The Retirement Savings Puzzle:
the Role of Gains in Housing Wealth and Home
Production
Kruijver, W.E.
Combined Thesis MSc. Economics & MSc. Finance
Abstract
Dutch retired households’ wealth patterns are at odds with the basic life-cycle model. Post-retirement consumption decreases and both financial and housing wealth are stable. Two potential explanations are checked: the marginal propensity to consume out of capital gains in housing wealth and home production. A MPC is found of 0.08 percent per percentage increase in Net Housing Wealth for retired home owners. This limited translation of gains to consumption confirm the conjecture that the liquidity of housing wealth is limited. No evidence is found for the interaction of non-market time and home ownership: home production.
University of Groningen Supervisor: Dr. R. van Ooijen Department of Economics, Econometrics Date: 27-6-2016
& Finance
1. Introduction
The basic life-cycle model (Modigliani and Brumberg, 1954) predicts decumulation of wealth from retirement until the end of life. In practice however, this decumulation occurs slower than predicted. This phenomenon is called the Retirement Savings Puzzle.
There are three categories of explanations for this phenomenon: longevity risk, health expenditure risk and the existence of a bequest motive. The present thesis researches the role of homeownership in the puzzle. Housing wealth makes up a large part of total wealth in the Netherlands and is especially interesting since it serves both as a consumption good and as an asset. In addition to that, theory suggests that housing and non-market time may be complements and the house can thus be used as an input for home production, affecting consumption.
First, the existing theory on consumption is synthesized to give an overview of the predictions of life-cycle models in terms of wealth and consumption patterns. Then a descriptive analysis of Dutch lifetime wealth and consumption is provided and compared to what is expected from theory. It is found that the consumption pattern is more in line with the predictions of the life-cycle model than the wealth pattern.
In order to further investigate the puzzling profiles of the elderly in the Netherlands, two possible explanations are examined which are all related to home-ownership. On the one hand the question is how households view their housing wealth when planning their consumption. What is the housing component in wealth and what is the role of its capital gains and losses?
Resulting from the descriptive analysis of wealth, consumption and time use, three stylized facts are formulated:
- The effect of capital gains on consumption is limited due to the limited ability to extract housing wealth.
- Capital losses in housing wealth result in depressed consumption due to the decrease in lifetime wealth.
- Home production affects consumption expenditure and this effect is different for home owners compared to tenants due to the interaction of housing and non-market time.
These stylized facts are checked via the empirical strategy. Consumption dynamics and its response to financial wealth, housing wealth and home production is modeled. To distinguish the role of homeownership, home owners and tenants are separated.
It is concluded that part of the Retirement Savings Puzzle in the Netherlands is explained by the limited extent to which gains in its wealth translate to increased consumption.
2. Consumption Theory
2.1 The Basic Life-Cycle Model and its Extensions
pointed at the fact that the share of income spent on consumption actually declines as income rises. Despite the fact that in the long run the percentage share that is spent on consumption is constant, short-term estimates showed declines. Following that, research evolved to intertemporal utility-maximization theory, better explaining individual consumption decisions. Starting with Modigliani and Brumberg (1954) their life-cycle Theory of Consumption and the Permanent Income Hypothesis of Friedman (1957). The life-cycle model features agents with a finite lifetime and a separate period for retirement, the permanent income model has infinitely living agents.
The life-cycle model of Modigliani and Brumberg (1954) serves as a framework to analyze the savings and consumption patterns. According to the theory, a utility maximizing agent is expected to smooth his consumption over lifetime due to the concave relationship between consumption and utility. Subsequently, the ability to consume is determined by the lifetime budget constraint. The slope of the budget constraint determines the tradeoff between consumption in period t and consumption in period t+1 and is equal to – where r is the real interest rate at which consumers lend and borrow (Parker, 2010). The position of the budget constraint is determined by the present value of lifetime earnings, which is defined as1:
(1)
where is the stock of wealth (human and nonhuman) from time zero, is the value of current nonhuman (financial or physical) assets, for t = 0, 1, 2, …, T is the expected stream of real labor income over the lifetime and r is the real interest rate.
This basic life-cycle model predicts that assets (A0) are accumulated during working-life when income is high, to be declined during retirement when income is low. In practice however, the elderly tend to dissave much more slowly than predicted (Ang, 2014). This phenomenon is called the Retirement Savings Puzzle.
One of the explanations is that individuals face longevity risk due to lifetime uncertainty. Consumers face uncertainty with regard to time horizon for which they plan their consumption and risk to outlive their assets. Other explanations include having a bequest motive and health expenditure uncertainty.
If the basic life-cycle model is extended with lifetime uncertainty and without a bequest motive, agents might still accidentally leave a bequest. Hurd (1989) formulates a formal definition of the mortality rate to be incorporated in the optimization of lifetime consumption2. The mortality rate is defined as the product of the probability of survival to period t conditional on attaining period 1 and the probability to die in period t+1 given survival to period t (instantaneous mortality rate).
The instantaneous mortality rate is defined as:
The probability of survival to period t conditional on being alive in period 1 is given by:
The mortality rate thus is equal to:
Mortality risk is a source of uncertainty for the horizon of the consumption planning and makes consumers impatient (i.e. preferring actual consumption over later consumption). Having defined mortality risk, we can now derive the Euler equation from the life-cycle model of Hurd (1989) with mortality risk, no intentional bequest motive and liquidity constraints. The assumption of liquidity constraints is made as it does not make sense to die indebted. Assets at the time of death should thus in any case be larger than zero.
2 The derivation of the life-cycle model with mortality risk, without a bequest motive and
In period 1, consumers face the following maximization problem of expected utility with respect to consumption with impatience and mortality:
Solving for the FOCs resulting from the Lagrangian multiplier, we obtain the consumption Euler equation. The equation shows the marginal utility given the liquidity constraint.
(2)
Where is the Kuhn-Tucker multiplier associated with the liquidity constraint. The consumption Euler equation implies the intertemporal rate of substitution of consumption (i.e. the slope of the consumption profile). The parameters affecting the slope are thus the interest rate (r) and the rate of time preference (ρ). If r>ρ, this implies an upward sloping consumption profile, as substituting present consumption for future consumption yields greater utility. If r=ρ, the lifetime consumption profile is flat. If r<ρ, the consumers are impatient and the profile is downward sloping. Hurd (1989) shows that if annuities are bought in order to insure against longevity risk consumption profiles can be steeper and there is no need to leave a buffer near the end of life. In practice however, due to administrative costs and the illiquidity of annuities, annuities are invest in to a limited extent (Hurd, 1989).
especially interesting as the country’s inhabitants have large private pension reserves and state pensions are provided and health care expenditure is covered extensively. Income uncertainty and uncertainty with regards to expenses post-retirement thus is significantly lower than for individuals in the United States. The following section elaborates on the different extensions of the life-cycle model and the sensitivity of their predictions to the type of data used (consumption versus wealth) and the period analyzed (capital gains or losses).
2.2 Sensitivity Preference Parameters Life-cycle Models
The parameter central in the analysis of the Retirement Savings Puzzle, together with the real interest rate and the rate of time preference, is that of relative risk aversion. If consumers are highly risk averse they will want to guard against having to consume at a low level should they live substantially past their life expectancy, and so they will save more than consumers who are less risk averse (Hurd, 1989). Estimated life-cycle models usually provide very different preference parameters depending on whether consumption data (e.g. Palumbo, 1999; Gourinchas and Parker, 2002) or wealth data is used (e.g. Hurd, 1989; De Nardi et al., 2010). In addition, its parameters are sensitive to the period of analysis. Hurd (1989) estimates a life-cycle model using data in which a housing boom occurs and De Nardi (2010) analyzes a period during which capital gains are relevant.
for relative risk aversion is similar, but that for time preference is smaller, in accordance with the flatter wealth pattern including housing.
De Nardi (2010) also uses wealth data, but finds a significantly higher value for relative risk aversion: between 3.66 and 3.81. The period of analysis ranges from 1994 to 2006, when capital gains were realized on financial assets. Comparing the estimates of the models for relative risk aversion, we expect to find a different response of consumption out of housing wealth gains compared to gains in financial assets.
The scholars that use consumption data in general find higher estimates of relative risk aversion. E.g. Palumbo (1999) uses the Panel Study of Income Dynamics (PSID) of US households and fits both a life-cycle model and a health uncertainty model. It follows that the health uncertainty model outperforms the life-cycle model for all specifications. Including health uncertainty reduces the estimate for relative risk aversion significantly, from 24 to 7 for couples and from 25 to 6.25 for singles. These findings also show the significant level of health expenses uncertainty for the United States.
Gourinchas & Parker (2002) also use Consumption Expenditure Survey (CEX) data for the US, but for the period from 1980 to 1993. Specifying a model including labor uncertainty for the entire lifetime (separated in cohorts) rather than just retirees, yield estimates significantly different to those of Palumbo (1999). According to the researchers, age-heterogeneity is key in consumption behavior resulting from the interaction between retirement and precautionary motives for savings at different ages. It is found that consumers in early-life behave as buffer-stock consumers, setting consumption growth equal to average labor income growth, regardless of tastes. As retirement nears, consumption behavior better reflects that predicted by the life-cycle hypothesis. The coefficient of Relative Risk Aversion for the average household is significantly lower with a value between 0.5 and 1.4.
than for Dutch households. Recently however, a wealth-based payment for nursing-home care was introduced in the Netherlands (Van Ooijen, 2015). Besides that this leads to wealth transfers between parents and children (Van Ooijen, 2015), it is part of the movement towards more own contribution of payment of (long-term) health care in the Netherlands. Therefore, medical expenditure uncertainty becomes more relevant for the Dutch case in light of health care system reforms to guarantee its sustainability.
2.3 The MPC (out of Housing Wealth)
money and risky financial assets contribute significantly to explaining consumer spending, while savings and real assets are of no value. Particularly relevant for this thesis is the Marginal Propensity to Consume out of Housing Wealth. Housing makes up a substantial part of total assets and one of the questions is how this translates into consumption in a life-cycle setting.
Engelhardt (1996) examined the empirical link between house prices and savings behavior of American households during the 1980s. The research uses cross-time and cross-regional variation in housing market conditions to identify behavioral savings effects. It is found that the marginal propensity to consume out of real housing capital gains is equal to 0.03 for the median saver household. The saving response to total and unanticipated real housing capital gains is different.
Campbell & Cocco (2007) find that house price changes may stimulate consumption by increasing households’ perceived wealth, or by relaxing borrowing constraints. It investigates the response of household consumption to house prices using UK micro data obtained from the British Household Panel Survey and the Family Expenditure Survey over the period 1988-2000. The largest effect is estimated for older homeowners, the smallest (insignificant) for younger renters. The estimated elasticity is 1.7 if controlling for interest rates, household income, and other demographic variables.
This thesis is a microeconomic study and studies the MPC out of Housing Wealth separately for working age households and retired households. Part of the reasoning to make this separation is the differences due to age heterogeneity. Buiter (2008) argues that younger households wish to trade up over their life-cycle in housing and are thus short in housing. On the other hand, older ones wish to trade down and are long in housing. The respective positions imply a different response to changes in house prices.
2.4 Non-Market Time Models
3. Dutch Household Behavior
For the analysis of Dutch (retiree) behavior in terms of consumption, the LISS Panel from CentERdata has been used. It is chosen to use these data as the wide range of Core Studies facilitates a simultaneous analysis of consumption, assets, housing assets and time use as is necessary for this thesis. The present section describes wealth, and expenditure patterns in order to explore what Dutch data say in comparison to theory. First, household portfolio components are displayed, granting an overview of the ownership rates and average values of financial and nonfinancial assets and liabilities. Table 3.3 shows household aggregates of sub-types of wealth. Finally, (housing) wealth, expenditure and time use cohort profiles indicate the age patterns and inter-cohort differences in terms of these variables.
3.1 LISS Panel
The data set used for analysis is the result of Longitudinal Internet Studies for the Social Science (LISS). CentER of the University of Tilburg is gathering these longitudinal studies annually since October 2007 and provides access for research purposes to these data. A true probability sample of households is drawn through an address frame of Statistics Netherlands. In order to guarantee participation of low income households and especially the elderly, households that cannot participate otherwise, are provided a computer with internet connection. The number of participating households is approximately 5,000, bringing together about 8,000 members. When a household joins the panel, the first questionnaire submitted is that regarding the background variables. Hence, for every household member, which is assigned a number, a household number, and multiple background characteristics are available. The panel is designed around ten Core Studies which together with the background variables provide a rich data set for analysis of household behavior.
3.1.1. Core Studies
Table 3.1.1. presents the four Core Studies used for the analysis of the present paper.
Time Use & Consumption Economic Situation**
Assets Housing
N*** 7,006 7,006 5,560 3,292
Average
age* 55.2 55.2 53.9 53.9
Key
variables Household Expenditure Personal Expenditure Total Expenditure Eating at home Expenditure on cleaning and gardening Household
Chores Net worth resulting from household portfolio components Types of wealth specified in Table 2 WOZ Self-reported house price Mortgage(s) Waves 2009, 2010, 2012, 2015 2009, 2010, 2012, 2015 2008, 2010, 2012, 2014 2010, 2011, 2012, 2014, 2015 Table 3.1: Relevant variables selected from LISS panel
Time Use & Consumption: in this Core Study, the participants are first presented questions on if and how much time they spend on ten activities related to work, leisure, care, household chores and administrative activities. In addition to that, ten categories of respectively private (individual) and public (household) expenditures are asked. The consumption expenditure figures were converted from nominal as given in the data set to real, by correction for inflation.
Economic Situation: data on financial and non-financial wealth is obtained from the ‘Economic Situation: Assets’ and the ‘Economic Situation: Housing’ questionnaires. Participants are asked to report the values of the composites of their net worth, including real estate investment not used as a (second) home. Housing describes the housing assets of the participants and their corresponding mortgages for home owners and the rent expenditure and characteristics of the dwelling for tenants. Asset and housing data were corrected for inflation using CPI indicators per year.
Resulting from first cleaning the data and removing the outliers, appending the Core Studies for the years available and merging these with the background variables, data sets are obtained for the descriptive analysis of wealth and portfolios in the Netherlands and the subsequent regression analyses related to the Retirement Savings Puzzle. The limitations that come apparent are:
- The housing questionnaire is not available for the year 2008.
3.2. Household Wealth and Portfolios in the Netherlands
For the present part, the ‘Asset’ component of the Core Study ‘Economic Situation’ presented at part of the households is used. The household head is asked to list assets that are registered under both names, as well as on his/her own name. The partner is asked to list the assets registered solely on his/her name. In order to generate total household assets, the personal asset and liability components were collapsed on household level per year. It is decided to treat missing values as missing values in this calculation, rather than zero’s, to avoid underreporting.3 Gathering figures for net worth, net financial wealth and housing equity is done according to the subsequent household balance sheet. Definitions are used from Table 9.2 “Definition of Asset and Debt Categories” by Alessie et al. in Guiso et al. (1999).
Assets Liabilities
Financial Assets
Savings (balance of current accounts, savings accounts,
term deposit accounts, savings bonds or certificates)
Single-premium insurance policy, life annuity insurance,
endowment insurance (not linked to a mortgage)
Investments (growth funds, share funds, bonds,
debentures, stocks, options, warrants, and so on)
Mortgage and home equity Loans for investment real estate Credit card balances
Other debt (student debt, home improvement loans, vehicle loans, unsecured credit lines, loans against pension and life insurance policies
Nonfinancial Assets
Primary residence Real estate investment Vehicles
Table 3.2: Overview of Household Balance Sheet with LISS data available
3 If the questionnaire was not filled in by one of the household members, total household assets
3.2.1 Household Portfolios
2010 (%) (€) (€) 2012 (%) (€) (€)
N=2,287 Own. Mean Median N=2,141 Own. Mean Median Assets Savings 93.7 39,979 8,770 94.4 47,214 10,000 Single-premium insurance 19.3 54,062 17,500 18.0 30,937 15,000 Investments 20.3 61,409 19,000 18.6 57,977 20,814 Real-estate 5.8 221,033 172,500 8.9 224,688 164,500 Vehicles 73.5 13,149 6,000 75.4 14,103 6,000 Loans 6.1 19,796 2,500 6.4 28,295 2,000 Primary Residence 61.6 294,170 245,000 61.5 278,546 245,000 Liabilities* Mortgage 37.0 150,471 135,000 38.4 149,881 132,000 Household Debt 9.9 71,557 7,000 8.6 99,303 8,500 Student Debt 3.1 12,310 8,375 2.5 15,243 11,048 (%) (€) (€) 2014 N=2,866
Own. Mean Median
Assets Savings 93.7 42,899 8,750 Single-premium insurance 19.3 37,450 18,000 Investments 20.3 63,084 16,900 Real-estate 5.8 186,282 116,000 Vehicles 73.5 12,766 5,000 Loans 6.1 29,238 2,500 Primary Residence 58.4 261,050 226,000 Liabilities* Mortgage 34.4 158,248 140,000 Household Debt 9.4 46,357 10,000 Student Debt 5.3 14,708 10,000
Table 3.3: Household Portfolio Composition 2010, 2012, 2014
3.2.2 Household Wealth
Total Assets the sum of financial assets and the value of the house
according to the municipal property appraisal (WOZ-waarde)
Net Financial Assets the sum of guaranteed minimum payout of
single-premium or life-annuity insurances, total value of investments (growth funds, share funds, bonds, debentures, stocks, options, warrants) minus household debt and student debt
Housing Wealth value of the house according to the municipal
property appraisal (WOZ-waarde)
<35 35-50 50-65 65-80 Total Total Assets Mean 88,807 189,613 241,418 270,665 215,627 Median 12,000 192,004 216,000 229,970 198,000 N 729 1,188 1,680 1,454 5,051 Financial Assets Mean 10,912 31,483 58,640 88,598 52,229 Median 3,160 8,000 16,000 21,499 11,000 N 570 780 1,058 906 3,314 Net Financial Assets Mean -504 21,422 50,647 72,691 39,985 Median 616 5,000 12,000 20,000 8,000 N 624 827 1,106 886 3,443 Housing Wealth Mean 200,413 249,013 287,243 295,545 273,120 Median 197,500 225,500 245,000 253,500 237,000 N 292 806 1,196 1,060 3,354 Mortgage Mean 197,542 197,302 148,419 103,788 153,815 Median 186,000 175,000 127,000 84,857 136,603 N 224 720 952 719 2,615 Net Housing Wealth Mean 45,595 66,872 161,153 220,364 145,640 Median 2,000 47,500 148,000 194,000 133,000 N 313 877 1,255 1,083 3,582
4. Cohort Analysis of Wealth and Consumption
The present section is a descriptive analysis of the accumulation of household wealth over life and the differences between cohorts. Besides that, lifetime consumption is presented in a cohort graph. The consumption pattern is analyzed and subsequently compared to the predictions of a basic life-cycle Model based on (housing) wealth. Then, the patterns of possible solutions to the puzzle are described: housing wealth and home production.
4.1 Cohort Analysis
In order to plot the changes in saving as people age, the data on assets and housing were separated into age cohorts inspired by Deaton and Paxson (1994). These cohorts are defined by the year of birth of the household head. Given the amount of data available after collapsing the assets and merging these with data on housing, cohorts of fifteen years were generated, in line with the grouping of Table 3.4.
The average of the variables of interest are computed and tracked over the survey years (2010, 2012 and 2014). Since the age of an individual is a construct of the year the data point is situated and the year of birth of the individual, the problem is to distinguish between age, time and year effects. The overlap between the cohort lines provides a solution however. Whereas the slope of the cohort line indicates the behavior of the cohorts as they age and the time effects, the difference between the cohort lines shows the cohort effects.
4.2 Evolution of Dutch Household Wealth over Lifetime
The basic life-cycle model predicts asset accumulation until retirement, decumulation after that and depletion nearing the end of life. The cohort graphs in Figure 4.1 for the Netherlands show no such pattern. Let us walk through this descriptive analysis clockwise.
Looking at the accumulation of Total Assets, we observe that wealth indeed increases until retirement. The jumps between the cohort lines indicate that cohort effects are positively influencing household wealth. Individuals between 30 and 35 years old born before 1980 have a higher level of wealth than those with the same age belonging to the post-1980 cohort. This could potentially be due to a higher return on assets, especially housing assets, which make up a large part of total assets, before the period between 2010 and 2014. The figures shows that each cohort is richer than the previous one, indicating positive cohort effects for each older cohort compared to the younger one. Looking at the slope of the cohort lines, it is furthermore visible that households accumulate wealth as they age until retirement. This pattern is in line with the predictions of the basic life-cycle model. After retirement however, wealth remains rather flat and no sign of decline is observed. Nearing the end of life, assets are far from depleted.
A possible explanation for the observed pattern is that capital gains in housing are spent on consumption to a limited extent. In addition, housing wealth might be less liquid than financial wealth. The bottom cohort plots show the accumulation of financial assets over the life-cycle. Also for this asset class aggregate, no strong decrease nearing the end of life is observed. It even appears that financial asset decumulation is slower for home owners than for the entire set of respondents.
4.3 Evolution of Net Housing Wealth
Figure 4.2: life-cycle net housing wealth
It is striking that whilst part of households under 35 even have negative housing wealth, the oldest cohort group has such high net housing wealth. Looking at the values in Table 3.4, it is visible that these households not only have valuable housing, but also the lowest amount of remaining debt of all cohorts. The mean amount of net housing wealth for households with heads between 65 and 80 years old is equal to €220,364 whereas the median is equal to €194,000. For the group of households with heads below 35 years old these figures are equal to €45,595 and €2,000 respectively. These results hint at large capital gains in housing wealth for home owners of the two oldest age categories, among which are many members of the demographic post-World War II baby boom4. Subsequently, the horizontal sloping trend of the cohort lines indicate that these gains are not realized. However, it must be noted that richer households have a higher life expectancy and thus in all the plots wealth gets a positive bias as mortality increases. This bias however does not explain the high levels of housing wealth remaining. The profiles show that retirees extract housing wealth to a limited extent and keep high levels of net housing wealth nearing end of life.
4.4 Lifetime Consumption
Life-cycle models predict a consumption path such that lifetime utility is maximized given the amount of lifetime income available and depending on the rate of time preference versus the interest rate. Income typically increases until retirement and drops after that. In order to smooth consumption over lifetime as to keep the marginal utility of consumption constant (at the maximum level) financial wealth is used. First money should be borrowed to finance early consumption given a low actual income, then should be saved to finance post-retirement consumption given a high actual income and a decreased income during retirement.
Figures 4.3 shows the consumption patterns of Dutch households over the age of the household head. The pattern is upward sloping for young households, indicating that either they are depressing consumption. Either since they are for instance accumulating housing wealth, or due to liquidity constraints.
The consumption pattern shows that consumption is rather flat from the age of 30 until retirement, with a slight decline over time (showing that the rate of time preference is higher than the real interest rate). Despite the fact that a clearly hump-shaped consumption pattern is not observed, it is visible that consumption drops post-retirement. This is particular given the high level of wealth of this cohort. In order to check whether the consumption decrease is not solely the result of a decrease in household size, the cohort plot was also generated for singles, which showed a similar pattern.
4.5 Time Use and Home Production Retirement yields the opportunity to spend time on activities that was earlier spent on working. Aguiar and Hurst (2007) show that the opportunity cost of time of middle age individuals is 40 percent higher than that of retirees. The increased time available (as shown in Figure 4.4) affects consumption via the elasticity between time and market goods.
After retirement, Dutch households spend more time on household chores, such as cleaning, laundry, shopping, cooking, gardening and odd jobs. This increased time available, possibly leading to increased home production, is an explanation for the decreasing pattern in consumption post-retirement. Not taking account of home production would thus lead to a false rejection of the life-cycle model for Dutch households.
In addition, as housing and non-market time might be complements in home production, this could also explain the absence of the decumulation of housing wealth. Aydilek (2016) finds for the United States that whilst the elderly do not sell their financial assets nor liquidate their housing wealth, still can keep their consumption stable despite lower consumption expenditure. Figure 4.5 however shows that there is no difference between home production for tenants and the entire sample of households. Thus, we find no evidence of a higher level of home production for home owners and thus that housing and non-market time are complements.
5. Empirical Strategy and Model Specification
The previous descriptive analysis of the saving and consumption behavior of Dutch households over their lifetime, results in the following observations:
- The basic life-cycle model is soundly rejected with the existence of a Retirement Savings Puzzle (i.e. wealth is not depleted nearing end of life). - Dutch households accumulate substantial amounts of financial and
housing wealth towards retirement and do not decumulate these during retirement.
- The amount of weekly hours spent on home production increases post-retirement.
Combining these observations with concepts available in the literature, the role of housing in the Dutch Retirement Savings Puzzle is researched, leading to the following explanations:
- If desired housing consumption is constant throughout retirement, this will cause housing wealth to decline more slowly than other kinds of wealth if the ability to extract housing equity is limited (Henderson and Ioannides, 1983)
- Home production is related to housing itself and it also affects housing decisions of the elderly (Aydilek, 2016). Housing and non-market time are complements in home production, this could also explain the slow decumulation of housing wealth.
From which the following stylized facts are formulated, as to be verified by the empirical strategy:
- The effect of capital gains in housing wealth on consumption is limited due to the limited ability to extract housing wealth.
The two stylized facts are checked by modeling consumption dynamics in response to changes in Net Housing Wealth, Net Financial Assets and Household Income. For the stylized fact on capital gains the MPC out of Housing Wealth is estimated, in order to measure the effect of capital gains in housing wealth on consumption. In addition, the response to losses in housing wealth is estimated to further explore how households adjust their consumption to changes in wealth and housing wealth specifically. These estimations are separated for all home owners and retired home owners. The latter group is expected to respond differently to changes in housing wealth due to the fact that these are more likely to be long in housing than short (i.e. decreasing their investment rather than increasing). Also, the MPC out of Housing Wealth for retirees is interesting as it tells us how this group incorporates its housing wealth in its consumption planning. From this, a statement can be made about the liquidity of gains in housing wealth for retired households in relation to the Retirement Savings Puzzle.
For the verification of the stylized fact on home production, the time devoted by the household head to household chores is incorporated in the consumption model. The model is estimated separately for home owners and tenants to capture the possible interaction effect between home ownership and home production with regard to consumption. According to the life-cycle hypothesis, households adjust their consumption to their lifetime wealth. With the estimation it is attempted to determine the influence of home production on consumption, ceteris paribus (income, wealth and background characteristics), If a negative relationship is found, it is concluded that given an implied consumption level by the life-cycle model, part of that consumption is produced at home using the house as an input.
5.1 Model Specification
This analysis is a modification of Christelis et al. (2014), extended with: - The separation of coefficient estimation for wealth gains and losses. - The inclusion of changes in home production.
A similar analysis framework is used by Parker (1999), Johnson et al. (2006), Agarwal et al. (2007) and Disney et al. (2010).
For home owners:
with ; ; ; ;
For tenants: same specification, without changes in housing wealth
Where i denotes the household and
is the error term.
5.2 Estimation Results
Table 5.2 shows the results of four different estimations used to verify the stylized facts. The coefficients are interpreted as elasticities for variables expressed in differences and semi-elasticities for level variables. This section summarizes the estimation results and the consequent findings with regard to the stylized facts.
Home owners’ consumption responds to losses in Net Housing Wealth and no evidence is found of a response to gains. At a ten percent significance level it is found that a 1 percent decrease in Net Housing Wealth is associated with a 0.024 percent decrease in household consumption, ceteris paribus. This is not a very strong relationship, lending support to the conjecture that consumption is rather irresponsive to changes in Housing Wealth. There are however two sources of heterogeneity in the relationship not yet controlled for. First, the effect may be significantly stronger for the financially vulnerable, households with high debt-to-housing ratios. Second, the effect may differ with the position in the housing market the households are in, either long or short. In order to add on this, the relationship is also estimated for the oldest age cohort.
Retired households respond the other way around, that is, by increasing consumption after increases in Net Housing Wealth and no evidence is found of a response to losses. It is found that a one percent increase in Net Housing Wealth is associated with a 0.08 percent increase in household consumption, a result significant at the ten percent level. This is evidence for the first stylized fact. Gains in Net Housing Wealth stimulate consumption either via higher perceived wealth, or via relaxed borrowing constraints. However, the effect is limited (e.g. Christelis et al. (2014) also find a value of 0.056 across specifications) and thus there is plenty of room to keep accumulating Net Housing Wealth.
negative correlation between home production and consumption expenditure, but no causal relationship is found between the differences in the two variables for our data.
Finally, a negative relationship is found between household consumption and Net Financial Assets for home owners and retired home owners in particular. These coefficients are however not economically significant, i.e. they are very small. Also, no relationship is found between household income dynamics and consumption dynamics for retirees. The income of these households consists of (state) pension income and thus probably has little unexpected shocks.
(1) (2) (3) (4)
Variables Home Owners Retired Home Owners Tenants Everyone (Including HP)
HHI
0.180** 0.203 0.147*** 0.272** (0.0868) (0.151) (0.0317) (0.107) NHW + 0.00111 0.0813* 0.0365 (0.00750) (0.0488) (0.0374)NHW
- 0.0240* 0.0442 -0.00634 (0.0122) (0.0915) (0.0338)NFA
+ -0.00120*** -0.00544 -0.000307 -0.0115 (0.000413) (0.00923) (0.000527) (0.00767) NFA - -0.00126*** -0.00156*** 0.000870 -0.00117*** (0.000474) (0.000314) (0.00514) (0.000265)Age Head -0.00166 -0.000879 6.74e-05
(0.00142) (0.00421) (0.00169) Household Size 0.0402 0.0442 0.0281 (0.0304) (0.0278) (0.0272) HP -0.000449 -0.000420 (0.0105) (0.00383) Constant -0.0260 -0.110 -0.0641 0.0126 (0.0883) (0.305) (0.0905) (0.0328) Observations 359 169 116 160 R-squared 0.033 0.067 0.075 0.048
Clustered standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
5.3 Sample Size
As the regression analysis results show, the sample size is rather limited compared to vast amount of households participating in the LISS panel. First of all, this is the result of on the one hand collapsing assets solely for households in which both the head and the partner have responded in case of a household consisting of more than one member (to avoid underreporting and unwanted changes in wealth over time due to different responses). Since the regression is estimated in first-differences, also solely households were included for which data was available for at least two years. Finally, merging the Time Use & Consumption data with those on Assets and Housing also reduced the sample size due to the fact that not every household was participating in both surveys.
6. Conclusions and Discussion
6.1 Conclusions
The descriptive analysis following in this thesis indicate a rejection of the basic life-cycle/Permanent Income Hypothesis. The consumption pattern is sloped downwards after retirement whilst asset decumulation of both Net Housing Wealth and Net Financial Wealth is slower than what would be expected based on the basic life-cycle model, especially for home owners.
As this indicates that home ownership might be a piece of the Retirement Savings Puzzle in the Netherlands, the following stylized facts are checked:
- The effect of capital gains in housing wealth on consumption is limited due to the limited ability to extract housing wealth.
- Capital losses in housing wealth result in depressed consumption due to the decrease in lifetime wealth.
- Home production affects consumption expenditure and this effect is different for home owners than for tenants due to the interaction of housing and non-market time.
in Net Housing Wealth is found of 0.08 (elasticity), significant at the ten percent level. Retirees thus spend their gains in Net Housing Wealth to a limited extent and appear to partly overlook these gains in their consumption planning. This can also be due to the difficulty to liquidate their housing wealth.
No evidence is found for the second stylized fact: we do not find that home production depresses consumption after retirement. Also no different effect is distinguished for home owners versus tenants.
It is concluded that there indeed is a Retirement Savings Puzzle in the Netherlands and that the house is piece of the puzzling profiles via the limited extent through which gains in its wealth are translated to consumption.
6.2 Discussion
The present thesis shows that housing wealth in the Netherlands is accumulated until retirement and preserved during retirement. One the one hand this phenomenon is sought to be explained by the limited liquidity of housing wealth. On the other hand, via the interaction of the house with non-market time to facilitate home production. There are however more explanations for the preservation of housing wealth and the results obtained.
For example, moving involves mental costs as people become attached to the house, environment and neighbors as time passes (Aydilek, 2013). She models these mental costs in the form of habit persistence in the consumption of housing goods and finds that this also plays a role in the preservation of housing. Besides that, increased home production is measured as the time input used for household chores. It might actually be that home production does not increase, but that productivity decreases as people age. The measure of home production used does not correct for that if that is indeed the case.
7. References
Agarwal, Sumit, Liu, Chunlin & Souleles (2007). The reaction of consumer spending and debt to tax rebates: Evidence from consumer credit data. Working
Paper. Federal Reserve Bank of Chicago, 2007-10
Aguiar, M., & Hurst, E. (2007). life-cycle prices and production. The American
Economic Review, 97(5), 1533-1559.
Ang, A. (2014). Asset Management, a systematic approach to factor investing. Oxford University Press.
Aydilek A., (2013). Habit formation in housing over the life cycle. Journal of
Economic Modelling 33, 2013, 858-866.
Aydilek A., (2016)., The allocation of time and puzzling profiles of the elderly,
Journal of Economic Modelling 53, 2016, 515-526.
Buiter, W. H. (2008). Housing wealth isn't wealth. National Bureau of Economic
Research.
Campbell, J. Y., & Cocco, J. F. (2007). How do house prices affect consumption? Evidence from micro data. Journal of Monetary Economics, 54(3), 591-621.
Carroll, C. D., Slacalek, J., & Tokuoka, K. (2014). The distribution of wealth and the marginal propensity to consume.
Christelis, D., Georgarakos D. & Jappelli T. (2014). Wealth Shocks, Unemployment Shocks and Consumption in the Wake of the Great Recession, CEPR Discussion Paper 10196.
Davies, J. B. (1981). Uncertain lifetime, consumption, and dissaving in retirement. The Journal of Political Economy, 561-577.
Deaton, A. S., & Paxson, C. (1994). Saving, growth, and aging in Taiwan. Studies in
the Economics of Aging 331-362.
De Nardi, M., French, E. & Jones, J.(2010). Why do the elderly save? The role of medical expenses. Journal of Political Economy, 118(1), 39-75
Disney, R., Gathergood, J., & Henley, A. (2010). House price shocks, negative equity, and household consumption in the United Kingdom. Journal of the
European Economic Association, 8(6), 1179-1207.
Elliott, J. W. (1980). Wealth and wealth proxies in a permanent income model.
The Quarterly Journal of Economics, 509-535.
Engelhardt, G. V. (1996). House prices and home owner saving behavior.
Friedman, M. (1957). A theory of the consumption function: A study by the National Bureau of Economic Research. Princeton University Press.
Ghez, G., & Becker, G. S. (1975). The allocation of time and goods over the life cycle. NBER Books.
Gourinchas, P. O., & Parker, J. A. (2002). Consumption over the life cycle.
Econometrica, 70(1), 47-89.
Guiso, L., Haliassos, M. & Jappelli, T. (1999). Household portfolios. The MIT Press. Henderson and Ioannides 1983
Henderson, J. V., & Ioannides, Y. M. (1983). A model of housing tenure choice.
American Economic Review, 73(1), 98-113.
Hurd, M. D. (1989). Mortality risk and bequests. Econometrica: Journal of the
econometric society, 779-813.
Johnson, D. S., Parker, J. A., & Souleles, N. S. (2006). Household expenditure and the income tax rebates of 2001. The American Economic Review, 96(5), 1589-1610.
Keynes, J. M. (1937). The general theory of employment. The quarterly journal of
economics, 209-223.
Kotlikoff, L. J. (1989). What determines savings?. MIT Press Books, 1.
Modigliani, F., & Brumberg, R. (1954). Utility analysis and the consumption function: An interpretation of cross-section data. Franco Modigliani, 1.
Modigliani, F. (1971), Consumer Spending and Monetary Policy: The Linkages. Federal Reserve Bank of Boston Conference Series 5.
Muellbauer, J. N. and R. Lattimore (1994), The Consumption Function: A Theoretical and Empirical Overview, Pesaran, H. and Wickens, M.R.
Handbook of Applied Econometrics.
Palumbo, M. G. (1999). Uncertain medical expenses and precautionary saving near the end of the life cycle. The Review of Economic Studies, 66(2), 395-421. Parker, J., 1999. The Reaction of Household Consumption to Predictable Changes in Social Security Taxes. American Economic Review, 89 (4), 959-973.
Poterba, J. M. (2000). Stock market wealth and consumption. The Journal of
Economic Perspectives, 14(2), 99-118.
Teppa, F. (2014). Consumption behaviour and financial crisis in the Netherlands.
Van Ooijen, R., & Van Rooij, M. (2014). Financial literacy, financial advice and mortgage risks. DNB Working Paper.
Van Ooijen, R., Alessie, R., & Kalwij, A. (2015). Saving behavior and portfolio choice after retirement. De Economist, 163(3), 353-404.
