The Relationship between House Prices and Stock Prices in the Netherlands: An Empirical Analysis

(1)

Bachelor Thesis

The Relationship between House Prices and Stock Prices in the Netherlands:

An Empirical Analysis

Name: Ali Azimi

Student number: 10617108

Study: Economics and Business Economics Specializations: Finance and Organization Field: house prices, stock prices

Number of credits thesis: 12

Assigned supervisor: Mr. S. Olijslagers Date: 30 June 2020

(2)

2

Statement of Originality

This document is written by student Ali Azimi who declares to take full responsibility for the contents of this document.

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

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

(3)

3

Abstract

The fluctuations in house prices and stock prices have a considerable impact on the wealth of homeowners and investors. The purpose of this research is to investigate the relationship between residential housing prices and stock prices in the Netherlands. Quarterly data is used for the period 1996 to 2020. For the analysis, the variables, real gross domestic product, consumer price index, and Interest are used to the model. First, linear regressions are conducted to establish the number of lags needed in the model. Second, the augmented Dickey Fuller test is conducted to investigate the stationarity. Third, econometric methods such as unrestricted var model and granger causality test are analyzed. The latter is used to investigate the direction between all variables mentioned. The results show that the combined variables SPI, RGDP, CPI and Interest Granger cause housing prices. Lastly, the variables HPI, RGDP, CPI, and Interest combined Granger causes the stock prices.

(4)

4

1. Introduction

During the last decade, there were immense highs and lows between the house prices and stock prices have been observed. Real estate is the primary asset held by households almost everywhere in the world. For the Netherlands, the total assets of all households are 1416 billion euros, of which 725 billion euros is mortgage debts. According to CBS (2019), fluctuations in housing prices have a significant impact on capital for households, and so on for their wealth. According to the research of Parlevliet and Kooiman (2015) from the DNB, Dutch households have an increase in home equity growth for the period 1990-2008. At the same time also, their mortgage debts increased. Because of their longer balance sheets, the Dutch people became more vulnerable to asset prices and interest rates fluctuational. Mortgage loans increased from 30% of GDP in 1982 to 109% in 2012. For portfolio diversification, it is essential to understand more about the relationship between the stock market and the housing market. Households have fewer opportunities for diversification because their home is a significant portion of their total assets. Investors, on the other side, have more diversification opportunities, for both the relationship and, more precisely, the direction between house prices and stock prices is essential to understand for making clear choices about placing their capital (Van den End and Kakes, 2002).

The difference between both markets is that stocks are investments with low transaction costs and high liquidity, while the housing market has high transaction costs and low liquidity. The contrary is that the stock prices are more volatile in comparison with the house prices. However, the two assets typically appear in homebuyers’ and investors’ portfolios (Lin and Fuerst, 2012). For a clear understanding of the relationship, it is essential to understand the direction between both markets. The motivation for this research question is to investigate the relationship between the housing market and the stock markets in the Netherlands and to add macroeconomic variables to the analysis. According to Van den End and Kakes (2002), the direction is mostly from the stock market through the housing market, with housing prices responding to stock prices with a delay of two to three years. Chen (2001) also found significant evidence that stock prices tend to lead to housing prices. This will support this relationship presented in the research from Van den End and Kakes. In addition to the direction, Sutton (2002) explained that stock prices could explain house price fluctuations in six countries (USA, UK, Canada, Ireland, the Netherlands, and Australia). Both markets are also influenced by other macroeconomic factors such as GDP, consumer trust, and interest rates.

(6)

6 In support of the research from Quan and Titman (1999), real estate prices also found to be significantly influenced by GDP growth rates. This thesis intends to explain more about the relationship between both markets in the Netherlands. From the insights of the hypotheses, there may be able to predict the movement and possible investment strategies for investors in the Dutch stock market or the housing market. Furthermore, explain the possible impacts for upswings or downturns in those markets. The research question is: What is the Relationship between House Prices and Stock Prices in the Netherland? In addition to this question, there could be more understanding of the direction between the macro variables that are used in the paper and the direction from those variables on the stock markets and housing market.

The thesis is structured in the following way. After the introduction, the literature review is presented where the theories are described for understanding more about the relationship between both markets. Furthermore, there is a review of the empirical methods that are used in the paper. For this analysis, there is made use of the Augmented Dickey-fuller test to check for stationarity. For the optimal lag selection, the Akaike information criterion is used. Lastly, the conclusion and the discussion will be presented based on the findings in the results. The results are presented in the following order. There is an explanation first of the descriptive returns and graphical analysis for the data. After that, the results of the ADF test and the AIC is presented. With the results from those tests, there are regressions presented, and the Granger causality is performed.

(7)

7

2. Literature Review

In this section, the existing literature is reviewed to give a clear understanding of the relationship between stock and housing prices. From the theory side, there can be made a distinction between the two transmissions. Firstly, the wealth effect and credit price effect are explained based on prior research. Secondly, there is more focus on the study’s that are done in the Netherlands.

2.1. Transmissions

The first transmission is the wealth effect. This effect assumes that houses are investments and consumption goods. When stock prices rise, investors and households will notice this in their perception of their growing wealth. This will lead to an increase in demand for housing and will shift the demand curve upward. Housing prices will rise in this case as a result (Kapopoulous and Siokis 2005; Yuksel 2016). The direction of this relationship is that stock prices should lead to house prices. The second transmission is the credit-price effect. This effect assumes that there is a constrain of loans for purchasing real estate. For investing in real estate, there is a need for capital. When housing prices are rising, firms and households have higher collateral values for their loans. The theory states that this would lead to getting more or higher loans for purchasing real estate. This increase in demand lead in turn that households and firms have more capital to invest in the stock market (Lean 2012, Kapopulous and Siokis 2005). Higher housing prices stimulate in this way the economic activity in the Netherlands. To find support in favor for this effect, house prices should lead to stock prices. In this thesis, this effect is more unlikely because the data that is used is only for residential real estate.

2.2 Prior Research In The Netherlands

In this section, prior research conducted on the Dutch markets will be discussed. Besides the relationship explaining the variables housing prices and stock prices, there is also an effect of other macro-economic variables. First, Quan and Titman (1999) did a cross-sectional analysis that included the Netherlands. They found a significant positive relationship between stock returns and changes in housing prices. Besides the significance, they also found that there is a significant effect between GDP growth rates. Furthermore, Sutton (2002) made a similar conclusion. Through VAR analysis, which also included the Netherlands, this paper found that there is a significant part of changes in housing prices that are explained through the stock prices. Besides the variables mentioned above in this analysis is the Interest rate included. More

(8)

8 precisely, Sutton (2002) found for the Netherlands through the VAR that by an increase of 10% in stock prices, there was a raise for about 2% in the housing prices over a three-year period. Lastly, Van den End & Kakes (2002) did investigate the coherence between stock prices and housing prices. In their analysis, they included six countries, which included the Netherlands. For the period 1976-2001, they found that the stock market and housing market have a long-term coherence. Part of this coherence could be explained through macro-economic variables such as lending rates, consumption expenditures, and interest rates. Their conclusion is that there is a direct relationship between stock and housing prices. The direction that is mentioned is from the stock prices to the housing prices with a delay of 2 to 3 years. This direction will support the wealth effect theory. In later research, Kakes and Van den End (2004) found that besides the directional effect, there is also support for a causal influence between the stock market and the housing market. This causal influence is only supported when research is done by including macro-economic variables such as interest rates and economic growth.

When looking at previous research, there are a few different statistical methods used to analyze this relationship. Batayneh & Al-Malki (2015) did investigate the relationship between house prices and stock prices in Saudi Arabia. They conducted an empirical analysis with the unrestricted VAR method and the Granger causality test. The results showed that the stock market and real GDP influence housing prices. For the analysis of this thesis, a similar approach is used. From Granger causality, the support for the direction stock prices to housing prices is supported by the research of Chen (2001). The general perspective is that the relationship between stock prices and housing prices is complicated. However, this should not be the basis for not doing research.

2.1 Hypothesis

In this thesis, there are four hypotheses conducted. The first and second hypotheses support the wealth effect and the credit-price effect in the Netherlands. The third hypotheses suggest that there is Granger cause in both directions. The last hypothesis is for not finding Granger cause between stock prices and house prices. For stock prices and house prices, the returns are investigated through various statistical analyses. The hypotheses are similarly conducted as the paper of Batayneh & Al-Malki (2015). In addition to the formulated hypotheses, there will also be examined if there are other Granger causalities between the macro-economic variables to the stock prices and house prices and vice versa.

(9)

9 (1) Stock prices (SPI) Granger causes house prices (HPI). This will support the wealth

effect theory.

(2) House prices (HPI) Granger causes stock prices (SPI). This will support the credit-price effect theory.

(3) Stock prices (SPI) and house prices (HPI) Granger causes in both directions. Both effects will be supported.

(4) No Granger Cause between stock prices (SPI) and house prices (HPI). Variables are independent.

3. Data

For this thesis, a time series analysis is conducted to explain the relationship between housing prices and stock prices in the Netherlands. There are three control variables used. Besides housing prices (HPI) and stock prices (SPI), there are three variables used that also influence this relationship. The real gross domestic product (RGDP) of the Netherlands, the consumer price index (CPI), and lastly, the long term ten years bond yield (Interest). These variables were chosen that are in line with prior research, such as Quan & Titman (1999). In this part, there is first an explanation about the data that is used for statistical analysis. There is a summary statistic table conducted to examine the data. Furthermore, there is an overview of the methods that are used to investigate this property.

Housing Price Index (HPI)

The first variable used in this paper is the housing price index of the Netherlands, which is made by a collaboration with the central bureau of statistics of the Netherlands (CBS) and Kadaster. The method that is used to calculate this index is the sale price appraisal ratio method (SPAR) The SPAR is a fixed ratio between the sales prices and appraisals in the base year period (Vries and Haan, 2008). This index contains the selling prices of existing owner-occupied homes, with the base year of 2005. The data consist of a quarterly period from 1996.Q2-2020.Q1. An important note is that this dataset contains only residential real estate. Furthermore, the variable is transformed to the natural logarithm of the variable from now LHPI.

Stock Price Index (SPI)

The stock price variable that is used is the Amsterdam stock exchange index (AEX). This index contains the 25 companies that give an indication of the Dutch economy. The returns of this index will be used to examine the influence of the impact on the housing prices and vice versa.

(10)

10 The data for this index is collected from Factset. Actual quarterly data is used from the same period, as mentioned in 1996Q2-2020.Q1. Furthermore, the variable is transformed to the natural logarithm of the variable from now on LSPI.

Real Gross Domestic Product (RGDP)

The first variable that is used is the real gross domestic product (RGDP) of the Netherlands. The GDP growth rate has a significant impact on housing prices and stock prices (Van den End, Kakes, 2002; Quan and Titman, 1999). The level form of this variable is used and transformed to the natural logarithm, LRGDP. The data is collected from Factset.

Interest

Secondly, there is made use of interest rates. Interest rates are determinants of housing prices and their changes (Sutton, 2002). Interest rates are set by monetary policy and will have an impact on the lending possibilities for households and investors. When there is a low-interest rate, more individuals and investors are able to buy and invest in housing. This will raise the demand for housing and their prices. The interest rate that is used is the ten years bond yield of the Netherlands. The data is collected from the European Central bank. The frequentation of the dataset is monthly; for this paper, the quarterly conversions of the interest rates is used. Interest rates can have negative values; the logarithm is, in that case, not possible. For the interest rate, the level form is used in this analysis.

Consumer Price Index (CPI)

The last variable that is used is the consumer price index (CPI) of the Netherlands. The consumer price index is a measurement for the percentage change in prices of consumer goods. Raymond (2001), found evidence that the expected inflation are determinants of the changes in stock prices. The CPI dataset is from Factset, levels are used, and the quarterly data is conducted. The base year of the index is 2015.

(11)

11

4. Research Methodology

This section will describe in detail the econometric method chosen to analyze the relationship between house prices and stock prices. Moreover, it will introduce and describe all methods and statistical tests that will be used throughout this analysis.

4.1 Time Series Econometric Analysis

The goal of this paper is to analyze the relationship between house prices and stock prices. Therefore, there will be made using one of the most popular econometric models. The Unrestricted Vector Autoregression (UVAR), used to capture the linear dependencies among the multiple time series. UVAR is a stochastic process model describing the evolution of a set of endogenous variables as a linear function of solely their past values, over the same sample period (Stock & Watson, 2015). This method is the generalization of the univariate Autoregressive (AR) model by allowing multiple evolving variables. More specifically, in a VAR model, lagged values of the variables are used to forecast future values of those variables. The UVAR model that will help to investigate the relationship between house prices and stock prices, both stochastic variables evolving over time and endogenous, can be described in the following way: 𝐋𝐇𝐏𝐈𝐭= 𝛼 + ∑ 𝛽𝑡−𝑥 𝜌1 𝑥=1 𝐋𝐒𝐏𝐈𝑡−𝑥+ ∑ 𝛾𝑡−𝑥 𝜌2 𝑥=1 𝐋𝐑𝐆𝐃𝐏𝑡−𝑥+ ∑ 𝜃𝑡−𝑥 𝜌3 𝑥=1 𝐋𝐂𝐏𝐈𝑡−𝑥+ ∑ 𝜂𝑡−𝑥 𝜌4 𝑥=1 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝑡−𝑥+ 𝜀1𝑡 (𝟏) 𝐋𝐒𝐏𝐈t= δ + ∑ 𝜙𝑡−𝑥 𝜎1 𝑥=1 𝐋𝐇𝐏𝐈𝑡−𝑥+ ∑ 𝜋𝑡−𝑥 𝜎2 𝑥=1 𝐋𝐑𝐆𝐃𝐏𝑡−𝑥 + ∑ 𝜔𝑡−𝑥 𝜎3 𝑥=1 𝐋𝐂𝐏𝐈𝑡−𝑥+ ∑ 𝜇𝑡−𝑥 𝜎4 𝑥=1 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝑡−𝑥+ 𝜀2𝑡 (𝟐)

Where the Equation (1) corresponds to the case where is investigated whether the stock prices have a statistically significant impact on the house prices or not and the Equation (2) corresponds to the case where is investigated whether the house prices have a statistically significant impact on the stock prices or not. In those equations, the HPIt represents the house

prices at time period t, the SPIt represents the stock prices at time period t, the RGDPt represents

the Real Gross Domestic Product at time t, and the Interestt represents the interest rate at time

period t. Moreover, the 𝛼 and δ in the equations (1) and (2), respectively, represent the constant terms in those equations. The second term in the Equation (1), the sum over all SPIt-x with its

corresponding coefficient 𝛽𝑡−𝑥 is basically the sum of all possible lagged values of the variable

SPI where the x in t-x represents the xth lag of the variables and as it can be seen from the lower limit of the summation, x = 1, the zero lags cannot be included, and only the higher than

(12)

first-12 order lagged values can be included in that Equation. This holds for all independent variables included in both regressions. Moreover, the 𝜌_𝑖 and 𝜎_𝑖 for i = {4}, the upper limits of the summations represent the maximum amount of lagged values per independent variable that is included in the model. Furthermore, 𝛽_𝑡−𝑥, 𝛾_𝑡−𝑥, 𝜃_𝑡−𝑥, and 𝜂_𝑡−𝑥 Equation (1) represents the coefficients of the xth lagged value of independent variables SPI, RGDP, CPI, and Interest, respectively. Similarly, 𝜙_𝑡−𝑥, 𝜋_𝑡−𝑥, 𝜔_𝑡−𝑥, and 𝜇_𝑡−𝑥 Equation (2) represents the coefficients of the xth lagged value of independent variables SPI, RGDP, CPI, and Interest, respectively. Finally, 𝜀1𝑡 and 𝜀2𝑡 in the equations (1) and (2) represent the error terms of the models where

the house prices HPI and stock prices SPI are used as dependent variables, respectively.

In this research, there will also be an investigation about the possible impact of the RGDP, Interest rate, and CPI on the stock prices and house prices. For this purpose, the following equations are added to the current model:

𝐋𝐑𝐆𝐃𝐏_𝐭= 𝛼₀+ ∑ 𝛽_{0,𝑡−𝑥} 𝜌1 𝑥=1 𝐋𝐒𝐏𝐈_𝑡−𝑥+ ∑ 𝛾_{0,𝑡−𝑥} 𝜌2 𝑥=1 𝐋𝐇𝐏𝐈_𝑡−𝑥+ ∑ 𝜃_{0,𝑡−𝑥} 𝜌3 𝑥=1 𝐋𝐂𝐏𝐈_𝑡−𝑥+ ∑ 𝜂_{0,𝑡−𝑥} 𝜌4 𝑥=1 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭_𝑡−𝑥+ 𝜈_0,𝑡 (𝟑) 𝐋𝐂𝐏𝐈_𝐭= 𝛼₁+ ∑ 𝛽_{1,𝑡−𝑥} 𝜌1 𝑥=1 𝐋𝐒𝐏𝐈_𝑡−𝑥+ ∑ 𝛾_{1,𝑡−𝑥} 𝜌2 𝑥=1 𝐇𝐏𝐈_𝑡−𝑥+ ∑ 𝜃_{1,𝑡−𝑥} 𝜌3 𝑥=1 𝐋𝐑𝐆𝐃𝐏_𝑡−𝑥+ ∑ 𝜂_{1,𝑡−𝑥} 𝜌4 𝑥=1 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭_𝑡−𝑥+ 𝜈_1,𝑡 (𝟒) 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐭= 𝛼2+ ∑ 𝛽2,𝑡−𝑥 𝜌1 𝑥=1 𝐋𝐒𝐏𝐈𝑡−𝑥+ ∑ 𝛾2,𝑡−𝑥 𝜌2 𝑥=1 𝐋𝐇𝐏𝐈𝑡−𝑥+ ∑ 𝜃2,𝑡−𝑥 𝜌3 𝑥=1 𝐋𝐑𝐆𝐃𝐏𝑡−𝑥+ ∑ 𝜂2,𝑡−𝑥 𝜌4 𝑥=1 𝐋𝐂𝐏𝐈𝑡−𝑥+ 𝜈2,𝑡 (𝟓)

Where the 𝛼₀, 𝛼₁, and 𝛼₂ refer to the constant terms. 𝛽_{𝑘,𝑡−𝑥} , 𝛾_{𝑘,𝑡−𝑥} , 𝜃_{𝑘,𝑡−𝑥} and 𝜂_{𝑘,𝑡−𝑥} for k= {0,1,2} represent the coefficients of the xth lagged value of the corresponding independent variables at time period t. Moreover, 𝜌_𝑖 for i = {4} has the same interpretation as in case of the earlier equations. Finally, 𝜈0,𝑡 , 𝜈1,𝑡 , and 𝜈2,𝑡 represent the error terms of the models described

by the equations (3), (4), and (5), respectively. One of the main assumptions of this model is that all involved variables should be non-stationary, namely the statistical properties of these variables, such as the mean, and the variance are all not constant over time. If the statistical test checking for the non-stationarity of the variables involved in the regressions shows that any of the variables are non-stationary, then there is need to use the first differences of the variables and transform the earlier equations to the following equations:

∆𝐇𝐏𝐈𝐭= 𝛼 + ∑ 𝛽𝑡−𝑥 𝜌1 𝑥=1 ∆𝐒𝐏𝐈𝑡−𝑥+ ∑ 𝛾𝑡−𝑥 𝜌2 𝑥=1 ∆𝐑𝐆𝐃𝐏𝑡−𝑥+ ∑ 𝜃𝑡−𝑥 𝜌3 𝑥=1 ∆𝐂𝐏𝐈𝑡−𝑥+ ∑ 𝜂𝑡−𝑥 𝜌4 𝑥=1 ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝑡−𝑥+ 𝜀1𝑡 (𝟔)

(13)

13 ∆𝐒𝐏𝐈_t= δ + ∑ 𝜙_𝑡−𝑥 𝜎1 𝑥=1 ∆𝐇𝐏𝐈_𝑡−𝑥+ ∑ 𝜋_𝑡−𝑥 𝜎2 𝑥=1 ∆𝐑𝐆𝐃𝐏_𝑡−𝑥 + ∑ 𝜔_𝑡−𝑥 𝜎3 𝑥=1 ∆𝐂𝐏𝐈_𝑡−𝑥+ ∑ 𝜇_𝑡−𝑥 𝜎4 𝑥=1 ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭_𝑡−𝑥+ 𝜀_2𝑡 (𝟕) ∆𝐑𝐆𝐃𝐏𝐭= 𝛼0+∑𝛽0,𝑡−𝑥 𝜌₁ 𝑥=1 ∆𝐒𝐏𝐈𝑡−𝑥+∑𝛾0,𝑡−𝑥 𝜌₂ 𝑥=1 ∆𝐇𝐏𝐈𝑡−𝑥+ ∑𝜃0,𝑡−𝑥 𝜌₃ 𝑥=1 ∆𝐂𝐏𝐈𝑡−𝑥+ ∑𝜂0,𝑡−𝑥 𝜌₄ 𝑥=1 ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝑡−𝑥+ 𝜈0,𝑡 (𝟖) ∆𝐂𝐏𝐈𝐭= 𝛼1+ ∑ 𝛽1,𝑡−𝑥 𝜌1 𝑥=1 ∆𝐒𝐏𝐈𝑡−𝑥+ ∑ 𝛾1,𝑡−𝑥 𝜌2 𝑥=1 ∆𝐇𝐏𝐈𝑡−𝑥+ ∑ 𝜃1,𝑡−𝑥 𝜌3 𝑥=1 ∆𝐑𝐆𝐃𝐏𝑡−𝑥+ ∑ 𝜂1,𝑡−𝑥 𝜌4 𝑥=1 ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝑡−𝑥+ 𝜈1,𝑡 (𝟗) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐭= 𝛼2+ ∑ 𝛽2,𝑡−𝑥 𝜌1 𝑥=1 ∆𝐒𝐏𝐈𝑡−𝑥+ ∑ 𝛾2,𝑡−𝑥 𝜌2 𝑥=1 ∆𝐇𝐏𝐈𝑡−𝑥+ ∑ 𝜃2,𝑡−𝑥 𝜌3 𝑥=1 ∆𝐑𝐆𝐃𝐏𝑡−𝑥+ ∑ 𝜂2,𝑡−𝑥 𝜌4 𝑥=1 ∆𝐂𝐏𝐈𝑡−𝑥+ 𝜈2,𝑡 (𝟏𝟎)

where the ∆ represents the first difference of the variable, for instance ∆𝐇𝐏𝐈_𝐭= 𝐇𝐏𝐈_𝐭− 𝐇𝐏𝐈_𝐭−𝟏 and ∆𝐇𝐏𝐈_𝐭−𝟏= 𝐇𝐏𝐈_𝐭−𝟏− 𝐇𝐏𝐈_𝐭−𝟐.

4.2 Stationarity Test

In order to check whether the variables involved in this analysis are stationary or there is non-stationarity in the dataset, there will be made using one of the most popular non-stationarity tests, the Augmented Dickey-Fuller (ADF) test. Moreover, for the ADF test for the variables SPI, HPI, CPI, RGDP, the log levels are used. For the variable Interest, the level form is used. Given that in the regression analysis, the log-levels of the variables should be included. Furthermore, there is also needed to check whether those first differences are stationary. Those should also be based on the log-level of the variables. Under the null hypothesis of the test is stated that a unit root is present in that time series sample, whereas under the alternative hypothesis is stated that there is no unit root, there is stationarity, in that time series sample, that is:

{ H0: there is a unit root H₁: there is a stationarity

The idea behind the test is that if the time series is a unit root process then the lagged value of the series yt-1, first lag level of the variable, will not provide any relevant information when

predicting the change in the yt with the exception of the case when the first difference is used

(Dickey & Fuller, 1979).Therefore, if one is unable to reject the null hypothesis of the ADF test, this suggests that the corresponding variable contains unit root process, which means that the variable is non-stationary. Meanwhile, if one is able to reject the null hypothesis of the ADF test, this suggests that the corresponding time series process is not a unit root process, which suggests that the alternative is true; that is, the variable is stationary.

(14)

14

4.3 Granger Causality Test

In order to identify the exact direction of the causality between the target variables, the Granger causality test will be executed. The Granger causality test is a statistical test for determining whether one sample of time series is useful in predicting another sample of time series, often also described as the test finding the predictive causality. The idea behind the test is that if one can show through a series of t-tests and F-tests on a lagged values of X and with lagged values of Y, that those X values supply statistically significant information about the future values of Y, then in order the time series X is Granger-cause Y (Granger, 1980).

4.4 Lag Order Selection

In order to determine the optimal number of lags that are needed to include in the time series analysis, the information criterion is conducted. Various orders of lagged values of the set of variables of the model are set and compared with the amount of information that is gained by including them in the model. There must be made a tradeoff between the amount of information that is gained by included each lag of the variable versus the amount of predictive power, forecasting accuracy that is lost when doing so. However, including too few lags can decrease forecasting accuracy, because valuable information is lost while including too many lags increases estimation uncertainty in the model. In order to decide the optimal number of lags that are needed to include in the analysis, the Akaike Information Criterion (AIC) will be used, which is a popular measure when investigating the amount of information gained when changing the settings of the model. Stock and Watson (2015) state that the use of the AIC is recommended instead of the BIC. They mention that having more lags over fewer lags are preferred. The formula for the AIC:

𝐴𝐼𝐶 (𝑘) = 𝑙𝑛 [𝑆𝑆𝑅(𝑘)

𝑇 ] + 𝑘 ∗ 2 𝑇

Furthermore, there is a set of linear regressions used in order to analyze the impact over different lags of the house price variable HPI on the stock price variable SPI. Moreover, there is another set of linear regressions used to analyze the different impact lags of the stock price variable SPI on the stock price variable HPI. The following two equations summarize those regressions:

∆𝐋𝐒𝐏𝐈𝐭= 𝛿0+ ∑𝜌𝑥=1𝜔0,𝑡−𝑥∆𝐋𝐇𝐏𝐈𝑡−𝑥+ 𝜈0,𝑡 (𝟏𝟏) ∆𝐋𝐇𝐏𝐈𝐭= 𝛿1+∑𝜌𝑥=1𝜔1,𝑡−𝑥∆𝐋𝐒𝐏𝐈𝑡−𝑥+ 𝜈1,𝑡 (𝟏𝟐)

(15)

15 where the 𝛿₀ and 𝛿₁ represent the constant terms in regressions corresponding to the equations (11) and (12), respectively. Moreover, the 𝜔0,𝑡−𝑥 and 𝜔1,𝑡−𝑥 represent the coefficients of the

xth lagged value of the variables HPI and SPI, respectively, at time period t.Furthermore, the 𝜌 refers to the maximum order of lagged values included in the model. Finally, 𝜈_0,𝑡 and 𝜈_1,𝑡 represent the error terms of the regressions described in equations (11) and (12), respectively.

(16)

16

5. Results

5.1 Descriptives Returns

Table 1 presents the descriptive statistics of the variables included in the model in their log-level forms, given that the models that are used in this analysis and tests can be conducted only on the log-level forms of the variables. For Interest, there is chosen to not take the log-level due to that Interest had no trend. The observation is that the average LSPI is 6.035, and the average LHPI is 4.566. Moreover, the average LRGDP in the data is 5.069, and the average LCPI is 4.474. Furthermore, from this table, there is also an observation that the average Interest is 3.125. Finally, there are no missing observations for any of the variables. Which is needed for time series analysis.

LSPI LRGDP LHPI LCPI Interest

Mean 6.035 5.069 4.566 4.474 3.215 Median 6.055 5.106 4.636 4.488 3.770 Maximum 6.510 5.250 4.924 4.673 6.390 Minimum 5.397 4.793 3.837 4.233 -0.380 Std. Dev 0.262 0.112 0.258 0.128 1.847 Skewness -0.233 -0.614 -1.375 -0.305 -0.365 Kurtosis 2.241 2.762 4.048 1.940 1.921 Jarque-Bera 5.392 5.811 19.802 15.039 16.525 Probability 0.067 0.055 0.000 0.001 0.000 Observations 96 96 96 96 96

Notes: (1). Summary statistics of the logarithm of Stock prices (SPI), Real Gross Domestic product,

House prices index (HPI), Consumer Price Index (CPI). Interest rate is the long-term bond yield of the Netherlands with a maturity of 10 years. All variables are quarterly data for the period 1996 – 2020.

Table 1: Descriptive returns

5.2 Graphical Analysis

Figure 1 plots the evolvement of levels of the variables included in all equations introduced in Section 4 over time. From the line corresponding to the variable Interest, the observation is that during the sample period 1996 - 2020, the interest rate has no trend. Moreover, from the line corresponding to the variable SPI, the observation is that during the sample period 1996 - 2020, the stock prices have been on and off increasing then decreasing. Finally, from the lines corresponding to the variable HPI, RGDP, and CPI, the observation is that during the sample period 1996 - 2020, the house prices, the CPI, and the real GDP have been experiencing increase but very slowly.

(17)

17 Graph 1: The plot describing how variables evolve over time

5.3 Stationarity Test Results

In Section 4 is mentioned that in order to perform UVAR estimation, there is a need to check whether the variables included in the models are non-stationary. Table 2 presents the results corresponding. To the ADF test, there are used both the variables and their first differences. As mentioned before, for the test is made use of the log-level forms of the variables and the first differences of the log-levels given that those are the variables that are needed to use in the UVAR analysis. The ADF test is conducted with 1 lag for all variables. Looking at the table, the observation is that for most of the variables such as LSPI, LHPI, ∆LHPI, LCPI, and Interest, the null hypothesis of the Dickey-Fuller test cannot be rejected. This suggests that these variables follow a unit root process, so they are not stationary. Moreover, from Table 2, there is also an observation that for the variables ∆LSPI, LRGDP, ∆LRGDP, ∆LCPI, and ∆Interest the null hypothesis of the Dickey-Fuller test can be rejected, suggesting that these variables do not follow unit root process, so they are stationary. These results suggest that it is needed to use Equation (5) – (10) from Section 4.1 for analyzing the relationship between the house prices, the stock prices, and the rest of the variables in the sample.

(18)

18

Test Variable ADF Intercept Intercept, trend

LSPI 0.778 0.781 ∆LSPI 0.006*** 0.173*** LRGDP 0.835* 0.253 ∆LRGDP 0.002*** 0.004*** LHPI 0.045 0.060 ∆LHPI 0.001 0.000 LCPI 0.043 0.238 ∆LCPI 0.006*** 0.008*** Interest -0.005 0.830 ∆Interest -0.061*** -0.061***

Notes: (1) The ***, **, * next to the intercept values describe the significance level at which the null hypothesis,

that the corresponding variable contains a unit root (is stationary) can be rejected, namely 1%, 5%, and 10% levels, respectively. (2) The ∆ refers to the first difference and their values. ADF test is conducted with order 1

for all variables.

Table 2: Augmented Dickey-Fuller (ADF) test results

5.4 Optimal Amount of Lags Results

First, the Akaike information criterion is calculated. For the regression where the differences of the log-levels with up to 4th order of lag values of the independent variables are included. The observation is that for the cases when there is included up to 4 lagged values of the variables, the corresponding AIC values suggest that no significant amount of information is being added to the model. Whereas in case 4 lagged values of the variables are being added to the model, a significant amount of information is being added to the model, and the AIC value is significant at a 10% significance level. These results suggest that the optimal amount of lagged values that should be included in the model for all variables is equal to 4.

As was mentioned in Section 4.4, there is also set of linear regressions used in order to analyze the impact of different lags of the house price variable HPI on the stock price variable SPI and the impact different lags of the stock price variable SPI on the house price variable HPI represented by the equations (11) and (12), respectively. It is worth mentioning that the number of lags included in the model per variable, namely 4 lagged values per variable, is based on the results of Section 5.4 where the optimal number of lagged values was determined.

Table 3 presents the results corresponding to the former case, where the impact of different lags of the house price variable HPI on the stock price variable SPI is analyzed, corresponding to

(19)

19 the Equation (11). In each case, corresponding to each Column, different lagged values of the first difference of log HPI have been used. In Column (4) the observation is, which corresponds to the regression where the ∆LSPI has been used as the dependent variable, and 4 lagged values of ∆LHPI have been used as independent variables, that is ∆LHPI1, ∆LHPI2, ∆LHPI3 and

∆LHPI4, the coefficient corresponding to the ∆LHPI4 is statistically significant at 5%

significance level. This suggests that the fourth lag of difference of the house price variable HPI has a positive and statistically significant impact on the stock price variable SPI. From Column (8), which corresponds to the regression where the ∆LSPI has been used as the dependent variable, and 4 lagged values of the variables ∆LHPI and ∆LSPI have been used as independent variables, that is ∆LHPI1, ∆LHPI2, ∆LHPI3, ∆LHPI4, ∆LSPI1, ∆LSPI2, ∆LSPI3, and∆LSPI4, the

coefficient corresponding to the ∆LHPI4 is statistically significant at 5% significance level as

in the previous case. This finding suggests that the fourth lag of difference of the house price variable HPI, as in the previous case, has a positive and statistically significant impact on the stock price variable SPI.

Table 4 presents the results corresponding to the former case, where the impact of different lags of the stock price variable SPI on the house price variable HPI is analyzed, corresponding to the Equation (12). From Column (1) which corresponds to the regression where the ∆LHPI has been used as the dependent variable, and 1st lagged value of ∆LSPI has been used as an independent variable, the observation is that the coefficient is statistically significant at 10% significance level suggesting that the first lagged value of the first difference of SPI has a statistically significant impact on ∆LHPI. Moreover, from Column (5) which corresponds to the regression where the ∆LHPI has been used as the dependent variable and 1st_{lagged value}

of ∆LSPI, 1st_{lagged value of}_{∆LHPI have been used as the independent variable, the}

observation is that the coefficients corresponding to both variables are statistically significant at 1% significance level, suggesting that the first lagged value of the first difference of SPI and HPI have a statistically significant impact on ∆LHPI. Furthermore, from three columns of Table 4, columns (6), (7), and (8), the observation is that the coefficients corresponding to the ∆LSPI1,

∆LSPI2, ∆LHPI1, and ∆LHPI2 first and second difference variables of house price and stock

price variables, respectively, are statistically significant. These results suggest that the first and second lagged values of the difference variables of SPI and HPI have a statistically significant impact on HPI.

(20)

20 Notes: (1) The ***, **, * next to the coefficients of the regressions describes the significance level at which the

null hypothesis, that the coefficient is equal to 0, can be rejected, at 1%, 5%, and 10% levels, respectively.

Table 3: Regression Analysis Results ∆LSPI ∆LSPI (1) ∆LSPI (2) ∆LSPI (3) ∆LSPI (4) ∆LSPI (5) ∆LSPI (6) ∆LSPI (7) ∆LSPI (8) ∆LHPI1 -0.211 0.309 0.017 0.266 -0.211 0.171 -0.620 0.529 (0.72) (1.37) (1.60) (1.65) (0.72) (1.39) (1.69) (1.68) ∆LHPI2 -0.694 -0.971 0.602 -0.576 -1.121 1.055 (1.22) (1.38) (1.36) (1.31) (1.42) (1.36) ∆LHPI3 0.645 2.142 1.328 2.381 (1.30) (1.57) (1.30) (1.49) ∆LHPI4 3.769** 4.575** (1.73) (1.77) ∆LSPI1 0.001 -0.014 -0.014 0.018 (0.14) (0.14) (0.14) (0.14) ∆LSPI2 0.027 0.051 0.004 (0.14) (0.14) (0.13) ∆LSPI3 0.098 -0.021 (0.15) (0.14) ∆LSPI4 -0.129 (0.12) Constant 0.004 0.005 0.005 0.008 0.004 0.005 0.005 0.007 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) R2 _0.001 _0.004 _0.006 _0.076 _0.001 _0.005 _0.015 _0.090 F 0.085 0.286 0.213 1.375 0.044 0.172 0.272 1.008 N 91 90 89 88 91 90 89 88

(21)

null hypothesis, that the coefficient is equal to 0 can be rejected, at 1%, 5%, and 10% levels, respectively.

Table 4: Regression Analysis Results ∆LHPI ∆LHPI (1) ∆LHPI (2) ∆LHPI (3) ∆LHPI (4) ∆LHPI (5) ∆LHPI (6) ∆LHPI (7) ∆LHPI (8) ∆LSPI1 0.024* 0.022 0.021 0.020 0.028*** 0.027*** 0.027*** 0.027*** (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) ∆LSPI2 0.024 0.023 0.021 0.017** 0.020*** 0.021*** (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) ∆LSPI3 0.028 0.027 0.009 0.009 (0.02) (0.02) (0.01) (0.01) ∆LSPI4 0.022 -0.001 (0.02) (0.01) ∆LHPI1 0.840*** 0.364*** 0.300** 0.294* (0.08) (0.11) (0.15) (0.16) ∆LHPI2 0.542*** 0.506*** 0.506*** (0.08) (0.09) (0.08) ∆LHPI3 0.102 0.107 (0.13) (0.16) ∆LHPI4 0.006 (0.14) Constant 0.011*** 0.010*** 0.010*** 0.010*** 0.001 0.001 0.001 0.001 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) R2 _0.030 _0.053 _0.089 _0.109 _0.741 _0.817 _0.826 _0.825 F 2.830 2.820 3.056 3.088 61.676 57.512 49.933 38.960 N 91 90 89 88 91 90 89 88

(22)

22

5.5 Regression results included additional variables

Tables 5 and 6 presents the results of another set of regression corresponding to the equations (11) and (12). Table 5 shows the comparison of two applications of Equation (11). Column (1) of Table 5 presents the results of the regression. The dependent variable is the difference of log stock prices, ∆LSPI, and as independent variables, 4 lagged values of ∆LSPI and ∆LHPI. Whereas, Column (2) of Table 5 presents the results of the regression where again the dependent variable the difference of log stock prices is used, ∆LSPI, and as independent variables not only the 4 lagged values of ∆LSPI and ∆LHPI but also 4 lagged values of the remaining variables included in the sample, the ∆LRGDP, ∆LCPI, and ∆Interest. From Column (1) the observation is that the coefficients corresponding to all lagged values of the difference of stock prices are insignificant and only the coefficients corresponding to the 3rd and 4th lagged values of difference of house price variable, ∆LHPI3, and ∆LHPI4, have statistically significant

coefficients at 10% significance level. These results suggest that the 3rd_{and 4}th_{lagged values}

of difference of house price variable have a statistically significant impact on determining the current difference value of the stock prices. From Column (2), in which the remaining variables with their 4 different lagged values are added to the model represented by equation (11), the observation is that the coefficients of the variables ∆LHPI3, ∆LRGDP2, ∆LRGDP3, and ∆LCPI1

are statistically significant. The results suggest that the 3rd lagged value of the difference of house prices has a statistically significant positive impact on the current difference value of the stock prices. Moreover, the results suggest that the 2nd_{and 3}rd_{lagged values of the difference}

of RGDP have a statistically significant and negative impact on the difference value of the stock prices. Finally, from those results, the conclusion is that the first lagged value of the difference of CPI has a statistically significant and negative impact on the difference in the stock price variable. When comparing the results of these two columns, the observation is that most of the coefficients of variables that were insignificant in Column (2) are also insignificant in Column (1), with the exception of the coefficient of ∆LHPI4 that becomes insignificant. Moreover, the

coefficients of some of the additional variables are significant, suggesting that adding this variable has a statistical justification. This observation is also verified by the comparison of the R2 values corresponding to the basic model, from Column (1), and the extended model, from column (2), where the observation is that the basic model explains the 5.5% of the variation in the target variable while the extended model explains the 22.8% of the variation in the target variable.

(23)

23 Table 6 shows the comparison of two applications of Equation (12). Column (1) of Table 6 presents the results of the regression where the dependent variable, the difference of log stock prices is, ∆LHPI, and as independent variables, 4 lagged values of ∆LSPI and ∆LHPI. Whereas, Column (2) of Table 6 presents the results of the regression where again the dependent variable the difference of log stock prices is, ∆LHPI, and as independent variables not only the 4 lagged values of ∆LSPI and ∆LHPI but also 4 lagged values of the remaining variables included in the sample, the ∆LRGDP, ∆LCPI, and ∆Interest. From Column (1) the observation is that the first and second lagged values of both house price and stock price variables have a statistically significant impact on the ∆LHPI. From Column (2) the observation is that the coefficients of variables ∆LHPI3, ∆LRGDP2, ∆LRGDP3, and ∆LCPI1 are statistically significant, suggesting

that they have a significant effect in determining the house price variable.

When comparing the results of these two columns, the observation is that most of the coefficients of variables that were insignificant in Column (2) are also insignificant in Column (1), with the exception of the coefficient of ∆LSPI2 that becomes insignificant. Moreover, the

coefficients of some of the additional variables are significant, suggesting that adding this variable might be useful and improve the statistical power of the model. However, this observation is not verified by the comparison of the R2 values corresponding to the basic model, from Column (1), and the extended model, from column (2), where the observation is that the basic model explains the 8.23% of the variation in the target variable. The extended model explains 8.53% of the variation in the target variable. So, there is a minimal improvement in the performance of the model when these extra variables are added to the model in case of the Equation (12).

(24)

null hypothesis that the coefficient is equal to 0 can be rejected, at 1%, 5%, and 10% levels, respectively.

Table 5: Regression Analysis Results ∆LSPI ∆LSPI (1) ∆LSPI (2) ∆LSPI1 0.030 -0.110 (0.14) (0.13) ∆LSPI2 0.054 0.029 (0.13) (0.18) ∆LSPI3 0.059 0.224 (0.15) (0.19) ∆LSPI4 -0.078 0.031 (0.13) (0.15) ∆LHPI1 -0.102 0.129 (1.77) (1.68) ∆LHPI2 0.416 -0.129 (1.43) (1.86) ∆LHPI3 2.632* 3.698** (1.53) (1.77) ∆LHPI4 -3.235* -1.867 (1.79) (1.56) ∆LRGDP1 1.895 (3.28) ∆LRGDP2 -5.934** (2.90) ∆LRGDP3 -3.197* (1.90) ∆LRGDP4 1.098 (2.64) ∆LCPI1 -5.593** (2.60) ∆LCPI2 -2.665 (2.86) ∆LCPI3 -3.565 (2.93) ∆LCPI4 -2.206 (2.39) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭1 -0.071 (0.06) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭2 -0.028 (0.06) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭3 -0.001 (0.05) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭4 -0.072 (0.05) Constant 0.008 0.066** (0.01) (0.03) R2 _0.055 _0.228 F 0.637 1.470 N 92 92

(25)

null hypothesis, that the coefficient is equal to 0 can be rejected, at 1%, 5%, and 10% levels, respectively.

Table 6: Regression Analysis Results ∆LHPI ∆LHPI (1) ∆LHPI (1) ∆LHPI1 0.320** 0.341** (0.15) (0.17) ∆LHPI2 0.499*** 0.465*** (0.08) (0.09) ∆LHPI3 0.081 0.118 (0.16) (0.18) ∆LHPI4 -0.002 -0.041 (0.13) (0.17) ∆LSPI1 0.025*** 0.017*** (0.01) (0.01) ∆LSPI2 0.017** 0.007 (0.01) (0.01) ∆LSPI3 0.007 -0.002 (0.01) (0.01) ∆LSPI4 -0.002 -0.007 (0.01) (0.01) ∆LRGDP1 0.201 (0.21) ∆LRGDP2 0.221* (0.13) ∆LRGDP3 0.024 (0.14) ∆LRGDP4 -0.179 (0.14) ∆LCPI1 0.270 (0.21) ∆LCPI2 -0.270 (0.18) ∆LCPI3 0.038 (0.18) ∆LCPI4 -0.209 (0.17) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭1 -0.002 (0.00) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭2 -0.003 (0.00) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭3 0.002 (0.00) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭4 -0.006* (0.00) Constant 0.001 0.000 (0.00) (0.00) R2 _0.823 _0.853 F 39.435 23.925 N 92 92

(26)

26

5.6 Unrestricted Vector Autoregressive Results

Table 7, in the appendix, presents the UVAR estimation results using the Ordinary Least Squares (OLS), where the independent endogenous variables of different lagged values of the differences are used. The following variables, LHPI, LSPI, LRGDP, LCPI, Interest, where each of these variables have been used as explained variables. Stated differently, table 7 presents the VAR results where 5 different endogenous variables in the system are defined as functions of various lagged values of all endogenous variables. The effect for the interpretations of explanatory variables on the dependent variables is, on average, ceteris paribus.

Column (1) contains the UVAR estimation results on the difference of the log of HPI, where all the coefficients of the lagged values of the endogenous variables and their corresponding standard errors are reported. Moreover, they also show whether the coefficient is statistically significant or not, and at which statistical level. From this Column, the observation is that the first and the second lags of the difference of LHPI, ∆LHPI1, ∆LHPI2, and ∆LSPI1, ∆LCPI1 have

a significant and positive short-run effect on the difference of HPI. Moreover, from the same Column, the observation is that the coefficients of ∆LCPI2 and ∆Interest4 are statistically

significant and have a short-run effect on ∆LHPI, but the effect is negative in this case.

From Column (2), the observation is that the coefficients corresponding to ∆LHPI3 and ∆LSPI3

are statistically significant and positive, suggesting that those endogenous lagged variables have a positive effect on the ∆LSPI in the short run. Furthermore, the same Column shows that the coefficients corresponding to the lagged variables ∆LRGDP2 and ∆LCPI1 are statistically

significant and negative, suggesting that in the short run, these variables have a negative effect on the ∆LSPI. Additionally, from Column (3), the observation is that the coefficients corresponding to the ∆LHPI1, ∆LHPI3, ∆LSPI1, and ∆LSPI2 are statistically significant and

positive, suggesting that they all have a positive impact on the LRGDP in the short run. Moreover, the coefficient ∆LCPI4 has a statistically significant and negative impact on LRGDP

in the short run. Furthermore, Column (4) shows that the coefficients of ∆LHPI3, ∆LRGDP2,

and ∆LCPI4 are all statistically significant and positive, suggesting that they all have a positive

short-run effect on the difference of LCPI. Moreover, the only negative coefficient corresponds to ∆LHPI1, suggesting that it has a statistically significant negative impact on LCPI in the short

run. Finally, from the last Column, the observation is that the coefficients corresponding to the ∆LSPI1, ∆LCPI1, and∆Interest1 are statistically significant and positive, suggesting that they

(27)

27 have a positive short-run effect on the ∆Interest. Furthermore, the coefficients corresponding to

the ∆LHPI3 and ∆LSPI3 show that they have a statistically negative significant impact on the

(28)

28

5.7 Granger Causality Test

Table 8 presents the results of the Granger non-causality test, where only the first differences of log-level variables have been used, described in the equations (6) – (10) in Section 4.1. The Chi-statistics from the Granger causality test indicates short-run effects. From the first part of the results corresponding to the Equation (6) where the first difference of log HPI has been used as the dependent variable, there is no observation of any statistically significant direct causal relationship the runs from the stock prices, RGDP, CPI, and the interest rate to the house prices. Moreover, from the second part of Table 8 which corresponds to the Granger causality test results where the first difference of log SPI has been used as the dependent variable, the observation is that the Chi-squared statistics is statistically significant at 10% significance level suggesting that there is a statistically significant direct causal relationship the runs from the real GDP (RGDP) to the stock prices.

Furthermore, from the middle part of Table 8, which corresponds to the Granger test results where the first difference of log RGDP has been used as the dependent variable. The observation is that the Chi-squared statistics are statistically significant at a 1% significance level for the variables ∆LHPI and ∆LSPI, suggesting that there is a statistically significant direct causal relationship that runs from the house prices and stock prices to the real GDP, respectively. Moreover, there are also observations from the lower part of Table 8, which corresponds to the Granger test results, where the first difference of log CPI has been used as the dependent variable. Therefore, the Chi-squared statistics are statistically significant at a 10% significance level for the variable ∆LHPI, suggesting that there is a statistically significant direct causal relationship that runs from the house prices to CPI. Finally, from the lowest part of the of Table 8, which corresponds to the Granger test results where the first difference of Interest has been used as the dependent variable, is that the Chi-squared statistics are statistically significant at 5% significance level for the variable SPI, suggesting that there is a statistically significant direct causal relationship the runs from the stock prices to the interest rate.

Lastly, the observation ALL from the first row where dependent variable log HPI is the Chi-square significant at 1%. That is that the variables LSPI, LRGDP, LCPI, and Interest combined Granger cause the housing prices.

(29)

29 In the second row, the observation ALL from where the dependent variable is log SPI is the Chi-squared significant at a 5% level. Meaning that the variables LHPI, LRGDP, LCPI, and Interest combined Granger cause the stock prices.

Dependable

Variable Excluded Chi-sq df Prob. value

∆LHPI ∆LSPI 7.538 4 0.110 ∆LRGDP 5.797 4 0.215 ∆LCPI 7.081 4 0.132 ∆Interest 4.746 4 0.314 ALL 49.862*** 16 0.000 ∆LSPI ∆LHPI 6.400 4 0.171 ∆LRGDP 8.039* 4 0.090 ∆LCPI 6.873 4 0.143 ∆Interest 5.415 4 0.247 ALL 26.404** 16 0.049 ∆LRGDP ∆LHPI 24.232*** 4 0.000 ∆LSPI 20.411*** 4 0.000 ∆LCPI 7.062 4 0.133 ∆Interest 1.280 4 0.865 ALL 65.339*** 16 0.000 ∆LCPI ∆LHPI 8.358* 4 0.079 ∆LSPI 1.276 4 0.865 ∆LRGDP 6.675 4 0.154 ∆Interest 1.505 4 0.826 ALL 24.501* 16 0.079 ∆Interest ∆LHPI 5.682 4 0.224 ∆LSPI 13.245** 4 0.010 ∆LRGDP 2.778 4 0.596 ∆LCPI 7.059 4 0.133 ALL 33.916*** 16 0.006

Notes: (1) The ***, **, * next to the Chi-sq test statistics describe the significance level at which the null

hypothesis can be rejected, namely at 1%, 5%, and 10% levels, respectively.

(30)

30

6. Conclusion

In this thesis, the dynamic relationship between the stock market and the housing market in the Netherlands is analyzed. Therefore, a UVAR model is conducted for analysis using a and set of different regressions to support the choice of model and corresponding variables involved. The variables SPI, HPI, CPI, and RGDP are transformed in their log-level form. For the interest variable, this is not the case because no trend was observed in the interest variable. In the model, the difference of the housing price variable is used, ∆LHPI, and the difference of the stock price variable, ∆LSPI. The additional variables are used in their differences. The difference of interest rate, ∆Interest, the difference of the CPI, ∆LCPI, and the difference of real GDP, ∆LRGDP. These variables are used in their difference form based on the stationarity test results presented in section 5.3. In section 5.4, the conclusions are that adding four lagged values of all included variables will add significant information to the model. Moreover, this section showed also that adding different lagged values of various variables in the datasets improves the predictive power of the model significantly, from 5.5% to 22.8% of improvement when Interest rate, CPI, and RGDP variables are added, when predicting the stock prices. Furthermore, the conclusion is that adding different lagged values of different variables in the datasets does not improve the predictive power of the model significantly, from 8.23% to 8.53% change when the three non-target variables are added, when predicting the house prices.

From the Granger non-causality test, the conclusion is that there is a statistically significant direct causal relationship the runs from the real GDP (RGDP) to the stock prices. Moreover, there is a statistically significant, direct causal relationship that runs from the house prices to the real GDP, and there is a statistically significant direct causal relationship that runs from the stock prices to the real GDP. Furthermore, from the Granger non-causality test, the observation is that there is a statistically significant, direct causal relationship that runs from the house prices to CPI. Additionally, the Granger results showed that there is a statistically significant direct causal relationship that runs from the stock prices to the interest rate. Finally, the Granger causality test showed no statistically significant direct causal relationship running from the house prices to the stock prices and vice versa. Only when ALL variables combined, the variables Granger cause LHPI and LSPI, respectively, at 1% and 5%.

These results can be summarized as follows: Firstly, there exist no short-run relationship among all five variables in this research. Secondly, the RGDP plays a significant role in determining

(31)

31 the changes in the house prices, and the opposite holds as well, that the house prices play a significant role in determining the changes in RGDP. This finding is in line with the findings of Miller et al. (2011) in their research about the relationship between house prices and economic growth. Furthermore, house prices play an important role in determining the changes in CPI and the interest rate. This finding is in line with the finding of Sutton et al. (2017) about the relationship between interest rates and house prices.

Based on those results the conclusion is that the hypothesis 1, 2, and 3 formulated in section 1, are rejected and only the hypothesis 4, stating that no Granger causality exists between stock prices (SPI) and house prices (HPI), can be considered as accepted. However, the variables SPI, RGDP, CPI, and Interest combined Granger causes housing prices. Lastly, the variables HPI, RGDP, CPI, and Interest combined Granger causes the stock prices.

6.1 Limitations

The examination of the relationship between house prices and stock prices is difficult. Various researches tried to analyses this with different statistical methods. The most common is the VAR analysis, Granger causality, and integration analysis. This thesis did not include a co-integration analysis. Furthermore, the problem is that this research cannot be generalized without limitations. The housing market and stock markets are different across borders. Besides that, they are dynamic and change over time. This should also be considered. A solution that is presented by Quan and Titman (1999) is that doing cross-sectional research. Nevertheless, there should be a consensus about the way to study this relationship. Furthermore, this thesis there is chosen to only use data from the Netherlands. From the perspective of data, the real estate data is only residential. In further study’s it could be an option to add commercial real estate as well. Ultimately, the study between the relationship between housing prices and stock prices is incredibly important because it has an immense impact on the wealth of people.

(32)

32

7. References

Batayneh, K. I., Al-Malki, A. M. (2014). Relationship between house prices and stock prices in Arabia: An empirical analysis. International Journal of Economics and Finance vol. 10 (2018).

Centraal Bureau voor de Statistiek. (2019). Welvaart in Nederland 2019. Retrieved from https://longreads.cbs.nl/welvaartinnederland-2019/vermogen-van-huishoudens/

Chen, N. (2001). Asset price fluctuations in Taiwan: evidence from stock and real estate prices from 1973 to 1992. Journal of Asian Economics, 12, 215-232.

Dickey, D., and W. A. Fuller. (1979). “Distribution of the Estimators for Autoregressive Time Series with a Unit Root.” Journal of the American Statistical Association 74: 427–431.

End van den, J., Kakes, J. (2002). De samenhang tussen beurskoersen en hun huizenprijzen. De Nederlandsche Bank, Afdeling monetair en economisch beleid.

Granger, C. W. (1980): “Testing for causality: a personal viewpoint.” Journal of Economic Dynamics and Control 2: pp. 329–352.

Kakes, J., End van den, J., (2004). Do stock prices affect house prices? Evidence for the Netherlands. Applied Economics Letters, 11:12, 741-744.

Kapopoulous, P., Siokis, F. (2005). Stock and Real Estate Prices in Greece: Wealth Versus ‘Credit-Price’ Effect.” Applied Economics Letters 12.2 (2005): 125–128.

Lean, H. H. (2012). Wealth effect or credit-price effect? Evidence from Malaysia. Procedia

Economics and Fiance. 1:259-268.

Lin, P. T., Fuerst, F. (2012). The integration of direct real estate and stock markets in Asia. Working paper series.

Miller, N., Peng, L., Sklarz., M. (2011). House Prices and Economic Growth. Journal of Real Estate Finance. Vol.42.

(33)

33 Parlevliet, J., Kooiman, T. (2015). De vermogensopbouw van huishoudens: is het beleid in balans? De Nederlandsche Bank.

Quan & Titman. (1999). Do real estate prices and stock prices move together? An international analysis, Real Estate Economics, 27, 183-207.

Raymond, Y. C. Tse. (2001). Impact of property prices on stock prices in Hong Kong.

Stock, J., & Watson, M. (2015). Introduction to Econometrics. Boston: Pearson.

Sutton, G. D. (2002) Explaining changes in house prices, BIS Quarterly Review, September, 46–55.

Sutton, G. D., Mihaljek, D., Subelyte., A. (2017). Interest rates and house prices in the United States and around the world. Bank for international settlements.

Vries de P., Haan, de J. (2008). A House Price Index Based on the SPAR Method.

Yuksel, A. (2016). The relationship between stock and real estate prices in Turkey: Evidence around the global financial crisis. Central Bank Review 16:33-40.

(34)

34

8. Appendix

∆LHPI (1) ∆LSPI (2) ∆LRGDP (3) ∆LCPI (4) ∆INTEREST (5) ∆LHPI1 0.341*** 0.129 0.243*** -0.138** 3.576 (0.10) (1.69) (0.08) (0.06) (3.57) ∆LHPI2 0.465*** -0.129 -0.105 -0.044 3.284 (0.11) (1.82) (0.08) (0.06) (3.85) ∆LHPI3 0.118 3.698** 0.153* 0.107* -8.824** (0.11) (1.80) (0.08) (0.06) (3.79) ∆LHPI4 -0.041 -1.867 -0.112 0.055 1.763 (0.11) (1.77) (0.08) (0.06) (3.73) ∆LSPI1 0.017*** -0.110 0.020*** -0.001 0.582** (0.01) (0.11) (0.01) (0.00) (0.23) ∆LSPI2 0.007 0.029 0.010* 0.003 0.417 (0.01) (0.12) (0.01) (0.00) (0.26) ∆LSPI3 -0.002 0.224* 0.007 0.004 -0.555** (0.01) (0.13) (0.01) (0.00) (0.27) ∆LSPI4 -0.007 0.031 0.007 0.001 -0.234 (0.01) (0.13) (0.01) (0.00) (0.28) ∆LRGDP1 0.201 1.895 0.101 0.002 -2.404 (0.15) (2.50) (0.12) (0.08) (5.29) ∆LRGDP2 0.221 -5.934** -0.116 0.208** 4.734 (0.15) (2.49) (0.11) (0.08) (5.26) ∆LRGDP3 0.024 -3.197 -0.104 0.006 4.942 (0.15) (2.48) (0.11) (0.08) (5.24) ∆LRGDP4 -0.179 1.098 0.071 0.027 3.123 (0.13) (2.24) (0.10) (0.08) (4.72) ∆LCPI1 0.270* -5.593** 0.065 -0.011 13.656** (0.15) (2.53) (0.12) (0.09) (5.35) ∆LCPI2 -0.270* -2.665 -0.141 -0.114 -0.781 (0.16) (2.62) (0.12) (0.09) (5.52) ∆LCPI3 0.038 -3.565 -0.124 -0.041 0.743 (0.16) (2.58) (0.12) (0.09) (5.44) ∆LCPI4 -0.209 -2.206 -0.254** 0.623*** 1.315 (0.16) (2.57) (0.12) (0.09) (5.43) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭1 -0.002 -0.071 -0.001 -0.001 0.325*** (0.00) (0.05) (0.00) (0.00) (0.10) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭2 -0.003 -0.028 -0.001 0.000 -0.073 (0.00) (0.05) (0.00) (0.00) (0.11) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭3 0.002 -0.001 0.001 0.000 -0.045 (0.00) (0.05) (0.00) (0.00) (0.11) ∆𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭4 -0.006* -0.072 0.001 0.002 -0.149 (0.00) (0.05) (0.00) (0.00) (0.10) Constant 0.000 0.066** 0.004*** 0.002 -0.178*** (0.00) (0.03) (0.00) (0.00) (0.07)

(35)

null hypothesis, that the coefficient is equal to 0, can be rejected, at 1%, 5%, and 10% levels, respectively.

The Relationship between House Prices and Stock Prices in the Netherlands: An Empirical Analysis

Bachelor Thesis