. GDP growth was also a lot higher than in emerging Europe (Kattel, 2007).

• Unemployment rate and rising wages. In 2000 unemployment in the Baltic states was much higher than in the EU-27. However, by 2007 unemployment rates in Estonia, Latvia and Lithuania were even below the unemployment rate of the EU-27. This is partly the result of GDP growth, and partly of the large outflow of workers to other EU countries, driven by the fact that the Baltics have become Europe’s worst place to work (Hudson, 2008). Rising employment is likely to lead to an increase in demand for houses, whereas the large outflow might decrease the demand for housing.

However, the inequality in Estonia and Latvia was quite high around 2000. This could indicate that the people leaving the country were already the poorer part of the

population, which did not own their houses. Furthermore, wages increased during the asset boom allowing more people to buy a house. Wages have increased

Ireland, Spain and the UK. This huge growth of wages led to higher income and more demand for houses.

• Availability of mortgage credit and low interest rate. The interest rate in the Baltic states decreased in this period and it became much easier to take on mortgage debt. The annual change in mortgage debt has increased tremendously in the Baltic states during the house price boom.

• Low property tax. Another reason for the real estate boom is the very low property tax in the Baltic States. The Baltic states do not register property prices properly. Latvia’s “Land Book” is based on Soviet-era registration prices and until recently did not take into account current sales prices. Homeowners only pay little tax on the rental

revenues and can capitalise these revenues into bank loans. This raises housing and commercial property prices (Hudson, 2008).

All these factors combined led to a huge increase in house prices in the Baltic states.

The growth in GDP is mainly thanks to large domestic demand and solid import demand from the Baltic states and the Commonwealth of Independent States (CIS).

In 2007, the Baltic credit boom passed its peak. Bank lending and mortgage loans decreased. Property prices started to fall due to expanding supply and declining demand. House prices in Talinn, Estonia’s capital, fell by 15% between April 2007 and January 2008.2 Prices in Riga slumped by up to 20%.3 One reason for this decline is the jump of the Euribor interest rate on Euro variable rate loans, from five percent in January 2007 to 10% in June 2007. Another cause is the imposition of a 25% tax on personal income from property sold within the year of

purchase. Further reasons for the bust are government’s policies against inflation and property speculation (findaproperty.com).

Table 2gives the Gini coefficient in Estonia, Latvia and Lithuania between 2000 and 2005. The Gini coefficient is measured in different ways in different reports. It is therefore

important to use the Gini for each year from the same source. Unfortunately, this is very hard to find for these countries. I use data from the UNICEF WIID2C database which

unfortunately lacks data on 2005 for Latvia and Lithuania and for 2006 for all three Baltic states.

4. Empirical research

In this chapter the economic models used to estimate the effect of house prices on consumption and income inequality are described.

4.1 Economic model. The effect of house prices on consumption.

The models to measure the wealth effect on consumption used in previous literature have already been described in the literature review.

Case et al (2005) used the following equation:

Δlog ct = aΔlog ct-1 + β1 log yt + β2 log stockt + β3 Δ log houset + γ[log ct-1 - log yt-1] + β3 Δ log

stockt-1 + fixed effects + εt (1)

whereas Aron, Muellbauer and Murphy (2006) use the following equation:

Δ log ct = α[α0 – α1r1t – α2Ө2 + α3EtΔymt-k + γ1NLAt-1/Yt + γ2IFAt-1/Yt + γ3HAt-1/Yt +

(log yt – log ct-1) ] + β1Δlog yt – β2(DBt-1/ Yt ) Δ log nrt + εt (2)

and let the parameters shift for the credit conditions.

The equation used by Catte, Girouard, Price and Andre (2004) is:

c = a + β(y) + ω(nfwr) + γ(nrwr) + δ(unr) + ф(irsr) + Ө(inf) + ect (3)

should be included in the consumption regression.

Aron et al (2006) define a as the speed of adjustment. This is not a factor in which I am interested and I will leave this out. I want to measure the effect on consumption growth and my dependent variable will be, in line with Case et al and Aron et al, the percentage growth in consumption. Case et al do not control for unemployment, while this has a negative influence on the income expectations of households and thus negatively affects consumption and is also an important factor in the consumption regression. Aron et al use asset to income ratios because it gives a better approximation to the underlying linear additive structure of human and non-human capital than does the log-assets formulation. The stock market wealth in my equation will be calculated as a percentage of GDP. Unfortunately I have only ratios of house prices and it is not possible to do this for housing wealth.

Catte et al do not include GDP in their consumption regression but labour income. This avoids doubling counting return on financial assets, since this is already included in the variables for house prices and stock market capitalisation. In line with Catte I will use a variable for labour income instead of GDP.

I expect the interest rate to have a negative influence on consumption. When the interest rises on for example mortgages there will be a negative effect on consumption. The relevant interest rate to use is then the long-term interest.

The equation I will use is:

Δlog cont = [β0 + β1IE + β2ES + β3GB + β4LV + β5LT] + β6 logliit + β7 Δ smcit + β8 log rit +

β9uit + β10 Δlog HPit + β11mgit + β12(Δlog HPi,t · Δmgit) + εi,t (4)

GB the dummy for the United Kingdom, LV for Latvia and LT for Lithuania. These dummies capture the specific effect. The coefficient of the intercept measures the country-effect of Estonia. The Baltic states were dealt with simultaneously in chapter 3 because the causes of their asset booms were quite similar. They do however have separate dummies because there are still substantial differences between them. The consumption per capita and growth in consumption per capita in Estonia is much higher than in Latvia and Lithuania. Population growth is also different across the countries (Eurostat) and there might be more differences that can cause a difference in intercept.

The variables and their sources are described in table 3.

It is very well possible that consumers react to a rise in house prices with some delay. Consumers might wait a while before they perceive the increase in housing wealth as a permanent increase income. It might also take some time to take on a mortgage, or to spend the increased spending power obtained through the mortgage. A lagged value should be included to measure the effect of a rise in house prices in the previous period on consumption growth in this period. However, including this data leads to multicollinearity between the variables. This will render the results to be useless and I will leave them out.

To control for credit conditions Aron et al (2006) include the credit conditions index, CCI. They use the CCI estimated by Fernandez-Corugedo and Muellbauer (2006), which is derived by modelling data for ten credit indicators, from which a common credit indicator and a risk indicator is extracted, after controlling for standard economic and demographic variables. Unfortunately this index is only available for the UK and the data needed to construct such an index is not available for all the countries I deal with in this paper. Bayoumi (1993) used for credit deregulation the ratio of total outstanding consumer credit to GDP, but data for

countries studied in the paper I expect financial deregulation to effect consumption through expanded credit with houses as collateral, and not also through borrowing for non-durable consumption, such as is very common in for example the US.

The interaction term measures the effect of the change in mortgages on the effect of house price growth on consumption growth. I expect the coefficient of this interaction term of house prices and mortgages to be positive, meaning that financial deregulation strengthens the effect of house prices on consumption. When the coefficient for mortgages is significant and

positive this indicates that the financial deregulation on itself has an effect on consumption. Consumers can more easily take on mortgage debt to increase consumption irrespective of the value of their collateral.

Hypothesis 1 states that I expect house prices to have a positive effect on consumption. I thus expect β10 to be significant and positive.

Hypothesis 1a states that the positive effect of house prices on consumption increases when the degree of financial deregulation increases. I therefore expect β12 to be positive and

significant as well.

There are now two independent variables included that influence each other. This was also the case in Aron et al who include both CCI and interaction terms of all variables and CCI. In paragraph 4.5 diagnostic tests will be performed to check for multicollinearity to see if this causes any problems.

financial wealth in a consistent way. Therefore I have used the index of house prices and stock market capitalisation as a percentage of GDP instead.

It would be interesting to include interaction effects of house prices and a country dummy to find if there are significant differences in the effect of house prices on consumption. These differences could exist because of differences in home ownership rates. When the home ownership rate is high more people will benefit from increased wealth and the effect on consumption will be higher. Differences in wealth tax systems could also lead to country differences in the effect of house prices on consumption. However, the interaction terms can not be included since this leads to problems of multicollinearity.

In Ireland, the UK and Spain, the asset boom started in 1995. I will therefore conduct my research for these countries for the years 1995-2005. In the Baltic countries, the asset boom began in 2000 and I will look at those countries between 2000-2006.

4.2. Economic model. The effect of house prices on income inequality

Gyimah-Brempong and Munoz de Camacho (2006) measure in their article the effect of corruption on income distribution. The equation they use is:

gini = γo + γ1y + γ2edu+ γ3y + γ4corrupt + γ5govcon + γjΣj dumj x corrupt + ξ (5)

j = Africa, Asia and Latin

skip this variable. Instead I will use the ratio of house prices (HP) to measure the effect of house prices on the Gini coefficient. Zhao (2005a) suggested that unemployment also influenced the Gini coefficient. Household in his study with an employed or out-of-labour head or spouse have a substantially lower income. High unemployment will thus increase the Gini coefficient. Therefore I will add the level of unemployment to the equation.

Instead of the level of the Gini coefficient I will measure the difference in the Gini coefficient to measure if an increase in house prices increases the Gini coefficient.

Again, the degree of financial deregulation is important. If financial deregulation is high, households will find it increasingly easy to get a mortgage to buy their home. If more people own their home, an increase in house prices will benefit more people and inequality will decrease. I will include an interaction term of house prices and the change in mortgages. This measures how the effect of house prices on the Gini coefficient depends on the change in mortgages. As in equation 4 the variables mortgages and house prices influence each other. Diagnostic tests in paragraph 4.5 will show if this is expected to cause any problems. Due to multicollinearity problems I leave out two variables; education and income. Looking at the data there is no indication that the Gini coefficient is dependent on the level of income. In 2000 Estonia had the highest Gini coefficient of all countries while it had the highest GDP per capita of the Baltics. The UK had higher GDP per capita than Spain and Latvia and also a higher Gini coefficient than those countries. I expect GDP per capita growth to be more important in the determination of the change of the Gini coefficient. Furthermore part of the difference in GDP per capita is captured in the intercept since I use a fixed effects model. Education is left out since I except the difference in income between an unemployed and an employed person to be higher than the skill premium. To make the equation more linear I will use the linear-log model.

Δginiit = [γo + γ1 ES+ γ2GB + γ3 IE+ γ4LV + γ5LT] + γ6 y + γ7 Δlog HP + γ8uit + γ9mgit +

γ10(Δlog HP · Δmg) + εi,t (6)

where i and t denote the country and year, respectively.

In this equation gini is the level of the Gini coefficient of income distribution, y is the growth in of per capita income, HP is the index of house prices, u is the unemployment rate and mg is the annual change in mortgages outstanding. ES is the dummy variable for Spain, GB the dummy variable for the United Kingdom, IE the dummy variable for Ireland, LV the dummy variable for Latvia and LT the dummy variable for Lithuania. γo captures the country effect of

Estonia. The variables and their sources are described in table 4 in the appendix. Again it would have been interesting to add an interaction term measuring the country

differences in the effect of house prices on the change in the Gini coefficient. When the home ownership rate is high a large part of the population benefits from an increase in house prices. In this case the effect of house price growth on the Gini coefficient will not be as large as in the case of a low home ownership rate. However, as in the consumption regression, I was refrained from doing so because it leads to multicollinearity.

4.3 An overview of the methodologies used in research

The data used is a panel series, which is a combination of time-series and cross-sectional data. It is a data set in which there is data on the same countries over several time periods. It

consists of house prices and the independent and controlling variables between 1995 and 2005 for Ireland, Spain en the UK, and between 2000 and 2006 for Estonia, Lithuania and Latvia. Using panel data has the following advantages, according to Brüderl (2005):

• Panel data are more informative because there is more variability, less collinearity and more degrees of freedom, which makes estimates more efficient

• It allows to study individual dynamics

• It gives information on the time-ordering of events • It allows to control for unobserved heterogeneity.

In the case of cross-country data, the estimator relies totally on a between-country

comparison. This means it is biased because of unobserved heterogeneity. Panel data uses a within-country comparison. This makes it possible to identify the true causal effect. Most previous literature on the effect of house prices on consumption or investment has used panel data, although there are also cases in which the effect is studied in one particular country, using time series. Most research on the effect of house prices on income inequality focused on one country using time series. Piketty (2003) and Piketty and Emanuel (2001) use time series for France and the US.

I chose to use panel data for both models based on the advantages stated above.

The constant coefficient model (or the pooled regression model) has constant coefficients for both slope and intercept. This model assumes neither significant country nor significant temporal effects. The data can be pooled and an ordinary least squares (OLS) regression model can be used. The disadvantage is that there is still bias due to unobserved

heterogeneity, because

u

it and xit are correlated.

In the fixed effects model the countries have constant slopes but there are differences in intercepts. The intercept varies only across countries and not over time. There are two estimation methods for the fixed effects model: entity-demeaned OLS regression and the Least-Squares-Dummy-Variables-Estimator (LSDV). The entity-demeaned OLS regression measures the deviations from the averages. Another way to treat panel data is to include i-1 dummies, where i is the number of countries and the equation is estimated by OLS. This is the LSDV. The disadvantage of the fixed effect model is that it is only practical when the number of the units is not too big, since otherwise too many dummies are required. Apart from being cumbersome this also often results in a loss in a large number of degrees of freedom (measured as the number of dummies – 1).

In the random effects model each country can have a different intercept parameter and it is assumed that the intercepts are random variables. Between country differences are assumed to be random, and within country estimates are correlated. An advantage of the random effects model is that it allows for time-invariant variables to be included among the regressors.

effects. Lagged variables can be used in either a fixed or a random effect model. An instrument estimator (IV) is used. An IV estimator is based on the assumption that the

instrument correlates with the corresponding explanatory variable, but not with the error term.

There are different techniques for the IV estimator, including the Two-Stage Least Squares (TSLS) method and the Generalised Method of Moments (GMM). The two stages of the TSLS refer to stage 1 where new dependent or endogenous variables are created to substitute for the original ones, and the second stage in which the regression is computed in OLS fashion, but using the newly created variables. The purpose of the first stage is to create new dependent variables which do not violate OLS regression's recursivity assumption.4 This model is used if the exogenous variables are assumed to be known at time t-1 and the errors are not

successively correlated.

The GMM is developed by Arellano and Bond (1991) and produces consistent estimates when there are dynamic and endogenous regressors. GMM requires nothing about the shape of the distribution of the error terms and is therefore usually robust to violations of homoskedasticity and normality. However, GMM is a relatively complicated technique and sensitive to outliers. Furthermore, its properties hold when N is large, so in panel data with a small number of cross-section units the results can be severely biased and imprecise (Bruno, 2005)

4.4 Choosing a methodology

In this subchapter I first will describe what type of model previous literature has used and

on the basis of this knowledge explain which model I am going to use and why.

4.4.1. Choosing a methodology. The effect of house prices on consumption.

Hogan and O’Sullivan (2007) use both an OLS regression and an IV model to research the wealth effect in Ireland. They first run an OLS regression of consumption on income, interest rates and housing wealth. However, there might be a bias because of the relationship between income and consumption. Consumption is a function of income, but aggregate consumption is a part of GDP so that income is also a function of consumption. When this bias is not taken into account this might bias the estimates of the Marginal Propensity to Consume (MPC ) out of housing wealth. To control for these variables they re-estimate the model using an IV model, using as instruments the lagged values of all variables, current real government consumption per capita, potential GDP per capita and lagged real interest rates. The results of the IV estimation are little different from the OLS model.

Case, Quigley and Shiller (2005) use a fixed effect model when looking at the wealth effect of house prices on consumption. They also use a error correction consumption model with lagged variables, but these results are not significant different than the results using the fixed effect model. Aron et al (2006) use the generalised least squares (GLS) model to estimate their equation. The GLS is a transformed model in which the variance over the transformed error is constant over the whole sample.

The equation of Catte et al (2004) is estimated with the ordinary least squares (OLS).

the GMM method may be severely biased in small samples (Smith, 1997). Since I only have data for 6 countries and 11 years it is likely that the GMM will give biased results. I expect that there are country differences in the intercept and I will therefore use the LDSV.

4.4.2. Choosing a methodology. The effect of house prices on income inequality Zhao (2005a) uses a fixed-effect panel data model to find the effect of stock market appreciation on income inequality.

Gyimah-Brempong and Munoz de Camacho (2006) use a dynamic panel (DPD) estimator. They use this partly because it produces consistent estimates in the presence of endogenous regressors and partly because they lack reasonable instruments for the endogenous regressors that can be excluded from the growth equation.

For the same reason as mentioned before, the fact that GMM may be severely biased in small samples, I will not use GMM. Since I expect there to be country differences in the intercept I will add country dummies and use the LSDV.

.4.5 Testing the assumptions

To evaluate the validity of the regression, a few tests have to be conducted. These tests will be described below.

4.5.1. Descriptive statistics

Descriptive statistics present the basic characteristics of the data. The mean, median, standard deviation, minimum, maximum and the Jarque-Bera statistic and probability for both

For the consumption regression the null hypothesis is rejected for mortgages, the interaction term (Δlog house prices * Δ mortgages), consumption growth (Δlog consumption) and house price growth (Δlog house prices). For the Gini regression the same variables are not normally distributed, apart from consumption growth.

A dot plot is used to detect far outliers. An outlier is an observation that is very different from the rest of the data. The regressions are run without the outliers to see if there is a significant difference but this is not the case.

However, one reason for concern are the mortgage to GDP ratios for Spain in 2004 and 2005. The base of how mortgage was calculated on the Instituto Nacional de Estadística website changed in 2004 and there is a large difference between the change in mortgage in 2003 and in 2004. The mortgage for 2003 was given in both new and old base. I calculate the ratio of old to new and apply this to the data on the new base. This is the data I will use.

4.5.2. Nonstationarity

In time series, the dependent variable is stationary if its mean and variance are constant over time, and the covariance between two variables from the series depends solitary on the time between the two values, and not on the actual times at which the variables are observed. When these criteria are not met, nonstationarity exists. In time series the Dickey-Fuller test is used to test for nonstationarity. Unfortunately this is not possible in a panel data set.

EViews has the possibility to test for six types of unit root tests.

Table 7 shows the unit root test for Δlog cont .Three out of four tests reject the hypothesis of

nonstationarity, so I conclude that now the dependent variable is stationary.

For the dependent variable in the gini-regression the null hypothesis is rejected in all four tests as be seen from table 8 and I conclude that the dependent variable is stationary.

4.5.3. Multicollinearity

Multicollinearity exists when some independent variables are highly correlated. In the case of multicollinearity the data might not be informative for the purpose and it might not be

possible to isolate the economic relationship of interest.

To test for multicollinearity I will compute the correlations between the independent and control variables. As a rule of thumb I conclude that multicollinearity exists when the correlation between 2 variables is higher than 0.7.

As mentioned before, when an interaction term is included measuring the country effect on the effect of house prices on consumption, multicollinearity exists and I had to remove those interaction terms.

Table 9 and 10 in the appendix show the correlation matrices. None of the variables have a correlation higher than 0.7 suggesting there is no multicollinearity. However, this test does not detect all multicollinearity. The Variance Inflation Factor (VIF) measures how much the variances of the estimated regression coefficients are inflated because multicollinearity exists. A VIF higher than 10 indicates the presence of multicollinearity .

4.5.4. Heteroskedasticity

Heteroskedasticity exists when the variances for all observations are not the same, violating the linear regression model. The unit root tests described in paragraph 4.5.2 tested whether the variances are constant over time. To test whether the variances are constant over countries I have to test for heteroskedasticity. To test for heteroskedasticity the White test can be used. However, EViews does not yet support the White test for panel data. To account for possible heteroskedasticity I will use the White Diagonal Covariance method. This provides consistent estimates of the coefficient in the presence of heteroskedasticity of unknown form. When no heteroskedasticity exists this method does not wrongly affect the results.

4.5.6. Autocorrelation

When using time series data there is the risk of successive errors being correlated with each other. One way to test for autocorrelation is using the Watson test. If the Durbin-Watson statistic ≈2 there is no evidence of positive autocorrelation.

If autocorrelation is found I will control for autocorrelation by using an autoregressive, an AR(1), model. In this model et depends on its lagged value et-1 plus another random

component that is uncorrelated over time and has zero mean and constant variance. The random component et is thus composed of two parts; one part that is the ‘carryover’ from the

random error in t-1, and a part that is a new shock to the variable. The AR(1) model confirms that shocks run through for more than one period.

4.5.7. Model misspecification

there is no misspecification. Unfortunately the Ramsey test is not yet available for panel data in EViews.

4.5.8 Reverse causality.

Correlation does not necessarily mean causation. It is possible that not only xi has an

influence on y but y also has an influence on one or more elements of xi , for example on x2i.

When y and x2 are simultaneously determined this is called reverse causality. If reverse

causality exists this might cause biased results.

The Granger test looks at how much of the current dependent variable can be explained by past values of this variable and to see whether adding additional lags of the independent variable can improve the explanation.

The dependent variable (y) is said to be Granger caused by the independent variable (x) if x helps in the prediction of y. The null-hypothesis is that there is no causality. When the p-value is smaller than 0.05 the hypothesis is rejected at the 5% significance level and there is

causality.

5. Results

5.1 Results for the consumption regression.

In table 13 in the appendix the result of the regression can be found. The Durbin Watson statistic is 2.2. It is close to 2 and thus not completely conclusive. Another test for

autocorrelation is the correlogram of residuals. When the autocorrelation in the correlogram does not pass the dotted lines there is no autocorrelation.The correlogram shows that there is no autocorrelation.

The adjusted R2 value is 0.289630 meaning that 29% of the variation in consumption growth is explained by the explanatory variables.

At the 5% significance level, no coefficient is significant. At the 10% level the coefficient for the intercept is significant and positive. This is the country-effect of Estonia. The coefficient for the country-dummies are insignificant, meaning that there is no difference in the model between the countries.

The coefficient of unemployment is significant and negative at the 10% level. This means that an increase in the unemployment rate will decrease consumption growth. This is as expected since higher unemployment lowers the income of people and increases income uncertainty. Labour income is also significant at the 10% level. It is negative, implying that an increase in labour income decreases consumption growth. This is not as expected, since we would expect that an increase in labour income increases consumption growth. However, it could be

explained by the fact that the marginal propensity to consume out of labour income is likely to decrease when labour income increases. At first, a small increase in labour income will

All the other coefficients are negative. There is no significant effect of house prices on consumption. The coefficient of the interaction term is also insignificant, meaning that the effect of house prices on consumption does not depend on the amount of mortgages.

5.2 Results for income inequality

First, I look at the development of the index GDP per capita per quartile and the index of rents in Ireland, Spain and the UK between 1995 and 2000. Rents are used because I do not have data on house price with 2000 as base year. Since the base year for GDP is also 2000 rents are easier to compare. It is a good proxy since house prices and rents move together.

The development of rents and income per capita per quartile is shown in graphs for each country. Quartile 1 is the lowest income quartile and quartile 4 is the highest income quartile. The base year for income per quartile and rents is 2000. The effect of the house price boom on disposable income is described. Disposable income is income per capita minus rents.

Graph 2shows the development of GDP per capita and rents in Ireland between 1995 and 2001. Between 1995 and 1998 GDP per capita rose at about the same rate for all quartiles and growth was higher than rent growth. For all quartiles disposable income increased. In 1999 the rent price decreased while GDP per capita increased for quartile 1, 2 and 3 and it

between 1995 and 2001 disposable income increased for Q1, Q2 and Q3 while it decreased in the fourth quartile. In conclusion, disposable income distribution became more equal.

Graph 7gives the increase in the index GDP per capita and rents in Spain between 1995 and 2000. 2001 is not included since the LABORSTA uses 2000 as the base year for 1995-2000 and 2001 for the base year afterwards. This gives a ratio of 100 for both 2000 and 2001 and will therefore not give a reliable view. For quartile 1, between 1995 and 1997 rents increased while income decreased. Disposable income decreased. Between 1997 and 1999 income increases much faster than rents and disposable income increases tremendously. In 2000 income increases a bit more than rents leading to increased disposable income.

For quartile 2, 3 and 4 disposable income decreases slightly in 1996 and stays the same in 1997. It decreases in 1998 and increases in 1999 and 2000. Overall, the disposable income of quartile one, the lowest income quartile, increased more than the disposable income of the other quartiles, decreasing disposable income inequality.

Graph 4gives the development of the indices of GDP per capita and rents in the UK between 1995 and 2001. There is no significant difference between the indices of GDP per capita of each quartile between 1995 and 2000. Overall the GDP per capita increased more in this period than did rents, raising disposable income. In 2001 the income per capita of quartile 1, the lowest income quartile, increases more than the other quartiles. In this period, disposable income increased more for this quartile than for the other quartiles and disposable income inequality decreased.

Unfortunately there is no data available on GDP per capita per quartile for the Baltic States.

so it not conclusive. To be sure the correlogram of the residuals is checked. This shows autocorrelation. To control for autocorrelation I use an autoregressive, an AR(1), model. The AR(1) model confirms that shocks run through for more than one period. Table 15 gives the regression using an autoregressive model.

The R2 _{is 0.19 and the Durbin-Watson statistic is 2.22. The total number of observations is 35.}

The Gini coefficient was not available in 6 observations and using an autoregressive model decreases the number of observations since lagged values are used.

The coefficient of mortgages is significant at the 5% level. It is positive, meaning that an increase in mortgages has a positive effect on the change in the Gini coefficient. A possible explanation is the effect of interest costs. With increased financial deregulation, more lower-end incomes are granted a mortgage and will have to pay interest costs. If these costs are higher than the rental costs they paid before this might raise income inequality.

The coefficient of house prices is not significant.A change in house prices has no direct effect on the change in the Gini coefficient.

The coefficient of the interaction term of house prices and mortgages is significant and negative, supporting my hypothesis. The effect of house prices on the Gini decreases with increased liberalisation. If house prices rise and the financial market is underdeveloped, this raises the wealth of the home owners. If this wealth gain is realised, income inequality increases. It becomes increasingly difficult for non-home-owners to buy a house and it is difficult and expensive for them to get a mortgage. If financial deregulation is high, more people can buy a house and share in the increased wealth.

much higher than in all other countries and the other country dummies should also be significant and negative. This is not the case.

6. Conclusion 6.1 Conclusion

The first research question in this thesis is:

“What is the effect of an increase in house prices on consumption?”

I estimated this effect by a regression in a fixed-effect model and included a variable for financial deregulation, measured as the amount of mortgages granted in a year. I used data of 6 different European countries during the period that they experienced a house price boom. For Ireland, the UK and Spain this was 1995-2005 and for Estonia, Latvia and Lithuania the sample was 2000-2006.

The results show that there is no direct effect of house prices on consumption. The interaction term is also insignificant, meaning that the effect of house prices on consumption does not depend on the degree of financial deregulation.

The answer to the research question is thus: there is no effect of an increase in house prices on consumption.

The second research question is:

“What is the effect of an increase in house prices on income inequality?”

quartile and the disposable income distribution becomes more equal. Thus when looking at disposable income an increase in house prices decreases income inequality.

Another measure of income inequality is the Gini coefficient. This measures income inequality on a scale from 0 to 1, where 0 implies perfect equality and 1 perfect inequality. I use a regression analysis to measure the effect of an increase in house prices on the change in the Gini coefficient. The same sample is used as in the consumption regression. The direct effect of house prices is positive not significant. However, the interaction effect of mortgages and the Gini coefficient is negative and significant. The effect of house prices on the Gini coefficient decreases with increased financial deregulation. Financial deregulation makes it easier for lower-income households to buy their own house and a larger part of the population benefits from an increase in house prices. The effect of house prices on income inequality decreases.

In conclusion, an increase in house prices has no direct effect on house prices. However, it has a negative effect on the Gini coefficient when financial deregulation is sufficiently high.

6.2 Limitations and further research

The most important limitation of this research is the sample size. 6 countries is not a very large sample and for the countries discussed it was hard to find data for all the variables in each year. Some years were therefore excluded decreasing the sample size even more. The research might be expanded by more countries to be able to make more general assumptions. It could also be expanded by including countries that are very different in their home

Another limitation is the use of mortgages as a proxy for financial deregulation. An increase in financial deregulation does not necessarily mean an increase in mortgages. If people do not take the opportunity to borrow more there is no effect of financial deregulation on mortgages. I have used it since it was the best proxy available to me. If more data is available another proxy could be used.

APPENDIX

HOUSE PRICES BOOMS IN 6 EUROPEAN

COUNTRIES AND THEIR EFFECT ON

CONSUMPTION GROWTH AND INCOME

INEQUALITY

Winnie van der Wal S1337165

Rijksuniversiteit Groningen Faculty of Economics and Business First supervisor: Dr. D.J. Bezemer

Graph 1. Global house price developments

Source: IMF

Table 1. The Gini coefficient in Ireland, Spain and the United Kingdom

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Ireland 0,33 0,33 0,33 0,34 0,32 0,30 0,29 0,30 0,31 0,31 0,32 Spain 0,34 0,34 0,35 0,34 0,33 0,32 0,33 0,31 0,31 0,31 0,32 UK 0,32 0,32 0,30 0,32 0,32 0,32 0,35 0,35 0,34 N/A 0,34 EU-15 0,31* 0,30* 0,29 0,29 0,29 0,29 0,29 N/A 0,30 0,30 0,30 *Eurostat estimate Source: Eurostat

Table 2. Gini coefficient in Estonia, Latvia and Lithuania

2000 2001 2002 2003 2004 2005

Estonia 0.39 0.39 0.39 0.40 0.38 0.36

Latvia 0.34 0.32 0.33 0.33 0.32 N/A

Lithuania 0.36 0.35 0.36 0.32 0.31 N/A

Table 3. Variable definition and their sources for the consumption regression. Economic concept Abbreviation Units Description and Source

Consumption Con(01) In million

euros

Final consumption expenditure per capita, Eurostat

Labour income li Index Labour compensation per employee, base 2000=100, OECD StatExtracts Stock market capitalisation smc Percentage of GDP World Bank

Long-term interest rate r Percentage OECD website, national bank and statistics websites

Unemployment rate u Percentage Labour Force Survey adjusted series, Eurostat

House prices hp Index For Ireland, UK and Spain: base year = 1985, Bureau for International Settlements. For Estonia, Latvia and Lithuania: base year = 2000. Sources: Estonia: Estonia Statistics

http://www.stat.ee/

Latvia and Lithuania: Arco Real Estate, UNECE Discussion Paper series 2005.5.

Mortgages mg Percentage of GDP

Annual change in mortgages. Lithuania: Hypostat

Latvia: Bank of Latvia www.bank.lv

Estonia: Bank of Estonia

www.bankofestonia.info

UK: Bank of England

www.bankofengland.co.uk

Ireland: www.cso.ie

Spain: Instituto Nacional de Estadística: www.ine.es

The data in Spain for 2004-2005 is measured on a new base on the website. There is an overlap for 2003 for the new and old base and I used the relationship between the new and old value for 2003 to calculate values for 2004 and 2005 that are consistent with the other years.

Country dummy Ireland

IE Takes value

Spain 0 or 1 Country dummy United Kingdom GB Takes value 0 or 1 Country dummy Latvia LV Takes value 0 or 1 Country dummy Lithuania LT Takes value 0 or 1

Table 4. Variable definition and their sources for the Gini regression.

Economic concept Abbreviation Units Description and Source Gini Gini Scale from 0 to 1 Eurostat for Ireland, the UK

and Spain and the Unicef WIID2C database for Latvia, Lithuania and Estonia

Growth of per capita income

y Percentage growth Eurostat

House Prices hp Index For Ireland, UK and Spain: base year = 1985, Bureau for International Settlements. For Estonia, base year = 2000, Latvia: Arco Real Estate, Lithuania: Arco Real Estate and Global Property Guide Unemployment rate u Percentage Labour Force Survey adjusted

series, Eurostat Income per capita by

quartile

Index Mean net income per capita by quartile. Eurostat

Table 5. Descriptive statistics for the consumption regression.

Δlog