• No results found

ABSTRACT RUBEN BOS

N/A
N/A
Protected

Academic year: 2021

Share "ABSTRACT RUBEN BOS"

Copied!
33
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

ABSTRACT

We investigate whether individual experiences of returns in the housing market differs the individuals’ willingness to take a direct position within the Dutch real estate assets class by owning a house. Additionally, we investigate if returns on housing influences the individuals’ housing tenure choice. Using data from the Household Survey of the Dutch National Bank from 1994 until 2012, results indicate that individual experienced returns have a relationship with housing tenure choice, but the influence of those returns on the individuals’ housing tenure choice seems to be little. Evidence might indicate that consumption motives exert a greater influence on housing tenure choice of individuals than investment returns do within the Dutch housing market.

Keywords: real estate, behavioral finance, experience effects, homeownership, logistic regression, discriminant analysis

JEL codes: D03, D14, D83, D84, G11, R20, R21

RUBEN BOS

S2188449 January, 2014 Master Thesis Finance

(2)

Introduction 2/33

Index

1 Introduction ... 3

2 Data and methodology ... 7

2.1 Returns based on the experience hypothesis...8

2.2 The arithmetic mean of all available information ...9

2.3 Control variables ... 10

2.4 Summary statistics ... 11

2.5 Methodology ... 12

2.6 Panel regression models ... 13

2.7 Random vs. fixed effects ... 14

2.8 Discriminant Analysis ... 15

3 Results ... 16

3.1 Exclusion of the age variables ... 20

3.2 Stepwise inclusion ... 20

3.3 Discriminant analysis on the explanatory variables ... 23

4 Discussion ... 26

4.1 How our results relate to mean-variance framework ... 28

4.2 An aggregate perspective ... 29

5 Conclusion... 30

6 Acknowledgements ... 31

(3)

Introduction 3/33

1 Introduction

At the end of 2007, the fall of Lehman brothers initiated a financial crisis. Stock prices dropped dramatically, credit supplies dried up, and the economies of the United States and the major part of Europe came to a standstill. The economies of the United States and the major part of Europe experienced a period of economic depression or even recession of which the magnitude is measured as more severe than the Great Depression of the 1930’s. Malmendier and Nagel (2011) suggest that risk attitudes of individuals change with a long lasting effect after the occurrence of a large economic shock, such as the crises of 2007 and 1930’s. They find that people differ in their level of risk taking when they live trough different macroeconomic histories and try to capture the risk taking behavior of individuals by analyzing the willingness to participate in a certain asset class. By personalizing the returns they experienced over their respective lifetimes, it was shown that individuals who have experienced low stock market returns are less likely to participate in the stock market. They invest a lower fraction of their liquid assets in stocks if they participate and are more pessimistic about future stock returns. Similar results are shown for the bond market. Malmendier and Nagel (2011) call this the experience hypothesis.

The results of Malmendier and Nagel (2011) are in counter to the assumptions for standard neo-classical economic modeling which assumes that individuals are endowed with stable risk preferences that are unaltered by economic experiences. Standard models also assume that individuals incorporate all available historical data when forming beliefs about risky outcomes, Malmendier and Nagel (2011). Malmendier and Nagel (2011) show that individuals base their risk preferences on what they experience during their lifetimes, rather than on all the available historical data, when they form their beliefs on risky outcomes. Furthermore, they state that risk preferences are unstable and are altered by economic events. Psychological studies of Nesbitt and Ross (1980), Weber et al. (1993), and Hertwig et al. (2004) support the findings of Malmendier and Nagel. They find that recent personal experiences exert a greater influence on personal decision making, than statistical summarized information does. These findings implicate that individuals with different ages should experience recent returns in different ways, since the level of risk taking is correlated with differences in lifetime experiences, Malmendier and Nagel (2011).

(4)

Introduction 4/33 The results of Malmendier and Nagel connect to the literature of reinforcement learning where it is stated that: ‘a individual’s choice of actions strongly depends on the payoffs they obtain from the same actions in the past, even if circumstances have changed’, Erev and Roth (1998) and Camerer and Ho (1999). For example, Kaustia and Knüpfer (2008) provide evidence that the returns of the investors on initial public offerings (IPOs) is positively correlated with subscriptions of future IPOs. In addition, reinforcement learning and savings literature, Choi et al. (2009), provide evidence that high personally experienced returns lead to an increase in the saving rates of the American pension plans. So if individuals have experienced a positive return on a certain action in the past they are willing to invest more in a particular product or they will execute a certain action at a higher rate than before.

First, I want to examine empirically whether individuals differ in their willingness to take risks in the housing market depending on the return history they have experienced over the course of their lives. Secondly I want to examine if individuals are influenced by returns when making their housing tenure choice. Willingness to take risks in the real estate market is expressed by housing tenure choice, owning or renting a house. Housing tenure choice is defined by Aarland and Norvik (2009) as a snap-shot of a dynamic path of housing consumption at one singly point in time. Literature on tenure choice shows a strong interest in identifying and explaining the variations in relative costs of owning a house versus renting a house, Jones (1995), Aarland and Nordvik (2009). Variations in tenure choice are attributed to the individuals’ preferences and beliefs formation, but also to variation of available opportunities of each individual at each moment of time, like (financial) knowledge, income, human wealth, social environment, and tax non-neutralities, Artle and Varaiya (1978), Follain and Ling (1988), Jones (1995), Hochguertel and Van Soest (2001), Guiso, Sapienza and Zingales (2004; 2008), Alesina and Fuchs-Schündeln (2007), Osili and Paulson (2008) and Aarland and Nordvik (2009). Tenure choices based on preferences and beliefs formation are rooted in cultural values, Aarland and Nordvik (2009). Due to the non-analytical nature of cultural aspects, these will not be included in this thesis. Tenure preferences based on age, human wealth and (financial) knowledge can be examined in a numerical sense if they are treated as variations in efficiency and capacity to operate and maintaining a housing unit, Weiss (1978).

(5)

Introduction 5/33 assuming perfect capital market conditions. Setting a lower bound to the real estate asset class influences the individuals’ interaction on the financial markets with their residual share of capital.

Next to the financial consequences, there are two major risks that are associated with homeownership. The first associated risk is the exposure to house price risk. Sykes and Young (1981) state that purchasing a house restricts the owner-occupier solely to gain returns from capital risk, because the occupied object produces no real rental income. Capital returns emanate from differences between the buying and selling prices. Hochguertel and Van Soest (2001) find that owner-occupiers therefore critically depend on the house price developments. In this regard, the returns for the owner-occupiers heavily depend on the moment of acquisition and liquidation to monetize their capital returns, Kraft and Munk (2011). Hochguertel and Van Soest (2001) state that housing as an investment good shows a strong relationship with house price variations. A second risk that is associated with purchasing a house is the committed expenditure risk, Fratantoni (1998). Committed expenditure risk is risk assumed by commitment of households to make mortgage payments over a long horizon with an uncertain income stream. Therefore, adding a house to the financial portfolio of individuals causes them to show behavioral changes, Fratantoni (1998). One of these changes is temperance, which is reflected by an increase in saving rates as a response to unavoidable risk taking since individuals need to comply to their liabilities. This behavioral change results in a diminished amount of expendable income or liquid assets available for investments, Engelhardt (1994; 1996) and Scheiner (1995). Furthermore, purchasing a house increases the holding of safe financial assets in response to unavoidable risks, Guiso et al. (1996) and Fratatoni (1998).

(6)

Introduction 6/33 individuals are negatively influenced by returns they experience in the housing market, since the house price development shows a bullish trend while the ownership proportion of housing stock distribution remains stable.

Figure 1: Construction Input Price Index (CIPI) from 1994 to 2012 is plotted against the Housing Stock distribution

For each year t we calculated the proportions of homeownership and renting of the total housing stock distribution from our own sample. We added the Construction Input Price Index (CIPI) into this picture as an instrument for house price development, in order to visualize the pattern between house prices and our own housing stock distribution. Our sampling period is from 1994 to 2012.

The aim of this study is first to continue on the work of Malmendier and Nagel (2011) by examining empirically whether individuals differ in their willingness to take risks in the Dutch housing market depending on the return history they experienced over the course of their lives. Secondly, we want to analyze if returns on housing exert an influence on the housing tenure choice of individuals. Here we study the experienced returns on a housing product, which is different in its nature compared to the securities used in the work of Malmendier and Nagel (2011). Sykes and Young (1981) state that the price of an investment product can be seen as the present value of all future expected income including growth expectations of that investment product. Acquisition of securities is therefore considered to be based on investment motives. Acquisition of a house is driven by dual motives, namely consumption and investment motives, Ranney (1981), Schwab (1982) and Brueckner (1997). Consumption motives are motives that fulfill an immediate need, such as hunger, thirst, transport, etc. The price of a consumption good is based on what the consumers are willing to pay for the fulfillment of their needs. The price of a house consists of two elements, the present value of future indirect rental income and the additional element that individuals are willing to pay for their consumption motives, like status, housing, flexibility, and location, Sykes and Young (1981) and Aarland and Norvik (2009). The dual price elements for a real estate object makes it interesting to

15 63 64 63 63 34 45 45 51 58 56 58 56 59 58 55 60 60 58 85 37 36 37 37 66 55 55 49 42 44 42 44 41 42 45 40 40 42 0 20 40 60 80 100 120 140 160 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 In d ex Year

(7)

Data and methodology 7/33 investigate the experience hypothesis of Malmendier and Nagel (2011) and important to analyze the influence of the returns on the housing tenure choice of individuals.

.In this thesis we will use the household surveys of the Dutch Central Bank to gain insights about which determinants influence the housing tenure choice. We will construct returns related variables to test if the willingness of individuals to take risks in the real estate market, expressed by ownership of a house, is affected by the experienced returns over their life course. We test the differences in experience by using a sophisticated weighting model for recent and distant experienced returns and controlling for a wide range of household characteristics and wealth controls. Supported by the existing literature on the housing tenure choice and the results of Malmendier and Nagel (2011), where they show that past experienced returns affect the willingness to take financial risks in securities, we will investigate if the experience hypothesis is applicable for the real estate asset class. Deriving from the literature we know that the housing tenure choice is based on a combination of investment and consumption motives. Therefore we analyze if the housing tenure choice of individuals is influenced by the returns which may arise from ownership.

The remainder of this thesis is organized as follows. Section I defines the control variables and the variables we use to investigate the individual experiences on returns. Section II covers the methodology of our regression models and the discriminant analysis. The results of the regression models and the discriminant analysis are shown in Section IV and will be discussed in Section V. We will conclude the thesis in Section VI.

2 Data and methodology

In this section, we define our dependent variable, independent variable of interest and control variables which we are going to use to investigate the research questions. Our source of data is the DNB Household survey, which provides psychological and economical data on household level. The data is collected over 19 annual waves starting from 1994 until 2012. Each wave collects data on 2000 households who participate in the CentERpanel. The CentERpanel is an internet panel that reflects the composition of the Dutch-speaking population. Our sample covers information of 21,181 unique households in total.

(8)

Data and methodology 8/33 homeowners includes households who bought a house, those who built one and the ones who inherited their house from their parents. The rest of the households are considered to be renters.

2.1 Returns based on the experience hypothesis

Our independent variable of interest is the experienced return variable that includes the limitation of information based on their life spans and the dynamic ability to alter experienced returns by assigning different weights to experienced returns. Panel data from the DNB Household survey is used to identify the individuals ages, and life spans accordingly. Following the findings of Malmendier and Nagel (2011) stating that individuals are influenced by experiences that occurred during their lives, we allow therefore our weighting of returns to be dynamic. We use a sophisticated weighted scheme, also used by Malmendier and Nagel (2011), to calculate the experienced returns of the households. The weighting scheme multiplies the annual returns with the different weights based on the individuals’ life span. Time series data on the Dutch real estate market need to date back to the year of birth of the oldest head of the household in the sample, which is 89 in 2003, meaning that the relative return history needs to begin at 1914. The Central Bureau of Statistics provides a Construction Input Price Index (CIPI) which runs back to 1914. This index is used in similar studies on Dutch real estate returns, like Elsinga et al. (2011), Vandevyvere and Zenthöfer (2012) and Bartelmans et al. (2012).

Malmendier and Nagel (2011) use a parsimonious specification of weights that allows them to decline, to be constant and to increase over time by inclusion of one additional parameter, . In this thesis we model for declining, constant, or increasing weights of returns, by including for increasing weights that follow a convex structure, for holding the weights constant and for a declining convex structure. For each household i in year t, the weighted average of past returns is calculated by,

where (1)

Where is the return in year . The weight depends on k, age, l, the second item

(9)

Data and methodology 9/33 19 returns, since is 20 years in 2012 minus one. The first factor to be summarized will be k=1, so are the returns of 2011. The return of 2011 will be given a weight of

divided by the sum of all the weights of (191.5 + 181.5 + 171.5+…..+11.5). Next, the returns of 2010 will be given a weight of divided by the sum of all the weights of (191.5

+ 181.5+…..+11.5) and so on. Graphically the weighting function can be displayed like Figure 2.

Using the weighting scheme of Malmendier and Nagel (2011) solves two issues. First, a separate inclusion of all the experienced returns would give a large number of included variables, which would make it impossible to estimate the model with any meaningful precision, Malmendier and Nagel (2011). And second, the number of explanatory variables would differ across households depending on their ages which makes it a daunting task to compare the estimated results of all individuals, Malmendier and Nagel (2011).

Figure 2: Weights on Experienced Returns for different values of for a 20-year-old Household

Head

For each year t we calculated the respective weights for a 20-year-old household head following the sophisticated weighting function of Malmendier and Nagel (2011). The figure visualizes the weighting patterns of each loading of λ. In this thesis we use the functions that include an increasing convex pattern whereby λ=1.5, a linear pattern whereby λ=0.0 and a decreasing convex pattern whereby λ=-0.5. The line of λ=1 is only inserted to function as a reference point for λ=1.5 in order to distinguish the increasing concave function 0< λ >1 from the increasing convex function whereby λ>1.

2.2 The arithmetic mean of all available information

As a control group for the returns related variables based on the experience hypothesis we calculated the arithmetic mean of all available information on returns. This method follows the assumptions of standard neo-classical economic modeling. We use a linear weighting function for construction of this return variable, since the risk preferences are assumed to be stable over time and

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 We igh t

Year (present year t=2012)

λ= 0 λ= 1 λ= 1.5

(10)

Data and methodology 10/33 unaltered by events. The return variable that incorporates the standard neo-classical economic modeling assumptions, , is calculated by Equation 2.

(2)

Here is the difference in years from our starting point in the time series data until the actual year t. The CIPI starts with the annual return of 1914, so . reflects the annual return of year t.

2.3 Control variables

Tenure choice literature suggests that both economic and demographic factors play an important role in the tenure choice process, Aarland and Norvik (2009) and Kan (2000). We control for both factors in our models. Starting with the economic factors, Aarland and Norvik (2009) state that housing is a good consumed jointly by all of the same households. Therefore housing consumption will depend on the needs and resources of all members. A typical household consists of (an) adult(s) and/or their spouse. Aarland and Norvik (2009) continue with stating that, although they live in the same house, it is assumed that young adults or children have sole control over their resources and consequently that parents do not take their children’s resources into account when making housing choices. Therefore only the incomes of the head household and his or her partner is taken into account when constructing the income variable

(INCOME)

, which reflects the economic situation of the household. Di Salvo and Ermisch (1997) provides extensive information on household demographic characteristics. Age related variables are usually included in tenure choice regressions to control for taste dispersion. We included two age related variables, the linear variable

(Age)

and the squared variable

(Age2)

, into the regression equations. The squared term of age controls for the concave relationship of age and homeownership, Aarland and Norvik (2009). Both Aarland and Norvik (2009) and Di Salvo and Ermisch (1997) find that the composition of the households is an important determinant in the housing tenure choice. So therefore we control for the household composition by including a married dummy

(Married)

and two variables for the number of children in each household. We included a linear continuous variable

(#Childern)

to control for the size effects and a squared term to control for the concave relationship between the number of children and time. Additionally, Di Salvo and Ermisch (1997) also control for human wealth controls by including high school diploma

(HS)

and college degree dummies

(College)

.

(11)

Data and methodology 11/33 debt holdings and amount of liquid assets. Personal debt holdings

(PersonalDebt)

is measured as the sum of private hold loans, extended lines of credit, outstanding debt with mail-order firms, credit card debts, loans of family and friends, debt on hire-purchase contracts and study loans, Fratantoni (1998). For the measure of liquid assets

(LiqAssetsLN)

we use the definition of Malmendier and Nagel (2011). The total amount of liquid assets is the log of the summarized stock and bond holdings, checking and savings accounts, the mutual funds, saving certificates and certificates of deposits.

Table 1

Summary statistics of the dependent, independent and control variables.

We gathered data of 21,281 households over a period of 19 years, starting in 1994 until 2012. We constructed an unbalanced panel whereby households cover a time span of 1 year up until the full 19 years. Our binary dependent variable values 1 for homeownership and 0 for renting. The experienced return variables with , 0.0 and -0.5 is

calculated, as in Malmendier and Nagel (2011) [see Eq. 1]. The all available returns variable is the

arithmetic mean of all available information on house price returns, [see Eq. 2]. The high school dummy and college degree dummy shows if the household head posses a high school diploma and/or a college degree. The married dummy shows if the household head is married. The liquid wealth variable is calculated as the sum of stock and bond holdings, checking and savings accounts, the mutual funds, saving certificates and certificates of deposits, as in Malmendier and Nagel (2011). The personal debt variable follows the definition of Fratantoni (1998) whereby personal debt equals the sum of private hold loans, extended lines of credit, outstanding debt with mail-order firms, credit card debts, loans of family and friends, debt on hire-purchase contracts and study loans. For all variables the number of observations, the mean, standard deviation, minimum and maximum values are listed.

Variable, Number

of obs. Mean

Standard

deviation Min Max Dependent variable

Binary ownership (Wo1) 18032 0.640 0.480 0 1

Independent variables

Experienced returns ( 21281 4.295 0.468 2.500 6.400

Experienced returns ( 21281 5.046 0.566 1.958 6.605

Experienced returns ( 21281 5.288 1.304 0.272 9.565

All available returns (AitALL) 21281 4.642 0.998 4.300 4.800

Control variables

Age 21281 52.314 14.218 18 94

High school diploma dummy 21281 0.549 0.498 0 1

College degree dummy 21281 0.300 0.171 0 1

#Children 21281 1.611 2. 731 0 12 Married Dummy 21281 0.475 0.499 0 1 Income (x1000) 21246 38.527 0.607 0.234 2,874.085 Liquid Assets (x1000) 16048 39.249 115.779 -349.621 3,700.825 Personal Debt (x1000) 19052 -4.229 25.854 -1,422.400 0 2.4 Summary statistics

(12)

Data and methodology 12/33 returns when we follow the dynamic weighting functions, that summarize the experienced returns from 1994 to 2012. Whereby the increasing convex weighting function shows an annual return of 4.295%, with a minimum of 2.500% and a maximum of 6.400%. The linear function shows an annual return of 5.046%, with a 1.958% minimum and a 6.605% maximum return. The decreasing convex weighting function shows an annual return of 5.288%, with a 0.272% minimum and a 9.565% maximum. Our control group, whereby the returns are calculated considering the standard neo-classical economic assumptions, show an annual return of 4.642%, with a minimum of 4.300% and a maximum of 4.800%. In our sample 54.5% of the household heads are listed as having a high school diploma and 30,0% as having a college degree and 47,5% is married. The average household has an age of 52, 1.61 children, an income of 38,640 euro, a liquid wealth of 39.249 euro and a personal debt of 4,229 euro.

2.5 Methodology

(13)

Data and methodology 13/33

2.6 Panel regression models

To investigate if the individuals differ in their housing tenure choice, we will use panel regression models. Since we limited the housing tenure choice to dichotomous outcome, we use binary panel regression models. Utilizing the full information advantage of panel data, we are bound to either logistic or probit panel regressions. In preparation for this thesis both regressions were ran. Since the interpretation of the logistic odds are more easy to understand than those of the probit regressions, the logistic regressions are preferred above the probit ones. In addition, the logistic regression also achieved a smaller negative maximum likelihood than the probit regression in status quo. A model that fits the data well, will have a log likelihood value close to zero. A perfect model would have a likelihood value of zero, a smaller negative number rules in favor of the logistic model, Hosmer and Lemeshow (2013). Logistic panel regressions provide two methodologies to analyze the problem by using either a random effects model or a fixed effects model, whereby the random effects model is more efficient than the fixed procedure, but the fixed is more precise. By executing a Hausman-test we will analyze which one we should use.

For the main estimation, the effect of experienced returns on housing tenure choice, we use the maximum likelihood to estimate following logistic panel regression panel.

(14)

Data and methodology 14/33 The binary indicator equals 1 for homeownership for household i at time t¸ where 0 equals renting. is the variable of interest reflecting the calculated (experienced) returns. The list

summarizes all the control variables, such as households characteristics, income and wealth controls. The end term, , is the new cross-sectional error term containing all the variance that is not explained

by the independent variables.

2.7 Random vs. fixed effects

From the literature we learned that preference heterogeneity is a well-known problem when estimating tenure choice. The greatest concern acknowledged by Kan (2000) and Aarland and Norvik (2009) is that unobserved owning preference may be correlated with one or more explanatory variables, for instance with income or age, and may therefore bias the relevant coefficients. The results of the regression models with random effects shows that rho differs statistically significant from zero (data not shown). When rho statistically differs from zero it means that the variance in the homeownership variable is partially explained by the individuals differences among households. Heterogeneity arises when the individual differences correlate with the explanatory variables. The possible correlation problem can be avoided by using a modeling procedure with fixed effects. In order to determine which effects we should apply, we do the Hausman-test. We use our main estimation of Equation 3 when testing for possible heterogeneity problems. The test shows statistically significant that we should reject the H0 hypothesis, stating that the unobserved individual effects are uncorrelated with the explanatory variables. This result implies that we should use the fixed-effects estimator instead of the random-effects estimator to overcome the detected heterogeneity problems, even though 7908 of our observations are dropped from the sample due to the zero in-between variance of certain variables over time.

Table 2 Hausman-test

The table shows the outcome of the Hausman-test. The Hausman-test tests if unobserved individual effects are uncorrelated with the explanatory variables use in the regression models. The Chi Squared value is one of interest. If the value is statistically significant the hypothesis stating that the unobserved individual effects are uncorrelated with the explanatory variables must be rejected. In case reject the hypothesis heterogeneity problems are detected and the fixed effects estimator must be used. The main estimation model is constructed as in Equation 3 with λ=1.5 is used for the Hausman-test. The stars indicate the significance of the Chi Sqaure score, ** p>0.01.

Dependent: Binary housing tenure choice variable

Equation 3 modeled with Random-effects Equation 3 modeled with Fixed-effects # Observations 11,890 3,982 Maximum Log-likelihood -3800.191 -799.965 # Explanatory variables 12 12 Chi Squared 676.99**

(15)

Data and methodology 15/33

2.8 Discriminant Analysis

By executing logistic panel regressions we are able to analyze if individuals differ in their housing tenure choice based on their experienced returns. In addition, we want to analyze also if the housing tenure choice is actually influenced by the experienced return variable. In order to investigate if our added variable has a significant influence in discriminating the homeowners group from the group who is renting, we execute a multiple discriminant analysis. A multiple discriminant analysis analyzes the individual discriminant power of each of the explanatory variables, a procedure that is recently used in other finance related studies, see Körs et al. (2012) and Pei-Lee (2010). The canonical variant of the discriminant analysis is closely related to the logistic panel regressions, since it tries to find a logistic prediction model via the maximum likelihood procedure. The logistic model prediction model is built in order to successfully predict the membership of each individual in a binary setting. Our equation is defined as;

(4)

Whereby, is the discriminant score, is the coefficient and prefixes all the included variables. The included explanatory variables are equal to the logistic panel regressions as shown in Equation 2. An explanatory variable with a strong discriminant power is able to separate the distinct groups. The analysis standardizes the variables and ranks the variables on their discriminating power. We enter each explanatory variable separate following stepwise order. We impose a requirement for the F-statistic of each variable that is about to enter our prediction model. Since we have in total 11 explanatory variables and more than 120 observations we impose 2.25 as our critical F-value. The discriminant analysis uses the Wilks’ Lamba statistics to analyze the contribution of each explanatory variable in our prediction model.

(16)

Results 16/33

3 Results

Table 3 reports the estimation results from the logistic panel regressions of four models differentiating in the weighting parameter . In model 1 the experienced return variable incorporates an increasing convex weighting function, meaning that recent returns carry a larger weight than the returns experienced earlier in life and the time span is limited by the individuals’ age. The returns are bound by the individuals’ age and with a increasing convex weighting function seem to have a significant effect on housing tenure choice. Therefore an increase in the experienced returns implies an increase in the probability to own a house.

Model 2 of Table 3 shows the results whereby the weighting function is linear, meaning that an individual weights all the returns equally. In this model the experienced return variable has also a positive and statistically significant relationship with housing tenure choice. The control variables have similar relationships as in Model 1. Model 3 shows the estimation results whereby the weighting function is decreasing and convex, meaning that returns that are experienced earlier in life carry a larger weight than those who experienced more recently. The coefficient for the experienced return variable with a decreasing convex weighting function is not statistically significant which implies that individuals do no weigh the returns that they experienced in their early years more heavily than those which occurred more recently. Model 4 of Table 3 shows the estimation results incorporating all the available historical data on housing market returns with equal weights. The returns variable based on all available data with equal weights shows no relationship with housing tenure choice.

(17)

Results 17/33

Table 3

Logistic Panel Regression Results on Housing tenure choice with including all control variables

The table shows the individual coefficients and standard errors (in parentheses) of the explanatory variables used in each model. The four models differentiate in their returns related variables and how they are constructed. Models 1, 2, and 3 follow the sophisticated weighting model of Malmendier and Nagel (2011), [ as in Eq. 1]. Model 4 is our control group, since the returns are calculated by the standard neo-classical economical assumptions [as in Eq.2]. The stars indicate the significance of the Z-scores, * p>0.05 ** p>0.01. Results of our variables of interest are bold.

Dependent: Binary housing tenure choice variable Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Model 1 2 3 4 Weights parameter 1.5 0.0 -0.5 0.0

Return data based on Lifetime Lifetime Lifetime Full history until

1914 Experienced returns 0.897** (0.293) 1.656* (0.705) 1.248 (0.888) 1.276 (0.991) Age 0.709** (0.071) 0.792** (0.078) 0.782** (0.083) 0.744** (0.072) Age2 -0.004** (0.001) -0.004** (0.001) -0.004** (0.001) -0.004** (0.001)

High school dummy 0.098

(0.180) 0.093 (0.180) 0.080 (0.179) 0.068 (0.180) College Dummy -0.531 (0.289) -0.551 (0.288) -0.592* (0.287) -0.618* (0.286) #Children -0.137 (0.091) -0.139 (0.091) -0.140 (0.091) -0.124 (0.091) #Children2 0.042** (0.009) 0.041** (0.009) 0.041** (0.009) 0.039** (0.009) Married Dummy 2.310** (0.181) 2.290** (0.180) 2.260** (0.179) 2.235** (0.178) Log Income -2.296* (1.222) -2.381* (1.219) -2.476* (1.211) -2.403* (1.215)

Log Income Squared 0.145*

(0.063) 0.148* (0.063) 0.152* (0.063) 0.149* (0.063)

Log Liquid Assets 0.037

(18)

Results 18/33

(19)

Results 19/33

Table 4

Logistic Panel Regression Results on Housing tenure choice without the age related variables

The table shows the individual coefficients and standard errors (in parentheses) of the explanatory variables used in each model but without the age related variables. The four model differentiate in their returns related variables and how they are constructed. Models 1, 2, and 3 follow the sophisticated weighting model of Malmendier and Nagel (2011), [ as in Eq. 1]. Model 4 is our control group, since the returns are calculated by the standard neo-classical

economical assumptions [as in Eq.2]. Compared to Table 3 the variables Age and Age2 are excluded. The stars indicate

the significance of the Z-scores, * p>0.05 ** p>0.01. Results of our variables of interest are bold.

Dependent: Binary housing tenure choice variable Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Model 1 2 3 4 Weights parameter 1.5 0.0 -0.5 0.0

Return data based on Lifetime Lifetime Lifetime Full history until

1914 Experienced returns -0.128 (0.216) -4.413** (0.458) -6.678** (0.646) -7.037** (0.727)

High school dummy

0.680** (0.166) 0.478** (0.167) 0.454** (0.169) 0.577** (0.167) College Dummy -1.359** (0.260) -1.263** (0.263) -1.248** (0.265) -1.099** (0.263) #Children -0.361** (0.081) -0.238** (0.081) -0.300** (0.082) -0.333** (0.082) #Children2 0.057** (0.009) 0.044** (0.008) 0.051** (0.009) 0.055** (0.009) Married Dummy 2.293** (0.163) 2.162** (0.165) 2.239** (0.167) 2.376** (0.167) Log Income -4.215** (1.176) -3.770** (1.215) -3.523** (1.197) -3.523** (1.225)

Log Income Squared

0.253** (0.060) 0.227** (0.062) 0.215** (0.061) 0.217** (0.063)

Log Liquid Assets

0.023 (0.035) 0.030 (0.036) 0.035 (0.036) 0.043 (0.036) Personal Debt (std.) -0.062 (0.033) -0.075* (0.035) -0.074* (0.035) -0.070* (0.036) # Observations 3,982 3,982 3,982 3,982 Maximum Log-likelihood -967.053 -914.942 -903.477 -909.511

(20)

Results 20/33

3.1 Exclusion of the age variables

Aarland and Norvik (2009) find important effects when they exclude the age and age2 from their estimations. In their estimations some coefficients changed in direction, while some other coefficients become insignificant, where they were significant before excluding the age factors. In order to investigate the effect of excluding the age related variables on the included explanatory variables in our study, we estimated the same models as in Table 3 but without the age and age2 variables. Table 4 shows the estimation results of the four models without the age and age2 variables. When excluding the age variables we find some interesting results regarding our main variable of interest. The effects of the experienced return variables in Table 4 are the opposite of what we find in Table 3. The increasing convex and linear weighting functions in Model 1 and Model 2 show no relationship with housing tenure choice, while the decreasing convex weighting function is now negatively and statistically significant. Model 4, which includes the full history of the returns with equal weights, also shows a negative and statistically significant relationship.

If we compare the results of Table 4 with the results of Table 3, we see in Table 4 that the possession of a high school diploma and a college degree seems to have positive relationships, which is similar to the relationships in Table 3. The important difference is that the coefficients of both variables are statistically significant when the variables age and age2 are excluded . The coefficient of the linear variable of the number of children also becomes statistically significant, whereby the coefficients of the personal debt variables of model 1 becomes statistically insignificant. The exclusion of the variables age and age2 clearly has a major impact on the coefficients and the standard errors of other variables.

3.2 Stepwise inclusion

(21)

Results 21/33

Table 5

Logistic Panel Regression Results based on stepwise inclusion of control variables

The table reports the individual coefficients and standard errors (in parentheses) of the explanatory variables used in each model. The five models do not differentiate in their returns related variable. The returns related variable is calculated in the sophisticated increasing convex weighting scheme, whereby , as in Malmendier and Nagel (2011) [see Eq. 1]. The five models differentiate in their included control variables. Model 1 shows the univariate model. Model 2 includes the demographic characteristics of the households. Models 3 has additional the income related control variables. Model 4 has additional control variables that control for the economic wealth of the

individual households. Last, Model 5 adds the variables Age and Age2. The stars indicate the significance of the

Z-scores, * p>0.05 ** p>0.01. Results of our variables of interest are bold.

Dependent: Binary housing tenure choice variable Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Coefficients (Std. error) Model 1 2 3 4 5 Weights parameter 1.5 1.5 1.5 1.5 1.5

Return data based on Lifetime Lifetime Lifetime Lifetime Lifetime

Experienced returns -1.047** (0.159) -0.329* (0.167) -0.322 (0.170) -0.128 (0.216) 0.897** (0.293) Age 0.709** (0.071) Age2 -0.004** (0.001)

High school dummy

0.781** (0.134) 0.713** (0.136) 0.680** (0.166) 0.098 (0.180) College Dummy -1.274** (0.206) -1.127** (0.210) (0.260) -1.359** -0.531 (0.289) #Children -0.423** (0.066) -0.426** (0.067) (0.081) -0.361** -0.137 (0.091) #Children2 0.063** (0.007) 0.064** (0.007) 0.057** (0.009) 0.042** (0.009) Married Dummy 2.089** (0.123) 2.186** (0.126) 2.293** (0.163) 2.310** (0.181) Log Income -4.550** (0.947) (1.176) -4.215** -2.296* (1.222)

Log Income Squared

0.267** (0.049) 0.253** (0.060) 0.145* (0.063)

Log Liquid Assets

(22)

Results 22/33

Maximum LLH -2027.154 -1643.869 -1592.873 -967.053 -786.525

(23)

Results 23/33 this means that the inclusion of additional control variables raises the predictive power of the total model. The most striking part is that age and age2 consume large parts of the magnitudes of the

control variables coefficients and increases the standard errors likewise. The inclusion of the age related variables causes the high school diploma and college degree dummy variables to become statistically insignificant. While the control variable for personal debt turns out to have a negative and

statistically significant relationship. The main results we can derive from the stepwise inclusion procedure is that the experienced return variable is highly sensitive to other variables.

3.3 Discriminant analysis on the explanatory variables

From the robustness check whereby the explanatory variables are included step by step, we learned that the age related variables have a great influence on the other variables. Since we are interested in the experienced return variable we need to investigate the individual discriminant power of our variable of interest and the other explanatory variables by executing a canonical discriminant analysis. By executing a canonical discriminant analysis we predict whether an individual would choose to rent or to own a house based on the inserted explanatory variables. The discriminant analysis starts with a Wilks’ Lambda test. The Wilks’ Lambda test checks if the predicted model for the upstate of the binary outcome is significantly different from the downstate prediction model The Wilks’ Lambda scores vary from 0.812 to 0.808 and all have a chi-square p-value of 0.000. Since the F-test statistics of Wilks’ Lambda test are all significant at 1% level (p>0.01), we can accept the alternative hypothesis that the two groups have different discriminant scores. This means that individuals who own a house have significantly different characteristics than those who rent one.

Table 6 shows that the Wilks’ Lambda scores varies from 0.812 to 0.808 which indicates that 18.8% to 19.2% of the variance in our grouping variable (which is the housing tenure choice) is explained by our explanatory variables in the four models. The stepwise matrix shows percentages of explained variance of those variables which are statistically significant and meet up with the critical F-value. The percentage of each variable sums up with all the other to the total of total variance explained . We see that each model that incorporates a return variables that is based on the sophisticated weighting schemes explains 0.1% of the total variances in our grouping variable. The returns variable based on the standard neo-classical economic assumptions explains 0.6% of the variance in our grouping variable. The results in Table 6 also indicates that variable #Children2, which reflects the concave relationship of the amount of children within the

(24)

Results 24/33 continues with analyzing how accurate the prediction models are by showing a cross-validated classification matrix. The cross-validated classification matrix is based on the in-between correlation values of each explanatory variable. Table 7 shows the structure matrix correlation values.

Table 6

Wilks’ lambda table percentage of the total variance in the grouping variable (housing tenure choice) that is explained by the individuals explanatory variables

The table shows the individual percentages of each explanatory variable after the stepwise inclusion procedure of the discriminant analysis. The variables that pass the F-test by meeting the critical value of 2.25 (11, +120) are entered into the prediction model. Each model has a Wilks’ Lamdba score that reflects the proportion of the unexplained variance in the grouping variable. The sum of the percentages of each individual explanatory variable is the difference between 1 minus Wilks’ Lambda. Results of our variables of interest are bold.

Grouping variable: Binary housing tenure choice variable % of variance explained by each variable % of variance explained by each variable % of variance explained by each variable % of variance explained by each variable Model 1 2 3 4

Return data based on Lifetime Lifetime Lifetime All available

information

Weights parameter 1.5 0.0 -0.5 0.0

Experienced returns 0.1% 0.1% 0.1% 0.6%

Age 0.1% Excluded Excluded Excluded

Age2 0.1% Excluded Excluded Excluded

High school dummy 0.7% 0.7% 0.7% 0.7%

College Dummy 0.1% 0.1% 0.1% 0.1%

#Children 0.7% 0.7% 0.7% 0.7%

#Children2 Excluded Excluded Excluded Excluded

Married Dummy 6.2% 6.2% 6.2% 6.2%

Log Income 0.2% 0.2% 0.2% 0.1%

Log Income Squared 10.4% 10.4% 10.4% 10.4%

Log Liquid Assets 0.3% 0.3% 0.3% 0.4%

Personal Debt (std.) 0.1% 0.1% 0.1% 0.0%

Total variance explained 19.1% 19.2% 19.2% 19.2%

Wilks’ Lambda 0.809 0.812 0.812 0.808

(25)

Results 25/33

Table 7

Structural Matrix of Correlation after the discriminant analysis on each returns related variable

The table shows the individual canonical in-between groups correlations of each explanatory variable after the stepwise inclusion procedure of the discriminant analysis. The variables that pass the F-test by meeting the critical value of 2.25 (11, +120) are entered into the prediction model. The stars indicate which variables meet of with the canonical correlation threshold of 0.300. This thresholds is a cutoff point for important and less important variables in explaining the variance within our grouping variable. For variables that have a correlation value below 0.300 it is suggested that they are weakly associated with the housing tenure choice but are a function of other unassessed factors. Results of our variables of interest are bold.

Dependent: Binary housing tenure choice variable Pooled within-groups correlation Pooled within-groups correlation Pooled within-groups correlation Pooled within-groups correlation Model 1 2 3 4

Return data based on Lifetime Lifetime Lifetime All available

information

Weights parameter 1.5 0.0 -0.5 0.0

Experienced returns -0.039 -0.038 0.126 -0.154

Age 0.016 0.016† -0.015† 0.015†

Age2 -0.017 -0.017† -0.037† -0.017†

High school dummy 0.270 0.274 0.273 0.269

College Dummy 0.008 0.008 0.009 0.008

#Children 0.138 0.140 0.140 0.138

#Children2 0.112† 0.113† 0.111† 0.113†

Married Dummy 0.637* 0.645* 0.643* 0.634*

Log Income 0.691* 0.700* 0.698* 0.688*

Log Income Squared 0.703* 0.712* 0.710* 0.700*

Log Liquid Assets 0.183 0.185 0.184 0.182

Personal Debt (std.) -0.077 -0.077 -0.078 -0.077

Note: * indicates that the variable meets the threshold of having a canonical correlation of at least 0.300, † indicates that this variable is excluded from the model, due to insignificant contribution to the model since the variable did not meet the threshold by having a F-value greater than 2.25.

(26)

Discussion 26/33 What we can derive from Tables 6 and 7 is that the experienced related returns as well as the returns, based on the standard neo-classical economic assumptions, have a significant but a very small contribution in discriminating the homeowner characteristics from the characteristics of individuals who prefer to rent. Table 6 indicates that we should use the results from Model 1 from Table 3 when assessing the individual differences in willingness when recent returns bearing a heavier weight than returns close to birth. The estimation results for Model 1 from Table 3 indicate that there is a positive and statistically significant relationship with experienced returns that follow an increasing convex weighting function and homeownership. Table 6 also indicates that we should use the results from Models 2, 3, and 4 from Table 4 when we are assessing their relationships with housing tenure choice. The estimation results of Model 2 and 3 from Table 4 indicate that there is statistically significant but negative relationship between homeownership and the experienced return variable that is equally weighted and bound to the individuals lifetime. We find a similar relationship when the returns are weighted by a decreasing convex function. Our control group, including the return variable which minds the standard neo-classical economic assumptions, shows a also a negative but statistically significant relationship with homeownership. In sum, the relationship between variables that reflects possible investment returns and homeownership seems to be ambiguous.

4 Discussion

In this thesis we examined empirically whether individuals differ in their willingness to take a direct position in the Dutch real estate asset class depending on the return history they experienced over the course of their lives and if individuals are influenced by returns on their housing tenure choice. Our estimation results on the experienced returns coefficients, as well as on the arithmetic returns coefficient, show a significant relationship which homeownership, this means that individuals differ in their willingness to take a direct position in the Dutch real estate asset class. The results of the discriminant analysis show that both return related variables, based on experiences as well as on standard neo-classical economic assumptions, have little ability to discriminate between characteristics of household groups that prefer to own or to rent a house. Although the returns related variables show a statistical significant relationships with homeownership, they exert a small influence on the housing tenure choice of individuals.

(27)

Discussion 27/33 evidence that individual are influenced by their experiences and weight them in a certain way, since all variables calculated with the sophisticated weighting scheme show statistically significant relationships. The statistically significant estimated coefficient of return variable calculated based on the standard neo-classical economic assumptions, indicate that individuals do not limit the time span to their life courses when gathering information on historical returns and forming their beliefs on risky outcomes. The results on our control group are therefore are in congruence with the standard neo-classical economic assumptions and likewise in conflict with Malmendier and Nagel (2011). From the results of the discriminant analysis we learn that all returns related variables have very weak predictive power when it comes to discriminate in the housing tenure choice of the Dutch households. Analyzing our results we can think of a couple of explanations for the findings that the experience hypothesis does not hold in the Dutch real estate market.

First of all, the fact that the experience hypothesis does not hold in the (Dutch) real estate asset class might be rooted in the differences of the investigated products. As Ranney, Schwab (1982) and Brueckner (1997) acknowledge that the housing consumption is driven by investment and consumption motives, rather than solely by investment motives. The low correlation value of the returns related variables might indicate that individuals are only very little influenced by investment risks when it comes to a housing tenure choice. The average individual in our sample might not incorporate the downside of capital risk and takes the upside for granted, since the Dutch housing market shows an annual growth rate in of 4.7 percent point in the house prices since 1914.

Secondly, the results on the stepwise inclusion procedure, shown in Table 5, provide an explanation for the interfering effects of age related variables on the coefficients and standard errors of other independent variables in our estimations. The interference effect of the age variables with other variables is also found by Kan (2000) and Aarland and Norvik (2009). The literature on housing tenure choice uses the age related variables to control for the taste dispersion on housing tenure choice over the life courses of individuals. Results of our the discriminant analysis show that the age related variables, hardly meet up with the critical F-values, stating that the age related variables are no statistically significant variables in discriminating the homeowners characteristics and characteristics of individuals who prefer to rent. The inclusion of age related variables must therefore be questioned in the first place when analyzing housing tenure choice in the Dutch market.

(28)

Discussion 28/33 relationship with the homeownership. This results is similar to other papers that are modeling housing tenure choice, see Jones (1995), Kan (2000) and Aarland and Norvik (2009), but also similar to Malmendier and Nagel (2011) who find a positive relationship between income and financial risk taking in general. Looking at the log linear variable of liquid assets we see no relationship with the housing tenure choice, which is consistent with the papers of Kan (2000) and Aarland and Norvik (2009) but in contrast with Jones (1995). The fact that we find that being married and income related variables show a positive and statistically significant relationship might have to do with the mortgage lending preference of the Dutch households, since mortgage lending was one of the main sources from 1994 to 2012, Bartelmans et. al (2012). Stating this we assume that mortgage lending facilities consider being married and having a stable income as necessities for providing a mortgage. Our finding that liquid assets has no relationship with homeownership might indicate that financial institutions does not consider the ability of households to overcome a down payment as a necessity when it comes to qualifying for a mortgage. This result underpins the existence of the committed expenditure risk when owning a house, which is found by Fratantoni (1998).

4.1 How our results relate to mean-variance framework

(29)

Discussion 29/33

4.2 An aggregate perspective

On micro household level we show with our data that individuals are very weakly influenced by housing market returns when it comes to their housing tenure choice within the Dutch market. This result implies that other determinants exert a greater influence on the housing tenure choice of individuals, than our control variables and the possible return from capital risk do. If this results holds we should find similar evidence when we look at the Dutch housing market in an aggregate perspective. In the aggregate perspective we analyze the correlation between the P/E ratios of the aggregate Dutch housing market and our experienced return variables. The P/E ratio is valuation ratio that is used by investors to assess other investors’ growth expectations of future payoffs. A high P/E indicates that investors are willing to pay a higher price than they would earn per share, because they expect earnings growth compared to comparable companies with a low P/E ratio. In the field of real estate we must be aware that growth in earnings can emanate from an increase in house prices or an increase in consumption benefits. So if we find that the average experienced return of our sample show no positive and statistically significant relationship with the P/E ratio’s of the aggregate Dutch housing market, this might indicate that individuals expect growth in other than returns from capital risks, which would underpin our finding that housing tenure choices mainly driven by consumption motives. The P/E ratios of the aggregate Dutch housing market are calculated by dividing the average house prices, provided by the Dutch Bureau of Statistics (CBS), with the indirect rental earnings based on rental income of similar objects, Sykes and Young (1981). The average experienced return is calculated over the individuals aging from 25 to 75 with a weight of lambda is 1.5, see Malmendier and Nagel (2011). Figure 3 visualize our findings.

The time series are not statistically significant correlated, (corr. 0.437, p>0.08). Meaning that the experienced returns could not provide an explanation for the increased earnings growth expectations of the Dutch housing market reflected by the bullish trends in the P/E ratios. The result of an non-significant relationship contradicts the outcomes of Malmendier and Nagel (2011). However, this result is in congruence with our findings on micro household level.

(30)

Conclusion 30/33

Figure 3: Average Experienced Real Housing Market Returns (lambda is 1.5) and the average P/E ratio of the Dutch Housing Market

For each year t we calculated the average P/E ration of the Dutch housing market, by dividing the average house prices within the Dutch market, from CBS, by the indirect earnings based on the equivalent rental income of similar objects, see Sykes and Young (1981). Additionally we calculated the average experienced returns of a subsample whereby the households heads age varies from 25 to 75, following the method of Malmendier and Nagel (2011). Combining the two measures allows us to visualize the pattern between the average P/E-ratio’s and the average experienced returns that follow an increasing convex pattern (λ=1.5). Our sampling period is from 1994 to 2010.

5 Conclusion

This thesis investigated how the individuals’ housing tenure choice is influenced by experienced returns on the housing market. The first result we find is that the Dutch households do not use their individual experiences on housing market returns when forming their beliefs about risky outcomes in the Dutch real estate asset class. Although the returns on housing do have a statistically significant relationship with the individuals’ housing tenure choice, the influence of those returns on the individuals’ housing tenure choice is little. Secondly, we find that income factors and the marital status show a strong relationship with the housing tenure choice. This result might indicate that how our financial lending system is structured exerts a greater influence on housing tenure choice than investment returns do.

The major insight that this thesis provides us, is that individuals in the Dutch housing market do not consider possible investment returns from capital risks as a major determinant in their housing tenure choice. 0.000 1.000 2.000 3.000 4.000 5.000 6.000 7.000 8.000 9.000 P /E R a ti os vs. A vg. E x p er ic en d re tu rn s (% )

(31)

Acknowledgements 31/33

6 Acknowledgements

In this paper use is made of data of the DNB Household Survey.

7 References

Aarland, Kristin and Viggo Nordvik. 2009. On the Path to Homeownership: Money, Family Composition and Low-Income Households. Housing Studies 24, 81-101.

Alesina, Alberto and Nicola Fuchs-Schündeln. 2007. Good-Bye Lenin (Or Not?): The Effect of Communism on People's Preferences. American Economic Review 97, 1507-1528. Andersen, Erling B. 1997. Introduction to the Statistical Analysis of Categorical Data. 265. Artle, Roland and Pravin Varaiya. 1978. Life Cycle Consumption and Homeownership. Journal of

Economic Theory 18, 38-58.

Bartelsman, E.; R. Beetsma; H. Benink; P. Boelhouwer; A. Boot; L. Bovenberg; D. Brounen; J. Conijn; P. Eichholtz; J. Frijns; H. Garretsen; K. Goudswaard; B. Jacobs; K. Koedijk; R. Van Der Ploeg; H. Priemus; D. Schoenmaker; E. Sterken; L. Stevens; C. De Vries; H. Wijffels and S. Van Wijnbergen. 2012. Naar Een Duurzame Financiering Van De Woningmarkt. Economic Mandate 02, 1-16.

Brooks, Chris. 2008. Introductory Econometrics for Finance; Introductory Econometrics for Finance. 648.

Brueckner, Jan K. 1997. Consumption and Investment Motives and the Portfolio Choices of Homeowners. Journal of Real Estate Finance & Economics 15, 159-180.

Burns, Robert B. and Richard A. Burns. 2008. Business Research Methods and Statistics using SPSS. 544.

Camerer, Colin and Teck-Hua Ho. 1999. Experience-Weighted Attraction Learning in Normal Form Games. Econometrica 67, 827-874.

Choi, James J.; David Laibson; Brigette C. Madrian and Andrew Metrick. 2009. Reinforcement Learning and Savings Behavior. Journal of Finance 64, 2515-2534.

Di Salvo, Pamela and John Ermisch. 1997. Analysis of the Dynamics of Housing Tenure Choice in Britain. Journal of Urban Economics 42, 1.

(32)

References 32/33 Engelhardt, Gary V. 1996. Consumption, Down Payments, and Liquidity Constraints. Journal of

Money, Credit & Banking (Ohio State University Press) 28, 255-271.

Engelhardt, Gary V. 1994. House Prices and the Decision to Save for Down Payments. Journal of Urban Economics 36, 209.

Erev, Ido and Alvin E. Roth. 1998. Predicting how People Play Games: Reinforcement Learning in Experimental Games with Unique, Mixed.. American Economic Review 88, 848.

Follain, James R. and David C. Ling. 1988. Another Look at Tenure Choice, Inflation, and Taxes. Journal of the American Real Estate & Urban Economics Association 16, 207-229.

Fratantoni, Michael C. 1998. Homeownership and Investment in Risky Assets. Journal of Urban Economics 44, 27.

Guiso, Luigi; Tullio Jappelli and Daniele Terlizzese. 1996. Income Risk, Borrowing Constraints, and Portfolio Choice. American Economic Review 86, 158-172.

Guiso, Luigi; Paola Sapienza and Luigi Zingales. 2008. Trusting the Stock Market. Journal of Finance 63, 2557-2600.

Guiso, Luigi; Paola Sapienza and Luigi Zingales. 2004. The Role of Social Capital in Financial Development. American Economic Review 94, 526-556.

Hertwig, Ralph; Greg Barron; Elke U. Weber and Ido Erev. 2004. Decisions from Experience and the Effect of Rare Events in Risky Choice. Psychological Science (Wiley-Blackwell) 15, 534-539. Hochguertel, Stefan and Arthur van Soest. 2001. The Relation between Financial and Housing Wealth:

Evidence from Dutch Households. Journal of Urban Economics 49, 374.

Hosmer, David W.; Stanley Lemeshow and Rodney X. Sturdivant. 2013. Applied Logistic Regression. 1-500.

Huberty, Carl J. 1994. Applied Discriminant Analysis. 466.

Jones, Lawrence D. 1995. Testing the Central Prediction of Housing Tenure Transition Models. Journal of Urban Economics 38, 50.

Kan, Kamhon. 2000. Dynamic Modeling of Housing Tenure Choice. Journal of Urban Economics 48, 46.

Kaustia, Markku and Samuli Knüpfer. 2008. Do Investors Overweight Personal Experience? Evidence from IPO Subscriptions. Journal of Finance 63, 2679-2702.

(33)

References 33/33 Kraft, Holger and Claus Munk. 2011. Optimal Housing, Consumption, and Investment Decisions Over

the Life Cycle. Management Science 57, 1025-1041.

Malmendier, Ulrike and Stefan Nagel. 2011. Depression Babies: Do Macroeconomic Experiences Affect Risk Taking?*. Quarterly Journal of Economics 126, 373-416.

Nisbett, Richard and Lee Ross. 1980. Human Inference : Strategies and Shortcomings of Social Judgment. 1-334.

Osili, Una Okonkwo and Anna L. Paulson. 2008. Institutions and Financial Development: Evidence from International Migrants in the United States. Review of Economics & Statistics 90, 498-517. Pei-Ling Lee; Roy W. L. Khong and Suganthi Ramasamy. 2010. A Profile of Companies Going

Private in the Malaysian Stock Market. IUP Journal of Applied Finance 16, 44-58.

Ranney, Susan I. 1981. The Future Price of Houses, Mortgage Market Conditions, and the Returns to Homeownership. American Economic Review 71, 323.

Schwab, Robert M. 1982. Inflation Expectations and the Demand for Housing. American Economic Review 72, 143.

Sheiner, Louise. 1995. Housing Prices and the Savings of Renters. Journal of Urban Economics 38, 94.

Sykes, Stephen G. and Michael S. Young. 1981. A Differential Approach to Income Property Valuation: A New Measurement Technique. Appraisal Journal 49, 214-233.

Vandevyvere, Windy and Andreas Zenthöfer. 2012. The Housing Market in the Netherlands. European Commission: Economic Papers 458, 1-31.

Weber, Elke U. and Ulf Bockenholt. 1993. Determinants of Diagnostic Hypothesis Generation: Effects of Information, Base Rates, and.. Journal of Experimental Psychology.Learning, Memory & Cognition 19, 1151.

Weiss, Yoram. 1978. Capital Gains, Discriminatory Taxes, and the Choice between Renting and Owning a House. Journal of Public Economics 10, 45.

Referenties

GERELATEERDE DOCUMENTEN

This hedonic price model examines how the composition effect and constant quality indices behave, when there are no differences in marginal contribution of housing

Begheyn argue that the space for housing development within the borders of the built environment in Alphen aan den Rijn is scarce, causing the need for housing development outside

I have conducted interviews with 10 freelance journalists who have been reporting in Conflict Affected and High-Risk Areas, and experienced potentially traumatic

line in the ALMA data, the disk-averaged l:c spectrum, as shown in the left panel of Figure 3, is fitted with a 3 rd -order polynomial to remove the low-frequency curvature of

Overall, little evidence is found for the existence of a ripple effect, but the results provide strong evidence on long-run housing price convergence towards an equilibrium

In this scenario, the flexibility provided by the ArenA storage system is used for peak shaving the load of the stadium and the EV charging points combined.. The resulting

The ritual dynamics of separation, transition and integration allow us to further scrutinise post-mortem relationships and, as I will argue, not simply to point to breaking

Series volumes follow the principle tracks or focus topics featured in each of the Society’s two annual conferences: IMAC, A Conference and Exposition on Structural Dynamics, and