Elderly Labour Force Participation in the
European Union
How Demographic Changes Will Contribute
JEL classification: C33, J11, J14, J21
July 2008
Master thesis by Bram Veenvliet Student number: 1281771 Supervised by Dr. J.P. Elhorst
Abstract
The upcoming ageing problems of many EU countries resulted in a renewed recognition of the importance of elderly labour force participation. Researching elderly participation policy has become very popular, and EU countries are designing policy measures to get older people to re-enter or stay on the labour market. In the light of several upcoming demographic changes, this thesis focuses on demography rather than policy. For this purpose, a simultaneous model is constructed to examine how changes in demography will affect elderly labour force participation, using data of 236 EU regions over the period of 1999-2006. A sensitivity analysis is performed for the Netherlands in order to see how demographic changes during the upcoming decade will translate to elderly labour force participation
Contents
1. Introduction ... 3
2. Related literature ... 5
2.1 Health and participation ... 5
2.2 Schooling and participation ... 6
2.3 Unemployment and participation ... 8
3. A model of elderly participation ... 9
3.1 The model ... 9
3.2 Sample and data description ... 11
4. Model outcomes and sensitivity analysis ... 14
4.1 Model outcomes ... 14
4.2 Sensitivity analysis ... 16
4.2.1 Scenarios regarding elderly schooling, health and unemployment ... 16
4.2.2 Sensitivity analysis outcomes ... 18
5. Conclusions ... 20
1.
Introduction
The Netherlands is one of the EU countries where ageing will be severe in the decades to come. Where there are currently 23 people over 65 years of age per 100 people between 20 and 65 years of age, implying a grey pressure1 of 23%, demographic outlooks predict that this will more than double to 47% in 20382. This development increases the number of elderly people3 in need of social services relative to the younger people who finance these services, which will result in major increases in the burden of the social services of the elderly on the young.
If the same quality of these social services is going to be maintained, costs will rise dramatically. However, if these costs can be absorbed by more hours of labour, it is easier to keep the current quality of the social services of elderly affordable. Higher employment levels and participation rates are therefore crucial in our ability to deal with the costs of ageing. The elderly themselves have an important position in this matter. If labour force participation rates among the elderly increase, benefits are twofold: there are more workers that can carry the burden of the ageing population, but there are also fewer elderly that demand social services. It is thus no surprise that the Dutch government put considerable effort in implementing policy to prevent early exits from the labour market during the last decade.
Although elderly participation seems to respond quite well to active labour market policy4, it is certainly not solely determined by it. There are other factors that are very rigid in terms of policy, but do influence the decision of the elderly to re-enter the labour market. Due to demographic changes, three of these factors are expected to change structurally in the decades to come and will be the focus of this thesis.
First of all, there is unemployment. How the elderly perceive their odds of succeeding at getting a satisfying job can be decisive in choosing whether or not to stay on or re-enter the labour market. The idea is that higher unemployment leads to lower expected wages, but also that difficulties of getting a job negatively influences the number of people that go and look for one5. Labour market tightness is therefore expected to lead to labour market withdrawal. This unemployment effect might be particularly relevant for older workers, as they have increased opportunities to retire early and maintain a sufficient income. Interestingly, the ageing society will increase the tightness of the labour market in the decades to come. This will put downward pressure on unemployment rates, and should encourage more elderly people to re-enter the labour market.
1
The grey pressure is the number of elderly people relative to the potential labour force.
2
Source: CBS StatLine
3
In the remainder of this thesis, the elderly people are all people between 55 and 64 years of age.
4
For example, the Dutch labour force participation rate among men between 55 and 64 years of age decreased from 80% in 1970 to 38% in 1994 (source: CBS StatLine), while that of other male age groups have remained approximately the same. This was largely the result of a policy that stimulated the elderly to retire early in order to provide vacancies to the youth. Elderly participation rates started to rise again as soon as this policy was reversed (Soede and Bijkerk, 2002).
5
The second factor is education. A lower level of education limits skills and the jobs for which one is capable, and thus influences the decision of whether or not to re-enter the labour market. Since younger generations are generally better educated, it is expected that the elderly will be better educated in the future, and thus more people who will look for a job or keep working.
The last factor is health. Bad health generally decreases capability and mobility, and therefore refrains people from re-entering or stay on the labour market. Due to better nutrition, hygiene and health care, elderly people are increasingly healthy. We can expect a healthier labour force in the future, and thus more people that are able and willing to participate on the labour market.
In the light of these three changes, the expected impact of the three factors, and the upcoming ageing problems, I attempt to quantify how changes in these three factors influence the elderly labour force participation rate. The analysis focuses on the Netherlands.
The model that is used is estimated by two-stage least squares, in which elderly labour force participation and elderly unemployment are considered endogenous. Data is mostly regional, and for the specific age group of interest (people who are between 55 and 64 years of age). Regional data comes from Eurostat. The sample consists of all NUTS-2 regions that are part of the EU and the OECD. Regional characteristics are accounted for by including region dummies, and time patterns are accounted for by year dummies. The model will be used to perform a sensitivity analysis that shows how different combinations of the three factors influence elderly participation in the Netherlands at the end of the upcoming decade (2008-2018 that is).
2.
Related literature
During the last decade or so, elderly participation in the EU has received a lot of academic attention. The OECD initiated the “Pensions at a Glance” paper series in 2003, started collecting health data in 20006 and wrote many papers on the incentives that elderly have to participate on the labour market. In 2004, the Survey of Health, Ageing and Retirement in Europe (SHARE) started collecting micro data on health, socio-economic status and social and family networks of more than 30,000 individuals aged 50 or over, to facilitate research in that area. Especially Netspar, the Network for Studies on Pension, Ageing and Retirement that was founded in 2005, and the Center for Economic Research, made use of the SHARE data to produce several interesting contributions.
In this chapter, literature from the sources mentioned in the previous paragraph, as well as other strands of related literature, are discussed to see what academic research has rendered so far. This is done separately for each factor.
2.1 Health and participation
As already mentioned, the SHARE database has greatly facilitated studies on the effects of health on elderly labour force participation. It is a cross-national database (including 11 EU countries) that contains information on different aspects of health and socioeconomic variables of people that are between 50 and 64 years of age. The most interesting feature of de data is that the different dimensions of the data allow researchers to see how the effects of the various aspects of health on elderly participation depend on socioeconomic characteristics, and on the country. Kalwij and Vermeulen (2005) were the first to utilize the SHARE data in such a fashion. Their main results are the following:
§ Men between 50 and 64 years of age who have experienced a severe health condition participate considerably less on the labour market. The effect is smallest for Germany, where having dealt with a severe health condition leads to a 13% lower probability of being active on the labour market, and largest for Spain (25%) and Austria (31%)7.
§ Women between 50 and 64 years of age who have experienced a severe health condition participate considerably less on the labour market as well. The effect is smallest for the Netherlands, where having dealt with a severe health condition leads to an 11% lower probability of being active on the labour market, but much larger for Denmark (18%) and France (28%).
§ Men between 50 and 64 years of age who experience daily living restrictions participate considerably less on the labour market. The effect is smallest for the Netherlands, where a
6
As did Eurostat.
7
severe health condition leads to a 10% lower probability of being active on the labour market, but much larger for Spain (26%).
§ Women between 50 and 64 years of age who experience daily living restrictions participate considerably less on the labour market as well. The effect is smallest for Germany and the Netherlands, where a severe health condition leads to a 9% lower probability of being active on the labour market, but much larger for Sweden (15%).
§ The impact of elderly health on elderly participation is significant at the 5% level in all of the 11 EU countries that are in the sample.
Thus, Kalwij and Vermeulen not only show that the impact of health on participation is quite substantial in EU countries, but also that this effect differs per country. In the analysis of this thesis, one health effect is estimated for all regions across 19 different countries. Chapter 3 shows how this problem is addressed8. A point of critique is that the analysis of Kalwij and Vermeulen is quite blunt. Results would be much more meaningful and accurate if regional data was available, and if the age group (50-64) would be divided into smaller parts.
2.2 Schooling and participation
Although it was not their main point of interest, Kalwij and Vermeulen also included education in their analysis, and found the following:
§ Men between 50 and 64 years of age who have attained tertiary education participate considerably more on the labour market. The effect is largest for France, where a tertiary degree leads to a 34% higher probability of being active on the labour market, but much smaller for the Netherlands (14%). In some countries, the effect was insignificant at the 5% level. This may be due to the bluntness of the analysis mentioned before.
§ Women between 50 and 64 years of age who have attained tertiary education participate considerably more on the labour market as well. The effect is largest for Italy, where a tertiary degree leads to a 47% higher probability of being active on the labour market, but much smaller for Sweden (11%)9.
§ Men between 50 and 64 years of age who have attained secondary education participate considerably more on the labour market. The effect is largest for France, where a tertiary degree leads to an 18% higher probability of being active on the labour market, but much smaller for the Netherlands (10%)10.
§ Women between 50 and 64 years of age who have attained secondary education participate considerably more on the labour market as well. The effect is largest for Spain, where a
8
The same goes for schooling and unemployment: it is fair to assume that the effects of these factors are different across countries. Again, this issue is further discussed in chapter 3.
9
In some countries, the effect was insignificant.
10
tertiary degree leads to a 22% higher probability of being active on the labour market, but much smaller for the Sweden (8%)11.
It is clear that education is very decisive in the participation decision of the elderly. These findings are confirmed by data on elderly labour force participation by educational level in tables 1 and 212, which show that participation rates differ greatly across different educational levels.
Table 1. Participation rate of men, aged 55-64, by education level, 2005 Country Primary Secondary Tertiary
Austria 35.1 40.4 58.8 Belgium 31.6 47.3 62.5 Denmark 53.6 68.8 79 France 35.8 43.9 64.1 Germany 52.1 57.6 71.7 Greece 60.1 56.8 70 Italy 37 51 76.2 Netherlands 52 59.1 68 Spain 60.8 65.3 70.3 Sweden 68.5 77.8 84.6 Switzerland 68.8 76 84.9
Table 2. Participation rate of women, aged 55-64, by education level, 2005 Country Primary Secondary Tertiary
Austria 19.5 23.3 44.8 Belgium 15.4 31.1 36.2 Denmark 39.1 59.8 72.1 France 33.3 39.2 52.1 Germany 33.6 43.4 60.2 Greece 26.3 22.9 45.2 Italy 14.7 36.5 55.3 Netherlands 26.6 40.6 57.1 Spain 24.1 42.2 61.1 Sweden 53.2 68.6 85.3 Switzerland 47.2 60.5 71.6
It is remarkable to see how the Kalwij and Vermeulen results differ in magnitude across countries. Before concluding that cross-country differences are indeed large, we should not forget that the lack of regional and age distinction and the subsample composition (see footnote 7) might also partly account for the estimated differences. However, it seems that large differences are supported by the data from
11
In some countries, the effect was insignificant.
12
table 1 and 2. For example, differences in participation rates of men between 55 and 64 years of age depend relatively little on educational level in countries where Kalwij and Vermeulen find insignificant results such as Greece and Spain, whereas the opposite holds for France. Also, for women between 55 and 64 years of age, differences in participation rates depend relatively little on educational level in France, Greece and Sweden, countries for which Kalwij and Vermeulen find insignificant results or low scores.
2.3 Unemployment and participation
As already mentioned, there are two main channels through which labour market tightness influences participation rates. First of all, people expect wages to decrease with lower tightness. Second, low tightness also leads to discouragement among workers: difficulties to get a job negatively influence the expected search duration, and thus the number of people that remain active in the labour market. The search duration argument seems especially relevant for the elderly: their unemployment rate is generally lower than the unemployment rate of the young, but the incidence of long-term unemployment is generally much higher (OECD, 2006). This indicates that although more older people work, the consequences of losing their job are more severe in terms of unemployment duration. In addition, the negative effect of unemployment is expected to be larger for the elderly, because many of them have relatively little costs of labour market inactivity. For example, in several European countries, early retirement arrangements and disability pensions are widely used as a precursor to the standard pension in order to be inactive while still receiving a sufficient income.
So we might expect low labour market tightness, generally proxied by the unemployment rate, to have a negative effect on labour force participation, especially for the elderly. As already mentioned, there is a heap of literature on this subject. However, a large part does not manage to deal with the simultaneous character of the relationship between unemployment and participation13, 14. Also, most studies are limited to national data. When regional disparities are not accounted for, the analysis might not be able to provide a sufficient understanding of labour force participation (Elhorst, 1996). In his own research, Elhorst (2007) takes both considerations into account. He finds a significant negative relationship between unemployment and elderly labour force participation for both males and females, with regression coefficients of −0.188 and −0.172 respectively. However, Elhorst uses the general unemployment rate in the participation equation whereas this paper uses the elderly unemployment rate. Elderly unemployment is expected to better proxy elderly discouragement and elderly job opportunities than general unemployment, as it specifically reflects the degree to which the elderly have difficulties at finding a job.
13
Firstly discussed in Fleisher and Rhodes (1976).
14
Smith (1999) finds that there is not only a dual relationship between unemployment and participation, but also between health and participation. Accounting for this is beyond the scope of this paper, and might be an
3.
A model of elderly participation
In this section, the elderly participation model is presented. This model must allow us to see how changes in the three factors influence the labour force participation rate of the elderly. The first section explains the construction and variables of the model, the second section describes the data with which the model is calibrated.
3.1 The model
The basis of the model is a linear equation that estimates the elderly labour force participation rate, including the three variables of interest: elderly schooling, elderly unemployment and elderly health. In order to prevent biased estimators, the equation contains a region dummy for all but one region to account for all kinds of region-specific differences15 and a year dummy for all years to account for time patterns.
However, as was mentioned before, there exists a simultaneous relationship between the unemployment rate and the labour force participation rate. That is, the unemployment rate is influenced by the labour force participation rate, while the labour force participation rate is influenced by the unemployment rate at the same time. This interdependence must be incorporated in the model to determine the relationship without measurement errors (Fleisher and Rhodes, 1976). It is therefore that the model is estimated bya two-stage least squares regression analysis, in which elderly labour force participation and elderly unemployment are considered endogenous. In the first stage, the expected value of the elderly unemployment rate is estimated by the exogenous variables at hand. Subsequently, these estimated values are substituted in the elderly labour force participation equation so that the regression coefficients can be estimated.
Besides region and year dummies, two instrument variables for elderly unemployment are added to the first-stage estimate. First we have a variable for general labour market tightness. This variable is expected to have a negative effect on the elderly unemployment rate: higher labour market tightness increases labour demand, which in turn decreases elderly unemployment. The second variable that will be included is the unemployment trap. The unemployment trap represents all monetary incentives to be unemployed. For example, if we consider the replacement rates for initially unemployed people16 in table 317, we see that there are large cross-country differences in monetary incentives to leave unemployment for a low-wage job. For example, single, Irish, low-wage earners more than double their net monetary receipts when flowing from initial unemployment into employment, whereas the same Luxembourgers gain less than 20%. This may explain part of the regional differences in unemployment. In addition, the elderly generally have more incentives to be
15
A dummy for one region is left out to prevent singularity problems.
16
Defined as the ratio of a population's average social initial unemployment benefits and its average corresponding income.
17
unemployed than other age groups, because of their closeness to the retirement age. A significant unemployment trap may therefore be even more relevant for the elderly than it is for other age groups.
Table 3. Net replacement rates at the initial phase of unemployment, 2005
67% of average wage
No children Two children
Single person One-earner married couple Two-earner married couple Lone parent One-earner married couple Two-earner married couple Austria 55 57 81 70 72 85 Belgium 77 67 81 75 71 83 Czech Rep. 56 57 76 63 57 85 Finland 70 81 80 87 85 85 France 75 70 87 83 83 87 Germany 60 61 89 78 78 93 Greece 49 52 67 62 65 71 Hungary 52 55 76 66 66 80 Ireland2 43 68 72 62 70 76 Italy 62 61 82 62 64 86 Luxembourg 85 83 91 90 90 94 Netherlands 70 84 84 84 86 85 Norway 66 68 83 94 74 86 Poland 74 77 75 99 69 79 Portugal 77 75 90 86 85 91 Slovak Rep. 61 58 84 60 57 85 Spain 76 75 88 78 77 89 Sweden 82 82 91 91 89 92 UK 58 58 60 72 70 65
The model is formulated as follows:
PART5564 = β11UNEM5564 + β12SCHOOL5564 + β13HEALTH5564 + β14REGIONDUM(1) +
… + β12+NREGIONDUM(N–1) + β13+NYEARDUM(1) + … + β12+N+MYEARDUM(M)
There are two issues concerning the model in this form, which have to be addressed. First, as was pointed out by Elhorst (1996), static regression equations with regional panel data are highly likely to display serial autocorrelation and/or heteroskedasticity. Depending on the values of the Durbin-Watson statistic and the Lagrange Multiplier heteroskedasticity statistic, this will be corrected for. Second, as was mentioned in chapter 2, even though the effects of the three factors are likely to be different across regions and countries, the model does not control for interaction effects among the explanatory variables. In chapter 1 we have seen that the effect of health on elderly participation depends on many other variables. This is something that has to be remembered when interpreting the results: the regression coefficients represent the entire sample, and are not necessarily representative for a single country or region. On the other hand, the degree to which the effects are different across countries and regions must not be exaggerated. In addition, the heterogeneity of the countries in the sample18 ensures that regional disparities in the effects of the independent variables are mitigated. Therefore, when performing the sensitivity analysis for the Netherlands in the next chapter, it seems reasonable to assume that the model outcomes are sufficiently representative for the Netherlands19.
3.2 Sample and data description
The system of equations is calibrated using unbalanced panel data of 19 EU OECD countries20. The countries are selected on the basis of data availability and homogeneity purposes. For all countries, data on all variables is available for 1999-2006, so that eight years of data are used. This leads to a total number of 1888 observations. The list below gives a description of the data that is used to proxy the model variables21.
PART5564 This variable is proxied by regional data on participation rates of people between 55 and 64 years of age. Data comes from Eurostat.
UNEM5564 This variable is proxied by regional data on unemployment rates of people between 55 and 64 years of age. Data comes from Eurostat.
SCHOOL5564 This variable is proxied by national data on the percentage of people that are between 55 and 64 years of age, who have attained at least upper secondary schooling. Data comes from OECD’s Education at a Glance.
HEALTH5564 This variable is proxied by national data on the average expected years to live of people that are 65 years of age. Data comes from Eurostat.
18
See next paragraph.
19
As it would be for any other country in the sample.
20
Austria, Belgium, Czech Republic, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxembourg, The Netherlands, Norway, Poland, Portugal, Slovakia, Spain, Sweden, and the UK.
21
LMTIGHT This variable is proxied by regional data on the unemployment rate of people between 25 and 54 years of age.22 Data comes from Eurostat.
UNEMTRAP5564 This variable is proxied by national data on the number of people between 55 and 64 years of age that is unemployed for more than a year as a percentage of the number of people between 55 and 64 years of age that is unemployed. Data comes from the OECD database.
The last proxy deserves further explanation. As was explained before, the unemployment trap represents all monetary incentives to be unemployed. If these monetary incentives are present, the elderly are induced to be unemployed for longer periods of time, or might use unemployment as a precursor to the general pension. The number of elderly people that are unemployed for longer periods of time as a percentage of the number of elderly people that are unemployed might therefore be a good proxy for all the incentives to be unemployed.
Finally, the tables below show descriptive statistics per variable, and how the variables are related. We will find that the relatively low standard deviation of the health variable will result in a relatively high regression coefficient for HEALTH5564. Also note that participation and unemployment lie between 0 and 1, and health and schooling do not. The relatively low cross-correlations in tables 5-7 make sure that cross-correlations between explanatory variables are not problematic in the two-stage least squares regression analysis.
Table 4. Descriptive statistics Mean SD UNEM5564 0.057 0.044 SCHOOL5564 49.64 19.87 HEALTH5564 16.15 1.207 PART5564 0.429 0.125 LMTIGHT 0.073 0.045 UNEMTRAP5564 56.25 13.53
Table 5. Cross-correlations participation equation UNEM5564 SCHOOL5564 HEALTH5564
UNEM5564 1 0.22 - 0.19
SCHOOL5564 0.22 1 - 0.09
HEALTH5564 - 0.19 - 0.09 1
22
Table 6. Cross-correlations unemployment equation PART5564 LMTIGHT UNEMTRAP5564
PART5564 1 - 0.52 - 0.37
LMTIGHT - 0.52 1 0.28
UNEMTRAP5564 - 0.37 0.28 1
Table 7. Cross-correlations exogenous variables
SCHOOL5564 HEALTH5564 LMTIGHT UNEMTRAP5564
SCHOOL5564 1 - 0.09 - 0.21 - 0.16
HEALTH5564 - 0.09 1 - 0.24 0.02
LMTIGHT - 0.21 - 0.24 1 0.28
4.
Model outcomes and sensitivity analysis
The first paragraph of this chapter presents and discusses the outcomes of the model that was introduced in the previous chapter. In the second paragraph the sensitivity analysis is explained, and its results are presented and discussed.
4.1 Model outcomes
In order to see whether or not autocorrelation and/or heteroskedasticity exist, the model is estimated without any correction for the two potential problems.
Table 8. 2SLS estimation results, not corrected for autocorrelation and heteroskedasticity Participation Coefficient t-value UNEM5564 - 0.259441 - 3.07 SCHOOL5564 0.001055 2.73 HEALTH5564 0.014632 3.11 Value Durbin-Watson 1.145 LM F-stat 85.01 R squared 0.958
All three variables of interest have signs that could be expected on the basis of chapter 1. That is, given the existing regional disparities and time effects, the unemployment rate of people between 55 and 64 is negatively related to the participation rate of people between 55 and 64, and the average expected years to live of people that are 65 years of age and the percentage of people that are between 55 and 64 years of age who have attained at least upper secondary schooling are positively related to the participation rate of people between 55 and 64. The t-values indicate that these effects are significant at the 1%-level for all three variables. In addition, the R squared is relatively high, indicating that nearly 96% of all changes in the dependent variable is explained by the set of independent variables.
The relatively low Durbin-Watson statistic also confirms what was to be expected: there is a probability of less than 1% that the regression residuals do not display serial correlation. Software package Eviews 6.1 features the first-order autoregressive model to deal with this problem. In this model, the error terms depend on their lagged value plus another random component that is uncorrelated over time and has zero mean and constant variance. After controlling for heteroskedasticity and autocorrelation, the results are as follows.
Table 9. 2SLS estimation results, corrected for autocorrelation and heteroskedasticity Participation Coefficient t-value UNEM5564 - 0.307953 - 2.62 SCHOOL5564 0.001287 2.04 HEALTH5564 0.013740 2.19 Value R squared 0.966
Correcting for heteroskedasticity and autocorrelation leaves results largely unchanged. The coefficients for unemployment, health and schooling have increased somewhat. Significance levels decreased but remain below the 5%-level. The R squared has increased somewhat due to higher efficiency and lower standard errors.
An important question is how the significance levels, signs, and magnitudes of the coefficients relate to results of previous studies. First, the UNEM5564 coefficient implies that if the elderly unemployment rate increases with one percent point, the elderly labour force participation rate decreases with 0,308 percent point. To my knowledge, there is no study that has attempted to quantify the impact of elderly unemployment on elderly participation. The study that comes closest is Elhorst (2008). This is a regional analysis including EU countries, in which the total unemployment rate is regressed on elderly labour force participation. This study finds a negative and significant effect as well. Note that while there is little comparable research, there is a huge quantity of literature finding a significant discouragement effect for other age groups and other regions.
Second, the SCHOOL5564 coefficient implies that if the percentage of people that are between 55 and 64 years of age, who have attained at least upper secondary schooling increases with one percent point, the elderly labour force participation rate increases with 0,1287 percent point23. Kalwij and Vermeulen (2005) include both secondary and tertiary in their analysis. They find coefficients for secondary education mostly below 0.1287, and coefficients for tertiary education mostly above 0.1287. This is an indication that the results are in line.
23
Third, the HEALTH5564 coefficient implies that if the life expectancy at age 65 increases with one year, the elderly labour force participation rate increases with 1.37 percentage point. This magnitude seems rather high, but is mainly a consequence of the fact that the range of the health variable is very limited24. Other studies on the effect of health on labour force participation mainly use micro data such as SHARE, ELSA and HRM. Although magnitudes are therefore not comparable, these studies show that health has a significant and considerable effect on labour force participation25.
4.2 Sensitivity analysis
With the coefficient estimates in table 9, it is possible to see how changes in UNEM5564, HEALTH5564, and SCHOOL5564 translate to changes in PART5564. This allows us to see how different scenarios regarding the three variables influence elderly participation in the future. This sensitivity analysis focuses at the Netherlands and the end of the upcoming decade (2018). To begin with, the first subparagraph attempts to forecast the most realistic scenarios regarding the three variables during the following decade. In the second part of the paragraph we will see how these scenarios translate to changes in the elderly labour force participation rate.
4.2.1 Scenarios regarding elderly schooling, health and unemployment Elderly health development
Historical health data is presented in table 10 and 11. The data shows that the elderly live longer than a decade ago, and also that this development seems to continue: there is no satiation whatsoever. However, it seems reasonable to account for the fact that such a process cannot last forever and satiation may come into effect in the near future. Therefore, 50%, 75%, 100%, and 125% of the average increase in life expectancy of 65 year-olds of the decade 1997-2006 are considered as the most likely scenarios for the upcoming decade.
Table 10. Life expectancy of 60,5 year-olds in the Netherlands26
Males Females 1971-1976 16.61 20.59 1976-1981 16.88 21.67 1981-1986 17.18 22.32 1986-1991 17.50 22.63 1991-1996 17.91 22.72 1996-2001 18.48 22.92 2001-2006 19.44 23.36 24
With a minimum of 12,9 and a maximum of 18,2, and a standard deviation of 1,2.
25
See for example Campolieti (2002), van Gameren (2007), Kalwij and Vermeulen (2005), Kerkhofs, Lindeboom and Theewes (1999), and Stern (1989).
26
Table 11. Life expectancy of 65 year-olds in the Netherlands27 Males Females 1995 14.7 19.2 1996 14.8 19.2 1997 15.1 19.3 1998 15.1 19.4 1999 15.2 19.2 2001 15.6 19.4 2002 15.6 19.3 2003 15.8 19.5 2004 16.3 19.9 2005 16.4 20.1 2006 16.8 20.3
Elderly schooling development
In attempting to forecast the development of SCHOOL5564 in ten years, the degree to which 45-54 year-olds have at least upper secondary education is of course crucial: they will be between 55 and 64 years of age ten years from now. Therefore, the 2018 percentage of 55-64 year-olds who have at least secondary education is expected to be equal to the 2008 percentage of 45-54 year-olds who have at least secondary education. This is the only scenario that is considered in the sensitivity analysis. Table 12 shows that the current 45-54 year-olds28 are far better educated than the current 55-64 year-olds. This will ensure a boost in the educational level of the elderly in the following decade.
Table 12. Projected education in the Netherlands, 200829
Age % with at least upp. sec. edu
55-64 63
45-54 73
35-44 78
25-34 82
Elderly unemployment development
As already mentioned, the ageing society will tighten the labour market. Employers will find it harder to find a suitable workforce, so that the odds of finding a job, as well the wages offered for them, increase. This will put downward pressure on the number of elderly people who are looking for a job, but also on the number of elderly people who are voluntarily unemployed. However, it remains impossible to reliably forecast unemployment rates for ten years from now. Therefore, a relatively broad range of unemployment rates is used in the sensitivity analysis: 1 to 4.5 percent. This range is based on historical elderly unemployment rates in the Netherlands, shown in table 13.
27
Source: Eurostat.
28
As well as all the younger generations.
29
Table 13. Elderly unemployment rates in the Netherlands30
Year Unemployment rate
1987 4.3 1988 4.0 1989 3.3 1990 3.7 1991 3.4 1992 2.7 1993 2.9 1994 3.5 1995 3.0 1996 3.2 1997 2.5 1998 2.3 1999 2.9 2000 2.1 2001 1.7 2002 2.0 2003 3.1 2004 3.8 2005 4.5 2006 4.4
4.2.2 Sensitivity analysis outcomes
Table 14 shows how the different scenarios that were selected in the previous subparagraph affect the elderly labour force participation rate. The forecasted increase in the percentage of 55-64 year-olds that have attained at least upper secondary education (from 63% in 2008 to 73% in 2018) is incorporated.
Table 14. Expected absolute change of the elderly labour force participation rate due to changes in elderly health, elderly unemployment, and elderly schooling, the Netherlands, 2008-2018
Increase in the life expectancy of 65 year-olds 2008-2018 Unemployment rate 2018 1 1,5 2 2,5 1 0.036 0.043 0.050 0.056 1,5 0.034 0.041 0.048 0.055 2 0.033 0.040 0.047 0.053 2,5 0.031 0.038 0.045 0.052 3 0.030 0.037 0.043 0.050 3,5 0.028 0.035 0.042 0.049 4 0.027 0.033 0.040 0.047 4,5 0.025 0.032 0.039 0.046 30
A first, important observation is the fact that the worst case scenario implies a 2.5 percentage point increase in the Dutch elderly labour force participation rate in 2018, compared to the current participation rate. However, this result is not limited to its time span. Recall that the changes in elderly health, elderly unemployment, and elderly schooling are expected to last much longer than one decade. Since there are no indications that the elderly will become less healthy, the labour market is expected to be relatively tight as long as the grey pressure is higher than now, and younger generations remain far better educated than older generations. Thus, the same demographic changes that force policy makers to take budgetary precautions are not only expected to drive up elderly participation rates by at least 2.5 percentage point in ten years, but they are also expected to last31 for several more decades at the least. This will provide valuable extra budgetary room in the difficult times to come.
Table 14 also shows that the elderly labour force participation is relatively sensitive to the different scenarios of elderly unemployment and health: the more optimistic scenarios show increases of at least 4 percentage points. Increases of this magnitude would be very welcome for the Dutch policy makers in the decades to come, especially in times of high grey pressure. Furthermore, the increases from the table require no policy effort, and can thus be considered as a bonus on top of policy achievements.
31
5.
Conclusions
The upcoming ageing problems of many EU countries resulted in a renewed recognition of the importance of elderly labour force participation. By spreading the ageing costs over more hours of labour, it is easier to keep the current quality of our welfare state affordable. People that are able to work must therefore be encouraged to actively participate in the labour force. For example, according to the Social-Economic Council of the Netherlands (2006), nearly half of the grey pressure burden32 can be absorbed by increasing participation rates. Elderly participation has a crucial role in this matter. Indeed, by increasing labour force participation rates among the elderly, one kills two birds with one stone: there are more workers that can carry the burden of the ageing population, and there are fewer elderly that demand social services.
In the light of several upcoming demographic changes, this thesis has focused on demography rather than policy. In many EU countries, elderly people are expected to be structurally healthier, to have a lower probability of unemployment, and to be better educated. The analysis of this paper has attempted to quantify how changes in these three factors influence the elderly labour force participation rate for a set of regions across 19 EU OECD countries. The results of the highly explanatory model show a negative significant coefficient for unemployment, and positive significant coefficients for health and education. These results have been incorporated into a sensitivity analysis for the Netherlands, which has been used to analyze how different scenarios regarding the three variables influence elderly participation in the future. This analysis has shown that the Dutch elderly labour force participation rate will increase by at least 2.5 percentage point during 2008-2018, but also that the elderly labour force participation is relatively sensitive to the different scenarios of elderly unemployment and health: the more optimistic scenarios show increases of at least 4 percentage points. We can thus conclude that the same demographic changes that force Dutch policy makers to take budgetary precautions are expected to drive up elderly participation rates by at least 2.5 percentage point in ten years. Moreover, due to the structural nature of the demographic changes, the participation increase is expected to last for several more decades at least. This will provide valuable extra budgetary room in the difficult times to come.
32
References
Campolieti, M. (2002), Disability and the Health of Older Men in Canada, Labour Economics, vol. 9, no. 3, pp. 405-432
Elhorst, J.P. (1996), A Regional Analysis of Labour Participation Rates across the Member States of the European Union, Regional Studies, vol. 30, no. 5, pp. 455-465
Elhorst, J.P. (2008), A Spatiotemporal Analysis of Aggregate Labour Force Behaviour by Sex and Age across the European Union, Journal of Geographical Systems, vol. 10, no. 2, pp. 167-190
Fleisher, B.W. and G. Rhodes (1976), Unemployment and the Labor Force Participation of Married Men and Women: a Simultaneous Model, The Review of Economics and Statistics, vol.58, no. 4, pp. 398-406
Gameren, E. van (2008), Labour Force Participation of the Mexican Elderly: the Importance of Health, Estudios Economicos, vol. 23, no. 1, pp. 89-127
Kalwij, A. and F. Vermeulen (2005), Labour Force participation of the Elderly in Europe: the Importance of Being Healthy, IZA discussion paper 1887, Bonn
Kerkhofs, M., M. Lindeboom, and J. Teeuwes (1999), Retirement, Financial Incentives and Health, Labour Economics, vol. 6, no. 2, pp. 203-227
OECD Ageing and Employment Policies (2006), Live Longer, Work Longer, Paris
Smith, J.P. (1999), Healthy Bodies and Thick Wallets: the Dual Relation between Health and Economic Status, Journal of Economic perspectives, vol. 13, no. 2, pp. 145-166
Sociaal-Economische Raad (2006), Welvaartsgroei voor en door Iedereen, publication number 8, Den Haag
Soede, W. and A. Bijkerk (2002), De VUT Voorbij?, Breukelen: Nyfer
