The Retirement-Consumption Puzzle: Theory and Empirical Evidence

Seminar Paper 2011 23 Pages

Business economics - Investment and Finance


Table of Contents

List of Figures

List of Abbreviations

1 Introduction

2 The Life Cycle Hypothesis
2.1 The Stripped Down Life Cycle Model
2.2 Consumption Smoothing

3 The Retirement-Consumption Puzzle
3.1 Household Characteristics and Life Cycle Consumption
3.1.1 Wealth and Income
3.1.2 Unplanned Events
3.1.3 Decomposing Consumption
3.2 Approaches to the Puzzle
3.2.1 Substitution Effects
3.2.2 Irrationality
3.3 Critical Assessment

4 Conclusion



List of Figures

Figure 1 The Stripped Down Life Cycle Model

Figure 2 Spending on Basic Items, Food, and Work-Related Goods and Services

List of Abbreviations

illustration not visible in this excerpt

1 Introduction

The literature on consumption behavior finds that households consistently reduce consumption at retirement.1 It documents a consumption dip of between seven and 17% on average. However, according to life cycle theory, households smooth marginal utility of consumption across time periods.2 The discrepancy between the predictions of theory and empirical findings is known as the retirement-consumption puzzle.

A deeper knowledge on retirement saving behavior is of interest for at least three reasons. First, it facilitates the testing of theoretical models like the life cycle hypothesis. Thereby, it helps to understand if and by how far individuals plan their retirement in a rational manner. Second, it gives us an idea of the financial situation of retired households.3 If the drop in consumption is due to insufficient saving behavior, advanced financial planning can help households to fill the gap. Third, governmental authorities have a particular interest in sufficient saving. In the last decades, several countries have introduced tax policies as incentives for private retirement provision.4 Thereby, households are more aware and responsible for their retirement resources. It, furthermore, helps governments to shift a part of the burden of pension payments to the private sphere, which is of interest due to a steadily growing portion of elderly people.

The consumption decline around retirement proves to be robust in the United States5, Great Britain6, Canada7, Japan8, and Italy9. In addition, the retirement-consumption puzzle has been investigated from several perspectives. Hamermesh is one of the first to observe a drop of expenses at retirement which is not in line with the life cycle model.10 Bernheim et al. suggest that the unexplained part of the decline is due to myopic financial planning.11 Furthermore, Banks et al. interpret the puzzle with insufficient retirement resources.12 Haider and Stephens show that workers generally expect decreasing expenses at retirement. However, even for expected retirement, consumption drops significantly.13 Despite that, Smith figures out that the magnitude of the consumption decline is strongest if retirement is involuntary.14 The detailed consumption analysis by Aguiar and Hurst reveals heterogeneous spending developments for consumption categories.15 They suggest that retirees substitute a significant amount of leisure time for consumption expenses (e.g., home production of food). This conclusion is reinforced by the estimations of Fisher et al.16 and Hurd and Rohwedder17. A totally different approach is chosen by Lundberg et al. who establish a martial bargaining model.18 It proposes that bargaining power shifts at retirement due to the changing income structure of spouses. Since wives have a longer life expectancy they are more inclined to save.

The task of this paper is to assess both theory and empirical evidence of the retirement-consumption puzzle. It, therefore, discusses the basic characteristics of standard life cycle theory in section 2. Section 3 examines the main determinants of the puzzle and perspectives from which the puzzle has been investigated. Section 4 concludes.

2 The Life Cycle Hypothesis

Models on consumption behavior were first established through the permanent income19 and the life cycle hypothesis20. The last-mentioned appears more applicable in the context of retirement consumption. That is mainly because the life cycle model assumes that people retire and die at a certain point in time and, therefore, save for retirement. In contrast, the permanent income model is based on the simplification of infinite life.

It is vital to understand why retirement consumption behavior is actually considered puzzling. If researchers had no theoretical framework describing life cycle behavior, there would be no puzzle. Therefore, we will introduce the basic notions of the standard life cycle model. In the following, we discuss the relationship between income, savings, consumption, and wealth of households over the life cycle. Thereafter, it is show why maximizing utility implies smoothed consumption streams across time periods.

2.1 The Stripped Down Life Cycle Model

A relationship between earnings, savings, consumption, and wealth over the life cycle is established in a basic way by the stripped down life cycle model.21 It assumes households to be in steady state (i.e., consumption and income are stable over the working life, interest rates are zero, and savings are exhausted at the time of death). The model, furthermore, assumes that retirement is a discrete event that happens at time Abbildung in dieser Leseprobe nicht enthalten from which on income from work declines to zero (see Figure 1 in the Appendix for a graphical representation). According to the model, households keep a consumption level, Abbildung in dieser Leseprobe nicht enthalten, that is below their income, Abbildung in dieser Leseprobe nicht enthalten, in each period Abbildung in dieser Leseprobe nicht enthalten before retirement (i.e., Abbildung in dieser Leseprobe nicht enthalten, Abbildung in dieser Leseprobe nicht enthaltenAbbildung in dieser Leseprobe nicht enthalten). Hence, households save and accumulate wealth until they retire. To assure a stable standard of living, retirees decumulate their wealth after Abbildung in dieser Leseprobe nicht enthaltensuch that consumption expenditures are smoothed. Thus, the wealth curve is hump-shaped.

The standard life cycle model is obviously based on some strong assumptions (e.g., zero interest rate, no bequest, steady state, etc.).22 Researchers have proposed various extensions and refinements to the life cycle model in order to adjust for specific phenomena, for instance, by correcting for precautionary savings.23 In addition, researchers usually control for demographic variables like the size of households or for time-specific discount rates in their models.24 However, a simple approach to the life cycle model is sufficient to illustrate the major implications for consumption, saving, and wealth accumulation behavior.25 Thus, we can regard this model as a theory on fundamental life cycle planning of households. Investigating consumption data in the framework of this theory reveals systematic inconsistencies, which are discussed in section 3.

2.2 Consumption Smoothing

We examine a basic approach which demonstrates that forward-looking, utility maximizing households smooth marginal utility of consumption across periods.26 Imagine households’ remaining lifetime utility in period Abbildung in dieser Leseprobe nicht enthalten is represented through

illustration not visible in this excerpt

where Abbildung in dieser Leseprobe nicht enthalten is consumption in period Abbildung in dieser Leseprobe nicht enthalten and Abbildung in dieser Leseprobe nicht enthalten is the subjective discount rate. As households’ consumption is limited by wealth and income it faces the budget constraint

illustration not visible in this excerpt

with Abbildung in dieser Leseprobe nicht enthalten and Abbildung in dieser Leseprobe nicht enthalten being wealth and income in period Abbildung in dieser Leseprobe nicht enthalten and Abbildung in dieser Leseprobe nicht enthalten representing the interest rate. As specified in Appendix 1, maximizing equation (1) subject to equation (2) yields the relationship for marginal utility of consumption between period Abbildung in dieser Leseprobe nicht enthalten and Abbildung in dieser Leseprobe nicht enthalten:

illustration not visible in this excerpt

If we assume that r equals Abbildung in dieser Leseprobe nicht enthaltenfor simplification, equation (3) reduces to

illustration not visible in this excerpt

Hence, the marginal utility of consumption in period Abbildung in dieser Leseprobe nicht enthalten equals the expected marginal utility in the subsequent period Abbildung in dieser Leseprobe nicht enthalten. Most researchers follow the approach of Hall which amongst others requires the utility function to be quadratic.27 Assuming that one-period utility is of the quadratic form

illustration not visible in this excerpt

marginal utility is linear in consumption

illustration not visible in this excerpt

Based on this, equation (4) becomes

illustration not visible in this excerpt

According to that, households do not only smooth marginal utility of consumption but also consumption levels. This eases the investigation since research-relevant surveys contain actual consumption data.

3 The Retirement-Consumption Puzzle

3.1 Household Characteristics and Life Cycle Consumption

The decline in consumption around retirement is considered puzzling in the course of life cycle theory. Although research agrees on this fact, the estimated magnitude of the consumption decrease varies considerably. To identify the main determinates of the consumption decline, we analyze expenditure structures of households depending on socio-economic factors, unplanned events, and for different consumption categories.

3.1.1 Wealth and Income

It is widely established that the socio-economic status has an impact on health and mortality.28 Therefore, the level of education, employment, income, or wealth changes when cohorts age because households with lower socio-economic status tend to die earlier. Segmenting cohorts by socio-economic factors reveals heterogeneous consumption patterns. In order not to go beyond the scope of this paper, we concentrate on households’ wealth and income levels exclusively.

In US survey data, Bernheim et al. measure considerably negative changes in consumption at retirement for all wealth and income quartiles.29 The observed decline negatively correlates with both wealth and income replacement rates (i.e., the fraction of pre-retirement to post-retirement non-asset income). While consumption of the wealthiest and of the highest income replacement quartile drops by 9.4% and 5.2%, respectively, the poorest quartile’s consumption decreases by 24%. The authors can rejected that the consumption decline is of equal magnitude for all wealth quartiles with 99.9% confidence. Apart from households that belong to both the top income and wealth quartile of the sample, all households experience a considerable drop at retirement. Even in an above average wealthy sample, Ameriks et al. show that about 90% of workers expected to spend at most the same amount after retirement with a mean expected decline of 11.3%.30 At the same time, 80% of retirees spend at most the same after retirement with a mean realized decline of 4.6%.

In consequence, the phenomenon of shrinking consumption is observed in the majority of households that pass into retirement. The results also rule out interpretations that consumption decreases solely due to insufficient retirement resources. That is because a considerable portion of wealthy households apparently reduce their consumption expenditures voluntarily.

3.1.2 Unplanned Events

In the life cycle framework, financial planning to offset income declines at retirement plays an important role.31 The model requires smooth consumption paths if households retire as expected. However, unanticipated events can force households to overthrow their current plans.32 Most cited reasons for unplanned retirement are shocks due to unemployment or ill health.33 Statistics on employment reveal that a substantial portion of people are not working before retirement.34 In addition, there is a considerable shift in employment rates in the years prior to retirement. In the British Household Panel Survey (BHPS), for instance, around 80% of people who expect to retire in five years are employed. However, this figure decreases to around 60% one year before retirement. That development is due to both increasing long-term sickness and job losses. Overall, about 41% of earlier retirements are a result of health issues or unemployment. In the Consumption and Activities Mail Survey (CAMS) of the Health and Retirement Survey (HRS), 29% of retirees consider health to be a primary factor for retirement.35 With regard to consumption development, those who report to have poor or at most fair health before and after retirement reduced consumption expenses by 20.5% on average. Conversely, retirees in a good state of health decreased expenditures by 12%.

Some factors may explain the observed differences in consumption declines. If surprising events negatively affect expected retirement wealth, the life cycle model requires a downward adjustment of expenditures, which entails a discrete drop in consumption.36 Furthermore, a negative shock can also render uncertainty on the timing of retirement resulting in more cautious spending behavior. On top of that, ill health likely results in lower appetite or shifting expenditures among consumption categories to health-related items or services. However, shifts among consumption categories are hard to identify as most surveys contain rough consumption data (primarily food data exclusively). In summary, shocks can indeed explain parts of the drop of consumption around retirement. Blau, for example, models negative shocks which force workers to retire earlier.37 He concludes that a log change in consumption expenses of 3% is consistent with the data from the HRS-CAMS survey. Thus, there remains an unexplained part of the consumption decline. As even healthy retirees systematically reduce their expenditures at retirement explanations purely based on shocks are problematic.

3.1.3 Decomposing Consumption

One main issue that is pursued in the literature on consumption is how expenditures on consumption categories change at retirement. Disaggregating total consumption allows a closer analysis of consumption behavior and figures out what is the core of the puzzle. While a sizeable part of research focuses on food consumption due to constrained survey data, there are also analyses using broader consumption data.38

Food and work-related expenditures drop considerably at retirement.39 Figure 2 in the Appendix compares expenditures on food, work-related (i.e., canteen and restaurant meals, transport, and clothing), and basic items (i.e., mainly food at home and energy) over the life cycle. Regarding the development of expenditures, one observes gradually increasing expenses in the first half of life, which peak at the age of 50 to 60. That is basically in line with findings of other research.40 The figure, furthermore, illustrates that work-related expenditures decrease heavily after peaking. Food expenditures also fall at and after retirement. The decline in basic items is mainly due to the included food expenditures. The sharp decline in work-related items is for obvious reasons reasonable when households step out of the labor market. However, explaining the consumption drop around retirement in the context of the life cycle model is difficult provided that food is indispensible to life and, thus, has a low income elasticity as compared to dispensable goods.41 In particular, recent investigations find that food spending drops at retirement, while expenditures on luxurious goods like entertainment items gradually increase.42 This issue is discussed in more detail in the next section.

Contrary to food consumption, analyses of consumption flows around retirement show no declines.43 Consumption flows are usually defined as a very broad measure, taking total expenditures on durable and non-durable goods as well as the rental equivalent on owned real estate into account. As around 80% of house owners paid off mortgages before they retire retirees obtain a substantial part of housing services. Renters’ housing rental equivalent is the actual rental fee. In the Consumer Expenditure Survey (CEX) data, consumption flows reduced by food, alcohol, tobacco, and work-related items even increase by 6% on average in the years around retirement.44 Similarly, Fisher et al. reveal that consumption flows including food and work-related expenditures increase by 2.6% on average.45 Consequently, by adding housing services to consumption the expenditure drop disappears. In this regard, the retirement-consumption puzzle figures out to be a “retirement-food consumption puzzle.” Hence, investigations of pure food consumption data are misleading. In addition, the findings contradict with the once prevailing generalization that food expenses are a strong predictor of total consumption expenses.46 Nevertheless, it still needs to be clarified why food outlay drops sharply around retirement while consumption on other goods stays constant or even increases.

3.2 Approaches to the Puzzle

3.2.1 Substitution Effects

A common attempt to explain the puzzle away is that retired households use their time surplus as a substitute for consumption.47 Examining whether retirees indeed substitute leisure for consumption, depends on their utility function. Consider the utility function Abbildung in dieser Leseprobe nicht enthalten, depending on both consumption, Abbildung in dieser Leseprobe nicht enthalten, and leisure, Abbildung in dieser Leseprobe nicht enthalten, in period Abbildung in dieser Leseprobe nicht enthalten. If the utility function is separable (i.e., marginal utility of consumption is not affected by leisure), the life cycle framework requires marginal utility of consumption to be constant throughout retirement. As retirement is a planned and discrete event for most workers new retirees experience a huge rise in leisure time. With respect to that, non-separable preferences between leisure and consumption argue for a change in consumption. The literature, in general, supports the view that households’ preferences are non-separable.48 Households may regard certain consumption categories as substitutes or as complements to leisure depending on preferences. Thus, if the substitution effect is dominant on an aggregate level, one observes consumption to decline at retirement.49 Retirees may use their additional spare time to produce goods and services which they used to purchase before retirement.


1 Cf. Fisher et al. (2005), p. 1.

2 Cf. Modigliani (1986), p. 300.

3 Cf. Smith (2006), p. 130.

4 Cf. Banks et al. (1998), p. 769.

5 Cf. Bernheim et al. (2001), p. 846.

6 Cf. Banks et al. (1998), p. 777.

7 Cf. Burbidge/Robb (1989), p. 538.

8 Cf. Wakabayashi (2008), pp. 999-1001.

9 Cf. Miniaci et al. (2003), p. 37.

10 Cf. Hamermesh (1984), p. 369.

11 Cf. Bernheim et al. (2001), p. 855.

12 Cf. Banks et al. (1998), p. 784.

13 Cf. Haider/Stephens (2007), p. 257.

14 Cf. Smith (2006), p. 142.

15 Cf. Aguiar/Hurst (2008), pp. 8-13.

16 Cf. Fisher et al. (2005), pp. 12-13.

17 Cf. Hurd/Rohwedder (2006), p. 19.

18 Cf. Lundberg et al. (2001), p. 17.

19 Cf. Friedman (1957), p. 20.

20 Cf. Brumberg/Modigliani (1954), p. 4.

21 Cf. and what follows Modigliani (1986), p. 300.

22 However, it should be mentioned that most assumptions can be relaxed in a more complex model and have little impact on the model’s implications (cf. Modigliani/Brumberg, (1954) , p. 7.)

23 Cf. Hubbard et al. (1994), pp. 175-176.

24 Cf. Banks et al. (1998), p. 775.

25 Cf. Miniaci et al. (2003), p. 2.

26 Cf. Haider/Stephens (2007), p. 248.

27 Cf. Hall (1978), p. 974.

28 Cf. Ryff/Singer, pp. 100-101.

29 Cf. and what follows Bernheim et al. (2001), p. 846.

30 Cf. Ameriks et al. (2007), pp. 269-270.

31 Cf. Modigliani (1985), p. 305.

32 Cf. Hurd/Rohwedder (2006), p. 9.

33 Cf. Tanner (1998), pp. 183-184.

34 Cf. and what follows Smith (2006), p. 135.

35 Cf. and what follows Hurd/Rohwedder (2006), pp. 18-19.

36 Cf. and what follows Hurst (2008), p. 21.

37 Cf. Blau (2008), pp. 60-61.

38 Cf. Hurst (2008), pp. 2-3.

39 Cf. Banks et al. (1998), p. 783.

40 Cf. Fisher et al. (2005), p. 10.

41 Cf. Hurst (2008), pp. 4-5.

42 Cf. Aguiar/Hurst (2008), pp. 2-3.

43 Cf. and what follows Fisher et al. (2005), p. 8.

44 Cf. Aguiar/Hurst (2008), p. 46.

45 Cf. Fisher et al. (2005), p. 28.

46 Cf. Hurst (2008), p. 11.

47 Cf. and what follows Hurd/Rohwedder (2006), p. 10.

48 Cf. Battistin et al. (2006), p. 15.

49 Cf. Smith (2006), p. 132.


ISBN (eBook)
ISBN (Book)
File size
513 KB
Catalog Number
Institution / College
University of Mannheim
Retirement Retirement Consumption Puzzle Puzzle Modigliani Hurst Friedman Kahneman Laibson LCH LCM Life Cycle Life Cycle Hypothesis Lebenszyklusmodel Lebenszyklushypothese Lebenszyklus




Title: The Retirement-Consumption Puzzle: Theory and Empirical Evidence