Microsimulation As A Decision Making Tool in Socia

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/254826860
Microsimulation as a Decision Making Tool In Social Security Policy
Article · May 2011
CITATIONS READS
6 410
1 author:
Jan-Maarten van Sonsbeek

CPB Netherlands Bureau for Economic Policy Analysis
20 PUBLICATIONS 148 CITATIONS
SEE PROFILE
All content following this page was uploaded by Jan-Maarten van Sonsbeek on 08 January 2016.
The user has requested enhancement of the downloaded file.

Microsimulation as a
Decision Making Tool
In Social Security Policy
De VU Public Controlling reeks is een uitgave van de postgraduate opleiding tot
controller in de publieke en non-profit sector van de Vrije Universiteit Amsterdam,
kamer 2A19, De Boelelaan 1105, 1081 HV Amsterdam.
© J.M. Van Sonsbeek, 2011.
Cover en lay-out: Gertjan van de Griendt, StudioGM2, The Hague

Print: Gildeprint, Enschede
ISBN/EAN: 9789461081537
NUR: 782
All rights reserved. No part of this book may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photo-
copying, recording, or otherwise, without the prior written consent of the publisher..
VRIJE UNIVERSITEIT
Microsimulation as a Decision Making Tool

In Social Security Policy
ACADEMISCH PROEFSCHRIFT
ter verkrijging van de graad Doctor aan

de Vrije Universiteit Amsterdam,
op gezag van de rector magnificus
prof.dr. L.M. Bouter,
in het openbaar te verdedigen
ten overstaan van de promotiecommissie
van de faculteit der Economische Wetenschappen en Bedrijfskunde
op woensdag 25 mei 2011 om 13.45 uur
in de aula van de universiteit,
De Boelelaan 1105
door
Johannes Martinus van Sonsbeek
geboren te Breda
promotoren: prof.dr. F.A.G. den Butter
prof.dr. R.H.J.M. Gradus
commissie: prof.dr. K.P. Goudswaard
prof.dr. Ph.R. de Jong
prof.dr. M. Lindeboom
prof.dr. J.H.M. Nelissen
dr. G.J.M. de Vries
paranimfen: dr. C.L.J.P. van Raalte

drs. W.S. Zwinkels
Preface
When I was still young I used to program my Commodore 64 computer for the sheer fun
of programming. One of my experiments involved forecasting the population of a country
by applying birth and mortality rates. I was fascinated by the nature of such a process, a
non-linear path to a future steady state. I also realized that the population was composed of
individuals and that the mortality rates could be applied by drawing random numbers for a
great number of individuals. I guess that was when the seed for my future interest in micro-
simulation modelling was planted.
This thesis is based upon studies conducted between 2002 and 2010 at the Ministry of Social
Affairs and Employment and VU University Amsterdam. I did not start developing microsim-
ulation models with the idea of later on writing a doctoral thesis on the subject. My job at the
Ministry involved the forecasting of disability benefit numbers and costs under different pol-
icy alternatives. As major reforms were initiated at that time, I realized that the microsimula-
tion approach could have an added value for this purpose. So I started to develop my first real
microsimulation model and one led to the other. Together with the director of my department
I wrote my first scientific paper and discovered what scientific research involves besides the
daily practice of work at the Ministry. It took until 2007 before I realized that writing a thesis
on the subject was within reach. And now, four years later, the project is finished. Sometimes
it was hard labour, but almost always I was inspired to continue, especially because I found
out that policy makers can learn a lot from scientists and vice versa. I enjoyed moving around
in these two worlds that all too often are far apart but that can and should be brought together
more often.
There are many people I would like to thank for their support in writing this thesis. In the first
place, I would like to express my sincere gratitude to my supervisors Frank den Butter and
Raymond Gradus for all their help and comments. Frank in particular broadened my view as
an economist and helped me place my research in a wider framework. Raymond got me into
this Ph.D. project by stimulating my scientific interests as a director of my department and is
currently professor at the Zijlstra institute that printed my thesis. I am also very grateful to the
subsequent directors and heads of my department at the Ministry, who allowed me to spend
part of my working hours on my thesis. I would like to thank two colleagues from the Minis-
try of Social Affairs and Employment, who also came to share the love for microsimulation
models, in particular: Marieke van der Werf who developed the first version of the SADNAP
model and Ridwan Alblas who greatly contributed to my final paper by preparing the data
and sharpening my econometric knowledge.
Furthermore I would like to express my gratitude to all the people who supported my re-
search through the years. The list of names would be too long to enumerate here, especially
since I would not want to forget any of them. It concerns in the first place many colleagues
at the Ministry of Social Affairs and Employment and at VU University Amsterdam. But
also many colleagues from other organizations both at home and abroad contributed to this
research. Besides, I would like to thank all the people who discussed my papers either at
various conferences and seminars through the years or as referees for some journals. All these
contributions greatly helped me to improve my research.
Finally, I wish to express my greatest thanks to my family and friends, especially to Astrid
and to my parents for their love and their continuous support. Without the support and friend-
ship of you all, this research could not have been performed.
Contents
Preface
Contents

1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.1 History of Economic Modelling in the Netherlands. . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Introduction to microsimulation models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3 Microsimulation models in the Netherlands and abroad. . . . . . . . . . . . . . . . . . . . . . 23
1.4 Introduction to ageing in the Netherlands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.1 Ageing in the Netherlands in numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.2 Retirement behaviour in the Netherlands. . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.4.3 Ageing in the Netherlands in international perspective . . . . . . . . . . . . . . . . 33
1.5 Introduction to disability insurance in the Netherlands. . . . . . . . . . . . . . . . . . . . . . 36
1.5.1 Disability insurance in the Netherlands. . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.5.2 Disability insurance in the Netherlands in international perspective . . . . . . 43
1.6 Joint analysis of ageing and disability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.6.1 The ageing-disability relationship in numbers . . . . . . . . . . . . . . . . . . . . . . . 45
1.6.2 Political context. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.7 Overview of the thesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2
A microsimulation analysis of the 2006 regime change in the Dutch disability scheme. 55
2.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2 Disability benefits in the Netherlands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.2.1 Current system of disability benefits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.2.2 Regime change proposed, from 2006 onwards. . . . . . . . . . . . . . . . . . . . . . . 60
2.2.3 Predictions of the regime change effects at the macro level. . . . . . . . . . . . . 62
2.3 The Microsimulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3.1 Microsimulation versus cell-based macro-forecasts. . . . . . . . . . . . . . . . . . . 63
2.3.2 Model choice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.3.3 Model structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.4 Implementation of the simulation model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.4.1 The base-data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.4.2 Imputation of missing data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.4.3 The simulation module. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4.4 The behavioural model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.4.5 Model validation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.5 Results of the simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.5.1 Macro results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.5.2 Micro results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
10
2.5.3 Sensitivity analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

2.5.4 Redistribution effects of the regime change. . . . . . . . . . . . . . . . . . . . . . . . . 91
2.6 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Appendix 2.A The behavioural model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Appendix 2.B The benefit formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3 Estimating the effects of recent disability reforms in the Netherlands. . . . . . . . . . . 103

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.2 The Dutch disability insurance schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.2.1 History and past policy measures until the end of the 1990s . . . . . . . . . . . 106
3.2.2 Recent policy changes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.2.3 Literature review. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.3 Disentangling the effects of inflow-related policy measures . . . . . . . . . . . . . . . . . 114
3.3.1 Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
3.3.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.3.3 Estimation results for inflow model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.3.4 Estimation results outflow model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.4 The microsimulation model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.4.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.4.2 Theoretical framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.4.3 Modelling inflow, outflow and transitions in grade of disability and
working status. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.4.4 Estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
3.5 The combined effect of the policy reforms in perspective. . . . . . . . . . . . . . . . . . . 142
3.5.1 Combined effect of all policy measures . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.5.2 Combined effect of outflow-related measures. . . . . . . . . . . . . . . . . . . . . . . 145
3.5.3 An integral long-term forecast of all disability schemes. . . . . . . . . . . . . . . 147
3.6 Conclusions and topics for future research. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Appendix 3.A Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
Appendix 3.B Parameter estimates for WAO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Appendix 3.C Parameter estimates for WIA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Appendix 3.D Parameter estimates for Wajong. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Appendix 3.E Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Appendix 3.F Detailed output of sector models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4 The Social Affairs Department of the Netherlands Ageing and Pensions model . . . 167
4.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.2 Background. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.2.1 The Dutch pension system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.2.2 Models currently in use. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
11
4.2.3 Microsimulation models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

4.2.4 The SADNAP model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.2.5 Comparison with other dynamic population microsimulation models. . . . 177
4.3 Modelling redistribution within the state pension system. . . . . . . . . . . . . . . . . . . . 178
4.4 Modelling the retirement decision of employees . . . . . . . . . . . . . . . . . . . . . . . . . . 182
4.5 Model results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.5.1 Budgetary results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.5.2 Redistribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.5.3 Retirement decision. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.5.4 Validation and comparison to other models . . . . . . . . . . . . . . . . . . . . . . . . 197
Appendix 4.A The SADNAP Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.A.1 The demographic model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.A.2 The household formation model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
4.A.3 The participation status model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Appendix 4.B Data sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.B.1 State pension data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.B.2 Company pension data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.B.3 Macro data sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
5 Microsimulations on the effects of ageing-related policy measures. . . . . . . . . . . . . . 221

5.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
5.2 The Dutch pension system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
5.2.1 Institutional setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
5.2.2 Current situation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
5.2.3 Macro forecasts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
5.2.4 Redistributive characteristics and actuarial fairness of the state
pension system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.3 Model and data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.3.2 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.4 Baseline scenario in a micro approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
5.4.1 State pension costs in the baseline scenario . . . . . . . . . . . . . . . . . . . . . . . . 231
5.4.2 Retirement age in the baseline scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
5.4.3 Redistribution and actuarial fairness in the baseline scenario. . . . . . . . . . . 233
5.5 Policy alternatives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
5.5.1 Overview of policy measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
5.5.2 Direct budgetary effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
5.5.3 Labour participation effects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
5.5.4 Redistributive effects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
12
5.5.5 Assessment of the policy alternatives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

Appendix 5.A Direct budgetary effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.1 Discussion on the added value of microsimulation models. . . . . . . . . . . . . . . . . . . . . . . 253
6.2 Conditions for successful integration of microsimulation models into policy analysis. . 259
6.3 Policy findings on ageing and disability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
6.4 Directions for future research. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
6.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
List of figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
List of tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
List of abbreviations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Samenvatting (Summary in Dutch). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
1
Introduction
This thesis is about microsimulation modelling and social security reform, in particular in the
fields of disability insurance and old age pensions. More specifically, the use of microsimula-
tion models as a tool for decision making in social security policy will be studied. The useful-
ness of microsimulation for such purposes is not to be taken for granted. On the one hand,
the outcomes of microsimulation models are known to provide richness in detail that gives
added value in forecasting and policy advice. On the other hand, microsimulation models are
notorious for their complexity and black-box nature, which limits their usefulness in a policy-
making context. Since 2000, two new microsimulation models have been developed at the
Dutch Ministry of Social Affairs and Employment for use in forecasting and policy advice,
one on disability benefit reform and one on pension reform. Both models and their results are
discussed extensively in this thesis. Based on the experiences with developing, maintaining
and using the models, and with disseminating the knowledge about these models, the central
research question answered by this thesis is: Should microsimulation models be integrated
into the policy analysis, and how? In order to answer this central research question, the added
value of microsimulation models as a forecasting and policy analysis tool in social security
policy will be investigated. The conditions that are required in order for microsimulation
models to have such added value will also be researched.
14 chapter 1
In this introductory chapter, section 1.1 gives a brief overview of the history of economic
modelling in the Netherlands. Section 1.2 contains a short history of microsimulation model-
ling and section 1.3 specifically focuses on microsimulation models in the Netherlands and
abroad. Sections 1.4 and 1.5 give a brief overview of ageing and disability insurance in the
Netherlands. Both historical and current developments are sketched in key numbers and
the Dutch situation is placed in an international perspective. In section 1.6 the relationship
between ageing and disability is explored and the recent reforms in both fields are placed in
a political context. Section 1.7 gives an overview of the chapters of this thesis and explains
how they relate to the research questions.
1.1 History of Economic Modelling in the Netherlands
The Netherlands has a long tradition of economic modelling, which is well documented (int.
al. Barten, 1986, Den Butter, 2010). Tinbergen pioneered economic models in the 1930s
and developed the first macroeconometric model of the Netherlands in 1936. Tinbergen also
became the first director of the Netherlands Bureau of Economic Policy Analysis (Centraal
Planbureau or CPB). CPB was founded in 1945 as an independent organization advising
government on economic policies. This task was explicitly assigned to CPB and not to Sta-
tistics Netherlands (Centraal Bureau voor de Statistiek or CBS), which was created in 1899
and was assigned the task of collecting independent and undisputed data for public use in the
Netherlands. In the early years of CPB there was an internal discussion on its role. Tinbergen
advocated making a clear distinction between the workings of the economy and the policy
goals, and proposed an advisory role for CPB. At the same time his colleague Van Cleeff, as
Tinbergen’s fellow 1969 Nobel Prize winner Frisch had done in Norway, proposed a more
normative role for economic modellers. In the Netherlands, Tinbergen won this battle. Con-
sequently, economic policy preparation in the Netherlands is organized in three autonomous
parts: data collection and statistics preparation by CBS; describing from a scientific perspec-
tive the workings of the economy through economic models by CPB; and political decision
Introduction 15
making by government in dialogue with unions, employers’ organizations and other associa-
tions of organised interest.
For a long time, model-based economic policy analysis was virtually a monopoly of CPB.
Den Butter (1991) distinguishes three generations of models based on economic theory in the
Netherlands. These economic theory models contrast with econometric time series models
that abstract from economic theory and developed into so-called vector autoregression (VAR)
models. The first generation of economic theory models was developed during the 1950s and
1960s. They were short-term demand driven models based on Keynesian analysis. They fitted
the economic situation of the 1950s and 1960s which was characterized by high economic
growth and full employment. The second generation was developed during the 1970s when
economic growth slowed down and unemployment became an issue. The Keynesian first-
generation models fell short at that time because their traditional recipe of higher government
spending to boost economic demand and activity caused higher unemployment and inflation.
Therefore, the second-generation models, like the VINTAF model of CPB, added the supply
side of the economy to the model using a vintage approach. According to these models, a rise
in real wages exceeding the rate of technical progress caused increased scrapping of capital
goods and therefore higher unemployment. The third generation of models was developed
during the 1980s. These models, like the FREIA model of CPB, added a full description of
the monetary sector to the models because of the increased importance of the crowding out
argument at that time. According to this mechanism, an increasing government deficit that is
financed on the capital market causes a higher long-term interest rate and therefore crowds
out private sector investments. In the Netherlands, these three generations of models played
an important and sometimes decisive role in the policy debate, resulting in a policy of wage
moderation and cutting government expenditure which, in retrospect, largely contributed to
the economic recovery of the Netherlands during the 1980s.
In 1976, the American economist Robert Lucas published an influential paper that changed
the thinking on economic models. Lucas argued that all empirical forecasting models suffer
16
from a common weakness. They are derived from observed relationships between various
macroeconomic quantities over time, but these relations differ depending on the macroeco-
nomic policy regime that is in place. As a consequence, one cannot predict the effects of a
new policy regime using an empirical forecasting model based on data from previous periods
with a different policy regime. Partly as a response to the Lucas critique, economists of the
1980s and 1990s, among them those at CPB, began to construct micro-founded macroeco-
nomic models based on rational choice, which have come to be called general equilibrium
(GE) models. Following the classification of Den Butter, these models could be considered
the fourth generation of economic theory models.
These general equilibrium models specify the agents that are active in the economy, such as
households and firms, as well as their preferences and constraints. Each agent is assumed to
make an optimal choice, taking into account prices and the strategies of other agents, both in
the current period and in the future. However, these models usually strongly simplify reality
by assuming that all agents of a given type are identical (a ‘representative agent’) and have
rational expectations. The two main types of GE models are dynamic stochastic general equi-
librium (DSGE) models and computable or applied general equilibrium (CGE/AGE) models,
which can be either static or dynamic. Typically, CGE models focus on long-run relation-
ships, making them more suited to studying the long-run impact of permanent policies like
the tax system or the social security system. DSGE models, for example, the one used at ECB
(Christoffel, Coenen and Warne, 2010), instead emphasize the dynamics of the economy over
time (often at a quarterly frequency), making them more suited for studying business cycles
and the cyclical effects of monetary and fiscal policy. The main Dutch general equilibrium
models, such as MIMIC and GAMMA (Don and Verbruggen, 2006), were developed at CPB.
MIMIC is a CGE model for the Dutch economy with a strong emphasis on the labour market.
GAMMA is an overlapping generations (OLG) model, a special class of CGE models which
is particularly fit for analysing ageing-related issues (Draper et al., 2005).
17
It should be noted that DSGE and CGE models need not be limited to representative agents
with rational expectations. Greater realism can be achieved by considering heterogeneous
agents with adaptive expectations. However, the more complicated these models become, the
harder they are to solve, even with the help of computers.
1.2 Introduction to microsimulation models
Microsimulation, on the other hand, is a modelling technique that operates at the level of in-
dividual units such as persons, households or firms. The concept of microsimulation was first
described by Orcutt (1957), an American macroeconomist. Orcutt recognized some funda-
mental limitations of existing macroeconom(etr)ic models. These included their rather limited
predictive usefulness and the fact that they predict only aggregates and not distributions.
Orcutt noted that there was an inherent difficulty in aggregating anything but the simplest re-
lationships about elemental decision-making units into comprehensible relationships between
large aggregative units. Only when relationships at the micro level are linear can aggregation
be done without loss of information. Given the fact that most relationships at the micro level
are nonlinear, aggregation by definition leads to loss of information.
Gilbert and Troitzsch (1999) place microsimulation models in a broader context of techniques
available for computer simulations. They classify these techniques based upon the number
of levels (one level meaning either the individual or the society, two levels meaning both in-
dividual and society and their interaction), the possibility of communication between agents,
the complexity of agents and the number of agents. In this classification, system dynamics
and queuing models are two techniques that operate at the system level. They are therefore
not capable of investigating emergent phenomena, complex systems and patterns arising out
of a multiplicity of relatively simple interactions. Microsimulation is related to a number of
other simulation techniques that operate at the individual level, most notably cellular au-
18 chapter 1
tomata and agent-based models. Cellular automata traditionally have a high number of agents
of low complexity that interact with each other. Microsimulation models traditionally have a
high number of agents of high complexity that do not interact with each other. Agent-based
models traditionally have a smaller number of agents of high complexity that do interact
with each other. However, it is currently recognized that microsimulation models add more
behavioural and spatial interaction between individual units, cellular automata add a grow-
ing range of individual attributes and start to incorporate spatial behaviours, and agent-based
models add both space and fiscal/demographic characteristics to their agents. Therefore, the
three approaches move towards a common ground (Williamson, 2007).
Microsimulation models are used in many different areas. Modelling of the distribution
of traffic flows over a street network is an example of an increasingly important use of the
approach. Other fields of use include health care (e.g. the spread of infectious diseases) and
defence (e.g. battlefield simulations). Of the broader range of simulation techniques, micro-
simulation is in particular fit for use in economics. Given the emphasis on changes in distri-
bution, microsimulation models are often used to investigate the impacts on social equity of
fiscal and demographic changes. Toder et al. (2000) distinguish four main model types that
can be used for analysing the effects of social security reforms, where traditional macroeco-
nomic models, like the ones described in section 1.1, fall short: general equilibrium models
that operate on the macro level; cell-based models that divide the population in relevant sub-
groups; representative agent models that operate on the individual level using a small number
of stylized agents; and microsimulation models that also operate on the individual level but
use a large number of realistic agents.
Dekkers and Belloni (2009) present a further classification of microsimulation models. Their
basic distinction equals Toder’s distinction between representative agent models and mi-
crosimulation models. The latter are first subdivided into static and dynamic models. Static
models can be used for distributional analysis of the immediate effects of tax-benefit poli-
cies. Dynamic models can also be used for long term analyses of dynamic processes like
Introduction 19
population ageing. The dynamic models are further subdivided into models with static and
dynamic ageing. In static ageing the characteristics of individuals do not change, but the
ageing is approximated by assigning different weights to records (e.g. when the number of
60 year olds increases the weight of records of individuals that are 60 years old is increased).
Dynamic ageing changes the characteristics of the individuals during the simulation (e.g.
an individual aged 60 turns 61 the next year and its characteristics, like the probability of
being unemployed or disabled, are adapted to those belonging to a 61 year old). Dynamic
ageing models can be subdivided in three ways; population models vs. cohort models, open
vs. closed models and cross-sectional ageing models vs. longitudinal ageing models. This
subdivision resembles those made by O’Donoghue (2001) and Spielauer (2002). Popula-
tion models start with a base population that is aged while new cohorts are added each time
period. Cohort models follow one cohort from the beginning to the end. Open models allow
individuals to enter the model (e.g. when someone marries), closed models limit the popula-
tion to the population of the model (once an individual marries, the partner will come from
inside the model population). An additional subdivision introduced by Dekkers and Belloni
(2009) is the explicit recognition of cross-sectional vs. longitudinal ageing and population
vs. cohort models. Cross-sectional ageing means that agents are all aged one year at the same
time, allowing them to interact while they are ageing. Longitudinal ageing means that agents
are aged one by one; each agent is aged until the end of its life (or the simulation). Typically,
but not necessarily, cross-sectional ageing is associated with population models and longitu-
dinal ageing with cohort models. The various classifications described above are summarized
in Figure 1.1.
20 chapter 1
Figure 1.1 Microsimulation models in perspective
Artificial intelligence Stochastic processes Micro econom(etr)ics Macro econom(etr)ics
Cellular Agent-based Simulation Economic Time series

Autiomata models models theory models models
Representative Microsimulation Cell-based General G1: cyclical Vector

agent models models (MSM) models equilibrium G2: structural Autoregression
models (GE) G3: monetary models (VAR)
Computable
Static MSM Dynamic MSM (applied) GE
models (CGE)
Static ageing Dynamic ageing Static CGE Dynamic CGE Dynamic

MSM MSM models models Stochastic GE
models (DSGE)
The most prominently differentiated products from the modelling industry are forecasts and
analyses of the effects of policy measures (Den Butter and Morgan, 1998). In the case of
forecasts and analyses for social security reform, models need as a minimum to have a certain
level of differentiation in order to identify relevant subgroups in a population. The shaded
boxes in Figure 1.1 highlight the four model types that meet those minimum criteria. Toder
et al. (2000) developed an assessment framework for evaluating these model types. They
find that a wide array of model characteristics should be considered in such a framework,
which span two major areas: the model’s substantive content and its technical features (like
run time, language etc.). The substantive content consists of major areas like the capacity
(programmatic detail), the outcomes, the population definition, the time horizon and the
integration of social economic research on social security parameters. Assessment criteria
include representativeness, ability to investigate small, targeted groups, theoretical and em-
pirical validity, outcomes capability, programmatic detail, ease of use and development and
maintenance time and cost. All assessment criteria and Toder’s scoring of the model types are
depicted in Table 1.1.
Introduction 21
Table 1.1 Assessment framework for microsimulation models
Assessment criterion Cell-based model MSM model CGE model

representativeness Variable Very high Stylized
cohort size n/a Large Variable
generalizability Limited High Limited
ability investigate targeted groups None Easy None
theoretical and empirical validity Low Variable High
autocorrelation n/a Variable Little
feedbacks Sometimes Variable Extensive
outcomes capability Limited Very broad Limited
programmatic detail Variable Very high Low
projection assumptions Variable Explicit rules Equilibrium
sensitivity analyses capability Variable Very high Possible
stochastic replications n/a Monte Carlo n/a
ease of use Easy Difficult Variable
accessibility Variable Variable Variable
development and maintenance time/cost Variable Substantial Variable
Source: Toder et al. (2000)
When considering Toder’s framework, cell-based models tend to be superior to both micro-
simulation models and CGE models in ease of use and development and maintenance time
and cost. Microsimulation models are superior to CGE models and cell-based models in
representativeness, cohort size, generalizability, ability to investigate targeted groups, pro-
grammatic detail, outcomes capability and sensitivity analyses capability. This is not surpris-
ing given the nature of microsimulation models which follow large cohorts of heterogeneous
individuals. However, there is a price to pay. CGE models are superior to microsimulation
models in theoretical and empirical validity and feedbacks. Not surprisingly, a body of re-
search is developing which attempts to combine the strengths of microsimulation models and
macro general equilibrium models. This so called micro-macro model, which uses outputs
from microsimulation models as inputs to macro general equilibrium models and vice versa,
is considered one of the three main types of microsimulation models. The other two are the
22 chapter 1
static cross-sectional tax-benefit model and the dynamic model in which population charac-
teristics are projected forwards through a process of regular updating of the characteristics of
the individuals within the model (Williamson, Zaidi and Harding, 2009).
The micro-macro model is indeed one of the most promising approaches in microsimulation
and is certainly in line with Guy Orcutt’s original vision. In theory, an integrated approach
can unify the strengths of microsimulation models and general equilibrium models while
limiting the weaknesses of both. This observation is in line with the large body of literature
on improving forecasts by combining different forecasting methods. For an overview of this
literature see Clemen (1989). However, the number of integrated models is still very small
worldwide. Recent examples of concrete applications include Arntz, Boeters and Güertzgen
(2008), Clauss and Schubert (2009) and Peichl and Schaefer (2009). Also, it appears that the
combination of microsimulation models and macro models is relatively widespread in analy-
ses of developing economies. Here the benefits are twofold; microsimulation models provide
important redistributional information and general equilibrium models tend to fall short on
their own because they require a stable steady state which developing economies by defini-
tion do not have. Examples of these analyses are summarized in Davies (2009) and include
combination models for Indonesia, the Philippines, Madagascar and South Africa.
The main reasons to link microsimulation and macro general equilibrium models are three-
fold, according to Anderson (1990): the outputs from a macro model can be used to align
the predictions of the microsimulation model; to enable general equilibrium feedbacks and
interactions among variables in the micro model; and to provide a microeconomic basis for
aggregate behaviour. Following these lines, Baekgaard (1995) identifies four methods for link-
ing micro and macro models. In the top-down approach, the micro model is adjusted to match
an exogenous macro aggregate. In the bottom-up approach, a change generated in the micro
model is used to adjust the macro model. In the recursive linkage or top-down bottom-up ap-
proach, there is interaction both ways between the models. This approach can be extended into
the iterative approach, in which the two models are solved simultaneously within each period.
Introduction 23
Davies (2004) adds to this the integrated approach in which the microsimulation model and
the CGE model are fully integrated in contrast to the other ‘layered’ approaches. In an inte-
grated model, in fact the standard representative households that are present in a CGE are re-
placed with real households from the micro data source. These linkage methods are discussed
and evaluated by Colombo (2010). She finds the top-down bottom-up approach to be the most
complete approach, as on the one hand it can include all the microeconometric estimates re-
quired to account for behavioural responses by individual agents, and on the other hand it can
take into account the feedback effects from the micro to the macro level of analysis.
1.3 Microsimulation models in the Netherlands and abroad
In spite of the longstanding tradition of economic modelling, the use of microsimulation

models in economics never really took off in the Netherlands. The microsimulation efforts
that were done were mostly limited to static microsimulation models. At CPB three different
static microsimulation models were in use during the 1990s which have now been replaced
by one model called MIMOSI (Romijn et al., 2008). At the Ministry of Social Affairs and
Employment, a static microsimulation model called Micros has been in operation since the
early 1990s (see Hendrix, 1993). Both models are extensively used for analysing the distri-
butional effects of tax-benefit policy measures for the yearly budget and for election pro-
grammes of political parties. On dynamic models, only few efforts have taken place, most
notably the dynamic microsimulation model, Nedymas, which was developed in the 1990s at
Tilburg University (Nelissen, 1991, 1993, 1995a, 1995b).
However, large dynamic population microsimulation models have never been adopted in the
Netherlands. Given the fact that since Tinbergen the Netherlands has been at the forefront
of economic modelling, this is rather remarkable. Don and Verbruggen (2006) describe how
CPB researched three different and competing modelling approaches in the second half of the
1980s: a microsimulation model focussing on empirical labour supply behaviour at the level
24 chapter 1
of individual households, a macro model featuring a substantial disaggregation of the labour

market, and a CGE model providing an explicit description of labour supply decisions as the
result of utility. The latter (which developed into the MIMIC model) won because it had bet-
ter potential, in particular on labour supply modelling, for studying alternative tax and social
security regimes.
However, in some other countries, large dynamic microsimulation models have been de-
veloped. Examples of countries with such models are Australia (Dynamod, University of
Canberra), Canada (Lifepaths, Statistics Canada), France (Destinie, French National Statistics
Institute), Norway (Mosart model, Statistics Norway), Sweden (Sesim, Swedish Ministry of
Finance), the UK (Sage, London School of Economics) and the USA (Corsim, Strategic Fore-
casting and Dynasim, Urban Institute). There are various factors that may explain the choice
of developing a large scale dynamic microsimulation model or not. These include practical
reasons like the investment required for developing such a model and expectations on the re-
turn on investment in terms of added value in forecasting and policy evaluation. Developing
a comprehensive dynamic population microsimulation model is indeed a demanding, time-
consuming and complicated task (Cassels, Harding and Kelly, 2006), with heavy demands
on the availability of micro data sources and, especially in the past, on computer time and
space. Institutional considerations may also play a role, like the structure and organization of
the model building industry (Den Butter and Morgan, 1998). Large differences exist between
countries in the structure of their model building industries. The main differences are between
a monopolistic and a competitive market structure and within a monopolistic market structure
in the limitations and focus of the market, like the positioning of the modelling organizations
(as part of or apart from the data collecting organization) and the main product focus of the
modelling industry (forecasting or policy analysis).
In order to assess the influence of the institutional context on the use of microsimulation
models we can use the overviews of the major dynamic microsimulation models that are in
operation around the world. Such overviews are given in several recent studies. Spielauer
Introduction 25
(2007) investigates 27 dynamic microsimulation models for which documentation is avail-

able. Dekkers and Belloni (2009) list 43 dynamic microsimulation models worldwide, 6
of which are known to be no longer in active use. Li (2010) provides the most extensive
overview with 48 dynamic microsimulation models. In the listing of Li, 8 models come from
the USA, 6 from Italy and the UK, 5 from Canada and Sweden and 4 from Australia. In the
listing of Dekkers and Belloni, 9 models come from the USA, 6 from Italy and the UK, 4
from Sweden and 3 from Canada. In the listing of Spielauer, 6 models come from the USA,
4 from Sweden, 3 from Australia and Canada and 2 from Italy and the UK. In all three, the
same six countries stand out in number of dynamic microsimulation models. They all share a
competitive market structure in which ministries, research institutes, universities and private
institutes participate.
However, microsimulation also flourishes in Norway, although Norway is, like the Neth-
erlands, an example of a country with a monopolistic market structure1, albeit different in
organization and product focus of the market. According to Den Butter and Morgan (1998),
these differences go back to the founding fathers of the economic modelling and the respec-
tive modelling agencies in the Netherlands and Norway; Tinbergen and Frisch, the first Nobel
Prize winners in economics. In the Netherlands two separate (monopolistic) organizations
are involved in data collection and economic analyses. CBS is involved in “collecting and
processing data in order to publish statistics to be used in practice, by policymakers and for
scientific research”. CPB is involved in “making independent economic analyses that are
both scientifically sound and up-to-date, and relevant for policymaking in the Netherlands”.
In Norway one organization (Statistics Norway) is involved in both data collection and eco-
nomic analyses. The placement in one organization of data collection and economic analyses
contributes to the actual use of those administrative datasets for economic policy analysis. An
additional factor that may have contributed to microsimulation flourishing in Norway (and
other Scandinavian countries) is the availability to researchers of administrative datasets over
a long time period, due to relaxed privacy laws.
1 It is worth noting that these are not formal monopolies on economic policy advice. However, in the Netherlands
CPB has a de facto monopoly in model-based economic advice because of the costs involved in investing in specific
knowledge and building up of a reputation in the institutions of policy preparation in the Netherlands (Den Butter,
2010).
26 chapter 1
On the other hand, no large dynamic microsimulation model is currently in operation in

Germany, which has a very competitive market structure with a large number of economic
research institutes. Indeed, Germany was among the frontrunners in microsimulation during
the 1980s. But none of the four main economic research institutes, Ifo (Munich), IfW (Kiel),
IWH (Halle) and RWI (Essen), which participate in the twice yearly joint analysis of the state
of the German and world economy, have developed a dynamic microsimulation model. The
main restricting factor in Germany is the availability of micro data sources. Possibly because
of German history, privacy regulations are very tight. Institutes that have administrative
data sources are not distributing those sources to others and because the administrative data
sources lack a common identity, individual matching of data is not possible anyway. Micro-
simulation efforts in Germany therefore are still limited to a number of smaller scale, mainly
static, models. For an overview, see Wagenhals (2004).
1.4 Introduction to ageing in the Netherlands
Population ageing is a shift in the distribution of a country’s population towards older ages.
This is usually reflected in an increase in the population’s mean age, a decline in the propor-
tion of the population composed of children, and a rise in the proportion of the population
that is elderly. Population ageing arises from two (possibly related) demographic effects;
increasing longevity and declining fertility.
1.4.1 Ageing in the Netherlands in numbers
A first impression of ageing in the Netherlands can be obtained from Figure 1.2, which
gives the (absolute) number of births per year and the average number of children per fertile
woman in the Netherlands since 1900.
Introduction 27
Figure 1.2 Number of births per cohort and average number of children per woman,
1900-2008
Source: CBS, Statline
Until 1945 we see a combination of a falling number of children per woman and a rather sta-
ble absolute number of births per year, which is a logical consequence of the increase of the
population in general. In 1946 the number of births skyrockets to an all-time high of 280,000.
This is what was later to be called the baby boom generation. Thereafter, the number of
births decreases but remains on a very high level of around 250,000 per year until 1970. The
number of births then starts falling, due to the increased use of contraceptives, to a new stable
level of just below 200,000 per year. The average number of children is 1.7, which is below
the replacement ratio of 2.1. This means that the population of the Netherlands, without im-
migration and emigration, is bound to fall. The fact that this has not happened is due to high
immigration levels from the 1970s to the 1990s.
28 chapter 1
Figures 1.3 and 1.4 highlight the development of life expectancy over time. The general
picture of life expectancy at birth is given in Figure 1.3 from 1860 onwards. A more specific
picture of the development of life expectancy at 65 is given in Figure 1.4 from 1950 onwards.
Figure 1.3 Life expectancy at birth, 1861-2006
The general picture reveals a steady rise of life expectancy at birth from less than 40 years
between 1860 and 1870 to almost 80 years nowadays. The dips in 1916–1921 and 1941–1946
are logical consequences of the Spanish flu and the Second World War respectively.
Introduction 29
Figure 1.4 Life expectancy at 65, 1950–2050
Source: CBS, Statline (projections based on CBS population forecast 2008–2050)
Figure 1.4 shows first the divergence and later on the convergence of life expectancy (at 65
years) of men and women. In 1950 both men and women had a remaining life expectancy of
15 years when turning 65. The remaining life expectancy of women started increasing rapidly
until 1985, whereas the remaining life expectancy of men stayed the same. The divergence
can be explained mainly by smoking habits. Partly due to decreasing smoking rates among
men and increasing smoking rates (in the past) among women, the remaining life expectan-
cies of both are now converging so that it is expected that around 2050 the difference will be
less than two years.
Now that the two main factors driving population ageing have been discussed, we may have a
look at Figure 1.5, showing the composition of the Dutch population by age in 2009.
30 chapter 1
Figure 1.5 Population by age, 2009
Source: CBS, Statline (projections based on CBS population forecast 2008-2050)
From this figure, we can clearly see the baby boom birth cohort from 1946, who were aged
63 in 2009 and will enter the state pension scheme in 2011 when they turn 65. We can also
clearly see the last big birth cohorts from around 1970, who are aged around 40 years now
and will turn 65 around 2035. Consequently, between 2011 and 2035 large cohorts of pen-
sioners will enter the state pension scheme (assuming unchanged policies). On the left-hand
side of the figure, the large cohorts of pensioners are replaced by much smaller cohorts of
young people entering the labour market. How this development works out, can be deducted
from Figure 1.6 which shows the development of two main population composition indica-
tors; the grey pressure (the number of over 65 year olds over the number of 20–64 year olds)
and the green pressure (the number of 0–19 year olds over the number of 20–64 year olds).
Total pressure is defined in this figure as grey pressure plus green pressure.
Introduction 31
Figure 1.6 Population indicators, 1950–2050
Source: CBS, Statline (projections based on CBS population forecast 2008-2050)
Grey pressure rises from around 14% in 1950 to around 25% in 2009 and to a maximum of
almost 50% in 2038. The total of green and grey pressure rises to 90% in 2038. However, this
is not an all-time high as the total pressure topped 90% around 1965 as well, at that time due
to the accumulation of the large birth cohorts from 1946–1965 causing a high value of green
pressure.
Population ageing has a large impact on public finance. This is most easily understood by
having a look at the common lifetime contribution profile of an individual. Such analyses
have been made for the Netherlands by Ter Rele (1998) and Bovenberg and Ter Rele (1999),
among others. Typically they show people to be net payers between the ages of 20 and 65
and net consumers before and thereafter. However, the cost peaks are concentrated at old
age when health costs and pension costs accumulate. Therefore the peak of green pressure in
1965 cannot be compared (in cost terms) with the peak of grey pressure in 2038.
32 chapter 1
1.4.2 Retirement behaviour in the Netherlands
As a consequence of rising wealth of the Dutch population, preferences for leisure have
increased over time. This has manifested itself in a decrease of both the average number of
working hours per week and the average retirement age. On top of that came rising unem-
ployment during the 1970s, which led governments to promote early retirement in order to
increase the number of jobs available to the young. Figure 1.7 shows how the effective retire-
ment age has developed in the Netherlands since 1970.
Figure 1.7 Effective retirement age, 1970–2007
Source: OECD estimates based on the results of national labour force surveys and the EU Labour Force Survey
The effective retirement age decreased from 67 in 1970 to around 60 for men from 1986 on-
wards and even lower for women. Only since around 2003 has effective retirement age been
increasing again. This has been mainly as a result of reforms of the extremely generous early
retirement schemes that became too expensive to sustain and, because of the notion of the
lump of labour fallacy, were revealed to be ineffective instruments for promoting the employ-
ment of the young.
Introduction 33
1.4.3 Ageing in the Netherlands in international perspective
During the last decade many international comparative studies on ageing of the popula-
tion have been done, for example; the European Commission (2007), the OECD (2009) and
Whitehouse (2008). The main picture resulting from these studies is that the ageing problem
in the Netherlands is substantial but about average when compared with other EU or OECD
countries. In the OECD countries ageing is causing the heaviest burden on public finance in
Japan and Korea. Turkey and the United States have the smallest problem. Figure 1.8 gives
the grey pressure (the number of over 65 year olds over the number of 20–64 year olds) for
Japan, the United States and the Netherlands. Unlike in Japan, where the old-age dependency
ratio continues to increase until 2050, population ageing in the Netherlands peaks in 2038.
After 2038 the grey pressure slowly decreases to stabilize at about 40%.
Figure 1.8 Grey pressure in selected OECD countries, 2000–2050
Source: OECD, Pensions at a Glance (OECD, 2009)

34 chapter 1
Figure 1.9 gives the fertility rates (number of children per woman) for OECD member states
and reveals one major underlying cause for population ageing. Korea and Japan are among
the countries with the lowest fertility rates. In eastern European and Mediterranean countries,
the fertility rates also tend to be very low, at below 1.5, where 2.1 is the required replacement
level for a stable population. Such high fertility rates are only common in Turkey, Mexico
and the United States. Among European countries, Scandinavian countries, with their exten-
sive child support policies, tend to have the highest fertility rates.
Figure 1.9 Fertility rates by OECD country, 2006
Figure 1.10 highlights the other underlying cause and gives the life expectancy at birth in the
OECD member states.
Introduction 35
Figure 1.10 Life expectancy at birth by OECD country, 2006
Again we see clear differences, Japan having the highest life expectancy, two years above the
OECD average, the United States having a life expectancy of 78 years, two years below the
OECD average. The Netherlands has a life expectancy of 80 years, only slightly above the
OECD average.
Ageing causes governments all over the EU and the OECD to reform their pension systems. An
overview of recent pension reforms is given in Whiteford and Whitehouse (2006) and Martin
and Whitehouse (2008). In the latter study, pension reforms are subdivided into two main cat-
egories; changing the pension parameters and changing the pension paradigm. Changes of the
pension parameters include the pension age, the retirement incentives, the calculation meas-
ures and the indexation practice. Changes of the pension paradigm include transformations of
defined benefit (DB) schemes to defined contribution (DC) or notional defined contribution
(NDC) schemes and transformations from pay-as-you-go (PAYG) schemes to funded schemes.
Also the linking of the pension age to the development of life expectancy falls in this category.
36 chapter 1
1.5 Introduction to disability insurance in the Netherlands
The development of European social security systems is well documented (int.al. Barr, 1992,
Den Butter and Kock, 2003) and is generally subdivided into three stages, starting with char-
ity as the main source of social protection. In the second stage, social insurance schemes are
introduced to cover the risk of old age, occupational disability and (later on in the twenti-
eth century) also unemployment. In the third stage, countries expand social protection to
cover almost all aspects of occupational and private life. The Netherlands developed along
the Bismarckian path of insurance-based social security. Typically, in such a Bismarckian
system, disability, unemployment and pension schemes are insurances against loss of income,
with benefits that are related to the previous wage, eligibility depending on contribution and
financing through premiums. Den Butter and Kock develop an assessment framework for
social security systems consisting of coverage and scope of the programme, the gatekeeper
function and the incentives of the system. The former two control the inflow into the system
and the latter also controls the outflow out of the system. A targeted policy, such as disability
or unemployment insurance, needs a proper design in order to be sustainable. Such a design
consists of clear entitlement criteria, a proper enforcement of those entitlement criteria and
incentives to discourage misuse of the system. As we will see, although developed with good
intentions, disability insurance in the Netherlands was a prime example of a poorly designed
social security scheme that therefore became unsustainable.
1.5.1 Disability insurance in the Netherlands
The development of the social security system of the Netherlands started in the beginning of
the twentieth century (int.al. Berghman et al., 2003, Goudswaard, de Kam and Sterks, 2000).
In 1901, a work injury scheme was introduced in the Netherlands (Ongevallenwet). It was
the first social insurance law in the Netherlands and covered work injuries and occupational
diseases. In 1913, a generic employees’ insurance covering other causes of disability was
Introduction 37
added (Invaliditeitswet). In 1919, voluntary old-age insurance was introduced. After the
Second World War those schemes were extended to cover the general population. In 1957 the
voluntary old-age pension insurance was extended to a general state pension law called AOW
(Algemene Ouderdoms Wet). In 1967 the two disability schemes were replaced by a new dis-
ability scheme covering all disability contingencies, regardless of being work-related or not.
This law was called WAO (Wet op de Arbeidsongeschiktheidsverzekering) and existed until
2006 when it was replaced by a new disability scheme called WIA (Wet Werk en Inkomen
naar Arbeidsvermogen). In 1976 the self-employed and the young handicapped (having a
disability at the age of 18, before entering the labour market) were also covered in a separate
scheme called AAW (Algemene Arbeidsongeschiktheidswet). In 1998 this scheme was split
into its two components, the WAZ (Wet Arbeidsongeschiktheid Zelfstandigen) covering the
self-employed and the Wajong (Wet Arbeidsongeschiktheid Jonggehandicapten) covering the
young handicapped. In 2004, the WAZ scheme was abolished for new entrants because the
then government decided, in cooperation with employers’ associations, that the self-employed
could just as well buy disability insurance on the private market. Extensive documentation
of the history of the disability schemes is given in Aarts, de Jong and van der Veen (2002). In
the remainder of this section we have a closer look at the key numbers involved.
In Figure 1.11, we first have a look at the development of the stock of disability benefit
recipients since the introduction of the WAO in July 1967. The number of beneficiaries grew
almost continuously until an initial peak of over 900,000 in 1993. As the working popula-
tion at that time consisted of 6 million persons, the disability rate was extremely high and the
cost of the disability benefits was causing a heavy burden on the Dutch public finances. After
political turmoil and a first major change of the disability law, the number of disability benefit
recipients decreased for a short while, but in the second half of the 1990s it started increasing
again, until around 2002 it reached almost 1 million.
38 chapter 1
Figure 1.11 Stock of disability benefit recipients, 1968–2009
Source: UWV, Statistische Tijdreeksen UWV 2008
Since 2002, fuelled by a series of reforms of the disability schemes, the number of beneficiar-
ies has been decreasing, down to its current level of just over 800,000. Figure 1.12 shows the
number of new benefits (inflow) and terminated benefits (outflow) per year since the intro-
duction of the WAO.
Introduction 39
Figure 1.12 Inflow and outflow of disability benefits (WAO/WIA schemes), 1969–2009
Until 1994, the yearly inflow exceeded the yearly outflow, causing a rising number of
benefits. In particular, from 1975 onwards, the yearly inflow rose very quickly to between
80,000 and 100,000 per year until 1993. The decrease of yearly inflow that followed in 1994
and 1995, caused by a tightening of the entrance criteria, was only short-lived. Yearly inflow
started rising again to about 100,000 per year around the turn of the century. Yearly outflow
took place at a more or less constant rate with absolute numbers rising at the same rate as the
disability stock.
From this perspective, the decrease of yearly inflow since 2002 is remarkable. Current inflow
levels are at an all-time low of around 35,000 per year. Consequently yearly outflow has also
decreased, but the stock of WAO and WIA beneficiaries is still shrinking. However, the op-
posite is happening in the Wajong scheme. This scheme for the young handicapped started in
1976, as part of the general disability scheme called AAW. For two decades the annual inflow
40 chapter 1
of young handicapped in this scheme was at a stable level of around 4,000 per year. Since
1997 the yearly inflow has been rising. First, this happened mainly because of administra-
tive corrections (1998, 2000) and because of an increase of the insured population from 1998
onwards (students who became disabled past the age of 18 but while engaged in study). Since
2004 the annual inflow has risen spectacularly to a current level of around 17,000 a year. One
could expect that when inflow rises at such a fast pace, the average new beneficiary is less
severely disabled and therefore has a better chance of recovery than in the past. Remarkably
though, annual outflow has fallen in recent years.
Figure 1.13 Inflow and outflow of disability benefits (Wajong scheme), 1977–2009
A more detailed picture of the annual inflow into the WAO (and WIA) scheme shows a
remarkable development of the inflow rate of women as opposed to men. This development is
depicted in Figure 1.14.
Introduction 41
Figure 1.14 Inflow rates WAO/WIA by gender, 1967–2009
Until 1985 men and women entered the WAO at similar rates. From 1985 onwards, the inflow
rate of women rose rapidly until reaching a level, around the turn of the century, about twice
as high as the inflow rate of men. Meanwhile, as entrance criteria have tightened, the inflow
rate of women has decreased to a level only slightly above that of men. Although during the
years of high female inflow rates a great number of research studies was conducted on the
subject (e.g. Bronswijk et al., 2003), no clear causal relationships could be found. Multiple
causes were thought to be involved, like health-related gender differences, differences in mo-
tivation for participation on the labour market, differences in working conditions, the more
passive role of women who are on long-term sick leave and the different attitude of medical
officers involved in reintegrating sick employees.
Moreover, it was not only the share of women in the disability population that rose; the aver-
age age of the new female beneficiaries was much lower than the average age of the males.
42 chapter 1
Figure 1.15 shows the development of the disability stock by age category.
Figure 1.15 WAO, WIA and Wajong beneficiaries by age category, 1968–2009
One could say that in the early years of the WAO, the average new beneficiary was an older
man, with mostly work-related physical problems. During the economic crisis of the late
1970s and early 1980s younger men with less severe physical problems started entering the
scheme. During the 1990s the typical new beneficiary was a younger woman, often with a
diagnosis pertaining to mental health. Now that the inflow into the WAO and WIA schemes
has returned to normal proportions, we see a new group of typical new beneficiaries coming
up: the young handicapped with less severe mental problems who enter the Wajong scheme at
an early age because they apparently cannot properly enter the labour market.
Returning to the assessment criteria for social security systems as formulated by Den But-
ter and Kock (2003), when considering the former WAO scheme, we can conclude that the
system provided extensive coverage, a poor gatekeeper function, generous benefits and poor
Introduction 43
incentives for both employers and employees to prevent inflow or stimulate outflow. Conse-
quently, the number of beneficiaries spiralled out of control. As De Jong (2008) concludes, the
WAO scheme was misused to accommodate social change. It was used first to provide a social
solution for the massive lay-offs of the 1970s and 1980s, when many industries were restruc-
tured, and second, to accommodate the change from a traditional breadwinner society to a
dual-earners model. One could wonder whether the high inflow rates of young handicapped
point to another social change that the Wajong scheme is being misused to accommodate.
1.5.2 Disability insurance in the Netherlands in international perspective
The Netherlands used to have one of the highest rates of disability benefit recipiency in the
world. An extensive overview of international disability schemes is given in OECD (2003).
In 1999, on spending on all disability-related programmes, the Netherlands take a second
place (after Sweden), spending 4.64% of GDP, which is almost twice the OECD average. The
Netherlands is also among the three countries that stand out as having far higher than average
inflow rates (together with Norway and the UK). The OECD concludes, among other things,
that gender differences are usually small but that women tend to report higher rates of dis-
ability, especially in the Netherlands and the Scandinavian countries. Also the population of
disability benefit recipients in the Netherlands is comparatively young and part-time employ-
ment is relatively rare among them.
The Dutch reforms that have taken place since 1998 have been internationally assessed
and recognized (e.g. OECD 2008). The OECD describes how inflow into disability benefit
dropped remarkably, from almost 1.2% in 2001 to 0.4% in 2007 and concludes that this suc-
cess is a consequence of a series of very comprehensive reforms, characterized by a shift of
responsibilities to employers and employees, a tightening in benefit eligibility and generosity,
and a (partial) privatization of hitherto public schemes. The reforms may fit into a broader
trend of convergence of European welfare states (see Caminada, Goudswaard and Van Vliet,
44 chapter 1
2010). For incapacity-related benefits they find a converging trend within the EU to a lower
level. This is consistent with the OECD (2003), which finds a trend among almost all its
members to expand integration policies in disability programmes and contract compensation
policies.
1.6 Joint analysis of ageing and disability
There is a strong relationship between the ageing of the population and the disability insur-
ance laws. This relationship consists of several components, some of them well studied and
some of them not. In the first place, ageing of the working population increases the number of
elderly workers with a high risk of disability enrolment. This is a problem most OECD coun-
tries face and it is recognized in the literature, as in Autor and Duggan (2006) for the USA.
On top of that, disability and ageing are related in a second way, because disability benefit
schemes can be misused as an early retirement route. Literature on this relationship includes
Lindeboom (1998) and Kapteyn and de Vos (1999), both of whom find a significant use of
the disability scheme as an early retirement pathway. In the third place, the ageing of the
population and the increasing grey pressure require policies to increase the labour force par-
ticipation. Labour force participation in the Netherlands is relatively low among the elderly
and among women, so the largest participation gains can be reached among those groups.
However, both the elderly and women used to have very high disability inflow probabilities
so that policies increasing the labour force participation of women and elderly could backfire
through the disability scheme. In the fourth place, as the large baby boom cohorts start to
retire from 2011 onwards, the disability stock will also decrease as outflow through transfer
of recipients to retirement pension will, at least temporarily, increase.
Introduction 45
1.6.1 The ageing-disability relationship in numbers
For the Netherlands the relationship between ageing and disability can be investigated by
considering some recent figures on both. Figure 1.16 first gives the composition of the work-
ing population by age and main occupation: studying, working, on benefit (mainly disability,
unemployment and social assistance), non-participating and early retired. From this figure the
problem of the Netherlands with both ageing and disability becomes clear. At the age of 64
only a small minority of the population is still working. The majority is either early retired or
on benefit. Of the 11% that is still working at 64 the large majority are men. Among women
labour participation until 65 is almost non-existent. Of the 27% that are on benefit, the large
majority (about two-thirds) are on disability benefit.
Figure 1.16 Working population by age and occupation, 2005

46 chapter 1
The largest group of 64 year olds are the early retired (39%). This underlines the risks of in-
creased benefit use when the early retirement schemes are reformed and labour participation
of the elderly grows. Figures 1.17 and 1.18 show how labour participation rates of the elderly
and of women are expected to grow in the coming decades.
Figure 1.17 Labour participation rate males by age category, 2000 vs. 2050
Source: CPB, projections based on Euwals and Folmer (2009)

Introduction 47
Figure 1.18 Labour participation rate of females by age category, 2000 vs. 2050
Source: CPB, projections based on Euwals and Folmer (2009)
In particular, the rise of labour participation rates among women aged 50 and over and among
men aged 60 and over is spectacular. Figures 1.19 and 1.20 show male and female disability
rates by age in 2000 and 2009.
48 chapter 1
Figure 1.19 Disability rates of males by age, 2000 vs. 2009
Source: CBS Statline, UWV microdata
Figure 1.20 Disability rates of females by age, 2000 vs. 2009
Source: CBS Statline, UWV microdata

Introduction 49
The figures show the danger of rising labour participation of women and the elderly if the
disability inflow rates of 2000 had persisted, and how much matters have improved in this
area in recent years. Note that in 2000 the disability rate among older women was still com-
paratively low because the high female inflow rates of the 1990s mainly concerned relatively
younger women.
Figure 1.21 finally highlights the composition of the current disability benefit stock by age.
This figure reveals that population ageing from 2011 onwards will decrease pressure on the
disability benefit scheme as large cohorts of disability benefit recipients will retire.
Figure 1.21 Population structure of disability beneficiary stock, 2000 and 2009
Source: UWV microdata
It can be concluded from the above analysis that a high rate of disability benefit recipiency
and an ageing population can be a potentially dangerous combination. However, it can also
be argued that apparently, judging from recent numbers, the Netherlands have been success-
ful in overcoming this problematic relationship. The use of disability as an early retirement
50 chapter 1
pathway has decreased and as the entrance to disability schemes is tightened the risk of draw-
ing larger groups of elderly and women to the labour market is decreasing as well. However,
it remains to be seen whether the reformed disability regime can stand the increasing pressure
which will arise when more elderly participate and alternative exit routes like early retirement
are less attractive than they were in the past.
1.6.2 Political context
The political context of both subjects is also interesting. Both ageing and disability policy
have a long history of political sensitivity. It takes a lot of political courage for governments
to attack problems related to both topics and, because of this, some reform proposals had a
serious backlash. When the number of disability benefit recipients approached 1 million in
the early 1990s, then Prime Minister Lubbers linked his continuation in office to successful
reform. Such a broad reform was agreed upon in 1993, but it almost caused the collapse of
the government since it consisted of the large Christian-Democrat and Social-Democrat par-
ties, which are typically wide apart on social and economic issues. In the 1994 election cam-
paign, a reform proposal of the Christian-Democrats to freeze state pensions was one of the
causes of a huge election loss and saw a new government of Social-Democrats and Liberals,
which are even wider apart on social and economic issues, together with the smaller Social
Liberal party, take office. This government managed to govern for eight years and especially
the first four years of this government are considered to have been successful. The Social-
Democrats and Liberals governed during a period of an economic boom allowing them to
realize their objectives without being forced to take painful measures like social security re-
form or tax hikes. Consequently, the 1993 disability reform, which was very successful in the
beginning, was relaxed under pressure from society and disability inflow started rising again
to new record levels. Both ageing and disability became more or less taboo subjects, although
during the first half of this eight-year government period, experience rating for employers
Introduction 51
in the disability scheme was introduced and the first year of sickness benefit was privatized,
whereas until 1996 the first year of sickness was paid for by a flat premium by all employers.
During the second half of this government, then State Secretary Hoogervorst successfully
shifted focus to the administration of the law and the enforcement of the gatekeeper function
and introduced the gatekeeper law specifying the rights and obligations of both employer and
employee when an employee suffers ill health. These measures could be agreed upon by the
Social-Democrat and Liberal parties because the beneficiaries were not harmed and the em-
ployers’ costs were redistributed, not increased. Although, in retrospect, the experience rating
and gatekeeper laws have been very successful reforms, major systemic reforms were stalled
during the 1994–2002 government periods.
In 2002, the Social-Democrat and Liberal cooperation ended because both parties lost heavily
in an election that was overshadowed by the murder of Pim Fortuyn. Consequently, a right-
wing government, consisting of Christian-Democrats, Liberals and Pim Fortuyn’s party, took
office, the latter being replaced within a year by the socio-economically right-wing tending
Social Liberal party. This coalition agreed upon major social security reforms and managed
to deliver them. The main reforms of this government are described in Bovenberg and Gradus
(2008), and include the introduction of the new disability scheme (WIA), re-examinations of
the stock of disability benefit recipients, abolition of the generous early retirement schemes
then existing in the Netherlands, cuts in unemployment benefit duration (from a maximum
of 7.5 years until 2003 to just over 3 years from 2006 onwards), privatization of the health
care insurance system and introduction of financial incentives for municipalities administer-
ing the social assistance law. The cooperation between Christian-Democrats, Liberals and
Social Liberals came to an end in 2006 when disagreements on issues like immigration policy
increased. In the election campaign ageing remained a taboo subject for the three largest par-
ties. The next government, which lasted from early 2007 until early 2010, consisted again of
Christian-Democrats and Social-Democrats together with the smaller Christian Union. This
government was plagued by the old fundamental differences in social and economic opinions
52 chapter 1
between Christian-Democrats and Social-Democrats and managed few reforms. Only after
the economic crisis at the end of 2008 did the need for structural reforms become clear and
the government parties agreed upon raising the retirement age from 65 to 67.
In the Dutch situation the role of the Social Economic Council (SER) is particularly impor-
tant and specific for the Netherlands (Den Butter, 2010). The SER is the main policy advisory
board for the government on social economic issues and consists of labour unions, employer
associations and independent members (mainly from science, including the directors of CPB
and the Dutch Central Bank). On disability reform, the Social-Democrat and Liberal govern-
ment in 2000 installed the Donner commission to investigate the subject. Because disability
reform was declared taboo by this government, this was the only way to take a step towards
reform. The 2002 SER advice on disability reform, which was based on the 2001 Donner re-
port, was the final breakthrough for achieving the later 2006 disability reform. More recently,
in 2009 the SER initially did not manage to agree on state pension reform, after which the
government decided to raise the retirement age from 65 to 67. However, in June 2010, trade
unions and employers organizations managed to agree on pension reform after all. Whether
their proposal will be taken over by the new government, which consists of Liberals and
Christian-Democrats and came to power in October 2010, is not yet clear.
1.7 Overview of the thesis
In this chapter we have seen how microsimulation models should be placed in relation to
other model types such as economic theory models and general equilibrium models. We have
studied how and from which origin these different model types have developed over time.
Also we have seen that the use of, in particular, dynamic microsimulation models in the
Netherlands is rather limited whereas in some other countries such models flourish. Further-
more, the issues of population ageing and disability benefit recipiency in the Netherlands
were studied, rather superficially, as well as the relationship between the two.
Introduction 53
The core of this thesis consists of four chapters that are based on papers, three of which have
been published in refereed journals. Chapters 2 and 3 are on microsimulation and disability
insurance and chapters 4 and 5 are on microsimulation, ageing and pensions. In chapter 2,
a microsimulation analysis is presented of the 2006 reform in the Dutch disability scheme.
Chapter 3 has a more retrospective character and analyses the effects of recent disability
reforms in the Netherlands. In chapter 4, the Social Affairs Department of the Netherlands
Ageing and Pensions model (SADNAP) is presented. Chapter 5 is on microsimulations on
the effects of ageing-related policy measures using the SADNAP model.
The main research question of this thesis is whether and how microsimulation models should
be integrated into the policy analysis. In order to answer this central research question, first
two subquestions will be researched concerning the added value of microsimulation models
as a forecasting and policy analysis tool in social security policy and the conditions required
to allow microsimulation models to have such added value. These research questions will be
discussed in the concluding chapter 6 of this thesis.
54 chapter 1
2
A microsimulation analysis
of the 2006 regime change
in the Dutch disability scheme2
This chapter introduces a microsimulation model that simulates the budgetary impact of the
2006 regime change in the Dutch disability scheme. A dynamic population model fits the
case of the disability benefits the best. As opposed to macro forecasts, a microsimulation
can answer questions about the individual or meso income effects, the exact distribution of
expenses among different benefits and the time path of the savings. The introduction of the
proposed system change decreases the number of disability benefits by more than 25% from
2020 onwards and reduces total costs by almost € 2 billion or 20%. Based on the better incen-
tive structure, participation will increase and boost GDP. Microsimulation can be used to pick
the winners and losers of the new system and give the time path of the savings. For almost all
partially disabled that are working, the total discounted income after the system change is as
large as or larger than before the system change, for the non-working total discounted income
is lower. Overall, the Gini-coefficient slightly decreases due to the regime change.
2 This chapter has been published in Economic Modelling, Vol. 23, pp. 427-456 (Van Sonsbeek and Gradus, 2006). As
this chapter was written and accepted for publishing before the introduction of the new WIA law that replaced the
WAO in 2006, throughout this chapter the WAO is referred to as the ‘current system’.
56 chapter 2
2.1 Introduction
An important benefit program in the Netherlands is the Disablement Benefits Act (WAO).
This disability insurance covers the employee’s loss of income due to long-term sickness and
disability. From its introduction in 1967 until the beginning of the nineties, there has been a
steady increase in the number of disability benefit recipients. This increase coincided with a
steady rise in unemployment and a drop in labour market participation (Westerhout, 2001).
This led to the belief that disability schemes have been used as a labour market exit route.
Between 1994 and 1996 the number of disability recipients decreased slightly. This decrease
was due to a tightening of the eligibility criteria in 1993 - in particular the definition of dis-
ability – and (re)assessments based on the stricter criteria. Between 1996 and 2003 the num-
ber of benefits rose again, this time coinciding with an economic boom. In 2003, the number
of disability recipients approached 1 million (13% of the labour force). The disability rate in
the Netherlands is among the highest in the world and especially women and older workers
have a higher chance of getting disabled (OECD, 2003). Therefore, from the point of view of
an ageing society and an increasing share of women in the labour force, it has been forecasted
that the amount of disability recipients will increase in the near future and will surpass the
level of one million recipients.
Contrary to other European countries the Dutch Disability Insurance Act covers both work-
related injury (risque professionnel) and non-work-related injury (risque social). It allows
access into the scheme without taking into account the recipient’s previous working history.
Moreover, its access level is low as employees already receive a benefit as soon as their earn-
ing capacity is reduced by only 15% due to long-term sickness or disability. In general the
disability benefit is permanent and at least as high as other social security benefits such as the
unemployment benefit and social assistance. This is why the disability insurance in the past
few decades has become a very popular arrangement for early retirement (see also Kapteyn
and de Vos, 1998). As a result, the government tried to reduce benefit levels in 1993, by limit-
A microsimulation analysis of the 2006 regime change in the Dutch disability scheme 57
ing the duration of wage-based benefits. In response, most employees had taken an additional
(private) insurance policy to cover the risk of this reduction in insurance benefit. Therefore,
the number of disability benefit recipients continued to increase at the end of the nineties.
At the beginning of 2001, a group of experts was commissioned. The recommendations
of this group, the so-called Donner-proposal, are based on the premise that recovery and
reintegration of available labour capacity has the highest priority with high responsibility
resting with both the employers and employees. In 2002 the Social-Economic Council (SER)
published an agreement accepting the basic ideas of the Donner-proposal. In September 2003
the Dutch government proposed a major regime change in disability benefits, mainly based
on the SER-agreement. The government decided to limit eligibility by tightening entry condi-
tions and reducing benefit levels for partially disabled. Moreover, applications for disability
benefits will be subjected to a more rigorous screening of the loss in earnings capacity, and
the primary responsibility of employers and employees for limiting sickness absence will be
extended from one year to two years. The new Disability Act will be in force from January 1,
2006 onwards whilst the extension of the responsibility of employers and employees for lim-
iting sickness absence has already been in force from January 1, 2004 and the more rigorous
screening of the loss in earnings capacity has been in force from October 1, 2004.
For our discussion it is important to comprehend what kind of impact the different parts of
this new Act will have. For this reason, a microsimulation model is developed. A microsimu-
lation model takes micro-level units, like individuals, as the basic units of analysis. Nelissen
(1991) lists various advantages of microsimulation like that no information is lost through ag-
gregation, that there is full representation and treatment of the heterogeneity in household and
individual characteristics and that the most complicated interrelationships can be described.
This chapter discusses the microsimulation model used to forecast the impact of the new Act.
The chapter is structured as follows. In section 2.2 the current system of disability benefits
and the proposed regime change are discussed. In section 2.3 microsimulation and the differ-
58 chapter 2
ent kinds of simulation models are discussed. Both model choice and design are explained. In
section 2.4 the implementation of the simulation model is discussed, including the prepara-
tion of the data, the behavioural assumptions and the validation of the model. Section 2.5 dis-
cusses the results of the simulation model. In this section, both macro-results (total number of
disability benefits, total cost) and micro-results (spread in changes in individual incomes) of
the model are presented as well as some sensitivity analyses. Here, the redistribution effects
of the system change are also discussed. Finally, section 2.6 contains a conclusion and some
proposals for future research on this topic.
2.2 Disability benefits in the Netherlands
2.2.1 Current system of disability benefits
At the moment, workers below the age of 653 in the Netherlands are entitled, in case of full
disablement, to a benefit of 70% of their previous earnings, called WAO4. After a couple of
years, depending on the age of the person, this benefit is reduced to a lower level (second
phase, see table 2.1). This follow-up benefit can be drawn until the age of 65. In practice,
most employers supply employees with additional benefits agreed upon in collective agree-
ments in order to narrow the gap between the lower level in the second phase and the higher
level in the first phase. Currently, for 80% of the employees, the level of the disability benefit
for the fully disabled is permanently 70% of the former wage.
In case of partial disablement, the same rules apply, but instead the amount of benefit is
proportional to the grade of disability. In case of partial disablement the disability benefit
can be increased with an unemployment benefit, called WW. This benefit lasts for a limited,
age-dependent period of time, ranging from 0.5 to 5 years, and is therefore less attractive than
the disability benefit (see table 2.1). When the WW benefit has come to an end, the partially
disabled are also entitled to another unemployment benefit, called IOAW (The Provision for
3 The disability benefit will expire when someone turns 65. All persons aged 65 or over are entitled to the same general
old-age pension called AOW. Furthermore, for most employees there is a second tier based on a funded pension as well.
4 The maximum benefit level is € 30,639 per year. This level is almost 2 times the minimum wage and is indexed by the
average wage increase each year.
Older and Partly Disabled Unemployed Employees). IOAW depends on the income of the
whole household, but – contrary to the normal social assistance called WWB – it does not
depend on household wealth5.
More than half of the partially disabled are actually working part-time. When the total
amount of disability benefit, unemployment benefit and wage is below the social minimum
level, an additional benefit can be supplied to top-up the benefit to the minimum level. This
benefit is called TW (The Supplementary Benefits Act). Entitlement to such a TW-allowance
is, similar to the IOAW-benefit, dependent on the income of the whole household, but not on
household wealth.
Table 2.1 Disability benefits and possible supplements in the Netherlands (2003)
Maximum (2003) Conditions on
Benefit Duration
(for a single) benefit
WAO Disability benefit – 1st phase 0 - 6 years € 30,639 none
WAO Disability benefit – 2nd phase permanent € 11,474 / 30,639 none
WW Unemployment benefit ½ - 5 years € 30,639 none
TW Supplement to WAO/WW depends on duration € 3,442 partner income
of WAO/WW
IOAW Conditional unemployment permanent € 11,474 partner income
benefit for the older and partially
disabled
WWB Social assistance permanent € 11,474 partner income,
household wealth
In the Netherlands there also is a benefit called WAZ, for the disabled self-employed and a
benefit called Wajong, for the young handicapped. Both provide an individual benefit of at
most the social minimum (€ 11,474, 70% of the minimum wage, which in 2003 was set at €
16,391). They are more attractive than social assistance because they do not depend on part-
ner income or household wealth. In 2003, 56,000 people received a WAZ-benefit and 138,000
people received a Wajong-benefit. The number of Wajong-benefits is expected to grow stead-
ily to over 200,000 in 2020.
5 If net household income exceeds the social assistance level, no IOAW benefit is granted.
60 chapter 2
The WAZ-benefit ceased to exist in august 2004 because, according to the government, the
self-employed are supposed to be able to insure themselves. Only the existing WAZ-cases
keep their entitlement to a disability benefit. In 2003, 786,000 people received a WAO benefit
in the Netherlands. In this period more than 13% of the labour force received a disability ben-
efit and when compared with other OECD-countries this level is extraordinarily high (OECD,
2003). The OECD stresses that the current system provides recipients with little incentive
to work and recommends therefore more emphasis on getting disabled persons back into the
workforce.
2.2.2 Regime change proposed, from 2006 onwards.
The new system aims at increasing the financial incentives for employees in order to avoid
disability and to stimulate labour participation in case of a partial disability. Instead of one
general disability benefit law, the population of disabled is subdivided into three categories.
1) People who are fully disabled and have no or only a slim chance of recovery will be
granted a permanent disability benefit of 70% of their previous earnings. The lower
second phase benefit is skipped for this group6. If certain conditions are met, the benefit
level can be increased to 75% of previous earnings in the future7.
2) People who are partially disabled (with a degree of disablement of at least 35%) or
have a good chance of recovery, will be granted either a supplement to the wage they
are earning (in case they are working) or qualify for an unemployment benefit with the
conditions that apply to it. After expiration of the unemployment benefit, the benefit
equals the share of the social minimum level that corresponds with their degree of
disability. So, in the new system disabled people who are partially working will be
rewarded more than in the old system.
3) People who are partially disabled with a degree of disability lower than 35% will not
have access to disability benefits anymore. These people are supposed to be able to find
6 However, most employees have taken out an additional insurance to narrow the gap between the benefit level in the first
and the second phase (see section 2.2.1).
7 According to the proposal, the benefit level will be raised from 70% to 75%, if in 2007 the number of fully disabled
entering the scheme will be less than 25.000.
their way through the labour market themselves, preferably with the same employer as
before. For this group, the incentives to work are also encouraged.
Furthermore, in order to keep the first category reserved for people that really are fully disa-
bled, the entrance examinations will be altered. The current medical examination has numer-
ous restrictions on the assessment of labour capacity. These restrictions allow a large number
of people to keep receiving a full benefit, some of whom are still able to work. In the new
system, it will be more difficult to access the first category and be entitled to full benefits.
The alteration of the entrance examinations has already been in force from October 1, 2004,
anticipating the new Disability Act.
Figure 2.1 Proposed regime change in disability benefits
Medical
examination
Partially disabled Particially disabled

Fully disabled (>=35%) (< 35%)
No chance of Chance of Working Not Working Not working

recovery recovery part-time working
Permanent Temporary Supplement to Unemployment Unemployment

disability benefit disability benefit Normal Wage benefit
Wage (WGA) benefit (WGA)
(IVA) (WGA)
Breadwinner / Second
single income
Minimum benefit / Minimum benefit

social assistance
62 chapter 2
2.2.3 Predictions of the regime change effects at the macro level
Macro forecasts of the effects of the regime change have been calculated by the Nether-
lands Bureau for Economic Policy Analysis (CPB, 20028). CPB expects a decrease in cost
of benefits of € 2 billion. They have used an approach in which the long-term cost of the
current system are calculated by subdividing the most recent inflow cohort into the relevant
subgroups (the shaded boxes from figure 2.1) and by multiplying the average group sizes
with their average benefit levels and average benefit durations. The direct effects of the new
system are quantified the same way. The behavioural effects of the new system are calculated
using a disability benefit elasticity. For each subgroup the average replacement rate (the aver-
age income during the period of disability as a percentage of the income before disability) is
estimated for both the current and the new system. The change in replacement rates in terms
of percentage is multiplied by the elasticity in order to forecast the volume effects in terms
of percentage of such a change. According to analyses made by CPB (CPB, 2001), a benefit
elasticity of 1.5 can be applied to measure these volume effects for disability benefits. This
implies that a 1% decrease in replacement rate leads to a 1.5% decrease of the number of
benefits. The number of working people will increase with the same number. This elasticity
is based on a study on a sudden increase in benefit levels in Canada (Gruber, 2000). In this
study Gruber finds a point estimate of the elasticity of labour force non-participation with re-
spect to disability benefits of about 0.3. Because the disabled constitute about one fifth of all
non-participants in Gruber’s dataset, the elasticity of the probability of receiving a disability
benefit with respect to an increase of these benefits would equal about 1.5.
Though useful for a global view of cost effects and therefore sufficient for the macro calcula-
tions of the coalition agreement, this method has some important limitations because all data
used are group averages for groups that are not homogeneous at all. In fact these groups can
be very diverse, for example with respect to income ranging from low incomes of breadwin-
ners who already receive other benefits to high incomes of people who also have an income
earning partner. The use of a microsimulation model can overcome the limitations that result
from the use of group averages.
8 These forecasts are based on the regime change proposals in the coalition agreement of the cabinet Balkenende I. These
proposals are slightly different from the current proposals.
2.3 The Microsimulation model
2.3.1 Microsimulation versus cell-based macro-forecasts
Usually, system changes are evaluated using cell-based macro-forecasts like the one dis-
cussed in section 2.2.3. These forecasts are reasonably adequate in predicting total structural
effects of policy measures but they have difficulties in answering commonly asked questions
about the individual or meso income effects for specific groups of disabled, the exact distri-
bution of expenses among different benefits and the time path of the savings. In theory, these
difficulties can be overcome by increasing the number of subgroups. But in practice, because
of the large number of subgroups that arise when taking into account all the relevant charac-
teristics, these cell-based approaches become problematic when the subgroup size becomes
very small. For example, when distinguishing between men and women, 5 age-classes, 8
classes of disability, working and unemployed, at least 3 different household types, the exist-
ence of employer’s supplements and the existence of enough working history to be entitled to
certain benefits, almost 2,000 subgroups would develop.
Microsimulation can have an important added value in answering questions like the ones
mentioned above, especially in a complex social security benefit structure. This is the case in
the Netherlands, where many combinations of benefits are possible, and where benefit levels
depend on many factors like age, working history and household income and wealth. Cald-
well and Morrison (2000) give some examples of analyses in which microsimulation should
be preferred above macro-forecasts. These include analyses of projected winners and losers
of alternative policies on a period-specific or lifetime basis, analyses simultaneously focused
on families and individuals, and quantification of incentives to work at particular life-course
or period junctures. These examples all apply to the analysis of the system change in disabil-
ity benefits. Therefore a microsimulation model has been developed using data sources on
personal level that only recently became available.
64 chapter 2
2.3.2 Model choice
There are many classes of microsimulation models. Spielauer (2002) distinguishes between
static and dynamic models, forecasting and explanatory models, generalised and specialised
models, cohort and population models. The choice of model should match its purpose. A
model built to forecast the effects of a change in the disability benefit system should evident-
ly be a specialised forecasting model. This leaves the choice between a static and a dynamic
model, a cohort and a population model respectively.
In a static model the characteristics of the persons in the simulation, like age, are fixed. In a
dynamic model the characteristics of the persons in the simulation change over time. Dy-
namic models consider both first- and second-order effects of policies as behavioural effects
may be included. A static model can be appropriate when system dynamics (like ageing) are
not relevant to the policy measures, e.g. in case of short-term effects of fiscal policy changes.
Otherwise, a dynamic model is preferred. For the case of the disability benefits, we are
interested in the long-term effects of the system change, including behavioural effects, thus
making a dynamic model the appropriate choice.
In a population model, a whole population is simulated. The model starts off with a sample
of the population and then models their life and earnings histories over some period of time
(O’Donoghue, 2001). In a population model people leave the simulation if, for example, they
die, and new cases, for example births, can be added to the simulation. In a cohort model, one
single cohort is simulated over its entire lifetime, without the possibility of new cohorts being
added. The cohort model starts off with a sample of individuals aged 0 at a particular point in
time and models each individual’s life and earnings history.
Often, a cohort model is chosen because its development time is much shorter than for a
population model. Also, the cohort model provides insight into the income consequences over
the entire lifetime of individual persons. This is considered important in assessing behaviour-
al effects of policy measures. When using a cohort model, by accumulating the first and the
second year of the cohort we obtain an estimation of the cost of the new system in the second
year. In the same manner, we can deduct the cost of the third year, the fourth year and so on,
till an estimation of the structural situation is reached. In fact, as Gilbert and Troitzsch (1999)
show, a cohort model can be used to obtain a proxy of the results of a population model.
However, when using a cohort model the size and composition of the cohorts should not fluc-
tuate too much over time. Broadly speaking, disability cohorts do not change as much over
time as compared with, for example, unemployment cohorts, but the effects of demographic
changes, like ageing and changes in participation levels of e.g. women and older people, are
missing. Moreover, when making forecasts for the whole population (both the old and the
new cases) a sample from the current population should be simulated in any way in order to
estimate the total number of disability benefits.
A population model, even though requiring a larger development time, being more time-con-
suming and generating larger databases, is needed to derive integral forecasts of a complex
system change like the one described above. Particularly, the short-term effects of the new
measures can be calculated directly and more accurately as compared with a cohort model. In
a cohort model these short-term effects have to be deducted, first by accumulating the cohorts
for every different cohort simulation and second by adding up the results of the different
accumulated cohorts. Also the database of a population model ends up with a large number
of cohorts, each of which can be analysed separately so that the typical ‘winners and losers’-
analysis can be executed. All in all, the dynamic population model appears to fit the case of
the disability benefits the best.
66 chapter 2
2.3.3 Model structure
The model is built using the SAS programming language and consists of four modules. In the
first module the base data set is transformed into a consistent set of data. A number of records
with missing or false data on gender, age, grade of disability, household type and wage are
omitted. Afterwards, a sample from the current population is drawn and another sample
from the cohort of new benefits is added for each year of the simulation, with corrections for
demographic developments.
The sample generated in the first module is used in the second module in which the base data
are enriched with imputed data on partner income, working status, working history and avail-
ability of employer’s supplements, so that each person in the sample has a set of nine charac-
teristics (five basic and four imputed) necessary to compute the appropriate benefit level.
The actual simulation takes place in the third module. In this module, a double simulation
takes place: one in which the current system of disability benefits is simulated, and one in
which the new system is simulated. Both simulations take place with exactly the same sample
but with a different set of accounting rules. Finally, the fourth module aggregates the data
from the two databases in order to provide various summary output tables on number of
benefits, cost of benefits and average benefit level for each forecast year and for every kind of
benefit by gender and working status. Scatter graphs in order to judge income mutations on
individual level are constructed from the original non-aggregated database.
Figure 2.2 Summary of modules WAO/WIA model
1. Base data module

- reading and cleaning base data
- drawing population sample
2. Base enrichment module

- imputing working status
- imputing working history
- imputing (existence of sufficient) partner income
- imputing (existence of ) additional private insurance
Accounting rules
3. Simulation module - disablility benefits
- Read acclounting rules - unemployment benefits
- Sumulate current system - wages
- Simulate new system - supplements
- social assistance
4. Reporting module
- Aggregate data
- Priduce output tables
68 chapter 2
2.4 Implementation of the simulation model
2.4.1 The base-data set
The base data sets consist of records of all 66,635 new disability benefits from 2003 and all
785,595 disability benefits that were being supplied at the end of 2003. Thus, a full adminis-
trative data source has been available for microsimulation. The data sets are supplied by the
Central Benefit Administration Office (UWV) about six months after year end. Cross tables
from these data sets are published in UWV (2004). The records contain basic characteristics
like age, gender, grade of disability, household type and wage. In the base-data module these
two datasets are transformed into consistent datasets. Records with incorrect ages (younger
than 18 or older than 64), missing grade of disability, missing household type and missing or
suspiciously low wages (less than € 5 a day) are removed. As a result of this 16% of the new
benefit records and 8% of all benefit records are removed, leaving datasets of 56,184 benefits
that started in 2003 and 722,646 that existed by the end of 20039. On all characteristics, the
distribution appears to be highly comparable with the distribution of the full dataset, so the
removed records do not cause relevant biases. Table 2.2 gives the basic characteristics of the
2003 cohort of new and existing disability benefits.
9 The large number of removals is mainly due to missing wages.

Table 2.2 Characteristics disability beneficiaries (2003)
New benefits All benefits

Gender
male 28,645 51% 410,123 57%
female 27,537 49% 3112,523 43%
Age (2003)
15-25 1,899 3% 4,025 1%
25-35 10,715 19% 53,828 7%
35-45 15,370 27% 126,427 18%
45-55 17,712 32% 220,010 30%
55-65 10,488 19% 318,356 44%
Grade of disability
15-25% 5,480 10% 53,828 7%
25-35% 5,369 10% 57,729 8%
35-45% 4,069 7% 46,283 6%
45-55% 4,150 7% 45,050 6%
55-65% 2,223 4% 22,558 3%
65-80% 2,179 4% 22,247 3%
80-100% 32,354 58% 475,348 66%
Household type
single 14,720 26% 170,951 24%
couples 41,464 74% 551,695 76%
Previous wage
< € 16,391 (minimum) 12,284 22% 113,924 16%
€ 16,391 - € 43,770 41,168 73% 579,273 80%
> € 43,770 (maximum) 2,732 5% 29,449 4%
Total 56,184 100% 722,646 100%
70 chapter 2
The table shows that, even though men outnumber women in the working population, women
have a higher chance of getting a disability benefit than men as the shares of inflow of men
and women are almost equal. This confirms findings in e.g. OECD (2003) that, though gender
differences are usually small, women tend to report higher disability prevalence rates in a
majority of countries, most especially in the Netherlands and the Scandinavian countries.
Also, as one would expect and as noted in OECD (2003), the chance of disability rises with
the age. In the Netherlands, the majority of disability benefits are full benefits (with a grade
of disability of 80-100%) and the proportion of low incomes (below the minimum wage for
full-time employed) is considerable. The latter is caused by both the large number of part-
time workers (mostly women) in the Netherlands and the higher chance of obtaining disabil-
ity benefits that women tend to have.
Using samples instead of whole cohorts adds to the uncertainty of the results. This uncer-
tainty can be reduced by increasing the sample size. However, simulating the entire cohort
(or a very large sample) is impractical because of time- and capacity-constraints. Therefore, a
balance between a workable sample-size and a sufficiently low level of randomness has to be
found. The randomness due to the sample size is one of various sources of randomness that
can be distinguished in microsimulation (Spielauer, 2002). Imperfection randomness applies
to both micro- and macro-models, whereas Monte-Carlo variability is inherent in microsimu-
lation.
In order to test for an adequate sample size, for four different sample sizes seven simulation
runs each are performed. Then, for two key simulation results (the number of disability ben-
efits in 2040 and the total cost of disability benefits in 2040) the standard deviation of those
seven runs is computed. The results are presented in table 2.3.
Table 2.3 Deviation in simulation results by sample size

Sample size Standard deviation Standard deviation
(absolute / % of mean) (absolute / % of mean)
disability benefit expenses 2040 number of benefits 2040
1% € 90 mln / 1.0% 9,000 / 1.1%
2% € 80 mln / 0.9% 7,000 / 0.8%
5% € 40 mln / 0.4% 4,000 / 0.5%
10% € 40 mln / 0.4% 3,000 / 0.4%
Of course, larger samples produce more accurate results, but also complicate quick data
analysis of the simulation data. The analysis shows that a sample size of 5% leads to small
enough standard deviations of less than 0.5% of the mean. In absolute terms a 5% sample
leads to a standard deviation in total expenses of € 30 million (so a 95% chance of deviating
less than € 60 million from the mean) and a standard deviation in total number of disability
benefits of 4,000 (so a 95% chance of deviating less than 8,000 from the mean). Doubling
the sample size to 10% does not add much in terms of the accuracy of results concerning the
whole population. Nevertheless, in order to secure more accurate analyses on smaller sub-
groups, a 10% sample is used.
Therefore, a 10% sample from the disability benefit population of 2003 is drawn. For all
years in the simulation (2004-2040) a 10% sample from the new benefits is drawn with an
adjustment for the expected demographic changes. The adjustments are based on projections
of the inflow from 2004-2040 as made by the UWV for a report by the Social-Economic
Council (see also SER, 2004). The UWV-projections are based on recent forecasts of the
workforce by CPB (Huizinga and Smid, 2004) in which the labour participation of especially
women and older workers rises10. The accumulation of a 10% sample of the existing benefits
and 37 cohorts of new benefits amounts to a base dataset of almost 400,000 records.
10 CPB presents four scenarios varying from a global economy, in which the labour force grows the most, to the opposite in
which the labour force shrinks. The Global Economy scenario was also used in earlier publications by CPB and SER. In
section 2.5.3 we will discuss the labour market scenarios in more detail.
72 chapter 2
2.4.2 Imputation of missing data
Other important characteristics like working status (after disability), working history (before
disability), availability of a supplement to the second phase disability benefit and partner in-
come (which determines entitlement to and the level of social assistance) are not included in
the base data set. Nevertheless, these data are known at aggregate level from other public data
sources of similar populations. Therefore, these characteristics are imputed using a technique
similar to the ‘random hot-deck method’ (as described in detail in Nelissen, 1991), in which
values are randomly imputed using a known distribution of these values.
Working status is known from the whole population of disability benefits in 2002 for sub-
groups of gender, age-class and grade of disability (see table 2.4). Working history is known
from a sample of the cohort of new disability benefits in 1999 for subgroups of gender,
age-class and grade of disability. Availability of a supplement to the second phase disability
benefit (through private insurance paid for by employers and employees together) is known
only as a general percentage (80%) of the whole working population from a survey among a
panel of employers (Van Deursen et al., 1997). For couples, the existence of a second income
is imputed using overall averages for men and women. These averages are based on a sample
of new benefits in 1999. This sample of new benefits is followed in time by UWV concerning
an epidemiological survey (called EPI) among people who receive a disability benefit.
Table 2.4 Imputation scheme WAO/WIA model

Gen- Age Grade of Source Distribution
der disability
Working status X X X All benefits 2002 (UWV) (0,1)
Working history X X X Sample of new benefits 1999 (UWV) (0,1)
Partner income (existence) X Sample of new benefits 1999 (EPI) (0,1)
Supplement second phase Research report 1997 (ZARA) (0,1)
disability benefit
2.4.3 The simulation module
In the simulation module, in order to measure structural effects, theoretically, the simulation
should run until all old cases are replaced by new cases. In the case of disability benefits this
would mean a simulation duration of 47 years as the minimum age that entitles to a disability
benefit is 18 and the maximum age is 65. However, the number of younger people receiving
a disability benefit is comparatively low. In 2040, less than 1% of the starting population is
still receiving a disability benefit. Therefore, simulation duration is fixed at 37 years, starting
2004 and ending 2040.
The simulation starts at the entry examination by imputing, for every person in the sample,
the grade of disability after the implementation of the new eligibility criteria given a grade
of disability before the implementation of the new criteria. The new criteria decrease the
chance of being allowed a full disability benefit and also raise the chance of being included in
a lower class of disability than would be the case at the moment. The new grade of disability
is imputed using a transformation matrix, supplied by UWV, which describes the chance that,
if the grade of disability equals (i), the grade of disability according to the new criteria will
equal (j). This transformation matrix is itself the result of a computer simulation by UWV of
the effects of the new criteria on a sample of new benefits in 200211.
The simulation loop12 consists of a set of deterministic functions that determine the various
benefit levels, and of a set of stochastic functions that determine the set of parameters to be
used the next year. For each case in the sample, total personal income from various sources
is determined for the first year of disability. The simulation then continues on to the next
year. Firstly, the occurrence of major events (recovery from disability, retirement or death)
is simulated. Individuals can quit the simulation in case of death, recovery from disability or
retirement. Chances of death and recovery from disability are determined using group aver-
ages based on age, gender and grade of disability. Occurrence of these events is determined
11 The transformation matrix is available upon request.

12 See Appendix 2.B for a more detailed description of benefits.
74 chapter 2
by a Monte Carlo process. A person becoming 65 years of age quits the simulation automati-
cally. If a person keeps a benefit for the new year, using the same method, a possible change
in the grade of disability and in the working status is simulated. Secondly, with the actualised
set of parameters, the new benefit levels are determined for the second year. In the end, a
simulation data set is constructed, consisting of all cases in the sample, the duration of the
simulation and seven types of income components (six different kinds of benefits plus wage
in case a person worked in a certain year). Figure 2.3 illustrates the simulation process. The
deterministic functions determine the different kinds of benefit levels given a set of parame-
ters belonging to person (i) in year (t). The stochastic functions determine whether the person
continues the simulation for one more year or not.
Figure 2.3 Simulation loop WAO/WIA model
All persons (i) All years (t) in the Deterministic functions:

in the sample simulation - WAOi,t = F1(agei; gdi,t; pwi)
- WAGEi,t = F2(wsi,t; gdi,t; pwi)
- WWi,t = F3(wsi,t; whi; gdi,t; pwi)
- TWi,t = F4(hhi; WAOi,t; WWi,t)
- IOAWi,t = F5(hhi; WAOi,t; WWi,t; TWi,t)
- WWBi,t = F6(hhi; WAOi,t; WWi,t; IOAWi,t; TWi,t)
Stochastic functions:
Next persons (i) Next year (t) in the - STATUSi,t+1 = T1(agei; gender; gdi,t)
in the sample simulation - STATUSi,t+1 Є (disabled, recovered, dead, pensioned)
gd = grade of disability Є (15-25%; 25-35%; 35-45%; 45-55%; 55-65%; 65-80%; 80-100%)

ws = working status Є (working; unemployed)
pw = personal wage when becoming disabled Є (< € 43,229)
hh = household type Є (single; 1-income couple; 2-incomes couple)
Not all personal characteristics are supposed to be dynamic. For simplicity the household
type is supposed to be fixed over time. Fixing household type is not very realistic when look-
ing at the real world where households are in fact dynamic. However, in the case of disability
benefits, fixing the household situation is a reasonable assumption because the majority of the
cohort is over 40 years of age and changes in household type for this age category are not as
frequent as compared with younger populations.
Generally speaking, increasing the dynamics of the model is not always preferable. A dy-
namic grade of disability and working status is clearly more realistic. For example, individu-
als suffering from a progressive disease will have an increasing grade of disability over time,
whereas individuals recovering from a disease will sometimes have a slowly decreasing
grade of disability over time. The working status of these individuals will be affected by the
changing grade of disability as well. In reality, people do not always work for a consecutive
period of many years, but may get hired for shorter periods of time or get fired after a while.
Introducing dynamics in grade of disability and working status makes it difficult to identify
large homogeneous groups. Instead of major groups like the working partially disabled and
the unemployed partially disabled, one ends up with a large in-between group of people who
work part of the time and are unemployed at other times. Depending on the extra value of
these dynamics for the macro results and the problems in identifying homogeneous groups,
one can choose for a simulation with or without dynamics in these characteristics.
2.4.4 The behavioural model
Besides the dynamics described in section 2.4.3, microsimulation also involves modelling
the behaviour of individuals, taking into account the effects of policy parameters such as tax
rates, eligibility rules for benefits and subsidies and compensation rates in the social security
system (Klevmarken, 2001).
Behavioural effects as a result of financial triggers are included in the simulation module of
our model. To measure these behavioural effects of the system change on the labour partici-
pation, two additional important assumptions are built into the model. First, the financial
incentives, which are calculated using the conventional methods applying benefit elasticities
(see section 2.2.3) to the change in replacement rate, are considered not to influence the num-
ber of new benefits but only the participation status of those benefits. Implicitly, it is assumed
that because of the stricter allowance criteria, the medical situation of a person is given and
76 chapter 2
the result of the new, stricter entrance examination cannot be influenced by financial trig-
gers. Since labour participation increases recovery chances, the number of benefits will be
influenced in our model through changes in duration of the benefit. So, if working is made
more attractive for the partially disabled, the total number of new benefits is assumed not to
change, but the share of labour participants in this total is assumed to rise13. Second, it is as-
sumed that if someone gets a lower disability grade because of the stricter allowance criteria,
the probability that someone participates on the labour market changes to the level currently
belonging to that new disability grade. This assumption reflects the viewpoint that the partial
disability benefit is an instrument of reintegration in itself (e.g. de Jong & Thio, 2002 and
Besseling, 2002).
In the model, the elasticity described above is used to estimate the chance of entering or
leaving the labour market. This reflects the viewpoint that for a large number of workers , the
labour supply decision is a discrete one as many jobs are defined as full- or part time, with
the hours of work more or less predetermined (Røed and Strøm, 1999). However, it is taken
into consideration that not all disabled workers will be able to find a job for the number of
hours they are able and willing to work. This is estimated in the following way. Given the
previous earnings and the grade of disability of a person, the earning capacity of a person can
be computed14. In the model, it is assumed that if a person participates on the labour market,
the wage of this person will not exceed the earning capacity. However, wages below the earn-
ing capacity are possible. As is known from data from UWV that a large majority of 95% of
working beneficiaries is able to earn at least the earning capacity level (Tweede Kamer der
Staten-Generaal, 2005), it is assumed that 95% of the working persons in the sample earns
exactly his earning capacity and the other 5% earns a share of the earning capacity that is ran-
domly distributed between 0 and the earning capacity. The latter can reflect either a smaller
number of hours worked or a lower wage earned per hour.
13 In the macro forecasts by CPB as described in section 2.2.3, each subgroup of disabled is seen as an independent entity
that interacts with the labour market only, whereas shifts to other subgroups are not taken into account. If, for example,
policy measures are implemented that cut the benefits of non-working partially disabled, they are assumed to resume
working fully, whereas a shift to working partially and still receiving a partial benefit is more plausible.
14 For example, if a person earned € 30.000 a year before disability and has a grade of disability of 50%, this person has an
earning capacity of € 15.000.
In the behavioural model is abstracted from the effects of the change in eligibility rules15.
These effects are diverse but hard to model. Part of the system change is that so called soft
diagnosis claims (disabilities that are based on subjective symptoms like chronic fatigue
syndrome and whiplash) will be examined more vigorously. However, data on those soft
diagnoses are not very reliable and spillovers can occur from one illness to another, e.g. if
whiplash is not eligible and depression still is, workers non-eligible in one category may seek
eligibility in another. For an overview of the behavioural model used, see appendix 2.A.
2.4.5 Model validation
Caldwell and Morrison (2000) find that validation methods for longitudinal dynamic micro-
simulation models are underdeveloped and underutilized. They address the validation of such
models from several perspectives like consistency of simulated results with administrative
data, consistency with survey data, comparability with results from other non-microsimula-
tion models and special validation challenges posed by the Monte Carlo nature of the micro-
simulation models. The latter is dealt with in section 2.4.1. The other validation perspectives
are used in the following way.
First, the distribution of the imputed data is compared with the original distribution as known
from separate aggregate data sources. Second, the model output for the base year is compared
with the realisations as known from public data sources by UWV. Third, the model results for
the forecasting years are tested on plausibility by comparing them both to long-term forecasts
from other public data-sources and by recognizing expected trends not yet quantified. Exam-
ples of such trends are the following:
1) A drop in average disability benefit level as people who entered the disability benefit
since 1993 qualify for a higher first phase benefit followed by a lower second phase
benefit, whereas people who entered the disability benefit before 1993 (who now slowly
phase out because of pensioning) keep the higher first phase benefit until they turn 65.
15 The modeling of the direct effects of the change in eligibility rules is described in the second paragraph of section 2.4.3.
78 chapter 2
2) A rise in use of social assistance, because the average benefit level drops and more low
part-time incomes entered the disability benefit in recent years.
3) A drop in partial unemployment benefit because the duration of the unemployment
benefit is decreased by 2 years as a consequence of a policy measure from 2003.
All these trends are recognised in the model output. Also, two interesting deviations appeared:
1) The numbers of unemployment benefits are predicted to be 20% higher than known
from micro data analysis. This can be attributed to so-called voluntary unemployment.
People make a trade-off between the level of the unemployment benefit and the require-
ments that come with it, like being obliged to be available to the labour market.
2) The numbers of social assistance benefits are predicted to be far higher than known
from realisations whereas the average level of the social assistance is far lower than
known from the same data sources. Apparently, if the social assistance someone is
entitled to is too low, people do not bother to get it because social assistance also has
strict requirements on availability to the labour market and strict wealth information re-
quirements. Because of this model validation result, the model is adapted so that social
assistance lower than a certain threshold value, is assumed not to be used. The thresh-
old value is set at a value that produced average benefit levels equal to the ones known
from other data analysis.
2.5 Results of the simulation model
This section contains the forecast results of the simulation model that are used to determine
the effects of the new system of disability benefits, as proposed to the Dutch Parliament
(Tweede Kamer der Staten-Generaal, 2005). The baseline is the structural situation of the cur-
rent system, which is the system as it was before the tightening of the eligibility criteria that
has been in force from October 1, 2004. Section 2.5.1 deals with the macro results, section
2.5.2 deals with the micro results. Section 2.5.3 contains some sensitivity analyses and sec-
tion 2.5.4 assesses the redistribution effects of the reforms.
2.5.1 Macro results
Like most other developed countries, the Netherlands is faced with an increasing share of
elderly persons that have a higher chance of getting disability benefits. The current system
leads to about 750,000 disability benefits in 2040, costing € 9.9 billion. In order to simulate
the new system, the changes in the rules are categorised and simulated both separately and
simultaneously. The main changes in the new system are the following:
1) Stricter allowance criteria that decrease the chance of being allowed a full disability
benefit and also raise the chance of being included in a lower class of disability than
would be the case at the moment. Because of the stricter allowance criteria, more jobs
will be found at the entrance examination. This implies that enough jobs can be found in
cases where in the current system, not enough jobs would have been found. If a job earns
enough compared with the former wage, the entitlement to a disability benefit can end.
2) Exclusion of the lower two classes of disability (15-25% and 25-35%);
3) Changes in the benefit regime:
a. A less attractive benefit regime for partially disabled that are unemployed (their ben-
efit level is limited to a share of the social minimum equal to the grade of disability);
b. A more attractive benefit regime for partially disabled that are working (they keep
their higher first phase disability benefit as long as they work);
c. A more attractive benefit regime for the fully disabled (they keep their higher first
phase disability benefit as long as they are fully disabled);
d. Extension of the duration of the higher first phase disability benefit by on average 1
year.
These changes can be simulated both separately and simultaneously. The stricter allowance
criteria have been in force from October 1, 2004. The exclusion of the lower two classes of
disability and the changes in the benefit regime are proposed to be in force from January 1,
2006. Sometimes the combined effect of two policy changes is greater than their separate ef-
fects. For example, because of the stricter allowance criteria, more people populate the lower
80 chapter 2
two classes of disability, so exclusion of these two lower classes has a larger effect. Some-
times the opposite is true as well. For example, when a lower benefit level is considered, the
effects will decrease when stricter allowance criteria yield fewer benefits. The new system
leads to about 550,000 disability benefits in 2040, costing € 8.0 billion. As mentioned earlier,
the old system leads to about 750,000 disability benefits costing € 9.9 billion. The number of
disability benefits is reduced by 27% and the total cost of all social security benefits (in-
cluding supplementary benefits from private insurance companies) by 19%. The amount of
income from wages of working disabled people rises from € 2.1 billion in the old system to €
3.8 billion in the new system. The projected development of the number of disability benefits
is shown in figure 2.4.
Figure 2.4 Projected number of disability benefits before (WAO) and after system change
(WIA), 2003-2042
The new system leads to a steady reduction in numbers of benefits as compared with the old
system16. Roughly, three stages can be distinguished for the old system:
1) From 2006-2015, the number of benefits drops slightly, mainly due to a high number of
pensioners from 2010 onwards.
2) From 2015-2030, the number of benefits grows rapidly to the maximum level of over
750,000. One obvious cause of this rise is the growth of the workforce but more sig-
nificant than its growth is the composition of the workforce. The share of older people
and women in the workforce rises and both the older and the women have a higher than
average disability risk.
3) After 2030, the number starts to fall slightly. The workforce, according to the current
projections, starts to fall after 2025 and with some delay this influences the number of
disability beneficiaries in later years.
The fall in number of disability benefits is a gross effect of the system change. The net effects
of the system change are more important. Both direct and indirect effects play an important
role here. Direct effects assume a steady state situation in which the behaviour of people does
not change. For example, someone who gets a disability benefit in the current system can
get no disability benefit in the new system because of the stricter allowance criteria. A direct
effect of such a change can be that the person considered will get an unemployment benefit
or social assistance instead. So, when direct effects are calculated the net effect of the system
change will be smaller than the gross effect.
Moreover, when behavioural effects are taken into account, the effects can increase. Using
the same example, taking into account that the unemployment benefit or social assistance are
less attractive than the disability benefit, the person considered can decide to work instead of
receiving a benefit. Because of such behavioural effects, a rise in the working population is to
be expected. This is exactly the intention of the new disability law. Following the behavioural
model described in section 2.4.4, the working population increases by more than 80,000 (see
figure 2.5), and the total wages earned increases by € 1.7 billion.
16 In 2005, the number of benefits drops sharply by about 50,000. This is caused by the extension in 2004 of the duration of
sickness benefit from one to two years. Sick employees stay with their employers for 2 years, before they can enter the
disability benefit scheme.
82 chapter 2
Figure 2.5 Projected number of beneficiaries working before (WAO) and after system
change (WIA), 2003-204217
Direct effects of the system change account for 70% of the total savings. The stricter allow-
ance criteria are the most important here, whereas the changes in the benefit regime actually
induce extra costs when only direct effects are taken into account. Behavioural effects ac-
count for 30% of the total savings. Here, however, the changes in the benefit regime do play
an important role in the savings because the new system increases the difference between
the benefits for the working partially disabled and the unemployed partially disabled, which
makes working more attractive.
Finally, figure 2.6 shows the development of the total cost of benefits for the disabled. This
includes disability benefits, unemployment benefits, social assistance and supplementary ben-
efits from private insurance companies. The total cost of the new system as compared with
the old system is reduced by 19% or € 1.9 billion in 2040.
The savings of the system change are slightly smaller than earlier used macro models predict-
ed. This is a common result when comparing micro to macro models. Usually if a policy meas-
17 Including for the WIA the ones that would have qualified for the WAO but do not qualify for the WIA.
ure generates savings, the income level below minimum are somehow compensated. If so, ap-
plying a set of accounting rules to group averages produces higher estimates than applying the
same set of rules to individuals, because for part of the individuals the policy measures have
no financial consequences. Of course, if using a macro model, one can take this aspect into
account by extending the number of relevant subgroups to include also the groups that will be
partially or fully compensated, but this will lead to increasingly complex macro models.
The development in time of the total expenses follows the same pattern as figure 2.4 (disabili-
ty benefit expenses). The ageing workforce leads to a 13% increase in total expenses between
2015 and 2030. After 2030 expenses stabilise. The expenses can be compared with the total
insured income of the working population, which at the moment amounts to € 175 billion
(2004 estimation). The expenses in figure 2.6 are net expenses, which should be increased by
about 25% for cost of executing the disability law, cost of reintegration etc. Taking into ac-
count the growth of the workforce, the burden of disability benefits then drops by about 1.4%
from 7.1% to 5.7%.
84 chapter 2
Figure 2.6 Projected total cost of all disability related benefits before (WAO) and after
system change (WIA), 2003-204218
Comparing the microsimulation model results to a macro cohort model used to predict
the time path of the savings, the expected savings of the system change in the early years
surprisingly appear to have been predicted too high. The net savings as a percentage of the
long-term savings after four years are predicted to be 29% with a simple macro model and
appeared to be only 21% in the microsimulation model. Again, the use of means appears
to have played an important role. Using average durations of benefits, the new popula-
tion seemed to replace the old population quite rapidly, whereas using individual recovery
chances that depend on the duration of the benefit, the new population appeared to build up
more slowly, so that savings come later than expected.
18 Including for the WIA the benefit costs of the ones that would have qualified for the WAO but do not qualify for the WIA.
2.5.2 Micro results
The population model indicates the macro-effects of system changes and their development
in time. Isolating one or more cohorts (so in fact reducing the population model to a cohort
model), an estimation can be given of the direct income effects of the system change for a
cohort of new benefits. This indicates the effects on the long-term incomes of individuals in
order to find the “winners” and “losers” of the system change and the amount they win or
lose. Winners in our article are defined as those individuals that improve financially in their
situation, without taking loss of leisure into account. For every person in the simulation, total
discounted individual income during the disability period is measured. The real discount rate
is fixed at 10% per year, based on assumptions used by the CPB (CPB, 2002). The accumu-
lating period is maximised to 20 years, because not much valuable information is gathered
from additional years since discounting weights decline. For all people that enter the new
benefit system from 2006-2020, total discounted individual incomes are estimated in both the
old and the new system. This number, divided by the accumulated discounted income before
disability, is the replacement rate as mentioned in section 2.2.3. Table 2.5 gives average re-
placement rates for the six main subgroups in the system (the grey boxes from figure 2.1).
Table 2.5 Projected replacement rates by subgroup before (WAO) and after system
change (WIA)
Subgroup Replacement rate Replacement rate Projected income
WAO WIA mutation
Permanently, fully disabled 72% 73% +1%
Temporarily, fully disabled 73% 74% +1%
Partially disabled, working 82% 84% +2%
Partially disabled, unemployed, one-income 66% 62% -4%
Partially disabled, unemployed, second-income 60% 47% -13%
Partially disabled, <35% 77% 65% -12%
All disabled 72% 72%
86 chapter 2
The development of the replacement rates clearly points out how the incentives of the new
system work out. The fully disabled and the partially disabled that are working increase
their replacement rate by one to two percentage points whereas the partially disabled who
are unemployed have a considerably lower replacement rate. This holds especially for the
unemployed with earning partners because they will not receive social assistance. These
figures show that there is an incentive for partially disabled to work more. This incentive to
work leads to an increase in the number of working disabled and a decrease in the number of
unemployed disabled. The average replacement rate for all disabled is therefore stable. This
is consistent with the macro results presented earlier, which include a € 1.9 billion drop in
benefit costs and a € 1.7 billion rise in wages19. Figure 2.7 shows the income development of
the partially disabled in the new system. The figure shows the total accumulated income for
each person in the simulation in the old system (x-axis) and the new system (y-axis)20.
Figure 2.7 Projected income development partially disabled before (WAO) and after
system change (WIA)
19 The replacement rate is stable although the drop in cost of benefits exceeds the rise in wages. This can be explained by
the time path of these mutations: apparently the rise in wages on average takes place earlier than the drop in benefits.
20 A dot on the x=y-axis means that no income mutation took place, dots to the right of the x=y-axis point to income de-
creases and points to the left of the x=y-axis to income increases.
Dots above the x=y-axis belong to the working partially disabled. The highest income
increases are among the partially disabled who would not be working in the current system,
but will be working in the new system because of its better incentive structure. But also the
partially disabled who would be working now and would do so in the new system as well can
belong to the winners of the system change. Every disability benefit is replaced by a lower
second phase disability benefit after a while, but for a large majority of employees the dif-
ference between both is compensated by collective insurances. However, for the small group
that is not compensated now, there will be an income rise in the new system as the lower
second phase disability benefit is abolished for the working.
Dots below the x=y-axis belong to the unemployed partially disabled. Two categories of
partially disabled in particular are at risk: the higher incomes and the two-income couples
which have a long duration of disability. The steepest falls in individual income occur in
these groups. For the partially disabled who are not working, there will be a first phase ben-
efit which is as high as in the current system. During this phase, people have the opportunity
to find a job. When first phase benefits end, the second phase benefit equals a share of the
social minimum level, corresponding to the grade of disability. High incomes will lose the
most; low incomes can often keep their current benefit level. That is, if they are breadwinners
or singles. When there is a second income in the household higher than the social minimum
level, supplementary social assistance is not available.
There is a considerable difference in the spread of income mutations among lower total
incomes as compared with higher total incomes. Lower total incomes belong to either low
incomes or short durations of disability. In both cases the new system will usually not cause
such a large difference. In the short run unemployment benefits tend to have about the same
level as disability benefits. Besides, in case of breadwinners and singles, if the total income is
below the social minimum, it will be topped up to the social minimum level, both in the cur-
rent system and the new system.
88 chapter 2
2.5.3 Sensitivity analysis
The simulation used in the budget forecasts is a baseline scenario with some important
assumptions concerning the development of the labour force, the benefit elasticity and the
discount rate. In order to test for the sensitivity of the simulation results a sensitivity analysis
is performed using different assumptions.
First, the development of the labour force is based on one of the four scenarios for the Neth-
erlands by CPB (Huizinga and Smid, 2004), called Global Economy. This scenario has the
strongest growth of the labour force. The scenario assumes a continuing economic growth
and a strong rise in labour participation among the elder working force and among women.
This scenario assumes successful international cooperation and a strong focus on individual
responsibilities. An alternative scenario, called Regional Communities, is one in which
economic growth stagnates and the labour force shrinks compared with the current level. This
scenario assumes national sovereignty instead of international cooperation and a strong role
for the public sector. All four scenarios contain underlying assumptions about the develop-
ment of the disability benefit law, so use of the scenario data in order to make projections
of the number of disability cases should be done with care. The Global Economy scenario
was previously used in partial macro computations of the system change (SER, 2004) and
assumes a strict disability benefit system with a decreasing number of disability benefits. Re-
placing Global Economy with another scenario and assuming no change in the disability sys-
tem reduces the expected number of disability cases in both the old and the new system and,
as a consequence, the savings of the new system. Indicative data, using the Global Economy
scenario as a base scenario with an index of 100%, are presented in the following table.
Table 2.6 Sensitivity analysis on labour market scenarios (2040)

Labour market Labour force Disability benefits Disability benefits Savings of system
scenario (x 1 million) WAO WIA change
Global Economy 9.0 (100%) 100% 100% 100%
(base scenario)
Transatlantic Market 7.9 (88%) 90% 92% 81%
Strong Europe 7.6 (84%) 89% 90% 85%
Regional Commu- 6.6 (73%) 79% 79% 76%
nities
The table shows that in the Regional Communities scenario, the other extreme, the labour
force is 27% smaller than in the Global Economy base scenario in 2040 (see table 2.6, 73%
compared with 100%). The number of disability benefits in the simulations of both the
current and the new system are 21% smaller. As a consequence, the savings of the system
change in the Regional Communities scenario decreases by more or less a similar amount
(about 24%). However, one key assumption in this scenario is a strong public sector and an
extensive system of social security excluding the stricter disability measures currently being
proposed. Therefore, this scenario is not consistent with the introduction of the WIA.
Second, in the baseline scenario it is assumed that the benefit elasticity is 1.5, based on a
micro economic Canadian study on two distinct Disability Insurance programs in Quebec
and the rest of Canada (Gruber, 2000). However, this may not be accurate for the current
Dutch situation. Recent results of a meso economic study done at the CPB (Van Vuren and
Van Vuuren, 2007) based on Dutch data from 1996-2000, suggest the elasticity to be at least
2.5. Therefore an alternative scenario with an elasticity of 3.0 is tested. On the other hand, the
level of the elasticity may well be dependent on the strictness of the entrance examinations.
If entrance criteria are loose, medical reasons are less important and financial considerations
can play a more significant role. At the moment, entrance criteria are implemented rather
strictly. The number of new disability benefits in 2003, the base year for the simulation, is
one third lower than the yearly number of new disability benefits at the end of the 1990’s,
the time period Van Vuren and Van Vuuren studied. Therefore, an alternative scenario with a
lower elasticity of 0.75 is also tested.
90 chapter 2
Table 2.7 Sensitivity analysis on benefit elasticity (2040)

Elasticity Number of working disabled Total cost of benefits after system
after system change change
1.5 (base scenario) 100% 100%
0.75 95% 102%
3.0 119% 95%
Using an elasticity of 3.0 increases the financial incentive to work. As a consequence, the
number of disabled that are working increases by 19% and the total cost of benefits are
reduced by 5% (see table 2.7, 95% compared with 100%). By using an elasticity of 0.75 the
opposite happens, the number of working disabled is reduced by 5% and the total cost of
benefits increases by 2%.
Third, the elasticity used is a uniform one, which is applied to all individuals in the simula-
tion. The value of 1.5 clearly points out sizeable efficiency losses from a rise in disability
benefits. However, the assumption of a uniform elasticity may be questionable, because some
groups will value the leisure effect different from other groups. It is, for example, well known
from the labour supply literature that two-income households have higher labour supply
elasticities than one-income households (e.g. Blundell, 1992). Therefore, in the sensitivity
analysis a scenario is tested in which for breadwinners and singles the lower elasticity of 0.75
is assumed and for second incomes the higher elasticity of 3.0.
Table 2.8 Sensitivity analysis on benefit elasticity heterogeneity (2040)

Elasticity Number of working disabled Total cost of benefits after system
after system change change
1.5 (base scenario) 100% 100%
0.75 / 3.0 110% 99.6%
An interesting outcome results from this sensitivity analysis. The number of working disabled
increases by 10% whereas the total cost of benefits remain almost stable. The lower elastic-
ity leads to a higher benefit use of breadwinners and singles that have, on average, higher
benefits. The lower elasticity leads to a lower benefit use of second-incomes, who have, on
average, lower benefits. Although the number of working rises, the costs of benefits do not
fall at all. The participation growth is a logical consequence of the higher elasticity being
applied to the group with the largest income decrease (see table 2.5, the unemployed partially
disabled second-incomes).
Fourth, the real discount rate is set at 10% based on estimates used in earlier studies by the
CPB (e.g. CPB, 2002). In these studies the real discount rate is a function of the expected
nominal interest rate, inflation, life expectancy and an uncertainty factor. The discount rate
influences the behavioural effects in the following way. A lower discount rate increases the
weight of income changes in later years. As the financial incentives of the new disability
regime are most visible in later years, using a lower discount rate would increase the changes
in replacement rate and increase the labour participation of the partially disabled further. An
alternative scenario with a discount rate of 5% is tested. This only implies minor changes in
the replacement rates.
The financial incentives are aimed especially at the non-working partially disabled. This
subgroup will, on average, notice the lower benefit after two years. Assuming an average
benefit duration of ten years, using a discount rate of 5% instead of 10% reduces the weight
of the first two years in the total income by five percentage points and increases the weight of
the later years by five percentage points. This change for only one subgroup appears to be too
small to substantially influence the main results21.
2.5.4 Redistribution effects of the regime change
In the previous section we found that the average replacement rate for all disabled is stable,
because the decrease in accumulated benefits is compensated by an increase in wages. The
simulation data can be used to assess the redistribution effects of the reform by calculating
21 The use of a very high discount rate would, of course, influence the results. For example, in a theoretical scenario with a
discount rate of 100%, the financial incentives would disappear altogether. In such a scenario, only the benefit change in
the first year would influence the replacement rate.
92 chapter 2
the Gini-coefficient and consider the effects of the regime change for different deciles or
quartiles of gross income like in e.g. Spadaro (2003). For this analysis Gini-coefficients are
computed before and after the reform, and also for the main parts of the reform separately.
The Gini-coefficients are computed on average discounted total incomes of each person in the
simulation22. Also, the quartiles of the income distribution are given (also computed on aver-
age discounted total incomes and as indices with the old system being 100). First, the results
are presented for the three main components of the new system as described in the beginning
of section 2.5.1. Also, the results are presented for the benefit increase for the fully disabled
which is scheduled for 2008 (see footnote 7).
Table 2.9 Changes in Gini-coefficients

Gini-coefficient Q1 (Index) Q2 (Index) Q3 (Index)
Old system 0.321 100.0 100.0 100.0
• Eligibility criteria +0.015 -3.4 -1.5 0.3
• Exclusion of lower two classes -0.005 -2.1 -1.6 -5.4
• Changes in benefit regime partially disabled -0.011 6.9 5.0 5.6
New system 0.320 101.4 101.9 100.5
• Benefit increase fully disabled -0.004 2.8 2.2 1.4
Future system incl. benefit increase 0.316 104.2 104.1 101.9
In general, inequality decreases slightly. The change in eligibility criteria has the largest re-
distribution effects. This measure increases inequality as benefit levels decrease for a substan-
tial part of the group of non-working while at the same time a substantial part of the group
of working is able to increase its total income. Here the first quartile of the income distribu-
tion decreases the most. Exclusion of the lower two classes of disability actually decreases
inequality. This is the result of the unemployed being more frequently compensated for their
loss of benefit by receiving additional social assistance than the working. As a consequence,
here the third quartile of the income distribution decreases the most. The changes in the
benefit regime for the partially disabled decrease inequality. The latter is the result of the
22 The Gini-coefficients are computed on the simulation population of disabled only, so not taking into account redistribu-
tion changes for the rest of the population (for example as a result of the decrease in the contributions to the disability
scheme).
behavioural effects of these changes. Part of the persons affected by the changes in the benefit
regime compensates the loss of income by starting to participate on the labour market. Also
the benefit increase for the fully disabled decreases inequality. This is a consequence of the
lower incomes being overrepresented among the fully disabled.
As stated in Bourguignon and Spadaro (2006), including behavioural response in a micro-

simulation model allows also for an explicit analysis of the equity-efficiency trade-off. It is
shown that the reforms boost efficiency, while slightly reducing inequality. When the fully
disabled get higher benefits, inequality will be reduced further.
2.6 Conclusion
In this chapter a microsimulation model is used to forecast the effects of the system change
in the Dutch disability benefit law, based on an administrative data source from 2003 and
the imputation of other important data. In the chapter different model choices are discussed.
It is shown that model choice depends on the type of problem and that a dynamic popula-
tion model fits the case of the disability benefits the best. It is shown that the introduction of
the proposed system change decreases the number of disability benefits by more than 25%
from 2020 onwards and reduces total costs by almost € 2 billion or 20%. Based on the better
incentive structure, participation will increase and boost GDP by a similar amount so that, on
average, replacement rates remain stable. Overall, the Gini-coefficient slightly decreases due
to the regime change. When the fully disabled get higher benefits, inequality will be reduced
more substantially. Moreover microsimulation can be used to pick the winners and losers of
the new system. It is shown that for almost all partially disabled that are working, the total
discounted income after the system change is as large as or larger than before the system
change and for the non-working it is the reverse. This clearly points out the direction of the
system change towards more participation.
94 chapter 2
The model provides an improvement as compared with earlier macro analyses, especially
for detailed short-term forecasts and for insight into long-term effects of policy measures for
subgroups of disabled. Two results from the microsimulation model lead to an alteration of
the forecasts. First, when comparing the microsimulation model to a macro cohort model, the
savings of the system change in the early years are overestimated and are reduced. Second,
comparing the microsimulation model to the macro model, the long-term savings appear to
be underestimated. It is concluded that microsimulation appears to be a superior forecasting
tool as compared with the conventional macro models in a complex and dynamic environ-
ment, especially because of the in depth information of its forecasts.
However, there are many topics for future research. A first important issue concerns the
parameters used in predicting the behavioural effects. The benefit elasticity is based on
estimation of the reaction on a system change in Canada. A Dutch study on data from the late
1990s suggests a higher elasticity. However, the validity of both elasticities for the current
Dutch situation is questionable as entrance criteria have tightened substantially already prior
to the reforms. Moreover, a uniform elasticity for all disablement beneficiaries is used, so
that, although our behavioural model takes leisure into account implicitly, the model cannot
differentiate between individual differences in valuation of labour and leisure. It would be
interesting to investigate whether different elasticities can be estimated for different groups of
disablement beneficiaries. For example, different elasticities are to be expected for breadwin-
ners and second incomes. It is shown in the sensitivity analyses section that the impact on
the labour participation of such a differentiation in elasticities can be substantial. Moreover,
measurement of these different elasticities will allow, like Creedy and Kalb (2005) suggest,
the effective use of a more general money metric welfare measure rather than simply an
income measure.
Second, the model is a partial equilibrium model that gives a static picture of the effect of
the regime change only. Not considered is the effect that extra participation will have on
the labour market (wages, unemployment) as in the model the labour market is taken to be
exogenous. Nelissen (1991) also remarks that microsimulation models cannot directly be
used to process these so-called third order effects and suggests constructing a macro model
around the micro model. Bourguignon and Spadaro (2006), however, give some suggestions
to extend a partial equilibrium model to a general equilibrium model.
A third topic for future research concerns the data used. Some exogenous variables, like size
and composition of the labour market, have a large influence on the absolute results of the
simulations, although the relative results (e.g. the savings of the new system as compared
with the old system) are affected less. Also in the current approach the existence of partner
income is imputed by a sample of new disability benefits with a high level of aggregation
because for the subpopulation of disabled more detailed data are not yet known. When more
detailed data become available, improvements can be made by using a more detailed imputa-
tion method (like in Nelissen, 1995b or Sutherland, Taylor and Gomulka, 2002) or by using a
model for the labour market situation of married and cohabiting partners (e.g. Henkens, Grift
and Siegers, 2002). The appropriate level of the discount rate is another subject for further
research but appears to be less important for the results of this particular system change.
96 chapter 2
Appendix 2.A The behavioural model
In this appendix, the behavioural model from section 2.4.4 is given. The behavioural model
can be described as follows. We start by distinguishing between the main groups in the
system: the fully disabled, the >35% partially disabled and the <35% partially disabled.
The fully disabled are assumed not to be able to work, the groups of partially disabled are
subdivided in working and non-working. Here we see, like Creedy and Kalb (2005) mention,
that behavioural microsimulation models have a lower degree of population heterogeneity
than non-behavioural models because econometric estimation of the important relationships
involve the use of a limited range of categories. Per group (j) the following variables are
defined:
εj : the income elasticity of group (j)
δt : the discount rate in year (t) = (1-δ)t
PIi : the pre-injury income of person (i), the annual salary earned by the person (i) at the
time of the injury
Ii,t : the total income of person (i) in year (t), the total of disability benefit, employer’s sup-
plement, wage, unemployment benefit, supplementary benefits and social assistance.
Nj : the number of benefits in group (j)
The aggregate behavioural effect for group (j) can be computed by computing the change in
group replacement rates and multiplying by the group elasticity:
(2.1) Total discounted income of group (j) in the current system:
(2.2) Total discounted income of group (j) in the new system:

98 chapter 2
(2.3) Total discounted pre-injury income of group (j):
(2.4) Average replacement rate in current system:
(2.5) Average replacement rate in new system:
(2.6) Number of benefits of group (j) in the new system:
Because of our assumption that the change in replacement rate does not influence the number
of benefits, the group totals have to be adjusted as follows:
(2.7)
(2.8)
(2.9)
If the replacement rate of group (j) decreases because of a policy measure (so if RRj’ < RRj),
the number of working partially disabled in group (j) increases by (Nj’-Nj ). As the total
number of benefits in group (j) does not change, the number of unemployed partially disabled
in group (j) decreases by the same amount.
Appendix 2.B The benefit formulas
In this appendix, the formulas from figure 2.3 are explained.
Disability benefit (WAO)

The level of the disability benefit for individual (i) at time (t) is determined in the first phase,
based on grade of disability (gd) and personal wage (pw) previously earned before disability:
(2.10)
Equation (2.10) represents the wage in the first phase of the benefit. The duration of this ben-
efit depends on the age of the recipient upon becoming disabled.
The level of disability benefit in the second phase is based on a supplement on top of seventy
percent of the minimum wage level. This supplement is equal to seventy percent times two
100 chapter 2
percent of the difference between personal wage and the minimum wage (mw) times the
number of years the recipient, on the day of sickness, is older than 15 years of age.
(2.11)
Unemployment benefit (WW)

The unemployment benefit is dependent on the working status (ws) of recipient (i) at time (t),
the age, the grade of disability (gd), the working history (wh) and the level of personal wage
when becoming disabled (pw):
(2.12)
The duration of the unemployment benefit depends on the working history as follows:
If a person has a grade of disability of at least 80% with a disability benefit, then he is
excluded from receiving an unemployment benefit. Recipients above the age of 65 are also
excluded. The working history is the total number of working years which determines the
duration of the benefit.
Supplementary Benefits Act (TW)

The level of supplementary benefit for individual (i) at time (t) is determined by the differ-
ence between the total amount of disability benefit/unemployment benefit/wage (toti,1) and
the social minimum, which equals 70% of the minimum wage (mw) for a single, 90% of the
minimum wage for a single with a child below the age of 18 and 100% of the minimum wage
for a breadwinner:
(2.13)
However, the supplementary benefit is limited by 2 ceilings. First, the supplement cannot
exceed 30% of the benefit itself. Second, the total benefit, including the supplement, cannot
exceed the previous wage.
Provision for Older and Partly Disabled Unemployed Employees (IOAW)

The level of IOAW benefit for individual (i) at time (t) is also determined by the difference
between the total amount of disability benefit/supplementary benefit/wage (toti,2) and the
social minimum:
(2.14)
102 chapter 2
3
Estimating the effects of recent
disability reforms in the Netherlands23
Disability rates in the Netherlands used to be among the highest in the world. Although
already during the 1980s and 1990s many different policy measures were tried, no solution
seemed in sight until 2002 when the number of disability recipients approached 1 million.
However, since then the number of disability cases has dropped remarkably due to a number
of very successful policy changes, the last of which was the introduction in 2006 of a whole
new disability law.
This chapter consists of two parts, both using administrative datasets of Dutch disability ben-
efit recipients from 1999 until 2009. In the first part, a regression approach using aggregated
data is used in order to estimate the immediate effects on disability inflow and outflow of the
recent policy measures. In the second part, a long-term forecast of the disability stock is made
using a microsimulation model in which recovery and mortality rates and transitions within
the disability schemes are estimated with multinomial logistic regression models. By compar-
ing microsimulation forecasts with different base years the long-term effect of the whole set
of policy measures can be estimated as well.
23 A short Dutch language version of this chapter has been published in Economisch Statistische Berichten, Vol. 95 (Van
Sonsbeek and Gradus, 2010)
104 chapter 3
It is shown that, without the reforms, the number of disability benefit recipients would have
grown to 1.2 million in the long run. Three policy changes introduced between 1998 and 2004
(experience rating, gatekeeper protocol and stricter examinations) have had a large combined
effect, reducing the long-term number of disability benefits by 600,000. In the long run,
the introduction of the new disability scheme (WIA) in 2006 further reduces the number of
disability benefits by almost 250,000. However, part of the effect of the scheme is artificial
because costs are transferred to the payroll of individual employers. Moreover, in recent years,
inflow into the disability scheme for the young handicapped (Wajong) has almost doubled,
which in the long run increases the number of Wajong benefit recipients by over 200,000, so
that the total number of disability benefit recipients will not decrease substantially anymore.
Furthermore, it is shown that the effects of re-examination of the disability stock are substan-
tial in the short run but may be overcompensated in the long run by the adverse effects of the
re-examination operation and the simultaneous changes in the re-examination periodicity.
3.1 Introduction
Disability rates in the Netherlands used to be among the highest in the world. In 2002 the
number of disability benefit recipients approached 1 million. At the time of writing, eight
years later, the Netherlands is considered one of the prime examples of effective policy re-
form in the disability scheme (Prinz and Tompson, 2009). The number of disability recipients
is still high, but yearly inflow into the scheme has dropped to one-third of the levels that were
common until 2001. Between 1998 and 2006 a series of reforms has taken place. Although it
is clear that the combined effect of these reforms is very substantial, up to now there has not
been any conclusive research investigating the individual and combined effects of the reforms
simultaneously. This chapter tries to fill this gap using administrative datasets of all disability
benefit recipients in the Netherlands from 1999 until 2009.
Estimating the effects of recent disability reforms in the Netherlands 105
In the first part of this chapter, the immediate effects of the individual inflow- and outflow-
related policy measures on disability inflow and outflow are assessed, controlling for business
cycle and seasonal effects, with a regression model using aggregated data. The effects of the
individual measures are quantified both in a general model and in age/gender- and sector-
specific models since the measures may affect different demographic groups and sectors in
different ways. In the second part of this chapter, a long-term forecast of the disability stock
is made using a microsimulation model with multinomial logistic regression estimates for the
transitions within the disability schemes based on the same administrative dataset. As all the
policy measures considered in this analysis were effective in 2009, the 2009 simulation gives
the most accurate long-term forecast under an unchanged policies scenario, thus including
the effects of all policy reforms considered in the first part of this chapter. The combined
long-term effects of the policy measures on the disability stock are assessed by comparing
the 2009 simulation with a simulation using administrative data from a base year before the
policy measures became effective.
The structure of this chapter is as follows. In section 3.2 a brief overview of the Dutch dis-
ability system and a history of the most important past policy measures are given. In section
3.3, the developments of the inflow and outflow rates are modelled in order to disentangle the
individual effects of the various policy reforms using the macro-regression approach. Sec-
tion 3.4 summarizes the microsimulation model that is used in this chapter. In particular, the
modelling and estimation of the inflow into and the outflow out of the disability schemes are
presented. In section 3.5 the long term forecast is presented and the combined effect of the
policy reforms is estimated. Section 3.6 concludes with some policy recommendations and
topics for further research.
106 chapter 3
3.2 The Dutch disability insurance schemes
In this section, a brief overview is given of the Dutch disability insurance schemes and the
major reforms that have taken place since their introduction. As the observation period of the
data used in this chapter runs from 1999 to 2009, the reforms are split into two parts. Section
3.2.1 focuses on the policy measures taken before the end of the 1990s, which are not includ-
ed in the analysis in this chapter. Section 3.2.2 focuses on the more recent policy measures,
starting with the introduction of premium differentiation for employers in 1998. Section 3.2.3
presents an overview of the literature on the effects of the recent reforms.
3.2.1 History and past policy measures until the end of the 1990s
The Netherlands used to have three separate disability schemes covering employees (WAO),
the self-employed (WAZ) and the young disabled (Wajong). The WAO was in force from
1967 onwards and the schemes for the self-employed and the young disabled (individuals
who become disabled before the age of 18 and therefore had no chance to access the regular
schemes for employees and self-employed) were both introduced in 1976. The WAZ scheme
was abolished in 2004 as the government then decided to consider disability a risk that the
self-employed could insure on the private market. Therefore in this chapter the WAZ will
be left out of the analyses. In 2006 the WAO was replaced with a new disability scheme for
employees, the WIA.
The WAO integrated schemes for occupational diseases and work injuries (risque profes-
sionnel) and other disability causes (risque social), in one scheme, combining the generosity
of risque professionnel schemes with a broad risk definition as is common in risque social
schemes. The number of beneficiaries in the WAO grew rapidly and soon the scheme was
found to be unmanageable. In fact, as De Jong (2008) concludes, the WAO was misused
twice to accommodate social change. Firstly, during the 1970s and 1980s the scheme support-
ed the change from an industrial society to a service-oriented society. Massive lay-offs in the
industrial sector were partly transferred to the disability scheme, which provided a generous
and permanent wage-related benefit. The WAO started to function as an early retirement route
(see also de Vos and Kapteyn, 1997). Secondly, according to De Jong, during the 1990s the
WAO supported Dutch households in their transformation from a traditional single breadwin-
ner type to a modern dual-earner type.
Already since the early 1980s several policy measures had been introduced in an attempt
to reduce the number of benefits. During the early 1980s the net value of the benefits was
decreased. Replacement rates fell from 90% to 75% (De Jong, 2008) but were still extremely
high by international standards (at least for non-work-related injuries). Until 1987, the partial-
ly disabled were allowed full benefits, accommodating a lot of hidden unemployment within
the scheme (Koning and Van Vuuren, 2010). From 1987, the partially disabled received only
partial benefits. However, during the whole history of the WAO the large majority of benefi-
ciaries were fully disabled. In 1991 a controversial scheme of premiums and fines was intro-
duced. Finally, in 1993, after political turmoil, a new law was adopted that proved to be the
first really successful reform. Because of this law, from then on at the entry examination all
possible job alternatives were taken into account. As a consequence, inflow into the disability
scheme dropped substantially from 93,000 in 1993 to 64,000 in 1995. Also, all the existing
disability benefit recipients were re-examined, those aged 45 and below against the new cri-
teria. Finally a two-phase benefit system was introduced, starting with a wage-related benefit
followed by a lower, age-dependent follow-up benefit. However, the new entrance criteria
were considered by many to be too strict and a lot of negative publicity caused the then State
Secretary to instruct the benefit administration office to relax the entrance criteria in practice.
Moreover the two phase benefit system never really worked in practice because employer and
employee organizations agreed on collective insurances, covering the gap between the initial
wage-related benefit and the follow-up benefit. From 1996 on, inflow started rising again
very rapidly until a new record high in 2001 of 104,000 new entries. Figures from the OECD
108 chapter 3
(2003) show that the Netherlands was among the OECD members with the highest rates of
disability benefit recipiency and inflow of new recipients. In particular, the high inflow rates
of women and the high overall benefit recipiency among the elderly were considered worry-
ing given the oncoming ageing of the population. The decrease since then has been impres-
sive. Current inflow levels are about one-third the level of around the turn of the century. The
OECD (2008) concludes that this success is a consequence of a series of very comprehensive
reforms, characterized by a shift of responsibilities to employers and employees, a tightening
in benefit eligibility and generosity, and a (partial) privatization of hitherto public schemes.
Figure 3.1 shows the development of the disability stock and inflow since 1968.
Figure 3.1 Disability (WAO+WIA) stock (left axis) and inflow (right axis), 1968–2009
Source: UWV, Statistische Tijdreeksen UWV 2008.

3.2.2 Recent policy changes
The main recent policy reforms are the following:

• Experience rating and private insurance (1998 onwards). Until 1998 all employers paid a
fixed premium rate for the disability scheme, regardless of the number of disability cases
they had. Since 1998 two separate premium rates exist, one differentiated rate covering
the first five years of disability and one fixed rate covering the cost of the later years.
The differentiated premium is based upon experience rating and calculated for year (t)
based upon the firm-specific disability costs of year (t-2). The experience rating system
was introduced in five steps, increasing the average differentiated premium to its struc-
tural level in 200324. Meanwhile the experience rating has been extended to ten years but
the permanently fully disabled are exempted.
• Gatekeeper protocol (2002 onwards). This protocol, in force from April 2002, speci-
fies the legal responsibilities of both the incapacitated employee and the employer.
After eight weeks of absence, the employer and employee together should have made a
vocational rehabilitation plan specifying the pathway to recovery (if possible). If later
on a disability benefit is claimed, a rehabilitation report should be provided containing
the original rehabilitation plan and the steps followed since. If reintegration efforts are
considered insufficient the disability claim is dismissed.
• Stricter examination criteria (2004 onwards). From 1 October 2004 onwards, the
entrance criteria to the WAO have been tightened. This change was originally meant to
be part of the WIA, but it came into force earlier, capturing the last quarter of the WAO
inflow. The changes are mainly technical, including decreasing the minimum number of
job alternatives that have to be found in order to determine the grade of partial disability
and increasing the number of job alternatives for part-time workers (until 2004 full-time
jobs were not taken into account in determining their grade of disability).
• Re-examinations of the disability stock (2004–2009). Using the same tightened exami-
nation criteria, a large part of the disability stock was re-examined between 1 October
24 In 2003, the experience rating system was abolished for small employers, for whom it was considered to be unfair.
However, the experience rating was re-introduced for small employers from 2006 onwards in the new WIA law.
110 chapter 3
2004 and 1 April 2009. Originally all disability benefit recipients younger than 55 in
2004 were meant to be re-examined, but under political pressure this age limit was de-
creased in two steps to all people younger than 45 in 2004. At the same time, from 1 Oc-
tober 2004 onwards, the re-examination procedure for the disability stock was changed.
Instead of ‘mechanical’ re-examinations after the first, fifth and tenth year of the benefit,
the benefit administration office was given complete freedom in determining whom to
re-examine and when.
• WIA scheme (2006 onwards). The WIA reform itself introduced two different schemes:
one generous scheme (called IVA) without medical re-examinations for the fully disa-
bled who had no chance of recovery and one less generous scheme (called WGA) for
the partially disabled and the fully disabled with some chance of recovery. The WGA
involves frequent medical re-examinations and consists of two phases. During the first
phase, which lasts as long as the unemployment benefit would have lasted for the indi-
vidual claimant, the benefit level equals the unemployment benefit. During the second
phase, a generous wage subsidy is paid to those who work to a sufficient level according
to their abilities, but only a small minimum benefit is paid to those who do not. Also the
minimum grade of disability in this scheme is 35% instead of 15% in the WAO. Preced-
ing the WIA, in 2004, the waiting period for entrance into the disability scheme was
extended from one to two years in order to increase the direct financial incentive for em-
ployers. During this period of incapacity, employers have full responsibility for paying
their employees, so there is a strong incentive for reintegration. Moreover, it was agreed
between government, employers’ organizations and trade unions to create a financial in-
centive for employees to return to work by limiting the sick pay during the second year
of incapacity to 70% of the wage instead of the average of 90% that was common until
then.
These are the major policy changes that have taken place since 1998. They have in com-
mon that they actually changed the disability insurance law and that they directly affect the
disability schemes. Their expected effectiveness was assessed in the financial section of the
respective laws. An overview of literature on their actual effectiveness is given in section

3.2.3. However, other policy changes may have played some role as well.
In the first place, there are some smaller policy changes which are also directly linked to the
disability schemes, but that did not really change the disability insurance laws. An example
of such a smaller policy change is the publication of inflow of disability benefit claimants per
employer for the largest employers in 2003, which tries to influence companies’ behaviour
through “naming and shaming”. This change came into force during the same period that the
gatekeeper protocol was introduced. Also, during this period, the premium differentiation
rates were still rising. Another example of a smaller policy change is the increasing amount
of attention paid to the benefit administration by then State Secretary Hoogervorst between
1998 and 2002. This shift in focus also coincided with the building up of the experience
rating system. The effects of such smaller policy changes are not included in our analysis
because of the difficulty of isolating their effects. This implies that in some cases the effect of
the major policy changes may be slightly overestimated, as combined effects may be meas-
ured.
Secondly, major system reforms may cause announcement effects prior to their introduction.
However, such announcement effects would tend to cause a temporary increase in disability
uptake prior to the system reform, followed by a sharp drop after the system reform, as hap-
pened, for example, when the duration of the unemployment benefit was cut in 2003 (Van
Ours and Tuit, 2010). Such patterns are not observed in disability uptake in the observation
period.
Finally, changes in social security schemes other than disability benefits may have influenced
claims for disability benefits. The unemployment scheme was significantly reformed in 2003
and 2006, reducing the maximum duration of the unemployment benefit from seven years to
three years and two months. Also early retirement schemes have been gradually reformed,
from very generous schemes providing a strong incentive for early retirement to actuarially
112 chapter 3
neutral schemes in most sectors. However, those changes all decreased the generosity of the
“competing schemes”, so they cannot explain a decrease in disability benefit uptake.
3.2.3 Literature review
There is a substantial body of literature on the effects of individual disability reforms. How-
ever, a joint analysis of the individual effects of all the recent measures on the disability rate
has not been made up to now. Previous attempts have been made to explain the development
of sick leave percentage from several policy and economic variables. Lindner and Veerman
(2003) explain the sick leave percentage from 1993 to 2002 from five policy variables plus
an economic indicator (unemployment rate). The problem with such analyses is that the time
series are usually too short for a reasonable explanation given the large number of explana-
tory variables caused by the series of policy measures adopted since the early 1990s. Stege-
man (2005), therefore, in a more elaborate analysis, explains the sick leave percentage from
1980 to 2003 (yearly data) and 1995 to 2003 (quarterly data) from the same policy variables,
plus an economic indicator and several indicators on the composition of the working popula-
tion. In the same study, Stegeman also tries to explain the development of WAO inflow from
the same set of variables. He finds a significant negative effect of the gatekeeper protocol and
a non-significant effect of experience rating, but considers his results to be not stable.
Van Sonsbeek and Schepers (2001) also concluded that it was, at that time, still too early to
find any impact of experience rating on the disability inflow. However, Koning (2004) finds a
substantial impact (15% reduction) of experience rating on WAO inflow, based on an analysis
of micro data from 2000 to 2002. Zwinkels (2007) finds an additional effect of the introduc-
tion of private insurers to the disability insurance market. Employers that opt out of the pub-
lic system reduce their inflow by 11%. However, as only about 20% of the employers opted
out of the public system, and most of them did so during the same years when the premium
differentiation rates were increased to their structural level, it is not possible to disentan-
gle the individual effect of the introduction of private insurers from the available data. The
impact of the gatekeeper protocol on WAO inflow is confirmed and estimated to be about half
of the 43% inflow reduction between 2001 and 2004 by De Jong, Thio and Bartelings (2005).
These researchers explain 35% of the 43% reduction from the gatekeeper protocol, 9% from
premium differentiation and 14% from the business cycle. The remaining 42% is unexplained
but attributed at least partly to elements of the gatekeeper protocol not taken into account in
their analysis. Therefore, the gatekeeper effect amounts to somewhere between 15% and 33%
according to this study. Their business cycle effect is rather surprising: disability inflow de-
creases when unemployment is higher. This is contrary to the experience in the Netherlands
during the 1980s but consistent with the findings of Stegeman (2005) on the relationship
between sick leave and the business cycle. The impact of the stricter examination criteria is
quantified by UWV (2009) and estimated to reduce inflow by 9%. The WIA itself (including
the extension of the sickness period) reduces inflow by 38% in terms of inflow level for 2004
or 21% in terms of inflow level for 2000 (Berendsen, Mulders and Van Loo, 2007), mainly
because of the lower number of claims because of the extension of the sickness period and
the increased entrance level.
On outflow-related policy measures, Van der Burg and Prins (2010) conclude that, between
2005 and 2009, 345,000 WAO beneficiaries were re-examined. For 110,000 beneficiaries the
re-examination caused a decrease or termination of the benefit.
On combined long-term effects of the reforms, Van Ewijk et al. (2006) estimate the budgetary
effects of the reduced inflow, in combination with its favourable effect in labour participation,
to improve the sustainability balance with 1.0% of GDP (€ 5 billion).
114 chapter 3
3.3 Disentangling the effects of inflow-related policy measures
In this section the immediate effects on disability inflow and outflow of the individual policy
measures since the late 1990s are disentangled. It starts with a description of the micro data
sources used in this chapter in section 3.3.1. In section 3.3.2 the models for analysing the
effects of the inflow-related and outflow-related policy measures are discussed. In sections
3.3.3 and 3.3.4 the estimation results of the two models are presented.
3.3.1 Data
The data used in this analysis – micro datasets of the disability schemes since 1999 – were
supplied by UWV, the benefit administration office. The datasets contain both the flows (all
benefits that started or terminated during the year) and the stock of benefits by the end of the
year. Although the administrative data sources are complete and accurate, the data are mostly
limited to the information that is needed for determining and paying the benefit. These in-
clude, for example, birth date, wage, grade of disability, start date of the benefit and termina-
tion date of the benefit.
The datasets for the WAO, WAZ and Wajong schemes are all available for 1999–2009.
The datasets for the WIA scheme are available for 2006–2009. Because the start dates and
termination dates of the benefits are known, quarterly data can be constructed from the yearly
datasets. This may help to overcome the problem of a time series that is too short given the
number of explanatory variables25. Also, the time series is corrected for administrative distur-
bances. This procedure is described in appendix 3.A. Figure 3.2 shows the development of
the corrected inflow rate (as a percentage of the working population) over time.
25 All regressions have been performed as well using monthly data instead of quarterly data, in order to extend the time
series further. However these analyses yielded similar results.
Figure 3.2 WAO+WIA inflow rate on cohort basis, 1999Q1–2009Q3
When studying the raw inflow data, some important trend breaks can be seen, especially in
2002 and 2003 and between 2004 and 2006. This suggests a large impact from the gatekeeper
protocol (that started in 2002Q2) and the WIA (that started in 2006Q1) and only minor
impact from experience rating and the stricter entrance examinations from the last quarter of
2004 onwards. When measured in fully disabled equivalents (fde), the WIA effect is smaller,
which is a logical consequence of part of this effect being mainly caused by the abolition of
the two lower classes of disability. From 2006 onwards, as can be seen from the figure, the
inflow rate in fde is almost equal to the inflow rate in persons, because of a significant rise
in the average grade of disability. Figure 3.3 shows the development of corrected outflow
rate (as a percentage of the DI stock) over time. Outflow because of demographic reasons
(pensioning and mortality) is excluded because this outflow is not influenced by the policy
reforms.
116 chapter 3
Figure 3.3 WAO+WIA outflow rate on cohort basis, 1999Q1–2009Q3
Outflow rates are stable from 1999 until 2003 and decrease slightly in 2004. Outflow rates
then increase substantially from the first quarter of 2005 onwards, which seems to be a logi-
cal consequence of the re-examinations taking place at that time. However, the effect of the
re-examinations seems to slow down after a while. Already during 2006, outflow rates drop
below the 2004 levels. In 2009, outflow rates are down to less than half the level of the begin-
ning of the century.
3.3.2 Model
We try to explain the inflow and outflow rates from a set of explanatory variables includ-
ing the policy variables, the business cycle and seasonal effects. Composition effects of the
working population are not corrected for. It is known that characteristics like gender, age and
ethnicity are relevant to inflow in the disability schemes, but including composition variables
(like the share of women and immigrants in the insured population and the average age of the
insured population) very much complicates the analysis as all three variables, like some of
the policy variables, have a slowly increasing trend over time. Instead separate analyses are
performed on the relevant demographic subgroups in order to examine whether the different
policy measures have different effects on men and women, or on younger and older workers.
Note that explaining inflow trends from composition variables is not trivial. Besides the op-
tion of assuming age-specific disability incidence to be constant (implying mean incidence to
rise because of ageing) one can, according to Aarts and de Jong (1992), also assume average
disability incidence to be constant (implying age-specific incidences to adapt to a changing
composition of the labour force).
The dependent variable in the inflow model is the inflow rate. Until 2004 the inflow rate in (t)
equals the inflow in (t) over the insured population in (t-1). From 2006 onwards, the inflow
rate in (t) equals the inflow in (t) over the insured population in (t-2) as the period of incapac-
ity is extended from one to two years in 2004, so the new beneficiaries in 2006 had their first
day of incapacity in 2004. The working population, excluding the self-employed, is used as a
proxy for the insured population26.
For the explanatory variables in the inflow model we first consider the four main inflow-relat-
ed policy measures, as described in section 3.2.2; experience rating, the gatekeeper protocol,
stricter examinations and the WIA law. The policy variables are 0 when the policy measure is
not yet effective and 1 when the measure is fully effective. In the case of experience rating,
the value of the policy variable is stepwise increased from 0 to 1 between 1998 and 2003
as the premium rates were increased stepwise to a structural level during that period. The
step sizes are based on the average premium level. The value of this policy variable is kept
constant at 1 after 2003 because it can be argued, from the viewpoint of the employers, that
experience rating has been institutionalized since 2003. However, mean premium rates fell
after 2003, mainly because of decreasing inflow. Therefore, a variable is added representing
26 They are almost but not exactly equal. Disability insurance is mandatory for all employees. However, some people
voluntarily join the disability insurance scheme. Also unemployment beneficiaries are insured.
118 chapter 3
the change in annual premium rate since 2003. In the case of the gatekeeper protocol, the
value of the policy variable is stepwise increased from 0 to 1 between the second quarter of
2002 and the second quarter of 2003 as this measure affects all sick employees, so those who
became incapacitated in the second quarter of 2002 (and entered the disability scheme in the
second quarter of 2003) were the first to be fully affected by the measure. The stricter exami-
nations and the WIA law became effective immediately.
The model allows for increasing or decreasing effectiveness of the gatekeeper protocol and
the WIA law by including cross-terms of the policy variables and the time. It can be argued,
for example, that it takes time for employers to get used to the gatekeeper protocol so that ef-
fectiveness increases over time, or that after using the protocol for a while attention decreases
so that effectiveness decreases over time.
Among the explanatory variables we include a business cycle indicator. The literature on
the relationship between worker incapacity and claims for disability benefit and the busi-
ness cycle is ambiguous. Benitez-Silva, Disney and Jimenez-Martin (2010) report strong
evidence of higher total enrolments, lower outflows from disability benefit rolls and, often,
higher inflows into disability rolls in regions and periods of above‐average unemployment,
suggesting the disability schemes being used to cover unemployment risks in bad economic
times. This confirms the experience of the Netherlands in the 1980s. However, in a recent
study in the Netherlands, Stegeman (2005) finds a procyclical relation where economic up-
swings go together with more absence due to illness. Possibly during good times people are
less afraid of losing their jobs and are thus more willing to take time off work when they feel
unwell. Stegeman (2005) tested five business cycle indicators for estimating the effect of the
business cycle on sick leave and found the unemployment rate to be the best indicator. For
sector specific models Stegeman suggested using the vacancy rate as an alternative indicator
as sector specific unemployment rates are not available. Two business cycle indicators are
tested: the unemployment rate, and a composite business cycle indicator as supplied by De
Nederlandsche Bank (DNB) which captures both unemployment and confidence indicators
of producers and consumers (Den Reijer, 2006). Because there is a considerable lag between
the first day of incapacity and the inflow into the disability scheme (one year until 2004 and
two years from 2006 onwards), in the inflow model the mean value of the indicator in the
quarter of the first day of incapacity and the quarter of entry into the disability scheme is used
as a proxy for the average labour market situation during the incapacity period preceding the
entrance to the disability scheme. Both business cycle indicators are found to perform about
as well, with slightly better performance (in terms of higher R2 and better Durbin-Watson test
statistic) of the DNB indicator. The better performance of the DNB indicator as compared
with the unemployment rate, which differs from Stegeman’s findings, may be attributed to
the difference between the disability scheme and the sick leave scheme, the former covering a
longer time span in which future expectations may play a role. As the DNB indicator includes
forward-looking confidence indicators, this is a plausible result. An overview of the values of
the policy variables and the business cycle indicator is included in appendix 3.E.
When using quarterly inflow data seasonal dummies have to be included because there is a
clear seasonal pattern, with the highest level in the first quarter and the lowest level in the
third quarter (see Figure 3.2). This may be a logical consequence of entry into the disability
scheme taking place exactly one year after the first day of sick leave. As sick leave levels
are higher in winter and lower in summer, this pattern also appears in the disability entries.
However, it appears from the data that the seasonal effect grows smaller over time and cur-
rently it is almost non-existent anymore. The seasonal effect grows smaller in particular after
the introduction of stricter examinations and the WIA. Therefore, cross-terms need to be
introduced for the seasonal dummies and the policy variable representing the introduction of
the stricter entrance examinations or the policy variable representing the introduction of the
WIA27. The model (3.1) results:
(3.1)
27 These two cross-terms differ in value in only one quarter (the last quarter of 2004). The policy variables representing
the introduction of the stricter examinations and the introduction of the WIA also differ in value only in the last quarter
of 2004. Therefore, it is not possible to include them all in the regression. As both stricter examinations and WIA
potentially reduce seasonal effects, and from observing the data it cannot be said which one influences the seasonal trend
the most, it is not clear which cross-term should be used in the model. Throughout this chapter, results are given for the
model that includes the cross-term of seasonal dummies and stricter examinations. Results for the model that includes
the cross-term of seasonal dummies and WIA are given in footnotes.
120 chapter 3
where It represents the inflow rate, Pt represents a vector of inflow-related policy variables
(experience rating, gatekeeper, stricter examinations and WIA), Ct represents the business
cycle variable, St represents a vector of seasonal dummies, PtSt represents the cross-term of the
seasonal dummies and the policy variable representing the introduction of the stricter entrance
examinations or the WIA and PtT represents the cross-term of the policy variables and the time.
The dependent variable in the outflow model is the outflow rate. Outflow through mortality
and pensioning are excluded as neither is influenced by disability-related policy measures. The
outflow rate in (t) equals the outflow in (t) over the disability stock in (t-1). The explanatory
variables include the outflow-related policy variables, a business cycle variable, seasonal dum-
mies and, because outflow through recovery is most likely to take place for recent entrants to
the scheme, a variable representing the average inflow rate of the previous eight quarters.
Of the inflow-related policy measures, only experience rating is included as an outflow-

related policy measure as well because employers can reduce their premium by re-integrating
disability benefit recipients. Experience rating indeed was intended both to reduce inflow
(by preventive measures) and to increase outflow (by re-integrative measures). An important
and controversial outflow-oriented policy measure is the re-examination of the disability
stock against the new tighter entrance criteria. This re-examination programme was running
between October 2004 and April 2009. At the same time the re-examination programme
started, the general re-examination procedure for the disability stock was changed. Instead
of ’mechanical’ re-examinations after the first, fifth and tenth year of the benefit, the benefit
administration office was given complete freedom in determining who to re-examine and
when. Although both measures were introduced at the same time and the re-examinations
were finished only just before the end of the observation period, it is possible to give an
indication of the separate effects. The re-examination operation happened to be politically
very controversial and its impact was reduced several times, most notably in 2007, when a
centre-left government took office and the age limit was reduced to 45 years. People aged
between 45 and 50 were re-examined after all, but based upon the old examination criteria.
Those who had been re-examined already could re-enter the disability benefit scheme. As
a consequence, from the second quarter of 2007 onwards, the benefit administration office
focused on correcting the re-examinations of beneficiaries aged 45 and over that already had
taken place so that the re-examinations more or less came to a halt for a while. Therefore, the
values of the policy variable are based on the number of re-examinations per quarter, with the
maximum value in the last quarter of 2005 when the most re-examinations were performed28.
Furthermore, a cross-term is included of the re-examination operation policy variable and the
time period in order to reflect changes in strictness during the re-examination operation, as it
is known from the benefit administration office (UWV, 2008) that the effects of the re-exami-
nations decreased during the operation.
The quarterly outflow data also show a clear seasonal pattern with peaks in the first and
third quarter of each year (Figure 3.3). A possible explanation for the third-quarter peaks
is the availability of seasonal labour in summer, which increases the chances of rejoining
the labour market for disability benefit recipients. Higher outflow rates in winter may have
mainly administrative reasons, like higher inflow in winter causing higher outflow in winter
because the first re-examination used to take place one year after inflow or labour capacity at
the benefit administration office being higher in the first quarter because of a lower number of
holidays. The model (3.2) results:
(3.2)
where Ot represents the outflow rate, Pt represents a vector of outflow-related policy vari-
ables (experience rating, re-examinations of DI stock, change of the re-examination system),
Ct represents the business cycle variable, St represents a vector of seasonal dummies, It-1;t-8
represents the average inflow rate of the previous eight quarters and PtT represents the cross-
term of the policy variable representing the re-examinations of the DI stock and the time.
28 The policy variable equals 0 in the last quarter of 2004. Although re-examinations already took place during that quarter,
benefits were only terminated a minimum of two months after the re-examination, so that almost no benefits were actu-
ally terminated during the fourth quarter of 2004.
122 chapter 3
3.3.3 Estimation results for inflow model
Regression analyses are performed for the inflow rate in both persons and fde-equivalents.
The latter allows for changes in the average grade of disability of the different inflow cohorts.
This way, policy measures that not only affect the number of disability cases but also the aver-
age grade of disability of the cases can be assessed in a better way. In both models, first-order
autocorrelation is present. OLS estimates yielded Durbin-Watson statistics of 1.04 and 1.08
respectively. Autocorrelation is corrected for using the Yule-Walker method of generalized
least squares (GLS). Results using the maximum likelihood method are consistent with the
OLS estimates. Table 3.1 gives the model estimates for the general models of the inflow rate.
Table 3.1 GLS estimates inflow model29

Persons Fde
Intercept 0.455 (0.017) *** 0.373 (0.014) ***
Experience rating -0.061 (0.023) * -0.086 (0.018) ***
Premium rate -0.007 (0.012) -0.001 (0.010)
Gatekeeper protocol -0.101 (0.020) *** -0.075 (0.017) ***
Gatekeeper protocol * Time -0.004 (0.003) -0.002 (0.002)
Stricter examinations -0.058 (0.014) *** -0.039 (0.011) **
WIA -0.105 (0.021) *** -0.059 (0.017) **
WIA * Time 0.004 (0.003) 0.003 (0.002)
Business cycle indicator 0.013 (0.009) 0.005 (0.007)
Quarter 2 -0.068 (0.004) *** -0.050 (0.003) ***
Quarter 3 -0.057 (0.005) *** -0.044 (0.004) ***
Quarter 4 -0.036 (0.006) *** -0.030 (0.004) ***
Quarter 2 * stricter exam. 0.057 (0.006) *** 0.039 (0.005) ***
Quarter 3 * stricter exam. 0.048 (0.008) *** 0.036 (0.006) ***
Quarter 4 * stricter exam. 0.027 (0.007) ** 0.023 (0.006) ***
N 39 39
R2 0.997 0.996
Autocorrelation order (1) (1)
Dw (P-value) 1.65 (0.06) 1.71 (0.08)
*** p-value < 0.001, ** p-value < 0.01, * p-value < 0.05
29 In the alternative model using cross-terms of quarter and WIA (see footnote 27), coefficients for stricter examinations
decrease to -0.031 and -0.016 respectively. The WIA coefficients increase to -0.132 and -0.082 respectively. All other
coefficients remain unchanged.
The regression on inflow in persons shows all four policy variables to have significantly
reduced the inflow rate. Estimates are 13% for premium differentiation30, 22% for the gate-
keeper protocol, 13% for the stricter examinations and 23% for the WIA scheme in the persons
model31. There is no effect of the premium rate in the experience rating system after 2003, nei-
ther is there a significant time trend in the effectiveness of gatekeeper and WIA. The signs of
the coefficients are negative (inflow-reducing) for gatekeeper and positive (inflow-increasing)
for WIA. The business cycle has no significant influence on the inflow. The seasonal dummies
all have significant negative values and as expected, because the seasonal trend seems to have
almost disappeared, the cross-terms of seasonal dummies and stricter examinations have signif-
icant positive values in the same order of magnitude. Similar results follow from the regression
on inflow in fde with slightly higher estimates for experience rating (23%) and slightly lower
estimates for gatekeeper (20%), stricter examinations (10%) and WIA (16%)32. The combined
policy measures have reduced inflow in persons by 71% and inflow in fde by 69%. The three
measures introduced up to 2004 reduced inflow in persons by 48% and inflow in fde by 53%33,
so already before the WIA inflow was down by half as a result of the earlier policy reforms.
The former analysis is repeated for age- and gender-specific subgroups. Four groups are distin-
guished; men aged below 45, men aged 45 and over, women aged below 45 and women aged
45 and over. Again, autocorrelation is corrected for using the Yule-Walker method of GLS. For
convenience, the results are limited to the inflow rate in persons. Results are shown in Table
3.2.
30 That is, 0.061 / 0.455, as 0.455 is the inflow rate in the first quarter when all policy measures are ineffective and the busi-
ness cycle indicator equals 0.
31 In the alternative model using cross-terms of quarter and WIA (see footnote 27), the effect of stricter examinations
would decrease to 7% and the effect of the WIA-scheme would increase to 29%.
32 In the alternative model using cross-terms of quarter and WIA (see footnote 27), the effect of stricter examinations
would decrease to 4% and the effect of the WIA-scheme would increase to 22%.
33 In the alternative model using cross-terms of quarter and WIA (see footnote 27), the inflow reduction caused by the three
measures preceding the WIA would decrease to 42% and 47% respectively.
124 chapter 3
Table 3.2 GLS estimates inflow models for men and women aged >= 45 and < 45
respectively34
Men >= 45 Men < 45 Women >= 45 Women < 45
Intercept 0.629 (0.023) *** 0.216 (0.006) *** 1.114 (0.043) *** 0.509 (0.020) ***
Experience rating -0.087 (0.031) *** -0.034 (0.008) *** -0.191 (0.055) ** -0.115 (0.027) ***
Premium rate -0.019 (0.019) -0.003 (0.005) -0.014 (0.040) -0.010 (0.016)
Gatekeeper protocol -0.113 (0.030) *** -0.038 (0.008) *** -0.288 (0.054) *** -0.123 (0.025) ***
Gatekeeper protocol * Time -0.006 (0.004) -0.002 (0.001) -0.011 90.007) -0.002 (0.003)
Stricter examinations -0.059 (0.022) * -0.019 (0.005) *** -0.151 (0.056) *** -0.063 (0.018) **
WIA -0.159 (0.031) *** -0.056 (0.008) *** -0.164 (0.068) *** -0.117 (0.026) ***
WIA * Time 0.007 (0.004) 0.002 (0.001) 0.011 (0.008) 0.002 (0.004)
Business cycle indicator 0.018 (0.012) 0.003 (0.003) 0.024 (0.024) 0.017 (0.011)
Quarter 2 -0.105 (0.007) *** -0.020 (0.002) *** -0.216 (0.021) *** -0.061 (0.005) ***
Quarter 3 -0.087 (0.008) *** -0.012 (0.002) *** -0.185 (0.021) *** -0.055 (0.006) ***
Quarter 4 -0.055 (0.008) *** -0.006 (0.002) ** -0.122 (0.022) *** -0.038 (0.006) ***
Quarter 2 * stricter exam. 0.082 (0.011) *** 0.018 (0.003) *** 0.185 (0.033) *** 0.054 (0.009) ***
Quarter 4 * stricter exam. 0.031 (0.012) * 0.006 (0.003) 0.096 (0.036) * 0.032 (0.010) **
N 39 39 39 39
R2
0.996 0.998 0.992 0.997
Autocorrelation order (1) (1) na (1)
Dw (P-value) 1.85 (0.12) 2.11 (0.36) 1.82 (0.07) 1.66 (0.05)
Both for experience rating and the gatekeeper protocol, we find a higher impact among
women, both older and younger women. This could be a logical consequence of the inflow
rates among women being well above those of men in the beginning of the observation
period, so that the highest gains from an active re-integration approach of employers could be
made among women. Also, the impact of the stricter examinations is higher among women.
This may be caused in particular by the tightening of the disability entrance possibilities
for part-time workers. The impact of the WIA is higher among men. This is in line with the
observation that men have a higher chance of getting a partial disability benefit as in the WIA
the lower two classes of partial disability were abolished. All four models have insignificant
values of the business cycle indicator and the time effects of gatekeeper and WIA.
decrease to -0.028, -0.014, -0.055 and -0.031 respectively. The WIA coefficients increase to -0.190, -0.062, -0.260 and
-0.149 respectively. All other coefficients remain unchanged.
Finally, the former analyses are repeated for specific sectors of economic activities. UWV
distinguishes 69 different sectors that can be aggregated into ten main groups, based on the
International Standard Industrial Classification of All Economic Activities used by Statis-
tics Netherlands, and for which detailed information on the insured population is available.
These sectors can be expected to have different incentives from the recent policy measures.
For example, the experience rating may be felt less by the public sector which experiences
a less direct and less accountable effect of higher premiums. The gatekeeper protocol can be
expected to affect sectors with larger employers the most, because they can more easily deal
with the administrative requirements of this law. Financial incentives of the WIA may be
felt less by the flexi work sector as flex workers are excluded from experience rating in the
WIA. The results of four typical sectors are presented in Table 3.3 (tables for other sectors are
included in appendix 3.F).
Table 3.3 GLS estimates for sector inflow models35

Construction Financial / Health care Public sector /
business education
Intercept 0.371 (0.010) *** 0.153 (0.012) *** 0.633 (0.032) *** 0.450 (0.059) ***
Experience rating -0.081 (0.011) *** 0.009 (0.012) -0.172 (0.034) *** -0.043 (0.054)
Premium rate 0.001 (0.008) -0.008 (0.007) -0.008 (0.020) 0.001 (0.029)
Gatekeeper protocol -0.035 (0.009) *** -0.042 (0.009) *** -0.164 (0.025) *** -0.165 (0.040) ***
Gatekeeper protocol * Time -0.005 (0.001) *** -0.004 (0.001) * -0.007 (0.004) 0.001 (0.006)
Stricter examinations -0.032 (0.012) *** -0.016 (0.009) -0.083 (0.030) ** -0.091 (0.033) *
WIA -0.079 (0.012) *** -0.051 (0.008) *** -0.069 (0.033) * -0.097 (0.043) *
WIA * Time 0.006 (0.002) ** 0.005 (0.002) * 0.009 (0.004) * 0.000 (0.007)
Vacancy rate 0.000 (0.000) 0.000 (0.000) -0.000 (0.002) -0.001 (0.003)
Quarter 2 -0.040 (0.004) *** -0.028 (0.003) *** -0.098 (0.011) *** -0.109 (0.012) ***
Quarter 3 -0.042 (0.004) *** -0.024 (0.004) *** -0.092 (0.011) *** -0.079 (0.012) ***
Quarter 4 -0.029 (0.005) *** -0.013 (0.004) *** -0.073 (0.012) *** -0.062 (0.011) ***
Quarter 3 * stricter exam. 0.031 (0.007) *** 0.018 (0.005) ** 0.087 (0.018) *** 0.072 (0.019) ***
Quarter 4 * stricter exam. 0.019 (0.008) * 0.009 (0.005) 0.069 (0.019) ** 0.052 (0.018) **
decrease to -0.013, -0.007, -0.015 and -0.039 respectively. The WIA coefficients increase to -0.098, -0.050, -0.137 and
-0.149 respectively. All other coefficients remain unchanged.
126 chapter 3
(continuation) Table 3.3 GLS estimates for sector inflow models

Construction Financial / Health care Public sector /
business education
N 39 39 39 39
R 2
0.997 0.993 0.992 0.984
Autocorrelation order na (1) na (1)
Dw (P-value) 2.27 (0.43) 1.82 (0.12) 1.63 (0.02) 1.91 (0.22)
# employees in sector 389,000 1,232,000 1,155,000 908,000
The results include some interesting findings. For the construction sector, the effect of experi-
ence rating is comparatively large and the effect of the gatekeeper is comparatively small.
Possibly, the employers in this sector are very sensitive to financial incentives but lack the
size or the capability to implement a bureaucratic protocol like the gatekeeper successfully.
A similar result is found for the hotels and restaurants sector (see appendix 3.F). On the other
hand, in the financial and business sector, which has the lowest rate of inflow of disability
recipients of all sectors, the effect of experience rating is non-existent, but the effect of the
gatekeeper protocol is substantial. Possibly, the comparatively large employers (banks, insur-
ers etc.) in this sector were not sensitive to financial incentives initially given their low dis-
ability costs, but were able to reduce long-term sickness and disability through the gatekeeper
protocol after all. In the health care sector, which is mainly populated by female workers, the
effects of experience rating and gatekeeper are both large, but the WIA effects are compara-
tively small, which corresponds to the finding in the general model that the WIA affected men
more than women. Another interesting result is the lack of effectiveness of experience rating
in the public sector, which may be explained by the low financial accountability of public
administration and educational organizations. However, the employers in the public sector,
which tend to be large and used to bureaucracy, have been able to implement the gatekeeper
protocol very successfully. The gatekeeper effect in the public sector is the highest out of all
sectors.
3.3.4 Estimation results outflow model
Again, for the general model regression analyses are performed for both the number of per-
sons and the fde-equivalents. As no autocorrelation is present OLS estimates are used. Table
3.4 gives the model estimates for the general model of the outflow rate (persons and fde).
Table 3.4 OLS estimates outflow model

Persons Fde
Intercept 1.131 (0.172) *** 0.797 (0.097) ***
Experience rating 0.391 (0.226) 0.029 (0.128)
Premium rate 0.169 (0.206) 0.083 (0.116)
Re-exam. DI stock 1.366 (0.157) *** 0.466 (0.089) ***
Re-exam. DI stock * Time -0.063 (0.030) * -0.029 (0.017)
Re-examination system -0.377 (0.151) * -0.219 (0.085) *
Business cycle indicator -0.050 (0.036) -0.045 (0.020) *
Mean inflow rate (t-1;t-8) 1.604 (0.735) * 0.523 (0.414)
Quarter 2 -0.195 (0.064) ** -0.103 (0.036) **
Quarter 3 -0.092 (0.065) -0.059 (0.037)
Quarter 4 -0.165 (0.066) * -0.075 (0.037)
N 43 43
R 2
0.858 0.851
Autocorrelation order na na
Dw (P-value) 1.93 (0.12) 1.78 (0.05)
The regression on outflow in persons shows, as expected, a strong upwards influence of the
re-examinations of the DI stock on the outflow rate. The effect of the re-examinations of the
DI stock decreases over time, although the coefficients of the cross-term are not significant in
the fde model. The change of the re-examination system (abolition of the periodic re-exami-
nations) has a significant negative impact on the outflow rate. It cannot yet be deduced from
the data whether this is caused by the change of the re-examination system itself (e.g. the
benefit administration office not being able adequately to select the cases to be re-examined)
128 chapter 3
or by side effects of the re-examination of the DI stock (e.g. lack of capacity to perform regu-
lar re-examinations or recovered beneficiaries waiting to report to the benefit administration
office until their re-examination). However, this result confirms earlier findings (UWV, 2006)
that the number of “spontaneous” recoveries decreased remarkably after the re-examinations
had started. If the result is due to the change in the system, so that when the re-examinations
of the DI stock are finished the significant negative influence of the abolishment of the pe-
riodic re-examinations persists, it might be that the temporary gains of the re-examinations
of the DI stock will be overcompensated by the losses from the abolition of the periodic re-
examinations. Decreasing inflow is another significant reason for the decreasing outflow from
the disability schemes. Finally, the seasonal dummies for the second and fourth quarter both
have significant negative values whereas the business cycle and the experience rating law
have no significant influence on the recovery rates.
Again, the former analysis is repeated for age- and gender-specific subgroups. For the older
men and women the variable on re-examinations of the DI stock is included as a small part of
this group (the 45–50 year olds) was re-examined. However, the cross-term of re-examina-
tions and time period is excluded as the exact numbers of re-examinations of the elderly over
time are not known so time effects cannot be attributed to effectiveness. For convenience, the
results are limited here to the inflow in persons. The results are shown in Table 3.5.
Table 3.5 OLS estimates outflow models for men and women aged >= 45 and < 45
respectively
Men >= 45 Men < 45 Women >= 45 Women < 45
Intercept 0.586 (0.099) *** 2.243 (0.408) *** 1.001 (0.152) *** 2.652 (0.654) ***
Experience rating 0.177 (0.133) 0.890 (0.526) -0.042 (0.209) 0.639 (0.754)
Premium rate -0.009 (0.123) 0.394 (0.491) 0.021 (0.188) 0.568 (0.675)
Re-exam. DI stock 0.052 (0.078) 4.184 (0.382) *** 0.010 (0.118) 6.215 (0.523) ***
Re-exam. DI stock * Time -0.219 (0.078) ** -0.439 (0.098) ***
Re-examination system -0.210 (0.086) * -0.651 (0.376) -0.348 (0.133) * -1.177 (0.490) *
Business cycle indicator -0.006 (0.022) 0.002 (0.085) -0.056 (0.034) -0.045 (0.119)
Mean inflow rate (t-1;t-8) 0.236 (0.280) 8.455 (3.822) * -0.158 (0.243) 2.012 (2.036)
Quarter 2 -0.192 (0.039) *** -0.423 (0.154) ** -0.162 (0.059) ** -0.175 (0.211)
Quarter 3 -0.115 (0.039) *** -0.123 (0.157) -0.111 (0.059) -0.052 (0.213)
Quarter 4 -0.182 (0.039) *** -0.389 (0.155) * -0.110 (0.060) -0.054 (0.216)
N 43 43 43 43
R2 0.743 0.862 0.668 0.856
Autocorrelation order na na na na
Dw (P-value) 1.60 (0.03) 2.09 (0.26) 1.18 (0.00) 1.99 (0.17)
When we consider the younger men and women, we see the same conclusions from the over-
all analyses coming back more pronouncedly. The re-examinations predictably boost outflow
in these groups, but their effect significantly decreases over time. The change in the re-exam-
ination system decreases outflow because of recovery, although this effect is significant for
younger women only. Among older men and women, there is no effect of the re-examinations
as these were mostly exempt from it. The change in the re-examination system significantly
decreases outflow because of recovery for both older men and women. Again, it should be
noted that it is not clear whether this effect is caused by the change itself or by side effects of
the re-examinations of the DI stock.
The effect of experience rating and the mean premium rate is not significant, although the
signs are positive except for older women. There is no influence of the business cycle on DI
130 chapter 3
outflow. The inflow variable is significant for younger men only, which is puzzling as in the
general model this variable is significant.
3.4 The microsimulation model
In this section the microsimulation model used for making long term forecasts of the disabil-
ity schemes is presented. It starts with a short introduction in section 3.4.1 and a theoretical
framework in section 3.4.2. In section 3.4.3 the modelling of the inflow into, outflow from
and transitions within the disability schemes is discussed. In sections 3.4.4 the estimation re-
sults of the multinomial logistic regression models used for the transitions within the schemes
are presented. They are input for the microsimulation model. The results from the microsimu-
lation model itself are presented in section 3.5.
3.4.1 Introduction
The use of microsimulation models in policy assessment and evaluation is becoming more
widespread (see e.g. Buddelmeyer, Freebairn and Kalb, 2006, Van Sonsbeek and Gradus,
2006, and Van Sonsbeek, 2010). This chapter uses the dynamic population microsimula-
tion model that is currently in use for making long-term budgetary forecasts of the disability
schemes at the Dutch Ministry of Social Affairs and Employment. The model was originally
described in Van Sonsbeek and Gradus (2006) and was developed for forecasting the effects
of the introduction of the WIA. In this model, a sample of the disability stock is taken and
samples of the most recent inflow cohort are added for each forecasted year. The size of the
inflow cohorts is corrected for demographic change. Outflow probabilities are cell-based rates
based on the most recent outflow cohort. The grade of disability and the working status do
not vary over time.
The original model is adapted in three ways in order to improve the forecasts. First, for the
inflow forecasts, not only the size but also the composition of the inflow cohorts is corrected
for demographic change. Second, the outflow rates are now estimated with a multinomial lo-
gistic regression model. Third, the grade of disability and the working status are now variable
over time. Together with the major outflow routes (recovery and mortality) these transitions
(from fully disabled to partially disabled and vice versa and from working to unemployed and
vice versa) are now also estimated with a multinomial logistic regression model.
3.4.2 Theoretical framework
A calculation model for simultaneously generating forecasts of the labour force and the dis-
ability benefit population is given in Aarts and de Jong (1992). In their model the flow models
of the labour force and the DI population are linked through the inflow into and the recover-
ies from the DI population. In our microsimulation model, the labour force is exogenous.
However, the transitions within the DI population are modelled in a very detailed way. We
start from a simple description of stock and flows of the DI population of a certain scheme.
(3.3) S_(t+1)=S_t+I_t-O_t
where St represents the stock of DI beneficiaries in year t, It the inflow and Ot the outflow.
We add to this the notion that the stock of disability benefit recipients of a scheme consists
of several relevant subgroups. A first important subdivision within the disability schemes is
between the fully and partially disabled. The partially disabled tend to have lower benefits,
higher recovery rates and lower mortality rates. As the grade of disability can change over
time, the transitions between fully and partially disabled and vice versa need to be modelled.
However, the number of cases in each individual class of partial disability is rather small.
Therefore, all classes of partially disability are aggregated. This is only a minor limitation as
132 chapter 3
a higher class of partial disability does not necessarily correspond with a more severe limita-
tion. The class of disability is determined by the loss of income as compared with the wage
before disability. This implies that higher incomes tend to populate the higher classes of dis-
ability regardless of the severity of the disability36. Indeed, micro data show that the share of
partially disabled that are working is about the same throughout all classes of partial disabili-
ty. Other important subdivisions follow from the WIA, like those between the IVA scheme for
the permanently fully disabled and the WGA scheme for the temporarily fully disabled and
the partially disabled, and within the WGA scheme between the partially disabled who are
working and those who are unemployed37. Therefore the stock in year t+1 from equation (3.3)
can be rewritten for each subgroup as the sum of the subgroup stocks, inflows and outflows
and the transitions between the subgroups in year t.
(3.4) S_(i,t+1)=S_(i,t)+ ∑_(i≠j)▒T_(j,i,t) -∑_(i≠j)▒T_(i,j,t) +I_(i,t)-O_(i,t)
where Si,t, Ii,t and Oi,t represent the stock, inflow and outflow of subgroup i in year t and
Ti,j,t represents the transition from subgroup i to subgroup j in year t.
Well-known determinants for disability inflow are age, gender, ethnicity, cohabitation status
and wage (Selten and Copinga, 2003). Employer-specific factors like sector and company
size are also important. However, the inflow data collected by the benefit administration
office include a limited set of personal characteristics that are relevant for determining and
paying the benefit; of the inflow determinants only age, gender, wage and sector are known.
Moreover, no reliable estimates on future development of wage and sector distribution are
known, so future inflow trends can be controlled for by age and gender only. This is shown in
equation (3.5).
(3.5) I_(i,j,t)=i_(i,j,0) L_(i,j,t-2)+∑_k▒〖r_(i,j,t-k) I_(i,j,t-k) 〗
36 E.g. when someone who earned € 80,000 because of a disability can only earn € 20,000 in another job (and/or for less
hours), his grade of disability will be 75% (class 65–80%), whereas someone who earned € 40,000 with the same dis-
ability can also earn € 20,000 in another job will have a grade of disability of 50% (class 45–55%).
37 When a WIA beneficiary is partially disabled, the benefit level is determined based on the working status. If that person
works, a benefit based on the former wage is supplied on top of the current wage, but if they do not work, a benefit based
on the legal minimum wage is supplied.
where for gender i and age j, Ii,j,t represents the inflow in year t, Li,j,t-2 represents the labour
force from two years earlier and ri,j,t-k represents the fraction of the inflow of k years earlier
that re-enters the benefit scheme in year t after being out for a while.
The main reasons for termination of a benefit are recovery, mortality and pension (when
someone turns 65, the legal retirement age in the Netherlands, the disability benefit is auto-
matically terminated). Other, less common reasons include detention and emigration. As their
numbers are too small to model them separately and their rates like recovery rates decrease
by age, they are included in the recovery rates. Therefore, outflow for each subgroup can be
defined by:
(3.6) O_(i,t)=R_(i,t)+M_(i,t)+P_(i,t)
where Ri,t, Mi,t and Pi,t represent the number of recoveries, mortalities and pensioners for
subgroup i in year t respectively. The number of pensioners follows from the stock of 64 year
olds and their rates of recovery and mortality.
(3.7) P_(i,64,t)=S_(i,64,t)-R_(i,64,t)-M_(i,64,t)
In sum, the above system is fully determined by the inflow I, the outflow components R and
M and the transitions Ti,j. In the next section, estimation of these functions will be described.
3.4.3 Modelling inflow, outflow and transitions in grade of disability and

working status
The age- and gender-specific development of the working population is taken from the most
recent forecast by CPB (Euwals, de Mooij and van Vuuren, 2009). This forecast projects
the development of the working population by five-year age cohorts, class and gender. The
134 chapter 3
changes in the composition are significant. The share of women in the working population
and the average age of working men and women all show an upward trend for the forthcom-
ing decades, suggesting upward pressure on future inflow as both women and the elderly are
known to have high inflow rates into disability benefit schemes. For the Wajong scheme, the
inflow comes from the general population, mostly from the population of 18 year olds. The
inflow equation is similar, but instead of the labour force from two years earlier the general
population from one year earlier is taken, based on the most recent population forecast by
Statistics Netherlands (CBS, 2009). The re-entry variables are important because re-entry
cases are known to make up over 10% of yearly inflow cohorts.
In the microsimulation model, new inflow cohorts are taken from a sample of the most recent
inflow cohort, with a correction for the changing weight of the corresponding age/gender
class in the working population or the general population. Re-entry cases are already part of
the sample. For the WIA scheme, the number of re-entry cases is still low because the WIA
has been in existence for just four years. Additional re-entry cases for future inflow cohorts
are added based on the re-entry pattern of the last full WAO inflow cohort from 2004.
For recovery, mortality and transitions between subgroups, it holds that if a scheme con-
sists of n subgroups, there are (n2+n) functions to be estimated: (n2-n) transitions from one
subgroup to another, n recovery functions and n mortality functions. At the individual level,
all transitions are binary choices, so a logistic regression suits this purpose. However, because
all transitions, including recovery and mortality, from a certain subgroup (the reference state)
can be considered competing risks, a multinomial logistic regression is chosen.
It is known from previous research that age, gender, duration of the benefit and disability
grade are major outflow determinants (Linder, 2005, Jehoel Gijsbers, 2007). Therefore, in all
models, age, gender and duration are used as dependent variables38 (disability grade needs
not be modelled because separate models are estimated for the fully and partially disabled).
The relevance of age, gender and duration of the benefit becomes clear from Figures 3.4 and
38 With the exception of the IVA scheme. As outflow from the IVA is almost always caused by mortality, which is mainly
age-dependent, duration is not used as a dependent variable. However, duration dummies for the first and second year in
the IVA scheme are used because part of the IVA inflow consists of people who are terminally ill.
3.5. Figure 3.4 gives recovery rates for men and women by duration of the disability benefit
in years for the WAO scheme in 200439. Benefit duration is indeed an important predictor of
recovery as recovery is concentrated in the early years of disability. Recovery rates are also
found to be consistently higher for women. For both men and women recovery rates peak in
the second year after inflow and decrease from then on, with small peaks after 5 and 10 years
because of the periodic re-examinations. Figure 3.5 gives mortality rates for men and women
by age for the WAO scheme in 2004. As expected, mortality rates are found to be consistently
higher for men and for both men and women to rise by age.
Figure 3.4 Recovery rate by duration of the benefit, WAO 2004
39 In this example 2004 data are used because 2004 was the last year that a new cohort (with benefit duration of 0) was
added to the WAO stock.
136 chapter 3
Figure 3.5 Mortality rate by age, WAO 2004
On top of this basic model, additional variables are added using a forward selection proce-
dure. The model with the lowest value of the Akaike information criterion (AIC)40 is chosen.
As additional variables that may enter the models, in the first place dummies for durations
of 0, 1, 5 and 10 years are used. Until 2004 all disability benefit recipients were periodically
re-examined. These re-examinations took place after the first and fifth year of disability and
consequently every 5 years. Figure 3.4 reveals that indeed recovery rates peaked in the years
the re-examinations took place. In particular for the Wajong scheme a dummy for the age of
18 years is also allowed to enter the model. This age dummy is not the same as the dummy
for duration of 0. The majority of beneficiaries in the Wajong enter the scheme at 18 years
of age but a sizeable minority enters the scheme at a later age. It is to be expected that the
kind of disabilities of the later entrants (and therefore their transition rates) differ from the
kind of disabilities of the early entrants (mainly congenital diseases). Secondly, duration2 and
age2 are used as quadratic relationships of transitions with duration or age may exist. Higher
grade polynomials are not used as there is not really a logical explanation for those structures
40 Alternatively, the Schwarz information criterion (SIC) could have been used. The AIC tends to select more complicated
models as the AIC penalizes additional variables less than the SIC. As predictive value is important for a forecasting
model and higher degree polynomials for age and duration are excluded anyway, we prefer to base the model selection
on the AIC.
and the interpretation of the coefficients becomes very difficult. Finally, also the cross-term
age*duration is used. This reflects the idea that benefit duration has a different impact at dif-
ferent ages, for example, because the distance to the labour market and the human skills of
the beneficiaries deteriorate more rapidly when they enter the disability scheme at a higher
age.
One transition in the WIA that cannot be modelled in the way described above is the transi-
tion from the WGA scheme for partially disabled and temporarily fully disabled to the IVA
scheme for permanently fully disabled. The fully disabled in the WGA scheme are regularly
re-examined. As soon as their disability is considered to be permanent, they qualify for the
IVA scheme. However, how many of the fully disabled in the WGA will eventually qualify
for the IVA and when they will do so is not yet clear, as the WIA has existed for just four
years to date. In earlier calculations of the effects of the WIA (SER, 2004) a duration of
five years is chosen as a cut-off point for permanent disability. However, there is no formal
rule allowing the fully disabled to enter the IVA scheme after five years of disability. In the
simulation it is assumed that after two, three or four years of being fully disabled, WGA
beneficiaries have a 10% chance of qualifying for the IVA. After five or more years of being
fully disabled they have a 50% chance of qualifying for the IVA. On average, in this scenario
WGA beneficiaries qualify for the IVA after being fully disabled for 5.5 years.
The above transitions can be represented in a transition matrix. The reference states are
fully disabled and partially disabled (whenever relevant further subdivided into working and
unemployed). The transition states equal the reference states plus two additional states for
outflow through recovery and outflow through mortality. Not all transitions have to be mod-
elled. In the WIA scheme, so far, almost all the outflow from the IVA is because of mortality.
Outflow through recovery is rare and transitions to the WGA scheme are almost non-existent,
just as was intended, as the IVA covers the permanently, fully disabled. The transition matri-
ces are summarized in Tables 3.6–3.8 with mean transition rates based on the 2009 data:
138 chapter 3
Table 3.6 Transition matrix, WAO (means for 2009)

State(t) Fully disabled Partially disabled Recovery Mortality
State(t-1)
Fully disabled 0.974 0.008 0.008 0.010
(N=359,461)
Partially disabled 0.034 0.948 0.014 0.004
(N=170,537)
Table 3.7 Transition matrix, WIA (means for 2009)

State(t) Fully disabled Partially Partially Recovery Mortality
disabled WGA disabled WGA
State(t-1) working unemployed
Fully disabled 0.934 0 0 0.003 0.057
IVA
(N=18,952)
Fully disabled 0.915 0.007 0.012 0.058 0.008
WGA
(N=52,346)
Partially 0.053 0.866 0.043 0.032 0.006
disabled WGA
working
(N=7,964)
Partially 0.080 0.098 0.796 0.021 0.005
disabled WGA
unemployed
(N=8,007)
Table 3.8 Transition matrix, Wajong (means for 2009)

State(t) Fully disabled Partially disabled Recovery Mortality
State(t-1)
Fully disabled 0.982 0.002 0.009 0.007
(N=189,532)
Partially disabled 0.056 0.921 0.022 0.001
(N=3,963)
Group sizes vary from very large (WAO scheme and fully disabled in the Wajong scheme)
to small (partially disabled in the WIA scheme and partially disabled in the Wajong scheme).
The small number of partially disabled in the WIA scheme is due to the WIA being in force
for just four years. In the Wajong scheme, the grade of disability is determined in the same
way as in the WIA scheme. However, the result of this process almost always amounts to full
disability for two reasons. First, as Wajong applicants are usually 18 years of age when they
apply for the benefit, they have no working history, and thus no wage to compare with the
loss of income. Therefore, the legal minimum wage is taken as a benchmark. Second, only
jobs on the regular labour market, that earn at least the legal minimum wage, are taken into
account in the process. This implies that either, the applicant is able to earn the regular mini-
mum wage, so his or her grade of disability will be 0%, or the applicant is not able to earn the
regular minimum wage, so his or her grade of disability will be 100%.
3.4.4 Estimation results
The models chosen are presented in Table 3.9. For the WAO scheme the duration dummies
for 0, 1 and 5 years enter the model in all years. In 2004, the dummy for 10 years is also
included. In the Wajong scheme, the dummy for 0 years and the dummy for age 18 enter the
model in five out of six cases (except for the partially disabled in 2009). Duration2 and the
cross-term of age and duration are generally included in the WAO and WIA models. Age2 is
not included in most models.
140 chapter 3
Table 3.9 Model selection for WAO, Wajong and WIA schemes, 2000, 2004 and 2009
Model Int. Gen- Dur. Age Dur 0 Dur 1 Dur 5 Dur Age Dur2 Age2 Age*
der 10 18 Dur
WAO 2000
Fully Disabled x x x x x x x x x x
Partially disabled x x x x x x x x x
WAO 2004
Partially disabled x x x x x x x x x x
WAO 2009
Fully Disabled x x x x x x x x x
Partially disabled x x x x x x x x x
WIA 2009
Fully Disabled IVA x x x x x x
Partially disabled WGA x x x x x x x x x
Part. work x x x x x x x
Part. Unempl. x x x x x x x
Wajong 2004
Fully Disabled x x x x x x x x x
Partially disabled x x x x x x x
Wajong 2009
Partially disabled x x x x
The full sets of parameter estimates are included in appendix 3.B (WAO scheme), 3.C (WIA
scheme) and 3.D (Wajong scheme). A closer look at the parameter estimates reveals the fol-
lowing:
• WAO: The differences between the 2000 and 2004 estimates are comparatively small.
The 2009 estimates differ slightly more from the 2000 and 2004 estimates. In most cases
the signs are consistent. In all years men have significantly higher mortality rates than
women. Also in all years, fully disabled men have a higher probability of becoming par-
tially disabled and partially disabled women have a higher probability of becoming fully
disabled. As expected, mortality rates rise by age and recovery rates decrease by age,
although this cannot be deducted straightforwardly from the coefficients as both age and
age2 are used in the model. Interesting are the differences in recovery rates between men
and women. In 2000 fully disabled men and partially disabled women have significantly
higher recovery rates. In 2004, women have higher recovery rates among both the fully
and partially disabled. In 2009, men again have higher recovery rates among both the
fully and partially disabled, the latter albeit not significant.
• WIA: Of course, estimation possibilities of the WIA model are still limited as the WIA
has only existed since 2006, so benefit durations of more than four years are not yet pos-
sible. However, the first indications are that estimates differ considerably from the WAO
estimates. For example, except for the fully disabled in the WGA scheme, there are no
longer significant gender differences in mortality rates. This may be attributed to the
entrance criteria for the WIA being much tighter than for the WAO. In the WAO, inflow
rates of women were much higher than for men, apparently implying that women in the
WAO were less severely disabled than men. Transitions from fully disabled to partially
disabled are still more common among men. However, there is no gender difference in
transitions from partially disabled working to unemployed and vice versa. In general,
recovery rates are lower in the WIA than in the WAO when controlled for age and dura-
tion. This may reflect the fact that inflow rates into the WIA are so low that the WIA
beneficiaries are more severely disabled than the WAO beneficiaries, resulting in lower
recovery rates and higher mortality rates overall.
• Wajong: There are substantial differences between the coefficients of the Wajong and
WAO/WIA schemes. As the Wajong scheme covers mainly congenital diseases, the
majority of beneficiaries enter the scheme at the age of 18. For all transitions the dummy
for age 18 is significant: those that enter the Wajong at 18 have significantly higher
recovery rates but no difference in mortality rates. There is a strong correlation between
recovery rate and age and/or duration of the benefit. After a while in the benefit, recov-
ery rates diminish, which may also explain why beneficiaries who enter the scheme at
age 18 still have higher recovery rates as they are still developing (e.g. in case of lighter
forms of mental disability). In the Wajong scheme the share of partially disabled is very
142 chapter 3
low, typically around 2% of the stock of beneficiaries. This complicates model estima-
tion with typically a very small number of significant explanatory variables remaining in
the models for the partially disabled.
3.5 The combined effect of the policy reforms in perspective
In this section, the microsimulation model described in section 3.4 is used for making long-
term forecasts of the DI-schemes. In section 3.5.1 the combined long-term effects on the
disability stock of all policy measures are derived by comparing microsimulations from 2000
(before the measures became effective) and 2009 (when all measures are effective). In sec-
tion 3.5.2 the combined long-term effects of the outflow-related policy measures are derived
by comparing microsimulations from 2004 (before the measures became effective) and
2009 (when all measures are effective). In section 3.5.3 an integral forecast of all disability
schemes is presented and effects are placed in perspective by including the Wajong scheme in
the analysis.
3.5.1 Combined effect of all policy measures
The combined effect of the policy reforms can be deducted by comparing the microsimula-
tion forecasts from subsequent years. This section specifically focuses on the forecasts from
2000, 2004 and 2009. The first year for which micro data are available is 1999 and the last
year is 2009. Because the data of year t-1 are needed for estimating the transition functions
for year t, 2000 is the first year that can be used for a simulation. This year can be considered
the baseline before the recent policy changes took effect. The last year in which new WAO
cases were allowed is 2004. Also, by the end of 2004, the re-examinations of the disability
stock had begun. So, by comparing the 2004 and 2000 simulations, the combined effect of
three inflow-related policy reforms is captured: experience rating, gatekeeper protocol and
stricter entrance criteria. According to the model results from section 3.3.3, these reduced in-
flow to the DI scheme by 42-48%. By comparing the 2009 and 2004 simulations, the effects
of the WIA and the re-examinations of the disability stock are captured. The latter of course
only has a temporary effect as the 2004 stock will eventually die out anyway.
Figure 3.6 gives the long-term development of the number of beneficiaries in the WAO and
WIA schemes for the base years 2000, 2004 and 2009. From the 2000 simulation, it is clear
why reform was inevitable. The number of disability cases would have reached 1.2 million in
the long run, which, as a percentage of the working population, would be about 17% in 2040.
This even excludes the young disabled and the disabled self-employed in the Wajong and
WAZ schemes.
Figure 3.6 Long-term forecasts, WAO/WIA scheme for 2000, 2004 and 2009 to 2040
144 chapter 3
The combined effect of the reforms between 2000 and 2004 is huge. The number of WAO
cases in the long run stabilizes at 600,000, 50% lower than the 2000 simulation. This is
similar to the inflow reduction estimated in section 3.3.3. This similarity is not trivial. When
the inflow reduction is translated into a similar stock reduction, the outflow patterns of the
smaller inflow cohorts are not significantly different from the larger inflow cohorts from the
past. From comparison of the 2009 and 2004 simulations it follows that the WIA scheme
caused another large decrease of the number of benefits. In the long run, the number of WIA
cases stabilizes around 370,000, which is 830,000, or almost 70%, lower than the 2000
simulation and 230,000, or almost 40%, lower than the 2004 simulation. One can wonder
whether the WIA inflow will remain at the current level in future years. Currently, WIA in-
flow is rising, which raises doubts about how stable the inflow levels of recent years will be.
However, recent research (Van Deursen and Van Loo, 2010) indicates that the rising inflow is
mainly due to structural factors like size and composition of the working population, which
are accounted for in the microsimulation model. Also, in the inflow model from section 3.3.3,
decreasing effectiveness over time of the WIA scheme could not yet be proven.
When the WIA inflow stabilizes at current levels, however, there are two limitations to the
WIA effects. First, it should be noted that part of the WIA effect is artificial because costs of
the second year of sickness are transferred to the payroll of individual employers. Second,
the WIA effect, when measured in fde, is smaller: the 2000 simulation stabilizes just below 1
million fde in 2040, the 2004 simulation at just above 450,000 fde and the 2009 simulation at
just above 330,000 fde.
3.5.2 Combined effect of outflow-related measures
On the outflow-related policy measures, in section 3.3.4 an ambiguous effect of the re-
examinations of the disability stock was found. The re-examinations of the disability stock
appear to have boosted outflow, but the change of the re-examination system (abolition of
the periodic re-examinations) seems to have had a negative influence on outflow rates. A
quantification of the underlying processes can be done by running two microsimulations; one
originating in 2004, using the transition functions from 2004 (Table 3.B.3/4) combined with
the known inflow realizations from 2005–2009 and one originating in 2009, using the transi-
tion functions from 2009 (Table 3.B.5/6). As both simulations follow exactly the same inflow
pattern, the difference between the two can only be explained by the direct effects of the
re-examinations (extra outflow during 2005–2009 that is not captured in the 2004 simulation)
and differences in the transition function estimates of 2004 and 2009. Of course, conclusions
only hold true under the assumption that the estimates of 2004 would have been (and the
2009 estimates will be) representative for future years and that the 2004 and 2009 data are
not biased upwards or downwards by specific circumstances. Figure 3.7 gives the results for
yearly outflow of such an analysis.
Figure 3.7 Long-term forecasts WAO outflow, 2004 and 2009 to 2040
146 chapter 3
The overall picture that results from comparing the 2004 and 2009 microsimulations is
surprising. As expected, the early years of the re-examination operation cause a significant
increase of the outflow. However, from 2007 onwards, outflow falls far below the level
predicted in the 2004 simulation. Apparently the effect of decreasing overall recovery rates
overcompensates for the temporary increase in recovery rates of the younger part of the stock
during the re-examination operation.
A possible explanation is that the abolition of the periodic re-examinations has been harmful
because a substantial number of recoveries are no longer detected. This may be caused by a
lack of re-examination capacity (because the benefit administration office was so occupied by
the re-examinations operation that the re-examinations of the rest of the population are not
performed with the intended strictness) or the benefit administration office not being able to
select the re-examination cases adequately. Also, the effect of the re-examinations themselves
may be more modest in net terms than it seemed because some of the beneficiaries who lost
the benefit after being re-examined would have recovered anyway. It can be argued that,
now the re-examinations have ended, recovery rates might rise again because additional re-
examination capacity becomes available. In that case, because the re-examinations operation
finished almost a year ago, such an effect should have shown in the outflow numbers since.
However, during 2009 it has not and in the first half of 2010 outflow rates indeed have been
rising slightly, but not enough to make a substantial difference.
Although much care has to be taken when interpreting these results, it can be concluded that,
especially when compared with the very effective inflow-related policy reforms, the net ef-
fects of the outflow related policy measures are minor at most.
3.5.3 An integral long-term forecast of all disability schemes
In order to complete the picture, the Wajong scheme needs to be included in the analysis.
Figure 3.8 gives the results of the three simulations of the Wajong scheme, originating in
2000, 2004 and 2009 respectively. The results are measured in persons only, as in the Wajong
scheme the difference between persons and fde is very small because of the overwhelming
majority of Wajong beneficiaries being fully disabled.
Figure 3.8 Long-term forecasts, Wajong scheme 2000, 2004 and 2009 to 2040
There are only minor differences between the 2000 and 2004 simulations. However, the
Wajong forecast has increased dramatically since 2004, topping 400,000 around 2040 by
current expectations. Garcia Gomez, von Gaudecker and Lindeboom (2009) conclude that
the Wajong programme is now becoming more a substitute for general social assistance than
a substitute for disability insurance. Following De Jong (2008), who argued that the WAO
scheme was misused in the past in order to accommodate social change, the Wajong scheme
148 chapter 3
can be argued to accommodate the transition to an information and technologically oriented

society, leaving large groups of poorly educated young people who are not productive enough
for the regular labour market on benefit. Therefore, the government has initiated a new
systemic reform for the Wajong scheme. Whether this reform is as successful as previous
reforms remains to be seen, especially as it lacks financial incentives for employers and the
entrance criteria of the scheme remain unchanged.
When the growth of the Wajong stock is taken into account, the total number of disability
benefit recipients now stabilizes at almost 800,000 (11% of the working population), still
substantially lower than the 2000 simulation (almost 1,400,000 or 20% of the working
population) but only marginally lower than the 2004 simulation. Apparently the gains from
the WIA are almost compensated by the losses from the Wajong. In terms of fde, correcting
for the grade of disability of the partially disabled, the 2009 forecast ends up substantially
higher than the 2004 forecast because the gains of the WIA for a substantial part come from
the lower classes of disability and the new Wajong cases are almost all fully disabled. An
actual causal relationship between the growth of the Wajong scheme and the decreasing use
of the WAO/WIA schemes might be studied in the future, but does not follow from logic as
the WAO/WIA beneficiaries are employees with a work history whereas the Wajong scheme
is intended for the young disabled who have no work history.
Table 3.10 Long-term (2040) effects of clusters of policy measures

(x 1,000) 2000 2004/2000 2004 2009/2004 2009 2009/2000
(exp. rating, (WIA) (all re-
gatekeeper, forms)
examin.)
WAO/WIA-benefits 1201 -603 598 -230 368 832
Wajong benefits 188 19 207 205 413 -225
Total benefits 1388 -583 805 -24 781 608
WAO/WIA-fde’s 994 -537 457 -124 333 661
Wajong fde’s 186 19 204 202 406 -221
Total fde’s 1180 -518 661 78 740 440
The conclusion of Table 3.10 is twofold. The combined long-term effect of the policy meas-
ures on the number of WAO/WIA benefits is huge: 832,000 (69%) less benefits and 661,000
(66%) less fdes. On the other hand, most of the “gains” due to decreasing WAO/WIA stock
are lost because of the rapidly increasing Wajong stock. The long-term projection of WAO,
WIA and Wajong together has decreased between 2004 and 2009 by only 24,000 benefit
recipients and has increased between 2004 and 2009 by 78,000 fde.
The long-term forecasts based on 2009 estimates give a higher number of benefits and there-
fore a slightly more pessimistic view of the future than an earlier projection based on 2008
estimates (Van Sonsbeek and Alblas, 2010). The long-term forecast of the Wajong scheme
is slightly increased because outflow rates in 2009 are lower than in 2008. The long-term
forecast of the WIA scheme is increased because inflow rates are slightly increased and out-
flow rates, especially from the IVA scheme, decreased in 2009 as compared with 2008. This
highlights one limitation of the model results. Long-term forecasts are based on continuation
of current inflow and outflow patterns with corrections for known trends like changes in the
working population and new policy reforms. When random shocks influence the inflow or
outflow patterns in a certain year, the impact on the long-term forecast will be large. None-
theless, the forecasts provide the best possible estimate and are therefore used as the base for
policy discussions.
3.6 Conclusions and topics for future research
Until 2001 disability rates in the Netherlands were among the highest in the world. Since
then, some very effective policy measures have been introduced. The effects of those policy
measures are estimated in this chapter by using administrative datasets of all disability benefit
recipients from 1999 to 2009. In the first part of this chapter, with a regression on aggregated
data, the effects on inflow and outflow rates of the individual policy measures are estimated.
In the second part of this chapter, with a microsimulation model, long-term forecasts of the
150
disability schemes are made, making it possible to deduce the combined long-term effect of
the reforms.
If the 2000 inflow rates had continued into the future, the number of disability benefits would
have risen to 1.2 million or 17% of the working population by 2040 in the WAO scheme or
1.4 million or 20% of the working population in the WAO and Wajong scheme together.
The introduction of experience rating has reduced inflow into the WAO scheme by 13 per-
centage points, the introduction of the gatekeeper protocol has reduced inflow by 22 percent-
age points and the tightening of the eligibility criteria has further reduced inflow by 7-13
percentage points. The combined effect of experience rating, gatekeeper protocol and tight-
ened eligibility criteria is 42-48 percentage points. The additional effect of the WIA is large
as well, resulting in a decrease of inflow by 23-29 percentage points. There is no conclusive
evidence for increasing or decreasing effectiveness of the policy measures over time.
As a consequence of experience rating, gatekeeper protocol and tightened eligibility criteria,

inflow into the WAO scheme had already halved before the WIA came into force, resulting
in the number of disability benefit recipients stabilizing at 600,000 or 8.5% of the working
population in 2040. The WIA causes the number of benefits to further decrease to 370,000
but the effect of the WIA is smaller in cost terms.
All those effective policy measures have one thing in common: they focus on preventing
inflow. Indeed, prevention is the best way in the long run of keeping claimant numbers low.
Only the re-examinations of the disability stock from 2004 to 2009 caused a significant
increase in disability outflow. However, this increase came at a cost. At the same time as the
re-examinations boosted outflow, the recovery rates of the population not affected by the
re-examinations decreased sharply, possibly because of the change in the re-examination
periodicity. In the long run, this may cause a higher disability stock in the WAO, causing net
losses instead of net gains.
151
Whereas the key to success appears to have been found for the disability schemes for em-
ployees, new problems have risen in the scheme for the young disabled whose numbers have
been rising rapidly during the past few years. When the current inflow numbers are contin-
ued into the future, the number of Wajong benefit recipients would grow to almost 450,000,
which is even more than the number of WIA benefits. Indeed, it seems that the “gains” in
terms of reduction in benefits from the introduction of the WIA are compensated by the rising
number of Wajong benefits. Again a disability scheme seems to be misused to accommodate
social change, in this case the transformation to an information oriented society requiring
higher education levels of its people and leaving behind large groups of people at the bottom
of the labour market.
Important topics for future research include the effects of the recent Wajong reform which
has been in force since 1 January 2010. Also, the possible relation between the decreasing
number of WIA benefits and the rising number of Wajong benefits needs to be investigated
further. Finally, it remains to be seen whether all policy reforms remain effective in the long
run, especially when pressure on the disability schemes might increase in the future when the
retirement age is raised and early retirement pathways are blocked.
152 chapter 3
Appendix 3.A Data
In section 3.3.1 the micro data sources used in this chapter are described, comprising all years
since 1999. The datasets are corrected for administrative disturbances. This correction proce-
dure is described in this appendix.
The dataset of new disability benefits of 1999 consists of 91,500 records. Just 74,100 (81%)
of them stem from 1999, the remaining 19% are from earlier years. There are several reasons
for this. First, there is a common overflow from one year to another. People who had their
medical examination at the end of 1998 may be included in the administrative database only
several months later, in early 1999. Moreover, people who raise objections to the decision
about their application may receive their final decision several months later than one year af-
ter their first day of sick leave. Benefits that are awarded that way will be paid retrospectively
for the missed months. However, another large part of the new benefits with a starting year
earlier than 1999 are people who were already in the disability scheme earlier and resumed
working for a while but suffered a recurrence of the same health problem. In such a case the
old benefit is resumed. These administrative influences are corrected for by constructing new
cohort datasets. The cohort 1999 consists of all records from the 1999 and 2000 databases
that originate in 1999. Records from later databases that originate in 1999 are considered to
be re-entry cases that will be left out of the analysis. This results in the following corrected
year totals on cohort basis:
Table 3.11 Inflow on cohort basis (WAO+WIA) 1999–2009

1999 2000 2001 2002 2003 2004 2006 2007 2008 2009
Persons
Original 91,517 100,193 104,142 92,387 66,335 59,171 21,051 22,160 24,947 29,341
Cohort 93,009 94,872 94,765 78,114 58,882 50,716 22,081 22,263 24,575 26,179
Difference - 1,492 5,321 9,377 14,273 7,453 8,455 -1,030 -103 372 3,162
Fde
Original 73,135 77,562 76,583 67,848 49,015 44,583 19,886 20,303 22,787 27,104
Cohort 73,422 72,563 69,680 57,613 43,713 38,152 20,780 20,412 22,482 24,194
Difference -287 4,999 6,903 10,235 5,302 6,431 -894 -109 305 2,910
In 2004, the sickness period before the benefit became payable was extended from one to two
years. Therefore, inflow into the disability scheme is extremely low in 200541. In the other
years, significant differences can be seen between the administrative inflow totals and the
corrected cohort totals. For example, according to the administrative data, between 1999 and
2001, a significant inflow rise of almost 13,000 persons seems to have taken place, but the co-
hort size of the 2001 inflow cohort is only 1,700 persons larger than the 1999 inflow cohort.
For outflow, the same procedure is followed as in Table 3.11 on inflow. The original numbers
are the numbers of outflow cases within a calendar year as reported by the benefit adminis-
tration office. The quarterly cohort totals are based on the termination dates as reported in
the micro data sets. All benefits that are terminated in years (t) and (t+1) are considered as
outflow cases also tend to have a lag of on average two to three months between the termina-
tion date of the benefit and the end of the administrative process. The corrected year totals on
cohort basis are presented in Tables 3.12 (including demographic outflow) and 3.13 (exclud-
ing demographic outflow).
Table 3.12 Outflow on cohort basis (WAO+WIA) 1999–2009

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Persons
Original 72,104 75,531 80,805 82,035 83,208 78,914 82,253 77,066 57,739 50,275 48,093
Cohort 71,643 75,198 77,963 80,465 84,125 77,714 78,005 71,696 56,185 49,113 46,034
Difference 461 333 2,842 1,570 -917 1,200 4,248 5,370 1,554 1,162 2,059
Fde
Original 55,930 57,409 59,136 58,912 60,662 58,552 55,590 51,286 42,110 37,906 37,938
Cohort 55,261 56,588 57,184 57,731 60,877 57,601 52,676 47,882 41,163 37,345 36,683
Difference 669 821 1,952 1,181 -215 951 2,914 3,404 947 561 1,255
41 Although not exactly zero. Because of the gatekeeper law from 2002, it is possible that a claim for a disability benefit
will be rejected because reintegration efforts of the employer are considered insufficient. In such cases, claims from
people who first became incapacitated in 2003 may be awarded, after some delay, in 2005.
154 chapter 3
Table 3.13 Outflow on cohort basis (excluding pensioning and mortality, WAO+WIA)
1999–2009
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Persons
Original 34,470 37,231 42,513 43,805 43,945 39,483 42,867 38,188 20,849 13,886 12,733
Cohort 34,257 36,880 39,629 42,460 44,665 38,273 38,913 34,686 19,444 12,788 11,213
Difference 213 351 2,884 1,345 -720 1,210 3,954 3,502 1,405 1,098 1,520
Fde
Original 21,821 22,804 24,893 24,993 25,963 23,964 21,468 17,981 10,968 8,037 9,100
Cohort 21,363 21,985 22,902 24,019 26,003 23,038 18,858 16,180 10,174 7,528 8,113
Difference 458 819 1,991 974 -40 926 2,610 1,801 794 509 987
Appendix 3.B Parameter estimates for WAO
Table 3.B.1 Multinomial logistic regression coefficients, WAO 2000 (reference group
fully disabled)
Partially disabled Recovery Mortality
Constant -6,296 (0,171) *** -1,867 (0,120) *** -7,104 (0,352) ***
Gender -0,538 (0,018) *** -0,115 (0,016) *** -0,548 (0,026) ***
Age 0,223 (0,008) *** 0,045 (0,006) *** 0,085 (0,015) ***
Duration -0,132 (0,015) *** -0,183 (0,012) *** -0,091 (0,015) ***
Age * Duration -0,004 (0,000) *** -0,003 (0,000) *** 0,000 (0,000)
Dummy Duration 0 -0,744 (0,037) *** -0,478 (0,030) *** -0,302 (0,062) ***
Dummy Duration 1 0,484 (0,028) *** 0,621 (0,024) *** 0,416 (0,048) ***
Dummy Duration 5 -0,118 (0,053) * -0,113 (0,048) * 0,012 (0,069)
Dummy Duration 10
Age ^2
Duration^2 0,008 (0,000) *** 0,010 (0,000) *** 0,003 (0,000) ***
partially disabled)
Fully disabled Recovery Mortality
Constant -3,814 (0,208) *** -3,308 (0,167) *** -6,043 (1,050) ***
Gender 0,192 (0,019) *** 0,161 (0,017) *** -0,513 (0,082) ***
Age 0,069 (0,010) *** 0,106 (0,008) *** -0,028 (0,044)
Duration -0,022 (0,012) -0,048 (0,013) *** -0,080 (0,045)
Age * Duration -0,002 (0,000) *** -0,004 (0,000) *** 0,000 (0,001)
Dummy Duration 0 -1,094 (0,044) *** -0,371 (0,033) *** -1,043 (0,193) ***
Dummy Duration 1 0,109 (0,031) *** 0,745 (0,026) *** -0,143 (0,139)
Dummy Duration 5 0,128 (0,040) ** 0,079 (0,044) 0,153 (0,139)
Dummy Duration 10
Age ^2
Duration^2 0,004 (0,000) *** 0,008 (0,000) *** 0,002 (0,001) ***
156 chapter 3
fully disabled)
Constant -6,566 (0,179) *** -2,502 (0,125) *** -8,018 (0,380) ***
Gender -0,140 (0,018) *** 0,220 (0,016) *** -0,514 (0,026) ***
Age 0,224 (0,008) *** 0,066 (0,006) *** 0,124 (0,016) ***
Duration -0,118 (0,013) *** -0,026 (0,010) * -0,095 (0,016) ***
Age * Duration -0,002 (0,000) *** -0,005 (0,000) *** 0,000 (0,000)
Dummy Duration 0 -0,717 (0,044) *** -0,345 (0,033) *** 0,329 (0,061) ***
Dummy Duration 1 0,492 (0,030) *** 0,535 (0,025) *** 0,883 (0,049) ***
Dummy Duration 5 0,511 (0,028) *** 0,520 (0,024) *** -0,045 (0,056)
Dummy Duration 10 0,460 (0,056) *** 0,329 (0,054) *** -0,079 (0,077)
Age ^2
Duration^2 0,004 (0,000) *** 0,006 (0,000) *** 0,003 (0,000) ***
partially disabled)
Constant -3,718 (0,226) *** -2,566 (0,155) *** -7,382 (1,054) ***
Gender 0,076 (0,020) *** 0,314 (0,016) *** -0,543 (0,073) ***
Age 0,046 (0,010) *** 0,064 (0,007) *** 0,026 (0,044)
Duration -0,032 (0,013) * 0,012 (0,012) -0,066 (0,045)
Age * Duration -0,001 (0,000) *** -0,005 (0,000) *** 0,001 (0,001)
Dummy Duration 0 -0,549 (0,051) *** -0,519 (0,035) *** -0,698 (0,208) ***
Dummy Duration 1 0,453 (0,034) *** 0,682 (0,025) *** 0,420 (0,123) ***
Dummy Duration 5 0,170 (0,035) *** 0,483 (0,026) *** -0,233 (0,131)
Dummy Duration 10 0,075 (0,063) 0,634 (0,054) *** 0,168 (0,151)
Age ^2
Duration^2 0,002 (0,000) *** 0,007 (0,000) *** 0,001 (0,000)
fully disabled)
Constant -11,151 (0,607) *** -3,499 (0,457) *** -7,068 (0,739) ***
Gender -0,178 (0,039) *** -0,280 (0,038) *** -0,549 (0,035) ***
Age 0,361 (0,026) *** 0,061 (0,021) ** 0,057 (0,030)
Duration -0,175 (0,027) *** -0,197 (0,023) *** -0,032 (0,024)
Age * Duration 0,002 (0,001) *** 0,001 (0,000) 0,000 (0,000)
Dummy Duration 0 0,950 (0,263) *** 1,890 (0,148) *** 0,015 (0,586)
Dummy Duration 1 0,635 (0,156) *** 0,757 (0,121) *** -0,042 (0,287)
Dummy Duration 5 -0,129 (0,077) -0,247 (0,072) *** 0,240 (0,078) **
Dummy Duration 10
Age ^2
Duration^2 0,000 (0,000) 0,003 (0,000) *** 0,002 (0,000) ***
partially disabled)
Constant -9,461 (0,501) *** -3,512 (0,592) *** -9,695 (2,219) ***
Gender -0,051 (0,028) -0,049 (0,044) -0,590 (0,091) ***
Age 0,285 (0,021) *** 0,076 (0,026) ** 0,053 (0,085)
Duration 0,032 (0,021) -0,196 (0,032) *** 0,157 (0,068) *
Age * Duration -0,002 (0,000) *** 0,002 (0,001) ** -0,003 (0,001) *
Dummy Duration 0 -0,831 (0,719) 2,106 (0,248) *** -11,020 (546,203)
Dummy Duration 1 -0,099 (0,230) 0,975 (0,184) *** -0,030 (1,014)
Dummy Duration 5 0,140 (0,052) ** 0,026 (0,076) 0,215 (0,174)
Dummy Duration 10
Age ^2
Duration^2 0,002 (0,000) *** 0,002 (0,000) *** 0,001 (0,001)
158 chapter 3
Appendix 3.C Parameter estimates for WIA
Table 3.C.1 Multinomial logistic regression coefficients, WIA 2009 (reference group
fully disabled IVA)
Partially disabled Partially disabled Recovery Mortality
WGA working WGA unemployed
Constant -4,906 (2,377) * -6,869 (1,008) ***
Gender -0,343 (0,269) -0,003 (0,064)
Age 0,043 (0,108) 0,132 (0,041) **
Duration
Age * Duration
Dummy Duration 0 -0,254 (0,323) 0,195 (0,080) *
Dummy Duration 1 -0,092 (0,315) 0,667 (0,073) ***
Age ^2 -0,001 (0,001) -0,001 (0,000) **
Duration^2
fully disabled WGA)
Constant -9,966 (1,260) *** -4,565 (0,740) *** -4,095 (0,360) *** -8,441 (1,283) ***
Gender -0,227 (0,106) * -0,330 (0,080) *** -0,092 (0,038) * -0,259 (0,100) **
Age 0,176 (0,051) *** -0,008 (0,031) 0,085 (0,016) *** 0,080 (0,050)
Duration 0,178 (0,398) -0,542 (0,270) * 0,122 (0,123) 0,661 (0,442)
Age * Duration 0,009 (0,005) 0,023 (0,005) *** 0,006 (0,002) *** 0,015 (0,006) **
Dummy Duration 0 1,113 (0,388) ** 1,641 (0,287) *** 1,044 (0,128) *** 0,439 (0,380)
Dummy Duration 1 0,727 (0,231) ** 0,678 (0,216) ** 0,156 (0,086) 0,414 (0,198) *
Age ^2 -0,002 (0,001) *** 0,000 (0,000) -0,002 (0,000) *** -0,001 (0,001)
Duration^2 -0,044 (0,063) -0,145 (0,050) ** -0,070 (0,021) *** -0,272 (0,069) ***
partially disabled WGA working)
Constant -5,000 (1,253) *** -0,529 (1,144) -7,547 (1,624) *** -29,796 (10,465) **
Gender -0,100 (0,104) -0,193 (0,116) -0,002 (0,136) -0,267 (0,314)
Age 0,071 (0,052) -0,100 (0,049) * 0,148 (0,066) * 0,644 (0,347)
Duration 0,871 (0,340) * 0,241 (0,313) 2,192 (0,546) *** 5,759 (2,119) **
Age * Duration 0,012 (0,006) * 0,008 (0,005) 0,020 (0,010) * -0,060 (0,032)
Dummy Duration 0
Dummy Duration 1
Age ^2 -0,001 (0,001) 0,001 (0,001) -0,003 (0,001) *** -0,004 (0,003)
Duration^2 -0,278 (0,050) *** -0,103 (0,050) * -0,483 (0,087) *** -0,513 (0,177) **
partially disabled WGA unemployed)
Constant -5,913 (0,927) *** -4,708 (0,784) *** -7,049 (1,608) *** -21,121 (8,840) *
Gender 0,240 (0,087) ** 0,120 (0,079) 0,259 (0,165) -0,519 (0,386)
Age 0,114 (0,039) ** 0,121 (0,035) *** 0,156 (0,073) * 0,449 (0,322)
Duration 1,233 (0,254) *** 0,948 (0,222) *** 1,291 (0,510) * 2,542 (1,729)
Age * Duration 0,006 (0,005) 0,004 (0,004) 0,029 (0,011) ** -0,003 (0,029)
Dummy Duration 0
Dummy Duration 1
Age ^2 -0,002 (0,000) *** -0,002 (0,000) *** -0,003 (0,001) *** -0,003 (0,003)
Duration^2 -0,265 (0,041) *** -0,270 (0,037) *** -0,425 (0,092) *** -0,453 (0,174) **
160 chapter 3
Appendix 3.D Parameter estimates for Wajong
Table 3.D.1 Multinomial logistic regression coefficients, Wajong 2000 (reference group
fully disabled)
Constant -2,687 (1,307) * -5,087 (0,317) *** -4,807 (0,494) ***
Gender 0,321 (0,134) * -0,874 (0,053) *** -0,221 (0,065) ***
Age -0,130 (0,085) 0,124 (0,018) *** -0,046 (0,026)
Duration 0,177 (0,060) ** 0,004 (0,016) 0,003 (0,027)
Age * Duration
Dummy Duration 0 -1,225 (0,604) * 0,370 (0,125) ** -1,122 (0,467) *
Dummy Duration 1
Dummy Duration 5 -0,031 (0,264) 0,422 (0,082) *** -0,017 (0,228)
Dummy Duration 10
Dummy Age 18 -1,149 (0,175) *** -0,761 (0,065) *** -0,106 (0,109)
Age ^2 -0,008 (0,002) *** -0,006 (0,001) *** 0,000 (0,001)
Duration^2 0,000 (0,001) -0,001 (0,000) *** 0,001 (0,000) ***
partially disabled)
Constant -1,354 (0,311) *** -3,687 (0,395) *** -7,718 (1,438) ***
Gender 0,108 (0,141) -0,573 (0,182) ** -0,376 (0,559)
Age -0,025 (0,008) ** 0,025 (0,007) *** 0,064 (0,024) **
Duration 0,005 (0,038) 0,133 (0,051) ** 0,000 (0,156)
Age * Duration
Dummy Duration 0 -1,140 (0,618) 1,580 (0,432) *** -9,850 (366,640)
Dummy Duration 1
Dummy Duration 5
Dummy Duration 10
Dummy Age 18
Age ^2 -0,001 (0,002) -0,008 (0,002) *** 0,001 (0,006)
Duration^2
fully disabled)
Constant -4,028 (0,382) *** -3,536 (0,133) *** -6,334 (0,283) ***
Gender 0,951 (0,119) *** -0,322 (0,041) *** -0,141 (0,064) *
Age -0,040 (0,013) ** 0,017 (0,004) *** 0,035 (0,008) ***
Duration -0,003 (0,038) 0,160 (0,014) *** -0,058 (0,016) ***
Age * Duration -0,001 (0,001) -0,005 (0,000) *** 0,001 (0,000) ***
Dummy Duration 0 -1,889 (0,522) *** 1,025 (0,091) *** -1,302 (0,518) *
Dummy Duration 1 -0,431 (0,239) 0,157 (0,095) -0,450 (0,302)
Dummy Duration 5 0,549 (0,171) ** 0,838 (0,066) *** -0,359 (0,288)
Dummy Duration 10
Dummy Age 18 -1,188 (0,133) *** -0,907 (0,049) *** -0,074 (0,098)
Age ^2
Duration^2
partially disabled)
Constant -1,148 (0,430) ** -2,261 (0,328) *** -9,447 (2,133) ***
Gender -0,281 (0,155) -0,069 (0,143) 0,857 (0,858)
Age -0,048 (0,015) ** -0,006 (0,009) 0,059 (0,047)
Duration 0,054 (0,041) 0,092 (0,042) * 0,096 (0,210)
Age * Duration
Dummy Duration 0 -0,165 (0,462) 1,978 (0,263) *** -9,397 (347,040)
Dummy Duration 1
Dummy Duration 5
Dummy Duration 10
Dummy Age 18 -0,115 (0,183) -0,503 (0,179) ** 1,062 (0,995)
Age ^2 -0,002 (0,001) -0,006 (0,002) *** -0,004 (0,007)
Duration^2
162 chapter 3
fully disabled)
Constant -4,924 (0,786) *** -3,426 (0,354) *** -5,992 (0,399) ***
Gender -0,371 (0,109) *** -0,728 (0,053) *** 0,052 (0,056)
Age 0,046 (0,049) 0,020 (0,022) -0,016 (0,022)
Duration -0,127 (0,034) *** -0,014 (0,015) 0,001 (0,019)
Age * Duration
Dummy Duration 0 -2,073 (0,519) *** -0,601 (0,150) *** -0,994 (0,397) *
Dummy Duration 1 -0,836 (0,205) *** -0,382 (0,092) *** -0,220 (0,205)
Dummy Duration 5
Dummy Duration 10 0,844 (0,243) *** 0,063 (0,137) -0,185 (0,263)
Dummy Age 18 -1,097 (0,138) *** -1,018 (0,064) *** -0,010 (0,084)
Age ^2 0,002 (0,001) -0,003 (0,001) *** 0,000 (0,000)
Duration^2 -0,001 (0,001) 0,000 (0,000) 0,001 (0,000) ***
partially disabled)
Constant -1,925 (0,331) *** -4,799 (0,434) *** -7,814 (3,038) *
Gender -0,103 (0,151) -0,216 (0,229) -0,028 (1,037)
Age -0,026 (0,013) * 0,043 (0,013) *** -0,052 (0,123)
Duration 0,002 (0,015) -0,029 (0,016) 0,176 (0,135)
Age * Duration
Dummy Duration 0
Dummy Duration 1
Dummy Duration 5
Dummy Duration 10
Dummy Age 18
Age ^2
Duration^2
Appendix 3.E Model parameters
Table 3.E.1 Model parameters inflow/outflow models

Quarter Exp. Pre- Gate Stricter WIA Re-exam Re-exam Business Business Mean
Rating mium keeper examin DI stock system cycle cycle inflow
rate inflow outflow (t-1,t-8)
1999Q1 0.36 0.23 0 0 0 0 0 0.627 0.794 0.398
1999Q2 0.36 0.23 0 0 0 0 0 0.707 0.852 0.398
1999Q3 0.36 0.23 0 0 0 0 0 0.781 0.908 0.398
1999Q4 0.36 0.23 0 0 0 0 0 0.847 0.965 0.398
2000Q1 0.65 0.29 0 0 0 0 0 0.907 1.019 0.397
2000Q2 0.65 0.29 0 0 0 0 0 0.957 1.061 0.396

2000Q3 0.65 0.29 0 0 0 0 0 0.992 1.077 0.395
2000Q4 0.65 0.29 0 0 0 0 0 1.006 1.048 0.395
2001Q1 0.70 0.05 0 0 0 0 0 0.989 0.958 0.394
2001Q2 0.70 0.05 0 0 0 0 0 0.929 0.797 0.393
2001Q3 0.70 0.05 0 0 0 0 0 0.819 0.562 0.392
2001Q4 0.70 0.05 0 0 0 0 0 0.654 0.260 0.390
2002Q1 0.89 0.19 0 0 0 0 0 0.435 -0.088 0.385
2002Q2 0.89 0.19 0.125 0 0 0 0 0.171 -0.456 0.378
2002Q3 0.89 0.19 0.375 0 0 0 0 -0.124 -0.811 0.367
2002Q4 0.89 0.19 0.625 0 0 0 0 -0.431 -1.121 0.352
2003Q1 1 0.11 0.875 0 0 0 0 -0.725 -1.362 0.335
2003Q2 1 0.11 1 0 0 0 0 -0.986 -1.517 0.317
2003Q3 1 0.11 1 0 0 0 0 -1.196 -1.581 0.298
2003Q4 1 0.11 1 0 0 0 0 -1.341 -1.560 0.278
2004Q1 1 -0.01 1 0 0 0 0 -1.415 -1.468 0.258
2004Q2 1 -0.01 1 0 0 0 0 -1.421 -1.325 0.245
2004Q3 1 -0.01 1 0 0 0 0 -1.365 -1.149 0.234
2004Q4 1 -0.01 1 1 0 0 1 -1.258 -0.955 0.223
2005Q1 1 -0.29 Na Na Na 0.45 1 Na -0.760 0.186
2005Q2 1 -0.29 Na Na Na 0.57 1 Na -0.557 0.159
2005Q3 1 -0.29 Na Na Na 0.75 1 Na -0.340 0.131
2005Q4 1 -0.29 Na Na Na 1.00 1 Na -0.100 0.103
2006Q1 1 -0.29 1 1 1 0.79 1 -0.649 0.170 0.085
2006Q2 1 -0.29 1 1 1 0.70 1 -0.425 0.474 0.073
2006Q3 1 -0.29 1 1 1 0.65 1 -0.173 0.803 0.058
2006Q4 1 -0.29 1 1 1 0.75 1 0.092 1.140 0.046
164 chapter 3
(continuation) Table 3.E.1 Model parameters inflow/outflow models

Quarter Exp. Pre- Gate Stricter WIA Re-exam Re-exam Business Business Mean
Rating mium keeper examin DI stock system cycle cycle inflow
rate inflow outflow (t-1,t-8)
2007Q1 1 0.11 1 1 1 0.63 1 0.347 1.453 0.058

2007Q2 1 0.11 1 1 1 0.43 1 0.578 1.713 0.069
2007Q3 1 0.11 1 1 1 0.28 1 0.770 1.880 0.080
2007Q4 1 0.11 1 1 1 0.17 1 0.910 1.920 0.092
2008Q1 1 -0.28 1 1 1 0.23 1 0.985 1.800 0.093
2008Q2 1 -0.28 1 1 1 0.24 1 0.990 1.505 0.094
2008Q3 1 -0.28 1 1 1 0.29 1 0.920 1.038 0.095
2008Q4 1 -0.28 1 1 1 0.38 1 0.787 0.435 0.096
2009Q1 1 -0.04 1 1 1 0.25 1 0.608 -0.238 0.099
2009Q2 1 -0.04 1 1 1 0.08 1 0.434 -0.845 0.101
2009Q3 1 -0.04 1 1 1 0 1 0.294 -1.291 0.102
Appendix 3.F Detailed output of sector models
Table 3.F.1 GLS estimates of sector inflow models42

Agriculture Manufacturing Trading
Intercept 1.154 (0.057) *** 0.331 (0.010) *** 0.285 (0.009) ***
Experience rating -0.371 (0.066) *** -0.061 (0.010) *** -0.055 (0.007) ***
Premium rate -0.041 (0.049) -0.008 (0.007) -0.008 (0.005)
Gatekeeper protocol -0.282 (0.050) *** -0.059 (0.007) *** -0.067 (0.005) ***
Gatekeeper protocol * Time -0.016 (0.009) -0.004 (0.001) ** -0.002 (0.001) *
Stricter examinations -0.109 (0.069) -0.023 (0.011) * -0.028 (0.007) ***
WIA -0.107 (0.066) -0.073 (0.011) *** -0.061 (0.008) ***
WIA * Time 0.022 (0.011) 0.005 (0.001) ** 0.003 (0.001) *
Vacancy rate 0.003 (0.002) 0.000 (0.000) 0.000 (0.000)
Quarter 2 -0.204 (0.024) *** -0.043 (0.004) *** -0.036 (0.003) ***
Quarter 3 -0.203 (0.036) *** -0.033 (0.004) *** -0.035 (0.003) ***
Quarter 4 -0.075 (0.031) * -0.023 (0.004) *** -0.024 (0.003) ***
Quarter 2 * stricter exam. 0.140 (0.039) ** 0.034 (0.006) *** 0.029 (0.004) ***
Quarter 3 * stricter exam. 0.147 (0.042) ** 0.028 (0.006) *** 0.031 (0.005) ***
Quarter 4 * stricter exam. 0.049 (0.042) 0.012 (0.007) 0.019 (0.005) ***
N 39 39 39
R2 0.989 0.996 0.998
Autocorrelation order na na na
Dw (P-value) 1.54 (0.02) 2.03 (0.19) 1.63 (0.02)
# employees in sector 121,000 892,000 1,276,000
42 In the alternative model using cross-terms of quarter and WIA (see footnote 27), coefficients for stricter examinations de-
crease to -0.061, -0.011 and -0.009 respectively. The WIA coefficients increase to -0.155, -0.086 and -0.080 respectively.
All other coefficients remain unchanged.
166 chapter 3
Table 3.F.2 GLS estimates of sector inflow models43

Transportation Hotels / restaurants Flexi-work
Intercept 0.323 (0.021) *** 0.346 (0.014) *** 0.375 (0.020) ***
Experience rating -0.050 (0.020) * -0.141 (0.011) *** -0.035 (0.028)
Premium rate -0.013 (0.011) -0.001 (0.001) -0.013 (0.019)
Gatekeeper protocol -0.078 (0.009) *** -0.045 (0.009) *** -0.107 (0.018) ***
Gatekeeper protocol * Time -0.004 (0.002) * -0.002 (0.002) 0.001 (0.003)
Stricter examinations -0.028 (0.015) -0.006 (0.016) -0.035 (0.029)
WIA -0.067 (0.017) *** -0.061 (0.014) *** -0.107 (0.028)
WIA * Time 0.006 (0.002) ** 0.003 (0.002) 0.002 (0.004)
Vacancy rate44 0.000 (0.000) -0.000 (0.000)
Quarter 2 -0.038 (0.005) *** -0.035 (0.008) *** -0.049 (0.011) ***
Quarter 3 -0.023 (0.006) *** -0.034 (0.010) ** -0.085 (0.011) ***
Quarter 4 -0.008 (0.006) -0.009 (0.008) 0.009 (0.011)
Quarter 2 * stricter exam. 0.029 (0.009) ** 0.021 (0.011) 0.040 (0.017) *
Quarter 3 * stricter exam. 0.013 (0.009) 0.022 (0.015) 0.068 (0.017) ***
Quarter 4 * stricter exam. 0.001 (0.010) -0.000 (0.012) -0.008 (0.018)
N 39 39 39
R2 0.993 0.992 0.976
Autocorrelation order na (2) na
Dw (P-value) 2.10 (0.27) 1.76 (0.05) 1.53
# employees in sector 445,000 268,000 519,000
43 In the alternative model using cross-terms of quarter and WIA (see footnote 27), coefficients for stricter examinations de-
crease to -0.027, -0.006 and -0.044 respectively. The WIA coefficients increase to -0.068, -0.060 and -0.099 respectively.
All other coefficients remain unchanged.
44 Not available for flexi-work sector.
4
The Social Affairs Department
of the Netherlands Ageing and
Pensions model45
This chapter describes a newly extended version of the dynamic microsimulation Social Af-
fairs Department of the Netherlands Ageing and Pensions model (SADNAP). SADNAP is
being developed for calculating the financial and economic implications of the ageing of the
population and of the ageing-related policy measures that are being proposed to cope with
ageing. The model uses administrative datasets of Dutch public pension payments and entitle-
ments for both public and private pensions. SADNAP has already been in use since 2007 for
forecasting the state pension expenditures and for analysing the budgetary effects of policy
changes.
The model has been extended in order to give a broader assessment of policy alternatives by
providing insight into other important evaluation indicators like income redistribution and the
retirement decision of workers. For the modelling of income redistribution a new micro data
source with individual data on private pensions is combined with differentiation of mortality
rates in order to gain a better insight in the income at the individual level within the popula-
45 This chapter has been accepted for publication in the International Journal of Microsimulation (Van Sonsbeek, 2011).
168 chapter 4
tion of pensioners. For the modelling of the retirement decision, an option value model is
developed in which key parameters vary at the individual level in order to benefit from the
microsimulation approach. These extensions greatly enhance the performance of SADNAP.
Besides the financial implications, additional insight can now be provided into the effects of
policy measures on a set of key indicators.
In this chapter both extensions are described in detail and a complete baseline projection of
all key indicators is discussed.
4.1 Introduction
The Netherlands, like most OECD countries, is facing an ageing population. Especially, this
is a complication for the state pension, known as AOW, which is financed through a pay-as-
you-go system. The state pension is the first of the three pillars of the Dutch pension scheme.
The second pillar consists of supplementary company or occupational pension facilities.
Employees are obliged to take part in those second-pillar pension programmes when avail-
able. The third pillar contains individual pension saving programmes in which participation is
voluntarily. Both second- and third-pillar pensions are fully funded.
The dynamic microsimulation model SADNAP is being developed for calculating the finan-
cial and economic implications of the ageing problem and of the policy measures considered.
A microsimulation model, as compared with macro-oriented models, can give more detailed
information on the ageing problem and on the redistributive effects of policy options, which
can be used in the evaluation of those options. The model uses administrative datasets of all
Dutch public pensions and entitlements for all public pensions and a large share of private
pensions.
The Social Affairs Department of the Netherlands Ageing and Pensions model 169
The structure of this chapter is as follows. Section 4.2 gives a brief overview of the Dutch
pension system, the forecasting models currently in use at the Ministry of Social Affairs and
Employment and a short general introduction to microsimulation models and the SADNAP
model. Sections 4.3 and 4.4 present in more detail two recent extensions of the SADNAP
model. Section 4.3 focuses on the modelling of incomes and redistribution within the state
pension system and in section 4.4 the modelling of the retirement decision using the option
value approach is described. In section 4.5 the main results of the model are presented. These
results are limited to the baseline scenario of unchanged policies. Section 4.6, finally, con-
tains conclusions and some topics for future research.
4.2 Background
4.2.1 The Dutch pension system
The Dutch government supplies a state pension called AOW to all persons aged 65 or over
when they become entitled. Inhabitants of the Netherlands build up a right to this pension by
living in the Netherlands while aged between 15 and 65. The state pension accrues at the rate
of 2% for every year this condition is fulfilled. Part of the population is only partially entitled
because they have not lived in the Netherlands continuously between the ages of 15 and 65.
This share of incomplete state pensions is rising because of the growing number of immi-
grants during the last few decades.
The state pension scheme provides a basic minimum income guarantee in case of a full
entitlement. Therefore the system makes a distinction between partners of a couple and single
persons. A single person is entitled to 70% of the net minimum wage46 and a member of a
couple is entitled to 50% of the net minimum wage. Until 2015, persons with a (non-work-
ing) partner younger than 65 can supplement their state pension of 50% with an allowance
46 The gross minimum wage in 2009 amounts to approximately € 18,000 per year. The gross AOW benefit for a single
person is approximately € 12,700, the gross AOW benefit for a couple is approximately € 8,700 for each partner. In net
terms this amounts to 70% and 50% of the net minimum wage respectively.
170 chapter 4
of another 50% to a combined maximum of 100% of the minimum wage. Partially entitled
persons can claim social assistance. Social assistance, however, is income and means tested.
The AOW is a pay-as-you-go arrangement, the current population of workers pay for the cur-
rent population of pensioners. Hence the AOW is financed through a premium paid by these
workers. The premium is fixed at a rate of 17.9% of the first two tax brackets (the limiting
income is approximately € 32,000 in 2009). Revenue from this premium is not sufficient to
cover all AOW costs, however. The government contributes the part of the AOW costs (cur-
rently about one third) that is not covered by the premiums. The government contribution is
financed by taxes, which are paid by pensioners as well.
The importance of second- and third-pillar pensions for the income position of the elderly is
growing as more people are saving for such pensions and their average savings are increas-
ing. Per person average second-pillar pension savings are almost equal now to the average
first-pillar state pension entitlements. In the future, it is to be expected that second- and third-
pillar pensions together will provide more than half of the average pension income. Although
there are many second-pillar pension funds in the Netherlands, each with its own rules on
contributions and pensions, broadly speaking one can say that pension funds try to supple-
ment the state pension to a total gross income level of 70% of the pensioner’s final wage.
Most pension funds have recently switched from a final wage system to a career average sys-
tem, but on average they still aim for a gross pension level of 70% of the final wage. Because
pensioners no longer have to pay state pension contributions, the net level of their first- and
second-pillar pensions together usually, in the case of a full pension, comes close to 90% of
the final wage. Other income sources, like third-pillar pensions can add to this income level.
4.2.2 Models currently in use
The Ministry of Social Affairs and Employment is responsible for preparing state pension
forecasts for the yearly budget. The budget horizon is six years (the current budget for 2010
contains forecasts from 2009 until 2014). Although beyond the budget horizon, the long-term
forecast of pension expenses is of great importance as well because government budgets are
also affected by the long-term sustainability of public finance. Besides the financial effects
for the government budget, the Minister of Social Affairs and Employment is also responsible
for income policies and labour participation policies. When new policy options are discussed,
a broad analysis will be required of both short-term and long-term financial effects, income
effects and labour participation effects. Moreover, in the case of ageing-related policy meas-
ures, income effects will not be limited to direct effects on purchasing power but will include
intra-generational and inter-generational redistribution issues as well. In order to assess all
these effects, a number of different models are used.
For state pension expenses, a simple macro model is used, using forecasts of the number of
pensioners for the most relevant subgroups of the state pension population (men and women,
singles and couples with and without a partner allowance, complete and reduced pensions).
These volume forecasts are supplied by the state pension administration office (SVB). The
macro model calculates the costs by multiplying the expected group sizes with the average
pensions for each group. As the SVB forecasts project ahead as far as 2024 and rely heavily on
extrapolating existing trends, for the long-term development of the state pension expenses the
Ministry relies on a macro AGE model of CPB. This model, called GAMMA (see Van Ewijk et
al., 2006), is used once every four years (in the run-up to the general elections) for a long-term
forecast of the whole Dutch economy. For income effects, the long running static microsimula-
tion model Micros (Hendrix, 1993) has been in use since the early 1990s. This model focuses
on the short-term income effects of complex sets of policy measures. Labour participation
effects are quantified on an ad-hoc basis using recent research papers by CPB and others. Re-
distribution effects are mostly abstracted from or quantified on an ad-hoc basis as well.
This approach has several shortcomings. Because different models from different internal
and external sources are used, it is very difficult to obtain a consistent picture of the effects of
policy measures. Besides, the Ministry is highly dependent on other institutions for supplying
172 chapter 4
information and forecasts. Therefore, it can be difficult to anticipate and respond quickly to
policy developments. Also, the quality and richness in detail of the forecasts can be improved
by using one consistent microsimulation model.
In the first place, information does get lost because the macro model uses only a small
number of groups sharing the same basic characteristics. Age groups are not included, for
example, although among the population of persons aged 65 and over, different ages may
have very different characteristics. Second, there are certain features of the AOW that cannot
simply be taken into account with macro models, such as changes in migration patterns and
changes in household situation. Migration affects the entitlements to the AOW because the
AOW entitlement depends on citizenship. Changes in the number and age of immigrants and
emigrants will affect the pension expenses later on. The AOW entitlement also depends on
household situation. Two single persons receive a higher pension than two persons in a cou-
ple, so when the number of singles among the population of pensioners rises, the cost of the
AOW will rise as well. Third, the macro models are limited to the state pensions that provide
the basic income level, whereas the main differences in the income positions of pensioners
are caused by private pensions. The Micros model, which is used for the income effects, is a
static model that is not capable of adequate long-term forecasts. Fourth, the effects on labour
participation and income redistribution are not captured at all by the current models in use at
the Ministry.
Therefore the Ministry has been developing the dynamic microsimulation model SADNAP
to handle the problems just mentioned. SADNAP is an integral ageing and pensions model,
including the income and redistributive effects of different policy measures. The purpose of
SADNAP is to provide consistent and integral forecasts of both short-term and long-term
effects of the baseline scenario of unchanged policies and various policy measures on the cost
of state pensions for government budget, the income position of the elderly, redistribution and
labour participation. SADNAP has already been used for budgetary forecasts since 2007.
4.2.3 Microsimulation models
Microsimulation basically is a modelling technique that uses large datasets containing data
on the individual level. Records on individual persons contain characteristics like birth year,
gender, ethnicity, income level, household status etc. Transition probabilities and institutional
rules are applied to simulate whether events will happen in the future to a specific record, e.g.
whether someone starts working or finishes a relationship. Calculation rules are used to apply
the probabilities and institutional rules to the micro data file. The result is an estimate of the
outcomes of applying these rules, including both the total aggregate change and the distribu-
tional nature of that change.
Microsimulation models can be subdivided in many different ways (O’Donoghue, 2001). The
most important one is between dynamic and static models. With dynamic microsimulation
the characteristics of a record can change over time. Static microsimulation does not allow
characteristics to change. Although in static simulations reweighing techniques can be used
to allow for changes in population composition, static microsimulation is usually seen as
more suited for short-term forecasts, like the short-term impact of fiscal measures, whereas
dynamic microsimulation is seen as more suited for long-term forecasts like the impact of
ageing.
Microsimulation is subject to Monte Carlo variability, resulting in different outcomes for

each individual simulation experiment. Of course, a larger sample can reduce the fluctuations
between different runs with the model, but not eliminate them. Moreover, in large dynamic
microsimulations sample size can still be limited due to disk capacity or computer speed. One
can deal with the Monte Carlo variability in several ways. First, several simulations can be
done and an average outcome can be calculated. The difference in average outcome between
the base situation and the policy alternative can then be accounted to the policy change. A
second approach is proposed by Klevmarken (2007), who describes a calibration technique in
174 chapter 4
which the simulation results are aligned to an a priori defined target, such as a macro forecast,
eliminating the variability. Third, Monte Carlo variance can be avoided completely by using
a fixed set of random numbers used to generate the events. This last method is useful to allow
for replication of model results and to compare policy alternatives with the base situation,
because when the random numbers are fixed, differences between two simulations can only
be caused by the policy change. For every individual a simulation of a policy alternative can
then be performed under exactly the same conditions as the simulation of the baseline sce-
nario. In SADNAP, both calibration and fixing of random numbers are used.
Microsimulation is very useful when information for specific individuals or groups of

individuals is needed. Information on specific groups can also be obtained by creating more
groups within cell-based macro forecasts. In practice, however, because of the large num-
ber of subgroups that arise when taking into account all the relevant characteristics, these
cell-based approaches become problematic when the subgroup size becomes too small (Van
Sonsbeek and Gradus, 2006).
However, construction of a dynamic microsimulation model can be very complex and time
consuming. This holds true especially for a dynamic population model, which requires
replacing the starting population with new cohorts over time. Cassels, Harding and Kelly
(2006) identify some success and failure factors and recommend that models should have
clear objectives, a modular design, be user friendly, produce timely output and be transparent.
With SADNAP these recommendations have been followed by initially limiting the model to
the budgetary impact of the state pensions only.
4.2.4 The SADNAP model
The Ministry of Social Affairs and Employment has been developing the SADNAP model
since 2006. As the model is modularly designed, attention was first focused on the demo-
graphic model and the state pension forecast. Therefore, since 2007 the SADNAP output has
been used in preparing the state pension budget forecasts of the Ministry. In later years, the
original demographic modules have been extended. The immigration and emigration code has
been improved in order to allow for the interdependency between the two (immigrants having
a higher emigration rate). Also, the take-up of state pensions by former Netherlands citizens
who have emigrated to other countries has been incorporated in the model. The household
formation code has been improved in order to provide reliable relationship patterns at the
micro level. In a new module, non-budgetary aspects (like income distribution and labour
participation/retirement decision) have been introduced in order to gain a more complete
picture of the pros and cons of different ageing-related policy measures.
In the early versions, the income was limited to the state pension (building up of entitle-
ments for the population aged 64 and below and pension payments for the population aged 65
and over). The income position has now been supplemented with private pension data, first
with rough estimates based on aggregate data and subsequently with a full micro data set on
private pension entitlements which has been supplied by Statistics Netherlands. A detailed
description of the demographic and income modules of SADNAP is given in appendix 4.A
and a detailed description of the micro and macro data sources used in SADNAP is supplied
in appendix 4.B. The flow diagram of the SADNAP model is given in Figure 4.1.
176 chapter 4
Figure 4.1 SADNAP model flow diagram
Population (t)
New births New immigrants
Mortality
Emigration
Household
formation
No Retired? Yes
Labour participation / wage Partner generation

( < 60 years old )
State / pricate pension

entitlement
State pension
Partner allowance
Private pension
Retirement decision
( >= 60 years old )
Population (t+1)
This chapter focuses on two extensions of the SADNAP model which have been imple-
mented recently. The first is the differentiation of mortality rates that is used to investigate the
redistribution within the state pension scheme, and which is described in section 4.3. The sec-
ond is the modelling of labour participation and the retirement decision, which is described in
section 4.4. Both extensions fill the gaps that are left in the assessment of policy alternatives
as described in section 4.2.2.
4.2.5 Comparison with other dynamic population microsimulation models
Within the Netherlands, SADNAP is the second attempt to develop a dynamic population
microsimulation model capturing ageing issues. The only comparable model in the Neth-
erlands is the NEDYMAS model (Nelissen, 1993), which was prominent during the 1990s.
Although SADNAP, as compared with some well known international simulation models, is
a comparatively simple and small-scale project, it shares some key characteristics with these
larger models. Cassels, Harding and Kelly (2006) present an overview of six large dynamic
population microsimulation models (Dynasim3 from the USA, Dynacan from Canada,
Mosart from Norway, Sesim from Sweden, Sage from the UK and Dynamod from Australia).
Like all the models but Dynamod, SADNAP is a discrete model with one-year time steps.
The development platform is SAS, as in Dynasim3. The sample size (1–2%) is comparable to
most models (e.g. Dynacan, Mosart, Dynamod). The time horizon (2080) is also comparable
to for example, Dynacan and Mosart (2100). In SADNAP results are aligned to targets taken
from macro data sources. As in Dynacan, for example, alignment targets include rates for
mortality, fertility, migration, marriage and divorce propensities. As in most of the models
mentioned, SADNAP contains modules on population, household formation, labour force
participation, benefits and taxation.
However, there are some simplifications as compared with the larger models. For example,
household formation in SADNAP is a binary choice between single and cohabiting, which
excludes, for example, children leaving home, people moving in and out institutions and
adults living in other households without family relations (cf. Mosart). Taxation is included
in SADNAP, as in most other models (except Dynacan), but is simplified to the main tariff
structure. Education and health are not included in SADNAP. Financial wealth and sav-
ings are also not included, but they are planned for an extension in the future. SADNAP is
comparatively narrow in scope, like for example Dynacan, so most effort is put into subjects
directly related to pensions and ageing. In the current SADNAP version, most effort has been
put into the retirement decision model, which consequently is comparatively elaborate.
178 chapter 4
4.3 Modelling redistribution within the state pension system
The Dutch state pension scheme can be classified as a “Beveridge”-style public pension pro-
gramme (Disney, 2004), characterized by significant departures from actuarial fairness and
significant provision of private retirement benefits, as opposed to “Bismarck”-style public
pension programmes, characterized by high “actuarial fairness” and limited private provision
of private retirement benefits. The Dutch scheme, with its flat-rate pensions for single persons
and cohabitants, therefore has a highly intra-generational redistributive character.
There is also redistribution from higher to lower incomes because higher income earners
contribute more to the scheme during their lifetime. However, this only holds true for income
differences up to the limiting income of approximately € 32,000 (in 2009). For the moment,
this kind of intergenerational redistribution is not included. Additional research has to be
done in order to identify which groups have a better balance of withdrawals as compared with
their contributions.
Typically, subgroups with lower life expectancies on average contribute more than they
withdraw from the scheme. Well known factors affecting life expectancy are gender, income,
marital status and ethnic background. Gender- and age-specific mortality rates are derived
from the CBS population forecast and have been used in SADNAP from the beginning. How-
ever, there are also notable differences in mortality rates between different income levels,
between single persons and couples and between different ethnicities. From a redistribution
perspective those differences are important although they are not easy to implement in a
simulation model because of alignment problems. Moreover, these possible causes correlate,
complicating the analysis of the ground cause of the differences in mortality.
On the relationship between income and life expectancy, in a large Finnish study Martikainen
et al. (2001) show the mortality rates of the lowest income decile on average to be 2.37 times
as high for men over 30 years of age and 1.73 times as high for women over 30 years of age
as those in the highest income decile. On the relationship between marital status and life
expectancy, de Jong (2002) shows the mortality rates of married people to be significantly
lower than those of single, divorced or widowed persons. The difference is larger for men
than for women, and increases in time for both men and women. However, the differences in
mortality rates are smaller in the higher age categories. On the relationship between ethnic
background and life expectancy, Bos et al. (2004) show mortality to be higher among three
of the four largest groups of immigrant males in the Netherlands. However, among Moroccan
males, mortality appeared to be lower and among females in general, inequalities in mortality
are small. Moreover, mortality rates are influenced by marital status and socio-economic sta-
tus, leaving a smaller influence of ethnic background in itself, except for younger age catego-
ries. This contrasts with data from SVB (2008) showing the mortality age of people not born
in the Netherlands to be significantly lower than that of people born in the Netherlands, with
differences in average mortality age of more than 10 years between natives and Turks and
Moroccans. On average, people with reduced state pensions, including most first-generation
immigrants, live four years less than people with full state pensions, according to this study.
In SADNAP, the differences in mortality rates by income are derived from the study of
Martikainen et al. (2001). The expected total private pension is used as a proxy for income.
This means that people do not move between income deciles, only one “lifetime” decile is
assigned per person. The estimation of the pension entitlements has been improved recently
because a detailed micro dataset of company pensions has become available. This dataset
is described in appendix 4.B. The wage level of the participants is known for the base year.
Their pension entitlements are based on continuation of their current wage level throughout
their working life. That means that the younger one is in the base year, the less accurate the
pension entitlement forecast is as wages are expected to rise during the working life. Wages
in the Netherlands are strongly correlated with age. Figure 4.2 presents the average wages
and expected pensions by age based on the 2005 micro dataset. The picture strongly resem-
bles earlier findings on age–earnings profiles, like those from a longitudinal Dutch survey
(see Alessie, Lusardi and Aldershof (1997) and more recently Kalmijn and Alessie (2008)).
180 chapter 4
They found that the age–earnings profile shows a steep profile for the young, subsequently
a moderate increase over the life cycle and finally a sharp decline well before the mandatory
retirement age of 65.
Figure 4.2 Average wage and expected second-pillar pension by age, 2005
Currently SADNAP lacks the more elaborate modelling of wages over the life cycle of, for
example, Borella and Coda Moscarola (2005). However, when wages and pension savings
are assumed to follow each other’s development over the life cycle, the replacement rates will
remain the same. Only, as earlier noted, the replacement rates for the youngest age cohorts
provide no good guidance for the replacement rates at older ages. Figure 4.2 suggests that
from age 30–35 onwards reasonably accurate projections of future income development can
be made.
By introducing this difference in mortality rates by income, differences in mortality rates

by household status and ethnic group are introduced at the same time, as single persons on
average have lower incomes than cohabitants and immigrants on average have lower incomes
than natives. This also results in higher mortality among people with reduced state pensions,
mainly immigrants, an observation that also is made in SVB (2008). A further difference in
mortality rates between single persons and cohabitants and between natives and immigrants
is introduced in order to increase the differences in life expectancy to the levels reported in
the studies of de Jong and Bos et al. respectively.
Table 4.1 shows the life expectancies (at age 65) for different subgroups of the population of
pensioners for the 2006–2045 cohorts of pensioners (the 1941–1980 birth cohorts). As the
simulation runs until 2080, the 2045 pensioner cohort is one of the last cohorts that will have
almost completely died out by 2080. Besides the familiar difference in life expectancy be-
tween men and women, there are also sizeable differences between different income groups,
between cohabitants and single persons and between natives and non-natives. The differences
between income groups are in line with recent Dutch research by Kalwij, Alessie and Knoef
(2009). The average expected age of the cohorts turning 65 is 86. This is consistent with CBS
(2009), in which life expectancy at 65 years old rises from 19.4 years (17.8 for men and 21.0
for women) in 2009 to 21.8 years (20.8 for men and 22.9 for women) in 2045. The differ-
ence in life expectancy between men and women is decreasing over time until a difference of
about two years is left. The differences in life expectancy between income quintiles, single
persons and cohabitants, and natives and first-generation immigrants are assumed to remain
constant over time.
182 chapter 4
Table 4.1 Projected average life expectancies at 65 of the 2006–2045 pensioner

cohorts
Subgroup Average mortality age Δ Average
By income
1st quintile 83.5 -2.5
2nd quintile 84.5 -1.5
3rd quintile 85.6 -0.4
4th quintile 86.8 +0.8
5th quintile 88.5 +2.5
By gender
Women 86.9 +0.9
Men 85.1 -0.9
By household status
Singles 84.4 -1.6
Cohabitants 86.7 +0.7
By origin
Natives 86.6 +0.6
Immigrants 84.0 -2.0
Total 86.0
The population decomposition used allows for an analysis of redistribution within the state
pension scheme by aggregating pension payments for each subgroup. Such an analysis is
presented in section 4.5.2.
4.4 Modelling the retirement decision of employees
In most current literature the retirement decision is modelled by using the option value model
by Stock and Wise (1990). More and more often, this approach is implemented in microsimu-
lation models (e.g. Dekkers, 2007). In the option value model, the individual chooses the
optimal retirement age R* by maximizing the expected lifetime utility from both consump-
tion (labour income) and leisure (retirement income). In this decision the expected value of
all current and future incomes Vt(R) at all possible retirement ages t is considered.
(4.1)
Here β (=1/1+ρ) represents the discount factor (with ρ the time preference parameter), p(s|t)
the survival probability, Uy the utility of consumption, Ys the labour income, γ the risk-
aversion parameter, Ub the utility of leisure, k the leisure preference parameter and Bs(R) the
income after retirement. Often, the option value model is simplified (Euwals, Van Vuuren and
Wolthoff, 2006) by fixing the parameters γ, k and ρ at some given values, but in a microsimu-
lation model, heterogeneity in the parameters can be implemented straightforwardly. Also
the peak value model, as proposed by Coile and Gruber (1998) and discussed by Samwick
(2001), can be considered a simplification of the option value model. In the peak value model
future earnings no longer play a role in the retirement decision. This approach chooses the
retirement age that maximizes the expected lifetime retirement income. Abstracting from
future earnings allows us to set the leisure preference parameter k to 1, which as Samwick
(2001) notes, seems counterintuitive but as peak value compares income flows only during
retirement, this assumption is not restrictive. The values of the option value parameters vary
widely in the literature and differ from the original estimates of Stock and Wise (γ = 0.63, k
= 1.25 and ρ = 0.28). Euwals, van Vuuren and Wolthoff (2006) propose parameter values for
the Netherlands of γ = 0.75, k = 1.7 and ρ = 0.04.
In SADNAP, assuming 60 to be the first and 70 to be the last possible retirement age, for each
individual the option value is computed for retirement ages 60 to 69. The utility functions Uy
and Ub equal labour and retirement income respectively. The model then depends on generic
gender-specific survival rates and the discount rate, leisure preference value, risk-aversion
value, labour income and retirement income that are all specific to the individual. The ex-
pected retirement age is set to the year (t) that maximizes the option value. In this retirement
184 chapter 4
decision the expected value of all current and future incomes Vt(R) is taken into account.
In the option value model, the role of the discount rate is important. In the original estimates
of Stock and Wise, based on utility rather than income, a very high discount rate of 0.28 (cor-
responding to a discount factor of 0.78) was estimated. In most later research (e.g. Börsch-
Supan, 2000, and Berkel and Börsch-Supan, 2004) much lower discount rates of 0.03 to 0.05
are used. In general, in the literature the estimates of the time preference parameter vary
within a wide range, as is shown in an overview by Frederick, Loewenstein and O’Donoghue
(2002). This suggests heterogeneity. Samwick (1998) notes that a distribution of preference
parameters like the discount rate should be assumed instead of a fixed value. Samwick finds
a median value of the discount rate of 0.08 for all ages (slightly lower for the 60–65 years
age group). He finds a distribution with 50% of discount rates between 0.03 and 0.15 but
also a large number of outliers with about 5% having negative discount rates of -0.15 and
below and 20% having discount rates of 0.2 and above. Also Gustman and Steinmeier (2005)
estimate a distribution of time preference with 40% between 0 and 0.05, 20% between 0.05
and 0.1 and a large group of over 25% having a very high time preference rate of over 0.5. In
SADNAP the findings from both studies are combined, taking advantage of the microsimula-
tion approach by applying a distribution of discount rates, with 20% having a discount rate
of 0, 20% uniformly distributed between 0 and 0.05, 20% uniformly distributed between
0.05 and 0.1, 20% uniformly distributed between 0.1 and 0.2 and 20% uniformly distributed
between 0.2 and 1.
The estimates of the leisure valuation parameter or rate of substitution between consump-
tion and leisure also vary widely. Stock and Wise estimate the parameter k at 1.25 whereas
Börsch-Supan et al. (2004) estimate k at 2.8. This may of course represent a difference in the
leisure valuation between the USA and continental Europe. However, most other studies, like
Bovenberg and Knaap (2005) who find an elasticity of substitution of 0.56 corresponding
to a k-value of 1.78, report values in between. On the difference between men and women,
Lise and Seitz (2007) report only a minor difference: they estimate k for men at 1.58 and for
women at 1.64. In the simulation model, the average is assumed to be 2.0 and a uniform dis-
tribution of leisure valuation rates between 1.0 and 3.0 is applied for both men and women.
Correlation between time preference and leisure preference was hypothesized and rejected by
Gustman and Steinmeier (2005).
The estimates of the risk-aversion parameter vary less. In general, people are risk averse in
their pension and retirement decisions. In the option value model, the lower the risk-aversion
parameter γ is, the earlier the retirement age will be. Stock and Wise estimate the parameter γ
at 0.63. Euwals, van Vuuren and Wolthoff propose 0.75. In a recent study, based on Austrian
data, Manoli, Mullen and Wagner (2009) estimate γ at 0.71 with a 95% confidence interval
between 0.49 and 0.81. In the simulation model, γ is assumed to have an average of 0.7 and a
uniform distribution between 0.5 and 0.9.
The future retirement incomes (both state pension and second-pillar pension) are easy to
predict at age 60, as most entitlements have been built up and mainly depend on institutional
parameters. However, the future labour income is more difficult to predict. A simple approach
would be to set the labour income for ages 61–70 equal to the labour income at age 60. For
the higher ages this may not be a good approach because of the decrease in productivity that
can be expected in combination with rising probabilities of becoming disabled or unem-
ployed, which the individual will take into account in his decision. Therefore the formula
for labour income in year (t+1) is specified as a function of labour income in year (t), the
expected yearly wage decrease τ due to productivity loss and the probability of becoming
disabled p(d|t) or unemployed p(u|t) during year t. It is assumed that both unemployment
and disability lead to an income loss of 30% as both disability and unemployment benefits
roughly equal 70% of the former wage47.
(4.2) (Y_(t+1) )= (1-τ)((1-p(d│t)-p(u│t)) Y_t+(p(d│t)+p(u│t))0.7Y_t )
47 With some exceptions: benefits for the permanently, fully disabled equal 75% of the former wage. Unemployment ben-
efits equal 75% of the former wage during the first two months of unemployment and 70% for the three years thereafter.
After those three years and two months, all people who become unemployed from age 60 onwards can claim a minimum
benefit to bridge the gap until retirement at 65.
186 chapter 4
For an indication of a plausible value for τ we can have a closer look at the age–earnings pro-
file of elderly workers. Figure 4.2 represents all wages including those of the self-employed
and of retirees working part-time and Table 4.8 represents the wages of the employees only.
From Figure 4.2, it appears that the average wage at age 64 is 38% lower than at age 59,
which corresponds to a value of τ of 0.09. From Table 4.8, it appears that the 60–64 year
olds earn almost the same as the 55–59 year olds, which corresponds to a value of τ of zero.
The latter intuitively corresponds to a society in which demotion is almost non-existent. The
wage decrease from Figure 4.2 reflects both overrepresentation of self-employed, who work
longer but earn less, and employees working less hours, either by preference or due to their
health. It can be concluded that people who work on until 65 will have no loss of income,
but that when also the employees who by preference or due to their health work less than
20 hours a week (who are considered retired) are taken into account, an income loss exists.
Average values of τ of 0 and 0.045 were tested and it is concluded that in the τ=0 scenario the
share of the population working until the last possible retirement age (of 70) is unrealistically
high, as compared with Euwals and Folmer (2009). In the τ=0.045 scenario, a close match
with Euwals and Folmer (2009) is made for males (10% of the 65–70 year olds participating
on average). Therefore τ is assumed to have an average of 0.045 and a uniform distribution
between 0 and 0.09.
In the present version of SADNAP the option value approach is used for the retirement
decision of the birth cohorts from 1946 until 1970. The 1946 cohort is aged 60 in 2006, the
starting year of the simulation. The 1970 cohort is aged 35 in 2006 and above it was noted
that wages and pensions are known with enough accuracy from about the of age 35 onwards.
The option value approach computes an expected retirement age based on a forward-looking
calculation. In reality, events like unemployment and disability will influence the retirement
decision. Therefore, after determining the optimal retirement age at 60, all individuals work
through until the optimal retirement age unless they become unemployed or disabled. As both
unemployment and disability can be considered absorbing states from age 60 onwards48, in
that case the year that one becomes unemployed or disabled is considered to be the year of
48 In the Netherlands, the unemployment benefit currently lasts for a maximum of five years for people aged 60 and over.
The unemployment benefit itself lasts for a maximum of three years and two months and the subsequent benefit for
people aged 60 years and over complements the time until retirement.
retirement. Unemployment and disability probabilities are observed in 2008 for the ages 60
through to 64. Unemployment and disability probabilities for age 65 onwards are considered
to be equal to those observed at 64. Whereas disability probabilities, even at higher ages, are
currently quite low because of the 2006 disability reform (see Van Sonsbeek and Gradus,
2006), unemployment probabilities rise up to 5% per year for 64 year olds in 2008, and that
year was still barely affected by the economic crisis. Table 4.2 summarizes the option value
parameters used in SADNAP.
Table 4.2 Option value parameters in the SADNAP model

Parameter Mean value Distribution
k (leisure preference) 2.0 U (1 , 3)
ρ (time preference) 0.17 0
U(0 , 0.05)
U(0.05 , 0.1)
U(0.1 , 0.2)
U(0.2 , 1.0)
γ (risk aversion) 0.7 U(0.5 , 0.9)
τ (expected wage decrease after age 60) 0.045 U(0 , 0.09)
Furthermore, mortality before age 70 is considered to be related to ill health at age 65, so
individuals who die before the age of 70 are not likely to retire past the age of 65. This as-
sumption was also made in the 2008 government proposal to introduce a retirement window
between the ages of 65 and 70, which still has to be discussed in parliament. This proposal,
which is designed in an actuarially neutral way, will still incur costs because of adverse
selection. People with a higher life expectancy are more likely to opt for delaying the state
pension. By excluding the people who died before the age of 70 from delaying their pension,
average life expectancy of those who did opt for delaying is about one year above the aver-
age, which is in line with findings on adverse selection in the German retirement system by
Kühntopf and Tivig (2008).
188 chapter 4
4.5 Model results
This section gives the results of the baseline scenario of unchanged pension policies. Section
4.5.1 focuses on the demographic and budgetary results. These results are up-to-date projec-
tions, using the demographic and budgetary modules of SADNAP that are already in use.
Sections 4.5.2 and 4.5.3 focus on the redistribution within the state pension system and the
retirement decisions of older workers. These results come from the new SADNAP modules
described in this chapter. Section 4.5.4 compares the SADNAP results with results from other
comparable models.
4.5.1 Budgetary results
The population of the Netherlands is not predicted to grow much in the future, but its com-
position will change significantly. The number of people aged 65 and over increases from
2.5 million in 2009 to 4.5 million in 2040. The number of people aged 20–64 decreases from
10.1 million in 2009 to 9.2 million in 2040. Therefore the so-called grey pressure (the num-
ber of persons aged 65 years and older as a percentage of the number of people aged 20–64
years) doubles from 25% in 2009 to 49% in 2040.
When pensions stabilize at the current level in real terms, the state pension costs will rise
from € 27.7 billion in 2009 to € 50.3 billion in 2040. In terms of GDP, assuming that GDP
also stabilizes at the current (2009) level, the state pension costs will rise from 4.8% in 2009
to 8.8% in 2040. The rise is huge, but still somewhat less than expected if constant pension
costs per pensioner were assumed. In that case state pension costs would rise to € 51.9 billion
in 2040 or 9.1% of current GDP. Apparently, the cost per person will decrease. This mitigates
the increasing pressure on the system from the newest population projection (CBS, 2009)
which predicts increasing longevity. If the former projection (CBS, 2007) had been used
instead, state pension costs would have risen to € 47.7 billion or 8.3% of GDP in 2040, 0.5%
less than the forecast based on the newest projection. Figure 4.3 gives the current SADNAP
projection in percentage of GDP decomposed in state pensions and partner allowances. It
shows how the state pension costs after the ageing peak around 2040 stabilize on about 8% of
GDP in the long run.
Figure 4.3 Projected state pension cost as % of GDP (including partner allowances),
2007–2080
In reality, of course pensions will increase in real terms, as GDP does. Van Ewijk et al. (2006)
assume for the oncoming decades state pensions to increase by 1.7% a year in real terms and
GDP to grow by 1.4% a year in real terms. If that assumption holds true, in terms of percent-
age of GDP, the state pension costs will rise from 4.8% in 2009 to 9.6% in 2040 as GDP
grows more slowly than the state pensions in real terms.
There are several reasons for the lower than expected rise of the state pension costs. From
the simulation results, it appears that not only does the size of the population of pensioners
change but that its composition changes as well. In particular, three trends are important.
190 chapter 4
First, when studying the composition of the pensioner population by origin, it appears that the
share of immigrants is rising. This holds true especially for the first-generation immigrants.
Their share in the population aged 65 and over almost doubles from 8.7% in 2009 to 15.6%
in 2040. Most first- generation immigrants have an incomplete state pension, unless they im-
migrated to the Netherlands before age 15. Emigrants will also have a reduced state pension,
and their number is growing as well. The number of reduced pensions is therefore rising,
from 15.0% in 2009 to 27.5% in 2040.
The second is the development of the share of cohabitants in the pensioner population.
This result is less clear-cut. In the short term the share of cohabitants among pensioners is
increasing. This reflects the trend seen in recent state pension realizations and is caused by
the increasing life expectancy. Partners live together for a longer time after reaching the age
of 65. For the same reason, in a recent study by Poos et al. (2008), a decrease in health care
costs is predicted for the forthcoming decades. However, after 2020 the percentage of singles
starts increasing again. This can be explained by the socio-economic trend that fewer people
become cohabitant or married. This trend finally overshadows the current trend of increasing
numbers of cohabitants because of the rising life expectancy. Already in 2040, the share of
single persons among pensioners is above the current level. After 2040 the share of singles
will stabilize at a level above the current and put additional pressure on state pension costs.
A third important trend is the rising labour participation over time, especially among women.
This influences the number of people qualifying for the partner allowance. These allowances
currently account for € 1.4 billion. A person qualifies for the partner allowance when he or
she turns 65 and has a partner who is younger than 65 and does not earn enough income of
their own49. Most of those who qualify for the partner allowance are men. Men tend to have a
wife who is on average three years younger, and labour participation among older women is
still particularly low. In fact, the majority of men turning 65 currently qualify for the partner
allowance. However, as the labour participation among women is rising, this number will be
decreasing in the future. Therefore, the costs of the partner allowances will grow slowly until
49 When the partner earns an income below 15% of the minimum wage, a full partner allowance of up to 50% of the mini-
mum wage is given. When the partner earns an income between 15% and 97.5% of the minimum wage (SVB, 2008),
a reduced partner allowance is given. When the partner earns more than 97.5% of the minimum wage, no allowance is
granted.
2013, then stabilize more or less on the same level and decrease slowly after 2035. In the
meantime the share of women in the age category 60–64 who participate in the labour market
will have doubled. In 2040 the costs of the partner allowance will be almost equal to 2009.
Figure 4.4 shows all three trends. In sum, the cost per person accounts for a reduction of state
pension cost equivalent to 0.3% of GDP. The rising share of reduced state pensions, mainly
because of the rising share of first-generation immigrants and the rising labour participation
of women, each account for 0.2%. The development in the share of single persons in the
population of pensioners has a small upward effect of 0.1% of GDP in 2040.
Figure 4.4 Changes in composition of pensioner population, 2006–2050
4.5.2 Redistribution
Redistribution within the scheme is investigated in detail by computing the share of lifetime
state pension income taken by different subgroups. The lifetime state pension income is
computed by accumulating incomes from the year a person turns 65 until the year a person
192 chapter 4
dies. For this analysis, the 2006–2045 pensioner cohorts (the 1941–1980 birth cohorts) are
aggregated. The average lifetime state pension income per person is around € 190,000, with
lifetime income per person decreasing for the later cohorts because of the rising number of
people with incomplete state pension entitlements. Table 4.3 shows a subdivision of the ac-
cumulated cohorts by subgroup, with the share of each subgroup in the cohorts of pensioners,
its share of lifetime state pension income and the ratio between the two.
Table 4.3 Share of lifetime state pension income compared with share of state pension
cohorts
Subgroup Share of cohorts Share of lifetime Ratio
turning 65 pension costs
By income
1st quintile 19.4% 15.4% 0.79
2nd quintile 19.8% 18.5% 0.93
3rd quintile 20.0% 19.4% 0.97
4th quintile 20.3% 21.8% 1.08
5th quintile 20.5% 24.9% 1.21
By gender
Women 49.4% 52.6% 1.06
Men 50.6% 47.4% 0.94
By household status
Singles 30.6% 34.0% 1.11
Cohabitants 69.4% 66.0% 0.95
By origin
Natives 73.5% 82.4% 1.12
Immigrants 26.5% 17.6% 0.66
The higher income quintiles receive an above-average share of total state pension because of
differences in life expectancy. This redistribution through life expectancy is substantial. The
first income quintile receives more than a third less than the fifth income quintile (a ratio of
0.79 vs. a ratio of 1.21). This is mainly due to the difference in life expectancy, but also to
the larger share of incomplete state pensions in the lower income quintiles. Women receive
6% more state pension from the scheme than their share in the cohort would justify. Singles
receive 11% more state pension from the scheme than their share in the cohort would justify.
This is because the lower life expectancy of singles is overcompensated by their higher state
pension. Immigrants receive 34% less state pension from the scheme than their share in the
cohort would justify. However, this large difference is primarily due to immigrants build-
ing up less entitlement during their life and only for a smaller part due to differences in life
expectancy.
4.5.3 Retirement decision
The participation transitions after age 60 in SADNAP are modelled through the behavioural
option value model described in section 4.4. The participation rates at age 60 are given by
the participation status model from appendix 4.A.3 and are similar to the participation rates
for people aged 60 as projected by CPB. The retirement decision is determined by the option
value model only for those still working at age 60. This excludes about 40% of the cohorts as
even in the long run some 30% of men aged 60 and 50% of women aged 60 will be in receipt
of benefits or not participating in the labour market at all.
When the distribution of individual retirement ages is studied, spikes are found at certain
pivotal ages. This is a well known phenomenon (e.g. Lumsdaine, Stock and Wise, 1995, and
Gustman and Steinmeier, 2005) which can be partly explained by retirement taking place
according to social-cultural norms, but also partly by economic reasons. As the models, like
the option value approach used, only take the latter into account, they usually underestimate
the spikes. For the Netherlands, Nelissen (2002) finds a strong preference for individuals to
retire either at the first or the last possible retirement age. In the Netherlands, the first possible
retirement age used to be 60 years in many sectors. Since the late 1980s for most employees
a generous early retirement scheme existed that guaranteed an income level of 70–80% of the
194 chapter 4
final wage without loss of pension accruals from 60 years of age onwards. As a result, most
people did indeed retire at age 60 (Euwals, de Mooij and van Vuuren, 2009). Gradually, the
generous early retirement schemes are being replaced by actuarially neutral schemes until,
from 2015 onwards, all schemes are fully actuarially neutral (Bovenberg and Gradus, 2008).
The last possible retirement age in the Netherlands for most employees is still 65. At that age,
the state pension starts being paid and most employees automatically have their employment
terminated. However, the Dutch government has sent a proposal to parliament to abolish the
automatic process of employees being fired at 65 and to allow delaying the state pension to
70 years of age instead of the current 65.
When the retirement decisions in SADNAP are evaluated, indeed, when the generous early
retirement scheme is in place, the majority of retirement decisions take place at 60, the earli-
est possible age. In a fully actuarially neutral scheme (assuming a last possible retirement
age of 70), the model predicts two spikes in retirement, a large one at 65 and a smaller one at
the last possible retirement age. The lines in Figure 4.5 show these retirement patterns. The
dashed line gives the retirement pattern of the birth cohorts 1946–1950 (those that turn 60
between 2006 and 2010) when the generous early retirement scheme would still have been in
place (assuming 80% of the working population to be covered by this generous ERS provid-
ing an income level of 75% of the wage at age 60). The solid line gives the retirement pattern
of the same cohort in an actuarially neutral early retirement scheme. The average retirement
age rises by 2.5 years for those who are working at least until 60 in the actuarially neutral
scenario. The bold line gives the retirement pattern of the birth cohorts 1966–1970 (those
who turn 60 between 2026 and 2030). The share of non-participants at age 59 decreases from
54% for the 1946–1950 cohorts to 39% for the 1966–1970 cohorts. The average retirement
age for the entire population rises by one year (from 61.9 to 62.8) because of the higher num-
ber of people working at least until 60.
Figure 4.5 Projected rate of non-participation in the labour force by age
The average retirement age increases with 2.5 years for the population still participating at
age 60. Results in the same order of magnitude are found by Kapteyn and de Vos (2004), who
simulated the effect of a change from the generous ERS that existed in the Netherlands at
the time to a more or less actuarially neutral scheme. They forecasted an increase in average
retirement age of four years for males and insignificant changes for women with the option
value model, which in the same study they found to perform better than the peak value model
in the baseline estimation. In 2007, retirement age had indeed increased by two years to 61.7
years from below 60 during the 1990s when the generous early retirement schemes were com-
mon (Advies Commissie Arbeidsparticipatie, 2008). However, even when the generous early
retirement schemes were common, a fair share of workers continued working until 65 or later.
This concerns mainly the self-employed and also employees not covered by collective agree-
ments on early retirement. On the other hand, 40% of the population is no longer participating
in the workforce at age 60, which still leaves important participation gains to be made.
The model predicts 26% of the people working at 60 to retire before 65, 38% to retire at 65
196 chapter 4
and the other 36% to retire past 65. Those retiring early are the ones with high time prefer-
ence, high leisure preference or high expected wage decrease, or they are risk-averse, or a
combination of the above. The influence of time preference and leisure preference seems to
be dominant. Also, disability is an important factor causing early retirement for about one in
six retirees who retire early. As the disability scheme in the Netherlands currently is so strict
that abuse of the scheme as a means of early retirement is virtually impossible, the unemploy-
ment scheme is nowadays often used as an early retirement pathway at all ages. Table 4.4
gives a characterization of the retirees per retirement age:
Table 4.4 Characteristics of retirees by retirement age (birth cohorts 1946–1970)

Retirement Share of Time Leisure Risk Wage Share of
age population preference preference aversion decrease disability
retiring
≤ 59 0.427
60 0.029 0.26 2.15 0.70 0.046 0.14
61 0.025 0.21 2.07 0.70 0.046 0.16
62 0.027 0.20 2.04 0.70 0.046 0.17
63 0.035 0.21 2.04 0.70 0.047 0.14
64 0.036 0.21 2.07 0.70 0.048 0.18
65 0.217 0.20 2.05 0.69 0.046 0.03
66 0.043 0.14 1.97 0.70 0.046 0.14
67 0.042 0.13 1.97 0.71 0.046 0.15
68 0.040 0.12 1.94 0.71 0.045 0.14
69 0.080 0.08 1.81 0.73 0.039 0.06
The SADNAP model rightly predicts a strong preference for retiring at 65, the year the state
pension (and partner allowance) start being paid. However, the number of people working on
past 65 is slightly higher than the levels currently seen, especially for women. The aboli-
tion of the automatic termination of employees at 65, which is expected soon, will probably
influence current retirement patterns. Moreover, it is known (Coile, 2004) that husbands’ and
wives’ retirement behaviour is influenced not only by their own financial incentives but also
by spill-over effects from their spouses’ incentives, which may explain why women’s retire-
ment age is overestimated by the option value algorithm. The SADNAP estimates may give
a good estimate of the retirement patterns that will be realized when the automatic firing of
employees at 65 has been abolished and when all early retirement schemes that are rewarding
early retirement are abolished. However, it remains to be seen whether such a substantial part
of the whole population of males and females will retire at the last possible retirement age
when that last possible retirement age is increased to 70.
4.5.4 Validation and comparison to other models
The demographic model results are benchmarked with the official population forecast of
the CBS. The SADNAP estimates in all years stay within a margin of 1% of the comparable
CBS estimates for main age groups. There is no exact match with macro population numbers
as only the yearly number of births and immigrants and mortality and emigration rates are
aligned to CBS forecasts. The grey pressure, a key indicator, equals 49% in 2040 in both
SADNAP and the CBS projection.
The financial forecast compares well with the earlier macro calculations of Van Ewijk et al.
(2006), who forecasted a rise in state pension cost from 4.7% of GDP in 2006 to 8.8% of
GDP in 2040, based on the 2004 population forecast of CBS. As in the 2008 population fore-
cast the number of people aged 65 and over in 2040 is 12% higher than in the 2004 forecast,
an update of the calculations of van Ewijk et al. based on the newest population projection
would lead to an estimate of 9.9% of GDP in 2040. The difference with the 9.6% estimated
by the SADNAP model is caused by the decreasing cost per person that was not taken into
account in the macro approach.
The short-term forecast of number of state pensioners and state pension costs gives compa-
rable results for 2009. The number of pensioners is 0.6% lower than the 2009 estimate of
SVB, the pension administration office. In later years the SADNAP estimates point to slightly
198 chapter 4
higher state pension costs and a slightly higher number of state pensioners. In 2024, the last
year of the SVB forecast, the number of pensioners is 0.2% higher than the SVB estimate.
Recent studies report mixed findings on whether people actually reach their target of a 70%
replacement rate (total pension as a percentage of the final wage). Statistics Netherlands
(CBS, 2008) finds that a gross income level of on average 73% of the final wage is to be
expected for the cohorts currently saving for their pension. The SADNAP results show the
same average replacement rate. Moreover, in a microsimulation study on the wealth of Dutch
pensioners (SZW, 2006), the income position of pensioners is found to improve substantially
as a consequence of more second- and third- pillar pension savings by younger generations.
The researchers show the average net income of the 65–69 year olds to rise to 102% of the
income of the 60–64 year olds in 2030 as compared with 92% in 2010. Knoef, Alessie and
Kalwij (2009) also report that between 2008 and 2020 mean equivalised household income
increases by 11–14% for the retirees. However, Euwals, de Mooij and van Vuuren (2009) find
that when the consequences of the career average system to which most pension funds have
recently switched are taken into account, the younger generations have lower replacement
rates, up to less than 50% for the birth cohorts 1972–1976 (weighted average of all cohorts
64%). The SADNAP results show rising replacement rates and are in line with the estimates
of SZW and Knoef, Alessie and Kalwij.
Few comparable studies are available on redistribution. Knoef, Alessie and Kalwij (2009)
report a Gini coefficient of 0.23 for the population of retirees, based on equivalised household
income. This is well below the Gini coefficient of 0.29 as reported by SADNAP, which is
based on individual income. However, a lower Gini coefficient when comparing household
income is plausible because higher income earners (mainly males) usually have younger part-
ners, so their income will generally be shared with a partner, decreasing the level of inequal-
ity measured.
The average retirement age increases by 2.5 years as compared with the benchmark scenario
of a generous early retirement system. Results in the same order of magnitude are reported
by de Vos and Kapteyn (2004), as described in section 4.5.3, and Euwals de Mooij and van
Vuuren (2009). Table 4.5 gives an overview of some key SADNAP results and some compa-
rable estimates.
Table 4.5 Overview of key SADNAP results compared with other models
Indicator Year SADNAP Comparable By
estimate estimate
Grey pressure 2040 49% 49% CBS 2008 population projection
State pension expenses 2040 9.6% 8.8% CPB based on 2004 population projection, would
be 9.9% based on 2008 population projection
Number of state pensions 2024 4.132 mln 4.125 mln SVB 2009-2024 projection
Expected replacement All 0.73 0.73 CBS
rate when retiring 0.64 Euwals, de Mooij, van Vuuren (2009)
Gini coefficient 2040 0.29 0.23 Knoef, Alessie and Kalwij (2009) on household
income (SADNAP estimate on individual income)
Retirement age 2007 61.9 61.7 Adviescommissie Arbeidsparticipatie (2008)
Retirement age increase 2015 2.5 4/0 Kapteyn and de Vos for men / women
(compared to generous 2 – 2.5 Euwals, de Mooij, van Vuuren (2009)
ERS)
Like most other OECD countries, the Netherlands is facing an ageing population, causing a
burden on the public finances. A significant part of the rise in public expenses will be caused
by the rise in costs for the state pensions. The old-age dependency ratio almost doubles from
now until 2040 (the peak of the ageing process in the Netherlands).
The dynamic population microsimulation model SADNAP is developed for calculating the fi-
nancial and economic implications of the ageing problem and of the relevant policy measures
considered. The model uses administrative datasets on state pension payments, state pension
200 chapter 4
entitlements and private pension entitlements. Life paths are constructed for a sample of the
Dutch population, including immigration and emigration, household formation and labour
participation. In this chapter, two extensions of the model are elaborated. First, because of
improvement of the income data sources and differentiation of mortality rates, redistribu-
tion within the pensions system can be analysed. Second, the retirement decision is modelled
based on the Stock and Wise option value approach, allowing for individual variation in the
main option value parameters based on literature review in order to make use of the added
value of microsimulation.
It is shown that in the baseline scenario the state pension costs rise less sharply than the
number of pensioners. The composition of the pensioner population is changing. The number
of immigrants with reduced state pensions is rising. During the forthcoming decades, the pro-
portion of single persons among the pensioner population is decreasing. However, this trend
will be reversed in the future. Also the rising labour participation of women decreases the
cost of partner allowances. These downward influences amount to 0.3% of GDP in 2040. This
partly compensates for the increasing longevity from the latest population forecast.
The intra-generational redistribution within the Dutch pension scheme is shown to be

substantial. The bottom income quintile receives less out of the system, a discount of more
than one-third, than the top income quintile, mainly because of lower life expectancy. Single
persons, however, get more out of the system than partners in a couple. Their higher pension
compensates for their shorter life expectancy.
The modelling of the retirement decision through the option value model confirms the retire-
ment patterns experienced in the Netherlands when a very generous early retirement scheme
was still in place. Average retirement age for those who are still working at 60 can rise by
2.5 years when the early retirement schemes have become fully actuarially neutral. In the
actuarially neutral scheme, a strong preference for either retiring at 65 or at the last retirement
age is suggested. The time preference and leisure preference parameters appear to be the most
important drivers for the retirement decision. An assumption of wage decrease is added to the
model in order to obtain more plausible results and less people working on until the last pos-
sible retirement age. Another important factor is the role of disability and unemployment. The
unemployment scheme especially is still used as an early retirement pathway.
Future research will focus on evaluating policy options with the model. New datasets on state
pensions and private pensions will become available for more recent years and more informa-
tion on third-pillar pensions and financial wealth will become available on the micro level.
Also, the availability of linkable datasets of different years will allow for an estimation of the
option value parameters and their distribution.
202 chapter 4
Appendix 4.A The SADNAP Model
This appendix gives a more detailed description of the SADNAP model. The model is pro-
grammed in SAS and consists of different modules on demographics, household formation,
labour participation and the retirement decision, which are described subsequently in sections
4.A.1 to 4.A.3.
4.A.1 The demographic model
Before the simulation starts, a base data file is created out of the source files described in
section 4.3. Three different data sources are combined into a single file containing a repre-
sentative sample of the Dutch population in the base year (2006). Aggregate CBS data on the
population aged 0–15 years are used in the base year. For the population aged 15–64 years,
the micro datasets from CBS on state pension entitlements and private pension entitlements
as described in section 4.3.2 are used. Finally the micro dataset from SVB on state pension
payments as described in section 4.3.1 is used for the population aged 65 and over. These
three datasets complete the population for the base year. The records for people younger than
15 years are constructed by using general demographic CBS statistics on the age, gender and
ethnic composition of the Dutch population. As state pension entitlements are zero until per-
sons turn 15, and children do not have wages or private pension entitlements, no additional
information on this group is needed.
The CBS file with entitlement data for people aged 15–64 has another year of origin (namely
2005) than the SVB file with payment data for people aged over 65 (namely 2006). There-
fore, the 2005 population aged 15–64 is simulated towards a population aged 16–65 in 2006.
A population can change by four demographic events; births, deaths, immigration, and emi-
gration. Births do not affect the population aged 15–64 in one year of simulation. Therefore,
the 2005 population is made subject to mortality, immigration, and emigration. After this first
simulation the 2005 population is aged 16–65. Because richer data are available for people
aged over 65 in 2006, persons aged 65 are deleted from the simulation. All age groups are
added together to complete the base dataset.
For each simulation year, records for the new births and the new immigrants are added, based
on the macro data sources as described in section 4.3.3. When applying the mortality rates
and emigration rates as derived from the CBS population projection, an accurate population
forecast results. Stocks, flows and rates from the CBS are available until 2050. After 2050,
the 2050 numbers are kept constant.
Although there is now an accurate population forecast, there is still a problem left relating to
the complexity of the modelling of immigration and emigration. Also “remigration” needs to
be modelled. This is important because on average during the last few years about 50% of the
emigrants are former immigrants and about 20% of the immigrants are former emigrants. By
not allowing for remigration, the share of immigrants in the population will clearly be overes-
timated. The former (immigrants having a greater emigration rate) can easily be implemented
by using different sets of emigration rates depending on whether the emigrant was born in
the Netherlands or not. The latter is more difficult and few of the larger simulation models,
notably Lifepaths and Sesim (Pennec and Keegan, 2007), allow for emigrants re-entering the
population. SADNAP ignores repatriation of emigrants, resulting at the micro level in a slight
overrepresentation of small entitlements as returning emigrants continue building up their
already existing rights, whereas SADNAP assumes these rights to consist of separate parts
belonging to two separate persons.
Moreover, still missing is a part of the population that will be entitled to a state pension in
the future, but is not living in the Netherlands anymore. As can be seen from Table 4.1, 9%
of the current population of pensioners is living abroad. Since emigration is modelled, the
204 chapter 4
model captures all future pensioners who live in the Netherlands in the base year, but will
emigrate in the future. However, we still miss the people aged between 15 and 64 in the base
year who built up state pension entitlements in the Netherlands in the past but emigrated
before the base year. To correct for this, records are added for former emigrants. As a starting
point, the youngest cohort of pensioners in the base year is used. Of this cohort also 9% of
the pensioners are living abroad. From the state pension entitlement, their year of emigration
can be estimated. Everybody missing one year of entitlement is assumed to have emigrated at
age 64, everybody missing two years at age 63 and so on. As in the simulation, people aged
64 in the base year can emigrate in the first year of the simulation, after that first year of the
simulation only the claimants living abroad who emigrated at age 63 or younger have to be
added. According to SVB (2008), non-take-up among people living abroad is common, hence
a correction is made, based on the assumption that the younger one emigrated, the less likely
one is to claim a Dutch state pension.
The whole process described above is represented in Figure 4.6. The filled boxes represent
the micro databases from the base years that are used in the simulation and the blank boxes
represent the micro data that are constructed from macro data sources in order to add new
cohorts to the base year data.
Figure 4.6 The simulation process of the population forecast

2005 2006 2007 ---- 2055 ---- 2080
State pension
payments (65+)
State pension
entitlements
(15-64)
Private pension
entitlements
(15-64)
Population
0-15
Immigration
cohort 2006
Immigration
cohort 2080
Birth cohort
2007
Birth cohort
2080
Emigrant
cohort 2007
Emigrant
cohort 2055
Figure 4.7 shows the composition of the simulation database, based on a 1% sample and
extrapolated to the whole population. The numbers add up to more than the population of
the Netherlands as the life paths of immigrants are taken into account before they immigrate
to the Netherlands and the life paths of emigrants are taken into account after they leave
the Netherlands. In 2080, the final year of the simulation, the base year micro data sets will
almost completely have phased out and will have been replaced by persons from the con-
structed datasets from 2006 onwards.
206 chapter 4
Figure 4.7 Composition of the simulated population, 2006–2080
4.A.2 The household formation model
In the next steps variables are added to the demographic model, such as household type.
From the databases of pension entitlements and pension payments, the household status of all
individuals aged 15 and over is known. SADNAP distinguishes between single persons and
cohabitants only50. The aggregated state pension for two singles is higher than the aggregated
state pension of two partners in a couple.
Age- and gender-dependent transition probabilities are used to determine whether single
persons remain single or start cohabiting and whether cohabitants separate and become single
or stay together. The transition probabilities can be derived from the age- and gender-specific
household forecast from CBS as described in section 4.3.3.
50 In the Netherlands, the state pension system treats (formal) cohabitation in the same way as being married.
When PS denotes the probability of being single and PC the probability of cohabiting, the
transition probabilities PSC (probability of a single person cohabiting the next year) and PCS
(probability of a cohabitant being single again the next year) can be defined as follows:
(4.3)
(4.4)
(4.5)
(4.6)
If the correction terms ε(age,gender) are set to zero, most individuals will have only one
lasting relationship during their lifetime. The higher the correction terms are set, the more
relationships will be started and finished each year. The correction terms can be used to align
the simulation to the information on household formation and dissolution from the CBS
household forecast. In the baseline scenario, the terms are set to zero.
However, by introducing differences in mortality rates (see section 4.3), a deviation is intro-
duced from the original population projection in the numbers of singles and cohabitants by
age. As the household formation model from appendix 4.A.2 is based on the original popula-
tion projection, the numbers need to be realigned in order to match the original population
projection again. Concretely, the equations (4.3) and (4.5) need to be adapted as the prob-
abilities of singles cohabiting need to decrease and the probabilities of cohabitants becoming
single again need to increase.
208 chapter 4
(4.7)
(4.8)
In some larger microsimulation models, the cohabiting process is very elaborate. Those mod-
els contain a formalized mate-matching module in which partners are found within the model
based on certain matching criteria (for an overview of methods, see Perese (2002) and for an
overview of models see Bacon and Pennec (2007)). SADNAP follows a simple approach, in
which the important characteristics of the partner are determined as soon as those character-
istics become relevant for the model calculations. In the ageing calculations the gender, age
and participation status of the partner are the most important characteristics. The gender of
the partner is assumed always to be the opposite of the gender of the other partner. From the
dataset of state pension payments, detailed information on the age difference between part-
ners of a couple is available. The age differences from the youngest cohort of this dataset (the
1941 birth cohort) are used, assuming that the distribution of age differences in relationships
will remain the same in the future. Given the gender and age of a partner, the corresponding
participation rate can be derived from the age and gender specific participation estimates as
described in section 4.3.3.
At this point enough information is available to calculate the costs of the state pension. Infor-
mation is available on the future population size and its division by age. Starting with current
state pension entitlements, the building up of entitlements in the future can be simulated. As
information on the household type is also available, by adding benefit levels to the model, the
future state pension benefits of all individuals can be simulated. The total costs for the state
pension can be calculated by aggregating the individual benefits. All calculations within the
model are done at the current price level.
4.A.3 The participation status model
From the database of pension entitlements, the labour market status of all individuals aged
15 to 64 is also known. Participants can be either employees or self-employed. The self-
employed, for whom we have no data on private pension savings, are treated by assuming
their pension savings on average to be equal to those of the employees. Non-participants can
be studying, receiving a benefit, early retired or not participating at all. Figure 4.8 shows the
distribution of the 2005 Dutch population by age and by participation category.
Figure 4.8 Composition of the 2005 population by age and participation category
One of the most striking conclusions from the above graph is that in the last couple of years
before the statutory retirement age of 65, only a small minority of the population is still work-
ing. This is mainly due to the popularity of early retirement schemes and the use of benefits,
especially disability and unemployment, as an early exit route (see e.g. de Vos and Kapteyn,
2004). For example, of the 64 year olds, only 11% are working, whereas 27% are on benefit
and 39% are early retired. However, the participation rate among the 60–64 year olds is cur-
210 chapter 4
rently rising due to policy changes, especially regarding early retirement schemes and disabil-
ity insurance (Euwals, de Mooij and van Vuuren, 2009).
In SADNAP, age and gender dependent transition probabilities are used to determine whether
participants continue participating or become non-participants, and vice versa. The transi-
tion probabilities can be derived from the age- and gender-specific forecast of participation
rates from CPB as described in section 4.3.3. Participation in SADNAP is a binary state.
Participants include employees, the self-employed and the involuntarily unemployed who are
actively seeking work. Non-participants are not available to the labour market and include
students, disability benefit recipients, the early retired and the voluntarily unemployed.
When PN denotes the probability of being non-participating and PP the probability of partici-
pating, the transition probabilities PNP (probability of a non-participant participating the next
year) and PPN (probability of a participant not participating the next year) can be defined as
follows:
(4.9)
(4.10)
(4.11)
(4.12)
Again, if the correction terms ε(age,gender) are set to zero, most individuals will have only
one uninterrupted period of participation during their lifetime. The higher the correction
terms are set, the more short periods of participation and non-participation will occur during
a life course. In the baseline scenario, the terms are set to zero. The transition probabilities
are used only until people turn 60. From that age on, the retirement decision (for those who
are still participating at age 60) is modelled through the behavioural option value model as
described in section 4.4.
212 chapter 4
Appendix 4.B Data sources
4.B.1 State pension data
From the state pension perspective the Dutch population can be separated into three groups;
people aged younger than 15, people aged between 15 and 64, and people older than 65.
People younger than 15 are not yet building up state pension rights. People between 15 and
64 are building up those rights when living in the Netherlands, and people aged 65 plus are
receiving a state pension if they have built up entitlements in the past. For the latter two
groups micro data files are available.
Two institutions provide the micro data sources. Statistics Netherlands (CBS) supplies a
micro data file concerning state pension entitlements. This file contains all of the over 11
million persons aged 15–64 who live in the Netherlands. Variables in this dataset include
birth date, nation of origin, gender, household status, and the number of entitled years for the
AOW. Micro datasets are available for 2004 and 2005. The Social Insurance Bank (SVB) is
the authority that accomplishes the payment of, among others, state pensions. The Ministry of
Social Affairs and Employment receives detailed data from the SVB on the current state pen-
sions, including a full administrative data file with information on all 2.6 million persons re-
ceiving AOW in 2006. The SVB data file contains information on gender, birth date, country
of residence, marital status, birth date of the partner, number of entitled years for the AOW
and entitlement to a partner allowance. Moreover, this file contains information on people
receiving AOW abroad whereas the CBS only delivers AOW information on inhabitants of
the Netherlands. Some characteristics of the two main data sources can be found in Table 4.6.
Table 4.6 Some characteristics of the two main data sources on state pensions
SVB file 2006 (65+) CBS entitlement file 2005 (15-64)
Number % Number %
Gender
Male 1,130,506 43.5 5,563,649 50.5
Female 1,469,568 56.5 5,458,248 49.5
Origin
Native51 9,656,948 87.6
Immigrant (1st gen) 1,364,949 12.4
Household status
Single 2,667,713 24.2
Cohabiting 8,337,017 75.8
State pension entitlement
Complete 2,200,342 84.6 9,443,153 85.7
Reduced 399,802 15.4 1,572,768 14.3
Country of residence
Netherlands 2,364,891 91.0
Abroad 235,183 9.0
Total 2,600,074 100.0 11,021,897 100.0
Because of their higher life expectancy, women outnumber men by 56.5% to 43.5% among
the population aged 65 and over. Among the population of 15–64 year olds, however, men
outnumber women by 50.5% to 49.5%. Over 12% of the population aged 15–64 are first-gen-
eration immigrants (not born in the Netherlands). Among this group, reduced AOW rights are
common because the majority of the group migrated to the Netherlands after the age of 15.
Almost 15% of the population aged 15–64 already has a reduced right. This is about the same
as the share of reduced rights among the current population of pensioners, but this percent-
age will rise in the future because, before reaching the age of 65, part of the population aged
15–64 who at the moment have a complete entitlement can lose some years of accrued rights
when they go abroad.
51 Including non-natives from the 2nd generation (born in the Netherlands)

214 chapter 4
4.B.2 Company pension data
Recently, Statistics Netherlands has also started providing micro data on (second-pillar)
company pensions. In 2009, a micro dataset became available based on data of a representa-
tive sample of pension funds including the large pension funds for civil servants and health
workers. This file includes individual data on company pension entitlements of 53% of the
population aged 15–64 and 67% of the Dutch employees. Some characteristics of the data
source on private pension entitlements can be found in Table 4.7.
Table 4.7 Some characteristics of the main data source on company pensions
State pension entitlements Private pension entitlements
Number % Number %
Gender
Male 5,563,649 50.5 3,151,280 54.2
Female 5,458,248 49.5 2,664,918 45.8
Origin
Native 9,656,948 87.6 5,288,526 90.9
Immigrant (1st gen) 1,364,949 12.4 527,672 9.1
Household status
Single 2,667,713 24.2 1,325,328 22.8
Cohabiting 8,337,017 75.8 4,484,873 77.2
Complete 9,443,153 85.7 5,074,905 87.3
Reduced 1,572,768 14.3 739,206 12.7
Participation status
Study 1,207,782 11.0 128,824 2.2
Work 6,916,620 62.9 4,474,776 76.9
Benefit 1,320,852 12.0 644,887 11.1
None 1,217,069 11.1 318,848 5.5
(Early) retired 342,141 3.1 246,483 4.2
Total 11,021,897 100.0 5,816,198 100.0
As can be seen from comparing the figures for state pensions and private pensions, the
distributions of the main characteristics differ somewhat between the full set of people aged
15–64 and the subset of those whose private pension entitlements are also surveyed. This
is caused by some people having no second-pillar pension entitlements at all, like students,
self-employed, people on social assistance and people who have never worked or who work
in low-wage jobs without building up entitlements.
Besides the common characteristics like gender, birth date, ethnicity and household sta-
tus, the data set contains a projection of the expected private pension if a person’s career is
continued with the same wage as in 2005. This means that the older a person is, the more ac-
curate the forecast will be, especially since most pension funds in the Netherlands changed to
a career-average pension instead of a final-salary pension around 2004. The average wages,
expected pensions and median replacement rates (expected state pension + expected private
pension/wage) by subgroup are presented in Table 4.8. They concern the almost 4.2 million
employees and exclude the self-employed, who usually build up their own pension savings in
the third pillar.
216 chapter 4
Table 4.8 Expected private pension by main subgroups, 2005

Wage (mean) Expected private pension Expected replacement rate
(mean) (median)
Age-category
15-19 € 12,546 € 3,740 1.283
20-24 € 20,605 € 7,005 0.857
25-29 € 25,580 € 9,128 0.779
30-34 € 28,623 € 10,445 0.754
35-39 € 30,294 € 11,323 0.747
40-44 € 31,855 € 12,194 0.729
45-49 € 33,582 € 13,020 0.702
50-54 € 34,911 € 13,668 0.675
55-59 € 36,016 € 12,460 0.638
60-64 € 36,594 € 10,335 0.575
Gender
Male € 37,892 € 13,965 0.644
Female € 21,984 € 8,275 0.839
Origin
Native € 31,008 € 11,714 0.743
Immigrant € 27,282 € 7,791 0.607
Household status
Single € 28,989 € 9,952 0.761
Cohabiting € 31,147 € 11,767 0.725
Complete € 30,536 € 11,628 0.751
Reduced € 32,062 € 9,484 0.576
All € 30,703 € 11,394 0.732
When interpreting the data, familiar patterns can be seen, like wages and pension entitlements
rising with age, and men, natives and cohabitants having considerably higher wages and pen-
sions than women, immigrants and single persons respectively. On average, employees have
expected second-pillar pension savings of € 11,394, which is more than the expected state
pension entitlements52. The data confirm the growing importance of company pensions as an
income component of the elderly.
52 In 2005, the maximum state pension for a single person was € 11,211 and for a partner in a couple € 8,008.
The expected private pension peaks in the age category 50–54. Past that age, decreases are
most probably due to people with high pension savings tending to retire earlier, although it
is also known that the younger cohorts tend to have higher second-pillar pension savings
(SZW, 2006). Also, there is a correlation between first- and second-pillar pensions. People
with high company pensions more often have a complete state pension, whereas people with
incomplete state pension entitlements more often have smaller or no company pensions.
Rather surprising, though, is that people with reduced state pension entitlements tend to have
higher wages than people with complete entitlements. However, this only holds true for the
population of employees, which indicates that people who have lived abroad for a while tend
to have higher than average wages. For the general population, immigrants with reduced state
pension entitlements tend to be on benefits more often.
The median replacement rate53 is 0.73, but there are sizeable differences between subgroups,
with women, natives and singles having higher replacement rates than men, immigrants and
cohabitants respectively. The high replacement rates of women are due to the large share of
women in the Netherlands who during their life course change to part-time jobs, whereas
pension savings up till then may have been based on full-time jobs. The higher replace-
ment rates of singles are due to the higher first-pillar state pension for which they qualify.
The replacement rates tend to decrease by age. This is partly due to younger cohorts having
higher second-pillar pension savings, although it should be kept in mind that in a career-
average system, replacement rates will go down with age, as wages tend to grow with age
and replacement rate is related to the final wage. Also the very high replacement rates among
the youngest age groups need explanation. In the Netherlands a full minimum wage can only
be earned at 23 years of age. Wages paid to workers below that age tend to be lower, whereas
expected private pensions take account of the wage increase due to legal requirements. But
because major career jumps have yet to take place at that age, the expected private pensions
for the younger cohorts are not reliable indicators anyway.
53 The average replacement rates are higher than the median replacement rates, but they are not a good indicator as some
people with high pension savings and very low wages (for example, because they worked for only part of the year) can
have very high replacement rates.
218 chapter 4
Statistics Netherlands is currently preparing a micro dataset on (third-pillar) individual ar-

rangements, based mainly on data files from insurance companies. This dataset is expected to
be available to researchers in 2009. These data are especially important for groups that cannot
participate in the company pension schemes, like the self-employed. Finally, Statistics Neth-
erlands is preparing a micro dataset on personal wealth, the so-called fourth pillar, from tax
office databases. This fourth pillar is known to play a role in the retirement decision (Bloe-
men, 2008) and can also be particularly important when intergenerational wealth transfers are
researched.
4.B.3 Macro data sources
In SADNAP, some macro data sources are also used. On demographic events, the most recent
population projection by age, gender and ethnicity by Statistics Netherlands (CBS, 2009) is
used. From this projection, which runs from 2008 to 2050, the number of births and the num-
ber of immigrants by age and gender per year are used. Also mortality rates and emigration
rates by age and gender are deducted from this projection. Although this official projection
is used in the baseline scenario, by including the underlying processes of emigration, im-
migration, births and deaths, the model allows for analyses of alternative scenarios in which
(policy-driven) changes, for example, in immigration or emigration levels, can be represent-
ed. In SADNAP, a narrow definition of ethnicity is used, distinguishing only between natives
and immigrants (people not born in the Netherlands), as distinguishing between second-gen-
eration non-natives and natives is not important for the state pension entitlements.
Information on household types is available for the most recent population projection. From
this projection, which runs from 2009 to 2050, the cohabiting rates by age and gender are
derived. The CBS data distinguish between single persons, couples, lone parents, children
living with their parents, people living in institutions, and a remainder group of people co-
habiting without having a relationship to each other (e.g. students). In SADNAP only single
219
persons and cohabitants are distinguished, based on the social benefits for which they qualify.
The remainder group – people living together without having a relationship – are considered
to be single persons. Of the people living in institutions, half are considered to be single and
half to be cohabiting.
On labour participation, the most recent labour participation rates as provided by CPB (Euw-
als and Folmer, 2009) are used. These are average participation rates for five-year age classes
by gender. From these five-year averages, participation rates for individual ages can be
deducted. These participation rates include all persons who are available to the labour market,
including the involuntarily unemployed who are actively seeking work. The non-participants
are mainly students, disability benefit recipients, the early retired and the voluntarily unem-
ployed (e.g. housewives). SADNAP currently lacks a model of benefit use during the life
course. However, benefit use plays an important role in determining the net effects of ageing-
related policy measures, like raising the retirement age. Therefore, in SADNAP the final state
of benefit use at the age of 64 is modelled. It is assumed that benefit use at such high ages is
an absorbing state (outflow rates are near zero). In 2005, based on the CBS dataset, 27% of
those aged 64 were in receipt of benefits: 17% from the disability insurance scheme, 2% from
the unemployment insurance scheme, 4% from social assistance and 4% from other benefits.
Because of the major reform of the DI scheme, the use of disability benefits will decline in
the future. Long-term disability rates are based on current long-term forecasts based on the
model described in Van Sonsbeek and Gradus (2006). In the meantime, the unemployment
scheme has also undergone major reforms, limiting the duration of the benefit from a maxi-
mum of 7.5 years in 2003 to a maximum of just over 3 years in 2006. However, as Euwals,
de Mooij and van Vuuren (2009) conclude that, as a result of the reforms of DI and early
retirement schemes, pressure on unemployment insurance may increase in the near future, it
seems plausible to keep long-term rates for the other benefits constant at the 2005 level. Table
4.9 presents an overview of the macro data sources used.
220
Table 4.9 Overview of macro data sources

Age Gender Ethnicity Household Participa- Income
status tion status
Birth cohorts n/a E E n/a n/a n/a
Immigrant cohorts E E E - - -
Mortality rates E E I I - I
Emigration rates E E E - - -
Cohabiting rates E E - n/a - -
Participation rates E E - - n/a -
Benefit rates E E - - E -
In Table 4.9, ‘E’ means that a characteristic is explicitly known from the data source itself,
‘I’ means that a characteristic is implicitly taken account of in the model, and ‘-‘ means that
a characteristic is not known or used. Finally, some combinations, like the age and household
status of newborns, are not applicable (n/a).
5
Microsimulations on the effects of
ageing-related policy measures54
In the Netherlands, like in most OECD-countries, the ageing of the population endangers the
sustainability of public finances. In this chapter a dynamic microsimulation model is used
for calculating the financial and economic implications of the ageing problem and the policy
measures considered. The model uses micro datasets of all Dutch pensions and pension enti-
tlements. The retirement decision is modelled by using an option value approach.
First, the chapter discusses the baseline scenario of unchanged policies. The microsimulation
results differ from previous macro CGE results. The state pension costs rise less sharply than
the number of pensioners. Also the microsimulation model is used to analyse the redistribu-
tive character of the Dutch pension system, both through differences in pension entitlements
and through differences in life expectancy, for different subgroups. The retirement decision is
analysed with an option value based behavioural model.
Secondly, the chapter discusses the effects of five policy measures aimed at reducing the state
pension costs and the sustainability gap: abolishment of the partner allowance (a measure that
is already decided about), raising the retirement age from 65 to 67 years of age, introduction
54 This chapter has been published in Economic Modelling, Vol. 27, pp. 968-979 (Van Sonsbeek, 2010).
222 chapter 5
of a flat rate state pension at the same level as the current pension for partners of a couple,
raising the taxation of wealthier pensioners by abolishing their tax exemption and introduc-
tion of a flexible retirement window with a high accrual to reward later retirement. For each
measure, the budgetary effects, labour participation effects and redistributive effects are
quantified and assessed.
5.1 Introduction
In the Netherlands, like in most OECD-countries, there is a large discussion on the costs of
the ageing population and on the policy measures needed to cope with this development.
According to the Netherlands Bureau of Economic Policy Analysis (CPB, 2007), mainly
because of the ageing of the population, the gap between government expenditures and
revenues is projected to be 2.2% of GDP in 2040. In order to keep Dutch public finances sus-
tainable, additional policy measures are necessary. Moreover, the need for reform has become
more apparent because of the current economic crisis, which has tripled the sustainability gap
to around 6% of GDP (Gradus, Beetsma and Smit, 2010).
The dynamic microsimulation model SADNAP (Social Affairs Department of the Neth-
erlands Ageing and Pensions model) is being developed for calculating the financial and
economic implications of the ageing problem and of the policy measures considered. A
microsimulation model, as compared with macro general equilibrium models, can give more
detailed information on the ageing problem and on the redistributive effects of policy options,
which can be used in the evaluation of those options. The model uses administrative datasets
of all Dutch public pension payments and of all entitlements for public pensions and a large
share of entitlements for private pensions.
In a baseline scenario, it is shown that the state pension costs rise less sharply than the num-
ber of pensioners because of several changes in the composition of the population of pension-
ers. The downward influences amount to 0.3% of GDP (Van Sonsbeek, 2011).
Microsimulations on the effects of ageing-related policy measures 223
Concerning the policy options, first the abolishment of the partner allowance is discussed.
This policy measure was already decided about in 1996, but will be in force from 2015
onwards. In the current political discussion, the major policy direction considered is raising
the retirement age. Current government in October 2009 decided to raise the retirement age
to 67. This reform is due to be discussed in parliament early 2010. This chapter considers as
a second measure raising the retirement age to 67 years of age and finds macro results on the
reduction of the state pension costs and the sustainability gap, which are in line with earlier
macro forecasts by CPB. The third measure considered is raising the taxation of the wealthier
pensioners by abolishing their tax exemption. This alternative was favoured by different left-
wing parties during the 2006 election campaign and was also proposed in an earlier advice of
the Social Economic Council. The fourth measure discussed is introduction of a flat rate state
pension at the same level as the current pension for partners of a couple. This comes down to
decreasing the pension for singles from 70% to 50% of the minimum wage. The fifth measure
considered is the introduction of a flexible retirement window between 65 and 70 years of age
with financial incentives for postponing retirement.
The structure of this chapter is as follows. Section 5.2 briefly overviews the Dutch pension
system and the earlier macro forecasts of state pension costs and ageing-related policy meas-
ures. Section 5.3 presents a brief description of the microsimulation model and the data it
uses. In section 5.4, a baseline scenario is presented on the development of the state pension
system. Besides the budgetary implications also the redistributive character of the system
and the retirement incentives are investigated. Section 5.5 then describes the model results
on budgetary, redistributive and labour participation effects of the policy alternatives. Section
5.6, finally, contains conclusions and some topics for future research.
224 chapter 5
5.2 The Dutch pension system
5.2.1 Institutional setup
The state pension called AOW is the first pillar in the Dutch pension scheme, which is based
on three pillars. The second pillar consists out of supplementary company or sector pension
facilities. Employees are obliged to take part in those second-pillar pension programmes.
The third pillar contains individual pension saving programmes which are voluntarily to
participate in. Unlike the first-pillar pensions, both second and third-pillar pensions are fully
funded. This three-pillar model has a long tradition in the Netherlands, and is becoming more
and more popular as a transitional system in which a small public PAYG scheme is kept
alongside a larger (private) funded scheme (Baroni, 2007).
Dutch government supplies a state pension called AOW to all persons aged 65 or over when
they are entitled. Inhabitants of the Netherlands build up a right to this pension by living or
working in the Netherlands while aged between 15 and 65. A right of 2% for the state pension
is built up for every year this condition is fulfilled. Part of the population is only partially
entitled because they have lived only temporarily in the Netherlands when aged between 15
and 65. The number of incomplete state pensions is rising because of the growing number of
immigrants during the last decades.
The state pension scheme aims to provide a basic minimum income guarantee in case of
a full entitlement. Therefore the system makes a distinction between partners of a couple
and singles. A single gets a benefit of 70 percent of the minimum wage55 and a partner of a
couple gets 50 percent of the minimum wage. Until 2015, pensioners with a (non-working)
partner younger than 65 can supplement their state pension of 50 percent with an allowance
of another 50 percent to a combined maximum of 100 percent of the minimum wage. Partly
entitled persons can lay a claim on social assistance. Social assistance, however, is income
and means tested.
55 The gross minimum wage in 2009 amounts to approximately € 18,000 per year. The gross AOW-benefit for a single is
approximately € 12,700; the gross AOW-benefit for a couple is approximately € 8,700 for each partner. In net terms this
amounts to 70% and 50% of the net minimum wage respectively.
The AOW is a pay-as-you-go arrangement, the current population of workers pay for the cur-
rent population of pensioners. The AOW is financed through a premium paid by these work-
ers. The premium is fixed at a rate of 17.9 percent of the first two tax brackets (the limiting
income is approximately € 32,000 in 2009). This premium revenue is not sufficient to cover
all AOW costs. The government contributes the part of the AOW costs that is not covered by
the premiums. The government contribution is financed by taxes, which are paid by pension-
ers as well.
5.2.2 Current situation
The ageing of the population is in particular a complication for the state pension, which is
financed through a pay-as-you-go system. In 2009, 2.8 million people were receiving a state
pension. The total cost of these pensions in 2009 amounted to € 27.7 billion. Currently, from
the total cost of state pensions of € 27.7 billion, about two third comes from premiums and one
third from taxes. The AOW expenses are rising sharply because of the ageing of the population
whereas the premium income has stabilized. This implicates that the state contribution is rising.
The importance of second and third-pillar pensions for the income position of the elderly is
growing as more people are saving for such pensions and their average savings are increas-
ing. Per person average second-pillar pension savings are almost equal now to the average
first-pillar state pension entitlements. In the future, it is to be expected that second and third-
pillar pensions together will provide more than half of the average pension income. Although
there are many second-pillar pension funds in the Netherlands, each with its own rules on
contributions and pensions, broadly speaking one can say that pension funds try to supple-
ment the state pension to a total gross income level of 70% of the final wage. Because pen-
sioners do not have to pay state pension contributions anymore, the net level of their first and
second-pillar pensions together usually comes close to 90% of the final wage. Other income
sources, like third-pillar pensions can add to this income level.
226 chapter 5
Recent studies report mixed findings on whether people actually reach their target of a 70%
replacement rate (total pension as a percentage of the final wage). In a microsimulation study
on the wealth of Dutch pensioners (SZW, 2006) the income position of pensioners is found to
improve substantially as a consequence of more second and third-pillar pension savings from
younger generations. Statistics Netherlands (CBS, 2008) finds that a gross income level of
on average 73% of the final wage is to be expected for the cohorts currently saving for their
pension. Also in Van Sonsbeek (2011), it is shown that median replacement rates average
73% and are higher for the younger cohorts. However, in this study it is also signalled that re-
placement rate averages are biased because of high replacement rates for subgroups combin-
ing comparatively high pension savings with low current incomes (for example women that
changed to working part-time during their career). Men on average only reach a replacement
rate of 64%. Euwals, de Mooij and van Vuuren (2009) find that when the consequences of
the career average system most pension funds recently switched to are taken into account, the
younger generations have lower replacement rates, up to less than 50% for the birth cohorts
1972-1976.
5.2.3 Macro forecasts
Ageing has been on the political agenda since the 1980’s when a government commission
(the Drees commission) investigated the consequences of ageing for public finance. This
commission reported that the cost of state pensions as a percentage of GDP would rise to
somewhere between 7.0% and 8.5% in 2030. This early macro forecast is still reasonably
accurate. Many macro analyses followed, their forecasts ranging from 5.4% to 10.0% in 2040
(see for an overview SVB, 2008).
The most recent study was published in the run up to the 2006 general elections. In this study,
CPB (Van Ewijk et al., 2006) provided an analysis of ageing and the sustainability of Dutch
public finances using the general equilibrium model GAMMA. The authors conclude that if
current budgetary arrangements are maintained in the future, the gap between government ex-
penditures and revenues will increase by 3.3% of GDP from a 0.7% surplus to a 2.6% deficit.
This so called sustainability gap has a money value of € 15 billion based on the current GDP
level. Present budgetary and social security arrangements are therefore considered to be not
sustainable. Sustainability is defined by the authors as satisfying Musgrave’s criterion for
intergenerational equity (Musgrave, 1986). This criterion implies that the net benefit from the
government (relative to their lifetime incomes) is constant for all generations starting from
now. According to this criterion, starting from now, all generations should pay the same tax
rates and have the same benefit from government expenditures.
According to Van Ewijk et al., the cost of state pensions rise by 4.1% of GDP and the health
care costs rise by 4.3% of GDP. These ageing related cost rises are the main causes of the
rising government expenditures. The rises are partially compensated because tax revenues
on pension income rise by 1.8% of GDP and tax revenues on consumption of the elderly by
2.3% of GDP. However, they do not fully compensate the cost rise, causing a sustainability
gap of 2.6%56 of GDP. Meanwhile the sustainability gap has tripled to around 6% of GDP
because of the economic crisis (Gradus, Beetsma and Smit, 2010).
Van Ewijk et al. calculate the effects of different policy measures on the sustainability gap.
Among others they consider raising the retirement age from 65 to 67 years of age and abol-
ishing the tax exemption for pensioners and conclude that both measures reduce the sustaina-
bility gap by 0.6%57. Both measures were again considered with more or less the same results
for the government commission (Advies Commissie Arbeidsparticipatie, 2008) that was to
propose policy measures to boost labour participation and that mentioned both as long term
solutions to the sustainability problem. In more recent studies, like CPB (2009b) the effect of
raising the retirement age to 67 is slightly increased to 0.7%.
56 In a later study, the authors update their earlier estimates and present a sustainability gap of 2.2% of GDP (CPB, 2007).
The decrease is the result of an upward correction as a consequence of newer life expectancy tables from Statistics
Netherlands (CBS, 2007) and a downward correction because of improvement of the primary EMU balance from 2006
to 2007 and inclusion of current government’s policies.
57 This reduction is the result of a general equilibrium model and concerns both direct and indirect effects on both
expenses (state pension and other benefits) and tax revenues.
228 chapter 5
5.2.4 Redistributive characteristics and actuarial fairness of the state pen-

sion system
The Dutch state pension scheme can be classified as a “Beveridge”-style public pension pro-
gramme (Disney, 2004), characterized by significant departures from actuarial fairness and
significant provision of private retirement benefits, as opposed to “Bismarck”-style public
pension programmes, characterized by high “actuarial fairness” and limited private provision
of private retirement benefits. The Dutch scheme, with its flat rate pensions for singles and
cohabitants, therefore has a highly intra-generational redistributive character.
Actuarial fairness requires that the present value of lifetime contributions to a scheme equals
the present value of lifetime benefits from a scheme. Actuarial fairness so relates to the entire
lifetime of contributions and benefits (Queisser and Whitehouse, 2006). The redistributive ef-
fects of a scheme can be considered to be the flip side of the actuarial fairness of the scheme.
In an actuarially fair scheme, there will be no redistribution, except redistribution through
random differences in life expectancy. However, no pension scheme can be considered fully
actuarially fair as redistribution effects are inherent to pension systems because of differences
in life expectancy between groups and pension schemes providing safety nets to the poor.
The Dutch state pension system adds to this different pension levels for singles (70% of the
minimum wage) and partners of a couple (each 50% of the minimum wage), causing redistri-
bution from couples to singles.
In this chapter intra-generational redistribution is assessed. Besides intra-generational redis-

tribution, also intergenerational redistribution (e.g. from the younger birth cohorts to the older
birth cohorts) can be considered, like in Pettersson, Pettersson and Westerberg (2006). This
kind of redistribution is particularly important when discussing the intergenerational fairness
of policy proposals. For a pension scheme, when comparing the total amount of contributions
to the scheme to the total amount of withdrawals from the scheme, internal rates of return can
be computed for consecutive birth cohorts in order to assess the intergenerational redistribu-
tion. Typically, given their rising life expectancy, benefit withdrawals in a public PAYG sys-
tem will overcompensate previous contributions for the current cohorts. Current cohorts can
be expected to have a positive internal rate of return: they receive better value from the state
pension plan than the cohorts that follow. In a microsimulation study of the Canadian Pension
Plan, Morrison (2009) shows that all current cohorts have positive internal rates of return.
This holds true even for subgroups with lower life expectancies that on an intra-generational
basis are subsidizing groups with higher life expectancies. Future extensions of the SADNAP
model are planned to allow for analyzing intergenerational redistribution.
5.3 Model and data
5.3.1 Model
The use of microsimulation models in policy assessment and evaluation is becoming more
widespread (see e.g. Buddelmeyer, Freebairn and Kalb, 2006, and specifically on pension
reform Oksanen, 2004 and Stensnes and Stolen, 2007). This chapter uses a microsimulation
model called SADNAP (Social Affairs Department of the Netherlands Ageing and Pensions
model) to analyse the budgetary, redistributive and labour participation effects of policy
measures. The model, which is described in Van Sonsbeek (2011), is a dynamic microsimu-
lation model that simulates life paths of a sample of the Dutch population using transition
probabilities on demographic events. The model uses administrative datasets on 2006 state
pension payments and 2005 state pension and private pension entitlements. The demographic
component of the model is aligned with the newest population projection of Statistics Neth-
erlands (CBS, 2009), taking into account the most recent birth and immigration numbers and
mortality and emigration rates. The SADNAP model has been modularly designed and its
main demographic and budgetary modules have been used in the budget forecasts of the Min-
istry of Social Affairs and Employment since 2007. Recently the model has been extended
with a module modelling the retirement decision. Also differentiation in mortality rates by
230 chapter 5
various characteristics has been introduced, so that the difference in life expectancy between
higher and lower incomes, between singles and cohabitants and between natives and immi-
grants is now accounted for. These extensions are described in detail in Van Sonsbeek (2011).
The main events modelled concern the household and participation status and the retirement
decision. The household and participation status are simplified to a binary choice model,
allowing the household status to be either single or cohabiting and the participation status to
be either participating or non-participating. The retirement decision is modelled by the option
value approach first suggested by Stock and Wise (1990), taking into account the individual
data on wages, state pension entitlements and private pension entitlements and combining
those with individually varied option value parameters (time preference, leisure preference
and risk aversion). Also, productivity loss at higher ages and age- and gender specific mortal-
ity-, unemployment- and disability rates are taken into account. An extensive description of
the model and its parameters is included in chapter 4.
5.3.2 Data sources
In SADNAP, three micro data sources are used. Statistics Netherlands (CBS) supplies a micro
data file on state pension entitlements. This file contains all over 11 million persons aged 15
to 64 who lived in the Netherlands in 2005. The Social Insurance Bank (SVB), the authority
that accomplishes the payment of state pensions, supplies a full administrative data file with
information on all 2.6 million persons receiving state pension in 2006. Recently, Statistics
Netherlands has also started providing data on (second-pillar) company pensions. In 2009, a
micro dataset has become available based on data of a representative sample of pension funds
including the large pension funds for civil servants and health workers. This file includes in-
dividual data on company pension entitlements in 2005 of 53% of the population aged 15-64
and 67% of the Dutch employees. A detailed description of these data sources is presented in
the separate chapter on the SADNAP model (Van Sonsbeek, 2011).
Besides the micro data sources, a number of macro data sources are used as well. These
include the most recent population projection by age, gender and ethnicity by Statistics
Netherlands, running from 2008 to 2050, the household projection from Statistics Nether-
lands, running from 2009 to 2050 and the most recent labour participation rate projection as
provided by CPB (Euwals and Folmer, 2009), also running from 2009 to 2050.
5.4 Baseline scenario in a micro approach
5.4.1 State pension costs in the baseline scenario
When pensions stabilize at the current level in real terms, the state pension costs will rise
from € 27.7 billion in 2009 to € 50.3 billion in 2040. Like in earlier CPB publications, the
year 2040 is focused upon throughout this chapter for the long-term effects as around 2040
the ageing in the Netherlands is at its peak. In terms of GDP, assuming that GDP also stabi-
lizes at the current (2009) level of € 573 billion, the state pension costs will rise from 4.8%
in 2009 to 8.8% in 2040. The rise is huge, but still somewhat less than expected. When also
constant pension costs per pensioner would be assumed, state pension costs would rise to €
51.9 billion in 2040 or 9.1% of current GDP. It appears that the cost per person will decrease
over time (apart from increases through indexation). These decreasing cost per person ac-
count for a 0.3% of GDP lower state pension cost. The rising share of reduced state pensions
(because of the rising share of first generation immigrants) and the rising labour participation
of women account for 0.2% each. The development in the share of singles in the population
of pensioners has a small upward effect of 0.1% of GDP in 2040. During the oncoming dec-
ades, the share of singles among the pensioners population is decreasing but this trend will be
reversed in the future.
232 chapter 5
5.4.2 Retirement age in the baseline scenario
In the Netherlands, the first possible retirement age used to be 60 years in many sectors.
Since the late 1980’s, for most employees a generous early retirement scheme existed that
guaranteed an income level of 70-80% of the final wage without loss of pension accruals
from 60 years of age onwards. Gradually, the generous early retirement schemes are being
replaced by actuarially neutral schemes until, from 2015 onwards almost all schemes will be
fully actuarially neutral (see Bovenberg and Gradus, 2008). This will lead to a sharp increase
of the effective retirement age. With the option value model that is incorporated in SADNAP
a retirement pattern is predicted for a baseline scenario with fully actuarially neutral schemes.
This pattern concerns only the population that is still working at 60. At that age, a lot of
people are already not available to the workforce anymore. Even according to the newest par-
ticipation rate forecasts (Euwals and Folmer, 2009), also in the future at age 60 almost half
of the women and 30% of the men will not be participating at the labour market anymore,
mainly due to being on benefit and for women also to voluntary unemployment. Of the ones
that are still working at 60, about 26% will retire before 65, about 38% at 65 and about 36%
past 65. As compared to the generous early retirement schemes from the past, the average
retirement age of the population still participating at age 60 has increased by 2.5 years.
The baseline projections give a plausible retirement pattern, especially for males, although
the number of people working past the age of 65 is larger than currently observed. This may
be due to the compulsory dismissal at age 6558. As government has meanwhile proposed to
abolish this compulsory dismissal, it is to be expected that the number of workers past the
age of 65 will grow in the future. The importance of abolishing the compulsory dismissal at
age 65 could be substantial in eliminating the ‘Age-65-retirement-effect’ which explains the
high retirement spikes usually found at 65 from the influence of custom or accepted practice
(Lumsdaine, Stock and Wise, 1995). In the SADNAP projections, the last possible retirement
age is 69. There is a substantial spike at this last possible retirement age, caused by people
58 In a later study, the authors update their earlier estimates and present a sustainability gap of 2.2% of GDP (CPB, 2007).
The decrease is the result of an upward correction as a consequence of newer life expectancy tables from Statistics
Netherlands (CBS, 2007) and a downward correction because of improvement of the primary EMU balance from 2006
to 2007 and inclusion of current government’s policies.
for who financial incentives make working until the last possible retirement age the most at-
tractive option. Although this may from the current viewpoint seem improbable, it should be
noted that at the moment already 13% of the men aged 65-74 are participating at the labour
market (Monden, 2008), albeit mostly as a self-employed or in smaller part-time jobs.
5.4.3 Redistribution and actuarial fairness in the baseline scenario
Redistributive effects of the various policy measures are assessed in two ways. These two
approaches may lead to different conclusions. Broadly speaking, the second, more detailed
approach can be used to explain the results from the first approach.
First, a general inequality indicator, the Gini coefficient, is used to judge whether the policy
measure increases or decreases income inequality. The Gini coefficient is computed on the
total individual pension income (state pension plus second-pillar pension) of all pensioners in
2040. In 2040, The Gini coefficient is 0.291 in the Baseline scenario. By using the 2040 data,
the same year is used for the effects on the Gini coefficient as for the budgetary effects. The
population of pensioners in 2040 consists of all people that retired between 2006 and 2040,
making 2040 the first year having a population of pensioners that almost completely follows
from the simulation. This is important because for the people that retired before 2006 the
second-pillar pension entitlements are not known and for the people retiring after 2040, the
second-pillar pension entitlement data become less reliable as they are based on the wages of
those people in 2005, so when they were still below 30 years of age.
Second, the redistribution within the scheme is investigated in more detail by computing
the share of lifetime state pension income taken by different subgroups. The lifetime state
pension income is computed by accumulating incomes from the year a person turns 65 until
the year a person dies. For this analysis, the 2006-2045 pensioner cohorts (the 1941-1980
234 chapter 5
birth cohorts) are aggregated. As the simulation runs until 2080, the 2045 pensioner cohort is
one of the last cohorts that by 2080 will have almost completely died out. Table 5.1 shows a
subdivision of the accumulated cohorts by subgroup, with the share of each subgroup in the
cohorts of pensioners, its share of lifetime state pension income and the ratio between the
two.
Table 5.1 Share of lifetime state pension income compared with share of state pension
cohorts
Subgroup Share of cohorts Share of lifetime Ratio
turning 65 pension costs
By income
1st quintile 19.4% 15.4% 0.79
2nd quintile 19.8% 18.5% 0.93
3rd quintile 20.0% 19.4% 0.97
4th quintile 20.3% 21.8% 1.08
5th quintile 20.5% 24.9% 1.21
By gender
Women 49.4% 52.6% 1.06
Men 50.6% 47.4% 0.94
By household status
Singles 30.6% 34.0% 1.11
Cohabitants 69.4% 66.0% 0.95
By origin
Natives 73.5% 82.4% 1.12
Immigrants 26.5% 17.6% 0.66
The average lifetime state pension income per person is around € 190,000, with lifetime in-
come per person decreasing for the later cohorts because of the rising number of people with
incomplete state pension entitlements. The higher income quintiles receive an above aver-
age share of total state pension because of differences in life expectancy. This redistribution
through life expectancy is substantial. The first income quintile receives more than a third
less than the fifth income quintile (a ratio of 0.79 vs. a ratio of 1.21). This is mainly due to the
difference in life expectancy, but also to the larger share of incomplete state pensions in the
lower income quintiles. Women receive 6% more state pension from the scheme than their
share in the cohort would justify. Singles receive 11% more state pension from the scheme
than their share in the cohort would justify. This is because the lower life expectancy of
singles is overcompensated by their higher state pension. Immigrants receive 34% less state
pension from the scheme than their share in the cohort would justify. However, this large
difference is in the first place due to immigrants building up less entitlement during their life
and only for a smaller part to differences in life expectancy.
5.5 Policy alternatives
In this section various policy alternatives are discussed. Dutch government in October 2009
decided to raise the retirement age to 67 years59. However, many alternatives to raising the
retirement age have been considered meanwhile and a final decision still has to take place as
the government proposal is still due to be discussed in parliament in 2010.
5.5.1 Overview of policy measures
The first measure considered is the abolishment of the partner allowance. This policy change
was already decided about in 1996 but will be implemented from 2015 onwards. Therefore,
this measure was not yet included in earlier macro forecasts of CPB. A state pensioner quali-
fies for the partner allowance when he or she has a partner that is younger than 65 that earns
not enough income of his own. As a consequence, most of the partner allowance recipients
are men turning 65 having a several years younger spouse who is earning no or too little
income herself. Therefore the partner allowance implicitly penalizes labour participation of
the younger spouses.
59 The retirement age will be raised in two steps, first to 66 in 2020 and then to 67 in 2025.
236 chapter 5
The second measure considered is the raising of the retirement age to 67 as current govern-
ment decided to do. Raising the pension age from 65 to 67 was already considered by some
smaller political parties during the 2006 election campaign, but was at that time avoided by
the main political parties. Meanwhile, not only the government parties, but also the main
liberal opposition party now advocates this measure (CPB, 2009a).
The third measure considered is raising the taxation of pensioners, by eliminating their social
security premium exemption. Pensioners do not pay contributions to the pension scheme
anymore. This tax exemption allows a reduction in the tax rate of 17.9% points in the first
two income tax brackets. This measure, which was favoured by most left-wing parties during
the 2006 election campaign, leaves the net state pension unchanged so affects only the people
with additional second or third-pillar pensions60.
The fourth measure concerns individualizing the state pension level to 50% of the minimum
wage for each pensioner. Currently singles get a higher pension than cohabitants (70% vs.
50% of the minimum wage). This is justifiable from the viewpoint of singles having higher
cost of living than cohabitants61, but not from the perspective of contributions to the scheme.
Moreover, second-pillar pension funds usually do not distinguish between singles and cou-
ples when setting the desired pension level. This results in singles ceteris paribus reaching a
higher replacement rate (income from first and second-pillar pension together as a percentage
of the final wage) than partners of a couple. Such an individualization was recently men-
tioned in Bovenberg and Gradus (2008) but has a far longer history and was previously pro-
posed already by the 1987 Drees commission that advised on how to cope with ageing (SVB,
2008), albeit at a higher level of 60% of the minimum wage for each pensioner, regardless of
the marital or cohabiting status.
A final measure described in this chapter is the introduction of a retirement window between
65 and 70 years of age with an accrual rate that is above actuarially neutral level in order to
60 Current government in 2007 decided to slowly phase out a (comparatively small) part of this tax exemption between
2011 and 2030.
61 The state pension aims to provide a basic minimum income guarantee, so a complete state pension entitlement guaran-
tees a net pension that is exactly equal to the social minimum level, which is 70% of the minimum wage for singles and
100% of the minimum wage for a couple.
stimulate employees to delay their retirement. This option is suggested by Den Butter and
Van Sonsbeek (2008) and can be seen as an adaptation of current government’s proposal
to introduce a retirement window between 65 and 70 years of age with actuarially neutral
adjustments (which is also due to be discussed in parliament in 2010). For this measure, an
accrual rate of 8% instead of 5% in the government’s proposal is assumed.
Raising the retirement age and introducing a retirement window can be considered generic
measures that affect all persons in the same way, albeit that some will be able to cope with
such changes better than others. The other measures are specific measures that affect spe-
cific groups and leave others unaffected. Such specific measures fundamentally change the
character of the state pension scheme. For example, because abolishing the tax exemption of
pensioners affects higher incomes (that contributed more to the scheme) more than lower in-
comes, actuarial fairness of the scheme will decrease, giving the scheme, in terms of Disney
(2004) a more ”Beveridgean” character. On the other hand, both abolishing the partner allow-
ance and decreasing the state pension for singles to the current level for cohabitants increase
actuarial fairness and give the scheme, in terms of Disney, a more “Bismarckian” character.
5.5.2 Direct budgetary effects
As SADNAP is not an equilibrium model, analysis of the budgetary effects is limited to the
direct effects. This underlines the case for combining microsimulation models with macro GE
models in order to get an optimal assessment of a measure (see e.g. Peichl, 2008). Direct ef-
fects incorporated in SADNAP include both the expenses (state pensions and social security
benefits) and the tax revenues.
All measures are assumed to be in force from 2015 onwards. The abolishment of the partner
allowance will be in force from 2015 onwards, as this measure is already decided about. For
238 chapter 5
comparability the other measures follow this scheme. The measures are assumed to affect
only the pensioner cohorts of 2015 onwards in order to prevent existing rights to be violated.
For abolishing the partner allowance and decreasing the singles pension, this means that all
people that turned 65 before 2015 keep their entitlements to the partner allowance and the
70% singles pension respectively. For decreasing the singles pension this implies that the
full effect of the measure will only be reached around 2050 when the cohorts that turned 65
before 2015 have died out. On the other hand, raising the retirement age by definition only
affects new cohorts and, from the state pension perspective, raising the taxation of pensioners
by abolishing their tax exemption has no effect at all as the higher taxation affects only the
second-pillar pensions.
It is assumed that the retirement age is increased immediately from 65 to 67 in 2015 and
that the tax exemption is abolished immediately in 2015 for all. Although, this is technically
possible, from a policy viewpoint, this may not be a realistic assumption. Both for raising the
retirement age and raising the taxation of pensioners a stepwise approach will be more plausi-
ble, like in the government proposal to raise the retirement age in two steps in 2020 and 2025
to 66 and 67 respectively. The financial effects of such a stepwise approach can of course be
easily deducted from the tables on the immediate approach. Table 5.2 summarizes the effects
of the different measures in 2040. A detailed table of direct budgetary effects is included in
appendix 5.A. Also the structural effect (2050 onwards) is given as decreasing the singles
pension will only be fully effective in 2050. For 2040, the effects of the five measures are
given as compared with the baseline scenario and to a baseline scenario in which the abolish-
ment of the partner allowance is incorporated. This especially affects the results of raising
the retirement age because when the partner allowance starts at a higher retirement age, the
costs of the partner allowance will rise as chances of a partner earning enough income decline
when that partner is also two years older.
Table 5.2 Projected budgetary effects of the policy alternatives (2040)

Effects 2040 (%GDP) Partner Retirement Raising Pension Ret. window
allowance age 67 taxation singles 50% Accrual 8%
Compared to baseline
State pension cost (1) -0.25 -0.64 +2.37 -1.19 +0.15
Other benefits (2) +0.02 +0.11 0 +0.10 0
Tax revenues (3) -0.02 -0.11 +3.35 -0.20 +0.03
Net effect (1+2-3) -0.20 -0.42 -0.98 -0.89 +0.12
Baseline incl. partner allowance
Net effect (1+2-3) -0.50 -0.98 -0.89 +0.12
Structural effect (2050)
Net effect (1+2-3) -0.50 -0.85 -1.04 +0.22
Direct budgetary effects of abolishing the partner allowance amount to 0.25% of GDP in
2040. As this measure is already decided about, the effect should in fact be part of a baseline
scenario. When added to the decreasing cost per person in the state pension scheme from
section 5.4.1, that reduces the baseline forecast by 0.3% of GDP, a 0.55% reduction of state
pension costs can be made as compared with the macro forecast of van Ewijk et al. This
compensates the effect of the rising longevity from the most recent population projection that
causes a 0.5% of GDP increase in the state pension cost forecast (Van Sonsbeek, 2011).
By far the largest direct budgetary effects can be achieved by reducing the state pension for
singles to 50% of the minimum wage and by raising the taxation of pensioners. Individuali-
zation of the pension levels reduces the state pension cost by 1.19% of GDP in 2040 and by
1.38% in 2050 when the full effect of this measure is reached. The cost of social assistance
will go up, but as most people have adequate second-pillar pension savings or personal
wealth, no social assistance can be claimed by most. When also lower tax revenues are taken
into account, a net effect of 1.04% of GDP results.
Abolishing the tax exemption of pensioners actually increases state pension costs as the
state pension is linked on a net basis to the social minimum income so that a removal of the
240 chapter 5
tax exemption would increase state pensions on a gross basis, yielding the same net dispos-
able state pension income as before (OECD, 2008). The gain from this measure of course
comes from the tax revenues that rise faster than the pension expenses. The direct net effect
of this measure (0.98% of GDP in 2040 and 0.85% in 2050) is slightly larger than mentioned
in previous studies, probably because future pension wealth increases for new cohorts of
pensioners. However, it should be noted that according to general equilibrium model analysis
like from De Hek and Van Erp (2009) a negative effect on labour supply is to be expected as
individuals supply less labour over their lifetime and the reduction in labour supply and the
participation rate is widespread across the different ages.
Raising the retirement age decreases state pension costs by 0.72% of GDP in 2040. However,
the costs of the partner allowances go up by 0.08%. Also expenses on other social benefits
(mainly disability, unemployment, social assistance and survivor’s pensions) will go up as
people that are on one of those benefits at age 64 will most probably use that benefit for
two years more. However, this leakage rate decreases over time because of the decreasing
disability rates that resulted from the 2006 reform of the DI scheme (see Van Sonsbeek and
Gradus, 2006). Tax revenues decrease only slightly as people that are working or on benefit at
65 and 66 will also pay taxes. The net direct effect is about half of that of reducing the singles
pension: 0.42% when the partner allowance is kept and 0.50% when the partner allowance is
abolished.
Introducing a retirement window with a high accrual increases state pension costs for two
reasons. In the first place, state pension costs rise because of adverse selection: people with
longer life expectancies are more likely to delay their pension. Evidence for adverse selection
comes from recent studies by Finkelstein and Poterba (2004) and Kühntopf and Tivig (2008).
In the second place, the high accrual rate means that lifetime state pension expenses rise for
all people delaying their retirement. State pension costs go up by 0.12% of GDP in 2040 and
0.22% of GDP in 2050.
Figure 5.1 shows how the year-by-year savings on state pension expenses of the four policy
measures build up over time (abolishing the tax exemption is abstracted from as state pension
costs are not affected). This concerns the total savings on state pension expenses of a measure
as compared with the baseline scenario excluding abolishment of the partner allowance.
Figure 5.1 Projected savings of policy alternatives 2006-2050
Savings of abolishing the partner allowance quickly rise till € 1.7 billion until around 2030,
but decrease over time as labour participation of women increases and cohorts of new
pensioners are growing smaller. Savings of raising the retirement age rise very quickly to a
structural level of around € 4 billion. When the size of the pensioner cohorts grows smaller
after 2036, the savings decrease. Savings of decreasing the singles pension increase slowly to
a structural level of around € 8 billion, which is reached around 2050 because all pensioners
from before 2015 are assumed to be not affected by this measure. Introducing a retirement
window with a high accrual first causes savings because early cohorts start delaying their
pension. However, as soon as they start claiming their higher state pension, savings disappear
242 chapter 5
and turn negative eventually from 2034 onwards. The balance is in particular negative around
2050 when the ageing of the population got past its top. At that time the size of the new co-
horts, part of which is delaying the state pension, is decreasing, while the larger cohorts that
delayed the state pension in the past still have to be paid higher state pensions.
5.5.3 Labour participation effects
Labour participation effects are measured by the effect on the effective retirement age of the
people that are still working at age 60. Their retirement age is forecasted by the option value
model described in section 5.3.1. Again it should be noted, that effects on participation over
the lifetime are not taken into account, nor are effects on the participation of other household
members. Table 5.3 summarizes the effects on the distribution of retirement ages of the vari-
ous policy measures. It should be noted that the baseline scenario is the current situation of
unchanged policies. This does not fully represent current retirement patterns as until recently
generous early retirement schemes were in force, that are slowly being phased out. Therefore
an additional benchmark scenario is added with generous early retirement schemes for most
of the population like existed during the 1990’s.
Table 5.3 Projected effect on retirement age of the policy alternatives

Retirement age Generous Baseline Partner Ret. age Raising Singles Accrual
ERS allowance 67 taxation 50% 8%
Not participating at 60 43 43 43 43 43 43 43
Between 60 and 64 39 15 15 15 14 15 15
At 65 7 22 19 5 22 21 19
Past 65 11 20 23 37 20 21 24
Average (conditional on 62.8 65.2 65.3 65.8 65.3 65.3 65.5
working at 60)
Δ Baseline (months) -29 1 7 0 1 3
Raising the retirement age till 67 causes the largest positive effect on the retirement age,
increasing it by 7 months. This corresponds with a literature overview by Euwals et al. (2008)
in which effects from raising the retirement age on the effective retirement age are mentioned
ranging from about 10% (2 months) from several ex-ante evaluations to 50% (1 year) from
an ex-post evaluation by Mastrobuoni (2006). The higher ex-post effect was explained by the
new retirement age becoming the new social-cultural norm, which cannot be captured in the
option value model. An increase of the effective retirement age will also increase tax rev-
enues. The effect of raising the taxation of pensioners and decreasing the pension of singles
on the retirement age is very small. Moreover, it should be kept in mind that raising the taxa-
tion of pensioners has a negative effect on labour supply over the lifetime. The small effect
on the retirement age by the large decrease of the pension of singles can be explained by the
comparatively small number of people immediately affected (at age 65, about 20% of men
and 30% of women are single) whereas when raising the retirement age to 67 the complete
cohorts of 65 and 66 years old are affected. The effect of abolishing the partner allowance on
the retirement age is also small, but may become more substantial when taking into account
that an important additional effect on labour participation may occur through the participation
decision of the younger partner. Abolishing the partner allowance also has an immediate ef-
fect on a comparatively large part of the cohort. Introducing a retirement window with a high
accrual substantially boosts labour participation. The increase is half of the increase when
raising the retirement age. This confirms findings from the US by Pingle (2006) who shows a
strong correlation between the accrual and the use of delaying retirement and from Vonkova
and Van Soest (2009) who find similar results from stated preference surveys in the Neth-
erlands. However, all effects of new policy measures are minor as compared with the huge
change in retirement behaviour and increase of the average retirement age (almost 2.5 years)
caused by abolishing the generous early retirement schemes.
244 chapter 5
5.5.4 Redistributive effects
The Gini coefficients of the different policy measures are presented in table 5.4. The abol-
ishment of the partner allowance and the raising of the retirement age affect the Gini coef-
ficient only marginally. Raising the taxation of the wealthier pensioners of course decreases
income inequality. The Gini coefficient decreases by over 2 percentage points. Decreasing
the pension of singles to 50% will increase income inequality by 1.5 percentage point. When
comparing state pensions only, one would expect the income inequality to decrease because
the difference between singles and couples is levelled out. However, singles have a lower
second-pillar pension (about € 2,000 on average) which mitigates the difference in state
pension level. When state pension levels are equalised, the difference between singles and
cohabitants in second-pillar pensions becomes larger than their difference in total pensions
in the baseline. Introducing the retirement window with a high accrual also increases income
inequality as higher incomes with longer life expectancy are more likely to delay their retire-
ment.
Table 5.4 projected effect on income inequality of the policy alternatives

Baseline Partner Retirement Raising Pension Ret. window
Gini coefficient 0.291 0.293 0.291 0.268 0.306 0.297
Difference +0.002 0 -0.023 +0.015 +0.006
In table 5.5, the ratio of the share of lifetime state pension income and the share of the cohorts
turning 65 is presented. The policy alternatives can be compared with the baseline which was
already presented in table 5.3. Raising the taxation of the wealthier pensioners has no effect
on the state pension level, so has no effect on this ratio as only the redistribution within the
state pension scheme is analysed. The table gives the redistributive effects of the measures
when they have reached their full effect so the transition period that plays an important role
with some measures has no influence on the results.
Table 5.5 Effect of policy alternatives on share of lifetime state pension income com-
pared with share of state pension cohorts
Subgroup Baseline Partner Retirement Raising Pension Ret. window
62
By income
1st quintile 0.79 0.79 0.78 0.79 0.78 0.79
2nd quintile 0.93 0.93 0.93 0.93 0.94 0.93
3rd quintile 0.97 0.97 0.97 0.97 0.97 0.97
4th quintile 1.08 1.07 1.08 1.08 1.07 1.08
5th quintile 1.21 1.22 1.23 1.21 1.22 1.22
By gender
Women 1.06 1.09 1.06 1.06 1.03 1.07
Men 0.94 0.91 0.94 0.94 0.97 0.94
By household status
Singles 1.11 1.14 1.09 1.11 0.85 1.11
Cohabitants 0.95 0.94 0.96 0.95 1.08 0.95
By origin
Natives 1.12 1.12 1.13 1.12 1.12 1.12
Immigrants 0.66 0.66 0.65 0.66 0.66 0.66
The different measures have little effect on the share of lifetime state pension income taken
by different income quintiles and by natives vs. immigrants. Raising the retirement age to 67,
decreasing the singles pension and introducing a retirement window with a high accrual all
slightly increase the share of lifetime state pension taken by the higher incomes because of
their higher life expectancy. Abolishing the partner allowance increases the share of pension
income taken by women (ratio of 1.09 vs. 1.06) as mostly men are receiving partner allow-
ances because they usually turn 65 first and have a higher chance of a non-working partner.
Of course, abolishing the partner allowance also increases the share of lifetime pension
income taken by singles (ratio of 1.14 vs. 1.11). Reduction of the singles pension to 50%
does exactly the opposite and has very large redistributive effects. Because of their lower life
expectancy, singles take a smaller share of lifetime pension income than partners of a couple
when their pension levels are equal (ratio of 0.85 vs. 1.11). It is worth noting that by combin-
62 Note that the numbers for this policy alternative exactly equal the baseline numbers. The net state pension in this alterna-
tive does not change as compared to the baseline. Financial gains in this alternative come from higher taxation of second
pillar pension income.
246 chapter 5
ing abolishment of the partner allowance and reduction of the singles pension their redistribu-
tive effects are mitigated somewhat, which can make this combination a more acceptable
option from a policy point of view. This can be even more so when such a combination of
measures is combined with the retirement window between 65 and 70 with a high accrual,
which allows both singles and partners of a couple to actuarially increase their pension to a
higher level. Singles may make more use of such an opportunity than couples because of the
importance of the social network belonging to their work. Moreover, it is known from Coile
(2004) that husbands’ and wives’ retirement behaviour is influenced not only by their own fi-
nancial incentives but also by spill over effects from their spouses’ incentives so that financial
incentives may be less effective for couples than for singles.
5.5.5 Assessment of the policy alternatives
Table 5.6 summarizes the effects on different indicators for the four policy alternatives. Actu-
arial fairness is not yet quantified with an indicator, but a qualitative analysis can be made.
Table 5.6 Overall assessment of the policy alternatives

Measure Partner Retirement Raising Pension Ret. window
Budgetary effect (Δ % of GDP) -0.2 -0.5 -0.9 -1.0 +0.2
Behavioural effect (Δ ret. age, months) 1 7 0 1 3
Redistributive effect (Δ Gini coefficient) +0.002 0 -0.023 +0.015 +0.006
Effect on actuarial fairness + 0 - + 0
From a budgetary point of view, reducing the pension for singles and raising the taxation of
pensioners have the largest direct effect (1.0% and 0.9% of GDP respectively). However,
both measures have only limited effect on labour participation as the retirement age is only
marginally affected. Moreover, both measures have large implications for income inequal-
ity and actuarial fairness. Raising the taxation by abolishing the tax exemption for pension-
ers decreases the Gini coefficient by over two percentage points but makes the system less
actuarially fair. Reducing the singles pension increases the Gini coefficient by 1.5 percentage
points but makes the system actuarially fairer. Also, it should be noted that raising taxation
has negative indirect effects on labour supply reducing the sustainability effect.
Raising the retirement age has a direct budgetary effect that is only half of the abovemen-
tioned measures. However, this alternative has the largest effect on labour participation as
the retirement age is expected to rise by seven months. Because such an increase in labour
participation will increase tax revenues, indirect budgetary effects will decrease the differ-
ence with the other measures. Raising the retirement age has no redistributive effects, which
can make the measure politically more attractive in a coalition government. Introducing a
retirement window with a high accrual has labour participation effects half the size of raising
the retirement age but increases state pension costs by 0.2% of GDP. Abolishing the partner
allowance has smaller but still substantial effects on both the government budget and labour
participation. This holds true especially when one takes into account the indirect effect of this
measure on the labour participation of the younger partners that is not part of this analysis. As
redistributive effects are minor and actuarial fairness of the scheme improves, this seems to
be the one no regret measure.
The ageing of the population jeopardises the sustainability of public finances in the Nether-
lands. Most political parties propose policy measures to restrain the sharp increase in pension
costs. The policy measures considered in this chapter are abolishment of the partner allow-
ance, raising the retirement age from 65 to 67, decreasing the pension of singles from 70%
to 50%, raising the taxation of wealthier pensioners and introduction of a flexible retirement
window with financial incentives for postponing retirement.
248 chapter 5
The microsimulation model SADNAP is used to calculate both the cost of the state pensions
and the consequences of the policy alternatives for the income distribution and the retire-
ment decision of pensioners. The redistributive character of the Dutch state pension system
can be assessed, taking into account differences in life expectancy between higher and lower
incomes, women and men, cohabitants and singles and between natives and immigrants. The
retirement decision is modelled based on the Stock and Wise option value approach, allowing
for individual variation in the main option value parameters in order to make use of the added
value of microsimulation. Therefore a broader judgment of policy alternatives can be made.
In the baseline scenario, by using the micro instead of the macro approach, it is shown that
the state pension costs rise less sharply then the number of pensioners. This compensates for
the increasing longevity from recent population forecasts.
Assessing the policy measures, from a budgetary point of view, reducing the pension for
singles and raising the taxation of pensioners appear to have the largest direct effect, 1.0%
and 0.9% of GDP respectively. However, both measures have large and opposite implications
for income inequality and actuarial fairness. From a labour participation point of view, raising
the retirement age has the largest effect, increasing effective retirement age by seven months.
This measure has no redistributive effects. Its direct budgetary effects are 0.5% of GDP, half
of that of the aforementioned measures. Abolishing the partner allowance has smaller but still
substantial effects on both the government budget and labour participation. As also actuarial
fairness of the scheme improves without causing redistributive side effects this seems to be
the one no regret measure. Introducing a retirement window with a high accrual has substan-
tial labour participation effects, as retirement age increases by three months, but increases
state pension expenses in the long run.
Future research on the model will focus on improving the model by using more recent data-
sets and estimating the option value parameters using consecutive datasets. The model will be
extended in order to assess the actuarial fairness of the scheme and of the policy alternatives.
This among others requires construction of life paths for all future pensioners by backwards
simulating the participation and cohabiting “careers” of the starting population. Finally, in
the ageing discussion the distribution of wealth over the generations is also important. In
the future, the model can be further extended, enabling a comparison of the expected future
wealth of generations currently contributing to the wealth of the generations currently using
the pensions. However, these intergenerational analyses are very complicated and still under-
developed.
250 chapter 5
Appendix 5.A Direct budgetary effects
Table 5.7 Direct budgetary effects baseline and policy scenarios 2009-2050
Baseline Partner Retirement Singles Tax Higher

allowance age pension exemption Accrual
mln. € %GDP mln. € %GDP mln. € %GDP mln. € %GDP mln. € %GDP mln. € %GDP
2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009
2009 27765 4.85% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00%
2010 28532 4.98% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00%
2011 29559 5.16% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00%
2012 30651 5.35% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00%
2013 31591 5.51% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00%
2014 32419 5.66% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00%
2015 33199 5.79% -259 -0.05% -962 -0.17% -25 0.00% 8956 1.56% -221 -0.04%
2016 33860 5.91% -554 -0.10% -2788 -0.49% -219 -0.04% 9135 1.59% -619 -0.11%
2017 34578 6.04% -831 -0.15% -3701 -0.65% -489 -0.09% 9329 1.63% -953 -0.17%
2018 35390 6.18% -1074 -0.19% -3817 -0.67% -700 -0.12% 9548 1.67% -1219 -0.21%
2019 36144 6.31% -1169 -0.20% -3810 -0.66% -918 -0.16% 9751 1.70% -1280 -0.22%
2020 36919 6.44% -1257 -0.22% -3811 -0.67% -1146 -0.20% 9960 1.74% -1187 -0.21%
2021 37730 6.59% -1358 -0.24% -3904 -0.68% -1385 -0.24% 10179 1.78% -1136 -0.20%
2022 38526 6.72% -1407 -0.25% -3965 -0.69% -1629 -0.28% 10394 1.81% -1081 -0.19%
2023 39401 6.88% -1440 -0.25% -4041 -0.71% -1887 -0.33% 10630 1.86% -1041 -0.18%
2024 40314 7.04% -1504 -0.26% -4191 -0.73% -2159 -0.38% 10876 1.90% -1016 -0.18%
2025 41177 7.19% -1555 -0.27% -4348 -0.76% -2442 -0.43% 11109 1.94% -978 -0.17%
2026 42023 7.34% -1557 -0.27% -4382 -0.76% -2741 -0.48% 11337 1.98% -925 -0.16%
2027 42896 7.49% -1578 -0.28% -4415 -0.77% -3058 -0.53% 11573 2.02% -815 -0.14%
2028 43795 7.64% -1618 -0.28% -4534 -0.79% -3364 -0.59% 11815 2.06% -705 -0.12%
2029 44703 7.80% -1654 -0.29% -4622 -0.81% -3663 -0.64% 12060 2.10% -589 -0.10%
2030 45513 7.94% -1647 -0.29% -4583 -0.80% -3963 -0.69% 12279 2.14% -457 -0.08%
2031 46178 8.06% -1611 -0.28% -4481 -0.78% -4257 -0.74% 12458 2.17% -345 -0.06%
2032 46895 8.19% -1650 -0.29% -4449 -0.78% -4551 -0.79% 12651 2.21% -254 -0.04%
2033 47647 8.32% -1684 -0.29% -4461 -0.78% -4856 -0.85% 12854 2.24% -120 -0.02%
2034 48354 8.44% -1659 -0.29% -4530 -0.79% -5175 -0.90% 13045 2.28% 44 0.01%
2035 48993 8.55% -1630 -0.28% -4581 -0.80% -5488 -0.96% 13217 2.31% 160 0.03%
2036 49548 8.65% -1617 -0.28% -4478 -0.78% -5790 -1.01% 13367 2.33% 264 0.05%
(continuation) Table 5.7 Direct budgetary effects baseline and policy scenarios 2009-2050
Baseline Partner Retirement Singles Tax Higher

allowance age pension exemption Accrual
2037 49933 8.72% -1591 -0.28% -4263 -0.74% -6076 -1.06% 13471 2.35% 387 0.07%
2038 50141 8.75% -1542 -0.27% -4040 -0.71% -6331 -1.10% 13527 2.36% 543 0.09%
2039 50272 8.77% -1478 -0.26% -3826 -0.67% -6576 -1.15% 13562 2.37% 705 0.12%
2040 50277 8.78% -1409 -0.25% -3666 -0.64% -6798 -1.19% 13564 2.37% 842 0.15%
2041 50210 8.76% -1353 -0.24% -3590 -0.63% -6983 -1.22% 13546 2.36% 920 0.16%
2042 50110 8.75% -1276 -0.22% -3547 -0.62% -7176 -1.25% 13519 2.36% 984 0.17%
2043 49977 8.72% -1210 -0.21% -3572 -0.62% -7349 -1.28% 13483 2.35% 1089 0.19%
2044 49842 8.70% -1176 -0.21% -3674 -0.64% -7493 -1.31% 13447 2.35% 1129 0.20%
2045 49756 8.68% -1160 -0.20% -3808 -0.66% -7626 -1.33% 13423 2.34% 1158 0.20%
2046 49608 8.66% -1148 -0.20% -3848 -0.67% -7735 -1.35% 13383 2.34% 1248 0.22%
2047 49311 8.61% -1095 -0.19% -3722 -0.65% -7817 -1.36% 13303 2.32% 1329 0.23%
2048 48966 8.55% -1054 -0.18% -3658 -0.64% -7861 -1.37% 13210 2.31% 1384 0.24%
2049 48649 8.49% -1045 -0.18% -3694 -0.64% -7894 -1.38% 13125 2.29% 1412 0.25%
2050 48337 8.44% -1064 -0.19% -3772 -0.66% -7912 -1.38% 13040 2.28% 1445 0.25%
252
6
Conclusion
In the concluding chapter of this thesis, the research questions formulated in chapter 1 will
be addressed. First the two subquestions will be covered. In section 6.1, the added value of
microsimulation models as a forecasting and policy analysis tool in social security policy
will be discussed. In section 6.2, the conditions that are necessary and favourable for micro-
simulation models to have such added value will be discussed. Policy recommendations from
the chapters 2–5 will be summarized and linked in section 6.3. Topics for future research are
covered in section 6.4. Finally section 6.5 contains an answer to the central research question:
Should microsimulation models should be integrated into the policy analysis, and how?
6.1 Discussion on the added value of microsimulation models
In the previous chapters we have followed the development and use of two dynamic popula-
tion microsimulation models. In chapter 2 a model is presented to forecast the effects of the
introduction of the WIA scheme. This model was further developed after the introduction of
the WIA, as presented in chapter 3, and currently consists of four submodels, one to forecast
254 chapter 6
each of the four different disability schemes now existing in the Netherlands (WAO, WIA,
Wajong and WAZ). In chapters 4 and 5 the SADNAP model is presented and used to study
the budgetary effects and the effects on income redistribution and labour participation of vari-
ous ageing-related policy measures.
The disability benefit model from chapter 2 was used for both policy analyses and budget
forecasts during the years preceding the introduction of the WIA. The budget forecasts from
the simulation model were accepted by the Ministry of Finance as the leading forecasts on
the disability regime change. As we recall from section 1.2, microsimulation models compete
with, among others, cell-based models and general equilibrium models as the tool of choice
in social security forecasting and policy analysis. The Ministry of Social Affairs and Employ-
ment and CPB both used a cell-based forecasting model for disability benefits before the
introduction of the WIA. The complexity of the regime change caused the cell-based models
to fall short of making a consistent forecast with enough detail in outcomes. The complexity
is caused in the first place by the change from one generic law (WAO) to a new law (WIA)
consisting of separate schemes for the permanently fully disabled (IVA) and the partially
and temporarily fully disabled (WGA). In the second place, the tighter entrance criteria for
the WIA cause spillovers to the unemployment scheme and to social assistance. The com-
plexity of those schemes, with additional supplementary arrangements and special schemes
for the older unemployed and partially disabled, further complicates quantification. In the
third place, the WIA includes a part of the benefit in the new public schemes, that before
was privately insured. For these reasons, it was almost impossible adequately to forecast
the individual schemes in a consistent way without turning to microsimulation. Instead of a
cell-based model, CPB used a macro approach to analyse the structural effects of the WIA
(Dekker and Suijker, 2005). The structural forecasts of both approaches cannot be compared
easily because of different assumptions63. What can be concluded is that the macro approach
63 One important difference between both analyses is that the effect of the WIA reform described in chapter 2 in fact
consists of three reforms: the extension of the sickness period from one to two years that has been in force from 1 Janu-
ary 2004 onwards, the stricter examination criteria that have been in force from 1 October 2004 onwards, and the WIA
scheme that has been in force from 1 January 2006 onwards. When discussing the package of all three reforms, we speak
of the broad WIA reform. Originally, however, the stricter examination criteria were intended to be part of the WIA but,
in order to generate additional savings earlier in time, these criteria were implemented earlier. CPB therefore limits its
analysis to the WIA itself.
Conclusion 255
only gives long-term steady-state results that cannot be used for budget forecasts and lack
the internal consistency of the microsimulation approach. What also can be concluded from
comparing chapters 2 and 3 is that the inflow-reducing effects of the WIA are larger than an-
ticipated. In particular, the yearly inflow of permanently fully disabled workers into the IVA
scheme is far less than expected. On the other hand, the inflow of temporarily fully disabled
workers into the WGA scheme is higher than expected.
The disability models from chapter 3 are currently being used for making budget forecasts at
the Ministry of Social Affairs and Employment. Each year, after a new administrative dataset
from UWV becomes available, a new long-term forecast from the base year until 40 years
thereafter is made. Unlike in chapter 2, where microsimulation was used ex-ante to forecast
the effects of a new law, microsimulation is also used in chapter 3 as an ex-post forecast-
ing tool, as the microsimulation forecasts concern existing laws and are compared with
and if necessary adapted to match the known realizations of the year in which the forecasts
are made. The forecasts contain a high level of detail, including various outcomes (inflow,
outflow, number of benefits, number of fde equivalents, mean benefit level per beneficiary or
per fde equivalent, total benefit cost) and various subdivisions per scheme (outflow by cause,
beneficiaries by benefit duration, beneficiaries by working status, benefit cost by fund). The
forecasts are shared with other stakeholders in budget forecasts like the Ministry of Finance,
CPB and UWV. The high level of detail and internal consistency of the numbers contribute to
acceptance of the forecasts by the outside world. An important added value of the models is
in the accurate and detailed forecasts of mean benefit levels. In all disability schemes, benefit
levels follow trends that are very difficult to predict with time-series analysis. One example is
the WAO benefit regime that consists of a higher benefit during the early years of the benefit
and a lower benefit during the later years for all beneficiaries that entered after 1993 and a
permanently high benefit for the beneficiaries that entered in 1993 or before. Another exam-
ple is the WGA benefit that, for the unemployed partially disabled, consists of a high benefit
during the period of unemployment benefit and a much lower benefit thereafter. The cohort
effects of such different regimes very much complicate forecasting a generic time trend.
256 chapter 6
The ageing and pensions model SADNAP, presented in chapters 4 and 5, is used for policy
analyses but is not yet the leading forecasting tool. Budget forecasts are very detailed but
have less added value compared with the cell-based approaches that are currently used by the
Ministry of Social Affairs and Employment, SVB and CPB. In the first place this is due to a
lack of regular updates of the administrative data sources. Unlike in the disability case, where
UWV supplies a yearly update of the micro data, a similar process has not yet been agreed
upon with SVB. The other micro data source for the pension model, on private pensions, is
available from CBS and is regularly updated, but is only available for on-site research and
cannot easily be entered into a microsimulation model outside CBS. Secondly, state pensions
are basically flat rate. There are composition effects that play a role in mean pension levels,
because single persons receive a higher state pension than couples and each year a pensioner
has lived outside the Netherlands reduces the state pension by 2%, but those composition
effects follow trends that are rather predictable in the short run. A clear added value in fore-
casting comes up in long-run budget forecasts, when the composition effects start deviating
from the short-run trend, and in more complex reforms like the introduction of a delayed
retirement credit and the abolition of the partner allowance. Also the recent proposal of trade
unions and employers’ organizations to link the state pension age to the development of life
expectancy and simultaneously increase the yearly indexation of the state pension is of such a
complexity that microsimulation offers added value, in particular when distributional effects
also have to be assessed. However, on other, simpler system reforms, like the raising of the
retirement age, the microsimulation approach in itself in the long run falls short of the general
equilibrium approach. Behavioural effects concerning saving and consumption decisions over
the life cycle start to play an important role here.
Therefore the micro-macro link discussed in section 1.2 remains an important issue. Such a
micro-macro link is implemented in both the disability models and the SADNAP model, but
to a different extent. In both models, the top-down approach is mostly used. This is most no-
tably the case in the SADNAP model, which uses various alignments to exogenous macro ag-
gregates like mortality, fertility, migration, marriage and divorce propensities. The top-down
Conclusion 257
approach is also used in the disability models, where aggregate values on labour participation
by age and gender from a CPB labour market model drive the alignment of inflow numbers.
However, the bottom-up approach has been used here as well, for one iteration, as informa-
tion on the development of the number of disability benefit recipients has been passed back
to CPB for use in the labour market model and consequently the labour market forecasts from
CPB are used as input for new runs of the disability microsimulation models.
When applying the assessment criteria from section 1.2 for forecasting purposes, we need to
distinguish between short-term, medium-term and long-term forecasts. Usually, in the context
of the Ministry of Social Affairs and Employment, the short term is defined as the current
budget year and the next budget year. The medium term consists of all budget forecast years,
which comes down to the current year and the next five years. Long-term forecasts can cover
the whole period until a system reaches a steady state, which in the case of disability ben-
efits lasts for 47 years (benefits can be received from age 18 till 64) and in the case of state
pension benefits at least 35 years (pensions can be received from age 65 onwards). For short
and medium term forecasts and policy analyses, representativeness, outcomes capability
and programmatic detail are especially important and microsimulation models have a clear
advantage over CGE models and over cell-based models on these criteria. In the case of the
WIA reform, only with a high degree of representativeness, a broad outcome capability and
a high level of programmatic detail, it is possible to forecast the replacement of one scheme
(WAO) with a range of other schemes. Clearly, the case for investing in microsimulation
models as a forecasting tool is the strongest for complex social security schemes or reforms.
In such cases the superiority of microsimulation models as a forecasting tool over standard
cell-based models makes the most difference. However, the gains in more accurate forecasts
need to be weighed against the uncertainty that is inherent in the forecasts in the first place.
This explains why microsimulation can be an excellent choice in case of disability insurance
or pensions but less so for unemployment insurance, as the influence of the business cycle is
so important for the latter, even in the short term, that the gains from more refined forecasts
through microsimulation are small compared with the uncertainty margins because of busi-
258 chapter 6
ness cycle fluctuations. On the other hand, current practice at the Ministry of Social Affairs
and Employment shows that macro models that abstract from explicitly defining the flows of
the unemployed into and out of the unemployment benefit scheme, cause inconsistent fore-
casts in the short run. From this perspective, microsimulation models, in which those flows
are explicitly defined, can still considerably improve short-term forecasts.
For long-term forecasts and policy analyses (that usually have a long term or steady state
focus) it remains to be seen whether the advantages of microsimulation outweigh its limita-
tions as compared with CGE models. Other assessment criteria like theoretical and empirical
validity and feedbacks become more important in the long run when behavioural responses,
also at the macro level, become more important. At the same time, microsimulation models
keep certain advantages that are especially important in policy evaluations, like the ability
to investigate distributional effects of policy measures for targeted groups. Good examples
of the latter are immigrants with incomplete state pension entitlements and disability benefit
recipients with a low income. This makes the case for combining microsimulation models
and CGE models to use the strengths of both in one integrated model.
One final note, after discussing various advanced forecasting methods, concerns the limita-
tions of forecasting the distant futures. Usually long-term forecasts, like the ones in this
thesis, start from a baseline scenario of unchanged policies. In the first place, the unchanged
policies scenario is not always clear at first sight and may include policy choices. For ex-
ample, CBO (2010) explicitly presents two baseline scenarios for the USA, one following
current law and one incorporating several changes to current law that are widely expected
to occur or that would modify some provisions of law that might be difficult to sustain for
a long period. Examples of such changes are the indexation of tax rates because when tax
brackets and tax rates are kept constant, like in a strict interpretation of current law, infla-
tion would cause steadily higher average tax rates. For the Netherlands, Van der Horst et al.
(2010) therefore choose to abstract from current law and present one baseline scenario only
including those widely expected changes.
Conclusion 259
More fundamentally, the choice of a baseline scenario assumes the future to be known,
whereas future changes might occur that are fully out of sight in the present. Therefore Van
Asselt et al. (2010) present a conceptual framework that helps to reflect critically on futures
studies. They advocate forecasting (defined as exploring the surprise-free future) only when
it is sensible to assume continuity and stability, whereas in case of uncertainty they advocate
either exploring multiple possible futures (foresighting) or even, when normative consensus
on the future cannot be taken for granted, reasoning backwards in time from one or more
desirable futures to which actions should be taken to realize such a desirable future. Micro-
simulation models can surely add to such foresighting or normative analyses, because the
richness in outcome detail allows for assessing normative issues like income redistribution or
intergenerational distribution of wealth.
6.2 Conditions for successful integration of microsimulation models into

policy analysis
Considering the above, we can answer the question: “Which conditions are necessary for the
development of dynamic microsimulation models and which conditions are favourable?”. This
will shed light on why large dynamic models are used in some countries and not in others.
The most important prerequisite is the availability of good micro data sources that are updat-
ed regularly. This condition explains, in the cases in this thesis, why the disability models are
fully integrated into the policy analysis whereas the SADNAP model is not yet. Also interna-
tionally, the strong position of microsimulation in Norway and Sweden is obviously related to
the long history of collecting automated administrative data in those countries. However, the
Netherlands is catching up quickly in this respect. Since 1998, administrative databases on
most social security schemes have been collected and distributed by CBS. In particular, the
development of the Social Statistical Database (SSB) provides major new opportunities for
researchers. Complete linkable databases are now available for 1998–2007 and the number of
260 chapter 6
characteristics included in the database is rapidly expanding. However, a major disadvantage

of using the Social Statistical database for microsimulation purposes is its accessibility, which
is limited to on-site or remote-access use on a CBS platform with tight privacy regulations
that prevent the generation of output on the individual level for use outside CBS.
In the past, the availability of a powerful computer environment for coping with the heavy
burden microsimulation models impose on computer time and space was also a prerequisite.
Until the 1990s limitations on computer speed and space seriously limited the size of cohorts
or populations that could be used in microsimulation models. However, as both computer
speed and space are rapidly expanding these limitations become less important.
An important factor remains the availability of good staff for developing and maintaining
the models. This point is certainly not trivial. In a way, microsimulation modelling can be
considered a craft more than a science. Users of microsimulation models in a policy environ-
ment tend to have an academic background in, for example, economics, but generally will
not be qualified to develop and maintain a complex simulation model. Developers that are
qualified for building and maintaining such models will generally lack the feeling for the
policy environment in which their models are used. Therefore, it can pay to use standardized
development platforms like the Modgen language (Statistics Canada, 2010) and employ staff
with a strong information technology background and also an interest in working in an eco-
nomic policy environment. A further complication in the case of microsimulation models is
that, from an academic perspective, microsimulation models fall short of general equilibrium
models because of their lack of theoretical validity.
Another important factor concerns the design of the model, which should be modular and
aimed at providing early results during the development process. Failing to supply early
results because of a too grandiose design will cause the vanishing of support and funding for
the development of the model. All the models discussed in this thesis are modularly designed,
but the combination of a comparatively simple design and the supply of early results greatly
Conclusion 261
contributed to the acceptance of the disability models, whereas the more complex design of
the SADNAP model complicated the supply of early results with added value for the policy
analysis.
The market structure of the model market also plays a role. From section 1.3 we learnt that
countries with a competitive market structure, like the USA, the UK, Canada and Australia,
tend to have more microsimulation models in operation than countries with a monopolistic
market structure like the Netherlands. However, not all countries with a competitive market
structure have a history of dynamic microsimulation models. Germany, for example, has
a very competitive market structure but lacks adequate data availability due to very strict
privacy regulations, limiting microsimulation models to smaller-scale static models. Also,
among the countries with large dynamic microsimulation models running are France and
Norway, which have more monopolistic market structures.
Therefore, the institutional context of government agencies involved in economic advice and
data may be a more important factor. This distinction between data collection and economic
advice is taken seriously in the Netherlands. CBS is dedicated to the collection and process-
ing of data and does not do any economic forecasting. CPB is dedicated to economic analy-
ses and does not perform any data collection. CBS distributes numerous micro data sources
to external users like CPB. However, for all users, the use of micro data sources can take
place only under restrictions because of privacy laws. Because traditionally CBS micro data
sources have not been available to CPB and still can be accessed with limitations only, the
development of a large-scale microsimulation model was not likely, neither at CPB or CBS.
In a country like Norway, where the tasks of data collection and economic advice are in one
hand, such a large-scale dynamic model has flourished.
A final factor concerns the product focus of the model market. The strengths of dynamic
microsimulation models (as compared with CGE models) are most prominent when used as
a forecasting tool or as a policy evaluation tool when distributional effects are important. The
262 chapter 6
limitations (again as compared with CGE models) are most prominent when used as a policy
evaluation tool in which behavioural responses are important. Behavioural responses on la-
bour participation decisions can be implemented in microsimulation models but behavioural
responses in a life-cycle context (such as on saving and consumption) are far more difficult to
implement.
6.3 Policy findings on ageing and disability
In this section the main policy findings from this thesis are summarized.
In chapter 2 it is shown in an ex-ante evaluation using microsimulation that the broad WIA
reform was expected to have large effects. These large effects are two-thirds due to the
stricter examinations that formally preceded the WIA, and one-third due to the WIA itself.
The extension of the sickness period from one to two years is considered part of the base-
line scenario. The effects of the extension are considered a transfer of disability costs from
the public scheme to the payroll of employers only. The number of benefits is expected to
decrease by 200,000 (27%) and the cost of the disability benefits is expected to decrease by €
1.9 billion (19%) in the long run as compared with continuation of the WAO scheme with-
out stricter examination criteria. Also, it is shown that because the financial incentives in the
WIA are directed towards participation, labour income of the disabled is expected to rise by
€ 1.7 billion. When this shift from benefits to wages is taken into account, replacement rates
are shown to remain stable on average, and income inequality is shown to decrease slightly.
Therefore, it can be concluded that the WIA reform increases social welfare.
In chapter 3 the individual effects of the stricter entrance criteria and the WIA scheme are
assessed simultaneously in an ex-post evaluation using aggregated micro data from the period
1999–2009. Also, two major earlier reforms, the experience rating scheme and the gatekeeper
law, are taken into account. It is shown that all four of these inflow-related policy measures
Conclusion 263
have large and significant downwards effects on inflow into the disability scheme despite
unfavourable external conditions (growing working population with a rising share of women,
elderly and immigrants, and competing schemes like early retirement and unemployment
also being reformed significantly). Inflow is reduced by 70% because of the four reforms.
In the long run this amounts to 830,000 fewer disability benefit recipients. The gains of the
broad WIA reform are very substantial, and exceed the microsimulation forecasts of chap-
ter 2. In assessing those effects, the extension of the sickness period is taken into account
because this reform had far larger effects than anticipated on reducing inflow. The extension
of the sickness period significantly reduces the number of WIA claims and therefore more
than just transfers costs from the public scheme to the payroll of employers. The combined
effect of experience rating, gatekeeper law and stricter examinations criteria amounts to a far
larger long-term reduction (600,000 less benefits) than that of the WIA scheme (230,000 less
benefits). Moreover, the WIA gains are considerably lower when measured in fully disabled
equivalents (fde) than in single benefits. This confirms the modest effects predicted for the
WIA reform itself in chapter 2 and should not be too surprising, because the gains made by
experience rating, gatekeeper and stricter examinations are so large that the additional effect
of the WIA scheme was bound to be smaller. If the WIA scheme had been implemented first,
the effects might have been larger. But in the end, it can be concluded that the key to reducing
disability numbers was found before the introduction of the WIA itself.
One core conclusion from chapter 3 is that policy measures that aim to reduce inflow are far
more effective than policy measures that aim to increase outflow. This confirms findings in
the international literature (Autor and Duggan, 2006). In fact, it seems that the earlier in the
sickness process the policy measure intervenes, the greater the effect. This leaves limited pos-
sibilities for further policy improvements, although some gains can still be made by recon-
sidering the level of sickness benefits, which is very high by international standards (usually
100% of the wage during the whole first year of sickness) and the position of beneficiaries
without an employer (mainly flexible workers and unemployed) who make up a large share
of the current WIA inflow.
264 chapter 6
When someone is on disability benefit for a longer period of time, it is very difficult to
reverse that process. In the first place, this is because the contact with the labour market and
an actual employer is vital in returning to the labour market; hence the effectiveness of the
gatekeeper law. Secondly, policy measures that are effective later on in the process of sick-
ness and disability tend to be more controversial than measures that are effective in the early
stages. The failing of the introduction of the lower second-phase disability benefit in 1994
and the controversy over the re-examination procedures of 1994 and 2005 are clear examples.
A striking result from chapter 3 is the ambiguous effect of the re-examinations of the dis-
ability stock. While clear gains were made among those who were re-examined, the recovery
rates of those who were not re-examined fell sharply. Whether this is caused by the abolition
of the periodic re-examinations system of this group is not clear yet. It can be concluded that
effective policy combines intervening as early as possible in the process of sickness with ef-
fective re-examinations of those that enter the disability schemes.
Although the WAO/WIA schemes for employees are quickly returning to normal propor-
tions, the rapid growth of the number of Wajong beneficiaries is worrying. In the past the
WAO was used to accommodate the social changes of transition from an industrial society
to a service oriented society and from a traditional breadwinner society to a dual earners
society (De Jong, 2008). It seems that the Wajong scheme currently is being used to accom-
modate another social change, namely the change towards an information or network society
and a knowledge economy. Such a society is very well suited for educated people with broad
networks who are able to function in a flexible way. However, as such a society becomes
increasingly complex, the demand for workers with low or no education to do the simplest
of jobs may decrease, causing increasing competition for such jobs. People with the lowest
abilities, including those with congenital mental or physical disorders that limit their capaci-
ties to performing those simplest of jobs, may be the first to fall victim to such a society.
Using the lessons from successful policy measures of the past, it is clear that policy meas-
ures that intervene as early as possible in the process of disability are the most effective in
reducing the number of disability benefits. In the case of the Wajong, where most disabilities
Conclusion 265
are congenital, early intervention means intervening in the special schools for the mentally
disabled that should focus on preparing their students for the labour market. The employers in
return, should provide enough permanent jobs for the low-skilled young handicapped, as well
as allowing some of their jobs to be tailored to the needs of the higher-skilled young handi-
capped. This can be achieved by enforcing quotas on the percentage of young handicapped
people employed by larger employers, or by sufficiently high wage subsidies, allowing the
young handicapped to be paid wages below the minimum wage in order to be competitive on
the labour market.
In chapters 4 and 5 the effects are shown of various ageing-related policy measures as com-
pared with a baseline scenario of unchanged policies. In the baseline scenario, it is shown
that in the long run changes in composition of the pensioner population, like the initially
decreasing and later increasing proportion of single persons and the rising proportion of im-
migrants and emigrants with incomplete state pensions, have considerable budgetary effects.
Also, in the baseline scenario it is shown that intra-generational redistribution within the state
pension scheme is substantial. Therefore, care should be taken when considering proposals
for differentiation of retirement age by income or education level. Whereas such proposals
may decrease redistribution from lower to higher incomes or from the lower to the higher
educated because of differential mortality, they may cause other forms of redistribution, for
instance, from men to women, to increase. Moreover, such a differentiation by education
level or income will decrease actuarial fairness of the state pension scheme and provide new
incentives for early retirement. Also, a classification based on income seems to be arbitrary
as differentiation by gender or ethnic background could (from the argument of differential
mortality) just as well be used to differentiate the retirement age. Clearly, such differentia-
tions will not be viable.
One clear finding when researching the option value parameters used in the microsimula-
tion model is that the heterogeneity in those parameters is very large, especially in discount
rate and leisure preference rate. Such a large difference in preferences implies that a pension
266 chapter 6
system that is as flexible as possible is preferable to a static system with a fixed retirement
age and pension level. The second pillar in the Dutch pension system provides such flexibil-
ity. The first pillar hitherto does not, but may be adapted to provide flexible use, although the
size of the second pillar in the Netherlands is so large that the necessity of making the first
pillar more flexible remains questionable. However, when making the first-pillar system more
flexible, the government can use the yearly accrual rate in case one works for one more year
to influence the retirement decision for labour participation purposes. A high accrual rate will
effectively increase the retirement age, although the costs of a higher accrual rate (in fact a
late retirement bonus) should be weighed against the gains from higher labour participation,
especially against the background of the large number of tax benefits for elderly workers
already in force in the Netherlands (see Euwals, de Mooij and van Vuuren, 2009).
One proposal that might both reduce intragenerational redistribution and increase social
welfare is individualization of the state pension, possibly in combination with a delayed
retirement credit. Individualization decreases the pension of single persons, who may be
more prone to working on longer and who thus may be more likely to gain from late retire-
ment bonuses. Of course, in such a system a safety net should be provided for those with no
second-pillar pension income.
6.4 Directions for future research
In this section the main directions for future research that follow from this thesis are summa-
rized.
One major research question on disability insurance that needs further investigation concerns
the role of financial incentives. In section 2.6 it is noted that the average value of the benefit
elasticity (of 1.5) used by CPB for the Netherlands is questionable as is the assumption of
homogeneity that follows from the macro approach. The value of 1.5 is deduced from Gruber
Conclusion 267
(2000), who finds a labour force non-participation elasticity of between 0.28 and 0.36 for a
Canadian “natural experiment” in which the benefit levels during the late 1980s temporarily
differed between Quebec and the rest of Canada. As about one-fifth of the non-participants
were on disability benefits, the benefit elasticity of 1.5 was deduced (see CPB, 2010). How-
ever, other research points to a wide range of benefit elasticity values. Campolieti (2004)
estimates the effect of disability benefits on labour supply using earlier Canadian data from
the early 1970s and finds no proof at all (so in fact an elasticity of 0) for the assumption that
increases in disability benefits lead to increases in non-participation. Bell and Smith (2004)
directly measure a benefit elasticity of 0.5 for the UK. Brinch (2009) finds a benefit elasticity
of 4.5 for Norway. Van Vuren and Van Vuuren (2007) estimate a benefit elasticity of between
2.5 and 3.0 for the Netherlands during the late 1990s.
There seems to be a clear pattern in which the benefit elasticity is higher in countries or in
times with less strict entrance criteria to the scheme. Indeed, Campolieti (2004) concludes
that “the effect of disability benefits on labour supply decisions will depend on the underly-
ing screening stringency for benefits” and Brinch (2009) concludes that: “The ‘marginal’
disability pensioner (with respect to stringency of requirement) should be more responsive
to economic incentives than the average disability pensioner because the pool of pensioners
include a substantial number of persons that are so disabled that they are in practice not able
to work”. In theory, when entrance examinations are so strict that entrance is limited to those
that are not at all able to work, as in the current IVA scheme in the Netherlands or during the
early 1970s in Canada, benefit elasticity should approach 0. This implies that benefit elastic-
ity theoretically can be written as a function of the stringency of the scheme.
Microsimulation models allow us to introduce heterogeneity in benefit elasticities. How-

ever, the current literature on benefit elasticities lacks detail on heterogeneity. Although Van
Vuuren and Van Vuren (2007) tested for heterogeneity between fully and partially disabled
claimants and found no difference, this does not mean that heterogeneity can be neglected.
In a scheme with very loose entrance criteria, as the WAO scheme used to be, the majority of
268 chapter 6
even fully disabled persons were able to work to some extent. Simple calculations underline
this. In 1999, 2000 and 2001 cohorts of about 100,000 new benefit recipients per year were
allowed in the WAO, about 60% of them fully disabled. Between 2006 and 2009, the yearly
number of fully disabled allowed into the WIA scheme was about 20,000, with only 5,000
classified as permanently fully disabled. Apparently, in a scheme with such loose entrance
criteria as the WAO, so many fully disabled may still have labour capacity that indeed, a dif-
ference in benefit elasticity at the margin cannot be expected anymore. However, the simple
fact that the fully disabled by definition include those without any labour capacity implies
that under maximally strict entrance criteria the benefit elasticity of the fully disabled ap-
proaches 0 while the benefit elasticity of the partially disabled may be much higher. Also, a
difference in benefit elasticities between men and women is plausible as major differences
are known from the literature on labour supply elasticities (for an overview, see Evers, De
Mooij and Van Vuuren, 2005). Furthermore, differences in benefit elasticity between differ-
ent income levels seem plausible as well. All in all, the case for further investigations seems
evident.
A final comment on benefit elasticities concerns the assumed linearity of the elasticity under
different external shocks. For example, a benefit increase may have different consequences
than a benefit decrease of the same size. Also, the applicability of benefit elasticities under
very large benefit changes is questionable. Such policy proposals are not rare at all. Rather
than opting for a common decrease of benefit levels by a small percentage, in certain reforms
policy makers may opt to decrease the benefit level for certain specific groups by a very large
amount (or even exclude those groups from benefits at all). Recent examples of such propos-
als are numerous and include the introduction of a means test in the WIA and Wajong schemes
and the decrease of the Wajong benefit for students from 75% of the minimum wage to 25%.
Theoretically, when assuming linear benefit elasticity, an elasticity of 1.5 would reduce the
number of beneficiaries to zero when benefits are decreased by two-thirds. Clearly, the benefit
elasticity can no longer be used in such situations. Apparently the benefit elasticity is not lin-
ear but can theoretically be written as a non-linear function of the income shock as well.
Conclusion 269
Chapter 3 discusses the separate and simultaneous effects of different disability-related policy
measures. One of the more striking results with respect to the long-term forecasts of the num-
ber of disability benefit recipients is the decrease in the WIA scheme being almost entirely
compensated by the increase in the Wajong scheme. One could suspect that some kind of
substitution takes place between the two, although common sense says that the group targeted
by the Wajong scheme (persons without a labour market history) is quite different from the
group targeted by the WIA scheme (persons who by definition do have a labour market his-
tory). However substitution can theoretically take place in two ways. First, because people
who fail in the labour market and end up receiving unemployment benefit or social assistance
or without any benefit can enter the Wajong scheme at later age as long as they can prove
having had a disability from age 18 onwards. Second, by knowing the increased difficulty of
entering the WIA scheme at a later age, incentives to enter the Wajong scheme at an earlier
age may have increased. Future research should further investigate possible substitution
between the WIA and Wajong schemes.
Although the importance of incentives for beneficiaries is beyond discussion, in chapter 3 we

have shown how incentives for employers have proven to be very effective as well. Howev-
er, financial incentives for employers have proven to be effective in particular sectors whereas
legislative incentives for employers have proven to be more effective in other sectors. Further
research should clarify how to target incentives for employers, not only in disability insur-
ance but also in unemployment insurance and in policies to employ the elderly. The latter are
particularly important because of the ageing-related policy measures discussed in chapters 4
and 5 that include raising the retirement age and introducing a delayed retirement credit.
In the SADNAP microsimulation model from chapters 4 and 5, heterogeneity is incorporated

in parameters related to the retirement decision. These parameters are based on plausible
estimates from current literature as actual estimates are still lacking for the Netherlands. The
availability of the first large dataset on second-pillar pension entitlements (for 2005) provided
the opportunity for creating a database of both first- and second-pillar pension entitlements
270 chapter 6
for use in a microsimulation model for the first time. Meanwhile similar datasets for 2006
and 2007 have been created and datasets for more recent years will follow. This will allow for
the first time research on the retirement decision of a large part of the Dutch population of el-
derly, taking into account essential parameters on income and wealth. Thus actual estimation
of the option value parameters for the Netherlands may be possible. Special attention needs
to be paid to the cohabitation status of individuals as it is known that the retirement decision
is often a joint decision of partners. In such a joint decision, the income and age difference
between partners may play an important role. The partner earning the highest income may be
more sensitive to financial incentives and the younger partner may prefer earlier retirement.
This can explain why men are more sensitive to financial incentives than women and why
women tend to prefer earlier retirement than men.
One promising direction for future research on ageing-related policies concerns the inter-
generational aspects of the Dutch pension system or even of the whole system of govern-
ment taxes and transfers. These intergenerational aspects are underdeveloped and are usually
studied with overlapping generations CGE models or representative agent microsimulation
models, most recently in Van der Horst et al. (2010). Efforts to use dynamic microsimulation
models for studying intergenerational aspects in detail are still scarce but are slowly gaining
ground. They require an adequate modelling of wage profiles over the career, something still
underdeveloped in SADNAP, and also backwards simulating the life paths of the starting
population (see, for example, Pettersson (2006) who used national accounts for “backcasting”
life paths until 1930). However, only such an approach will give an adequate insight into the
actuarial fairness of the schemes, both within and between generations. Moreover, the role
of inheritance of wealth of younger generations can be taken into account, a topic which has
up till now rarely been studied, but which may prove to be important in the assessment of
intergenerational fairness of the baseline situation and policy measures.
A final direction for future research concerns the development of an integrated micro-macro
model of the Dutch labour market. Currently, in the Netherlands there are lots of smaller ad-
Conclusion 271
hoc models, be it microsimulation models, cell-based models or simple spreadsheet models

that are used for forecasting purposes. These models may share some macro inputs from
general equilibrium models of CPB, but there are no feedback loops and there is no standard-
ization. This situation is not satisfactory because the quality and validity of so many ad-hoc
models is questionable. This is illustrated by the scarce microsimulation initiatives that are
operational in the Netherlands, which generally lack standardization and linkage to the GE
models of CPB. A radical solution to the above problem would be the development of an
all-purpose integrated microsimulation-CGE model central to government forecasts. Such a
model would represent the whole population including the young, workers, benefit recipients
and the retired. Such a microsimulation model needs not be aligned to macro forecasts as it is
the core of the forecasting system itself. The all-purpose microsimulation-CGE model would
require the responsible institution, probably CPB, to develop a large-scale microsimulation
model and integrate it to a CGE. It needs to have full access to the relevant micro data. This
is not likely in the short run, given the investments done in CGE modelling already and the
separation between the data function at CBS and the modelling function at CPB.
Therefore, having seen the added value of the macro-micro linkage, a compromise solution
might be considered. In such a modelling environment, a CGE model of CPB remains the
core of the economic modelling industry. However, a larger number of specific microsimula-
tion models can be linked to this CGE using a top-down bottom-up approach of communi-
cation with the CGE model. Such models do not need to be developed by CPB but can be
developed and operated in different institutions, although a common development platform
is preferable. Models should be transparent and there needs to be a rigorous quality check of
CPB on the models before allowing them to be part of the modelling structure. On the other
hand, there needs to be a complete and open provision of CGE outputs to be used as inputs to
the microsimulation models and a clear common specification for the required output formats
of those simulation models so that their outputs can be used as inputs again to the CGE
model. Logical components of such a system would be the following:
272 chapter 6
• A general equilibrium model of the Dutch economy, based at CPB.

• A demographic model of the Dutch population, based at CBS, which follows the long-
term population projection of CBS currently used in many Dutch economic models.
• A labour supply model, compliant with the CGE model, so preferably also based at
CPB, which combines the population projection with (at least) age- and gender-specific
labour participation rates and information on wages and hours worked.
• Microsimulation models on social security programmes, preferably based at the Ministry
of Social Affairs and Employment, including a disability insurance model, possibly an
unemployment insurance model, a social assistance model, a survivor benefit model and
a model on state and private pensions.
• A tax model, preferably based at the Ministry of Finance.
The above framework is limited to social-economic policy, but other fields can be integrated.
If so, the role of other institutions that are active in model building in their respective special-
isms should be considered. In the first place this concerns institutions like the Dutch Cen-
tral Bank (DNB), the Netherlands Institute for Social Research (SCP) and the Netherlands
Environmental Assessment Agency (PBL). CPB, SCP and PBL together are the three applied
policy research institutes of the Dutch government. DNB currently operates the macroecono-
metric model MORKMON, SCP operates a health forecasting model and a forecasting model
on the public sector, and PBL operates regionally specified population and housing models.
Other institutions that might be considered include the Research and Documentation Centre
of the Ministry of Justice (WODC) that operates crime-related models, the Netherlands Inter-
disciplinary Demographics Institute (NIDI) that operates, among others, a firm-level micro-
simulation model and the research institute TNO that operates the spatial general equilibrium
model RAEM.
The gains from this compromise solution would mainly come from consistency in the as-
sumptions used in the various models and in two-way linkage (following the top-down
bottom-up approach) of the CGE models and the microsimulation components. Moreover,
Conclusion 273
because model-building institutes need to cooperate, discussion between these institutes can
gain in terms of consistency.
6.5 Concluding remarks
In the final section of this thesis, the central research question – “Should microsimulation
models be integrated into the policy analysis, and how?” – is revisited. Based on the cases
investigated in chapters 2–5, it can be concluded that microsimulation models should be
integrated into the policy analysis. Microsimulation models have a clear added value in
some important respects of both forecasting and policy analysis. In forecasting, the ability
of microsimulation models to provide consistent forecasts for the short, medium and long
term with a high level of detail in outcomes gives microsimulation models a clear edge as
compared with competing modelling techniques. The more complex a scheme is, in terms of
system complexity, the higher the added value of microsimulation models. In policy analy-
ses, microsimulation models allow for assessing distributional effects of policy measures,
which are of great importance in the policy decision-making process and in welfare econom-
ics in general. This allows for a broader assessment of policy measures which currently is
often limited to budgetary effects and income effects for some representative households at
most. However, microsimulation models lack the theoretical validity and the macroeconomic
feedback mechanisms of general equilibrium models and should not replace them. Therefore,
there is a strong case for integrating microsimulation models and general equilibrium models
into a consistent modelling environment, in order to combine the strengths of both modelling
techniques.
In order for microsimulation models to flourish, certain conditions have to be met. The most
important prerequisite is the availability of accessible, regularly updated, high quality data
sources for use as input to the models. Preferably, the development of microsimulation mod-
els should follow a modular approach in a standardized development platform with adequate
274 chapter 6
staff. Although the institutional context of the Netherlands in some respects complicates the
development of large dynamic microsimulation models, the integration of different models
from different institutes into a single modelling framework, with a central role of CPB, is a
long-term perspective that is worth considering.
References
Aarts, L., de Jong, Ph. (1992), ‘Economic aspects of disability behavior’, Amsterdam: North-Holland.
Aarts, L., de Jong, Ph., van der Veen, R. (2002), ‘Met de beste bedoelingen: WAO 1975-1999, Trends, onderzoek en
beleid’, The Hague: Elsevier Bedrijfsinformatie.
Advies Commissie Arbeidsparticipatie (2008), ‘Naar een toekomst die werkt’, CPB Notitie 2008/20 Effecten van
participatiebeleid, 16 juni 2008.
Alessie, R., Lusardi, A., Aldershof, T. (1997), ‘Income and Wealth over the Life Cycle, Evidence from Panel Data’,
Review of Income and Wealth, Vol. 43, No. 1, March 1997, pp 1-32.
Arntz, M., Boeters, S., Güertzgen, N. (2006), ‘Alternative Approaches to Discrete Working Time Choice in an AGE
Framework’, Economic Modelling, Vol. 23, pp. 1008-1032.
Asselt, M.B.A. van, Faas, A., Molen, F. van der, Veenman, S.A. (2010), ‘Uitzicht, Toekomstverkennen met Beleid’,
Wetenschappelijke Raad voor het Regeringsbeleid, Amsterdam Univeristy Press.
Autor, D.H., Duggan, M.G. (2006), ‘The Growth in the Social Security Disability Rolls: A Fiscal Crisis Unfolding’,
Journal of Economic Perspectives, Vol. 20, No. 3, pp. 71-96.
Bacon, B., Pennec, S. (2007), ‘APPSIM – Modelling Family Formation and Dissolution’, NATSEM Working Paper
No. 4, November 2007.
Baroni, E. (2007), ‘Pension Systems and Pension Reform in an Aging Society. An Introduction to the Debate’, Institute
for Futures Studies, Arbetsrapport 2007:6.
Barr, N.A. (1992), ‘Economic theory and the welfare state: a survey and interpretation’, Journal of Economic litera-
ture, Vol. 30, No. 2, pp. 741-803.
Barten, A.P. (1986), ‘The History of Dutch Macroeconomic Modelling (1936-1986)’, In W. Driehuis, M.M.G. Fase, H.
Den Hartog (Eds.) Challenges for macroeconomic modelling, Amsterdam, North-Holland, pp. 39-88.
Bell, B., Smith, J. (2004), ‘Health, disability insurance and labour force participation’, Bank of England Working
Paper 218.
Benitez-Silva, H., Disney, R., Jimenez-Martin, S. (2010), ‘Disability, capacity for work and the business cycle: An
international perspective’, Economic Policy, Vol. 25, No. 63, pp. 483-536.
Berendsen, E., Mulders, H., Loo, J. van (2007), ‘Het eerste jaar WIA’, Economisch Statistische Berichten, Vol. 92,
No. 4505, pp. 132-135.
276
Berghman, J., Nagelkerke, A., Boos, K., Doeschot, R., Vonk, G. (2003), ‘Honderd Jaar Sociale Zekerheid in Neder-
land’, Delft: Eburon.
Berkel, B., Börsch-Supan, A. (2004), ‘Pension Reform in Germany: the Impact on Retirement Decisions’, FinanzAr-
chiv, Vol. 60, No. 3, pp. 393-421.
Besseling, J. (2002), ‘WAO: niet alle oude schoenen zijn slecht’, Tijdschrift voor Bedrijfs- en Verzekeringsge-
neeskunde, Vol. 10, No. 4.
Bloemen, H.G. (2008), ‘Private Wealth and Job Exit at Older Age: A Random Effects Model’, Tinbergen Institute
Discussion Paper TI 2008-025/3.
Blundell, R. (1992). ‘Labour Supply and Taxation: A Survey’, Fiscal Studies, Vol. 13. No. 3, pp. 15-40.
Borella, M., Coda Moscarola, F. (2005), ‘Distributive Properties of Pension Systems: A Simulation of the Italian Tran-
sition from Defined benefit to Defined Contribution’, CERP Working Paper 42/05.
Börsch-Supan, A. (2000), ‘Incentives of Social Security on Labour Force Participation: Evidence in Germany and
across Europe’, Journal of Public Economics, Vol. 78, pp. 25–49.
Börsch-Supan, A., Schnabel, R., Kohnz, S., Mastrobuoni, G. (2004), ‘ Micro Modelling of Retirement Choices in Ger-
many’, in: J. Gruber and D. Wise (Eds.), Social Security Programs and Retirement around the World, Chicago,
pp. 285-343.
Bos, V., Kunst, A., Keij-Deerenburg, I.M., Garssen, J., Mackenbach, J.P. (2004), ‘Ethnic inequalities in age- and cause-
specific mortality in the Netherlands’, International Journal of Epidemiology, Vol. 33, No. 5, pp. 1112-1119.
Bourguignon, F., Spadaro, A. (2006), ‘Microsimulation as a Tool for Evaluating Redistribution Policies’, Journal of
Economic Inequality, Vol. 4, No. 1, pp. 77-106.
Bovenberg, A.L., Ter Rele, H. (1999), ‘Generational Accounts for the Netherlands’, in A. Auerbach, L. Kotlikoff en W.
Leibfritz (Eds.), Generational Accounting around the world, The University of Chicago Press.
Bovenberg, A.L., Knaap, T. (2005), ‘Ageing, Funded Pensions and the Dutch Economy’, CESIFO Working Paper No.
1403, Category I: Public Finance.
Bovenberg, A.L., Gradus, R.H.J.M. (2008), ‘Dutch Policies Towards Ageing’, European View, Vol. 7, pp. 265–275.
Brinch, C.N. (2009), ‘The effect of benefits on disability uptake’, Statistics Norway, Research Department Discussion
Papers No. 576.
Bronswijk, E., Meester, R., Hosemans, M., Filius, A. (2003), ‘Voorstudie Vrouwen in de WAO’, Stecr Studierapport.
Buddelmeyer, H., Freebairn, J., Kalb, G. (2006), ‘Evaluation of Policy Options to Encourage Welfare to Work’, Aus-
References 277
tralian Economic Review, Vol. 39, No. 3, pp. 273-292.
Burg, C. van der, Prins, R. (2010), ‘Employment instead of benefit receipt? Process and outcomes of reassessment of
Dutch disability benefit recipients’, European Journal of Social Security, Vol. 12, No. 2, pp. 144-155.
Butter, F.A.G. den (1991), ‘Macroeconomic modelling and the policy of restraint in the Netherlands’, Economic Mod-
elling, Vol. 8, pp. 16-33.
Butter, F.A.G. den (2010), ‘The industrial organisation of economic policy preparation in The Netherlands’, in P. Wein-
gart en J. Lentsch (Eds.), Between Science and Policy: Quality Control and Scientific Policy Advice, Cambridge
University Press, Cambridge, pp. 227-284.
Butter, F.A.G. den, Kock, U. (2003), ‘Social security, economic growth and poverty ‘, Ch. 14 in K. Marshall and O.
Butzbach (Eds.), New Social Policy Agendas for Europe and Asia: Challenges, Experiences, Lessons, World
Bank, Washington DC, pp. 249-263.
Butter, F.A.G. den, Morgan, M. (1998), ‘What makes the models-policy interaction successful?’, Economic Modelling,
Vol. 15, pp. 443-475.
Butter, F.A.G. den, Sonsbeek, J.M. van (2008), ‘Opties voor flexibilisering van de AOW’, Economisch Statistische
Berichten, Vol. 93, pp. 742-745.
Caldwell, S., Morrison, R.J. (2000), ‘Validation of longitudinal dynamic microsimulation models: experience with
CORSIM and DYNACAN’, Ch. 9 in L. Mitton, H. Sutherland and M.J. Weeks (Eds.), Microsimulation model-
ling for policy analysis: challenges and innovations, Cambridge University Press.
Caminada, K., Goudswaard, K.P., Van Vliet, O.P. (2010), ‘Patterns of Welfare State Indicators in the EU: Is there
Convergence?’, Journal of Common Market Studies, Vol. 48, No. 3, pp. 529-556.
Campolieti, M. (2004), ‘Disability Insurance Benefits and Labor Supply: Some Additional Evidence’, Journal of
Labor Economics, Vol. 22, No. 4, pp. 863–889.
Cassels, R., Harding, A., Kelly, S. (2006), ‘Problems and Prospects for Dynamic Microsimulation: A Review and Les-
sons for APPSIM’, NATSEM Discussion Paper no. 63.
CBS (2007), ‘Bevolkingstrends, 1e kwartaal 2007’.
CBS (2008), ‘Streefpensioen van 70 procent niet voor iedereen haalbaar’, Webmagazine vrijdag 27 juni 2008.
CBS (2009), ‘Bevolkingstrends, 1e kwartaal 2009’.
Christoffel, K., Coenen, G., Warne, A. (2010), ‘Forecasting with DSGE models’, European Central Bank, Working
Paper Series No. 1185.
278
Clauss, S., Schubert, M. (2009), ‘The ZEW Combined Microsimulation-CGE Model: Innovative Tool for Applied
Policy Analysis’, ZEW Discussion Paper No. 09-062.
Clemen, R.E. (1989), ‘Combining forecasts: A review and annotated bibliography‘, International Journal of Forecast-
ing, Vol. 5, No. 4, pp. 559-583.
Coile, C., Gruber, J. (2001), ‘Social security incentives for retirement.’, In Themes in the Economics of Aging, ed. D.
Wise, pp. 311–341. Chicago: University of Chicago Press.
Coile, C. (2004), ‘Retirement Incentives and Couples’ Retirement Decisions’, Topics in Economic Analysis & Policy,
Vol. 4, No. 1, Article 17.
Colombo, G. (2010), ‘Linking CGE and Microsimulation Models: A Comparison of Different Approaches’, Interna-
tional Journal of Microsimulation, Vol. 3, No. 1, pp. 72-91.
CBO (2010), ‘The Long-Term Budget Outlook June 2010’, The Congress of the United States - Congressional Budget
Office.
CPB (2001), ‘Analyse van de voorstellen van de Adviescommissie Arbeidsongeschiktheid’, CPB-Notitie 01/16.
CPB (2002), ‘Analyse van het SER-advies: werk maken om arbeidsongeschiktheid te voorkomen’, CPB-notitie 02/35.
CPB (2007), ‘Actualisatie Economische Verkenning 2008-2011’, CPB document 151.
CPB (2009a), ‘Budgettaire en economische effect AOW-plan VVD’, CPB notitie 7 januari 2009.
CPB (2009b), ‘Houdbaarheidseffect voorstel verhoging AOW-leeftijd’, CPB notitie 4 december 2009.
Creedy, J., Kalb, G. (2005), ‘Behavioural Microsimulation Modelling for Tax Policy Analysis in Australia: Experi-
ences and Prospects’, Australian Journal of Labor Economics, Vol. 8, No. 1, pp. 73-110.
Davies, J.D. (2009), ‘Combining Microsimulation with CGE and Macro Modelling for Distributional Analysis in De-
veloping and Transition Countries’, International Journal of Microsimulation, Vol. 2, No. 1, pp. 49-65.
Dekker, P., Suijker, F. (2005), ‘Structurele effecten van de WIA’, CPB Memorandum No. 114.
Dekkers, G.J.M. (2007), ‘The Financial Implications of Working Longer: An Application of a Micro-Economic Model
of Retirement in Belgium’, International Journal Of Microsimulation, Vol. 1, No. 1, pp. 10-25.
Dekkers, G., Belloni, M. (2009), ‘Micro simulation, pension adequacy and the dynamic model MIDAS: an introduc-
tion.’, Project AIM - Deliverable 4.10.
Deursen, C.G.L. van, Schellekens, E.I.L.M., Huurne, A.G. ter, Fux, R.A.P., Veerman, T.J., Zwol, W.I. van (1997),
References 279
‘ZARA-werkgeverspanel rapportage 1997’, Ministerie van Sociale Zaken en Werkgelegenheid.
Deursen, C. van, Mulders, H. (2009), ‘Schatting effect aangepaste Schattingsbesluit op aandeel afwijzingen WIA’,
UWV Kennismemo 09-13.
Deursen, C. van, Loo, J. van (2010), ‘Analyse stijging WIA-instroom II’, UWV Kennismemo 10-07.
Disney, R. (2004), ‘Are contributions to public pension programmes a tax on employment?’, Economic Policy, Vol.
19, No. 39, pp. 267–311.
Don, H., Verbruggen, J. (2006), ‘Models and methods for economic policy; 60 years of evolution at CPB’, Statistica
Neerlandica, Vol. 60, No. 2, pp. 145-170.
Donders, J., Graafland, J. (1998), ‘CPB models and employment policy in the Netherlands’, Economic Modelling,
Vol. 15, No. 3, pp. 341-356.
Draper, N., Nibbelink, A., Bonenkamp, J., Rosenbrand, R. (2005), ‘Description Gamma Model, version pension
study’, CPB Memorandum 115.
European Commission (2007), ‘Pensions Schemes and Projection Models in EU-25 Member States’, European Econ-
omy Occasional Papers Number 35.
Euwals, R., van Vuuren, D., Wolthoff, R. (2006), ‘Early Retirement Behaviour in the Netherlands: Evidence from a
Policy Reform’, IZA discussion paper No. 1992.
Euwals, R., Erp, F. van, Hek, P. de, Rele, H. ter (2008), ‘Effect van verhogen pensioengerechtigde leeftijd op bruto
participatie’, CPB Memorandum No. 198.
Euwals, R., Folmer, K. (2009), ‘Arbeidsaanbod en gewerkte uren tot 2050: een beleidsneutraal scenario’, CPB Memo-
randum No. 225.
Euwals, R., Mooij, R. de, Vuuren, D. van (2009), ‘Rethinking Retirement. From participation towards allocation’,
CPB Book.
Evers, M., Mooij, R. de, Vuuren, D. van (2005), ‘What Explains the Variation in Estimates of Labour Supply Elastici-
ties?’, CPB Discussion Paper no. 51.
Ewijk, C. van, Draper, N., Rele, H. ter, Westerhout, E. (2006), ‘Ageing and the Sustainability of Dutch Public Fi-
nances’, CPB Book.
Finkelstein, A., Poterba, J. (2004), ‘Adverse Selection in Insurance Markets: Policyholder Evidence from the UK An-
nuity Market’, Journal of Political Economy, Vol. 112, No. 1, pp. 183-208.
280
Frederick, S., Loewenstein, G., O’Donoghue, T. (2002), ‘Time Discounting and Time Preference: A Critical Review’,
Journal of Economic Literature, Vol. 40, No. 2, pp. 351–401.
Garcia Gomez, P., Von Gaudecker, H.M., Lindeboom, M. (2010), ‘Health, Disability and Work: Patterns for the Work-
ing Age Population’, Netspar Panel Paper 17.
Gilbert, N., Troitzsch, K.G. (1999), ‘Simulation for the social scientist’, Open University Press.
Goudswaard, K.P., de Kam, C.A., Sterks, C.G.M. (2000), ‘Sociale zekerheid op het breukvlak van twee eeuwen’,
Samsom/Kluwer, Alphen a/d Rijn.
Gradus, R., Beetsma, R., Smit, R. (2010), ‘Op weg naar houdbare overheidsfinanciën’, Economisch Statistische Beri-
chten, Vol. 95, pp. 70-73.
Gruber J. (2000), ‘Disability Insurance Benefits and Labor Supply’, Journal of Political Economy, Vol.108, No. 6,
pp. 1162-1183.
Gustman, A., Steinmeier, T. (2005), “The Social Security Early Entitlement Age in a Structural Model of Retirement
and Wealth”, Journal of Public Economics, Vol. 89, pp. 441-463.
Hek, P. de, Erp, F. van (2009), ‘Analyzing labour supply of elderly people, A life-cycle approach’, CPB document
No. 179.
Hendrix, P.C.M. (1993), ‘Statische en dynamische microsimulatie, Toepassingen in de praktijk’, Ministerie van So-
ciale Zaken en Werkgelegenheid, Onderzoeksmemorandum AEI No. 93/04.
Henkens, K., Grift, Y., Siegers, J. (2002), ‘Changes in female labour supply in the Netherlands 1989-1998: The case of
married and cohabiting women’, European Journal of Population, Vol. 18, pp. 39-57.
Horst, A. van der, Bettendorf, L., Draper, N., van Ewijk, C., de Mooij, R., ter Rele, H. (2010), ‘Vergrijzing verdeeld.
Toekomst van de Nederlandse Overheidsfinanciën’, No. 86, June 2010.
Huizinga, F., Smid, B. (2004), ‘Vier vergezichten op Nederland, productie, arbeid en sectorstructuur in vier scenario’s
tot 2040’, CPB No. 55.
Jehoel-Gijsbers, G. (2007), ‘Beter aan het werk. Trendrapportage ziekteverzuim, arbeidsongeschiktheid en werkher-
vatting’, SCP-publicatie 2007/22.
Jong, A. de (2002), ‘Gehuwden leven het langst’, CBS-webmagazine, 2002a.
Jong, Ph. de, Thio, V. (2002), ‘Donner versus Veldkamp, over nut en nadeel van gedeeltelijke WAO-uitkeringen’,
Tijdschrift voor Bedrijfs- en Verzekeringsgeneeskunde, Vol. 11, pp. 169-172.
References 281
Jong, Ph. de, Thio, V., Bartelings, H. (2005), ‘UWV als poortwachter. Fase III: WVP en de instroom in de WAO’, Ape,
September 2005.
Jong, Ph. de (2008), ‘Recent changes in Dutch disability policy’, Ape, September 2008.
Jong, Ph. de, Lindeboom, M., van der Klaauw, B. (2010), ‘Screening Disability Insurance Applications’, Journal of the
European Economic Association, forthcoming.
Kalmijn, M., Alessie, R. (2008), ‘Life course changes in income: An exploration of age- and stage effects in a 15-year
panel in the Netherlands’, Netspar Panel Paper 10.
Kalwij, A., Alessie, R., Knoef, M. (2009), ‘Individual Income and Remaining Life Expectancy at the Statutory Retire-
ment Age of 65 in the Netherlands’, Netspar Paper, September 9, 2009.
Kapteyn, A. Vos, K. de (1998), ‘Social Security and Labor Force Participation in The Netherlands’, American Eco-
nomic Review, Vol. 88, No. 2, pp. 164-167.
Kapteyn, A., Vos, K. de (1999), ‘Social Security and Retirement in the Netherlands’, In: J. Gruber and D. Wise (Eds.),
Social Security Programs and Retirement Around the World, Chicago, University of Chicago Press, pp. 269–303.
Klevmarken, N.A. (2001), ‘Micro simulation – a tool for economic analysis’, Department of Economics, Uppsala
University, Working Paper 2001:13.
Klevmarken, N.A. (2007), ‘Simulating the future of the Swedish baby-boom generations’, Department of Economics,
Uppsala University, Working Paper 2007:26.
Knoef, M., Alessie, R., Kalwij, M. (2009), ‘Changes in the income distribution of the Dutch elderly between 1989 and
2020: a microsimulation’, Netspar Paper, September 9, 2009.
Koning, P.W.C. (2004), ‘Estimating the Impact of Experience Rating on the Inflow into Disability Insurance in the
Netherlands’, CPB Discussion Paper 37.
Koning, P.W.C., Vuuren, D.J. van (2010), ‘Disability Insurance and Unemployment Insurance as Substitute Pathways’,
Applied Economics, Vol. 42, No. 5, pp. 575-588.
Kühntopf, S., Tivig, T. (2008), ‘Early Retirement in Germany: Loss of income and lifetime?’, Thünen-Series of Ap-
plied Economic Theory, Working Paper No. 85.
Li, J. (2010), ‘Jinjing Li’s Website’, retrieved from http://www.jinjingli.com/microsimulation on September 20, 2010.
Lindeboom, M. (1998), ‘Microeconometric analysis of the retirement decision: The Netherlands’, OECD Economics
Department Working Paper 107.
282
Linder, F. (2005), ‘Dynamiek in de WAO, WAZ en Wajong: een longitudinale analyse van personen met een ar-
beidsongeschiktheidsuitkering, 1999–2002’, Sociaal-economische trends, 1e kwartaal 2005.
Lindner, H., Veerman, T.J. (2003), ‘Nederland wordt beter. De effecten van beleid en conjunctuur op het ziektever-
zuim’, AStri, Leiden, oktober 2003.
Lise, J., Seitz, S. (2007), ‘Consumption Inequality and Intra-Household Allocations’, The Institute for Fiscal Studies
Working Paper 09/07.
Lucas, R. (1976), ‘Econometric Policy Evaluation: A Critique’, in K. Brunner and A. Meltzer (Eds), The Phillips Curve
and Labor Markets, Carnegie-Rochester Conference Series on Public Policy, 1, New York, American Elsevier,
pp. 19–46.
Lumsdaine, R.L., Stock, J.A., Wise, D.H. (1995), ‘Why are Retirement Rates so high at Age 65?’, NBER Working
Paper No. 5190.
Manoli, D., Mullen, K., Wagner, M. (2009), ‘Pension Benefits & Retirement Decisions: Income vs. Price Elasticities’,
September 28, 2009.
Martikainen, P., Mäkeläa, P., Koskinend, S. and Valkonena, T. (2001), ‘Income differences in mortality: a register-
based follow-up study of three million men and women’, International Journal of Epidemiology, Vol. 30, pp.
1397-1405.
Martin, J.P., Whitehouse, E. (2008), ‘Reforming Retirement-Income Systems: Lessons from the Recent Experiences of
OECD Countries’, OECD Social, Employment and Migration Working Papers No. 66.
Mastrobuoni, G. (2006), ‘Labor Supply Effects of the Recent Social Security Benefits Cuts: Empirical Estimates Using
Cohort Discontinuities’, CEPS Working Paper 136.
Monden, C.W.S. (2008), ‘Grote verschillen in de groep werkende 65plus-mannen’, Economisch Statistische Berichten,
Vol. 93, No. 4531, pp. 183-185.
Morrison, R.J. (2009), ‘Rates of Return in the Canada Pension Plan: Sub-populations of Special Policy Interest and
Preliminary After-tax results’, in A. Zaidi, A. Harding, P. Williamson (Eds.), New Frontiers in Microsimulation
Modelling, European Centre Vienna, pp. 307-332..
Musgrave, R. (1986), ‘A Reappraisal of Financing Social Security’, In R. Musgrave (Ed.), Public Finance in a Demo-
graphic Society, Vol. II., Brighton, Wheatsheaf Books, pp. 103-122.
Nelissen, J.H.M. (1991), ‘Household and Education projections by means of a microsimulation model’, Economic
Modelling, Vol. 8, pp 480-511.
References 283
Nelissen, J.H.M. (1993), ‘Labour market, income formation and social security in the microsimulation model NEDY-
MAS’. Economic Modelling, Vol. 10, No. 3, pp. 225-272.
Nelissen, J.H.M. (1995a), ‘Lifetime income redistribution by the old-age state pension in The Netherlands. ‘, Journal
of Public Economics, Vol. 58, pp. 429-451.
Nelissen, J.H.M. (1995b), ‘Demographic projections by means of microsimulation, The NEDYMAS Model, Part A’,
Worc Report, Work and Organization Research Centre, Tilburg University.
Nelissen, J.H.M. (2002), ‘Early retirement: The impact of changes in the benefit level’, Tinbergen Institute Discussion
Paper TI 2002-031/3.
O’Donoghue, C. (2001), ‘Dynamic Microsimulation: A Methodological Survey’, Brazilian Electronic Journal of Eco-
nomics, Vol. 4, No. 2.
OECD (2003), ‘Transforming disability to ability, policies to promote work and income security for disabled people’,
OECD Book.
OECD (2008a), ‘OECD Economic Surveys: Netherlands Volume 2008 Issue 1’, OECD Book.
OECD (2008b), ‘Sickness, Disability and Work: Breaking the Barriers (Vol. 3) – Denmark, Ireland, Finland and the
Netherlands’, OECD Book.
OECD (2009), ‘Pensions at a Glance 2009 Retirement-Income Systems in OECD Countries’, OECD Book.
Oksanen, H. (2004), ‘Pension reforms: An Illustrated Basic Analysis’, CESifo Economic Studies, Vol. 50, No. 3, pp.
569-625.
Orcutt, G. (1957), ‘A new type of socio-economic system’, Review of Economics and Statistics, Vol. 58, pp. 773-797.
Ours, J.C. van, Tuit, S. (2010), ‘How Changes in Unemployment Benefit Duration Affect the Inflow Into Unemploy-
ment’, CentER Discussion Paper, 2010-07, pp. 1-20.
Peichl, A. (2008), ‘The Benefits of Linking CGE and Microsimulation Models: Evidence from a Flat Tax Analysis’,
IZA Discussion Paper No. 3715.
Peichl, A., Schaefer, T. (2009), ‘FiFoSiM - An Integrated Tax Benefit Microsimulation and CGE Model for Germany’,
International Journal of Microsimulation, Vol. 2, No. 1, pp. 1-15.
Pennec, S., Keegan, M. (2007), ‘APPSIM - Modelling Migration’, NATSEM Working Paper No. 5. September 2007.
Perese, K. (2002), ‘Mate Matching for Microsimulation Models’, Technical Paper Series Congressional Budget Office
2002-3.
284
Pettersson, T., Pettersson, T., Westerberg, A. (2006), ‘Generationsanalyser - omfördelning mellan generationer i en
växande välfärdsstat’, Finansdepartementet, ESS 2006:6.
Pingle, J.F. (2006), ‘Social Security’s Delayed Retirement Credit and the Labor Supply of Older Men’, Federal Re-
serve Board, Financial and Economics Discussion Series 2006-37.
Poos, M.J.J.C., Smit, J.M., Groen, J., Kommer, G.J., Slobbe, L.C.J. (2008), ‘Cost of Illness in the Netherlands 2005’,
RIVM rapport 270751019.
Prinz, C., Tompson, W. (2009), ‘Sickness and disability benefit programmes: What is driving policy convergence?’,
International Social Security Review, Vol. 62, No. 4, pp. 41-61.
Queisser, M., Whitehouse, E. (2006), ‘Neutral or Fair? Actuarial Concepts and Pension-System Design’, OECD So-
cial, Employment and Migration Working Papers No. 40.
Reijer, A. den (2006), ‘The Dutch business cycle: which indicators should we monitor?’, De Nederlandsche Bank,
Working Paper No. 100/2006.
Rele, H. ter (1998), ‘Generational accounts for the Netherlands’, De Economist, Vol. 146, pp. 555-584.
Røed, K., Strøm, S. (1999), ‘Progressive Taxes and the Labour Market – Is the Trade-Off between Equality and Ef-
ficiency Inevitable?’, Department of Economics, University of Oslo Memorandum 19/99.
Romijn, G., Goes, J., Dekker, P., Gielen, M., Es, F. van (2008), ‘MIMOSI Microsimulatiemodel voor belastingen,
sociale zekerheid, loonkosten en koopkracht’, CPB document No. 161.
Samwick, A.A. (1998), ‘Discount rate heterogeneity and Social Security reform’, Journal of Development Economics,
Vol. 57, No. 17, pp. 7–46.
Samwick, A.A. (2001), ‘Comment on Coile and Gruber: Social security incentives for retirement.’ In Themes in the
Economics of Aging, ed. D. Wise, Chicago: University of Chicago Press, pp. 341–54..
Selten, R., Copinga, M. (2003), ‘Wie komen in de WAO’, CBS 1 mei 2003, projectnummer CBO-202271.
SER (2004), ‘Advies verdere uitwerking WAO-beleid – reactie op enkele kabinetsvoornemens, Bijlage 8: Berekening
UWV wijziging WAO’, Februari 2004.
Sonsbeek, J.M. van (2010), ‘Microsimulations on the effect of ageing-related policy measures’, Economic Modelling,
Vol. 27, pp. 968-979.
Sonsbeek, J.M. van (2011), ‘The Social Affairs Department of the Netherlands Ageing and Pensions Model’, Interna-
tional Journal of Microsimulation, Vol. 4, No.1, forthcoming.
References 285
Sonsbeek, J.M. van, Alblas, R. (2010), ‘Langetermijn raming arbeidsongeschiktheidsregelingen Januari 2010‘, Min-
isterie van SZW.
Sonsbeek, J.M. van, Gradus, R.H.J.M. (2006), ‘A microsimulation analysis of the 2006 regime change in the Dutch
disability scheme’, Economic Modelling, Vol. 23, pp. 427-456.
Sonsbeek, J.M. van, Gradus, R.H.J.M. (2010), ‘De Beleidsmaatregelen voor de WAO gekwantificeerd’ , Economisch
Statistische Berichten, Vol. 95, pp. 586-588.
Sonsbeek, J.M. van, Schepers, J.H. (2001), ‘Evaluatie Arbeidsongeschiktheids- en Ziekteverzuimmaatregelen sinds
1992’, Ministerie van SZW, Oktober 2001.
Spadaro, A. (2003), ‘Micro-simulation and Normative Policy Evaluation: an Application to some EU Tax-Benefits
Systems’, DELTA, Working Paper No. 2003-23.
Spielauer, M. (2002), ‘Dynamic Microsimulation of Health Care Demand, Health Care Finance and the Economic Im-
pact of Health Behaviour’, International Institute for Applied Systems Analysis Interim Report IR-02-032/April,.
Spielauer, M. (2007), ‘Dynamic Microsimulation of Health Care Demand, health Care Finance and the Economic
Impact of Health Behaviours: Survey and Review’, International Journal of Microsimulation, Vol. 1, No. 1, pp.
35-53.
Statistics Canada (2009), ‘Modgen Version 10.1.0 Developer’s Guide’.
Stegeman, H. (2005), ‘De conjunctuurgevoeligheid van ziekteverzuim’, CPB Document 99.
Stensnes, K., Stolen, N.M. (2007), ‘Pension Reform in Norway. Microsimulating effects on government expenditures,
labour supply incentives and benefit distribution.’, Statistics Norway, Research Department, Discussion Papers
No. 524.
Stock, J.H., Wise, D.A. (1990), ‘Pensions, the option value of work and retirement.’, Econometrica, Vol. 58, pp.
1151–80.
Sutherland, H., Taylor, R., Gomulka, J. (2002), ‘Combining household income and expenditure data in policy simula-
tions’, Review of Income and Wealth, Vol. 48, No. 4, pp. 517-536.
SVB (2008), ‘The Dutch State Pension, Facts and Figures’, Sociale Verzekeringsbank, International Affairs and Cor-
porate Communications Departments.
SZW (2006), ‘De toekomstige inkomenspositie van ouderen’, werkdocument SZW.
Toder, E., Favreault, M., O’Hare, J., Rogers, D., Sammartino, F., Smith, K. (2000), ‘Long Term Model Development
for Social Security Analysis’, The Urban Institute, January 31, 2000.
286
Tweede Kamer der Staten-Generaal (2005), ‘Wet werk en inkomen naar arbeidsvermogen, Memorie van Toelichting’,
Kamerstukken 30034, No. 3.
UWV (2004), ‘Maandoverzicht arbeidsongeschiktheidsuitkeringen januari t/m december 2003’.
UWV (2006), ‘UWV Kwartaal Verkenning 2006-I’, Kenniscentrum UWV, April 2006.
UWV (2008), ‘UWV Kwartaal Verkenning 2008-I’, Kenniscentrum UWV, April 2008.
UWV (2009), ‘Statistische tijdreeksen UWV 2008, Wetgeving en volume-ontwikkeling in historisch perspectief’.
Vonkova, H., Soest, A. van (2009), ‘How Sensitive Are Retirement Decisions to Financial Incentives: A Stated Prefer-
ence Analysis’, Netspar Paper, September 8, 2009.
Vos, K. de, Kapteyn, A. (2004), ‘Incentives and Exit Routes to Retirement in the Netherlands’, Chapter in J. Gruber
and D.A. Wise (Eds.) Social Security Programs and Retirement around the World: Micro-Estimation, NBER,
pp. 461 - 498.
Vuren, A. van, Vuuren, D. van (2007), ‘Financial incentives in Disability Insurance in the Netherlands’, De Economist,
Vol. 155, No. 1, pp. 73-98.
Wagenhals, G. (2004), ‘Tax-benefit microsimulation models for Germany: A Survey’, Hohenheimer Diskussionsbe-
iträge, nr. 235/2004.
Westerhout, E. (2001), ‘Disability Risk, Disability Benefits, and Equilibrium Unemployment’, International Tax and
Public Finance, Vol. 8, pp. 219-243.
Whiteford, P., Whitehouse, E. (2006), ‘Pension Challenges and Pension reforms in OECD Countries’, Oxford Review
of Economic Policy, Vol. 22 No. 1, pp. 78-94.
Whitehouse, E. (2007), ‘Pensions Panorama. Retirement-Income Systems in 53 Countries’, The World Bank.
Williamson, P. (2007), ‘The Role of the International Journal of Microsimulation’, International Journal of Micro-
simulation, Vol. 1, No. 1, pp. 1-2.
Williamson P., Zaidi, A., Harding, A. (2009), ‘New Frontiers in Microsimulation Modelling: Introduction’, in A. Zaidi,
A. Harding, P. Williamson (Eds.), New Frontiers in Microsimulation Modelling, European Centre Vienna, pp.
31-50.
Zwinkels, W. (2007), ‘Effectiviteit van re-integratie: onbenutte potenties van privatisering’, Tijdschrift voor Arbeids-
vraagstukken, Vol. 23, No. 2, pp. 121-131.
List of figures
Figure 1.1 Microsimulation models in perspective.......................................................................................... 20
Figure 1.2 Number of births per cohort and average number of children per woman, 1900-2008.................. 27
Figure 1.3 Life expectancy at birth, 1861-2006................................................................................................ 28
Figure 1.4 Life expectancy at 65, 1950–2050.................................................................................................. 29
Figure 1.5 Population by age, 2009.................................................................................................................. 30
Figure 1.6 Population indicators, 1950–2050................................................................................................... 31
Figure 1.7 Effective retirement age, 1970–2007.............................................................................................. 32
Figure 1.8 Grey pressure in selected OECD countries, 2000–2050................................................................. 33
Figure 1.9 Fertility rates by OECD country, 2006............................................................................................ 34
Figure 1.10 Life expectancy at birth by OECD country, 2006........................................................................... 35
Figure 1.11 Stock of disability benefit recipients, 1968–2009........................................................................... 38
Figure 1.12 Inflow and outflow of disability benefits (WAO/WIA schemes), 1969–2009................................. 39
Figure 1.13 Inflow and outflow of disability benefits (Wajong scheme), 1977–2009........................................ 40
Figure 1.14 Inflow rates WAO/WIA by gender, 1967–2009.............................................................................. 41
Figure 1.15 WAO, WIA and Wajong beneficiaries by age category, 1968–2009............................................... 42
Figure 1.16 Working population by age and occupation, 2005.......................................................................... 45
Figure 1.17 Labour participation rate males by age category, 2000 vs. 2050.................................................... 46
Figure 1.18 Labour participation rate of females by age category, 2000 vs. 2050............................................. 47
Figure 1.19 Disability rates of males by age, 2000 vs. 2009.............................................................................. 48
Figure 1.20 Disability rates of females by age, 2000 vs. 2009........................................................................... 48
Figure 1.21 Population structure of disability beneficiary stock, 2000 and 2009.............................................. 49
Figure 2.1 Proposed regime change in disability benefits................................................................................ 61
Figure 2.2 Summary of modules WAO/WIA model......................................................................................... 67
Figure 2.3 Simulation loop WAO/WIA model................................................................................................. 74
Figure 2.4 Projected number of disability benefits before (WAO) and after system change (WIA), 2003-2042..
......................................................................................................................................................... 80
288
Figure 2.5 Projected number of beneficiaries working before (WAO) and after system change (WIA), 2003-
2042................................................................................................................................................. 82
Figure 2.6 Projected total cost of all disability related benefits before (WAO) and after system change (WIA),
2003-2042........................................................................................................................................ 84
Figure 2.7 Projected income development partially disabled before (WAO) and after system change (WIA).86
Figure 3.1 Disability (WAO+WIA) stock (left axis) and inflow (right axis), 1968–2009.............................. 108
Figure 3.2 WAO+WIA inflow rate on cohort basis, 1999Q1–2009Q3........................................................... 115
Figure 3.3 WAO+WIA outflow rate on cohort basis, 1999Q1–2009Q3......................................................... 116
Figure 3.4 Recovery rate by duration of the benefit, WAO 2004................................................................... 135
Figure 3.5 Mortality rate by age, WAO 2004................................................................................................. 136
Figure 3.6 Long-term forecasts, WAO/WIA scheme for 2000, 2004 and 2009 to 2040................................ 143
Figure 3.7 Long-term forecasts WAO outflow, 2004 and 2009 to 2040......................................................... 145
Figure 3.8 Long-term forecasts, Wajong scheme 2000, 2004 and 2009 to 2040............................................ 147
Figure 4.1 SADNAP model flow diagram...................................................................................................... 176
Figure 4.2 Average wage and expected second-pillar pension by age, 2005................................................. 180
Figure 4.3 Projected state pension cost as % of GDP (including partner allowances), 2007–2080............... 189
Figure 4.4 Changes in composition of pensioner population, 2006–2050..................................................... 191
Figure 4.5 Projected rate of non-participation in the labour force by age...................................................... 195
Figure 4.6 The simulation process of the population forecast........................................................................ 205
Figure 4.7 Composition of the simulated population, 2006–2080................................................................. 206
Figure 4.8 Composition of the 2005 population by age and participation category....................................... 209
Figure 5.1 Projected savings of policy alternatives 2006-2050...................................................................... 241

List of tables
Table 1.1 Assessment framework for microsimulation models...................................................................... 21
Table 2.1 Disability benefits and possible supplements in the Netherlands (2003)........................................ 59
Table 2.2 Characteristics disability beneficiaries (2003)................................................................................ 69
Table 2.3 Deviation in simulation results by sample size............................................................................... 71
Table 2.4 Imputation scheme WAO/WIA model............................................................................................. 72
Table 2.5 Projected replacement rates by subgroup before (WAO) and after system change (WIA)............. 85
Table 2.6 Sensitivity analysis on labour market scenarios (2040).................................................................. 89
Table 2.7 Sensitivity analysis on benefit elasticity (2040).............................................................................. 90
Table 2.8 Sensitivity analysis on benefit elasticity heterogeneity (2040)....................................................... 90
Table 2.9 Changes in Gini-coefficients........................................................................................................... 92
Table 3.1 GLS estimates inflow model......................................................................................................... 122
Table 3.2 GLS estimates inflow models for men and women aged >= 45 and < 45 respectively................. 124
Table 3.3 GLS estimates for sector inflow models........................................................................................ 125
Table 3.4 OLS estimates outflow model....................................................................................................... 127
Table 3.5 OLS estimates outflow models for men and women aged >= 45 and < 45 respectively................... 129
Table 3.6 Transition matrix, WAO (means for 2009).................................................................................... 138
Table 3.7 Transition matrix, WIA (means for 2009)..................................................................................... 138
Table 3.8 Transition matrix, Wajong (means for 2009)................................................................................ 138
Table 3.9 Model selection for WAO, Wajong and WIA schemes, 2000, 2004 and 2009.............................. 140
Table 3.10 Long-term (2040) effects of clusters of policy measures.............................................................. 148
Table 3.11 Inflow on cohort basis (WAO+WIA) 1999–2009......................................................................... 152
Table 3.12 Outflow on cohort basis (WAO+WIA) 1999–2009....................................................................... 153
Table 3.13 Outflow on cohort basis (excluding pensioning and mortality, WAO+WIA) 1999–2009............ 154
Table 4.1 Projected average life expectancies at 65 of the 2006–2045 pensioner cohorts........................... 182
Table 4.2 Option value parameters in the SADNAP model.......................................................................... 187
Table 4.3 Share of lifetime state pension income compared with share of state pension cohorts................ 192
290
Table 4.4 Characteristics of retirees by retirement age (birth cohorts 1946–1970)...................................... 196
Table 4.5 Overview of key SADNAP results compared with other models................................................. 199
Table 4.6 Some characteristics of the two main data sources on state pensions........................................... 213
Table 4.7 Some characteristics of the main data source on company pensions............................................ 214
Table 4.8 Expected private pension by main subgroups, 2005..................................................................... 216
Table 4.9 Overview of macro data sources................................................................................................... 220
Table 5.1 Share of lifetime state pension income compared with share of state pension cohorts................ 234
Table 5.2 Projected budgetary effects of the policy alternatives (2040)....................................................... 239
Table 5.3 Projected effect on retirement age of the policy alternatives........................................................ 242
Table 5.4 projected effect on income inequality of the policy alternatives................................................... 244
Table 5.5 Effect of policy alternatives on share of lifetime state pension income compared with share
of state pension cohorts................................................................................................................. 245
Table 5.6 Overall assessment of the policy alternatives................................................................................ 246
Table 5.7 Direct budgetary effects baseline and policy scenarios 2009-2050.............................................. 250
List of Abbreviations
AAW Algemene Arbeidsongeschiktheidswet (General Incapacity Act)
AGE Applied General Equilibrium
AOW Algemene Ouderdoms Wet (General Old Age Pensions Act)
CBS Centraal Bureau voor de Statistiek (Statistics Netherlands)
CGE Computable General Equilibrium
CPB Centraal Planbureau (Netherlands Bureau of Economic Policy Analyses)
DI Disability Insurance
DNB De Nederlandsche Bank (Dutch Bank)
DSGE Dynamic Stochastic General Equilibrium
ERS Early Retirement Scheme
Fde Fully disabled equivalent
IOAW Inkomensvoorziening voor Oudere en Arbeidsongeschikte Werknemers (Older
and Partially Incapacitated Former Employed Persons Income Scheme Act)
IVA Inkomensvoorziening voor Volledig en duurzaam Arbeidsongeschikten (Income
support scheme for persons incapable of work)
NIDI Nederlands Interdisciplinair Demografisch Instituut (Netherlands Interdisciplinary
Demographic Institute)
OECD Organisation for Economic Cooperation and Development
PBL Planbureau voor de Leefomgeving (Netherlands Environmental Assessment
Agency)
PEMBA Premiedifferentiatie En Marktwerking Bij Arbeidsongeschiktheid (Premium dif-
ferentiation and Market competition in Disability insurance)
SADNAP Social Affairs Department of the Netherlands Ageing and Pensions model
SAS Statistical Analysis System
SCP Sociaal Cultureel Planbureau (Netherlands Institute for Social Research)
SER Sociaal Economische Raad (Social Economic Council)
292
SVB Sociale Verzekerings Bank (Social Insurance Bank)

TBA Terugdringing Beroep op de Arbeidsongeschiktheidsregelingen (Act on Reducing
Disability Claims)
TW Toeslagenwet (Supplementary Benefits Act)
UWV Uitvoeringsorgaan Werknemers Verzekeringen (Employees Insurance Administra-
tion Office)
VAR Vector Auto Regression
Wajong Wet Arbeidsongeschiktheid Jonggehandicapten (Incapacity Insurance (Young
Disabled Persons) Act)
WAO Wet op de Arbeidsongeschiktheidsverzekering (Incapacity Insurance Act)
WAZ Wet Arbeidsongeschiktheid Zelfstandigen (Incapacity Insurance (Self Employed
Persons) Act)
WGA Werkhervattingsregeling Gedeeltelijk Arbeidsgeschikten (Work resumption ben-
efit for persons partially capable of work)
WIA Wet werk en Inkomen naar Arbeidsvermogen (Work and Income (Capacity for
Work) Act)
WODC Wetenschappelijk Onderzoek- en Documentatie Centrum (Research and Docu-
mentation Centre of the Ministry of Justice)
WW Werkloosheidswet (Unemployment Insurance Act)
Samenvatting
(Summary in Dutch)
Microsimulatiemodellen in sociale zekerheidsbeleid
In dit proefschrift wordt het gebruik van microsimulatiemodellen bestudeerd op het ter-
rein van de sociale zekerheid. De centrale onderzoeksvraag van het proefschrift is of en hoe
microsimulatiemodellen geïntegreerd kunnen worden in de beleidsanalyse. Ten behoeve van
de beleidsanalyse op het gebied van arbeidsongeschiktheid en AOW/pensioenen zijn op het
Ministerie van Sociale Zaken en Werkgelegenheid in het afgelopen decennium twee microsi-
mulatiemodellen ontwikkeld. Deze twee modellen worden in dit proefschrift beschreven en
gebruikt.
Economische modellen en microsimulatiemodellen
Nederland heeft een lange traditie in het ontwikkelen van economische modellen: Nobelprijs-
winnaar Jan Tinbergen stond aan de wieg van het macro-economische modelleren. In de loop
van de tijd zijn economen erin geslaagd steeds geavanceerdere benaderingen van de economi-
sche werkelijkheid te vinden. De onderkenning van het feit dat economieën zijn opgebouwd
uit individuen en bedrijven die zich ieder afzonderlijk verschillend kunnen gedragen, heeft
294
geleid tot de opkomst van zowel algemene evenwichtsmodellen als microsimulatiemodellen.

Algemene evenwichtsmodellen zijn geworteld in de micro-economische theorie en beschrij-
ven het gedrag van economische agenten. Bij deze modellen wordt een evenwichtstoestand
van de economie afgeleid waarbij de heterogeniteit van een samenleving wordt gerepresen-
teerd door opname van een aantal standaardhuishoudens en standaardbedrijven in het model.
Rekentechnisch is voor deze modellen echter alleen een sterk vereenvoudigde versie van de
werkelijkheid behapbaar. Microsimulatiemodellen benaderen de werkelijkheid veel dichter
door met grote hoeveelheden individuen een samenleving te beschrijven, maar het gedrag van
economische agenten wordt vaak slechts summier beschreven. Er worden geen evenwichten
berekend en de modellen werken in de basis mechanisch. Omdat beide aanpakken zowel
voor- als nadelen hebben, ontstaat er momenteel steeds meer belangstelling voor de combi-
natie van beide modelleertechnieken. Voor het in kaart brengen van de effecten van complexe
belastingen sociale zekerheidsmaatregelen is microsimulatie een goed instrument. Microsi-
mulatiemodellen hebben echter in Nederland nog slechts weinig ingang gevonden. Karakte-
ristieken van de Nederlandse situatie, zoals de gescheiden en nauwkeurig gedefinieerde rollen
van het Centraal Planbureau (CPB) en het Centraal Bureau voor de Statistiek (CBS), kunnen
hierbij een rol gespeeld hebben.
Sociale zekerheidshervormingen
In het afgelopen decennium hebben in Nederland enkele grote hervormingen in de sociale

zekerheid plaatsgevonden. Eén van de meest in het oogspringende was de hervorming van
de Wet op de Arbeidsongeschiktheid, de WAO. Arbeidsongeschiktheid is lange tijd een open
zenuw geweest in de Nederlandse politiek. Het percentage arbeidsongeschikten in Neder-
land behoorde tot de hoogste ter wereld en begin deze eeuw naderde het aantal arbeidson-
geschikten het miljoen, maar hervormingspogingen liepen tot dan veelal stuk op politieke
en maatschappelijke weerstand. Pas begin deze eeuw zijn belangrijke hervormingen van de
Samenvatting 295
arbeidsongeschiktheidsregelingen van de grond gekomen. Deze hebben de instroom van

nieuwe arbeidsongeschikten drastisch beperkt. Nederland is daardoor volgens de Organisatie
van Economische Samenwerking en Ontwikkeling (OESO) van een van de slechtste jongetjes
van de klas een voorbeeld van succesvol hervormingsbeleid geworden. Wel baart de sterke
groei van de Wajong, de arbeidsongeschiktheidsregeling voor jonggehandicapten, zorgen.
De laatste jaren staat vooral de hervorming van de AOW en pensioenen in het middelpunt
van de belangstelling. Ook hier is de noodzaak voor hervorming evident, maar in tegenstel-
ling tot de arbeidsongeschiktheidsproblematiek is vergrijzing een fenomeen dat zich in de
gehele westerse wereld in meer of mindere mate voordoet. De grote geboortegolf van na de
Tweede Wereldoorlog zal vanaf 2011 met pensioen gaan. Deze ouderen hebben een sterk
gestegen levensverwachting vergeleken met vroeger. De beroepsbevolking gaat afnemen en
in sectoren als zorg en onderwijs zullen hierdoor tekorten ontstaan. Pensioenhervormingen
worden dan ook in vrijwel alle westerse landen doorgevoerd of voorbereid en het maatschap-
pelijke verzet tegen deze hervormingen is aan het wegebben.
Arbeidsongeschiktheid en vergrijzing hebben veel met elkaar te maken. De vergrijzing leidt

eerst tot veroudering van de beroepsbevolking en daardoor tot een hoger risico op arbeidson-
geschiktheid. Later leidt te vergrijzing tot een toename van de AOW-lasten, die door een slin-
kende beroepsbevolking moeten worden gedragen. Dat noodzaakt tot participatie van zoveel
mogelijk mensen, dus ook van groepen met een verhoogd arbeidsongeschiktheidsrisico.
De invoering van de WIA
In hoofdstuk 2 van dit proefschrift wordt het microsimulatiemodel beschreven dat is ontwik-
keld en gebruikt voor de analyse van de introductie van de Wet WIA in 2006. De WIA is de
opvolger van de WAO die in 1967 werd geïntroduceerd. De WIA wijkt op een aantal cruciale
296
punten af van de WAO. In plaats van één algemene regeling voor alle arbeidsongeschikten,
bestaat de WIA uit twee regelingen: een voor duurzaam volledig arbeidsongeschikten (IVA)
en een voor gedeeltelijk en tijdelijk volledig arbeidsongeschikten (WGA). De WGA kent
bovendien verschillende uitkeringsregimes voor werkenden en niet-werkenden. Daarnaast
is de toetredingsdrempel in de WIA verhoogd ten opzichte van de WAO: van 15% naar 35%
arbeidsongeschikt. Ook zijn de keuringscriteria aangescherpt zodat gemiddeld genomen een
lagere mate van arbeidsongeschiktheid wordt toegekend. Ook bestaande WAO’ers beneden
een bepaalde leeftijdsgrens zijn onderworpen aan deze striktere keuringscriteria. Tenslotte
is de periode van loondoorbetaling bij ziekte door de werkgever verlengd van een naar
twee jaar. Dit samenstel aan wijzigingen maakt het ramen van de effecten van de WIA een
bijzonder complexe aangelegenheid. Voor mensen die tot de WIA worden toegelaten, zijn de
uitkeringsregels anders dan ze in de WAO zouden zijn geweest. Van de mensen die niet tot de
WIA worden toegelaten, zal een deel erin slagen om te blijven werken. Voor de anderen is het
van belang of er recht bestaat op andere sociale zekerheidsuitkeringen zoals de WW of de bij-
stand. Door de complexiteit van de WIA-hervorming is microsimulatie een interessante optie
om de effecten ervan te ramen. Het is, als uitkeringsgerechtigden in zoveel verschillende
regelingen terecht kunnen komen, vrijwel onmogelijk om met macromodellen een consistente
en gedetailleerde raming van alle betrokken regelingen te maken. Microsimulatiemodellen
zijn echter per definitie consistent, omdat individuen in een simulatie op één bepaald tijdstip
maar in één bepaalde toestand kunnen verkeren. Omdat het UWV in 2001 gedetailleerde
microdatabestanden over arbeidsongeschiktheidsuitkeringen ter beschikking is gaan stellen,
werd microsimulatie een haalbare optie.
Het aantal WIA-uitkeringen komt volgens de voorafgaand aan de WIA gedane microsimula-
tie-analyse op de lange termijn meer dan 25% lager uit dan het aantal WAO-uitkeringen bij
ongewijzigd beleid zou zijn uitgekomen. De kosten van de arbeidsongeschiktheidsuitkerin-
gen dalen met bijna € 2 miljard. Een aanzienlijk deel van de mensen die wel recht op WAO
maar geen recht meer op WIA hebben en van de gedeeltelijk arbeidsongeschikten in de WIA
Samenvatting 297
gaat aan het werk. Gemiddeld genomen blijft de inkomenspositie van arbeidsongeschikten
voor en na de stelselwijziging gelijk en slagen arbeidsongeschikten er dus in hun verlies aan
inkomen uit uitkeringen te compenseren met inkomen uit arbeid. De WIA kan daarom een
hervorming genoemd worden die de maatschappelijke welvaart verhoogt.
De arbeidsongeschiktheidshervormingen in retrospectief
In hoofdstuk 3 wordt een regressieanalyse gemaakt van de effecten van het samenstel aan
maatregelen dat is genomen om het aantal arbeidsongeschikten te reduceren. Naast de WIA
en de kort daaraan voorafgaand geïntroduceerde strengere keuringscriteria gaat het hier om
de introductie van premiedifferentiatie voor werkgevers (de Wet Pemba) in 1998 en de Wet
Verbetering Poortwachter uit 2002. In die laatste wet zijn de verplichtingen van werkgevers
en werknemers bij langdurige ziekte nauwkeurig omschreven evenals de sanctionering als
die verplichtingen niet worden nageleefd. Met behulp van microsimulatiemodellen wordt het
toekomstig verloop van de arbeidsongeschiktheidsregelingen bepaald vanuit verschillende
basisjaren van voor en na de hervormingen, zodat ook het gecombineerde effect van de maat-
regelen in kaart wordt gebracht.
Voor de instroomontwikkeling is een regressiemodel geschat waarin de instroom in de WAO/

WIA wordt verklaard uit de conjunctuur, seizoensinvloeden en beleidsmaatregelen. Uit het
model blijkt dat alle vier de genoemde hervormingen een significant neerwaartse invloed
op de instroom in de arbeidsongeschiktheid hebben gehad: Pemba met 13%, Poortwachter
met 22%, het nieuwe Schattingsbesluit met 13% en de WIA met 23%. Samen hebben deze
vier hervormingen de instroom met 71% verminderd. Nadere analyse leert dat de effecten
verschillend uitpakken voor mannen en vrouwen, jongeren en ouderen en voor verschillende
sectoren. Zo blijken Pemba en Poortwachter effectiever voor vrouwen dan voor mannen,
waarschijnlijk omdat juist de instroom van vrouwen in de WAO zo buitenproportioneel hoog
298
was. De WIA blijkt echter effectiever bij mannen die vaker gedeeltelijk arbeidsongeschikt
zijn. Ook blijken de financiële prikkels van Pemba zeer effectief in een sector als de bouw,
maar nauwelijks effectief in de publieke sector. Poortwachter is het effectiefst in sectoren met
veel grote organisaties zoals de publieke sector en de zakelijke en financiële dienstverlening.
Voor de uitstroomontwikkeling is ook een regressiemodel geschat waarin de uitstroom uit

de WAO/WIA is verklaard uit de conjunctuur, seizoensinvloeden en beleidsmaatregelen.
Het gaat hierbij vooral om de herbeoordelingoperatie WAO die tussen 2005 en 2009 heeft
gelopen. De herbeoordelingen hebben de uitstroom uit de WAO bevorderd voor degenen die
herbeoordeeld werden. Maar onder degenen die niet herbeoordeeld werden, traden veel min-
der herstelgevallen op dan op basis van hun kenmerken verwacht had mogen worden. Het is
daarom maar de vraag of de per saldo effecten van de herbeoordelingoperatie wel zo gunstig
zijn geweest als ze op het eerste gezicht lijken.
De instroomreductie als gevolg van de hervormingen in de WAO en de WIA levert op de lan-

ge termijn een enorme daling van het aantal arbeidsongeschiktheidsuitkeringen op: van 1,2
miljoen als er geen hervormingen zouden zijn geweest naar de actuele schatting van minder
dan 400.000. Deze spectaculaire resultaten worden de laatste tijd echter overschaduwd door
de snelle stijging van het aantal Wajong-uitkeringen. Waar op basis van de in- en uitstroom-
patronen uit 2004 nog een stabilisatie rond de 200.000 Wajong-uitkeringen verwacht werd,
wordt nu gerekend op een stabilisatie op boven de 400.000 uitkeringen op de lange termijn.
Analoog aan de analyse dat vroeger de WAO ten onrechte gebruikt is als vervroegde uittreed-
regeling voor slachtoffers van de massaontslagen in de industrieën, kan gesteld worden dat de
Wajong momenteel ten onrechte gebruikt wordt als vangnet voor slachtoffers van de transfor-
matie die Nederland doormaakt naar een kenniseconomie waarin steeds minder plaats is voor
de laagst opgeleiden. Uit de analyse van de WAO-hervormingen kan worden opgemaakt dat
alleen beleid gericht op instroombeperking met effectieve prikkels voor zowel werkgevers als
uitkeringsgerechtigden hiervoor soelaas kan bieden.
Samenvatting 299
Het modelleren van vergrijzing en pensioenhervormingen
Hoofdstuk 4 bevat een beschrijving van het microsimulatiemodel SADNAP dat ontwikkeld is
voor de beleidsanalyse van vergrijzinggerelateerde vraagstukken en de raming van de AOW.
Een microsimulatie-aanpak van vergrijzingvraagstukken is veel complexer en omvangrijker
dan van arbeidsongeschiktheidshervormingen omdat een beschrijving van de gehele bevol-
king noodzakelijk is. In SADNAP worden levenslopen geconstrueerd van een dwarsdoorsne-
de van de Nederlandse bevolking, gebaseerd op SVB-microdata van mensen van 65 en ouder
die al recht op AOW hebben, CBS-microdata van mensen van tussen de 15 en 65 die AOW-
en pensioenrechten aan het opbouwen zijn en bijgeschatte gegevens over kinderen van onder
de 15 die nog geen rechten hebben opgebouwd. Om de beschrijving van de toekomstige
Nederlandse bevolking te completeren worden ook cohorten van toekomstige geboorten en
immigranten toegevoegd, evenals cohorten van emigranten uit het verleden die Nederlandse
AOW-rechten opgebouwd hebben.
In SADNAP zijn de sterftekansen gedifferentieerd naar persoonskenmerken. Hogere inko-

mens blijken langer te leven dan lage inkomens, vrouwen leven langer dan mannen, samen-
wonenden leven langer dan alleenstaanden en autochtonen leven langer dan allochtonen. De
beslissing om te stoppen met werken en met pensioen te gaan wordt individueel gemodel-
leerd met behulp van het optiewaardemodel zoals oorspronkelijk ontwikkeld door Stock en
Wise. Volgens dit model maken individuen een calculatie van hun optimale uittreedleeftijd
waarbij het te verwachten inkomen uit werk en het te verwachten inkomen uit AOW en
aanvullend pensioen bepalend zijn, in combinatie met de persoonlijke voorkeuren voor vrije
tijd, risicoaversie en tijd. Het te verwachten inkomen is voor elk individu af te leiden uit
microdatabestanden en van de persoonlijke voorkeurparameters is uit divers wetenschappe-
lijk onderzoek bekend hoe ze verdeeld zijn over de bevolking en ook dat die verdeling zeer
heterogeen is. De microsimulatie-aanpak leent zich daarom prima voor een modellering van
de uittreedbeslissing.
300
Effecten van vergrijzing bij ongewijzigd en gewijzigd beleid
De resultaten van het SADNAP-model voor de basisprojectie komen goed overeen met be-
kende macro schattingen van CBS en CPB. De grijze druk verdubbelt in 2040 tot bijna 50%
en de AOW-lasten verdubbelen dan eveneens tot bijna 9% van het bruto binnenlands product.
In vergelijking met de macro schattingen komt de microsimulatie analyse enkele tienden
procentpunten lager uit omdat meer mensen, voornamelijk immigranten, een onvolledige
AOW zullen hebben in de toekomst en omdat de kosten van de AOW-partnertoeslag afnemen
omdat meer vrouwen zullen werken. Het budgettaire verloop in de tijd kan met microsimula-
tie goed in kaart gebracht worden. Aanvankelijk worden de AOW-lasten bijvoorbeeld gedrukt
doordat het aandeel samenwonende gepensioneerden stijgt, maar in de loop van de tijd wordt
die trend omgebogen en gaat het aandeel alleenstaande gepensioneerden juist toenemen.
Eveneens geeft het model inzicht in herverdeling in de AOW vanwege verschillen in levens-
verwachting. Vooral de herverdeling tussen lage en hoge inkomens springt in het oog. De
top 20% hoogste inkomens ontvangt anderhalf keer zoveel AOW als de laagste 20%. Al-
leenstaanden compenseren hun kortere levensverwachting ruimschoots omdat ze meer AOW
krijgen dan samenwonenden (70% in plaats van 50% van het minimumloon). Ook wordt het
effect van beleidsmaatregelen op de pensioneringsleeftijd berekend. Zo blijkt de gemiddelde
uittreedleeftijd met 2,5 jaar te stijgen als gevolg van het afschaffen van de lucratieve VUT- en
prepensioenregelingen uit het verleden.
In hoofdstuk 5 wordt met behulp van het SADNAP-model een inschatting gemaakt van de ef-
fecten van een aantal beleidsvoorstellen. Het gaat hierbij om verhoging van de AOW-leeftijd
naar 67 jaar, volledige fiscalisering van de AOW, individualisering van de AOW (uitgewerkt
als verlaging van de AOW voor alleenstaanden naar het niveau van samenwonenden), af-
schaffing van de partnertoeslag en introductie van een flexibele AOW waarin mensen tussen
65 en 70 kunnen kiezen wanneer zij hun AOW laten ingaan. Uitstel van de AOW wordt in die
variant beloond met een hogere AOW voor de rest van het leven. Microsimulatie maakt het
mogelijk om een maatregel op een consistente manier op verschillende aspecten te beoorde-
Samenvatting 301
len. Zo blijkt verhoging van de AOW-leeftijd naar 67 de AOW-uitgaven met 0,5% van het
BBP te verlagen zonder de inkomensverdeling tussen verschillende groepen belangrijk aan
te tasten. Volledige fiscalisering van de AOW heeft nog grotere directe budgettaire effecten,
maar heeft een sterk nivellerende werking. Bovendien wordt het verband tussen ingelegde
premies en uitgekeerde pensioenen (in netto termen) verder verstoord. Individualisering van
de AOW verlaagt de AOW-uitgaven het meest maar veroorzaakt grote negatieve inkomensef-
fecten voor alleenstaanden. Wel wordt bij deze maatregel het verband tussen ingelegde pre-
mies en uitgekeerde pensioenen sterker. Afschaffing van de partnertoeslag voor onvoldoende
verdienende jongere partners van AOW’ers heeft kleinere effecten, maar wel allemaal in de
beleidsmatig wenselijke richting: de kosten dalen, de uittreedleeftijd stijgt, de herverdeling
wordt nauwelijks aangetast en het verband tussen ingelegde premies en uitgekeerde pensioe-
nen wordt versterkt. Flexibilisering van de AOW-leeftijd verhoogt de arbeidsparticipatie van
ouderen, maar daarvoor moet de beloning voor uitstel van de AOW wel voldoende hoog zijn,
zodat de AOW-uitgaven juist gaan stijgen.
Een conclusie over microsimulatiemodellen
In het slothoofdstuk van het proefschrift worden enkele conclusies getrokken over de
toegevoegde waarde van microsimulatiemodellen en de condities waaronder microsimula-
tiemodellen het best tot hun recht komen. De microsimulatiemodellen die voor de arbeidson-
geschiktheidsregelingen zijn ontwikkeld, zijn volledig geïntegreerd in de beleids- en begro-
tingscyclus. Deze modellen worden ook extern gewaardeerd en jaarlijkse actualisering van de
data is gewaarborgd. Het SADNAP-model wordt meer incidenteel ingezet voor analyse van
ingewikkelde beleidsvraagstukken. De complexiteit van het model en de gedateerdheid van
de data verhinderen op dit moment volledige integratie in de beleids- en begrotingscyclus.
Dit sluit aan bij een internationale analyse over de condities waaraan voldaan moet zijn om
microsimulatiemodellen succesvol te integreren in de beleidsanalyse. Goede, regulier geactu-
302
aliseerde microdatabronnen, helder ontwerp van het model, adequate personele bezetting en
een competitieve markt van modelontwikkelingsorganisaties zijn hier de belangrijkste voor-
beelden van. In de toekomst kan zowel in termen van kwaliteit als consistentie van model-
ramingen en –analyses winst geboekt worden door te werken aan een geïntegreerd platform
waarin een inhoudelijke koppeling wordt gemaakt tussen modellen van alle organisaties in
Nederland die economische modellen ontwikkelen. Algemene evenwichtsmodellen van het
CPB zouden hier een centrale plaats kunnen innemen, maar ook zou een prominentere plaats
voor beleidspecifieke microsimulatiemodellen voor sociale zekerheid en fiscaliteit wenselijk
zijn.
View publication stats

Microsimulation As A Decision Making Tool in Socia

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Microsimulation As A Decision Making Tool in Socia

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Microsimulation as a Decision Making Tool In Social Security Policy

Article · May 2011

Jan-Maarten van Sonsbeek

The user has requested enhancement of the downloaded file.

© J.M. Van Sonsbeek, 2011.

Cover en lay-out: Gertjan van de Griendt, StudioGM2, The Hague

Microsimulation as a Decision Making Tool

ter verkrijging van de graad Doctor aan

Johannes Martinus van Sonsbeek

paranimfen: dr. C.L.J.P. van Raalte

2.5.3 Sensitivity analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

3 Estimating the effects of recent disability reforms in the Netherlands. . . . . . . . . . . 103

4.2.3 Microsimulation models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

5 Microsimulations on the effects of ageing-related policy measures. . . . . . . . . . . . . . 221

5.5.5 Assessment of the policy alternatives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

1.1 History of Economic Modelling in the Netherlands

1.2 Introduction to microsimulation models

Figure 1.1 Microsimulation models in perspective

Artificial intelligence Stochastic processes Micro econom(etr)ics Macro econom(etr)ics

Cellular Agent-based Simulation Economic Time series

Representative Microsimulation Cell-based General G1: cyclical Vector

Static ageing Dynamic ageing Static CGE Dynamic CGE Dynamic

Table 1.1 Assessment framework for microsimulation models

Assessment criterion Cell-based model MSM model CGE model

1.3 Microsimulation models in the Netherlands and abroad

In spite of the longstanding tradition of economic modelling, the use of microsimulation

of individual households, a macro model featuring a substantial disaggregation of the labour

(2007) investigates 27 dynamic microsimulation models for which documentation is avail-

On the other hand, no large dynamic microsimulation model is currently in operation in

1.4 Introduction to ageing in the Netherlands

1.4.1 Ageing in the Netherlands in numbers

Source: CBS, Statline

Figure 1.3 Life expectancy at birth, 1861-2006

Source: CBS, Statline

Figure 1.4 Life expectancy at 65, 1950–2050

Source: CBS, Statline (projections based on CBS population forecast 2008–2050)

Figure 1.5 Population by age, 2009

Source: CBS, Statline (projections based on CBS population forecast 2008-2050)

Figure 1.6 Population indicators, 1950–2050

Source: CBS, Statline (projections based on CBS population forecast 2008-2050)

1.4.2 Retirement behaviour in the Netherlands

Figure 1.7 Effective retirement age, 1970–2007

1.4.3 Ageing in the Netherlands in international perspective

Figure 1.8 Grey pressure in selected OECD countries, 2000–2050

Source: OECD, Pensions at a Glance (OECD, 2009)

Figure 1.9 Fertility rates by OECD country, 2006

Source: OECD, Pensions at a Glance (OECD, 2009)

Figure 1.10 Life expectancy at birth by OECD country, 2006

Source: OECD, Pensions at a Glance (OECD, 2009)

1.5 Introduction to disability insurance in the Netherlands

1.5.1 Disability insurance in the Netherlands

Figure 1.11 Stock of disability benefit recipients, 1968–2009

Source: UWV, Statistische Tijdreeksen UWV 2008

Source: UWV, Statistische Tijdreeksen UWV 2008

Source: UWV, Statistische Tijdreeksen UWV 2008

Figure 1.14 Inflow rates WAO/WIA by gender, 1967–2009

Source: UWV, Statistische Tijdreeksen UWV 2008

Source: UWV, Statistische Tijdreeksen UWV 2008

1.5.2 Disability insurance in the Netherlands in international perspective

1.6 Joint analysis of ageing and disability

1.6.1 The ageing-disability relationship in numbers

9 The large number of removals is mainly due to missing wages.

11 The transformation matrix is available upon request.