Tilburg University
Studies on labour supply, spending and saving before and after retirement
Bonekamp, Johan DOI: 10.26116/center-lis-2110 Publication date: 2021 Document Version
Publisher's PDF, also known as Version of record
Link to publication in Tilburg University Research Portal
Citation for published version (APA):
Bonekamp, J. (2021). Studies on labour supply, spending and saving before and after retirement. CentER, Center for Economic Research. https://doi.org/10.26116/center-lis-2110
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
Studies on labour supply
, spending and saving befor
e and after r
etir
ement
Johannes L.M. Bonekamp
Studies on labour supply, spending and
saving before and after retirement
Studies on labour supply, spending and
saving before and after retirement
P
ROEFSCHRIFTter verkrijging van de graad van doctor aan Tilburg University op
gezag van de rector magnificus, prof.dr. W.B.H.J. van de Donk,
in het openbaar te verdedigen ten overstaan van een door het
college voor promoties aangewezen commissie in de Aula van
de Universiteit op woensdag 23 juni 2021 om 10.00 uur
door
J
OHANNESL
EONARDUSM
ARIAB
ONEKAMPprof.dr. T.E. Nijman
LEDEN PROMOTIECOMMISSIE: prof.dr. E.K. Nyhus
prof.dr. S. Thorp
Preface
This thesis would not have been possible with the help, support and encour-agement of many people. To start with, I would like to thank my supervisors. Arthur, I am afraid that I will never be able to sufficiently express my gratitude. Your questions and suggestions have helped me to get back on track more than once. For some reason, the door of your office was always open and you always managed to squeeze in a meeting if necessary. None of the (sometimes far-fetched) ideas I pitched got rejected immediately. Instead, you inspired me to work out my existing ideas and encouraged me to continue asking new questions. I learned a lot from you, not only academically but also personally. Theo, in addition to your valuable comments and discussions, I would also specifically like to thank you for engaging me in the Dutch pension debate. Nowadays, I cannot read a single (news) article without asking myself who might benefit from it. I also greatly enjoyed working on the various projects you introduced me to, and learned a lot about how I could ensure that my research would not only be of interest from an academic point of view, but would be accessible and relevant to a broader audience as well.
I would also like to thank Bram, Marike, Hazel and Ralph for the pleasant collaborations on some of the projects that make up this thesis. Special thanks go to Jennifer and, even though not being a co-author, Lieven. Your hospitality made visiting Sydney one of the highlights of these past five years.
thesis and providing useful feedback. The individual chapters in this thesis greatly benefitted from your constructive comments.
My thanks also go to Netspar (Network for Studies on Pensions, Aging and Retirement), whose funding enabled some of the data collection for this thesis. Furthermore, the thesis benefitted from the feedback we received at the many Netspar conferences and workshops where we were given the chance to present our research.
I am also grateful to Sylvia, Wilma and Kristel for all the help and support while being seconded at Netspar, as well as Anja, Anja, Monique and Korine for all the help and support related to the Department of Econometrics & Operations Research. Furthermore, I would like to thank my roommates Suraj and Lieske. From day one, we had a perfect balance between working in silence and discussing the important things in life. Outside of the academic world, there are many people who I would like to thank (in no particular order): Anneke, Ronald, Monique, Mike, Martijn, Mitchell, Willem, Kevin, Marco, Nico, and everyone who has been a member of S.K.V. Melmac during the last 10 years.
The biggest thanks of all go to my family. Mom and Dad, without you none of this would have happened. You always let us choose our own path, supported us in every decision, believed in what we did, and were simply there at the times we needed you. Mom, thank you for all the moments we exchanged thoughts about the projects that I have been working on. It helped me organize my thoughts, and, more importantly, helped me to realize there is always a different perspective on the matter. I think that we both would have wanted that Annie Spanjers, your mother, my grandmother, would have been able to witness this achievement. Unfortunately, life is not always fair. Dad, this thesis does not suggest by any means that I no longer need your help when it comes to anything that requires a practical solution or physical labour. To my little sister, Anne, thank you for just being the best sister I could have ever asked for. Leonie, my love, thank you for your support, your endless patience whenever I have been a bit less enjoyable to be around, and for ensuring that I kept a - at least somewhat - healthy work-life balance. I love you. The last few years have been amazing and I cannot wait for all the years still to come.
Contents
Preface v
1 Introduction 1
1.1 Saving Preferences after Retirement . . . 2
1.2 Spending from regulated retirement drawdowns . . . 4
1.3 Evidence of Behavioural Life-Cycle Features in Spending Patterns 6 1.4 The effect of heterogeneous shocks in health on wealth . . . 8
1.5 Labour supply effects of inter-vivos transfers . . . 9
2 Saving Preferences after Retirement 13 2.1 Background and motivation . . . 13
2.2 Saving motives of the elderly . . . 15
2.3 The experimental survey . . . 19
2.3.1 Vignette characteristics . . . 20
2.3.2 Differences across vignettes (“treatments”) . . . 23
2.4 Data and descriptive analysis . . . 24
2.5 Modelling the importance of each saving motive . . . 28
2.6 Predicted importance of saving motives . . . 31
2.7 Conclusion . . . 35
2.8 Acknowledgements . . . 38
2.A Pre-test to identify a short list of saving motives . . . 39
2.A.1 Background and methodology . . . 39
2.A.2 Results . . . 39
2.B Base vignette . . . 42
2.C Additional summary statistics and predictions . . . 43
2.D Design of vignette choice sets . . . 45
2.E Background variables: description and descriptives . . . 49
2.F Estimation results . . . 51
2.G Additional results excluding self-gratification in estimation . . . . 55
3 Spending from regulated retirement drawdowns 61 3.1 Introduction . . . 61
3.2 Survey design . . . 65
3.2.1 Survey overview . . . 67
3.2.2 Choice Task . . . 67
3.3 Data and descriptive statistics . . . 72
3.3.1 Dependent variables . . . 74
3.3.2 Control variables . . . 78
3.4 Estimation results . . . 79
3.4.1 For whom is implied endorsement more effective? . . . . 80
3.4.2 Who is more sensitive to the implied endorsement prompt? 84 3.4.3 Robustness checks . . . 90
3.5 Conclusion . . . 97
3.6 Acknowledgements . . . 99
3.A Dutch and the Australian pension systems . . . 100
3.B Summary of the vignette design . . . 102
3.C Additional estimation results . . . 108
3.C.1 Explanation of the positive sign of self-assessed (pension) capability . . . 108
3.C.2 Sensitivity to the definition of reduced sample . . . 109
3.C.3 Sensitivity to the underlying distribution assumptions . . 113
3.C.4 Country-specific robustness . . . 113
4 Evidence of Behavioural Life-Cycle Features in Spending Patterns 123 4.1 Introduction . . . 123
4.2 The Dutch pension system . . . 129
4.3 Vignette study . . . 130
4.3.1 The experimental task . . . 131
4.3.2 Descriptives . . . 133
4.4 Empirical model . . . 136
4.5 Estimation results . . . 139
4.5.1 Preferences under the standard life-cycle model . . . 140
4.5.2 The importance of liquid wealth . . . 141
4.5.3 The middle response alternative . . . 143
4.5.4 Model predictions . . . 144
4.5.5 The importance of illiquid wealth and the tendency to choose the middle response . . . 146
4.6 Conclusion . . . 147
4.A Screenshot of the vignettes . . . 151
4.B Additional tables . . . 153
4.B.1 Survival probabilities . . . 153
4.B.2 Sensitivity to the importance of illiquid wealth parameter 154 4.C Additional figures . . . 155
5 The effect of heterogeneous shocks in health on wealth 157 5.1 Introduction . . . 157
5.2 The Dutch pension- and health care system . . . 162
5.3 Economic framework . . . 163
5.3.1 Setup of the model . . . 163
5.3.2 The impact of varying types of health shocks . . . 166
5.3.3 The empirical analysis in relation to the economic frame-work . . . 167
5.4 Methodology . . . 168
5.5 Data . . . 170
5.5.1 Dataset construction . . . 171
5.5.2 Selection of disease groups . . . 174
5.5.3 Descriptive statistics . . . 177
5.6 Results . . . 178
5.6.1 Main results . . . 179
5.6.2 Heterogeneity . . . 180
5.6.3 Robustness & sensitivity . . . 183
5.7 Discussion and Conclusion . . . 189
5.8 Acknowledgments . . . 193
5.A Derivations of the economic model . . . 194
5.A.1 Setup of the model . . . 194
5.A.2 Relationship between B1 and B2 . . . 197
5.A.3 Setup of the model without intended bequest motive . . 197
5.B Non-parametric event study estimates . . . 199
5.C Additional tables and figures . . . 202
5.D Disability and mortality for 14 ICD chapters . . . 210
5.E Additional tables and figures for 14 ICD chapters . . . 215
5.F International shortlist for hospital morbidity tabulation . . . 236
6 Labour supply effects of inter-vivos transfers 241 6.1 Introduction . . . 241
6.2 The gift tax in the Netherlands: 2012-2014 . . . 245
6.2.1 Exemptions on the gift tax . . . 245
6.2.2 Compensation schemes . . . 246
6.3 Data . . . 247
6.3.1 Source data and sample selection . . . 247
6.4 Econometric models . . . 253
6.4.1 Probit reduced form . . . 254
6.4.2 Tobit reduced form . . . 255
6.5 Estimation results . . . 256
6.5.1 Household labour income, mortgage debt and housing wealth . . . 257
6.5.2 Personal labour income . . . 259
6.5.3 Personal labour income when partner receives the gift . . 262
6.5.4 Robustness . . . 266
6.6 Conclusion . . . 271
6.7 Acknowledgements . . . 272
6.A Timeline for the exemptions on gift taxes . . . 273
6.B Additional figures . . . 275
6.C Dataset . . . 276
6.C.1 More on sample selection . . . 276
6.C.2 Individuals who are in the rush hour of life . . . 277
6.C.3 Analysis sample . . . 278
6.D Additional tables . . . 278
6.D.1 Instrument relevance . . . 278
6.D.2 Estimation results for CF, 2SLS, RE2SLS . . . 285
6.D.3 Inclusion of squared and cubic residuals . . . 292
6.E Estimation results using MLE . . . 294
List of Tables
2.1 Saving motives used in the vignettes . . . 16
2.2 Advised spending pattern by treatment (percent): the Nether-lands and Australia . . . 26
2.3 Changes in advised spending pattern between treatments (per-cent): the Netherlands and Australia . . . 27
2.4 Mean importance of saving motives by treatment and country . 29 2.5 Average expected importance of saving motives by treatment and country . . . 34
2.6 Ranking of 19 possible saving motives . . . 41
2.7 Text of the base vignette . . . 42
2.8 Distribution of the ranking per motive (%) . . . 43
2.9 Average expected importance of saving motives by retirement income policy per spending pattern . . . 44
2.10 Categorisation of gross household income into income groups for the Netherlands (Australia) . . . 45
2.11 Vignette household (yearly) consumption in euros per spending pattern and participant’s income . . . 47
2.12 Remaining vignette household wealth in euros (x1000) at ages 65, 75, 85 and 95 per spending pattern and retirement income policy design . . . 48
2.13 Description of the covariates . . . 50
2.14 Descriptive statistics . . . 51
2.15 Estimation results. Rank ordered logit . . . 52
2.16 Average expected importance of saving motives by retirement income policy per spending pattern excluding self-gratification in estimation . . . 55
2.18 Estimation results excluding self-gratification. Rank ordered logit 57 3.1 Saving motives used in the flexible and regulated drawdown
vi-gnettes . . . 73 3.2 Description of the relevant variables . . . 75 3.3 Descriptive statistics for the dependent and control variables . . 76 3.4 Propensity to change advised spending pattern . . . 83 3.5 Propensity to increase advised spending pattern . . . 85 3.6 Main results for the importance of the “government advice” and
“peer effect” prompts. Average marginal effects based on Logit estimates . . . 91 3.7 Robustness for the prompts using the analysis sample . . . 94 3.7 Robustness for the prompts using the analysis sample (continued) 95 3.7 Robustness for the prompts using the analysis sample (continued) 96 3.8 Share (%) of participants that find the implied endorsement
prompts important using the analysis sample. . . 109 3.9 Results for the importance of the “government advice” and “peer
effect” prompts with objectively measured and self-assessed com-binations . . . 110 3.10 Robustness analysis for the propensity to change advised
spend-ing pattern . . . 111 3.11 Robustness analysis for the propensity to increase advised
spend-ing pattern . . . 112 3.12 Robustness with respect to the underlying distribution . . . 114 3.12 Robustness with respect to the underlying distribution (continued)115 3.12 Robustness with respect to the underlying distribution (continued)116 3.13 Robustness for the prompts using the Australian sample . . . 117 3.13 Robustness for the prompts using the Australian sample
(contin-ued) . . . 118 3.13 Robustness for the prompts using the Australian sample
(contin-ued) . . . 119 3.14 Robustness for the prompts using the Dutch sample . . . 120 3.14 Robustness for the prompts using the Dutch sample (continued) 121 3.14 Robustness for the prompts using the Dutch sample (continued) 122 4.1 Background characteristics . . . 131 4.2 Vignette parameters (income, wealth and spending plan) per
income group . . . 133 4.3 Wealth at the start of age 77 per spending plan, vignette and
income group . . . 134 4.4 Estimation results for the main model specifications. . . 141 4.5 Translation of Figure 4.4. . . 152 4.6 Probability of surviving for another x ∈ [0, 19] years for a 77
LIST OF TABLES 4.7 Predicted advised spending plan per vignette for different varying
importance of illiquid wealth. . . 154 5.1 Summary statistics (averages) by disease group . . . 178 5.2 Summary statistics (averages) by disease for 14 ICD chapters . . 213 5.3 Summary statistics (averages) by disease for 14 ICD chapters
including financial wealth . . . 214 5.4 ISHMT combined with the share of diseases within a disease
group for the analysis sample . . . 236 5.5 Number of observations underlying Figure 5.1 . . . 239 6.1 Summary statistics . . . 252 6.2 The effect of receiving a gift on log(gross hhold. labour inc.),
log(mortgage debt) and log(gross housing wealth) . . . 260 6.3 Instrument relevance using the analysis sample for the
sec-ond stage dependent variable: log(gross hhold. labour inc.), log(mortgage debt) and log(gross housing wealth) . . . 261 6.4 The effect of receiving a gift on log(gross personal labour income)
for both males and females . . . 263 6.5 The effect of receiving a gift on log(gross personal labour income)
for the breadwinner’s and non-breadwinner’s sample . . . 264 6.6 The effect of the partner receiving a gift on log(gross personal
labour income) for both males and females . . . 265 6.7 The effect of receiving a gift of at least 40 000 Euros on log(gross
hhold. labour inc.), log(mortgage debt) and log(gross housing wealth) . . . 268 6.8 The effect of receiving a gift for primary home on log(gross hhold.
labour inc.), log(mortgage debt) and log(gross housing wealth) 269 6.9 The effect of receiving a gift for primary home on reduction of
the mortgage . . . 271 6.10 Timeline of the for this paper relevant reforms in the gift tax of
the Netherlands . . . 273 6.11 Instrument relevance using the analysis sample for the second
stage dependent variable: log(gross personal labour income) for both males and females . . . 279 6.12 Instrument relevance using the analysis sample for the second
stage dependent variable: log(gross personal labour income) for the breadwinner’s and non-breadwinner’s sample . . . 280 6.13 Instrument relevance using the analysis sample for the second
6.14 Instrument relevance when receiving a gift of at least 40 000 Euros using the analysis sample for the second stage dependent variable: log(gross hhold. labour inc.), log(mortgage debt) and log(gross housing wealth) . . . 282 6.15 Instrument relevance when receiving a gift for primary home
us-ing the analysis sample for the second stage dependent variable: log(gross hhold. labour inc.), log(mortgage debt) and log(gross housing wealth) . . . 283 6.16 Instrument relevance when receiving a gift for primary home
us-ing the analysis sample for the second stage dependent variable: mortgage debt . . . 284 6.17 The effect of receiving a gift on log(gross personal labour income)
for both males and females . . . 285 6.18 The effect of receiving a gift on log(gross personal labour income)
for the breadwinner’s and non-breadwinner’s sample . . . 286 6.19 The effect of the partner receiving a gift on log(gross personal
labour income) for both males and females . . . 287 6.20 The effect of receiving a gift on log(gross hhold. labour inc.),
log(mortgage debt) and log(gross housing wealth) . . . 288 6.21 The effect of receiving a gift of at least 40 000 Euros on log(gross
hhold. labour inc.), log(mortgage debt) and log(gross housing wealth) . . . 289 6.22 The effect of receiving a gift for primary home on log(gross hhold.
labour inc.), log(mortgage debt) and log(gross housing wealth) 290 6.23 The effect of receiving a gift for primary home on mortgage debt 291 6.24 The effect of receiving a gift on log(gross hhold. labour inc.),
log(mortgage debt) and log(gross housing wealth) . . . 292 6.25 The effect of receiving a gift on log(gross hhold. labour inc.),
log(mortgage debt) and log(gross housing wealth) . . . 293 6.26 The effect of receiving a gift on log(gross hhold. labour inc.),
List of Figures
2.1 Simplified overview of the different components within the survey. 21 2.2 Spending patterns for a household in the lowest income group
with moderate income and moderate wealth (as defined in Sec-tion 2.3.2) . . . 22 2.3 Visualization of the first two columns in Table 2.5 . . . 32 2.4 Example comparison set to elicit best/worst saving motives . . . 40 3.1 Simplified overview of the different components within the survey. 69 3.2 Base text for each vignette . . . 103 3.3 Screenshot of task 1 (Part A) for the flexible drawdown vignette
for the lowest income group in the Australian version of the survey.104 3.4 Screenshot of task 2 (Part B) in the flexible drawdown vignette
for the lowest income group in the Australian version of the survey.105 3.5 Screenshot of task 1 (Part A) for the regulated drawdown vignette
for the lowest income group in the Australian version of the survey.106 3.6 Screenshot of task 2 (Part B) for the regulated drawdown vignette
for the lowest income group in the Australian version of the survey.107 4.1 Distribution of the advised spending plan per vignette per income
group . . . 135 4.2 Predicted advised spending plan per vignette. . . 145 4.3 Predicted advised spending plan per vignette for different
be-havioural life-cycle features. . . 148 4.4 Screenshot (in Dutch) for the high-low pension income vignette
for a respondent with gross household income between 52 000 and 73 500 euros. . . 151 4.5 Predicted advised spending plan per vignette. . . 155 4.6 Predicted advised spending plan per vignette for different
5.1 Mortality and disability across disease groups . . . 175 5.2 Survival and the eligibility for home care and nursing home care
before and after a hospital admission. . . 176 5.3 Impact of hospitalization on gross household income for selected
ICD chapters that vary in mortality and disability . . . 181 5.4 Impact of hospitalization on savings account for selected ICD
chapters that vary in mortality and disability . . . 181 5.5 Impact of hospitalization on savings account, individuals whose
savings are above 80th percentile at t − 1, for selected ICD chapters that vary in mortality and disability . . . 184 5.6 Impact of hospitalization on savings account, individuals whose
savings are below 80th percentile at t − 1, for selected ICD chapters that vary in mortality and disability . . . 184 5.7 Impact of hospitalization on savings account, singles, for selected
ICD chapters that vary in mortality and disability . . . 185 5.8 Impact of hospitalization on savings account, two-person
house-holds, for selected ICD chapters that vary in mortality and disability185 5.9 Impact of hospitalization on savings account, having children,
for selected ICD chapters that vary in mortality and disability . . 186 5.10 Impact of hospitalization on savings account, without having
children, for selected ICD chapters that vary in mortality and disability . . . 186 5.11 Impact of hospitalization on savings account, without
comor-bidities, for selected ICD chapters that vary in mortality and disability . . . 190 5.12 Impact of hospitalization on savings account, with comorbidities,
for selected ICD chapters that vary in mortality and disability . . 190 5.13 Impact of hospitalization on gross household income (event
study approach) . . . 200 5.14 Impact of hospitalization on savings account (event study
ap-proach) . . . 201 5.15 Impact of hospitalization on gross household income, additional
demographics . . . 202 5.16 Impact of hospitalization on savings account, additional
demo-graphics . . . 203 5.17 Impact of hospitalization on gross household income, matching 204 5.18 Impact of hospitalization on savings account, matching . . . 205 5.19 Impact of hospitalization on gross household income, unbalanced
sample . . . 206 5.20 Impact of hospitalization on savings account, unbalanced sample207 5.21 Impact of hospitalization on gross household income
LIST OF FIGURES 5.22 Impact of hospitalization on savings account (difference-in-difference
approach) . . . 209 5.23 Eligibility for home care in the years before and after a hospital
admission (normalized at t − 1) . . . 210 5.24 Eligibility for nursing home care in the years before and after a
hospital admission (normalized at t − 1) . . . 211 5.25 Probability of dying in the years after a hospital admission . . . . 212 5.26 Impact of hospitalization on gross household income . . . 215 5.27 Impact of hospitalization on savings account . . . 216 5.28 Impact of hospitalization on savings account, individuals above
80th percentile at t − 1 . . . 217 5.29 Impact of hospitalization on savings account, individuals below
80th percentile at t − 1 . . . 218 5.30 Impact of hospitalization on savings account, singles . . . 219 5.31 Impact of hospitalization on savings account, two-person
house-holds . . . 220 5.32 Impact of hospitalization on gross household income, singles . . 221 5.33 Impact of hospitalization on gross household income, two-person
households . . . 222 5.34 Impact of hospitalization on savings account, having children . . 223 5.35 Impact of hospitalization on savings account, without having
children . . . 224 5.36 Impact of hospitalization on savings account, without comorbidities225 5.37 Impact of hospitalization on savings account, with comorbidities 226 5.38 Impact of hospitalization on gross household income, k = 3 . . . 227 5.39 Impact of hospitalization on savings account, k = 3 . . . 228 5.40 Impact of hospitalization on financial wealth . . . 229 5.41 Impact of hospitalization on financial wealth, without financial
assets at t = 0 . . . 230 5.42 Impact of hospitalization on financial wealth, with financial
assets at t = 0 . . . 231 5.43 Impact of hospitalization on savings account, singles with
chil-dren . . . 232 5.44 Impact of hospitalization on savings account, singles without
children . . . 233 5.45 Impact of hospitalization on savings account, two-person
house-holds with children . . . 234 5.46 Impact of hospitalization on savings account, two-person
6.3 Average amount given conditional on receiving a tax registered gift (x1000) . . . 251 6.4 Tax-exempt gift amounts for an annual gift from parent to child 274 6.5 Tax-exempt gift amounts for a one-off gift from parent to child
for general purpose . . . 274 6.6 Tax-exempt gift amounts for a one-off gift from parent to child
for the purchase or renovation of primary home, or repayment of mortgage debt . . . 274 6.7 Fraction of individuals that received a general purpose one-off gift275 6.8 Fraction of individuals that received a one-off gift for the
1
Introduction
Exploiting stated preference has become more common in the economics of ageing literature in recent years (see, e.g., Brown et al. 2017, 2019, 2008b, and Elsayed et al. 2018). Stated preferences are particularly helpful for studying preferences for choice options that are not (widely) available or known by individuals. Admittedly, a drawback is that people might not fully take into account the experimentally varied hypothetical scenario, a difficulty in all hypothetical experiments. The stated preference experiments in this thesis are administered to random samples of the Dutch (and Australian) population of ages 50-64 who did not yet retire. We restrict the experiments to this age range, as individuals aged 50-64 are likely to already have thought at least somewhat about retirement (De Bresser and Knoef 2015). This mitigates the potential drawback that individuals might not be involved and thus would not fully think through the hypothetical consumption decisions during retirement as much as they would do in real-life decisions. This does not mean that individuals do not make mistakes, but increases confidence that the systematic patterns in the stated choices resemble those in real-life choices. Moreover, literature shows that stated preference are often well in line with revealed preferences and help to predict actual behaviour (see, e.g., Michaud et al. 2020).
The final two chapters use administrative data instead of survey data on stated preferences. Chapter 5 studies the effect of a hospital admission for a specific condition on spending for the Dutch elderly. In the final chapter of this thesis, the focus shifts from preferences for spending and saving of individuals close to retirement or retirees, to personal finance of couples in the “rush hour of life”. Chapter 6 investigates the effect of tax-registered parental transfers on the recipients’ labour income in the year of the gift. Each essay is an independent paper. The remainder of this introductory chapter serves to link the various chapters and to present the principal findings of each paper.
1.1
Saving Preferences after Retirement
Saving Preferences after Retirement | Section 1.1 to hypothetical recently retired households (described in vignettes) and to rank the importance of saving motives that justify their advice. We vary wealth and income patterns to reflect different pension systems - from full annuiti-zation to complete flexibility - allowing us to analyse how the importance of saving motives varies with pension system design. We present a broad menu of saving motives founded on economic theory, as well as motives informed from economic psychology. A better understanding of motives for continued asset accumulation (or slower than expected decumulation) is important for analysing the efficiency of retirement saving and decumulation policy design.
The vignette methodology allows participants to make choices in different situations, including counterfactual settings that do not reflect the institutions of their own country. It appears that changing liquidity affects the advised consumption pattern in both countries. Dutch participants become less conser-vative (i.e. advise a high spending pattern more often) if their pension wealth is more liquid than in the vignette corresponding to the actual Dutch pension system. Correspondingly, Australian participants become more conservative in a setting with lower availability of wealth and higher annuity income than in the vignette corresponding to their actual institutional setting. On the other hand, our results seem to suggest that the importance of most saving motives is not substantially affected in the short run, by variation in experimental retirement drawdown arrangements that only affects the respondents in the long run. This could indicate considerable inertia caused by respondents’ national context, which, in turn, may hamper the evaluation of major changes in pension policy as it may take a long time before the full effect of a major change is noticeable in practice. Our results show that expected major health shocks or mortality of the partner have an impact on the advised spending pattern and saving motives. Not surprisingly, expecting a health shock in the near future raises the desire to hold precautionary savings for health. Similarly, the expectation that one of the household members dies within 10 years after retirement significantly raises the desire to ensure that if they die, their partner is able to maintain his or her standard of living - i.e. the intra-household bequest motive.
of the hypothetical household are able to enjoy life now as well as later - i.e. the self-gratification motive. In contrast to the results for the United States, for example in De Nardi et al. 2016, our results suggest that the desire to leave a bequest to the dependents or estate and life-span risk are less important, irrespective of the country of residence, advised spending pattern, or the insti-tutional setting. Based on this chapter, we find that lifetime consumption and saving decisions are complex choices for individuals. In order to accommodate for the observed heterogeneity and to protect individuals from suboptimal decisions that reduce expected lifetime utility, one option is to use choice archi-tecture to set drawdown decisions for a substantial proportion of individuals (Benartzi and Thaler 2007). This is not addressed in the current essay, but is investigated in chapter 3.
1.2
Spending from regulated retirement drawdowns:
the role of implied endorsement
Spending from regulated retirement drawdowns | Section 1.2 advice from the government or peers (a social norm comparison) and whether the nudge actually influence the vulnerable groups it is designed to target. We investigate this using an online survey.
Chapter 2 and chapter 3 use the data from the same experimental survey fielded in the Netherlands and Australia. In this essay, our experimental set-up uses two vignettes to present short descriptions of hypothetical retiree households with a given income similar to the state pension level, accrued retirement savings and an expectation to be in good health at age 70. One vignette allows flexibility of drawdown (flexible drawdown vignette), while a subsequent vignette includes a regulated drawdown from pension wealth set by government (regulated drawdown vignette). For both vignettes, participants are asked to advise a hypothetical retiree household on a spending pattern and rank the importance of a set of saving motives consistent with that spending advice. This design enables us to compare stated retirement spending and saving behaviour while simultaneously ranking motivations for such behaviour with, and in the absence of, an implied endorsement nudge.
1.3
Evidence of Behavioural Life-Cyle Features in
Spend-ing Patterns after Retirement
In this chapter, we continue to investigate the role that nudges could play to help individuals deciding how to spend down wealth during retirement. The most effective way to nudge individuals to make certain choices without forcing them or creating real economic incentives seems to be setting a default. Even though individuals can easily deviate from the default option in principle, they overwhelmingly tend to follow the default in practice. In the accumulation phase, they stick to the default for various key decisions, even if it harms their retirement savings (Dobrescu et al. 2018). After retirement, the default can explain the difference in take up rates of annuities versus a lump sum (Bütler and Teppa 2007). Consequently, policy-makers and pension plan providers carry great responsibility in carefully designing defaults, framing and other nudges implied by the choice architecture. Therefore, it is of key importance to understand how individuals value different assets and how they select different wealth decumulation and consumption strategies to maximize utility.
Effectively, the default in the Netherlands is to transform accrued pension rights into a life-long flat-rate annuity. Individuals can deviate from the default at the start of retirement by choosing a high-low (or low-high) pension income. This implies a high (lower) pension in the first years after retirement and a lower (high) pension in later years, in such a way that the expected net present value remains constant. Rather than focusing on the effectiveness of this already available option, our first contribution of this chapter is to investigate, using a utility framework, the behavioural implications of these alternative pension income schemes for expenditure choices, since the (exogenously given) income pattern acts as a frame for the (endogenous choice of) expenditure patterns. Our second contribution is in the importance of the (illiquid) net present value of the life-long annuity compared to liquid wealth. In the standard life-cycle model these are commonly assumed to be equally important, whereas in real-life this is ambiguous, since the complex setting of the real-life-cycle choice problem and the way it is presented to the individual may make one form of wealth more salient than the other (Levin 1998).
Evidence of Behavioural Life-Cycle Features in Spending Patterns | Section 1.3 hypothetical, but realistic, retirement scenarios. The scenarios varied in pension income, based on a high-low, a low-high and a flat-rate annuity design. We use the responses to investigate the behavioural implications of these differ-ent plans using a stylized model based upon life-cycle utility maximization that accounts for some behavioural features. First, we allow for framing as individuals can make systematically different choices depending on how their pension income is distributed over time. Second, in contrast with the standard life-cycle model, we allow for the possibility that individuals do not value illiq-uid wealth (the expected value of future annuity income) and liqilliq-uid wealth (discretionary wealth and the remainder of a lump sum paid at retirement) as equally important. Third, we account for the tendency of respondents to choose the middle option among the five consumption expenditure patterns in each choice set. Our findings are twofold. First, choices of preferred spending for a large majority of individuals depend on how income is annuitized. Second, in contrast to the prediction of the life-cycle model, liquid and illiquid wealth ten years into retirement affect chosen consumption patterns in very different ways. These findings combined suggest that most individuals do not behave as rational economic agents and policy makers should take this into account when they design choice architecture and communication for the decumulation of pension wealth.
(e.g. Davidoff et al. 2005). Nonetheless, the voluntary take-up of annuities in retirement is low. Researchers have given various explanations for this puzzle, such as uncertain medical expenses (Ameriks et al. 2011), hyperbolic discounting (Schreiber and Weber 2016), lack of cognitive skills in valuing complex annuities (Brown et al. 2019) and framing (Brown et al. 2008b). Our findings suggest that individuals who do not annuitize their pension wealth in real-life, may do so because they under value illiquid wealth, in the form of the net present value of future annuities, compared to the liquid wealth of a lump sum.
1.4
The effect of heterogeneous shocks in health on
wealth
Labour supply effects of inter-vivos transfers | Section 1.5 Using a stylized two-period model, we show that individuals are expected to adjust their savings after a health shock in the absence of income risk and out of pocket medical expenditures. In our model, consumption depends on health through survival, the marginal utility of consumption, and the strength of the bequest motive. For example, a health shock that has a strong negative effect on survival, but not on disability or reliance on care, will decrease wealth holdings as the expected time left to consume lifetime wealth decreases sharply. Similarly, health shocks with small effects on mortality, but large effects on disability, might increase savings as they could induce people to leave a strategic bequest. This may motivate relatives to provide informal care when needed (Zweifel and Strüwe 1996). We find that, regardless of the level of mortality and disability related to the hospital admission for a particular disease, there is no economically relevant effect on the savings account amount. This finding is consistent among different subgroups and across different model specifications. Based on these results, our overall conclusion is that, at least in the social system in place in the Netherlands, older individuals stick to a pre-determined level of consumption regardless of their health.
1.5
Labour supply effects of inter-vivos transfers in the
rush hour of life
bequeathing wealth if this becomes more attractive from a fiscal point of view (Poterba 2001).
Chapter 6 presents an analysis of the income effect of non-labour income using a gift tax reform in 2013 in the Netherlands. A temporary change in gift taxes generated exogenous variation in gifts, which we exploit to estimate the causal effect of non-labour income on labour supply. The gift tax rate in the Netherlands depends on both the value of the gift and the relationship between the recipient and the donor. There are a few specific rules that need to be accounted for. First, each child may receive tax-free gifts of roughly 5000 euros from their parents per year. All gifts above this threshold need to be declared at the tax authority. Second, the annual tax-exempt amount for children can be increased to roughly 25 000 euros once. If the gift is intended to finance a primary home purchase, renovation or to pay-off mortgage debt of the primary home, the exemption is approximately 50 000 euros. We observe the tax-registered and one-off gifts in our data. For both of the one-off tax-exempt gifts, age restrictions of the recipient apply. As of 2013, the age restriction is loosened from 18-35 to 18-40. In the final quarter of 2013 and in 2014, the age restriction for the one-off gift for the primary home was temporarily abolished. In addition, the tax-exempted gift amount temporarily increased from 50 000 to 100 000 euros. Thus, in 2012 individuals aged 36-40 were not eligible for the increased tax-exempted gift, while in 2013 and 2014 individuals aged 36-40 were eligible.
2
Saving Preferences after Retirement
This chapter is based on Alonso-García, Bateman, Bonekamp, van Soest, and Stevens (2018).
2.1
Background and motivation
The life-cycle model predicts that individuals and their households accumulate wealth during their productive careers and draw down their assets after retire-ment. Still, in spite of large differences in pension systems across countries and substantial pension reforms, recent empirical studies for different countries show that many retirees hold on to their assets or even keep on saving well into old age; see Love et al. (2009) for the United States, Van Ooijen et al. (2015) for the Netherlands, and Asher et al. (2017) for Australia. Several explanations for this stylised fact have been suggested, including motives informed by economic theory such as precautionary saving or the utility of leaving a bequest, and more abstract motives such as the fear of losing autonomy or the desire to feel secure.
of both partners. We then ask the survey participants to advise the retiree household on a spending pattern and to rank the importance of a set of saving motives justifying this advice. The answers reveal the survey participants’ stated preferences and not their behaviour in actual situations. Admittedly, the experiment is cognitively challenging and some of the results may reflect that participants do not fully take into account the experimentally varied hypothetical situation, a difficulty in all hypothetical experiments. Still, the literature shows that stated preferences are often well in line with revealed preferences and help to predict actual behaviour; see, e.g., Whitehead et al. (2008) or Michaud et al. (2020).
Saving motives of the elderly | Section 2.2 We find that the most important reasons for spending and saving after retire-ment are self-gratification, autonomy, liquidity (for the Dutch only) and security (for Australians). Life-span risk is not an important saving motive in either country. In absence of expected major health shocks or mortality of the partner, precautionary health, autonomy, liquidity (for the Dutch) and self-gratification (for Australians) are particularly important for those who advise conservative spending. Motives to spend and save after retirement are not sensitive to (ex-perimental) changes in the institutional pension settings, i.e., to whether there is full mandatory annuitisation or full flexibility. Awareness of the potential risks faced in the actual institutional setting (Dutch or Australian) appears to be more important for the ranking of saving motives than the experimental setting, suggesting that retirees may only slowly adjust their saving and spend-ing patterns after an actual policy shift. On the other hand, the importance of saving motives is modified in the event of an expected deterioration in own or the partner’s health: this makes the precautionary health saving motive more important.
The remainder of this paper is structured as follows. Section 2.2 provides a brief review of explanations for why individuals hold on to their wealth in retirement in the economic and psychology literature. Section 2.3 presents the experimental design and the structure of the experimental survey. Section 2.4 describes the data. Section 2.5 describes the estimation models. Section 2.6 presents the results and Section 2.7 concludes.
2.2
Saving motives of the elderly
not important whereas from a psychological point of view the framing of the decision matters.
In the experiment analysed in this paper, we restrict ourselves to ten possible saving motives, avoiding the cognitive demands to the survey participants
imposed by extensive lists of saving motives.1 To select the ten motives, we
used a pre-test described in detail in Appendix Section 2.A. In the remainder of this section we focus on the literature on the ten selected saving motives in the context of the elderly, see Table 2.1, distinguishing between those founded on economic theory and more abstract motives informed by economic psychology.
Table 2.1: Saving motives used in the vignettes
Name Text in vignette (The household . . . )
Economic
precautionary wants to ensure that they will be able to finance any unforeseen expenditures other than health and aged care expenditures precautionary health wants to ensure that they will be able to finance unforeseen health
and aged care expenditures
life-span risk wants to ensure that they will not outlive their wealth
intended bequest wants to ensure that they will be able to leave a bequest to their dependents or estate
liquidity wants to ensure that they have enough cash on hand at any time intra-household bequest wants to ensure that if they die, their partner is able to maintain
his/her standard of living
Psychological
autonomy wants to ensure that they remain financially independent security wants to ensure that they have enough money to have peace of
mind
self-gratification wants to ensure that they are able to enjoy life now as well as later political risk wants to ensure that they are protected against a change in the
superannuation / pension rules
Economic saving motives
The economic motives we consider are related to precautionary savings (sub-divided into precautions for health expenditures and general expenditures), bequests (intra-household and intergenerational), life-span risk, and liquidity. The work surveyed in De Nardi et al. (2016) suggests that the economic saving motives of the elderly essentially break down into two categories: precautionary savings, mainly for the risks implied by life-span uncertainty (the life-span risk
1Alternatively, selecting for each participant only a small subset from an extensive list would
Saving motives of the elderly | Section 2.2 motive) and uncertain medical - out of pocket - expenditures (the precautionary (health) motive) and bequest motives (the (intra-household) bequest motive). The literature related to savings for life-span risk has a long history, dating back to Yaari’s seminal paper (Yaari 1965) and empirical studies since then confirm the role of uncertain lifetimes on precautionary saving (Davies 1981, De Nardi et al. 2009, Kennickell and Lusardi 2004). De Nardi et al. (2009), using U.S. data on people aged 75 and older, show that individuals deplete their net worth by the end of their certain lifetime whereas individuals facing an uncertain life-span still have significant asset holdings towards the end of their lives, even when facing poor survival prospects. The marginal utility of consumption may vary with health (Finkelstein et al. 2013, Kools and Knoef 2019), implying that health also affects the optimal level of life-cycle savings, something that many studies do not take into account. The role of savings for health expenditures, however, is undisputed (De Nardi et al. 2016).
The importance of intergenerational transfers, both inter-vivos and in the form of bequests, has gained considerable attention in the economics literature (e.g. Masson and Pestieau 1997). There is no consensus on why people leave a bequest. Some argue that bequests are mainly accidental (Hurd 1989) as the elderly keep a buffer as a result of life-span risk. Lockwood (2018) argues that accumulated wealth during retirement serves the double purpose of building a bequest and covering large health care expenditures. Others find that bequests are intentional and motivated by inter-generational altruism or the joy of giving (De Nardi and Yang 2014, Laferrère and Wolff 2006).
costly or unavailable, the willingness to stay in the same place may be a reason for the elderly to save during retirement (Venti and Wise 2004).
Psychological saving motives
An increasing number of studies in the economics literature emphasise the importance of abstract explanations for savings (Canova et al. 2005, Shefrin and Thaler 1988). The psychology literature suggests that individuals find more abstract saving goals more important than concrete motivations, or save as a buffer against social risks (Engelberg and Sjöberg 2007). Canova et al. (2005) identified fifteen salient abstract motives for saving. These include autonomy, self-gratification and security, which are among the ten most important saving motives according to our pre-test. An explanation for why they are perceived as important is that these saving motives are likely to be the target of other saving motives (including ‘economic’ explanations) and linked to other goals. This aligns with the early work of Yamauchi and Templer (1982) who identify, using an experimental setting, three dimensions to explain the attitude towards money. The first is “power and prestige” – purchasing items or accumulating wealth to impress others and increase your self-esteem; the second and third are “time-retention” and “security”, which can be interpreted as placing value on preparing for future goals or security. Moreover, the psychology literature suggests that there is a tendency to view saving as a protection against the kind of vulnerability that is inherent to social involvement (Engelberg and Sjöberg 2007, Yamauchi and Templer 1982), explaining the importance of the motives autonomy and security. Risks could include the loss of trust and confidence in others, or loss of autonomy and, as a consequence, dependence on other people.
The experimental survey | Section 2.3 population, Van Dalen and Henkens (2018) find that the Dutch have reduced their trust in pension funds, banks and insurance companies after the global financial crisis, possibly affecting saving attitudes and actual saving behaviour.
2.3
The experimental survey
Individuals from representative samples in the Netherlands and Australia were invited to participate in an online experimental survey on spending patterns and saving motives after retirement. The experimental task focused on two main objectives: To investigate the effects of receiving a lump sum at retire-ment (“liquid wealth”) or a lifetime annuity (“illiquid wealth”) and to assess the effect of (expected) health problems on preferred spending patterns and
saving motives.2 We use stated-choice questions for hypothetical households,
so-called vignettes.3 Vignettes have often been used in economics since Van
Beek et al. (1997). Our vignettes comprise short descriptions of income, wealth and health status for retired couples. We use hypothetical households so that participants can be asked to make choices in an institutional setting that differs
from that in their own country.4 After each stated choice question, we collected
information on the participant’s motives for their decisions. In addition, we collected information on personal and household characteristics, personality traits, and financial competence, which may be associated with saving motives and spending preferences. For example, Asebedo et al. (2019) find that person-ality traits affect saving motives and therefore indirectly affect saving behaviour. The Dutch survey was fielded in December 2016 and the Australian survey in
March 2017.5
Participants in the Netherlands were recruited from two well-established ongoing panels - the LISS panel and the CentERpanel, including over 5,000 households in total. Invited panel members are a representative sample of
2A third objective, to analyse the role of government prescribed drawdown patterns (that is,
implied endorsement), is analysed in Alonso-García et al. (2021).
3See Louviere et al. (2000) for advantages and drawbacks of stated preference methods. 4In earlier experiments where we asked respondents to imagine themselves in a hypothetical
situation, we often got the comment that this situation was completely unrealistic for them. This is avoided using the vignettes.
5Static copies of the questionnaires are available in the supplementary materials as ‘Dutch
Dutch households, selected by Statistics Netherlands. Participants respond to surveys on a regular basis (biweekly for CentERpanel, monthly for LISS). We selected individuals aged 50 - 64, working for pay or with a partner working for pay. The Dutch sample comprised 1,798 eligible individuals. Participants in Australia were recruited from the commercial web panel ‘TEG rewards’ (which
includes over 1,000,000 panel members).6 The Australian sample comprised
1,004 randomly selected from all panel members aged 50-64 and not yet retired. The median time for completion of the survey was 31 and 30 minutes for the Dutch and Australian samples, respectively.
The flow of the survey is illustrated in Figure 2.1. It consists of some preliminary questions and four modules. The preliminary questions cover marital status, household income, and age and employment status of participant and partner (if applicable). This information is used to select the sample and to
allocate the participant to one of four household income groups.7 Module 1 is
the experimental task, explained in detail in Subsections 2.3.1 and 2.3.2 below. The other modules have questions on risk and time preferences, financial literacy
and pension knowledge, personality8 and other background characteristics that
may help to explain savings and consumption patterns behaviour and saving motives.
2.3.1
Vignette characteristics
Each participant is shown eight different vignettes, describing eight hypothetical retired couples that differ in expected health status (four alternatives), the institutional retirement income arrangement (“liquidity” - the combination of annuity income and freely available wealth; three alternatives), and implied endorsement (two alternatives). For each vignette the participant (A) advises one preferred spending pattern, and (B) ranks the importance of five saving motives for the given advice in two rounds of best / worst choice sets. Income
6LISS panel members were paid €5 for completing the survey; CentERpanel members
only receive a small compensation for internet use.Australian participants were paid A$4 on completion of the survey.
7These questions were not included in the Dutch survey, as the relevant information was
already available; Dutch participants with missing information on gross household income were randomly allocated to an income group.
8We follow the recent practice to add psychological personality tests such as the Big Five to
The experimental survey | Section 2.3 and wealth of the vignette households are based on (median) wealth and
the participant’s income group.9 The net present value of the total wealth at
retirement is always the same for participants in a given income group. The liquidity of retirement saving in each vignette is discussed in detail in Section 2.3.2; The text of the first (base) vignette is displayed in Table 2.7 in the appendix. Section 2.D in the Appendix explains in detail how the amounts in the vignettes (wealth and income, consumption expenditures) are determined.
Figure 2.1: Simplified overview of the different components within the survey.
Module 1: Vignettes
varying retirement income policy design varying health status
1 2 3 4 5 6 7 8 liquidity high liquidity moderate liquidity low regulated drawdown both healthy healthy + disabled healthy + dead disabled + dead Presentation of vignettes 1-3 randomized Presentation of vignettes 4-8 always in order Choice task
Part A: What spending pattern (from five alternatives) do you advise
the household to choose (based on your preferences)?
Part B: Indicate motivation for Part A choice by ranking five saving
motives from most to least important
Survey questions
Module 2: Questions on retirement planning, CFC (consideration of future
consequences), risk attitude and personality traits elicited using the ten-item personality inventory
Module 3: Questions on financial competence, including financial literacy,
numeracy and pension knowledge
Module 4: Questions on demographics and personal characteristics Preliminary screening
Age 50-64, not retired Allocated to 1 of 4 income groups
A: Advising a spending pattern
For each vignette, the participant is asked to advise one spending pattern out of five alternatives. To reduce complexity (Iyengar and Kamenica 2010), all five
9Wealth and income groups are aligned with gross household income in the Dutch LISS
alternatives have constant expenditures over time as long as financial wealth is positive; after that, expenditures are equal to income. For a given participant, the same five spending patterns are presented in all vignettes; the patterns are aligned to the participant’s household income group only. To help participants understand the implications of choosing a spending pattern, we remind them of the vignette household’s lifetime income (presented earlier in the vignette) and include information on remaining wealth at ages 65, 75, 85, and 95 for
all possible choices.10 Figure 2.2 shows an example screen shot of the five
spending patterns presented.
Figure 2.2: Spending patterns for a household in the lowest income group
with moderate income and moderate wealth (as defined in Section 2.3.2)
B: Saving motives used to advise a spending pattern
Informed by the economics and psychology literature, we identified 19 possible saving motives for retirees. Using a pre-test we reduced the list to ten saving motives that were most often considered important; see Appendix Section 2.A. Table 2.1 lists these ten motives, with their explanations given to the participants. These are the motives we used in our experiment. In the empirical analysis, we take them at face value, though some of them may be correlated and point at the same underlying concept. For convenience, we have divided them into economic and psychological motives, though admittedly, this categorization is not fully unambiguous. (For example, the political motive could be reflect genuine uncertainty about future policy changes and reflect economic arguments.)
10Remaining wealth at ages 65, 75, 85 and 95 per income group, spending pattern and
The experimental survey | Section 2.3 To reduce respondent burden, each participant ranks the importance of a randomly selected set of five motives (one set of five for the first four vignettes
and another set for the last four vignettes).11 The order in which these five are
presented is randomised.
2.3.2
Differences across vignettes (“treatments”)
In the first three vignettes, both partners are 65 years old and just retired. They are in good health and expect to remain so until at least age 70. They own the house they live in (without a mortgage) and do not intend to move or sell the house. If one of them dies, the widow(er) receives less pension income, but also spends less. The reduction in income is roughly equivalent to the expected decrease in expenditures needs. Based on the participant’s income group (Table 2.10), the couple has a given level of total resources (liquid wealth plus net present value of annuities, excluding housing wealth). Vignettes 1, 2 and 3 differ in the extent to which retirement savings are liquid – i.e., liquid wealth versus lifetime annuity. Vignette 1 has high wealth and low income, vignette 2 has moderate wealth and moderate income, and vignette 3 has low wealth and
high income.12 The order in which the first three vignettes are presented is
randomised.
An important characteristic of the experimental design is the relation be-tween the liquidity treatments in vignettes 1 – 3 and country specific pension rules. In the Netherlands, second pillar pension contributions are always con-verted into a lifetime income stream (full annuitisation; vignette 3), whereas Australian retirees can choose to take lump sums or phased withdrawals from their superannuation accumulation, mimicked by vignette 1. In the experiment, we explicitly mention lifelong income and wealth at retirement and point out that if wealth is exhausted, the household has to adapt their spending to their income, so there are no additional resources of income or wealth. The interme-diate arrangement (vignette 2) corresponds to a potential future direction for both retirement systems, with more flexibility than the Dutch system but less than the Australian system, offering more longevity protection than the current rules (Treasury 2016). Alternatively, vignette 2 could also be interpreted as a
system in transition from defined benefit to defined contribution, which is, for example, representative of the U.S. and the U.K.
In vignettes 1-3, both household members expect to remain in good health
until at least age 70 (shorthand notation: {H70, H70}). Vignette 4 is not used
in the current paper (it introduces a feature specific to the Australian system). Vignettes 5 to 8 change the couple’s health status (from good to poor). In
vignette 5, both expect to remain healthy until at least age 75 ({H75, H75}).
The household in vignette 6 expects that one of them will have difficulties
with activities of daily living (ADL) within ten years ({H75, M75}), while the
household in vignette 7 expects that one of them will pass away within 10 years
({H75, D75}). In both cases, the surviving spouse expects to remain healthy until
at least age 75. The couple in vignette 8 expects that one member of the house-hold will pass away within 10 years, and that the survivor will develop some
ADL limitations ({M75, D75}). For vignettes 5-8, each participant is randomly
assigned to liquidity treatment 1 [liquidity = high] or 3 [liquidity = low]. To keep the flow of the survey manageable, we did not randomize the order of the vignettes 5-8. We felt that presenting healthy and unhealthy persons in random order would increase respondent burden, leading to noisier answers. For the first three vignettes, we analysed whether order effects are important, adding dummies for five of the six different orderings for each motive except the benchmark to the model (see next sections). The null hypothesis of no order effects (coefficients on all these dummies are zero (5 x 9 = 45 restrictions)) was not rejected (p-value 0.1254).
2.4
Data and descriptive analysis
Data and descriptive analysis | Section 2.4
(see Table 2.13 in the Appendix).13 This reduced the initial samples to an
analysis sample of 1,437 Dutch and 983 Australian participants.
Background variables that may drive the importance of saving motives are defined in the Appendix Table 2.13, with descriptive statistics in Table 2.14. There are some substantial differences between the two countries, with, for example, much higher average pension income replacement rates in the Netherlands than in Australia.
Table 2.2 presents the distribution of advised spending patterns by vignette (“ treatment”) and by country. Note that spending patterns are ordered from high spending (s = 1) to low spending (s = 5; cf. Figure 2.2). Irrespective of the treatment, spending patterns s = 2 and s = 3 are the most popular. There is a salient difference between the countries: Australians tend to choose a more conservative spending pattern than the Dutch in all vignettes. This corresponds to the real-world context in the two countries. In the Netherlands, individuals have higher replacement rates, and 70% final pre-retirement earnings is still the social norm (Knoef et al. 2016), corresponding to spending pattern 2. In Australia, replacement rates are lower (OECD 2015a), and pension adequacy is generally communicated in terms of absolute rather than relative expenditure levels (ASFA 2018). These are usually lower than 70% of final earnings. There are also some interesting differences between the vignettes. For example, the death of a partner leads to a choice for a higher consumption level in both countries.
Table 2.3 shows how participants change their advised spending pattern if liquidity of retirement wealth or future health status change. Between 50 and 75% of the participants advise the same spending pattern as in the base treatment, with somewhat higher persistence among the Dutch than among the Australians. For columns 1 and 2, such persistence was expected, as vignettes 1-3 give the same total lifetime wealth. For vignettes 5-8, expecting a deterioration of health often implies lower advised spending.
Table 2.4 shows the average ranking for each of the ten saving motives by vignette and country, where the ‘most important’ motive gets value 5 and the ‘least important’ motive value 1. The importance of each motive is fairly
13Retirees were not supposed to participate in the survey experiment because we wanted to
consistent across vignettes. In both countries, the motives self-gratification and autonomy are often ranked as very important. The motives life-span risk and security are considered quite important in Australia, while intra-household
bequestand liquidity are more important for the Dutch. Intra-household bequest
is moderately important for both Australian and Dutch participants, while
intended bequestis not so important. Liquidity of retirement wealth only affects
the ranking of the saving motives in the case of an expected deterioration
in future health (cf. Table 2.4, columns [low, {H75, M75}], [high, {H75, M75}]
and [low, {M75, D75}], [high, {M75, D75}]). Expecting future health problems
increases the importance of the precautionary health motive, as one would expect.
Table 2.8 in the Appendix presents the distribution of the ranking per motive. It shows that intended bequest is considered the least important motive among
the five shown in a majority of all cases.14 Self-gratification is evaluated as the
most important motive in 39% of all cases. It might be seen as a general motive for saving that could be considered as the default. We therefore also conducted our main analysis after excluding self-gratification. This leads to results for the other motives that are very similar to those for the main analysis including
self-gratification(see Appendix Section 2.G).
2.5
Modelling the importance of each saving motive
We model the importance attached to each saving motive underlying the ad-vised spending patterns, focusing on the effects of the participant’s country of residence, liquidity of wealth, and future health expectations while controlling for other covariates. The importance of a given motive in a given vignette can be seen as an unobserved variable; we only observe its rank amongst five motives (a random subset of the ten motives in Table 2). We explain the ranking using a Rank-Ordered Logit model (Beggs et al. 1981). To be precise, the importance
14Since the hypothetical household owns the house they live in without a mortgage, an
of motive m (m = 1, . . . , 10) in vignette s (s = 1, . . . , 3, 5, . . . 8) for individual i
(i = 1, . . . , N) is given by:15
Uims = Vims + "ims
Uims consists of a systematic component Vims , and a random component "sim. We
assume that the "s
imare mutually independent and follow a type I extreme value
distribution. The systematic part Vs
imis determined by individual characteristics
Xi (see Table 2.13 in the Appendix), dummies for the observed advised
spend-ing patterns in each vignette (Ss
i), dummies (L
s) differentiating between the
retirement income policies in the vignettes (low, moderate or high liquidity)
and dummies (Hs) to capture the variation in health status across vignettes. In
addition, we include interactions S Ls
i of S
s
i and L
s to allow the effect of advised
spending pattern to vary with retirement income policy. Thus,
Vims = (β1,mLs+ β2,mSsi + β3,mS Lsi+ β4,mHs+ β5,mXi).
We use the precautionary motive (m = 1) as reference category, so β1,1=
· · · = β5,1= 0 and Vis1= 0.
The probability of a given ranking ys
i = (y
s
i1, . . . , yis5)0 where yi js is the
motive that is given rank j (given the set of motives and all vignette and individual characteristics), is now given by the product of four multinomial logit probabilities: P(Ui ys i1 > Ui yi2s > . . . > Ui yi5s) = 4 Y j=1 exp(Vi ys i j) P5 l=jVi ys il
In our quasi likelihood, we treat the rankings for different vignettes as inde-pendent. We cluster standard errors at the individual level to account for the correlation between rankings for different vignettes presented to the same participant.
15For notational convenience we do not include a subscript for the set of motives. Note that
