Social Security Wealth and Household Consumption: Evidence from recent pension reforms in the Netherlands

Social Security Wealth and Household Consumption:

Evidence from recent pension reforms in the Netherlands

Luuk Dieteren

January 19, 2018

Abstract: The Dutch pensions system was reformed between 2010 and 2015. One of the major

changes is the gradual increase of the mandatory retirement age from 65 to 67. Using household consumption data provided by the LISS Panel, this paper sets out to investigate the displacement effect pension wealth has on household consumption. Pension wealth will be based on the estimated social security wealth for households, which is entitled to all Dutch citizens under the General Old Age Act. Empirical evidence points out that there is less than perfect offset between pension wealth and consumption. This study corresponds to these findings and finds a less than perfect offset for Dutch households, where the displacement effect differs across households belonging to different levels of education or age cohorts. Changes in pension wealth are not fully incorporated by Dutch household into their consumption behavior. The variation in the displacement effect across households has policy implications, as changes in pension wealth due to a pension reform affect the consumption behavior of different household groups and generations in different ways.

Keywords: life-cycle model, consumption behavior, social security wealth, displacement effect JEL Classification: D11, D12, D15, H55

a _{This paper is submitted as thesis for the MSc. Economics for the Faculty of Economics and Business of the}

University of Groningen (RUG), supervised by prof. dr. R.J.M. (Rob) Alessie.

b _{In this paper use is made of data of the LISS (Longitudinal Internet Studies for the Social sciences) panel}

administered by CentERdata (Tilburg University, The Netherlands).

c _{I would like to thank Rob Alessie for all his help and support during the process of writing this paper and Mike}

2

1. Introduction

The combination of an ageing population and the continuous increase in the lifetime expectancy has and will remain to put pressure on pension systems across the world. This is also the case for the Netherlands, which has a pension system that has always been regarded as one of the most balanced, yet generous pension systems. However, even well-structured pension systems do need to adapt to demographic changes. Thus, the ageing population in combination with the increasing lifetime expectancy, did put severe pressure on the Dutch pension system. Their pension system, while praised for being generous, seemed to become unstable due to these demographic changes. Especially during and after the financial crisis in the first decade of the 21st century, the instability of the Dutch pension system became highlighted. Social security benefits in the Netherlands are financed with a pay-as-you-go system. Therefore, for a large part, the current working population in the Netherlands is financing the pension benefits for the elderly. The remaining part of these social security benefits are financed by the government, which indirectly makes the elderly partially pay for their own benefits as the government budget is mainly financed by taxes paid by the Dutch population. The combination of an ageing population and the increasing lifetime expectancy thus induced a rise in the total pension benefits necessary to provide pension benefits for all elderly. This has put pressure on both parts of the required contributions for the pension system, the part financed by the Dutch working population and the part financed by the Dutch government. Especially during the financial crisis, when households already experienced a considerable drop in their purchasing power and the government had to implement severe budget cuts, the instability of the Dutch pension system was highlighted. Moreover, the aftermath of the financial crisis triggered a severe drop in the interest rate in the Netherlands, which fell from 4.3% in 2007 to less than 1% in 2015. Antolin et al. (2011) point out that a protracted period of low interest rates is a feasible scenario for countries such as the Netherlands. Therefore, in 2010, former Dutch Minister of Social Affairs and Employment, Piet Hein Donner, shared the findings of Commissie Goudswaard. Commissie Goudswaard (2010) found that the Dutch pension system was insufficiently future-proof and that the system was in need of a reform.

3 Moreover, the decreasing interest rate diminished expected returns on pension contributions and thus severely worsened the solvency ratio of Dutch pension funds. The lower solvency rates gave rise to a growing uncertainty regarding future pension benefits. All these changes led to a comprehensive pension reform of the Dutch pension system. One of the major changes in the Dutch pension system is the gradual increase of the mandatory retirement age from 65 to 67. Further aspects of the reform will be discussed in more detail in Section 3. Subsequently, the increase in the mandatory retirement age has led to changes in expected pension wealth for households in the Netherlands. These changes in future pension entitlements might incentivize households to alter their current consumption and saving behavior. This paper is set out to study the effect that changes in future pension benefits will have on current consumption and saving behavior.

To investigate the impact of these changes, this paper will employ the life-cycle model. The life-cycle model was first introduced by Modigliani and Brumberg (1954). This well-known model explains the behavior of individuals when it comes to consumption choices during their lifetime. Ever since its introduction, the life-cycle model has been at the basis of empirical studies that tried to investigate the effect of changes in pension wealth on consumption choices.

As mentioned, this paper is set out to investigate the changes in expected pension wealth that the recent pension reforms had in the Netherlands and how these changes affect consumption and saving behavior of Dutch households. To narrow the investigation, this study will focus only on social security wealth, i.e. the Dutch first pillar in its pension system. This first pillar consists of a pay-as-you-go flat-rate defined benefit which is accrued at 2 percent per year an individual lives in the Netherlands. The benefit is linked to the net minimum wage, does not depend on income but does depend on marital status. In order to receive their accrued social security benefits an individual needs to reach the mandatory retirement age. Since the recent pension reforms are set out to increase the mandatory retirement age in the coming years, individuals will alter their expectations on future pension wealth. In accordance with the life-cycle model, this change in lifetime resources will influence current consumption and saving behavior. This paper thus sets out to answer the following question:

‘Did Dutch households alter their consumption and saving behavior in response to changes in their expected social security wealth due to the Dutch pension reforms?’

4 University. The LISS panel offers individual and household data on consumption choices. In addition, it contains a large set of background variables to accompany this consumption data. The time period of this dataset is 2009-2016, which provides a clear opportunity to analyze the effects of the pension reforms which were discussed and implemented during this time period.

The research question this study tries to answer has social policy implications. The life-cycle hypothesis suggests that individuals accumulate wealth by saving during their working years. The accumulated wealth is used to maintain consumption after retirement, thus after income dropped to zero. This way, an individual tries to smooth consumption over lifetime and thus tries to maintain the same level of consumption in each period irrespective of income in that period. Governments are able to intervene in this self-insurance process by obliging individuals to pay for social security contributions during their working years and to receive social security benefits after retirement. This takes away the incentive to self-insure for retirement, i.e. financial wealth is being substituted for pension wealth. This substitution effect could lead to crowding out of retirement savings, where pension wealth is just a direct substitute for financial wealth saved for retirement. However, there is wide empirical evidence that the trade-off between financial wealth and pension wealth is not one for one. In other words, a drop in future expected pension wealth does not result in the same, though properly discounted, increase in current wealth accumulation. Throughout the paper, this trade-off between pension wealth and financial will be presented as either the replacement rate or displacement effect. According to Alessie et al. (2013), for policymakers it is crucial to understand the effect that pension reforms will have on household and national saving, especially the effect of changes in pension wealth on private wealth, in order to correctly assess the welfare effects of a pension reform.

5 to investigate the displacement effect of pension wealth for different household groups within the Dutch society. Once again, the replacement rate of pension wealth is less than 100% for all different groups of households, but we are unable to find values of the replacement rate that are significantly different from 0. The results do indicate significant differences in the displacement effect for different groups of households. The displacement effect of pension wealth seems to be larger for households with a higher level of formal education and for households in the old age group. The different results for different groups of households lead to some policy implications. In line with other empirical evidence, the displacement effect of pension wealth appears to be lower for groups of households with a larger dependency on social security wealth. The Dutch pension reform therefore could have a larger negative effect on income after retirement for the lower classes of society, who already struggle to make ends meet after the financial crisis. Thus, policy makers should be aware that the displacement effect of pension wealth is far from a perfect offset and is lower for households with a larger dependency on social security wealth.

6 only have an impact on social security wealth, but also on lifetime income as individuals will have more working years and thus more years of income. Therefore, this study analyzes the effect of the 2012 Dutch pension reform on two different sources of wealth that affect consumption and saving behavior. This study is first in analyzing the effects of a pension reform, which increases the mandatory retirement age, on the replacement rate of pension wealth by employing surveyed household consumption data. Therefore, this paper could be a stepping stone into further research on the effect of similar pension reforms around the world. In line with the Dutch pension reforms, multiple other (European) countries are implementing or at least debating equivalent changes in the mandatory retirement age in order to combat the growing imbalance in their pension systems.

The paper will be organized in the following order: Section 2 will present the theoretical model and guides us through the derivations in order to reach the econometric model used for the empirical analysis. Section 3 will give an overview of the Dutch pension system and a detailed description of the recent pension reform. Section 4 will discuss the data and the assumptions that have been made in order to estimate lifetime income and expected pension wealth. Section 5 will showcase the results, which will be further discussed in Section 6. Section 7 concludes.

2. Theoretical model

7 extended by other empirical studies (e.g. Diamond and Hausman [1984] and Hubbard [1986] on US data and Brugiavini [1987] and Jappelli [1995] on Italian data). These studies all find a significant degree of substitutability between pension wealth and financial wealth, though the degree of substitutability varies in each of these studies.

Alessie et al. (2013) point out that these earlier studies on the displacement effect typically regress non-pension wealth on cash earnings and pension wealth, including some control variables. However, Gale (1998) shows that these regressions usually estimate degrees of the displacement effect with a downward bias. He introduces an age-specific adjustment factor, known as Gale’s Q, to adjust pension wealth based on an individual’s position in its life cycle. Attanasio and Rohwedder (2003) use this adjustment factor to analyze the effect of pension reforms in the UK, while Attanasio and Brugiavini (2003) apply the same method for a major pension reform in Italy. Both papers conclude that the displacement effect of pension wealth differs significantly across households. Kapteyn et al. (2005) investigate the determinants of wealth holdings for different cohorts in the Netherlands, while Alessie et al. (1995) determine that wealth holdings across Dutch elderly are unevenly distributed. There is limited evidence that the displacement effect in the Netherlands is less than perfect offset (e.g. Alessie et al. [1997] and Kapteyn et al. [2005]). Moreover, Alessie et al. (1997) introduce a set of formulae used to derive social security wealth for Dutch household. This study will utilize an updated version in the calculations of social security wealth for our household panel.

2.1 The life-cycle model

8 The basis of the simple life-cycle model is an isoelastic utility function with Constant Relative Risk Aversion (CRRA). The consumer tries to maximize this utility function subject to the lifetime budget constraint:

max 𝑐𝜏 ∑ (1 + 𝜌) 𝑡 − 𝜏 _𝑎 𝜏𝑡 𝑐_𝜏1 − 𝛾 1 − 𝛾 𝐿𝑚𝑎𝑥 𝜏 = 𝑡 (1) s.t. ∑ (1 + 𝑟)𝑡 − 𝜏_𝑐 𝜏 𝐿𝑚𝑎𝑥 𝜏 = 𝑡 = (1 + 𝑟)𝐴𝑡−1+ ∑ (1 + 𝑟)𝑡 − 𝜏 𝑦𝜏 𝐿𝑚𝑎𝑥 𝜏 = 𝑡 (2) ∑ (1 + 𝑟)𝑡 − 𝜏_𝑐 𝜏 𝐿𝑚𝑎𝑥 𝜏 = 𝑡 = (1 + 𝑟)𝐴𝑡 − 1+ ∑(1 + 𝑟)𝑡 − 𝜏 𝐸𝜏 𝑠−1 𝜏 = 𝑡 + ∑ (1 + 𝑟)𝑡− 𝜏 _𝐵 𝜏 𝐿𝑚𝑎𝑥 𝜏=𝑠 (3)

where cτ represents consumption at age τ, yτ denotes income at age τ and consists of At – 1, which is accumulated wealth up to period t, Eτ, which are pre-retirement earnings, and Bτ, which are the future pension benefits. Furthermore, ρ denotes the discount rate, Lmax is the maximum age of the agent, s is the mandatory retirement age and γ is the coefficient of relative risk aversion. The optimization problem does include survival probabilities, 𝑎_𝜏𝑡_{, which gives an individual’s} survival probability up to period τ given period t has been reached.

Combining and rearranging the first-order conditions gives a representation of the consumption path for the agent:

𝑐𝜏 = 𝑐𝑡[( 1 + 𝑟 1 + 𝜌) (𝜏−𝑡) 𝛾⁄ (𝑎𝜏𝑡) 1 𝛾⁄ ] (4)

The consumption path given in Eq. (4) can be implemented in the lifetime budget constraint given in Eq. (2). This yields the following equation:

9 Using Eq. (3), Eq. (5) could be rewritten:

𝑐_𝑡 = ( ∑ 𝜆𝜏−𝑡_(𝑎 𝜏 𝑡₎1 𝛾⁄ 𝐿𝑚𝑎𝑥 𝜏=𝑡 ) −1 [(1 + 𝑟)𝐴_𝑡−1+ ∑(1 + 𝑟)𝑡−𝜏 𝑠−1 𝜏=𝑡 𝐸_𝜏 + ∑ (1 + 𝑟)𝑡−𝜏_𝐵 𝜏 𝐿𝑚𝑎𝑥 𝜏=𝑠 ] (6)

where ∑𝑠−1𝜏=𝑡(1 + 𝑟)𝑡−𝜏𝐸𝜏 denotes total earnings from period t up till the mandatory retirement age s and ∑𝐿𝜏=𝑠𝑚𝑎𝑥(1 + 𝑟)𝑡− 𝜏 𝐵𝜏 denotes accumulated pension wealth at age t, i.e. the present value of expected pension benefits.

Expression (6) leads to the following empirical equation which will be used to estimate the model for all retired and non-retired households:

𝑐_𝑖,𝑡 = 𝛽₀+ 𝛽₁𝜙_𝑖,𝑡𝐴𝑊_𝑖,𝑡 + 𝛽₂𝜙_𝑖,𝑡𝐿𝐼_𝑖,𝑡+ 𝛽₃𝜙_𝑖,𝑡𝑆𝑆𝑊_𝑖,𝑡 + 𝛽₄𝑋_𝑖,𝑡+ 𝜀_𝑖,𝑡 (7)

where

𝑐_𝑖,𝑡 = annual household non-durable consumption at period t 𝜙𝑖,𝑡 = (∑𝐿𝜏=𝑡𝑚𝑎𝑥𝜆𝜏−𝑡(𝑎𝜏𝑡)1 𝛾⁄ ) −1 𝐴𝑊_𝑖,𝑡 = (1 + 𝑟)𝐴_𝑡−1 (Accumulated Wealth) 𝐿𝐼_𝑖,𝑡 = ∑𝑠−1(1 + 𝑟)𝑡−𝜏 𝜏=𝑡 𝑦𝑡 (Lifetime Income)1 𝑆𝑆𝑊_𝑖,𝑡 = ∑ (1 + 𝑟)𝑡−𝜏_𝐵 𝜏 𝐿𝑚𝑎𝑥

𝜏=𝑠 (Social Security Wealth)2

𝑋𝑖,𝑡 = vector of demographic household characteristics that might affect consumption

In this econometric model, the coefficient of interest for this study is β3. The coefficient β3 reflects the replacement rate between pension wealth and financial wealth. The standard life-cycle model predicts the value of this coefficient to be exactly minus 1. There is a perfect tradeoff between pension wealth and financial wealth, i.e. a €1 drop in expected pension wealth would lead to a €1 increase in financial wealth. This implies that the agent will save more to accumulate the extra financial wealth and thus will reduce its consumption path. This is the result of solving the life-cycle hypothesis, which indicates that the agent tries to smooth consumption over lifetime. However, government interventions like state provided social security schemes for retirees, distort the tradeoff effect of the life-cycle model. Moreover, there is wide empirical evidence that the replacement rate is not minus 1 but instead lies between minus 1 and 0. This

10 indicates a less than perfect tradeoff between pension wealth and financial wealth and thus indicates that individuals are not able or not willing to fully incorporate changes in future wealth into their current saving and consumption behavior.

Contrary to other empirical studies on the replacement rate of pension wealth and financial wealth, this study uses household consumption data rather than household saving data. We assume a perfect tradeoff between household saving and consumption, i.e. every Euro consumed cannot be saved and vice versa. Thus, in our empirical model, we will test the replacement rate β3 with the following hypothesis:

𝐻

: 𝛽

= 1

𝐻

: 𝛽

< 1

The expected sign of β3 will be lower than 1, or to be more precise between 0 and 1. This would imply a less than perfect tradeoff between pension wealth and financial wealth. In line with other empirical evidence this study thus suggests that households are unable and/or unwilling to fully adjust current consumption and saving behavior in response to changes in expected future pension wealth. However, as mentioned, this study diverts from earlier empirical work by using consumption data on households rather than saving data.

3. The Dutch pension system

3.1 The three pillars of the Dutch pension system

The Dutch pension system consists of three separate pillars and, albeit considered one of the best in the world, appeared to become unstable as a results of demographic developments. According to Delsen (2016), the much faster than expected increase in life expectancy, the financial crisis and subsequent protracted period of low interest rates, the policy responses to them and the lack of trust have raised concerns on the sustainability of the pension system. Moreover, several reports have questioned the current balance between funded and pay-as-you-go pensions, i.a. De Deken and Maarse (2013) and Beetsma et al. (2015).

3.1.1. First pillar

11 earnings history, but does depend on years lived in the Netherlands between the age of 15 and 65. For each year an individual lives in the Netherlands it accrues 2% of total AOW pension benefits, thus maximum AOW benefits are obtained by residing at least 50 years in the Netherlands. Furthermore, the first pillar benefits depend on the type of household. Individuals living in a single household will receive 70% of the net minimum wage while in a shared household each spouse will receive 50% of the net minimum wage. The majority of shared households consists of married couples or at least couples who are officially registered as partners. However, there are also cases where two individuals are not officially registered as partners but do have a shared household. In such cases, both individuals are most likely also entitled to only 50% of net minimum wage. In the Netherlands, the Social Insurance Bank (Sociale Verzekeringsbank, SVB) provides guidelines for individuals to figure out their exact future pension entitlements. Furthermore, the SVB issues regular inspections to verify that elderly are receiving the correct amount of AOW benefits.3

The AOW benefits are based on a pay-as-you-go system and thus are financed by contributions made by the current working force in the Netherlands. Elderly receiving AOW benefits are no longer obliged to pay these contributions. These contributions are levied as part of the income tax system, where the contribution rate was 17.9% of annual income up to a maximum of €6,012 in 20154_{. However, the contributions levied on income earned by the current} working population are insufficient to fully finance the AOW pension benefits. Therefore, the gap between these contributions and benefits is covered by the Dutch government, out of the general state budget. Consequently, the elderly do indirectly pay a small part of their own pension benefits since part of the AOW benefits are financed by general taxes levied by the government. This gap is one of the main reasons the Dutch pension system became increasingly unstable in recent years. According to De Deken and Maarse (2013), out of the €33 billion needed to provide the pension benefits in 2013, only €23 billion was covered by contributions levied on income of the current workforce. Therefore, it is clear to see that this first pillar is susceptible to the combination of aging in the Netherlands and the increasing life expectancy. This has led to a larger share of retirees among the Dutch population, which all need additional years of pension benefits as they live longer than was anticipated. In order to meet the growth in total AOW benefits, either the AOW contributions by the working population need to increase or a larger part of the general state budget needs to be addressed to close the gap between pension

12 contributions and benefits. The former will create an even larger sense of legacy debt: the elderly receive larger amounts of pension benefits while they have contributed for smaller amounts during their working years, since the life expectancy was significantly smaller at the time. The latter will put extra pressure on the general state budget, while the Dutch government already implemented serious budget cuts during the financial crisis in order to mitigate the increasing budget deficit.

3.1.2 Second pillar

The second pillar of the Dutch pension system consists of occupational pensions schemes. Since the second pillar is based on an individual’s lifetime earnings, this pillar provides income maintenance during retirement. These occupational schemes are collective agreements negotiated by the social partners. And while formally these pension schemes are voluntary contracts, the fact that they are invariably subject to administrative extension by the government make them de facto mandating (De Deken and Maarse, 2013). Thus, most employers offer an occupational pension scheme and when they offer a scheme participation is mandatory. Therefore, around 90% of the Dutch employees participate in a second pillar pension scheme (Delsen, 2016).

The second pillar pension schemes are fully funded systems: current savings fully finance future benefits. The pension contributions made are invested in assets and the returns will not be granted as benefits until retirement. Most schemes are labeled as defined benefit (DB) systems, where future pension benefits are predetermined by a formula. This will yield a certain rate of return on contributions and the pension funds will bear the risk. However, the financial crisis, extremely low interest rate and underestimation of the life expectancy have led to an underfunding of these pension schemes. Antolin et al. (2011) point out that the combination of a low interest rate environment and the underestimation of the life expectancy further worsens the solvency situation of annuity providers and DB pension funds, as the low interest rates magnify the present value of future increases in longevity. Therefore, in accordance with recommendations by the Commissie Goudswaard, the labor unions and employers agreed on a shift towards defined contribution (DC) systems, shifting the financial risks from the pension funds towards the participants (Delsen, 2012).

3.1.3 Third pillar

13 of individual annuity insurances or pension arrangements which are used to top up statutory pensions or to compensate for interruptions in residence that reduce basis state pension entitlements (De Deken and Maarse, 2013). Contributions to these individual pension schemes are mostly tax deductible, provided that the third pillar pension scheme does not increase future pension benefits over 70% of final wage. However, most second pillar occupational pensions already have an ambition level of 70% of final wage and around 90% of Dutch employees participate in such occupational pension schemes. Thus, this leaves little room for the third pillar. In fact, Mastrogiacomo and Alessie (2011) show that third pillar free pension savings are of little importance to the median Dutch citizen.

3.2 Pension reforms

First talks about major reforms to the Dutch pension system started back in 2009 by the Balkenende IV cabinet. The statutory retirement age had been fixed at the age of 65 ever since the introduction of the AOW in 1957. However, in 2009, the government offered the proposition to increase the mandatory retirement age with a two-step approach: the mandatory retirement was to be increased to 66 by 2020 and 67 by 2025. The increase in the mandatory retirement age was aimed at reducing public expenditures in order to mitigate the increasing budget deficit during the financial crisis. In fact, the expected changes would improve the sustainability of the pension system by 0.7% of GDP, which equaled public savings of € 4.5 billion, and would considerably improve employment rates (CPB, 2010: 109-110). Moreover, as the mandatory retirement age was to increase, second pillar pension accruals would improve in line with the extra years of employment. In order to counteract this effect, the maximum fiscally facilitated accrual rate for second pillar pensions was to be lowered from 2010 onward: from 2.25% to 2.15% for average pay schemes and from 2% to 1.9% for final pay schemes (Delsen, 2016). However, the Balkenende IV cabinet fell in the spring of 2010 and thus the reforms were postponed.

14 2016). Therefore, the Rutte I minority cabinet together with opposition parties came to the so-called Spring Agreement (Lenteakkoord). The statutory retirement age would by gradually increased to 66 in 2019 and 67 in 2023. Furthermore, after 2023, the statutory retirement age would be linked to the life expectancy which would suggest an even further increase in the retirement age after 2023. Eventually, the law to increase the statutory retirement age was implemented in June 2012. However, in June 2015, the Rutte II cabinet implemented an acceleration in the gradually increasing retirement age. The statutory retirement age would already be 66 in 2018 and 67 in 2021, and after 2021 would still be linked to the life expectancy. The acting Minister of Social Affairs and Employment has to announce any further increases in the statutory retirement age at least 5 years prior to the actual change. This has recently been announced for 2022, where the statutory retirement age will be 67 years and 3 months. Table 1 shows the exact statutory retirement age in the Netherlands for the years 2013-2022. According to Delsen (2016), the estimated cumulated savings of this acceleration of the increasing mandatory retirement age will be around €2.9 billion in the period 2016-2024; cumulative tax revenues and contributions will increase with €770 million.

2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Statutory retirement age 65 + 1 month 65 + 2 months 65 + 3 months 65 + 6 months 65 + 9 months 66 66 + 4 months 66 + 8 months 67 67 + 3 months

Table 1. Statutory retirement age in the Netherlands in the period 2013-2022

The gradually increasing statutory retirement age also has implications for second pillar pension plans. The AOW pension benefits are received once the statutory retirement age is reached. It still is possible for individuals to retire earlier, i.e. to retire before the statutory retirement age. However, these individuals will receive zero AOW benefits up to the statutory retirement age and will have to compensate this drop in income. Therefore, retirement ages in the second pillar will have to be kept in line with the increasing mandatory retirement age. According to Delsen (2016), the increase in the retirement age also implies a reduction in the accrual rates that may incite to retire early. In fact, the Rutte I cabinet did reduce the accrual rates for occupational pension schemes. The accrual rate for average wage plans reduced from 2.25% in 2013 to 1.875% in 2015 and the accrual rate for final wage plans reduced from 2% in 2013 to 1.675% in 2015.5 This would mean people will have to contribute for more years to their second pillar plan in order to accumulate a full second pillar pension, or the replacement rate of a full

15 second pillar pension will become much lower. The reasoning for the reduction in accrual rates was the increase in the mandatory retirement age: people will have to work longer and thus are able to contribute for a longer period and would still receive a respectable pension. Moreover, this would allow for a reduction of the contribution rates which was one of the recommendations from the Commissie Goudswaard.

The gradual increase in the mandatory retirement age will have a major impact on people’s pension wealth. Individuals form expectations about their future pension benefits, and these expectations will influence current consumption and saving behavior of households. Therefore, this study will take the gradually increasing statutory retirement age as the pension reform that will influence expected pension wealth. Subsequently, this study tries to analyze the effect these changes in expectations will have on current consumption behavior. Section 4 will give a more detailed description of the data used and how the increasing retirement age will be influential in this analysis.

4. Data

As mentioned, there is already a wide variety of empirical studies on the replacement rate between pension wealth and financial wealth. However, studies like Attanasio and Brugiavini (2003) and Attanasio and Rohwedder (2003) use a household’s saving rate as dependent variable, while studies like Alessie et al. (2013) use non-pension wealth in order to study the substitutability between pension wealth and non-pension wealth. This study differs as it is set out to investigate this effect with household consumption data, rather than data on saving or accumulated wealth. Using household consumption data to investigate the effect of the Dutch pension reform, i.e. the gradually increasing statutory retirement age, has not yet been done before. Moreover, there is generally a lack of empirical evidence on the substitutability of pension wealth and financial wealth using consumption data. This might be explained by the fact that it is difficult to derive sufficient high quality data on household consumption.

16 she gets paid, in order to reduce potential errors in measurement. Moreover, additional household data is provided by one member of every household. The regularity of the online questionnaires in combination with the offered compensation induces individuals to fill out each survey with care. Therefore, the LISS Panel provides data on the individual and household level of both high quality and frequency.6

The surveys for the LISS Panel are conducted by CentERdata at the University of Tilburg. The current period of the longitudinal surveys is from November 2007 to October 2017, though for this study the relevant period will be from 2009 to 2015. Part of the interview time for the LISS Panel is devoted to their Core Study, which is conducted on a yearly basis to keep up with each panel member’s life course and changes in living situation. In addition to this Core Study, the LISS Panel also provides smaller samples of longitudinal panel data to further amplify different research areas. One of these Assembled Studies is a longitudinal survey on “Time Use and Consumption”. This survey currently consists of five waves of data: 2009, 2010, 2012, 2015 and the recently added 2017. This study will implement the first four waves of this survey, as the most recent wave of 2017 was added during the research period. The “Time Use and Consumption” survey provides ample data on consumption expenditures, both on the individual level as well as aggregated to the household level. Furthermore, the LISS Panel provides a monthly survey on “Background Variables” which provides data on background characteristics, again on both the individual and the household level. The “Background Variables” panel also includes information of net monthly income on the household level. In addition, one of the surveys of the Core Study, “Economic Situation: Assets”, provides information on wealth accumulation by households up until the period of each survey. These three panel datasets will be used to estimate part of the relevant variables used to study the effect of the reform on the replacement rate of pension wealth and financial wealth.

17 Recall the relevant econometric model that will be used to study this effect, Eq. (7):

𝑐

= 𝛽

+ 𝛽

𝜙

𝐴𝑊

+ 𝛽

𝜙

𝐿𝐼

+ 𝛽

𝜙

𝑆𝑆𝑊

+ 𝛽

𝑋

+ 𝜀

𝑐𝑖,𝑡 = annual household non-durable consumption at period t 𝜙_𝑖,𝑡 = (∑ 𝜆𝜏−𝑡_(𝑎 𝜏 𝑡₎1 𝛾⁄ 𝐿𝑚𝑎𝑥 𝜏=𝑡 ) −1 𝐴𝑊𝑖,𝑡 = (1 + 𝑟)𝐴𝑡−1 (Accumulated Wealth) 𝐿𝐼_𝑖,𝑡 = ∑𝑠−1(1 + 𝑟)𝑡−𝜏 𝜏=𝑡 𝑦𝑡 (Lifetime Income)

𝑆𝑆𝑊𝑖,𝑡 = ∑𝐿𝜏=𝑠𝑚𝑎𝑥(1 + 𝑟)𝑡−𝜏𝐵𝜏 (Social Security Wealth)

𝑋_𝑖,𝑡 = vector of demographic household characteristics that might affect consumption

The remainder of this section will provide a more detailed description of each of the relevant variables which are included in the econometric model of Eq. (10).

Annual household non-durable consumption (𝒄𝒊,𝒕) – In line with i.a. Teppa (2014), annual household consumption is determined by the self-reported monthly household expenditures on non-durable goods, where the reported values are multiplied by twelve to determine the annual household consumption. Only consumption of non-durable goods is included, which provides enough relevance for the scope of this study. The variables used from the LISS Panel and the survey questions used to report these variables will be explained in more detail in Appendix A.

Accumulated wealth (𝑨𝑾_𝒊,𝒕) – Accumulated wealth is determined by estimating the present

18 A will give more details about the variables and corresponding survey questions used to estimate the household total accumulated assets.

Lifetime income (𝑳𝑰_𝒊,𝒕) – Since this study is primarily interested in the effect of a change in

pension wealth on consumption, the estimation for lifetime income has been simplified. The assumption is made that reported income at time t is a proper indicator for income at time τ (where 𝜏 > 𝑡). In other words, the assumption is made that agents will have a flat income path during their working years. However, especially for high-educated individuals in younger age cohorts, this assumptions might seem too broad. It would be interesting to take more assumptions into account in determining lifetime income, e.g. income uncertainty or a more hump-shaped income path. For the scope of this study however, the abovementioned assumption about lifetime income will be sufficient to estimate the model and investigate the effect of changes in pension wealth on consumption. Therefore, reported monthly net household income from the “Background Variables” surveys provided by the LISS Panel are used as an approximation for income in each period τ (where 𝑡 < 𝜏 < 𝑠7). For each household, the mean value of reported monthly net household income is multiplied by twelve in order to reach annual net household income. Furthermore, Appendix B will show the calculations used to derive at each household’s lifetime income from period t up until retirement at age s. Since the calculations include the statutory retirement age, lifetime income will be affected by the pension reform as the statutory retirement age is gradually increasing from the year 2012 onward.

Social Security Wealth (𝑺𝑺𝑾_𝒊,𝒕) – The determinant for pension wealth in this study will be the

Social Security Wealth (SSW). Detailed information about the Dutch pension system in Section 3 showed that the Dutch pension system consists of three pillars. As shown by Mastrogiacomo and Alessie (2011), the third pillar is of little importance to the median Dutch citizen and thus will not be included in our estimations of pension wealth. Though the second pillar is of much more importance, pension benefits from occupational pension schemes will not be included in the estimations for pension wealth. This exclusion has been made since the predictable effect of the pension reform on second pillar pension benefits on current consumption seems to be ambiguous. On the one hand, people will work for a longer period during their lifetime and thus will have more years of contributions to their pension schemes. Also, assuming individuals will reach an exogenous maximum age of Lmax, accumulated pension wealth is used to smooth consumption over a shorter retirement period since the mandatory retirement age has increased.

19 This would imply an increase in expected pension wealth and thus an increase in current consumption. On the other hand, the accrual rates for occupational pension schemes have been lowered. Therefore, it is not evident that the accumulated pension wealth would become larger or even would remain the same due to the extra working years. This ambiguity would be difficult to deal with, especially since this study does not have detailed data on an individual’s income history at its disposal as for example Alessie et al. (2013), who do include wage and job history in their pension wealth estimations.

20

Household No. of

SSW (new) SSW (old)

type Year observations Mean (Std. Dev.) Median Mean (Std. Dev.) Median Single 2009 793 _{(33 893)} 165 015 163 983 _{(33 893)} 165 015 163 983 2010 1019 _{(37 576)} 168 713 168 283 _{(37 576)} 168 713 168 283 2012 1055 _{(38 839)} 164 770 159 671 _{(38 127)} 176 493 179 113 2015 1479 _{(39 299)} 166 537 165 401 _{(40 132)} 179 845 181 999 Couple 2009 1808 _{(35 263)} 249 858 247 981 _{(35 263)} 249 858 247 981 2010 1911 _{(38 640)} 254 642 253 112 _{(38 640)} 254 642 253 112 2012 1989 248 759 _{(45 415} 243 075 _{(42 120)} 266 895 267 127 2015 2205 _{(45 258)} 258 058 253 563 _{(43 849)} 278 402 278 522

Table 2. Descriptive statistics of the SSW calculations for singles and couples in the Netherlands

The descriptive statistics in Table 2 clearly underline the prediction that the increase in the statutory retirement age will reduce SSW, both for singles and for couples. Moreover, controlling for age, the estimated values of SSW are higher for females than for males. This makes sense, since the SSW is calculated using survival probabilities. Appendix B also provides the exact formulas used in the calculation of these survival probabilities. The calculated probabilities stem from the average life expectancy, which is generally larger for females than for males. Thus, females will have a larger survival probability than males, i.e. 𝑎_𝜏𝑡_{(𝑓𝑒𝑚𝑎𝑙𝑒) ≥ 𝑎}

𝜏

21 cohorts of 50 years or older. Their findings also suggest a slightly left-skewed distribution of SSW, as median values are larger than mean values for most age cohorts.

The calculations of SSW for each household will be used in the econometric model as the measure of pension wealth, i.e. ∑𝐿_𝜏=𝑠𝑚𝑎𝑥(1 + 𝑟)𝑡−𝜏𝐵_𝜏. As shown by the detailed calculations of SSW in Appendix B, the calculated pension wealth is already discounted, with help from the interest rate r, to its net present value.

Adjustment factor (𝝓_𝒊,𝒕) – This adjustment factor stems from the closed-form solution for

consumption at time t, as derived in Section 2. The adjustment factor shows similarities to “Gale’s Q”, an age-specific adjustment factor proposed and applied by Gale (1998) to remove the downward bias on the estimated displacement effect of pension wealth. Alessie et al. (2013) apply “Gale’s Q” as adjustment factor in their life-cycle model. In their paper, the adjustment factor is defined as 𝑄(𝜆, 𝑡) =∑ 𝜆 𝜏−1 𝑡 𝜏=1 ∑𝐿 𝜆𝜏−1 𝜏=1 , where 𝜆 = ((1 + 𝑟) (1 + 𝜌))⁄ 1 𝛾⁄ _{⁄(1 + 𝑟)} . The model used by Alessie et al. (2013) starts at period 𝜏 = 1, while the model used in this study starts at period 𝜏 = 𝑡 and takes accumulated wealth up to period t (𝐴_𝑡−1) as given. Therefore, in our model, the numerator of the adjustment factor 𝑄(𝜆, 𝑡) becomes ∑𝑡_𝜏=𝑡𝜆𝑡−𝑡 = 1. Moreover, in addition to the model used by Alessie et al. (2013), our model incorporates the survival probability of the agent, 𝑎𝜏𝑡, which is incorporated in the denominator of the adjustment factor. Thus, to deal with the downward bias in estimating the displacement effect, our model incorporates a slightly altered version of Gale’s Q, an age-specific adjustment factor to discount all components of an agent’s wealth denoted as 𝜙_𝑖,𝑡. Further details on the calculation of the adjustment factor will be provided in Appendix B.

Vector of demographic household characteristics (𝑿_𝒊,𝒕) – To control for household specific

22 consumption and age is not likely to be linear. As pointed out by Thurow (1969) and graphically presented by i.a. Carroll and Summers (1991) and Attanasio et al. (1999), an individual’s consumption path over its life cycle is most likely to be hump-shaped, or as an inverted “U”. Therefore, the squared term of age of the household head is also added as a control variable in order to reflect this non-linear relationship. It is important to take this vector of household characteristics as control variable, since consumption may vary between households due to differences in household characteristics.

(Deaton and Paxson) Time dummies – Time dummies will be added to the estimation of the

econometric model to control for any year effects. Kapteyn et al. (2005) discuss the use of time dummies to control for any macro shocks and possible restrictions for time dummies. The econometric model will be estimated with two different approaches. First, a Pooled OLS model is estimated. This estimation implements the time dummies to control for year fixed effects. The second estimation approach will be a Fixed Effects (FE) model. The FE model requires a slight alteration to the time dummies used. We would like to distinguish the effects of age, cohort and time on consumption. However, as mentioned by Kapteyn et al. (2015), this will lead to the following identification problem: Calendar year is equal to an individual’s birth year plus its age. While some authors, e.g. Attanasio (1998), simply acknowledge the identification problem, others have slightly transformed the time dummies. The approach used in our FE model has been introduced by Deaton and Paxson (1994), who assume the coefficients of the time dummies will add up to zero. Moreover, they are orthogonal to a time trend. The alternative time dummies developed by Deaton and Paxson (1994) will be used in the estimation of the FE model to control for time effects.

23

Variables No. of obs Mean Std. Dev. Min. Max.

Annual household non-durable consumption

10 514 21 656 14 710 0 195 816

Annual household income 11 031 32 517 16 569 0 138 000

Accumulated wealth 6 561 68 607 164 428 -180 000 2 933 958

Lifetime income 11 204 399 201 433 763 0 3 939 128

Social security wealth 11 885 224 200 57 147 27 340 374 329

Adjustment factor φ 11 885 0.045 0.028 0.021 0.498

Age of the household head 11 885 54.23 14.92 25 97

Number of adults 11 885 1.656 0.475 1 2

Number of children 11 885 0.673 1.042 0 6

Low formal education 11 885 0.640 0.480 0 1

Medium formal education 11 885 0.251 0.434 0 1

High formal education 11 885 0.109 0.312 0 1

Table 3. Descriptive statistics of all variables used in the estimation of the econometric model.

Min. and Max. provide the minimum and maximum value of each variable respectively.

5. Results

Before discussing the results of the estimations of the econometric model, Table 4 presents correlations between the variables of lifetime income, level of formal education and the annual household consumption of non-durable goods.

Correlations Lifetime income Education

Lifetime income 1.000 .

Education 0.1568*** 1.000

Annual non-durable consumption 0.1498*** 0.2398***

Table 4. Correlations between lifetime income, level of formal education and annual household

non-durable consumption. ***=1% statistically significant

24 Next, Table 5 presents the estimated coefficients of the econometric model outlined in Eq. (10), labeled Model 1. The model is estimated by two alternative specifications: by Pooled Ordinary Least Squares (POLS) and by Fixed Effects (FE). Estimation of the model with Ordinary Least Squares (OLS) would lead to biased results and strong autocorrelation, due to the paneled structure of the dataset. Therefore, the first column of specifications in Table 5 are estimates of the POLS regression. The second column of specifications represent the FE regression. The FE regression is included to control for unobserved heterogeneity and any omitted variable biases.

The pooled OLS regression presents slightly more significant results. Consumption is increasing with number of adults present in the household. On average, couples spend €3390.36 more per year on non-durable consumption than singles. This finding is significant at the 1% level. Though the coefficient of number of children is also positive, we are unable to reject that this coefficient is significantly different from 0. One would assume a positive coefficient for the number of children present in a household, as every child would raise consumption of the household on non-durable goods. As mentioned however, we are unable to validate this argument with our estimation results. Moreover, in line with the correlation findings of Table 4, consumption is positively related with the degree of formal education. On average, households where the head has a college degree spend €2052.79 more on annual non-durable consumption goods than households with no or a low level of formal education. For households with an university degree, the amount spend on consumption is on average €4771.02 higher than households with no or a low level of formal education. Both coefficients are significant at the 1% level. Furthermore, consumption is increasing with age of the household head. The estimation results imply an increase of €312.64 in non-durable consumption for every year the head of the household is getting older. This coefficient is significant at the 1% level. As mentioned in Section 4, age of the household head squared is added to the model to control for the hump-shaped curve of consumption over lifetime. This assumption would imply a negative coefficient of age of the household head squared. While this is the case, the estimated coefficient is not significantly different from 0 and thus we are unable to verify this assumption.

25 significant at the 5% level. Moreover, consumption is significantly increasing with age, on average by €1712.76 per year. The FE estimation also finds a significant negative coefficient for age of the household squared, which indicates the hump-shaped consumption path over lifetime. Based on the results of the FE regression, we are unable to conclude anything from the other coefficients of the control variables as they do not significantly differ from 0. Finally, both models show to be properly fitted as an F-test would fail to reject the specifications of both models.

Model 1 Model 1

Variables Pooled OLS FE

Accumulated wealth 0.072* (0.0419) 0.0326 (0.0314) Lifetime income 0.3583*** (0.0423) 0.2443*** (0.0857) Social security wealth 0.1315

(0.213)

0.4314 (0.3213) Age of the household head 312.64***

(111.95)

1712.76*** (401.59) Age of the household head (squared) -1.262

(0.993) -11.34*** (3.689) Number of children 403.61 (278.55) -427.14 (814.23) Number of adults 3390.36*** (789.31) 1385.02 (1892.29) Education: college 2052.79*** (483.12) 3080.70 (3255.21) Education: university 4771.02*** (813.02) 12014.52** (5309.04) Constant -2404.57 (3220.02) -46225.27*** (11710.05) (Deaton and Paxson) Time dummies Yes Yes

N 6086 6086

Household clusters 3246 3246

R2 0.0915 0.0244

Prob > F 0.0000 0.0000

26 The FE regression does not find a significant result for the effect of accumulated wealth on consumption. The Pooled OLS regression does find a rather small impact of accumulated wealth, as the coefficient implies an increase of €0.07 cents in non-durable consumption per €1 increase in accumulated wealth. However, this effect is only significant at the 10% level. On the other hand, both regressions do find a significant, positive effect of lifetime income on consumption. An increase in lifetime income of €1 will increase annual consumption of non-durable goods with €0.36 cents and €0.24 cents, for the pooled OLS and FE regression respectively. These findings are both significant at the 1% level. However, these coefficients do not embody a perfect offset, i.e. a coefficient of 1. Thus, changes in lifetime income are not fully incorporated in household consumption and saving behavior. The real coefficient of interest is β3 in the econometric model presented in Eq. (10). Both regressions also show a positive coefficient for social security wealth on annual consumption that lies between 0 and 1. However, for both estimation methods, we are not able to conclude that the coefficient for social security wealth is statistically different from 0.

The life-cycle model does predict a perfect offset between pension wealth and financial wealth, i.e. every increase in pension wealth is followed by an equal, though properly discounted, decrease in financial wealth (current savings). Since the model assumes income is either saved or consumed, this also implies an increase in consumption equal to the discounted decrease in savings. However, the results of both regressions do not seem to support the perfect offset of the life-cycle model. Instead, both estimation methods find no or only a very limited offset between pension wealth and financial wealth. These findings are to some extent in line with other empirical evidence on the replacement rate between pension wealth and non-pension wealth. Other studies also indicate a replacement rate between pension and financial wealth of less than 100%, though most studies do find significant replacement rates between 0% and 100%. However, the empirical findings on the exact percentage of the replacement rate exhibit large variations across studies. Section 6 will provide a further discussion of our findings in comparison with other empirical evidence.

27 coefficients on the null hypothesis of Eq. (8), 𝐻0: 𝛽𝑖 = 1. In addition, the test results for the significance of each of the variables are also included in Table 8.

Pooled OLS FE Variables AW _{(𝑖 = 1)} LI _{(𝑖 = 2)} SSW _{(𝑖 = 3)} AW _{(𝑖 = 1)} LI _{(𝑖 = 2)} SSW _{(𝑖 = 3)} 𝛽_𝑖 0.072* _(0.0419) 0.3583*** _(0.0423) 0.1315 _(0.213) 0.0327 _(0.0314) 0.2443*** _(0.0857) 0.4314 _(0.3213) p-Value 𝛽𝑖 = 1 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0769* p-Value 𝛽_𝑖 = 0 0.0857* 0.0000*** 0.5370 0.2990 0.0044*** 0.1796 Table 6. Pooled OLS and FE estimation results and p-values of the F-tests for the three components of lifetime earnings: accumulated wealth (AW), lifetime income (LI) and social security wealth (SSW). Dependent variable is annual household non-durable consumption. Standard errors in parentheses. ***=1%, **=5% and *=10% statistically significant.

As Table 6 clearly shows, the coefficient for LI is significantly different from zero for both regression models. The coefficient for SSW is not significant in both estimation models, while the coefficient for AW is only significant in the pooled OLS model, albeit at the 10% level. The other F-test is to test if the coefficients are significantly different from 1, i.e. different than 100% and thus do not represent full offset. For AW and LI, the coefficients are significantly different from 100% for both estimation models. These findings are all significant at the 1% level. Our coefficient of interest is the coefficient for SSW and whether we are able to reject perfect offset as assumed by the theoretical lifecycle model. Table 6 presents the results for this test and shows that we are able to reject the null hypothesis presented in Eq. (8); 𝐻₀: 𝛽₃ = 1 for the estimated replacement rate for both estimation methods. While this result is significant at the 1% level in the pooled OLS regression, it is only significant at the 10% level in the FE regression. These findings are in line with other empirical evidence which also rejects perfect offset between pension wealth and financial wealth.

5.1 Sensitivity analysis

5.1.1 Differences in level of education

28 Table 7. Pooled OLS and FE estimation results for Eq. (10), SSW coefficients are estimated per level of formal education: 1) no or low formal education, 2) college degree and 3) university degree. The dependent variable is annual household non-durable consumption. Reported R2 for the FE estimation is the within R2_{. ***=1%, **=5% and *=10% statistically significant. Standard errors are clustered by} household and are given in the parentheses.

Model 2 Model 2

Variables Pooled OLS FE

Accumulated wealth 0.0619 (0.0415) 0.0337 (0.0315) Lifetime income 0.3759*** (0.0423) 0.2403*** (0.0858) Social security wealth

(no or low formal education)

0.0877 (0.2175)

0.4848 (0.3476) Social security wealth

(college degree)

0.3165 (0.2100)

0.5071 (0.4135) Social security wealth

(university degree)

0.3968 (0.2628)

-0.2201 (0.5991) Age of the household head 275.67**

(111.53)

1720.88*** (401.81) Age of the household head (squared) -0.9866

(0.9963) -11.3506*** (3.6897) Number of children 386.45 (278.80) -465.11 (815.16) Number of adults 3178.95*** (780.44) 1652.35 (1905.19) Education: college -305.24 (1062.69) 2786.05 (4566.99) Education: university 1731.66 (1642.45) 16824.27** (6682.95) Constant -601.09 (3197.05) -47339.27*** (11857.49) (Deaton and Paxson) Time dummies Yes Yes

N 6086 6086

Household clusters 3246 3246

R2 0.0930 0.0250

29 The econometric model in Eq. (10) will be used to estimate the model, again both by a pooled OLS regression as well as a FE regression. Table 7 presents Model 2, which estimates the effect of pension wealth on annual household consumption for three different subsets of the panel. These subsets are created based on formal level of education. Table 4 already presented a significant, positive correlations between consumption and formal level of education, were we concluded that households with a higher level of formal education tend to have a higher level of annual consumption. Table 7 presents the effect of pension wealth on consumption for households with different levels of formal education.

The results in Table 7 suggest a difference in the replacement rate of pension wealth to financial wealth by level of formal education. However, like Model 1, the coefficients of SSW do not significantly differ from 0 for each of the three educational groups. And again, this is the case for both estimation methods. Since the estimated coefficients for each educational group are not statistically significant, we implement each of the coefficients in a couple of Wald-tests in order to draw some conclusions from the estimation results. The SSW coefficient of each educational group will be tested on the null hypothesis from Eq. (8) and these results have been presented in Table 8. In addition to the standard tests of the coefficients being significantly different from 0 and 1, we perform tests on differences in the coefficients between the different educational groups. These test results will be used to further explain differences in the replacement rate between different groups of households based on educational attainment.

Table 8. Pooled OLS and FE estimation results and p-values of the F-tests for social security wealth (SSW) of three different household groups: for households with no or low formal education (low), households with a college degree (mid) and households with an university degree (high). Dependent variable is annual household non-durable consumption. Standard errors in parentheses. ***=1%, **=5% and *=10% statistically significant.

Pooled OLS FE

Variables SSW(low) _{(𝑖 = 1)} SSW(mid) _{(𝑖 = 2)} SSW(high) _{(𝑖 = 3)} SSW(low) _{(𝑖 = 1)} SSW(mid) _{(𝑖 = 2)} SSW(high) _{(𝑖 = 3)}

30 The results in Table 8 are to a large extent in line with the main results of the SSW coefficient in Model 1. Both estimation methods are unable to find a replacement rate that is significantly different from 0 for any of the different educational groups. However, the Pooled OLS model does find coefficients for SSW which are significantly different from 1. These results are at the 1% significance level for the household group with no or low formal education and for the household group with a college degree. For households with an university degree the coefficient is significantly different from 1 at the 5% level. Unfortunately, the FE model does not find similar results, as we are only able to conclude that the SSW coefficient for households with an university degree is significantly different from 1. This is again at the 5% level. To further test for any variation in the displacement effect of pension wealth on different groups of households, we investigate any significant differences between the SSW coefficients of all three educational groups. For the pooled OLS regression, the SSW coefficient, i.e. the displacement effect, for households with a college or university degree is significantly larger than the SSW coefficient is for the household group with no or low formal level of education. There is, however, no significant difference between the displacement effect between households with a college degree and households with an university degree. However, these results do indicate different degrees of the displacement effect for households based on their educational level. Moreover, the results suggest higher educational attainment will lead to a larger offset between pension wealth and financial wealth.

31 differences between any of the SSW coefficients and thus is unable to conclude any differences in the displacement effect in correlation with the level of education.

5.1.2 Differences in age groups

In the estimations of SSW and the adjustment factor ϕ, age plays an important part. The estimation of SSW is based on discounted future AOW benefits, starting by the number of years it takes to reach the first year of AOW benefits i.e. the statutory retirement age. Moreover, Appendix B.4 shows that the adjustment factor ϕ has a strong positive correlation with age and thus adjusts lifetime income and social security wealth more heavily for younger cohorts. Therefore, the econometric model will be estimated based on three different age groups. The age groups are defined by a young cohort, a middle cohort and an old cohort. The young cohort consists of households where the head is between 25 and 44 years old. The middle cohort consists of households where the head is between 45 and 64 years old. The old cohort consists of households where the head is between 65 and 99 years old. In other words, the old cohort consists entirely of retirees, while the young and middle cohort consist entirely of working individuals, or at least non-retirees. The econometric model of Eq. (10) will once again be the basis of the estimations, where social security wealth is now implemented for the three different age groups separately. Table 9 presents the estimation results, again for a pooled OLS regression as for a FE regression.

32

Model 3 Model 3

Variables Pooled OLS FE

Accumulated wealth 0.0763* (0.0416) 0.0330 (0.0315) Lifetime income 0.3647*** (0.0414) 0.2559*** (0.0862) Social security wealth

(age group 25-44)

0.1914 (0.3085)

0.1199 (0.4311) Social security wealth

(age group 45-64)

-0.1514 (0.2673)

0.2273 (0.3637) Social security wealth

(age group 65-99)

0.0530 (0.2224)

0.3620 (0.3271) Age of the household head 656.34***

(144.98)

1721.9*** (411.45) Age of the household head (squared) -4.023***

(1.190) -11.425*** (3.732) Number of children 303.61 (281.45) -451.11 (815.50) Number of adults 3619.34*** (829.25) 1665.69 (1908.15) Education: college 2061.98*** (479.83) 3047.93 (3255.63) Education: university 4876.70*** (806.50) 11820.74** (5312.62) Constant -11412.2*** (3932.99) -45489.24*** (12043.61) (Deaton and Paxson) Time dummies Yes Yes

N 6086 6086

Household clusters 3246 3246

R2 0.0945 0.0250

Prob > F 0.0000 0.0000

Table 9. Pooled OLS and FE estimation results for Eq. (10), SSW coefficients are estimated per age group: 1) age between 25 and 44, 2) age between 45 and 64 and 3) age between 65 and 99. The

33 Table 10. Pooled OLS and FE estimation results and p-values of the F-tests for social security wealth (SSW) for three different household groups: for households in the age group 25-44 (25-44), households in the age group 45-64 (45-64) and households in the age group 65-99 (65-99). Dependent variable is annual household non-durable consumption. Standard errors in parentheses. ***=1%, **=5% and *=10% statistically significant.

34

6. Discussion

The results presented in the previous section are to a large extent in line with predictions made about the replacement rate between pension wealth and financial wealth and how this replacement rate may differ across households. The general estimation results estimated by Model 1 show a less than perfect offset between pension wealth and financial wealth, i.e. the replacement rate is lower than 100%. These findings seem to be in line with most empirical evidence on the displacement effect of pension wealth. For instance, Alessie et al. (2013) report a displacement rate of 47.1% and 60.9%, using robust regression and median regression respectively, for a panel set of households for a number of European countries. Unfortunately, in contrast to earlier empirical evidence, our estimations do no find a value of the replacement rate that is significantly different from 0 and thus we are unable to draw any more conclusions on the degree of the replacement rate. Moreover, in line with other empirical evidence (e.g. Attanasio and Rohwedder [2003] and Attanasio and Brugiavini [2003]), the replacement rate does show variations across households when households are grouped based on age or formal educational level.

35 wide empirical evidence that financial literacy plays an important role in consumption and saving decisions by individuals. Most empirical evidence states individuals with low levels of financial literacy are unable to correctly apprehend how reforms will affect their personal financial situation and therefore are unable to properly adjust their financial behavior accordingly. Thus, it seems questionable if the Dutch pension reforms are comprehendible for all Dutch citizens. There has been an extensive debate on the pension reforms in the Netherlands and, as discussed earlier, the pension reforms even were put on ice and were implemented by a succeeding government. While one could assume the pension reform and increasing statutory retirement age had to be known by most Dutch citizens, the recent acceleration in the statutory retirement age may have been unknown for some, or perhaps still is. Furthermore, the increase of the statutory retirement age leads to multiple changes in both pension wealth as well as financial wealth. This study already shows the differential impact the reform will have on lifetime income and social security wealth, while we do not even include second pillar pension benefits in our estimation model. It seems that it would be difficult for households to correctly forecast how the pension reform will influence their personal financial situation, especially for households with a low level of financial literacy. Thus, the variation in the displacement effect of pension wealth between household groups of different educational levels could be explained by the inability to comprehend changes in personal social security wealth.