DNB W O R K I N G P A P E R. DNB Working Paper

Transcription

1 DNB Working Paper No. 115/September 2006 Allard Bruinshoofd and Sybille Grob DNB W O R K I N G P A P E R Do changes in pension incentives affect retirement? A stated preferences approach to Dutch retirement consideration

2 Do changes in pension incentives affect retirement? A stated preferences approach to Dutch retirement considerations Allard Bruinshoofd and Sybille Grob * * Views expressed are those of the individual authors and do not necessarily reflect official positions of De Nederlandsche Bank. Working Paper No. 115/2006 September 2006 De Nederlandsche Bank NV P.O. Box AB AMSTERDAM The Netherlands

3 Do changes in pension incentives affect retirement? A stated preferences approach to Dutch retirement considerations W. Allard Bruinshoofd *, Sybille G. Grob De Nederlandsche Bank, P.O. Box 98, 1000 AB Amsterdam, the Netherlands This version: 25 September 2006 Abstract The empirical connection between financial incentives and retirement decisions largely derives from revealed preferences in cross-sectional settings. This raises the issue to what extent unobserved tastes for retirement which may correlate with job selection and through that route with financial incentives play a role and can be controlled for. Using a stated rather than a revealed preferences approach, we contribute to this debate. Fielding a survey questionnaire in the Dutch DNB Household Survey we derive empirical estimates of pension adjustment and pension wealth effects. Our main finding is that retrenchments of pension arrangements to the effect of raising the standard retirement age by 1 year induce people on average to postpone retirement by about half a year. Retirement postponement varies across people, depending prominently on earnings and non-pension wealth; affluent people are more likely to capitalize on increased pension wealth through earlier retirement, whereas they more readily accept a lower pension benefit in case of a decrease in pension wealth. JEL classification: D12; D80; J26 Keywords: (Early) retirement; Financial incentives; Survey results * Corresponding author. Tel.: ; fax: address: w.a.bruinshoofd@dnb.nl (W.A. Bruinshoofd) Acknowledgements We are grateful to Jan-Marc Berk, Diederik Dicou, Peter van Els, Ralph de Haas, Lex Hoogduin, Jessica du Marchie Sarvaas, Henriëtte Prast, Maarten van Rooij, Frederica Teppa, Corrie Vis, Peter Vlaar and Jasper de Winter for their comments and advice regarding the questionnaire and/or previous versions of the paper. We also thank seminar participants at NETSPAR and De Nederlandsche Bank. The remaining errors should be attributed to us alone. The opinions set out in this paper do not necessarily reflect those of De Nederlandsche Bank or the European System of Central Banks.

4 1. Introduction Since people may self-select into jobs with pension schemes that fit individual tastes for retirement it is important to control for unobserved heterogeneity in these tastes when studying the role of pension incentives in individuals retirement decisions. Using repeated observations on pension incentives and retirement expectations Chan and Stevens (2004) illustrate this point empirically. They measure how changes in pension incentives generate systematic updates of retirement expectations while alternately controlling and not controlling for any individual fixed retirement tastes. Relative to not controlling they find that controlling for individual fixed effects slashes the estimated effect of pension incentives on retirement expectations by half. Conditioning on initial expectations and preferences, an experimental setup may alternatively be well-suited to avoid the contaminating effect of unobserved heterogeneity in retirement tastes. This is what we set out to do in this paper, fielding an extensive survey questionnaire with hypothetical changes in pension incentives in the Dutch DNB Household Survey. Additionally, we tailor our experiments so as to distinguish pension incentives that pertain to changes in accumulated pension wealth and those that pertain to changes in pension accrual from additional years of work. The (stated) behavioral response to the former is labeled the pension wealth effect and that to the latter the pension adjustment effect. Let us briefly discuss two recent (proposed) Dutch pension reforms to illustrate the relevance of these separate pension incentives in the redesign of a pension system. The first illustration concerns the implementation of and increases in the accrual of pension wealth within occupational pension schemes for additional years of work and the corresponding diminution of pension wealth following early retirement. These reforms were inspired by the recognition of perverse labor supply effects embedded in previous pension schemes and their labor participation effects strongly depend on pension adjustment effects. The second illustration concerns contemporaneous political debates revolving around the affordability of social security in an aging society. A frequently proposed reform to ensure sustainability is to broaden the tax base so that the growing numbers of retired contribute to the funding of social security payments. In a stylized setting, this implies a retrenchment in terms of net social security payments, which corresponds to a reduction in expected lifetime pension wealth. Hence the labor participation effects of such a reform lean heavily on pension wealth effects. The price we pay by using survey questionnaires to collect our data lies in the potential sensitivity of survey results to framing issues. We perform several robustness checks to ensure that our main results are not driven by framing-related issues. Specifically, we check the sensitivity of our results to deliberate variations in the way we present financial incentives to the panel. Despite this potential drawback the DNB Household Survey provides a unique opportunity to relate individual retirement expectations to various individual and household background characteristics such as household composition and household finances in an experimental setup. This allows for an extensive mapping of which types of respondents seem most sensitive to the distinguished pension incentives. 1

5 As we use a different dependent variable the willingness to postpone retirement rather than actual retirement our results do not easily compare in quantitative terms with results in related papers. Nevertheless, the relative importance of the pension adjustment and the pension wealth effects in individual retirement expectations provides an important piece of information to assess the labor participation effects of various reforms to the pension system. Our results summarize as follows. We find that the response to retirement incentives is as expected and similar in qualitative terms to findings in related literature. More specifically, people tend to postpone retirement when larger shares of the consequent gains accrue to them in terms of pension accrual (i.e. we find a positive pension adjustment effect). In addition, people tend to postpone retirement when they are faced with a decrease in their pension wealth (i.e. we find a negative pension wealth effect). While the sign of the pension adjustment effect and the sign and size of the pension wealth effect are robust to various ways of presenting these incentives to the panel, the size of the pension adjustment affect appears rather sensitive to framing. We also find that the responsiveness to retirement incentives differs substantially between individuals, most prominently between members of more and less affluent households. For example, the more affluent a household, the more likely are its members to capitalize on increases in pension wealth through early retirement, while they display on average a stronger tendency to accommodate decreases in pension wealth through lower pension benefit levels. These results reiterate the redistributional aspects of pension system redesign. The next section of the paper gives a brief tour through the literature on the factors that drive individual retirement expectations and decisions. Section 3 describes the survey data used and presents initial results on retirement expectations. Section 4 presents and discusses our empirical measures of the pension adjustment and pension wealth effects and section 5 concludes. 2. Pension incentives and retirement: a concise survey of the literature It is generally acknowledged that individual retirement decisions are underlain by financial considerations (e.g. Gruber and Wise, 1997). The emergence of Social Security and institutionalized pension plans in the US, for example, is linked up with early retirement and falling participation rates of the older segments of the labor market (e.g. Sheshinski, 1978; Boskin and Hurd, 1978; and Crawford and Lilien, 1981). Also, the design of the pension system may seem to have considerable effects on the behavior of employees. Although relative to individual pension arrangements collective arrangements provide fewer options to retire very early, Disney et al. (1994) shows that the retirement hazard of UK employees with collective arrangements rises steeply around the pensionable age. In this concise survey of the literature we distinguish pension incentives that pertain to changes in accumulated pension wealth and those that pertain to the pension accrual from additional years of work. The (stated) behavioral response to the former is labeled the pension wealth effect and that to the latter is labeled the pension adjustment effect. 2

6 2.1. Pension adjustment effects The pension adjustment effect refers to the change in (stated) retirement behavior following a change in the accrual to pension wealth from additional years of work. Figure 1 provides a highly stylized representation of the pension adjustment effect. In it, the horizontal and vertical axes represent lifetime consumption of leisure and lifetime consumption of goods and services, respectively. We assume that leisure can only be consumed upon retirement. That way, changes in the consumption of leisure can be interpreted as changes in the timing of retirement. 1 The slope of the section XY of the budget constraint reflects the total proceeds from working, i.e. the wage received while working, as well as discounted pension entitlements accruing from working. Initially, given preferences U 1, the optimal division of lifetime to work and leisure results in the consumption of L 1 of leisure and lifetime consumption of goods and services C 1. Figure 1 A stylized representation of the pension adjustment effect Lifetime consumption of goods and services (C) X U 1 U 2 C 2 X 2 C 1 C Y L 1 L 2 L 1 Z Lifetime leisure (L) An increase in the accrual to pension wealth from additional years of work tilts the XY section of the budget constraint for example to X Y. In figure 1 this results in a decrease in the consumption of leisure from L 1 to L 2 (i.e. postponed retirement) as a result of a Hicksian substitution effect L 1 -L 1 and an opposing income effect L 1 -L 2. This representation agrees with Sheshinski (1978) who demonstrates theoretically that retirement may be postponed when benefits are connected positively to retirement age, i.e. the pension adjustment effect is positive. Up until recent years many pension schemes were based on accrual rates that from an actuarial point of view did not sufficiently compensate participants for additional years worked. In their theoretical analysis of the connection between retirement decisions and 3

7 pension accrual Crawford and Lilien (1981) conclude that the effects associated with [ ] deviations from actuarial fairness probably tend to advance retirement (p. 505). Empirical studies by and large confirm a positive pension adjustment effect. Samwick (1998) finds that the accrual of pension wealth from working just one additional year is highly significant in retirement decisions of US employees. Berkel and Börsch-Supan (2003) analyze the sensitivity of German retirement decisions to pension adjustment. Simulations demonstrate that the effective retirement age goes up when the implicit tax on continuing to work is reduced. Duval (2003) looks at all OECD countries and arrives at a similar conclusion. The Dutch, too, are motivated by financial considerations; Mastrogiacomo et al. (2004) find that incentives and pension options are major determinants of retirement decisions. More specifically, comparing changes in early retirement incentives in selected Dutch pension schemes Euwals et al. (2005) conclude that a higher accrual to pension wealth from additional years of work induces workers to postpone retirement Pension wealth effects The pension wealth effect refers to the change in (stated) retirement behavior following a change in accumulated pension wealth or guaranteed pension benefits. 2 Figure 2 provides a stylized representation of the pension wealth effect. Starting from the same initial situation as in figure 1 above, figure 2 demonstrates the effects of an upward shift of the budget constraint to X Y Z. At given preferences this shift of the budget constraint implies a higher lifetime consumption of goods and services (C 2 ) as well as a higher lifetime consumption of leisure (L 2, i.e. earlier retirement). Clearly, therefore, the pension wealth effect relates to the financial feasibility of retiring (early). In this regard and referring to the US social security system Boskin (1977) concludes that the income guarantee has an enormous impact on retirement decisions, advancing retirement. In agreement with this conclusion Meghir and Whitehouse (1997) demonstrate that higher earnings on the job tend to reduce the retirement hazard of UK employees, while higher social security benefits tend to enhance it. 3 Using Health and Retirement Study data Coile and Gruber (2000) confirm a negative pension wealth effect in retirement decisions of older Americans. Bloemen (2006) obtains a negative pension wealth effect from the Dutch Socio-Economic Panel. While many studies thus conclude that a negative pension wealth effect is an important element in individual retirement decisions, Chan and Stevens (2004) argue that such findings may partly derive from the use of revealed retirement preferences data in combination with unobserved heterogeneity in tastes for retirement. Specifically, applied to the Health and Retirement Study data they show that 1 This implicitly disregards the possibility of part-time retirement. 2 Of course, financial and housing wealth may substitute for pension wealth. Blake (2004) shows for the UK that in addition to pension wealth, housing and other financial non-pension wealth also stimulates early retirement. Van Els et al. (2004) find indications that for Dutch households, too, housing wealth may substitute for pension wealth. We therefore include information on total household non-pension wealth among the explanatory variables when relating stated pension wealth effects to individuals background characteristics in the next sections. 3 Along these lines Diamond (2001) points out that an effective pension system reform which confronts people more explicitly with the actual costs of (early) retirement needs supporting measures that block the abuse of alternative ways of withdrawing from the labor force (disability or unemployment schemes). 4

8 properly controlling for unobserved individual-specific tastes for retirement may slash the estimated wealth effect by half. 4 Figure 2 A stylized representation of the pension wealth effect Lifetime consumption of goods and services (C) X U 1 U 2 X 2 C 2 C 1 1 Y Y L 1 L 2 Z Lifetime leisure (L) One way to reduce the confounding effect of unobserved heterogeneity in individual tastes for retirement on estimated pension wealth effects is to focus on selected pension schemes. From the assumption that individual-specific tastes for retirement via job selection result in participation in individually preferred pension schemes namely, it follows that the tastes for retirement of participants in specific pension schemes may be relatively homogenous. In that regard, using observed retirement data in pension schemes for civil servants Euwals et al. (2005) find that lower pension wealth induces Dutch civil servants to postpone retirement, i.e. confirming the negative pension wealth effect. Another way around the problem is to use stated preferences data in an experimental framework, as we will do in this study. Then, retirement expectations are conditional on individual-specific contemporaneous financial incentives while hypothetical changes in financial incentives generate stated behavioral effects. Hence even if respondents have selected into jobs with pension arrangements to match individual tastes for retirement, the stated behavioral effects nevertheless reflect the stated responses to incremental financial incentives. Van Soest et al. (2006) estimate a structural life-cycle model on stated retirement preferences data using survey results from Dutch households. Their results 4 To interpret the magnitude of effects, note that Chan and Stevens (2004) find that without controlling for unobserved individual-specific tastes for retirement, pension incentives on average serve to raise the self-assessed probability to continue working beyond the age of 62 with four percentage points for women and six percentage points for men. Properly controlling for unobserved individual-specific tastes for retirement reduces these effects to two and three percentage points, respectively, for women and men. 5

9 suggest a negative pension wealth effect; reductions in pension wealth tend to lead to postponement of retirement. 3. The survey data and initial retirement expectations 3.1. The data For the purpose of this analysis we have fielded an additional survey questionnaire in the DNB Household Survey (DHS), a representative panel of Dutch households that is being consulted annually and extensively on household finances, retirement issues and various other items. Our additional questionnaire contains detailed direct and indirect questions about preferences and expectations regarding postponing retirement and the role played by changes in financial incentives. The additional questionnaire was fielded in the weekend of January 8, 2005 among the DHS panel members of age 18 and above who were at that time working for pay, searching for work, studying, or retired. The questionnaire was presented to the 1,994 panel members who satisfy these criteria, of which 1,614 filled it out entirely. The response rate is therefore 80.9%. For the purpose of the present study we use only questions filled out by non-retirees which are 1,212 respondents. 1,081 of them are working for pay, 50 are looking for work and the remaining 81 are studying. Appendix A contains the main survey questions. While the additional questionnaire allows us to tailor questions specifically to the needs of our analysis, the connection with the information from the regular consultations of the panel provides useful individual and household background characteristics. Using regression analysis, we can thus check whether individual responsiveness to financial incentives is equal for all individuals or varies systematically with individual or household background characteristics. Individual characteristics are the respondent s age, gender, educational attainment, and when working the self-assessed physical and mental strain of the job. Household characteristics firstly describe the composition of the household in terms of whether a partner is present in the household, whether the respondent is household head or not and the number of dependent children in the household. Additionally, the regular DHS questionnaires deal with many aspects of the financial position of Dutch households (e.g. Alessie et al., 2002) which allows for the construction of monthly household income and household wealth. Lastly, we construct indicators of the respondents knowledge of their own pension provisions, as well as their impatience and risk tolerance when it comes to taking decisions. 5 Impatience gives an indication of the time horizon of the respondent s decision, a high degree of impatience meaning that the respondent tends to value short-term proceeds relatively more than proceeds which will take longer to realize. The risk-averse, however, may similarly prefer short-term to longer-term proceeds because the latter are more uncertain. In the following analyses, therefore, when we control for impatience we always take on board risk 5 We derive an individual s level of pension knowledge from the answers to four questions about specific aspects of the respondent s own pension arrangements (cf. Van Els et al., 2004) using factor analysis. Similarly, impatience derives from the degree to which respondents agree with twelve statements about the future, in combination with the answers to two questions on spending habits. Lastly, risk tolerance is gleaned from the measure of agreement evinced by the respondent with six questions about saving and risk-taking. Refer to Appendix B for the technical details. 6

10 tolerance as well, so as not to confuse the two when interpreting the results. The background characteristics of the respondents are defined as in table 1. Table 1 Some background characteristics All respondents Working for pay Variables Mean # Obs. Mean # Obs. Gender Age Education - Primary (%) - Lower vocational (%) - Intermediate vocational (%) - Higher vocational (%) - Professional (%) - University (%) Household head Partner Children Income - Income < 1,150 (%) - 1,150 Income < 1,800 (%) - 1,800 Income < 2,600 (%) - 2,600 Income (%) Home ownership Wealth - Wealth < 18,000 (%) - 18,000 Wealth < 120,000 (%) - 120,000 Wealth < 240,000 (%) - 240,000 Wealth (%) Pension knowledge Impatience Risk tolerance Physical strain Mental strain Notes: mean values use observations weighted by income and home ownership. The sample that we consider consists of respondents working for pay, respondents seeking work for pay, and students. Gender is a dummy variable that takes value 1 for males and 0 for females; age is measured in years; education denotes qualified educational attainment where 1 = primary education, 2 = VMBO (lower vocational education), 3 = HAVO/VWO (general /intermediate vocational education), 4 = MBO (higher vocational education), 5 = HBO (professional education), 6 = WO (university education); household head is a dummy variable that takes value 1 for household heads and 0 otherwise; partner is a dummy variable that takes value 1 for respondents with a live-in partner and 0 otherwise; children is a dummy variable that takes value 1 for respondents whose household includes dependent children and 0 otherwise; income denotes net monthly household income where 1 = income below EUR 1,150, 2 = income between EUR 1,150 and EUR 1,800, 3 = income between EUR 1,800 and EUR 2,600, 4 = income above EUR 2,600; home ownership is a dummy variable that takes value 1 for home owners and 0 otherwise; wealth denotes net household wealth (including net housing wealth) where 1 = wealth below EUR 18,000, 2 = wealth between EUR 18,000 and EUR 120,000, 3 = wealth between EUR 120,000 and EUR 240,000, 4 = wealth above 240,000; pension knowledge captures the degree of knowledge about one s own pension provisions, a higher value indicating more knowledge; impatience captures respondents emphasis on the present in decision-making and action-taking, a higher value indicating a higher degree of impatience; risk tolerance measures the degree to which respondents are prepared to accept risk in exchange for a higher return; physical strain denotes the self-assessed physical burden of the respondent s job, on a scale of 1 (very light) to 5 (very strenuous); mental strain denotes the self-assessed mental burden of the respondent s job, on a scale of 1 (very light) to 5 (very demanding). 7

11 Table 2 The determinants of expected retirement age Explanatory Estimated parameters (absolute t-values in parenthesis) variables (1) (2) (3) (4) (5) (6) Age 0.03 (2.5) 0.04 (2.0) 0.02 (0.8) 0.02 (1.4) 0.02 (1.1) Household head 1.00 (2.5) 2.04 (3.3) 0.88 (1.4) 1.30 (2.3) 0.70 (1.0) Education 0.31 (3.2) 0.31 (2.6) 0.20 (1.7) 0.15 (1.3) 0.15 (1.4) Partner (3.0) (2.5) (1.4) (1.8) (2.4) Children 0.77 (2.6) 0.54 (1.5) 0.30 (0.8) Gender (3.3) (2.2) (2.7) (2.1) Gender Partner Income Wealth Pension knowledge Impatience Risk tolerance Physical strain Mental strain (1.0) (0.9) (2.3) (3.0) (2.0) 0.11 (0.5) (0.2) (0.4) (1.2) (0.4) 0.01 (0.7) 0.47 (0.7) 0.21 (1.7) (1.8) 0.26 (0.7) 0.76 (1.4) 1.13 (1.3) (2.7) (2.0) 0.09 (0.4) (0.1) (0.4) (1.2) (0.3) (1.9) (2.3) (1.4) 0.30 (0.6) 1.41 (1.7) (1.7) (2.3) (1.5) # observations Adj-R Notes: OLS estimation results. Dependent variable is expected retirement age. All variables are defined as before. Absolute t-values are in parentheses An outline of initial retirement expectations To get a flavor of the data in general and initia l retirement expectations in particular we first turn to selfreported expected retirement age and its determinants. On average, DHS panel members expect to retire at an age of 62.1 years, only slightly higher than recent estimates of the effective Dutch retirement age of 62 years. 6 Next, we link retirement expectations to individual and household characteristics using regression analysis. Table 2 presents the results. The results in column 1 indicate that older 7, more highly educated and male respondents tend towards a somewhat higher expected retirement age. We discuss the gender effect more carefully shortly. The results furthermore demonstrate that household composition has substantial explanatory power. Specifically, household heads expect to retire on average one year 6 See for example Ageing and Employment Policies for the Netherlands, speech by Ms. Martine Durand, Deputy Director of OECD Directorate for Employment, Labour and Social Affairs, September 2005 ( 7 A squared term on age has also been considered. This does not modify our conclusions, though, while the joint statistical significance of age and squared age is rather low. 8

12 later. Having a partner in the household more than offsets this effect, though, while the presence of dependent children in the household tends to raise the expected retirement age. The latter effect may suggest that raising children is costly and through the wealth channel increases the expected retirement age. Indeed, once we explicitly control for household wealth (column 2) the dependent children effect loses much of its statistical significance. Wealth and income have strong and negative effects on expected retirement age, which again supports the hypothesis that the expected retirement age is influenced in part by financial feasibility (cf. Chan and Stevens, 2004; De Nederlandsche Bank, 2004). Using the parameter estimates in column 2 a move from the lowest to the highest income bracket (defined as per table 1) lowers the expected retirement age by 0.8 years. The effect of wealth is even stronger; moving from the lowest to the highest wealth bracket (defined as per table 1) lowers the expected retirement age by 1.6 years. The results in column 3 show that little additional explanatory power is gained from adding either our measures of knowledge, impatience and risk-tolerance or the self-assessed mental strain of the job. There is some negative effect, though, from the self-assessed physical strain of the job. In column 4 the gender effect is more carefully explored. There, the results show that a sizeable part of the gender (and household head) effect is captured by the interaction of gender and partner. After removing redundant variables, the estimates in columns 5 and 6 show that the gender effect all but disappears once this interaction effect is taken into account. 8 This suggests that males typically expect to retire later when they are part of a household that also contains a partner. 4. Pension design as a tool to extend working lives: a stated preference approach Having had a first glance at initial retirement expectations, we now turn to the sensitivity of these stated expectations to changes in financial incentives, delineated into changes in pension adjustment and changes in pension wealth. Since in a stated preferences setup we measure directly the stated response in retirement expectations to hypothetical changes in financial incentives, we feel that this setup is particularly suited as a way around the possible existence of heterogeneous tastes for retirement. At the same time, however, we have to bear in mind that in a stated preferences setup the results may be sensitive to the way in which the hypothetical changes in incentives are framed. Specifically, prospect theory describes a framing effect in the editing phase of a choice process and this may affect the subsequent phase of evaluation (Kahneman and and Tversky, 1979). Framing effects may arise from several features of individual evaluation procedures. For example, individuals may be risk averse in the gains domain and risk seeking in the loss domain (Kahneman and Tversky, 1979). Also, preferences may not be well-defined in combination with an aversion to opt for extreme choice options (Bernartzi and Thaler, 2002; Van Rooij et al., 2006). Furthermore, individuals may discount hyperbolically, feeding a status-quo bias and procrastination (Choi et al., 2003). It is therefore good practice to test the robustness 8 Alternatively, interactions of gender with household head, gender with education and gender with dependent children were considered. These could not explain the gender effect. 9

13 of preferences by deliberate attempts to frame a decision problem in more than one way (Kahneman and Tversky, 1983, p. 344) and this is exactly what we aim to do in our survey design. First, using two lists of questions, we seek to obtain direct measures of the pension adjustment (section 4.1) and pension wealth effects (section 4.2) in retirement decisions. We deliberately vary the formulation of these questions so as to get an indication of the extent to which our results are driven by framing issues. Using a third list of questions, we obtain indirect measures of pension adjustment and pension wealth effects (sections 4.3 and 4.4). Comparison of directly and indirectly measured stated preferences provides further insight into the robustness of our results to framing issues Pension adjustment effects The recent redesign of occupational pension schemes in the Netherlands aiming among other things to discourage early retirement was strongly based on the idea that pension accrual should better reflect actuarial adjustment of the pension benefit to the timing of retirement. Staring from an initial situation where early retirement was implicitly subsidized these reforms thus built strongly on a positive pension adjustment effect. In this section we analyze direct estimates of stated pension adjustment effects and assess whether people indeed take pension adjustment into consideration when thinking about the timing of their retirement. Furthermore, we relate the extent to which they do so to a set of individual and household background characteristics. We obtain our (direct) measure of the pension adjustment effect by asking the members of the DHS about the effort they are willing to exert to increase their annua l pension benefits by a given amount. Specifically, the panel was first asked whether it would be willing to postpone retirement in exchange for a 5 percentage point increase in the annual pension benefit. Alternative to postponing retirement respondents could state their willingness to pay a higher pension contribution, or a combination of a higher pension contribution and postponed retirement. Lastly, respondents could indicate that they were neither prepared to postpone retirement nor to pay a higher pension contribution. The results of this query are in panel A of table 3. More than half of the panel states that it is willing neither to postpone retirement nor to pay a higher pension contribution so as to raise the annual pension benefit by 5 percentage points. Still, about one third of the panel is willing to postpone retirement partly in combination with a higher pension contribution while a further 10 per cent is prepared only to pay a higher pension contribution. As a first check on the importance of framing effects in these results, the panel was subsequently asked whether it would accept a 5 percentage point reduction in the annual pension benefit in exchange for early retirement, a lower pension contribution, or a combination of early retirement and a lower pension contribution, see also panel B of table 3. In line with our earlier results, more than half of the panel indicates that it cannot be enticed along these lines to accept a lower pension benefit. In contrast with the above results, however, of those respondents for whom a lower pension benefit is somehow acceptable, a larger share now requires a lower pension contribution as part of the compensation, while a smaller share requires full compensation in terms of early retirement. This is 10

14 already suggestive of the role of financial feasibility in the decision to postpone or advance retirement relative to initial expectations. We will return to this issue shortly. Table 3 Changing the length of working lives: pension adjustment effects PANEL A What effort would you make to increase your pension benefit by 5 percentage points? Response Average effort (as % of total) # years %-points Work longer Pay a higher pension contribution Work longer and pay a higher pension contribution There is no satisfactory effort along these lines T-test of equal effort 3.87 ** 2.77 ** PANEL B What compensation do you demand for a decrease of your pension benefit by 5 percentage points? Response Average compensation (as % of total) # years %-points Retire earlier Pay a lower pension contribution Retire earlier and pay a lower pension contribution There is no satisfactory compensation along these lines 52.8 T-test of equal compensation 2.06 * 2.62 ** Notes: Observations (1,121 in total) are weighted by income and home ownership. Additional pension contributions were presented to the panel as a fraction of gross wages. T-tests of equal effort/compensation assume equal variances. Statistical significance at the 5% and 1% error level is indicated by * and **, respectively. The panel was subsequently asked to quantify its willingness to work and/or pay for a higher pension benefit as well as its required compensation for a lower pension benefit, see the last two columns of table 3. Respondents willing to postpone retirement in exchange for a higher annual pension benefit indicate that they are prepared to do so for slightly more than two years. Those willing to pay a higher pension contribution are willing to pay an additional 2½ percentage points of their gross wages on average. Respondents who opt for a combination of working longer and paying a higher contribution turn out to be willing to postpone retirement by about 1.7 years, and to pay a higher contribution averaging 2 percentage points of gross wages. When opting for a combination of work and pay, then, the willingness to work as well as the willingness to pay seems slightly lower than when the effort is expressed solely in willingness to work or willingness to pay. This feature is confirmed statistically by 11

15 the large t-test statistics of equal effort in the bottom of panel A. 9 Turning next to the quantification of the required compensation for a lower pension benefit in panel B of the table, note that typically this compensation exceeds the effort that the panel is willing to exert for a higher pension benefit. This is most pointedly the case when compensation is required in the form of lower pension contributions; a 4¼ percentage point cut in pension contributions is required to compensate for a decrease in pension benefits, whereas a mere 2½ percentage points are being offered for the same increase in pension benefits. Similarly, respondents seeking compensation in a combination of earlier retirement and lower pension contributions wish to retire 2¼ years earlier and to receive a 4 percentage point cut to pension contributions. 10 The results in table 3 allow us to construct the following estimates of the pension adjustment effect. Since in the case of an increase in pension wealth 21 per cent of the panel postpones retirement by 2.05 years while another 13 per cent postpones it by 1.68 years (the latter in combination with an increased pension contribution), the average pension adjustment effect is 0.6 years. Off course, we may not feel entirely comfortable with the stated willingness to postpone retirement by those respondents who are willing to do so in combination with higher pension contributions particularly as no trade-off between the willingness to work and the willingness to pay is detected in their responses (see footnote 9). Alternatively therefore, we may assume that these respondents if pressed would choose only to pay a higher pension contribution while not postponing retirement at all. Then, our estimated pension adjustment effect is 0.4 years. Likewise, we may assume that these respondents ultimately choose in the same way as those who only postpone retirement (by 2.05 years). In that case, the estimated pension adjustment effect is 0.7 years. To summarize, a pension accrual of 5 percentage points motivates people on average to extend their working lives by 0.4 to 0.7 years. Reasoning similarly, a decrease in pension benefits of 5 percentage points is acceptable to the panel if on average it is allowed to advance retirement by 0.4 to 0.8 years. As the pension adjustment effect may differ across individuals we subsequently confront the results in table 3 with individual and household background characteristics using logistic regression analysis. 9 Nevertheless, we do observe a positive correlation between the willingness to pay in the form of postponed retirement and the willingness to pay in the form of higher pension contributions. While this may indicate that for these respondents an increase in pension benefits is indeed extremely valuable, it may alternatively suggest that the succession of the questions may not be fully understood. Specifically, respondents opting for the combination of postponing retirement and paying higher pension contributions are first asked how many years they are prepared to postpone retirement and subsequently how much extra pension contribution they are willing to pay. If these questions are interpreted as complements as intended in the survey design it may typically be expected that respondents expressing a stronger willingness to postpone retirement also express a weaker willingness to pay extra pension contributions. Put differently, we then expect a negative correlation between the number of years respondents are prepared to postpone retirement and the amount of extra pension contributions they are willing to pay. If, however, these questions are interpreted as substitutes a positive correlation in their respective answers may follow. 10 Again we observe a positive correlation between the required compensation in terms of early retirement and reduced pension contributions. In addition, respondents opting for the combination of retiring early and paying lower pension contributions demand a larger advancement of retirement than respondents requiring full compensation through earlier retirement. 12

16 Table 4 The determinants of pension adjustment effects PANEL A An increase in pension benefits by 5 percentage points Explanatory variables is valued is valued in the form of longer work higher combination contribution (1) (2) (3) (4) Age Squared age ( 100) Education Partner Children Gender Income Wealth Mental strain -3.3 (3.5) 3.2 (2.8) 2.0 (1.9) -9.6 (2.8) 6.5 (2.2) -8.5 (3.2) (2.8) -8.4 (1.7) (3.0) 6.3 (1.7) 5.2 (1.7) 7.1 (1.8) 3.3 (1.2) -4.6 (1.5) 5.0 (2.3) 6.6 (2.0) # observations Pseudo-R PANEL B A decrease in pension benefits by 5 percentage points Explanatory variables can be can be compensated in the form of compensated early retirement lower combination contribution (1) (2) (3) (4) Age Squared age ( 100) Education Gender Income Wealth Pension knowledge Physical strain 3.4 (2.1) 3.7 (1.6) -8.5 (2.6) 3.3 (2.3) -3.4 (1.9) -4.7 (2.2) 5.5 (2.1) -4.4 (2.4) 5.3 (1.8) -6.1 (2.2) -6.2 (1.7) 6.3 (1.5) (3.3) 12.7 (3.0) 10.1 (2.0) 6.0 (1.9) # observations Pseudo-R Notes: Respondents are said to value an increase in pension benefits if they are willing to work longer and/or pay a higher pension premium for a hypothetical increase in pension benefits of 5 percentage points. Similarly, we infer that respondents can be compensated for lower pension benefits if they are willing to accept a shorter working life and/or a lower pension contribution for a hypothetical decrease in pension benefits of 5 percentage points. The correlation with individual background characteristics is explored using logistic regression analysis. The correlation of a specific effort/compensation (in the form of a change in the length of the working life, a change in the pension contribution, or both) with individual background characteristics is explored using multinomial logistic regression analysis. Marginal effects (in percentage points) of changes in explanatory variables on the chance of a positive answer are reported in the table. For 0-1 dummies, the marginal effects refer to discrete changes in the dummy variables from 0 to 1, for the other independent variables the sample average is used as the reference value. Absolute t-values are shown in parentheses. Reported specifications result from a selection among the variables age, squared age, household head, education, partner, children, gender, income, wealth, pension knowledge, impatience, risk tolerance, physical strain, mental strain (all defined as before). See the main text for details on the variable selection process. 13

17 The parsimonious regression results are in table Panel A of the table first characterizes the type of respondent that values an increase in pension benefits at all (column (1)). It turns out that it is mainly the young, male, single and highly educated respondent who is willing to work and/or pay for an increase in pension benefits. 12 From among these respondents columns (2) - (4) characterize which ones prefer to postpone retirement, pay a higher pension contribution, or opt for a combination of the two. Those willing to postpone retirement typically have lower educational attainment, are less affluent and have jobs with relatively low levels of self-assessed mental strain. In contrast, those willing to pay higher pension contributions are typically more affluent (in terms of both household income and household wealth) with relatively mentally straining jobs. These results already suggest that financial feasibility strongly determines whether individuals can be stimulated to postpone retirement in exchange for higher pension benefits. Interestingly, individuals from households with dependent children are much less willing to pay higher contributions for a higher pension benefit. Opting for a combination of postponing retirement and paying a higher pension contribution are typically older, well-educated and wealthy respondents. Similar patterns emerge when we look at what drives whether and in what way respondents indicate that they can be compensated for a reduction in pension benefits (panel B of table 4). Wealthy respondents are more likely to indicate that they may accept a lower pension benefit. They would typically opt for a combination of early retirement and lower contributions as compensation. The less affluent respondents already exhibit a lower probability to accept a reduction in pension benefits and if they do, they opt for lower contributions rather than early retirement. Also, older respondents are more likely to seek compensation in terms of early retirement than younger respondents, who require lower contributions or a combination of early retirement and lower contributions Pension wealth effects Another way to stimulate extension of working lives is to consider increases in the statutory retirement age. As changes in the statutory retirement age ultimately affect the present value of expected social security as well as occupational retirement benefits, its long-run effects in terms of the effective retirement age lean heavily on a negative pension wealth effect. In this section we analyze direct estimates of the pension wealth effect and relate individual pension wealth effects to a set of individual and household background characteristics. We obtain our (direct) measure of the pension wealth effect by confronting the members of the DHS with a change in pension wealth and subsequently asking them how they expect to respond to it. Specifically, we ask them to what extent they think they will accommodate wealth changes through 11 Initial regressions contain age, gender, education, partner and job, all defined as before. We then alternately add three blocks of variables. The first block consists of income and wealth. The second block contains awareness, impatience and risk tolerance. The third block contains physical and mental strain. We also ran regressions that include all three blocks of variables simultaneously. We proceed with those variables that are statistically significant in any one of these specifications and from this set sequentially remo ve redundant variables to arrive at the parsimonious specifications reported in table 4. 14

18 changes in their working lives on the one hand and changes in their pension benefit levels on the other (see questions 3 and 4 in appendix A). As a first check on possible framing effects, we confront the panel with an increase as well as a decrease in pension wealth. The results are summarized in Table 5. Table 5 Changing the length of working lives: pension wealth effects Weight on adjustment of length of working life relative to adjustment in pension benefit level following a change in pension wealth 0 (no adjustment through change in length of working life) 0 < weight < < weight < 1 1 (full adjustment through change in length of working life) Increase in pension wealth Decrease in pension wealth (response as % of all respondents) Notes: Observations (1,121 in total) are weighted by income and home ownership. Mean weight on adjustment through a change in the length of the working life following an increase (decrease) in pension wealth is 0.54 (0.53). Starting with a hypothetical increase in pension wealth, we ask respondents to indicate whether they respond by retiring earlier, enjoying a higher pension benefit, or a combination of the two. 40 per cent of the respondents indicate that they accommodate the wealth increase entirely through early retirement, while another 6 per cent accommodate it mainly though not entirely through early retirement. In response to a hypothetical decrease in pension wealth, similar shares of the respondents accommodate entirely (42 per cent) or mainly (5 per cent) through postponed retirement. Indeed, the average pension wealth effect the average weight the panel assigns to changes in the length of working lives following changes in pension wealth is nearly identical for increases (0.54) and decreases (0.53) in pension wealth. The symmetry may be slightly surprising and contrary to the prevailing impression that most people cannot wait to stop working. 13 The pension wealth effect may nevertheless differ substantially and predictably across individuals. We therefore confront the individual pension wealth effects to individual and household background characteristics using logistic regression analysis. Since the individual pension wealth effects are clustered at 0, 0.5, and 1 we assign individuals to three groups, distinguishing individuals who put the largest weight on changing the benefit level, those who apply equal weights, and those who put the largest weight on changes in the length of the working life. 14 Our model selection procedure is as described before in footnote 11 and the parsimonious regression results are reported in table While we have attempted to explain the gender effect through interactions with household head, education, children and partner, none of them proved successful. 13 Where social security is concerned, De Nederlandsche Bank (2004) concludes that the Dutch prefer a lower benefit to a postponed benefit. 14 Alternatively, we distinguish individuals who put full weight on changing the benefit level, those who put full weight on changes in the length of the working life, and those who accommodate changes in wealth in some way through both a change in the length of the working life and a change in the pension benefit level. The results are broadly the same. 15

19 Table 6 The determinants of pension wealth effects PANEL A Following a hypothetical increase in pension wealth Largest weight on Equal weights higher benefit Age -3.5 (1.8) Squared age ( 100) 4.4 (2.0) Education 1.7 (1.9) Income -5.7 (2.4) Wealth -6.7 (2.9) Impatience 1.8 (1.3) Risk tolerance 1.6 (1.1) # observations Pseudo-R PANEL B Following a hypothetical decrease in pension wealth Largest weight on Equal weights lower benefit Age -0.5 (2.0) -0.1 (1.7) Squared age ( 100) Education 1.9 (2.8) Wealth 6.8 (3.0) Largest weight on early retirement 3.4 (1.6) -4.1 (1.7) 6.6 (2.6) 5.0 (2.1) Largest weight on postponed retirement -3.0 (1.5) 4.3 (1.9) -8.5 (3.7) # observations Pseudo-R Notes: Multinomial logistic regression results. Dependent variables are defined by the weights given by respondents adjustment of length of working life relative to adjustment in pension benefit level following a hypothetical change in pension wealth (also see table 5). Specifically, largest weight on higher benefit indicates the respondents who attached up to 0.5 weight on adjustment of the length of the working life, equal weight the respondents who attached precisely a 0.5 weight, and largest weight on early retirement the respondents who attached a weight higher than 0.5. Dependent variables in panel B are defined similarly. Alternative definitions have been considered (treating respondents with weights that are unequal to 0, 0.5, and 1 differently, excluding respondents with weights of 0.5 or unequal to either 0 or 1 entirely) but this does not materially affect the presented results. Marginal effects (in percentage points) of changes in explanatory variables are reported in the table. For 0-1 dummies, the marginal effects refer to discrete changes in the dummy variables from 0 to 1, for the other independent variables the sample average is used as the reference value. Absolute t-values are shown in parentheses. Reported specifications result from a selection among the variables age, squared age, household head, education, partner, children, gender, income, wealth, pension knowledge, impatience, risk tolerance, physical strain, mental strain (all defined as before). See the main text for details on the variable selection process. Panel A of the table presents the results for the case of a hypothetical increase in pension wealth. Standing out among the results is the difference in wealth effects between the more and the less affluent respondents. Specifically, more affluent respondents accommodate a larger share of an increase in pension wealth through early retirement, whereas less affluent respondents accommodate through a higher pension benefit level. Similar patterns emerge when we look at what drives the way in which respondents indicate that they accommodate a decrease in pension wealth (panel B of the table). More affluent respondents are more likely to accommodate a decrease in pension wealth through a lower pension benefit level, while less affluent respondents tend to accommodate it more strongly through postponed retirement. These findings reiterate the role of financial feasibility in the decision to retire 16

20 earlier or later than planned. 15 Another noteworthy result is the difference in the pension wealth effect between younger and older respondents. Typically, older respondents tend to accommodate a larger part of a wealth increase through early retirement, but also a slightly larger part of a wealth decrease through postponed retirement. Through the wealth effect therefore, older workers can only be induced to extend their working lives via politically unattractive cuts in pension entitlements (e.g. limited indexation, specific tax measures, etc.). Table 7 Indirectly stated pension adjustment and pension wealth effects PANEL A Pension benefit levels by hypothetical pension schemes and retirement age Scheme Retirement age < > 65 A B C D E F < 55 < 51 < 48 < 52 < 48 < > 88 > 95 > 102 > 83 > 88 > 93 Mean 95% C.I. # t-test of equal retirement age w.r.t. Obs. IERA A B C D E PANEL B Retirement age by scheme: all respondents Expected: Initial expected retirement age (IERA) 62.5 ( ) 1099 Retirement age in scheme A scheme B scheme C scheme D scheme E scheme F ( ) ( ) ( ) ( ) ( ) ( ) *** 7.4 *** 8.1 *** 11.2 *** 11.3 *** 12.2 *** *** 4.5 *** 4.5 *** *** 4.4 *** 5.5 *** 3.8 *** 3.7 *** *** Notes: (Expected) pension benefit levels are expressed as percentages of final pay. Mean values use observations weighted by income and home ownership. Initial expected retirement age (IERA) is truncated at age 58 and 66 to facilitate comparison with stated retirement age in schemes A through F (for that reason alone it contrasts with the average expected retirement age of 62.1 mentioned before in the main text). T-tests of equal retirement age assume equal variances. Statistical significance at the 10%, 5% and 1% error level is indicated by *, **, and ***, respectively Robustness: indirectly stated preferences To check the robustness of the results we have obtained so far, we analyze retirement preferences in six different pension schemes (see panel A of table 7). Schemes A, B and C are calibrated such that upon retirement at the age of 62 the pension benefit replaces 70 per cent of final earnings. The schemes differ in terms of the costs (gains) of early (delayed) retirement. These costs (gains) are the lowest in scheme A, where rounded 6% of the pension benefit is relinquished (gained) by retiring one year earlier 15 This conclusion is further bolstered by the (unreported) observation that less affluent respondents have a lower self-assessed life expectancy than more affluent respondents. The result that less affluent respondents on average retire later (table 2), are less inclined to capitalize on increases in pension wealth through early retirement and are 17

21 (later). 16 They are the highest in scheme C where rounded 10% of the pension benefit is relinquished (gained) by retiring one year earlier (later). In scheme B these costs (gains) are 8%. Schemes D, E and F have the same pension adjustment structure, but differ from A, B and C in that they are calibrated around a standard retirement age that is one year higher, viz. 63. The panel members are asked to state their desired retirement age for every one of these schemes. The results are reported in panel B of table 7. The resulting indirect estimates of the pension wealth and the pension adjustment effects are discussed in turn. Compared with schemes A through C, schemes D through F differ only in the assumed level of pension wealth. Hence comparing the average retirement age in scheme D with that in scheme A (see panel B of the table) gives an indirect estimate of the wealth effect that we may compare with the direct estimate obtained before in section 4.2. There, we concluded that a reduction in pension wealth to the effect of raising the standard retirement age by 1 year resulted in postponed retirement by about half a year. Since the difference in pension wealth levels between schemes A and D is exactly to the effect of raising the standard retirement age by one year, we would expect the panel to retire about half a year later in scheme D than in scheme A. In fact the average retirement age in scheme A is 63.2 while in scheme D it is Hence our indirect estimate of the pension wealth effect is 0.4 years and reasonably similar to our direct estimate of about half a year. 17 Indirect estimates of the pension wealth effect at different levels of pension adjustment obtained by comparing the average retirement age in schemes E and F with those in schemes B and C confirm this result. Hence we conclude that the pension wealth effect is robust to indirect measurement in both its sign and its size. 18 Indirect estimates of the pension adjustment effect can be derived from the difference in the average retirement age in schemes A through C, which differ in pension adjustment while having the same standard retirement age. Compared to scheme A the average retirement age in schemes B and C is higher by 0.0 and 0.1 years only (see panel B of the table). This suggests that there is not much of a pension adjustment effect. 19 Note that relative to scheme A continuing to work in the 64th year in scheme C yields an addition to the pension benefit of 3 percentage points. Hence when compared with the direct estimate of the pension adjustment effect where respondents would change the length of their working lives by 0.4 to 0.7 years in exchange for a 5 percentage point increase in pension benefits the more inclined to postpone retirement to accommodate decreases in pension wealth (table 6) is therefore more likely to indicate financial constraints than a particularly zealous attitude towards work. 16 In scheme A, continuing to work at 62 yields a 4 percentage point increase in the pension benefit, from 70% to 74%; this is an increase of 5.7%. 17 In addition, the t-test statistic of equal retirement in schemes A and D is soundly rejected at conventional confidence levels. 18 Alternatively, financially illiterate respondents may not have detected the pension incentives embedded in schemes A through F at all and simply ticked the entry corresponding to a 70% replacement rate, thus feeding a spuriously large (small) indirect estimate of the pension wealth (adjustment) effect. Unreported results fail to reveal a negative relation between self-assessed financial literacy and the indirectly estimated pension wealth effect. 19 Note that the pension adjustment incentives contained in scheme C overcompensate respondents for their additional years of work; the Nederlandsche Bank s pension model PALMNET provides for an actuarially neutral remuneration for an extra year s work of roughly 8% (Van Rooij et al., 2004), while in scheme C (and F, too) the pension benefit is raised by rounded 10% for every extra year worked. According to this measure, schemes B and E contain incentives which are actuarially neutral, while schemes A and D compensate contributors insufficiently, in actuarial terms, for the extra years worked. 18

22 indirectly measured pension adjustment effect is substantially lower. This result is confirmed by the comparison of the average retirement age in schemes D through F. Hence we conclude that the pension adjustment effect is robust to indirect measurement in its sign, but not its size. 4.4 Reconciling initial retirement expectations and indirectly stated preferences A striking result in table 7 is that relative to the initial expected retirement age (IERA), retirement is on average 0.7 to 0.8 years later in schemes A through C. On the one hand, this gap may suggest that the pension adjustment effect is in fact considerably larger than we concluded from the mutual comparison of schemes A-C and D-F, respectively, inter alia moderating our conclusion that it would not be robust in size to indirect measurement. This reading implicitly assumes that the IERA embodies the same pension arrangements as do schemes A through C in terms of the normal retirement age (62 years) and replacement rate (70% of final earnings), but without any pension adjustment to changes in the timing of retirement. On the other hand, however, the gap may signal that the IERA is underlain by pension arrangements that do not match these implicit assumptions. In the remainder of this section we will therefore look into these assumptions in more detail. In advance of the analysis we will argue that we cannot be sure that the IERA is implicitly underlain by the assumptions needed to make a meaningful comparison with retirement in schemes A through C. More specifically, the gap between the IERA and retirement in schemes A through C probably derives to a considerable extent from a subset of respondents in the DHS panel with relatively poor knowledge of its own actual pension arrangements. Embedded in schemes A through C is a normal retirement age of 62 years where upon in retirement the pension benefit replaces seventy per cent of final earnings. 20 The comparison of retirement in schemes A through C with the IERA thus assumes that initial retirement expectations incorporate similar expectations regarding the pension replacement rate. In case the panel initially expected higher pension replacement rates it may have felt compelled to postpone retirement once confronted with schemes A through C, implying that the gap reflects pension wealth effects in addition to any pension adjustment effects. 21 We can check this explanation using the subset of respondents for whom we have an expected pension replacement rate (as a percentage of final pay) in the regular consultation of the DHS in Panel A of table 8 contains a summary of all relevant statistics for this selection of respondents. From Panel A it follows, first of all, that this selection of respondents anticipates a pension benefit of just below seventy per cent of final pay. For this selection of respondents, therefore, our implicitly assumed replacement rate seems adequate. Second, the gap between the IERA and retirement in schemes A through C has narrowed to 0.4 to 0.5 years. This confirms the suspicion that a considerable 20 Although most schemes have now been converted to average-pay schemes, in the survey the pension benefit was nevertheless expressed as a percentage of final pay. We suspect that this formulation is largely in line with the way respondents still think about their pension benefits. 21 On the one hand, Van Soest et al. (2006) remark that the Dutch are quite positive about their prospects to retire early at generous replacement rates. On the other hand, our description of schemes A through F in the survey is not specific about whether or not third pillar pension provisions are included in the presented replacement rates or not (see appendix A). They are not intended to, but the panel may have responded to our questions as if they did. 22 We have this information for only 29% of the respondents; the remainder did not know or was not consulted in the DHS

23 part of the initial gap was driven by the fact that our assumed pension replacement rates did not match the (implicit) expectations that part of the panel may have had about them. The remaining gap nevertheless remains well above our earlier indirect estimate of the pension adjustment effect, so that this cannot be the entire story. Table 8 Robustness checks on indirectly stated pension adjustment and pension wealth effects Mean 95% C.I. # t-test of equal retirement age w.r.t. Obs. IERA A B C D E PANEL A Retirement age by scheme: reported expected pension benefit level Expected: IERA Pension benefit level ( ) ( ) Retirement age in scheme A scheme B scheme C scheme D scheme E scheme F ( ) ( ) ( ) ( ) ( ) ( ) ** 2.4 ** 2.4 *** 5.3 *** 5.9 *** 6.4 *** *** 4.0 *** 4.6 *** *** 3.9 *** 4.4 *** PANEL B Retirement age by scheme: known adjustment factors in pension arrangements Expected : IERA Pension benefit level Upward adjustment 1 Downward adjustment ( ) ( ) ( ) ( ) Retirement age in scheme A scheme B scheme C scheme D scheme E scheme F ( ) ( ) ( ) ( ) ( ) ( ) ** 2.2 ** 2.6 ** * 2.0 ** 2.5 ** * 2.2 ** 2.7 *** 3.2 *** *** * PANEL C Retirement age by scheme: no knowledge of adjustment factors in pension arrangements Expected : IERA Pension benefit level Upward adjustment 1 Downward adjustment n/a n/a ( ) ( ) Retirement age in scheme A scheme B scheme C scheme D scheme E scheme F ( ) ( ) ( ) ( ) ( ) ( ) * 1.6 * 1.8 * Notes: (Expected) pension benefit levels are expressed as percentages of final pay. Mean values use observations weighted by income and home ownership. Initial expected retirement age (IERA) is truncated at age 58 and 66 to facilitate comparison with stated retirement age in schemes A through F (for that reason alone it contrasts with the average expected retirement age of 62.1 mentioned before in the main text). T-tests of equal retirement age assume equal variances. Statistical significance at the 10%, 5% and 1% error level is indicated by *, **, and ***, respectively. 1 Median values and binomial confidence intervals reported. Upward (downward) adjustment refers to the percentage increase (decrease) in pension benefits when retiring 1 year later (earlier). 20

24 Another potentially contaminating factor is that for some respondents schemes A through F may contain pension adjustment factors that are already embedded in their own actual arrangements, while for others they are not. For example, respondents whose IERA already incorporated adjustment factors embedded in their actual arrangements may end up not postponing retirement in our hypothetical schemes at all, while they may nevertheless be sensitive to pension incentives. Alternatively, respondents may not at all have been aware of the pension adjustment factors embedded in their actual pension arrangements. The analysis is further narrowed down to these groups of respondents in panels B and C of table 8. Since the progressively stricter criteria in terms of knowledge of actual pension arrangements increasingly narrow down the number of relevant observations, our conclusions from this exercise can only be tentative. 23 Panel B looks into the stated retirement age in schemes A through F for respondents who indicate that their pension arrangements contain pension adjustment factors when retirement is advanced or postponed and additionally, have knowledge of the size of the respective pension adjustment factors. For this subset of the panel, we find that the expected pension replacement rate is about seventy per cent, while upward and downward pension adjustment factors are within the range of adjustment factors embedded in our hypothetical schemes A through F. Stated retirement in schemes A through C reflects a weak pension adjustment effect, while retirement in these schemes is only 0.1 to 0.3 years later than initial retirement expectations. When the underlying assumptions are appropriate, therefore, the comparison of the IERA with retirement in schemes A through C agrees with a very weak indirect estimate of the pension adjustment effect. Panel C looks into retirement expectations of those respondents who do not know whether pension adjustment factors are already embedded in their actual pension arrangements or not. For this subset of the panel, too, the expected pension replacement rate is about seventy per cent, while stated retirement in schemes A through C confirms the weak indirect estimate of the pension adjustment effect. Most importantly, however, the gap between the IERA and retirement in schemes A through C is 0.3 years. To the extent that this gap is at the high end of the interval that resulted from panel B it gives some further credence to our suspicion that the initial gap estimated at 0.7 to 0.8 years in panel B of table 7 originates to a considerable extent from a subset of the panel that has relatively poor knowledge of its own actual pension arrangements. Lastly, note that throughout table 8 an indirect estimate of the pension wealth effect of 0.3 to 0.5 years is confirmed. 5. Conclusions This paper investigates the relationship between retirement expectations and hypothetical changes in incentives as derived from an extensive survey questionnaire fielded in a representative sample of Dutch 23 When we alternatively narrow down the analysis to those respondents who indicate that their actual pension benefits are not adjusted following either earlier or postponed retirement, the resulting number of relevant 21

25 households. The experimental setup of the analysis enables us to circumvent the problem of unobserved heterogeneity in tastes for retirement. Moreover, this setup allows for a distinction between pension wealth and pension adjustment effects in retirement considerations. The main results of our analysis are the following. First, we find that people respond to pension incentives. Both changes in pension adjustment and changes in pension wealth affect the way people feel about the timing of retirement. Second, regarding the pension adjustment effect we must stress the sensitivity of its size to the framing of the hypothetical change in pension incentives. Specifically, our direct estimate of the pension adjustment effect suggests that an accrual of wealth of five percentage points elicits an average postponement of retirement by about half a year. Measured indirectly, however, the pension adjustment effect is very small and indistinguishable from zero in a statistical sense. In addition to this, we observe systematic cross-sectional variation in the pension adjustment effect; members of more affluent households tend to be less sensitive to pension adjustment in their retirement considerations. Third, we find a negative pens ion wealth effect that is robust in both sign and size to deliberate variations in the way that the hypothetical change in pension incentives is framed. On average retrenchments of pension arrangements to the effect of raising the standard retirement age by 1 year induce people to postpone retirement by about 0.4 to 0.5 years. The stated wealth effect nevertheless differs systematically between the more and less affluent members of the panel. Relative to members of less affluent households, members of more affluent households seem more inclined to capitalize on increases in pension wealth through earlier retirement while they more readily accept lower pension benefits in case of a decrease in pension wealth. Increases in pension wealth therefore disproportionately trigger early retirement by members of more affluent households, whereas decreases in pension wealth disproportionately compel members of less affluent households to postpone retirement. These results suggest that reforms in the pension system may have sizeable effects on the participation of older workers. Of course, in an ageing society such as the Dutch society, strengthening the financial incentives in pension arrangements may be used as a means to discourage early retirement, raise the effective retirement age, and along those lines the labor participation of older workers. Let us briefly discuss two recent (proposed) reforms to the Dutch pension system to illustrate the implications of our findings. The first illustration concerns the implementation of and increases in the accrual of pension wealth within occupational pension schemes for additional years of work and the corresponding diminution of pension wealth following early retirement. These reforms were inspired by the recognition of perverse labor supply effects embedded in previous pension schemes. However, the labor participation effects of such pension reforms strongly depend on the pension adjustment effect while the results in this paper indicate that the pension adjustment effect may be limited. The second illustration concerns contemporaneous debates in politics revolving around the affordability of social security in an aging society. A frequently proposed reform to ensure sustainability is to broaden the tax base so that the growing numbers of retired contribute to the funding of social security payments. In a stylized setting, this implies a retrenchment in terms of net social security payments, which corresponds to a reduction in observations is too low to draw any meaningful conclusions. 22

26 expected lifetime pension wealth. Given the clearly negative estimate of the pension wealth effect in this paper such a reform may illicit substantial increases in the effective retirement age. Since the results in this paper strongly suggest that decreases in pension wealth disproportionately compel less affluent workers to postpone retirement, a trade-off between desirable labor participation effects and unintended distributional effects looms. Appendix A. Primary additional survey questions used This appendix contains the questions that are most relevant to this paper. The questions were fielded on January 8, 2005 among the DHS panel members of age 18 and above who were at the time working for pay, searching for work, studying, or retired (although none of the questions in this appendix were posed to retired panel members). 1a) Would you value an increase in your pension benefit by 5 percentage points (for example from 70 percent of final pay to 75 percent of final pay)? [1] Yes, I would be willing to extend my working life. [2] Yes, I would be willing to pay a higher pension contribution. [3] Yes, I would be willing to partly extend my working life and partly to pay a higher pension contribution. [4] No, I am satisfied with my current pension benefit level. If the answer to question 1a) was either [1] or [3]: 1b) By how many years would you be willing to extend your working life? [1] By half a year [2] By a year [3] By a year and a half [4] By two years [5] By two and a half years [6] By three years [7] By more than three years If the answer to question 1a) was either [2] or [3]: 1c) How much additional contribution, as a percentage of your wages, would you be willing to pay? [1] 1% of my gross wages [2] 2% of my gross wages [3] 3% of my gross wages [4] 4% of my gross wages [5] 5% of my gross wages 23

27 [6] 6% of my gross wages [7] More than 6% of my gross wages 2a) Can you imagine an adequate compensation for a decrease in your pension benefit level by 5 percentage points (for example from 70 percent of final pay to 65 percent of final pay)? [1] Yes, I should be able to retire earlier. [2] Yes, I should be paying a lower pension contribution. [3] Yes, I should be able to partly retire earlier and partly pay a lower pension contribution. [4] No, I cannot imagine an adequate compensation in terms of early retirement and/or a lower pension contribution. If the answer to question 2a) was either [1] or [3]: 2b) By how many years should you be able to retire early? [1] By half a year [2] By a year [3] By a year and a half [4] By two years [5] By two and a half years [6] By three years [7] By more than three years If the answer to question 2a) was either [2] or [3]: 2c) How much should your contribution, as a percentage of your wages, be reduced? [1] 1% of my gross wages [2] 2% of my gross wages [3] 3% of my gross wages [4] 4% of my gross wages [5] 5% of my gross wages [6] 6% of my gross wages [7] More than 6% of my gross wages 3-4) The accrual rate in a pension scheme determines how many years its takes to build up a full pension. A pension scheme that pays a benefit of 70 percent of final pay, for example, may be built up in 35 contribution years when the accrual rate is 2.00 (35 times 2.00 equals 70%). Should the accrual rate be 1.75 instead, it would take 40 contribution years to build up the same pension benefit. Questions 4 and 5 below assume that the contribution payments remain unchanged. 24

28 3a) Suppose that your pension scheme applies an accrual rate of Should this accrual rate be increased from 1.75 to 2.00 you have a choice. If you do not alter your retirement age, then you receive a higher pension benefit when you retire. If, alternatively, you choose not to alter the pension benefit level, you would be able to retire earlier. What would your choice be in the circumstances? [1] I choose for earlier retirement at an unchanged pension benefit level. [2] I choose for a higher pension benefit level at an unchanged retirement age. [3] I choose a combination of earlier retirement and a higher pension benefit level. If the answer to question 3a) was [3]: 3b) Could you indicate, on a scale from 0 to 100, to what extent you would accommodate this change to your pension scheme by means of a higher pension benefit level? 0 indicates that you do not wish to change your pension benefit level; you instead exploit fully the option to retire earlier. 100 indicates that you only wish to change your pension benefit level; you do not change you retirement age in this case. 4a) Suppose now that your pension scheme applies an accrual rate of Should this accrual rate be decreased from 2.00 to 1.75 you once again have a choice. If you do not alter your retirement age, then you receive a lower pension benefit when you retire. If, alternatively, you choose not to alter the pension benefit level, you would have to retire later. What would your choice be in the circumstances? [1] I choose for later retirement at an unchanged pension benefit level. [2] I choose for a lower pension benefit level at an unchanged retirement age. [3] I choose a combination of later retirement and a lower pension benefit level. If the answer to question 4a) was [3]: 4b) Could you indicate, on a scale from 0 to 100, to what extent you would accommodate this change to your pension scheme by means of a lower pension benefit level? 0 indicates that you do not wish to change your pension benefit level; you instead retire later. 100 indicates that you only wish to change your pension benefit level; you do not change you retirement age in this case. 5a-f) We will now present to you six pension schemes. Each scheme presents the pension benefit level as a percentage of final pay. The pension benefit level in each case depends on the retirement age that you select yourself. If you retire early your pension benefits level will be relatively low, whereas if you retire late your pension benefit level will be relatively high. For each of the presented pension schemes we ask you at what age you would like to retire. When answering the question, assume that from the moment of retirement onwards you will receive the pension benefit for the rest of your life. Additionally assume that in case you are currently working part time, you will continue to do so up until your retirement while in case you are currently working full time, you will also continue to do so up until your retirement. 25

29 Pension scheme 5a Retirement age < > 65 Pension benefit level (% of final pay) < > 88 5a) Based on pension scheme 5a above, at what age would you like to retire? [1] Before age 59 [2] At age 59 [3] At age 60 [4] At age 61 [5] At age 62 [6] At age 63 [7] At age 64 [8] At age 65 [9] After age 65 Questions 5b) through 5f) are the same, save for the presented pension schemes: Pension scheme 5b Retirement age < > 65 Pension benefit level (% of final pay) < > 95 Pension scheme 5c Retirement age < > 65 Pension benefit level (% of final pay) < > 102 Pension scheme 5d Retirement age < > 65 Pension benefit level (% of final pay) < > 83 Pension scheme 5e Retirement age < > 65 Pension benefit level (% of final pay) < > 88 Pension scheme 5f Retirement age < > 65 Pension benefit level (% of final pay) < > 93 Appendix B. Construction of pension knowledge, impatience and risk tolerance This appendix alternately explains the construction of pension knowledge, impatience, and risk tolerance. In order to maximize the precision of our proxies for these three characteristics, we do not 26

30 restrict the samples used in this appendix to include only respondents who also filled out the additional questionnaire to which the main text refers. B.1. Pension knowledge Individual pension knowledge can be gleaned by means of four questions about specific aspects of the respondents own pension arrangements, see box B1 (cf. Van Els et al., 2004). We assume that a respondent has knowledge of an aspect of his/her pension arrangements if he/she does not answer the corresponding question with don t know. With the aid of factor analysis, the knowledge of these four aspects is subsequently summarized in a single figure showing how knowledgeable the respondent is about his/her own pension arrangements. Analysis shows that knowledge of all four aspects shares a single common, unobserved factor, which loads positively on the knowledge of each of the four individual aspects. We therefore label this factor pension knowledge. Box B1 Knowledge of four aspects of pension arrangements 1. What type of pension arrangement do you have? 1 final pay 2 average pay 3 defined contribution 4 other - don t know 2. Pension arrangements may contain a provis ion that the level of pension entitlements, pensions already being paid, or both, will be adjusted. Such an adjustment may be on the basis of a price or wage index, or a combination of both. This is called index-linking. Is your (future) pension index-linked? 1 yes 2 no - don t know 3. What pension entitlements have you accumulated as per 1 January 2002, according to the pension statement, at the pension fund of your current/last employer? These entitlements are the amount paid to you annually, if you had stopped working from then on. Note: The answer to this question should be based on your current marital status. No allowance should be made for your entitlement to a state old age pension. amount in euro - don t know 4. What percentage of your final net pay do you expect your net pension (including the state old age pension) to be? (For those who have already retired early, this question refers to final pay before early retirement.) percentage of final net pay before retirement - don t know # answers We perform the same routine using only questions 1, 2 and 4, thereby raising the number of useful observations by over one third. Here, too, factor analysis reveals a single, unobserved, common factor in the answers to these three questions. This factor also displays positive loading on the knowledge of each 27

31 of the three individual aspects of pension arrangements 24 and correlates strongly with the pension knowledge factor based on all four individual knowledge variables ( ρ = 0.93 ). This factor, which we call pension knowledge, can be computed for 1,247 respondents, has an average value of 0 and varies between a minimum of and a maximum of 0.83 with a standard deviation of Tables B1 and B2 show the correlation between a set of personal and household characteristics and pension knowledge and its underlying components, respectively. The results are consistent with those in Van Els et al. (2004). Table B1 The determinants of pension knowledge Independent variable Estimated parameters Gender 0.28 (5.0) Age 0.03 (8.7) Education 0.05 (2.2) Job 0.15 (0.9) Pension 0.10 (0.6) Partner (1.5) Income 0.13 (4.1) Home ownership 0.13 (2.0) Share holdings 0.12 (1.9) Constant term (8.9) Number of observations 645 Adj-R Notes: OLS estimation results. The dependent variable is pension knowledge. Absolute t-values are shown in parentheses. Table B2 Knowledge of individual aspects of pension arrangements (outcomes and marginal effects in percentage points) how the indexation pension pension is entitlements built up 28 the ultimate pension benefit level % of respondents aware of Gender Age Education Job Pension Partner Income Home ownership Share holdings 11.9 (3.4) 1.1 (6.3) 2.9 (2.6) 5.7 (0.6) -3.0 (0.3) -3.1 (0.9) 4.9 (3.0) 3.9 (1.1) 4.2 (1.2) 17.9 (4.3) 1.3 (6.1) 1.5 (1.1) 11.4 (0.9) 21.4 (2.1) -2.4 (0.5) 9.2 (4.2) 10.5 (2.2) 8.3 (2.0) 0.8 (0.9) 1.5 (5.1) 4.5 (2.5) -2.0 (0.1) 13.8 (0.8) -4.0 (0.6) 6.1 (2.0) 12.5 (2.1) 0.0 (0.0) 14.9 (3.6) 0.8 (3.6) 4.0 (2.7) 2.9 (0.3) 2.9 (0.2) -4.8 (1.0) 5.1 (2.2) 8.6 (1.8) 9.7 (2.1) Number of observations Pseudo-R Notes: Respondents are aware of aspects of their pension arrangements if they answer anything but don t know to questions concerning these arrangements (see Box B1). Mean awareness values use observations weighted by income and home ownership. Marginal effects of a change in the independent variable on the chance of a positive answer are reported, using logit analysis. For 0-1 dummies, the marginal effects refer to discrete changes in the dummy variables from 0 to 1, for the other independent variables the sample average is used as the reference value. Absolute t-values are shown in parentheses. 24 The likelihood ratio statistic indicates that the single-factor model looks similar in a statistical sense to the fully saturated model.

32 Box B2 Indicators of impatience Statements about the future Please indicate, on a scale of 1 (indicating total disagreement) to 7 (indicating total agreement), to what extent you agree with the following statements. 1. I think about how things could be in the future, and try to influence them in my daily life. 2. I often think about matters which will only begin to have consequences in several years time. 3. I only think about current matters, and assume that things will work out in the future. 4. I always think only about the immediate consequences (in terms of days or weeks) of whatever I am doing. 5. Convenience is a major consideration in my decisions and actions. 6. I am prepared to sacrifice my current well-being for certain results in the future. 7. I believe it is important to take warnings about the negative consequences of my actions seriously, even if these negative consequences were to make themselves felt only in the distant future. 8. I believe it is more important to think about matters which will have major consequences in the future, than about matters which have immediate but less important consequences. 9. I generally ignore warnings about future problems because I think they will solve themselves. 10. I consider it unnecessary to make sacrifices for future matters because they can always be sorted out later. 11. I react only to immediate problems, on the assumption that I will deal with later problems when they present themselves. 12. I get clear results in my daily work; this is more important to me than working with vague results. Questions on spending behavior 1. Some people spend all the money they receive immediately. Others save for a nest egg. Please indicate on a scale of 1 to 7 what you do with the money left after you have paid for food, the rent, and other daily necessities, with 1 indicating that you intend to spend your money immediately and 7 that you intend to save as much as possible. 2. People use different time horizons when deciding how much of the ir income to spend or to save. Which of the time horizons set out below is MOST IMPORTANT to you when making spending and saving decisions? 1 the next couple of months 2 the next year 3 the next few years 4 the next five to ten years 5 more than ten years ahead B.2. Impatience Individual impatience can be gleaned from the degree to which respondents agree with twelve statements about the future, in combination with the answers to two questions about their spending habits (see box B2). Someone is impatient when he/she is more concerned with issues with which he/she will be confronted within the near future. An impatient person will therefore be more inclined to agree with statements 3, 4, 5, 9, 10, 11 and 12 and less with the other statements. An impatient person will also score low on the two questions about spending habits. Initial factor analysis results show a somewhat mixed picture regarding the optimal number of factors that should be retained. Depending on whether we look at scree plots or use the Akaike or Schwartz information criterion, the optimal number of retained factors is 2, 5, or 7, respectively. The 7- factor solution is discarded as it presents a Heywood solution, i.e. boundary values of uniqueness are not respected. We therefore continue exploring the results of the 2- and 5-factor solutions. What immediately 29

33 catches the eye is that in both cases we obtain a very strong first factor, with an Eigenvalue more than twice that of the second retained factor. We therefore also take on board a single-factor solution. Table B3 presents the rotated loadings on the first factor in analyses that retain 1, 2, and 5 factors, respectively. Regardless of the number of additional factors retained, the factor loadings of the variables on impatience always have the expected signs. Also, the reported loadings in the 2- and 5- factor solutions (columns 2 and 3 of table B3) differ little. Indeed, these two measures of impatience exhibit a strong correlation of Table B3 Rotated factor loadings on impatience (factors have been rotated using orthogonal varimax) # Factors retained Statement 1 Statement 2 Statement 3 Statement 4 Statement 5 Statement 6 Statement 7 Statement 8 Statement 9 Statement 10 Statement 11 Statement 12 Question 1 Question The loadings in column 1 do differ from those in the other two columns. 25 Specifically, the negative loadings tend to be larger in column 1 than in columns 2 and 3. Put differently, while the measures of impatience presented in columns 2 and 3 seem particularly good at identifying the really impatient individuals, the factor in column 1 seems more effective in covering the entire range of patience from very impatient to very patient individuals. This is not to say that patient individuals are disregarded in the analysis that produced the results in columns 2 and 3, though. Unreported results show that in both analyses the second-strongest factors seem to identify specifically patient individuals. While insightful when the focus of the analysis is on the respective determinants of patience and impatience, the analysis at hand favors a single factor that captures both strong patience and strong impatience. We therefore apply the loadings in column 1 to construct impatience. Impatience can be determined for 2,083 respondents; it has an average value of 0 and varies between a minimum of and a maximum of 2.78 with a standard deviation of This also reflects in the correlation of the resulting factor with the measures of impatience based on columns 2 and 3, which is 0.88 and 0.85, respectively. 30

34 Box B3 Indicators of risk tolerance Statements about saving and risk -taking Please indicate, on a scale of 1 (indicating total disagreement) to 7 (indicating total agreement), to what extent you agree with the following statements. 1. I prefer safe investments and a guaranteed return to taking risks in the hope of achieving the highest return. 2. I shall never invest in shares because I consider the risk too great. 3. If I believe an investment will be profitable, I am prepared to borrow to make that investment. 4. I want my investments to be sound. 5. I am increasingly convinced that I need to take more financial risk if I wish to improve my financial position. 6. I am prepared to risk loss, in order to make money. Characteristics of investment behavior Investing in shares: 1 = invests in shares or investment funds; 0 = does not invest in shares or investment funds. Investing in derivatives: 1 = owns/writes put options, call options, falcons, warrants; 0 = does not own/write derivatives. Question1 Please describe, on a scale of 1 (I haven t exposed myself to any risk at all) to 5 (I have often exposed myself to substantial risks), the risks that you have exposed yourself to in terms of your investment choices in the recent past? Preferences as to pension arrangements and risk -taking Pension1 Which of the following statements applies to you most? 1. I prefer to pay less contribution for a pension which is expected to be equally high on average, but where the ultimate pension benefit may be higher or lower, as a result of the investment form chosen (pension assets invested largely in shares). (risk tolerant) 2. Don t know. (risk neutral) 3. I prefer to pay more contribution for a guaranteed pension (pension assets invested largely in bonds). (risk averse) Pension2 Imagine the following two pension arrangements: A: your pension is guaranteed to be 70% of final pay. B: your pension is: A chance of 1 in 10-50% of final pay A chance of 2 in 10-60% of final pay A chance of 4 in 10-70% of final pay A chance of 2 in 10-80% of final pay A chance of 1 in 10-90% of final pay Suppose the standard pension arrangement is B. How much extra contribution are you prepared to pay monthly for the security of arrangement A? 1. No extra contribution, preference for arrangement B. (risk tolerant) 2. Don t know. (risk neutral) % to over 2% of gross wages. (risk averse) 31

35 B.3. Risk tolerance The measure of risk tolerance determines what weight a respondent assigns to risk and return in a (financial) decision. The greater the risk tolerance, the more utility people derive from return and the less disutility they experience from the uncertainty of this return. The greater the risk aversion, however, the greater the disutility from uncertainty, and consequently a lower the return is accepted in exchange for a more certain return. We glean the individual measure of risk tolerance from the measure of agreement evinced by the respondents with six questions about saving and risk-taking, in combination with information about their investment behavior and preferences with regard to pension arrangements, see box B3. Table B4 reports the (orthogonal varimax) rotated factor loadings on three factors capturing risk tolerance. The reported factors differ in the domains of risk tolerance included and, consequently in the number of observations. The presented results all derive from factor analyses that retain only a single factor. Nevertheless, in each case we found indications that more than a single factor might be retained. These indications were explored and yield the following insights. Separate factors tend to identify individuals who in the statements on savings tend to relate to upside potential (score relatively high on statements 3, 5, and 6) and those who relate to downside risks (score relatively high on statements 1, 2, and 4). Adding stock investment information feeds a factor that captures individuals appetite for the stock market. Pension variables elicit a factor that captures risk preferences specific to the pension domain. If the focus of the analysis had been on risk preferences in different domains (e.g. Van Rooij et al., 2006), these results would certainly validate the retention of various factors capturing risk preferences. In the analysis at hand, however, a single index that captures overall risk tolerance is preferred. Table B4 Rotated factor loadings on risk tolerance (factors have been rotated using orthogonal varimax) Statement 1 Statement 2 Statement 3 Statement 4 Statement 5 Statement 6 Share holdings Holdings of derivatives Question 1 Pension1 Pension # observations correlation with

36 The risk tolerance factors presented in table B4 build on increasingly more variables, but also capture increasingly fewer observations. The table further shows that the loadings of included variables do not change too much when additional variables are taken on board. Moreover, the high correlations presented in the bottom lines of the table suggest that all three factors capture a very similar common factor. This allows us to select as the preferred measure of risk tolerance simply the factor that retains the largest number of observations, i.e. the one described by column Recall that the motivation to construct a measure of risk tolerance is to be able to control for risk tolerance when assessing the effects of impatience in our analysis. Specifically, if we do not control for risk tolerance, we may erroneously ascribe its effects to impatience as both highly impatient as well as highly risk intolerant individuals may favor short-term benefits over longer-term benefits. That is to say, we suspect that impatience and risk tolerance correlate negatively. Our suspicions in that regard are validated as risk tolerance and impatience correlate with a coefficient of References Alessie, R.J.M., Hochguertel, S., and A. van Soest (2002), Household portfolios in the Netherlands, in: L. Guiso, M. Haliasos and T. Japelli (eds.), Household portfolios, Cambridge (MA): MIT Press, Berkel, B., and A. Börsch-Supan (2003), Pension reform in Germany: the impact on retirement decisions, NBER Working Paper Series 9913, National Bureau of Economic Research, Cambridge (MA). Bernartzi, S., and R. Thaler (2002), How much is investor autonomy worth? Journal of Finance 57, Blake, D. (2004), The impact of wealth on consumption and retirement behaviour in the UK, Applied Financial Economics 14, Bloemen, H. (2006), The impact of wealth on job exit rates of elderly workers in the Netherlands, NETSPAR Discussion Paper , NETSPAR, Tilburg. Boskin, M.J. (1977), Social security and retirement decisions, Economic Enquiry 15, Boskin, M. J., and M. D. Hurd (1978), The effect of socia l security on early retirement, Journal of Public Economics 10, Chan, S., and A.H. Stevens (2004), Do changes in pension incentives affect retirement? A longitudinal study of subjective expectations, Journal of Public Economics 88, Choi, J.J., Laibson, D., Madrian, B.C., and A. Metrick (2003), Optimal defaults, AEA Papers and Proceedings 93, Coile, C., and J. Gruber (2000), Social security and retirement, NBER Working Paper Series 7830, National Bureau of Economic Research, Cambridge (MA). 26 At the same time, however, this measure of risk tolerance displays weaker correlation with its alternatives than do the alternatives amongst themselves. The addition of stock market information apparently augments the features of extracted risk tolerance. Indeed, unreported results show that risk tolerance defined by column 1 correlates with our stock market variables only about half as strongly as do the alternative measures of risk tolerance in columns 2 and 3. 33

37 Crawford, P. C., and D. M. Lilien (1981), Social security and the retirement decision, Quarterly Journal of Economics 46, De Nederlandsche Bank (2004), Financial behaviour of Dutch households, Quarterly Bulletin September 2004, Diamond, P.A. (2001), Social security reform with a focus on the Netherlands, De Economist 149, Disney, R., Meghir, C., and E. Whitehouse (1994), Retirement behaviour in Britain, Fiscal Studies 15, Duval, R. (2003), The retirement effects of old-age pension and early retirement schemes in OECD countries, OECD Economics Department Working Paper 370, Organisation for Economic Cooperation and Development, Paris. Euwals, R., Van Vuuren, D., and R. Wolthoff (2005), Early retirement in the Netherlands: Evidence from a policy reform, CPB Discussion Paper 52, CPB Netherlands Bureau for Economic Policy Analysis, The Hague. Gruber, J., and D. Wise (1997), Social security programs and retirement around the world, NBER Working Paper Series 6134, National Bureau of Economic Research, Cambridge (MA). Kahneman, D., and A. Tversky (1979), Prospect theory: an analysis of decision under risk, Econometrica 47, Kahneman, D., and A. Tversky (1983), Choices, Values, and Frames, American Psychologist 39, Mastrogiacomo, M., Alessie, R.J.M., and M. Lindeboom (2004), Retirement behaviour of Dutch elderly households: diversity in retirement patterns across different household types, Journal of Applied Econometrics 19, Meghir, C., and E. Whitehouse (1997), Labour market transitions and the retirement of men in the UK, Journal of Econometrics 79, Samwick, A. A. (1998), New evidence on pensions, social security, and the timing of retirement, Journal of Public Economics 70, Sheshinski, E. (1978), A model of social security and retirement decisions, Journal of Public Economics 10, Van Els, P.J.A., End, W.A. van den, and M.C.J. van Rooij (2004), Pensions and public opinion: a survey among Dutch households, De Economist 152, Van Soest, A., Kapteyn, A., and J. Zissimopoulos (2006), Using stated preferences data to analyze preferences for full and partial retirement, DNB Working Paper 081/2006, De Nederlandsche Bank, Amsterdam. Van Rooij, M.C.J., Kool, C.J.M., and H.M. Prast (2006), Risk-return preferences in the pension domain: are people able to choose? Journal of Public Economics, forthcoming. Van Rooij, M.C.J., Siegmann, A.H., and P.J.G. Vlaar (2004), PALMNET: a pension assets and liabilities model for the Netherlands, DNB Research Memorandum WO 760/2004, De Nederlandsche Bank, Amsterdam. 34

38 Previous DNB Working Papers in 2006 No. 81 No. 82 No. 83 No. 84 No. 85 No. 86 No. 87 No. 88 No. 89 No. 90 No. 91 No. 92 No. 93 No. 94 No. 95 No. 96 No. 97 No. 98 No. 99 No. 100 No. 101 No. 102 No. 103 No. 104 No. 105 No. 106 No. 107 No. 108 No. 109 No. 110 No. 111 No. 112 No. 113 No. 114 Arthur van Soest, Arie Kapteyn and Julie Zissimopoulos, Using Stated Preferences Data to Analyze Preferences for Full and Partial Retirement Dirk Broeders, Valuation of Conditional Pension Liabilities and Guarantees under Sponsor Vulnerability Dirk Brounen, Peter Neuteboom and Arjen van Dijkhuizen, House Prices and Affordability A First and Second Look Across Countries Edwin Lambregts and Daniël Ottens, The Roots of Banking Crises in Emerging Market Economies: a panel data approach Petra Geraats, Sylvester Eijffinger and Carin van der Cruijsen, Does Central Bank Transparency Reduce Interest Rates? Jacob Bikker and Peter Vlaar, Conditional indexation in defined benefit pension plans Allard Bruinshoofd and Clemens Kool, Non-linear target adjustment in corporate liquidity management: an endogenous thresholds approach Ralph de Haas, Monitoring Costs and Multinational-Bank Lending Vasso Ioannidou and Jan de Dreu, The Impact of Explicit Deposit Insurance on Market Discipline Robert Paul Berben, Kerstin Bernoth and Mauro Mastrogiacomo, Households Response to Wealth Changes: Do Gains or Losses make a Difference? Anne Sibert, Central Banking by Committee Alan Blinder, Monetary Policy by Committee: Why and How? Céline Christensen, Peter van Els and Maarten van Rooij, Dutch households perceptions of economic growth and inflation Ellen Meade, Dissents and Disagreement on the Fed s FOMC: Understanding Regional Affiliations and Limits to Transparency Jacob Bikker, Laura Spierdijk, Roy Hoevenaars and Pieter Jelle van der Sluis, Forecasting Market Impact Costs and Identifying Expensive Trades Cees Ullersma, Jan Marc Berk and Bryan Chapple, Money Rules Jan Willem van den End, Indicator and boundaries of financial stability Zsolt Darvas, Gábor Rappai and Zoltán Schepp, Uncovering Yield Parity: A New Insight into the UIP Puzzle through the Stationarity of Long Maturity Forward Rates Elisabeth Ledrut, A tale of the water-supplying plumber: intraday liquidity provision in payment systems Ard den Reijer, The Dutch business cycle: which indicators should we monitor? Ralph de Haas and Iman van Lelyveld, Internal Capital Markets and Lending by Multinational Bank Subsidiaries Leo de Haan and Elmer Sterken, Price Leadership in the Dutch Mortgage Market Kerstin Bernoth and Guntram Wolff, Fool the markets? Creative accounting, fiscal transparency and sovereign risk premia Hans de Heij, Public feed back for better banknote design Carry Mout, An Upper Bound of the sum of Risks: two Applications of Comonotonicity Lennard van Gelder and Ad Stokman, Regime transplants in GDP growth forecasting: A recipe for better predictions? Froukelien Wendt, Intraday Margining of Central Counterparties: EU Practice and a Theoretical Evaluation of Benefits and Costs Jan Kakes, Financial behaviour of Dutch pension funds: a disaggregated approach Jacob Bikker and Jan de Dreu, Pension fund efficiency: the impact of scale, governance and plan design Bastiaan Verhoef, Pricing-to-market, sectoral shocks and gains from monetary cooperation Leo de Haan, Hubert Schokker and Anastassia Tcherneva, What do current account reversals in OECD countries tell us about the US case? Ronald Bosman and Frans van Winden, Global Risk, Investment, and Emotions Harry Garretsen and Jolanda Peeters, Capital Mobility, Agglomeration and Corporate Tax Rates: Is the Race to the Bottom for Real? Jacob Bikker, Laura Spierdijk and Paul Finnie, Misspecification of the Panzar-Rosse Model: Assessing Competition in the Banking Industry

39