The effects of the increase in the retirement age on Dutch household consumption

(1)

The effects of the increase in the retirement age on

Dutch household consumption

Bachelor Thesis

Specialization Economics and Finance

Daphne Kok

11023198

Supervised by

Stan Olijslagers

University of Amsterdam

25 June, 2018

(2)

2

Statement of Originality

This document is written by Student Daphne Kok 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

This thesis examines the impact of the increase in the retirement age on 1 January 2014 on the consumption behavior of Dutch households. Using three main theories in the literature and the performance of a regression on consumption, two effects are examined. It is argued that current income did rise due to the increase in the retirement age, resulting in more

consumption. Moreover, it is found that households realize they will have a higher expected future income and therefore consume more. Hence these results show that consumption did change significantly after 1 January 2014.

1. Introduction

In the Netherlands, the pension system consists of three pillars. The first pillar is a flat-rate public pension, called the AOW, available to all residents of the Netherlands between 15 and 65 years (van der Zwan, 2015). This pillar is based on a pay-as-you-method, which means that the current active working population provides the pensions for the retired part of the population. According to Anderson and Kaeding (2015), the AOW provides on aggregate about half of the retirement income, while the rest is accounted for by the second and third pillar. The second pillar are occupational pensions, in which an employee pays a fixed percentage of his salary for his future pension. The third pillar consists of voluntary personal pensions, which are only about 5% of the retirement income (van der Zwan, 2015). Therefore, most of the retirement income is due to the AOW and the occupational pensions.

Current research is mostly focused on possible reforms of the pension system and the effects on welfare and the tax burden of future generations. So, their focus is more on what will happen in the future due to changes regarding the pension system. For example, Beetsma, Bettendorf, and Broer (2003) examined the effects of a reduction in benefits, an increase in the retirement age and smoothing of the AOW contribution rate on the welfare of future generations in the Netherlands. The last part means that the AOW contribution rate is held constant at 11.4%. This has different effects on the welfare of generations. However, those studies are based on assumptions about future demographic developments, of which the true values were unknown at the time these studies have taken place. Also, researches by

European Commission (2018), which do indeed report actual values of the labour force, potential GDP and dependency ratios aim at providing information about fiscal sustainability for countries as a whole. Hence, those studies aim at describing effects of implementing a potential policy change on the country as a whole. In this case, actions from the perspective of households are not discussed. Therefore, it is useful to examine how households reacted after

(5)

5

a policy change was indeed implemented via an empirical analysis. Since consumption by households is an important part of GDP which influences economic growth and employment, I would like to examine whether there is a significant change in consumption of households in the Netherlands after the increase of the retirement age on 1 January 2014.

First of all, several theories about consumption behavior will be elaborated and will be used as a way of explaining how consumption can be affected by the increase in the

retirement age. Next, the results of studies in which those theories are tested, are discussed. After that, section 3 is about the methodology for the empirical analysis of this thesis, including a description of two potential effects on consumption due to the increase in retirement age. Then, the results are presented in the analysis part. Finally, this thesis ends with a discussion and conclusion on the research done regarding the main question.

2. Findings of the literature

In the literature, there are different contradicting theories about what drives consumer behavior. Several studies have been conducted with different results. Therefore, in the first part of this section, three well-known early theories about consumption will be elaborated. Next, a literature review will be provided in which the results of multiple studies testing those theories are compared.

2.1. Theoretical framework

2.1.1. John Maynard Keynes consumption function:

The first model which is discussed, is the standard Keynesian theoretical model which states that consumption is determined by autonomous consumption, the amount of disposable income and the marginal propensity to consume (Keynes, 1936). The last part is influenced by subjective and objective factors, such as changes in the monetary value of wealth of

households and changes in the wage-unit. However, changes in consumption would be mainly influenced by changes in the amount of disposable income, since the propensity to consume is a relatively stable function between zero and one, stated by Keynes (1936).

Furthermore, changes in fiscal policy can influence the amount of consumption. Namely, income taxes influence the amount of the resulting disposable income and hence determine consumption. Therefore, changes in fiscal policy could potentially influence household consumption. This is especially relevant for this thesis if we consider the increasing government debt as a result of the ageing problem and the potential effects for fiscal policy.

(6)

6

Moreover, according to Keynes (1936), the effects of changes in the interest rate on the consumption behavior of households are not certain. In the long run, it is likely that substantial changes in the interest rate will influence the way households spend their income, thus the propensity to consume. However, it is not sure in which direction this is affected. Furthermore, fluctuations of the interest rate over a short period are not likely to direcly affect consumption much. Hence, Keynes (1936) thinks that the interest rate is relatively

unimportant and is therefore not directly included in the consumption function as a separate variable.

2.1.2. Franco Modigliani and the life-cycle hypothesis

Another theory on consumption behavior is the life-cycle hypothesis of Ando and Modigliani. The starting point for this hypothesis is the utility function of an individual consumer (Ando & Modigliani, 1963). This function consists of both current aggregate consumption and future aggregate consumption. The idea behind the life-cycle hypothesis is that an individual consumer maximizes its life time utilityby smoothing its consumption over the course of its life. According to Ando and Modigliani (1963), income varies over a persons life because of retirement. When people have a job and therefore earn income, they can also afford a certain level of consumption. However, when they retire, their income will fall and they would not be able to maintain that level of consumption. Therefore, people try to smooth their life time resources to maintain the same level of consumption each year by saving some amount to compensate for the decrease in income when they retire. Those resources are current and future discounted life time earnings and the initial level of wealth (Ando & Modigliani, 1963). Finally, the individual consumption functions can be aggregated to get the consumption function for the country as a whole, which depends on income and wealth.

There are some underlying assumptions to the theory of Ando and Modigliani (1963) which affect the level of consumption. First of all, when an individual consumer receives one additional dollar of resources, consumption will change in the same proportion in which the total resources were allocated before this extra dollar was received (Ando & Modigliani, 1963). In addition, according to them, at any point of time, the individual consumer will use its total life time resources evenly over the rest of his remaining life. Moreover, they state that the individual consumer is not expected to receive or leave any bequests to its children. Their fourth assumption is that the total life span is set at 50 years and the earnings span is set at 40 years. Finally, the rate of return on assets is constant and will remain that way (Ando & Modigliani, 1963).

(7)

7

2.1.3. Milton Friedman’s permanent-income hypothesis

The third theory is the permanent-income hypothesis of Milton Friedman. According to Friedman (1957) income consists of a permanent component and a transitory component. The permanent component is the expected average income over a long period of time

(Friedman, 1957). The transitory part reflects all other factors that occur temporarily and are not perceived to persist in the future. Hence, the transitory component can be seen as the deviation from the expected average income. It can be caused by for example cyclical fluctuations in economic activity, or a one time winning in the lottery (Friedman, 1957).

At the same time, consumption can be represented by a permanent and transitory part too. According to Friedman (1957), the ratio between permanent income and permanent consumption does not depend on the size of the permanent income. The ratio between those two depends on the rates of interest at which the consumer can borrow or lend, on the ratio of financial wealth to income and the preferences for each individual consumer for either

consumption or additions to wealth (Friedman, 1957). Moreover, there is zero correlation between the permanent and transitory component for both consumption and income.

A more important assumption is the one that states that the transitory part of

consumption is uncorrelated with the transitory part of income (Friedman, 1957). This means that if a consumer receives a one time extra amount of income, such as an inheritance, the consumer is expected not to change his consumption. He is expected to save this amount rather than spend it on consumption. In contrast, if a consumer receives a permanent raise in the permanent income component, he is expected to spend almost all of it on consumption (Friedman, 1957). Essentially, only changes in permanent income will result in changes in consumption and changes in transitory income will be used to either save or dissave.

2.1.4. Comparison of the three theories

The three theories discussed in the previous sections have different starting points for what determines consumption of an individual consumer. The theory of Keynes (1936) states that consumption is only based on current income, while both the life-cycle hypothesis of Ando and Modigliani (1963) and the permanent-income hypothesis of Friedman (1957) argue that consumption should not be based on current income only. Hence, according to Keynes (1936), consumption will change if current income changes.

At the same time, according to Friedman (1957) current income can differ between years, while Ando and Modigliani (1963) assume consumers to earn the same amount of income each year. So according to the theory of Friedman (1957) consumption will increase if

(8)

8

the change in current income is permanent. On the other hand, Ando and Modigliani (1963) argue that consumption increases if there is an increase in wealth or if you would expect to retire later, which results in more life time earnings. Furthermore, Keynes (1936) and

Friedman (1957) argue that consumption changes due to a change in current income, but not if expected future income changes. In contrast, Ando and Modigliani (1963) state that consumption depends on expected future income. Thus according to their theory, if expected future income increases, consumption will also increase.

2.2. Literature review

2.2.1. Results on Keynes’ theory

The first theory described in the previous paragraph was the one of Keynes and this one is critized by Holden (1938). He thinks that the interest rate is far more important in determining consumption than Keynes (1936) states in his work. According to Holden (1938), a reduction in the interest rate can not only lead to a lower amount of savings out of current income, but it can even result in dissaving, depending on the preferences of individual

consumers. He implies that consumption can be higher than current net income, while Keynes (1936) stated that the propensity to consume was lower than unity, which means that

consumption would be less than current income for any interest rate. Hence, the interest rate may influence consumption more than implied by the theory of Keynes (1936).

2.2.2. Results on the life-cycle hypothesis of Ando and Modigliani

The second theory discussed was the life-cycle hypothesis of Ando and Modigliani (1963). First of all, the consumption-wealth channel, which includes the variables of housing and financial wealth and does also depend on the interest rate, has an effect on consumption. Since lower real interest rates will lead to higher housing prices, this would mean that wealth of households increases and consumption will increase (Macdonald, Mullineux, & Sensarma, 2011). Moreover, they state that lower real interest rates will increase the price of financial assets, resulting in higher financial wealth and more consumption. Nonetheless, D’Orlando and Sanfilippo (2010) write that current consumption closely follows current income and its fluctuations and is not much influenced by future income, consumer wealth and permanent income. Subsequently, this is not consistent with the life cycle-permanent income hypothesis and the study of Macdonald et al. (2011), which find that consumer wealth does influence consumption. On the other hand, D’Orlando and Sanfilippo (2010) do mention that wealth influences the marginal propensity to consume, but not in a linear way.

(9)

9

In addition, Carroll (1994) tested the implication that current consumption should depend on expected future income of the life cycle hypothesis of Ando and Modigliani (1963). He found little evidence in line with this implication. In his study, current income seemed to play a role in determining consumption, rather than expected future income. This result is consistent with the Keynesian consumption function. This would mean that there is only a significant change expected in consumption due to changes in current income, but not because of changes in expected future income. Furthermore, Carroll (1994) showed that consumption is influenced by uncertainty in future income, which implies that consumers are rational and do indeed care about the future and their corresponding income. This corresponds to Leland (1968) who points out that people reduce their consumption and save more to compensate for future uncertainty. Subsequently, Carroll (1994) implies that a variable should be included in the consumption function, to account for this uncertainty in future income, which does not fit the theory of Keynes (1936).

Moreover, the life cycle-permanent income hypothesis implies that consumers choose their current consumption on the basis of the available information about their current and future income. This means that only unexpected changes in policy will lead to a change in consumption (Hall, 1978). He reasons that only new information about changes in future policy will have an effect on consumption, because any information that is known today, will already be incorporated in the decision about current consumption. Furthermore, these unexpected changes in future policy will only influence consumption if those changes are permanent and as a result influence the permanent component (Hall, 1978).

In comparison, Flavin (1981) used a structural model in her study instead of the reduced-form consumption equation used by Hall (1978) to also test the life cycle-permanent income hypothesis. In contrast to Hall (1978), she did not find evidence in favor of this hypothesis. Namely, consumption did not only react upon unpredictable changes of income, but also on predictable variation in income (Flavin, 1981). Wilcox (1989) elaborated on those two studies with his research on the impact of changes in social security benefits on

consumption. Changes in those benefits have been announced at least six weeks prior to the actual payment (Wilcox, 1989). So according to the life cycle-permanent income hypothesis, those changes should be incorporated in the consumption expenses at the time of the

announcement and not at the time of the payment. However, his results are in line with Flavin (1981) in the fact that the life cycle-permanent income hypothesis is rejected. Although changes in social security benefits were fully anticipated, consumption did not only change at the time of the announcement, but it did also change at the time when the benefits were paid

(10)

10

(Wilcox, 1989). This is not consistent with the findings of Hall (1978), who would predict no change in consumption at the time of the actual payment. Therefore, it is not sure how

consumers will behave when there are potential changes in future policy.

Another factor that influences economic growth and thus has an effect on consumption, is a change in demographics. This is especially important with the ageing problem and the resulting raise in the retirement age. Leff (1969) found that there is a

significant negative effect of the old-age dependency ratio on the savings rate. This finding is supported by Edwards (1996), who also found that the coefficient of the old-age dependency ratio with respect to savings is significantly negative, which is in line with the life-cycle hypothesis. This ratio equals the amount of people older than 64 years to the amount of the working population in the age of 15-64 (Muszyńska & Rau, 2012).

However, Adams (1971) and Goldberger (1973) do not agree with the way Leff (1969) has conducted its research. They found cases in which the effect of the old-age dependency ratio on savings was insignificant or even positive. According to Goldberger (1973), the data constructed by Leff (1969) were internally inconsistent, which could account for the

discrepancies which were found. Hence, there is a debate about the relationship between the old age dependency ratio and savings. Moreover, there could be a relationship between consumption and the old-age dependency ratio. As a result, it is suggested by Leff (1969) to build a model in which individuals also maximize utility, but with respect to consumption in which the old-age dependency ratio is included.

2.2.3. Results on the permanent-income hypothesis of Friedman

Houthakker (1958) tested the permanent-income hypothesis of Friedman and used city class, occupation and age of the head of the household as independent variables. The test results did not support the theory and it is suggested by Houthakker (1958) that this could mean that the marginal propensity to consume out of transitory income is not equal to zero, which was an important feature of the permanent-income hypothesis of Friedman (1957). So if a consumer receives an extra amount of income he will not use all of it for savings, but he will use some of it for consumption expenses, which is in contrast with the theory of

Friedman.

However, Eisner (1958) argues that Houthakker (1958) did not execute the test in a correct way, which resulted in the rejection of the theory. According to Eisner (1958), the results would be consistent with the theory of Friedman, if the test is performed correctly. Subsequently, Eisner (1958) found that consumption is indeed unrelated to the transitory

(11)

11

components of income, in accordance with Friedman (1957). Hence, according to Eisner (1958), consumption will not change if there is a change in the transitory part of income. All in all, it can be shown that there is much discussion in the literature about the effects of the different variables on consumption. Therefore, it is not clear which results to expect based on the regression which will be performed later on in this thesis.

3. Methodology

In this paragraph, the empirical model is described to test whether there is a significant change in the consumption of households after the increase of the retirement age on 1 January 2014 in the Netherlands. When looking at the three theories, it becomes clear that there are two potential results of this potential increase. So this section is started with a description on the first effect, how it will be investigated and which data are used. Then, the second effect is elaborated, including the methodology on the analysis of this effect and the dataset used.

3.1. First effect

3.1.1. Methodology on the first effect

After the increase of the retirement age, it is expected that current income of people between the age of 65 and 66 will increase, since they have to work longer now, so they will receive wages for an additional period. According to Keynes (1936), Ando and Modigliani (1963) and Friedman (1957), this increase in aggregate current income of households should lead to higher aggregate consumption of households. Therefore, to be able to investigate this effect, a rough analysis will be provided in the fourth section on whether people of the age of 65 and 66 indeed decided to work longer and if this shows in the amount of aggregate current income and aggregate consumption.

For the first part, a Wald test is performed on quarterly data of the period 2003-2017 on the net labour participation rate of 65 and 66 years old residents. This test will provide results on whether there is a structural break in the time-series data after the increase of the retirement age on 1 January 2014. Namely, under the null hypothesis, there is no structural break. The alternative hypothesis states that there is indeed a structural break at a known break date in this case.

According to Piehl, Cooper, Braga, and Kennedy (2003), these hypotheses are shown in the following way: : = , meaning that the parameter of interest does not change between periods; ( ): = , = 1, … ,

(12)

12

point of structural change, 1 January 2014 in this case, T equals the sample size and Tπ is the time of the change. With the Wald test, the actual test statistic is compared to a chi-squared distribution. Hence, when the value of the test statistic is higher than the critical chi-squared value, the null hypothesis will be rejected and there is enough evidence to suggest that there is indeed a structural break in the data. This would mean that relatively more people of the age of 65 and 66 would have a paid job, leading to more income and potentially higher

consumption.

For the next part of the analysis, a rough indication will be given whether aggregate current income and aggregate consumption of households did indeed increase after 1 January 2014. This is done by plotting both aggregate current income and aggregate consumption against time and indicating whether an increase is visible. This should partly answer the research question of this thesis whether consumption of households did indeed increase after the raise of the retirement age on 1 January 2014.

3.1.2. Data on the first effect

First of all, it is expected that relatively more 65 and 66-year olds will continue to work longer. To measure this expectation, data about the net labour participation for people of the age of 65 and 66 are obtained from Central Bureau for Statistics (2018). It is expected that this rate will increase after 2014, because of the increase in the retirement age and this should result in higher aggregate current income. Next, data about aggregate current income of households are obtained from OECD (2018). According to theory of Keynes (1936), Ando and Modigliani (1963) and Friedman (1957) it is expected that an increase in aggregate current income will lead to an increase in aggregate consumption. Therefore, data about aggregate consumption of households are found on Eurostat (2018). For both aggregate current income and aggregate consumption, those data are yearly for the period 1995-2017 using constant prices with 2010 as a base year.

3.2. Second effect

3.2.1. Methodology on the second effect

As already mentioned, there are two potential results of the increase in the retirement age. Besides the increase in aggregate current income which should lead to higher aggregate consumption, Ando and Modigliani (1963) argue that households try to smooth their lifetime earnings to attain a certain level of consumption. Hence, the working population will realize that it is going to retire at a later date, so its expected future income will rise. This results in a

(13)

13

higher level of current consumption, since more lifetime earnings are available to smooth its income. So, according to Ando and Modigliani (1963) there is an additional increase in consumption, even after controlling for the income effect described in the previous part.

To measure this second effect, a regression will be carried out on the dependent variable, which is aggregate household consumption. Besides the control variables that determine consumption, namely the real interest rate, housing wealth, financial wealth, disposable personal income, the unemployment rate and the old-age dependency ratio, a binary variable will be created to account for this second effect. It can only take two values, namely 1 for a quarter after the retirement age is raised and it equals 0 for a quarter before the retirement age is raised. Hence, according to Ando and Modigliani (1963), the coefficient on this binary variable should be significant, while following the theories of Keynes (1936) and Friedman (1957) it should be insignificant. This contributes to answering the research question of this thesis.

Now that all the regressors determining aggregate consumption are known, a

regression can be carried out, which is represented by the following regression equation: =

+ ∗ + ∗ + ∗ + ∗ + ∗ + ∗ +

∗ + . Here Retiret represents the binary variable to measure the effect of the

increase in expected future income, while Ct represents the consumption and εt is the error

term for the time period. Moreover, tests on multicollinearity, non-stationarity,

heteroskedasticity and autocorrelation of the data will be carried out, to make sure that the results are accurate.

3.2.2. Data on the second effect

First of all, all data are quarterly for the time period of 1995 until the fourth quarter of 2017. To come up at the real interest rate, the nominal short term interest for that period is first obtained from Oxford Economics (2018). Next, the Fisher equation states that the real interest rate is approximately equal to the nominal interest rate less the rate of inflation (Karatzas, Shubik, Sudderth, & Geanakoplos, 2006). To represent this rate of inflation, the Consumer Price Index with changes over the different quarters is used and also retrieved from Oxford Economics (2018). Finally, the real interest rate is found by subtracting this rate of inflation from the nominal short term interest rate.

As stated in the previous part, household wealth also influences the amount

households are going to consume. Therefore, to represent housing wealth, the price-to-rent ratio is included as a regressor and data are obtained from OECD (2018). According to

(14)

14

OECD, this ratio is a measure of the profitability of house ownership. If this ratio goes up, this could mean that the house prices have increased relatively more than rents and indicating more wealth to households (André, Gil-Alana, & Gupta, 2014). Furthermore to represent financial wealth, the AEX index is included. To represent current income, data about disposable personal income, which is the amount of money available to households for consumption and savings after taxes have been deducted, are obtained from Oxford Economics (2018).

Next, data of the unemployment rate for the different quartiles during the period are also obtained from Oxford Economics (2018) to take the unemployment risk into

consideration. To measure the old-age dependency ratio, data about the active working population are retrieved from Eurostat (2018) and data about the population aged 65 and older are found on Oxford Economics (2017). These are subsequently used to calculate the old-age dependency ratio. Finally, data of final consumption expenditure of households are found on Eurostat (2018) and are used to represent the dependent variable.

In table 1, some key statistics of the independent variables are displayed, in which all variables are in percentages except for disposable personal income and consumption, which are measured in billions of euros and for the AEX which is an index. Many variables were influenced by the financial crisis of 2007-2008. For example the real interest rate, which was fluctuating around three percent before the end of 2008. However, after the fourth quarter of 2008, in which the interest rate equaled 4.61% it dropped sharply to 1.56% in the first quarter of 2009, represented by the minimum and maximum values in table 1. The real interest rate even reached negative values for most of 2016 and the whole year of 2017. This negative real interest rate after 1 January 2014 should have a positive effect on consumption.

Moreover after the first quarter of 2014, a downward trend is visible in the unemployment rate. This downward trend in the unemployment rate could indicate that households view the future as less uncertain, which reduces their savings, resulting in a higher amount of money available for spending, with a potential increase in consumption as a

consequence. Hence, this downward trend in the unemployment rate should also have a positive effect on consumption after 1 January 2014.

(15)

15

Table 1. Summary statistics

Variable N Mean Minimum Maximum Std. Dev.

Real interest 92 .0178 -.02 .05 .01628 House 92 1.1910 .7611 1.496 .21641 AEX 92 405.6036 178.92 675.44 117.0859 DPI 92 67.7583 41.88 88.46 13.02289 Unemployment 92 .0619 .03 .09 .01611 Dependency 92 .2959 .27 .37 .03165 Consumption 92 62.501 39.94 80.27 11.01308

4. Empirical analysis and discussion

In this section, the results of the two potential effects on consumption are elaborated. Moreover, tests on multicollinearity, non-stationarity, heteroskedasticity and autocorrelation are carried out to formulate an accurate answer on the research question of this thesis.

4.1. Findings on the first effect

It was already outlined that aggregate current income should increase because some part of the population is expected to work additional years. Then, following the theories of Keynes (1936), Ando and Modgliani (1963) and Friedman (1957), this increase in aggregate current income should lead to higher aggregate consumption. To examine whether this expectation holds, it is first tested whether people of the age of 65 and 66 years old

significantly increased their labour supply by using the Wald test, which was already outlined in the methodology part. The results can be found in table 2.

Table 2. Wald test on structural break in time-series data on net labour participation

Regressor χ2_value _df _P-value

NLP 65 years old 76.0787 59 .0000

NLP 66 years old 21.6633 59 .0000

For both the net labour participation rate of people with the age of 65 or 66, the chi-squared value has a p-value of 0.0000, indicating that it is significant at the five percentage level. This means that there is indeed a structural break in the data on the net-labour

(16)

16

significantly after 1 January 2014 means that relatively more people of the age of 65 and 66 have a paid job after this date and current income is expected to have been increased as well.

The next step is to investigate whether this increase in net labour participation did indeed result in more aggregate current income. But, first of all it is necessary to mention that this is ceteris paribus, so under the assumption that all factors that influence aggregate current income and aggregate consumption of households do not change. In reality it is not likely that this will hold, but to be able to say something about whether this increase in net labour

participation did lead to an increase in aggregate current income and subsequently in aggregate consumption, it is necessary to make this assumption.

Since data about aggregate current income of 65 and 66-year-olds specifically are not available, total aggregate household income is used to still be able to give an indication of the potential increase in consumption due to the increase in current income caused by the increase in the retirement age. In graph 1, aggregate household current income is displayed on the y-axis for the period 1995-2017, which is put on the x-y-axis. From graph 1 it is shown that current income fluctuated between 2000-2013 around 276 billion, but after that period, there is a slight increase in current income visible for the period 2014-2017. This increase in aggregate current income could partly be caused by the fact that relatively more 65 and 66-year-olds have a paid job after the increase of the retirement age on 1 January 2014. However, this increase in current income could also be due to other factors influencing current income. Hence, it is not sure if the increase in aggregate current income visible after 1 January 2014 is caused by the fact that relatively more people of the age of 65 or 66 have a paid job due to the increase in the retirement age, or if this increase is caused by other factors.

Graph 1. Aggregate current income of households 1995-2017

2 2 0 0 0 0 2 4 0 0 0 0 2 6 0 0 0 0 2 8 0 0 0 0 3 0 0 0 0 0 C u rr e n ti n c o m e 1995 2000 2005 2010 2015 Time

(17)

17

According to Keynes (1936), Ando and Modigliani (1963) and Friedman (1957), an increase in aggregate current income should lead to an increase in aggregate current

consumption. Again, no data is available on aggregate consumption of people of the age of 65 or 66 specifically, so total aggregate consumption of Dutch households is considered. In graph 2, aggregate household consumption is plotted against time, for which the years during the period 1995-2017 are used. Just as in the case with total aggregate current income, total aggregate consumption of households fluctuates between 2000-2013. Furthermore, it increases after this period for the years 2014-2017. The fact that the increase in aggregate current income is indeed accompanied by an increase in aggregate consumption, is in line with Keynes (1936), Ando and Modigliani (1963) and Friedman (1957). However, the fact that aggregate consumption increased, could also be due to other factors and is not necessary caused by the effects on labour participation and current income as a result of the increase in the retirement age on 1 January 2014.

Graph 2. Aggregate consumption of households 1995-2017

4.2. Findings on the second effect

4.2.1. Estimated results of the regression on the second effect

In this part, the results of the regression of household consumption on the different regressors are discussed to answer the research question of this thesis and these results can be found in table 3. Most of the outcomes on the coefficients of the control variables are in line with theory, except for the unemployment rate. It turns out that the unemployment rate coefficient is significant and positive, which is not what was expected according to Leland (1968) and Carroll (1994). However, it was expected that more money would be used for

2 0 0 0 0 0 2 2 0 0 0 0 2 4 0 0 0 0 2 6 0 0 0 0 2 8 0 0 0 0 C o n s u m p ti o n 1995 2000 2005 2010 2015 Time

(18)

18

savings by households to account for more uncertainty about their future income, represented by the increase in the unemployment rate.

Furthermore, it is more useful to look at the coefficient on the binary variable, because this coefficient contributes to partly answering the research question of this thesis. In the regression a coefficient of -1.0082 was found, so the increase in retirement age contributed to a decrease in household consumption of 1.0082 billion. However, it is not statistically

significant, which is displayed in table 3. So for the time period from 1995 until 2017, this would indicate that there is not a significant change in consumption due to the increase in the retirement age on 1 January 2014. If you compare this result with the three theories outlined in part two, this gives mixed outcomes. According to Keynes (1936), current consumption is mostly based on current income. Hence, it is expected that current income for elderly people would change and so would consumption. Namely, people who are 65 or 66 know that they will need to work longer and will have a higher current income. This would result in higher current consumption according to Keynes (1936). However, this effect is caught in the coefficient of disposable personal income and has nothing to do with the binary variable which was created. So the fact that the coefficient on retirement is not significant, is in line with Keynes (1936).

Next, the life-cycle hypothesis of Ando and Modigliani (1963) was discussed and they also view current income as an important factor in determining consumption. So, the increase in retirement age would also lead to higher current income and higher current consumption, just like Keynes (1936) argues. Furthermore, they state that consumers will recognize that they will have to work longer, so their life time earnings will increase, resulting in more consumption. This effect is captured with the indicator variable on retirement. Hence, following the reasoning of the life-cycle hypothesis of Ando and Modigliani (1963), the coefficient on retirement should be positive. However, the coefficient on retirement is negative, yet not significant, which does not fit the theory of Ando and Modigliani (1963). Finally, Friedman (1957) stated that only changes in permanent income would lead to changes in consumption. This is the same effect mentioned in the case of Keynes (1936) and so is captured in the variable of disposable personal income, representing current income. Hence, it is in line with Friedman (1957) not to expect the coefficient on retirement to be significant.

All-in all, the most important outcome of this regression is the fact that the coefficient on the binary variable is negative, but not significant. It was indeed expected by Keynes (1936) and Friedman (1957) that the coefficient should be insignificant. However, this is not in line with the life-cycle hypothesis of Ando and Modigliani (1963).

(19)

19

Table 3. Regression results of aggregate Dutch household consumption

Regressor (1) (2) (3) (4) (5) (6) Real interest -.2413* (.1230) -.5097** (.1671) -.4793** (.1274) -.4810** (.1276) -.4793 (.1891) -.4810 (.1885) House .2833** (.0225) .0987** (.0135) .2922** (.0931) .2354** (.0781) .2922** (.0860) .2354** (.0672) AEX .0037** (.0014) .0032** (.0019) .0005 (.0024) .0009 (.0023) .0005 (.0019) .0009 (.0019) DPI .2409** (.0512) .6681** (.0285) -.0739* (.0433) -.0689 (.0431) -.0739 (.0291) -.0689 (.0282) Unemployment .3500** (.1161) .2368 (.1617) -.4979 (.3915) -.4461 (.3893) -.4979 (.3609) -.4461 (.3712) Dependency 2.1576** (.2374) .9812 (.8782) .9811 (.9025) Retirement -1.0082 (0.6613) 2.6398** (.7359) -.3231 (.3202) -.1541 (.2827) -.3231) (.2753) -.1541 (.2241) Intercept -54.4781** (6.4949) 3.1670** (1.9634) .3068** (.1363) .3849** (.1172) .3068** (.1073) .3849** (.0890)

Notes Estimated using an OLS-regression. Retirement is a binary variable that equals 1 for a quarter after 1 January 2014 and 0 otherwise. The standard errors are reported in parentheses below the estimated coefficients. * Significantly different from zero at the 10% significance level

** Significantly different from zero at the 5% significance level

4.2.2. Multicollinearity

It is possible that the previous results of the regression are biased, due to

multicollinearity (Lavery, Acharya, Sivo, & Xu, 2017). As can be seen in table 4 in the appendix, there is a high correlation between the binary variable of retirement and the old-age dependency ratio of 0.8477. Moreover, there is a relatively high correlation between the real interest rate and the old-age dependency ratio of -0.7940. Therefore, it is expected that there is multicollinearity in this model due to the old-age dependency ratio. This can lead to large variances in the coefficients of the regression function. But more importantly, it is possible that the actual relationship between the explanatory variables with the dependent variable is not completely shown in this regression (Lavery et al., 2017).

To test for potential multicollinearity, the variance inflation factor is calculated, which shows the magnitude of correlation between the different explanatory variables. According to

(20)

20

Lavery et al. (2017), when the VIF value is higher than 10, this would indicate a high level of multicollinearity for this predictor variable, which would bias the outcomes of the regression. In table 5 in the appendix, the values of the variance inflation factors are displayed. It is found that the old-age dependency ratio has the highest VIF with 45.74, which would indicate multicollinearity. One solution to solve this multicollinearity, is to remove this variable from the consumption function. In the literature, it was already uncertain whether the old-age dependency ratio would have a clear effect on consumption, so it can be argued whether it is useful to include this ratio from the start. Therefore, to prevent that the regression results will be biased, the old-age dependency ratio is excluded from the regression.

So next, a regression on consumption will be performed on the remaining independent variables. The results of this second regression can be found in table 3. In this case, the regression coefficient on retirement equals 2.6398 and is statistically significant at the five percent significance level. Hence, due to the increase in the retirement age on 1 January 2014, consumption of households has increased significantly by an additional 2.6398 billion euros. As already mentioned, according to Keynes (1936) and Friedman (1957), the coefficient on the binary variable was not expected to be significant. They both argued that the increase in the retirement age would lead to a change in current income for some part of the population, but this effect would already be captured in the coefficient of disposable personal income. Therefore, the fact that the coefficient on retirement turns out to be significantly positive, is not in line with Keynes (1936) and Friedman (1957).

However, the life-cycle hypothesis of Ando and Modigliani (1963) does fit the outcome of this second regression. They agree with Keynes (1936) and Friedman (1957) that current income for some part of the population will change, but they also argue that expected future income will change. Namely, households will realize that they need to work longer, because of the increase in retirement age, and therefore will receive wages for these additional working years. According to Ando and Modigliani (1963), they will try to smooth these additional earnings over their remaining life span, resulting in more consumption. This effect can be found in the significantly positive coefficient of retirement. All in all, the outcome of the second regression is in line with Ando and Modigliani (1963) and indicates that there is indeed a significant change in consumption after the increase in the retirement age on 1 January 2014.

(21)

21

4.2.3. Non-stationary data and spurious regression

When considering time series, it is useful to look at the stationarity in the series. In this case, the distribution of the variables does not change over time, indicating a constant mean, constant variance and constant covariances. However, if the variables are non-stationary, this means they appear to be near random walks (Phillips, 1986). In this thesis, this could hold for the AEX index, since it is found in many studies that prices in the stock market fluctuate randomly (Cooper, 1982).

Moreover, non-stationarity in data leads to the risk of a spurious relationship (Granger & Newbold, 1974). Namely, they found that in most time series regression equations, there is a high degree of fit, measured by R2. However, this could be caused by the fact that an independent variable and a dependent variable follow a trend over time, but are actually unrelated to each other. So, the high value of the degree of fit is misleading. This could also be present in this regression, since consumption and disposable personal income are both increasing on average. Furthermore, Granger and Newbold (1974) argue that spurious regression could lead to estimates of regression coefficients which are inefficient and

significance tests on those coefficients which are invalid. Hence, the coefficient found on the binary variable in the regression of the second effect could be biased and not show the true effect on consumption.

Therefore, the augmented Dickey-Fuller test is used to detect non-stationarity (Stadnytstka, 2010). The simplest case of this test starts with the equation: = +

ℎ = − . The null hypothesis then tests if there is a unit root present, indicating that the variable is non-stationary. The alternative hypothesis states that the time series is stationary. This results in the following testable hypotheses: : = 0 : < 0.

In table 6 in the appendix, the test statistics and the critical values for the one, five and ten percent significance level for the different variables in the regression are displayed. If the absolute value of the test statistic is above the absolute critical value, this would mean that the null hypothesis is rejected and that there is enough evidence to suggest that the variable is stationary. It was found that the value of the test statistic for the real interest rate, the price-to-rent ratio and consumption of households is above the critical value of the ten percent

significance level. This means that the null hypothesis is rejected and there is enough

evidence to suggest that these variables are stationary. However, for all variables, the value of the test statistic is below the critical value of both the one and five significance level, meaning that the null hypothesis is not rejected. Thus, there is not sufficient evidence to suggest that all variables are stationary at the one and five percent significance level.

(22)

22

Therefore to transform non-stationary time series data into stationary data, the method of differencing is used (Strang, Didomenica, Berg, & Mcgorry, 2013). This means that a new time series is derived by subtracting the previous observation from the current observation, so in other words: = − . Hence each variable is integrated of order one for the whole period of 1995 until 2017, namely yt ~ I(1). Next, the augmented Dickey-Fuller test is

performed on the first differences of each variable. It is shown in table 7 in the appendix that for each variable, the value of the test statistic is higher than the critical value of both the one, five, and ten percent significance level. This means that the null hypothesis is rejected and there is enough evidence to suggest that the data are stationary. Finally, a regression of the first difference of consumption of households on the first differences of all independent variables can be executed. This is done for the case including the old-age dependency ratio and for the case without the old-age dependency, because of the multicollinearity found earlier. The results can be found in table 3 under regression 3 and 4.

One remarkable outcome is the fact that the coefficient on the first difference of disposable personal income is negative and significant at the ten percent significance level. This does not fit the theories of Keynes (1936), Ando and Modigliani (1963) and Friedman (1957). Furthermore, three control variables turn out to be not significant. But for this thesis, the result on coefficient of the binary variable representing the increase in retirement age is more useful. For both regression 3 and 4, this coefficient turns out to be negative, but not significant, meaning that there is no significant change in consumption after the increase of the retirement age.

As already mentioned, Keynes (1936) argues that only current income influences consumption. Therefore, the coefficient on the binary variable should be insignificant, which is in line with the findings of regression 3 and 4. However, Ando and Modigliani (1963) wrote that people base their consumption on life time earnings. These earnings will increase, since those people recognize that they have to work longer, which results in more

consumption. For that reason, a positive significant coefficient on the binary variable was expected, so the results of regression 3 and 4 do not fit the theory of Ando and Modigliani (1963). Finally, Friedman (1957) stated that changes in consumption are only based on changes in permanent income, which is already represented by disposable personal income. Therefore it was expected that the coefficient on the binary variable would be insignificant, which is indeed the outcome of the regressions. All in all, the coefficient on the binary variable is insignificant, indicating that there is no significant increase in consumption due to the increase in retirement age, which fits the theories of Keynes (1936) and Friedman (1957).

(23)

23

4.2.4. Heteroskedasticity and autocorrelation

In the previous regressions, it is assumed that the variance of the errors is constant, namely they are homoscedastic. In addition, it is assumed that the errors are uncorrelated with each other, thus no autocorrelation. However, if they turn out to be not constant or not

uncorrelated, this could lead to unbiased, but inefficient parameter estimates (White, 1980). Moreover, if there is heteroskedasticity or autocorrelation, this could lead to wrong results when hypotheses are tested. Namely, the standard errors could be biased, which leads to biased test statistics (White, 1980). Hence, the outcome of the regression with respect to the coefficient on the binary variable and thus the answer on the research question could be incorrect.

To test for heteroskedasticity, a White test is performed on the data used in regression 3 and 4, which is a general test that does not require normality. It is tested under the null hypothesis whether the variance of the errors is constant, while the alternative hypothesis states they are not constant (White, 1980). So : = : ≠ . It is found in table 8 in the appendix that in both regressions, the test statistic is significant at the one percent significance level. This means that the null hypothesis is rejected at this level and there is enough evidence to suggest that there is indeed heteroskedasticity present in both regressions at the one percent significance level.

Next, to test for autocorrelation, a Breusch-Godfrey LM test is performed on the data used in regression 3 and 4. The null hypothesis states that there is no autocorrelation, while the alternative hypothesis states there is autocorrelation (Huo, Kim, Kim, & Lee, 2017).

Hence : , = 0 ≠ : , ≠ 0 ≠ . The results on the

Breusch-Godfrey LM test for both regressions are displayed in table 9 in the appendix. In both cases the test statistic is significant at the one percent significance level, meaning the null hypothesis can be rejected at this level. So, there is enough evidence to suggest that there is indeed autocorrelation present in regression 3 and 4 at the one percent significance level.

To account for both the heteroskedasticity and the autocorrelation, the Newey-West robust standard errors are used in regression 5 and 6. Namely, Newey and West (1987)

developed a covariance matrix estimator which is consistent when both heteroskedasticity and autocorrelation are present. In table 3, the results on regression 5 and 6 can be found. The coefficients remain the same, but it is shown that less of the control variables are significant when the Newey-West standard errors are used. Even when the standard errors are accounted for heteroskedasticity and autocorrelation, the coefficient on the binary variable is still

(24)

24

insignificant in both regression 5 and 6. This means that the main results of regression 3 and 4, which are described at the end of previous section on stationary data and spurious

regression, are still the same. Namely, that there is no significant change in consumption due to the increase in the retirement age, which is in line with Keynes (1936) and Friedman (1957), but was not expected by Ando and Modigliani (1963).

5. Conclusion

In this thesis, it is tested whether consumption of households in the Netherlands changed significantly after the increase of the retirement age on 1 January 2014. Based on three main theories in the literature, two effects influencing consumption due to the increase in the retirement age were expected. First of all, according to Keynes (1936), the life-cycle hypothesis of Ando and Modigliani (1963) and the permanent income hypothesis of Friedman (1957), the increase in the retirement age would lead to an increase in the current income of people with the age of 65 and 66 years old, because it is expected that they will work longer. Subsequently, the increase in the aggregate current income would lead to an increase in the aggregate consumption of households.

The first part of this reasoning is investigated by using a Wald test on data about the net labour participation rate. For both 65 and 66-year-olds, it is found that there was a structural break after 1 January 2014, indicating that relatively more people with those ages had a paid job. The next step was to examine whether this structural break was also coupled with an increase in aggregate current income and subsequently an increase in aggregate household consumption. When current income increases, this would also increase

consumption, according to Keynes (1936), Ando and Modigliani (1963) and Friedman (1957). It was shown in two graphs that both aggregate current income and aggregate household consumption did have an upward trend after 1 January 2014, potentially caused by the increase in the retirement age.

However, those effects are not exactly measured in this thesis, because a strong assumption regarding the factors influencing aggregate current income and consumption is made. When examining the effect of the increase in retirement age on aggregate current income and consumption, it is assumed that all other factors influencing aggregate current income and consumption will remain constant. However, this assumption will not hold in reality, so this is a shortcoming of this thesis. Therefore, a suggestion for further research would be to examine if there is indeed an increase in aggregate current income, due to the fact

(25)

25

that people who are 65 and 66 years old are working for an additional period and if this led to an increase in aggregate consumption.

Moreover, according to Ando and Modigliani (1963), it was expected that households would realize that they had to work longer and would earn higher lifetime earnings. This would make it possible for them to attain a higher level of consumption. To test whether this second effect did indeed happen, a binary variable was created to account for this effect and a regression on consumption was performed using this variable and other control variables. It was found that the coefficient on the binary variable was not significant, which did not suit Ando and Modigliani (1963), but was in line with Keynes (1936) and Friedman (1957). However, after controlling for multicollinearity, the coefficient on the binary variable was significant and positive, indicating that there was a significant change in consumption due to this second effect, being in line with Ando and Modigliani (1963). But, when controlling for non-stationarity, heteroskedasticity and autocorrelation, the coefficient on the binary variable was not significant again, implying that consumption did not increase due to the increase in the retirement age.

All in all, it is hard to tell whether consumption of Dutch households have increased after the increase in the retirement age, because there are many different factors influencing consumption and the regressions possibly suffer from ommited variable bias. Moreover, following the reasoning of Hall (1978), it is possible that households already changed their consumption at the moment that the potential increase in the retirement age came up in the news and not when it was indeed implemented. Therefore, future research could be aimed at examining whether the findings of Hall (1978) also apply to the consumption behavior of Dutch households with respect to the change in the retirement age.

(26)

26

References Adams, N. A. (1971). Dependency Rates and Savings Rates: Comment. The American

Economic Review, 61(3), 472-475.

Anderson, K. M., & Kaeding, M. (2015). European Integration and Pension Policy Change: Variable Patterns of Europeanization in Italy, the Netherlands and Belgium. British Journal of Industrial Relations, 53(2), 231-253. doi: 10.1111/bjir.12030

Ando, A., & Modigliani, F. (1963). The “life cycle” hypothesis of saving: aggregate implications and tests, The American Economic Review, 53(1), 55-84.

André, C., Gil-Alana, L. A., & Gupta, R. (2014). Testing for persistence in housing price-to- income and price-to-rent ratios in 16 OECD countries. Applied Economics, 46(18), 2127-2138. doi: 10.1080/00036846.2014.896988

Beetsma, R., Bettendorf, L., & Broer, P. (2003). The budgeting and economic consequences of ageing in the Netherlands. Economic Modelling, 20(5), 987-1013.

doi: 10.1016/S0264-9993(02)00062-7

Carroll, C. D. (1994). How does future income affect current consumption? The Quarterly Journal of Economics, 109(1), 111-147. doi: 10.2307/2118430

Central Bureau for Statistics. (2018, May 15). Arbeidsdeelname; ouderen (Data file). Retrieved from: http://statline.cbs.nl/Statweb/

Cooper, J. C. B. (1982). World Stock Markets: Some Random Walk Tests. Applied Econometrics, 14, 515-531. doi:1080/0003-6846/82/050515

D’Orlando, F., & Sanfilippo, E. (2010). Behavioral foundations for the Keynesian consumption function. Journal of Economic Psychology, 31(6), 1036-1046. doi: 10.1016/j.joep.2010.09.004

Edwards, S. (1996). Why are Latin America’s savings rates so low? An international comparative analysis. Journal of Development Economics, 51(1), 5-44. doi: 10.1016/S0304-3878(96)00424-5

Eisner, R. (1958). The Permanent Income Hypothesis: Comment. The American Economic Review, 48(5), 972-990.

Flavin, M. A. (1981). The Adjustment of Consumption to Changing Expectations about Future Income. Journal of Political Economy, 89(5), 974-1009. doi: 10.1086/261016 Friedman, M. (1957). A Theory of the Consumption Function. Princeton: Princeton University

(27)

27

Goldberger, A. S. (1973). Dependency Rates and Savings Rates: Further Comment. The American Economic Review, 63(1), 232-233.

Granger, C. W. J., & Newbold, P. (1974). Spurious regressions in econometrics. Journal of Econometrics, 2(2), 111-120. doi: 10.1016/0304-4076(74)90034-7

Hall, R. E. (1978). Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: Theory and Evidence. Journal of Policital Economy, 86(6), 971-987.

doi: 10.1086/260724

Holden, G. R. (1938). Mr. Keynes’ Consumption Function and The Time-Preference

Postulate. The Quarterly Journal of Economics, 52(2), 281-296. doi: 10.2307/1881735 Houthakker, H. S. (1958). The Permanent Income Hypothesis, a review article. The American

Economic Review, 48(3), 396-404.

Huo, L., Kim, T., Kim, Y., & Lee, D. J. (2017). A residual-based test for autocorrelation in quantile regression models. Journal of Statistical Computation and Simulation, 87(7), 1305-1322. doi: 10.1080/00949655.2016.1262371

Karatzas, I., Shubik, M., Sudderth, W., & Geanakoplos, J. (2006). The inflationary bias of real uncertainty and the harmonic Fisher equation. Economic Theory, 28(3), 481-512. doi: 10.1007/s00199-005-0648-z

Keynes, J. M. (1936). The General Theory of Employment, Interest and Money. London: Palgrave Macmillan

Lavery, M. R., Acharya, P., Sivo, S. A., & Xu, L. (2017). Number of predictors and multicollinearity: What are their effects on error and bias in regression? Communications in Statistics – Simulations and Computation, 0(0), 1-12. doi: 10.1080/03610918.2017.1371750

Leff, N. H. (1969). Dependency Rates and Savings Rates. The American Economic Review, 59(5), 886-896.

Leland, H. E. (1968). Saving and Uncertainty: the Precautionary Demand for Saving. The Quarterly Journal of economics, 83(2), 465-473. doi: 10.2307/1879518

Macdonald, G., Mullineux, A., & Sensarma, R. (2011). Asymmetric effects of interest rate changes: the role of the consumption-wealth channel. Applied Economics, 43(16), 1991-2001. doi: 10.1080/00036840902950572

Muszyńska, M., & Rau, R. (2012). The Old-Age Healthy Dependency Ratio in Europe. Journal of Population Ageing, 5(3), 151-162. doi: 10.1007/s12062-012-9068-6

Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55(3), 703-708.

(28)

28

doi: 10.2307/1913610

Organisation for Economic Co-operation and Development (OECD). 2018. Analytical house prices indicators. Paris: OECD. Centre for Educational Research and Innovation. Oxford Economics. (2018, May 2018). Unemployment rate (Data file).

Retrieved from: https://www.oxfordeconomics.com/

Phillips, P. C. B. (1986). Understanding spurious regressions in econometrics. Journal of Econometrics, 33(3), 311-340. doi: 10.1016/0304-4076(86)90001-1

Piehl, A. M., C, S. J., Braga, A. A., & Kennedy, D. M. (2003). Testing for Structural Breaks in the Evaluation of Programs. Review of Economics and Statistics, 85(3), 550-558. doi: 10.1162/003465303322369713

Stadnytska, T. (2010). Deterministic or Stochastic Trend: Decision on the Basis of the

Augmented Dickey-Fuller test. Methodology: European Journal of Research Methods for the Behavioral and Social Sciences, 6(2), 83-92. doi: 10.1027/1614-2241/a000009 Statistical Office of the European Communities. (1990). EUROSTAT: Regional statistics:

Reference guide. Luxembourg: Eurostat

Strang, A. J., Didomenico, A., Berg, W. P., & Mcgorry, R. W. (2013). Assessment of

differenced center of pressure time series improves detection of age-related changes in postural coordination. Gait & Posture, 38(2), 345-348.

doi: 10.1016/j.gaitpost.2012.11.006

Van der Zwan, N. (2017). Financialisation and the Pension System: Lessons from the United States and the Netherlands. Journal of Modern European History, 15(4), 554-578. White, H. (1980). A Heteteroskedasticity-Consistent Covariance Matrix Estimator and a

Direct Test for Heteroskedasticity. Econometrica, 48(4), 817-838. doi: 10.2307/1912934

Wilcox, D. W. (1989). Social Security Benefits, Consumption Expenditure, and the Life Cycle Hypothesis. Journal of Political Economy, 97(2), 288-304. doi 10.1086/261604

(29)

29

Appendix

Table 4. Correlation matrix of the independent variables

Variable Real House AEX DPI Unempl Dep Retire

Real 1.0000 House .1276 1.0000 AEX .1670 .2105 1.0000 DPI -.6721 .4078 .1038 1.0000 Unempl -.5651 -3806 -.5208 .3222 1.0000 Dep -.7940 -.3111 -.0048 .7224 .5883 1.0000 Retire -.5966 -.2878 .1941 .4374 .5536 .8477 1.000

Table 5. Variance inflation factor for the consumption function

Variable VIF 1/VIF

Real 3.25 .307963 House 19.19 .052107 AEX 2.06 .486463 DPI 36.07 .027724 Unempl 2.84 .352571 Dep 45.74 .021861 Retire 5.15 .194236

Table 6. Augmented Dickey-Fuller test for a unit root Variable Test statistic 1% Critical Value 5% Critical value 10% Critical value P-value Real interest -2.687 -3.523 -2.897 -2.584 .0764 House -2.793 -3.523 -2.897 -2.584 .0592 AEX -2.273 -3.523 -2.897 -2.584 .1809 DPI -1.148 -2.369 -1.662 -1.291 .1270 Unemployment -1.171 -3.523 -2.897 -2.897 .6860 (Dependency) -1.146 -4.060 -3.459 -3.155 .9209 Consumption -1.341 -2.369 -1.662 -1.291 .0916

(30)

30

Table 7. Augmented Dickey-Fuller test for a unit root on first difference

Variable Test statistic 1% Critical Value 5% Critical value 10% Critical value P-value ΔReal interest -11.019 -3.524 -2.898 -2.584 .0000 ΔHouse -3.690 -3.524 -2.898 -2.584 .0043 ΔAEX -9.972 -3.524 -2.898 -2.584 .0000 ΔDPI -26.714 -3.524 -2.898 -2.584 .0000 ΔUnemployment -4.072 -3.524 -2.898 -2.584 .0011 Δ(Dependency) -4.167 -3.524 -2.898 -2.584 .0007 ΔConsumption -17.760 -3.524 -2.898 -2.854 .0000

Table 8. White test on heteroskedasticity

Regression χ2 value df P-value

3 56.27 34 .0095

4 47.28 26 .0065

Table 9. Breusch-Godfrey LM test on autocorrelation

Regression χ2 value df P-value

3 27.232 1 .0000

The effects of the increase in the retirement age on Dutch household consumption