The Thesis
Based on a review of literature in the field of pension adequacy evaluation, focusing in particular on the replacement rate as the dominant indicator in this area, we formulate the thesis that the replacement rate is not a sufficient pension adequacy measure in cross-country studies, as it does not capture the complexity of the measured phenomena. This complexity is reflected in the various dimensions of pension adequacy. In the empirical part of our study, we will focus primarily on the first two dimensions, namely protecting against poverty (dimension I) and consumption smoothing (dimension II), treating disproportions in pension adequacy between men and women (dimension III) only as a complementary, rather than the main dimension. The vital character of the first two dimensions is not only due to technical reasons (evaluation of dimensions I and II enables evaluation of dimension III), but also due to substantive reasons. Assessment of poverty among pensioners and pension income in fact renders possible an almost comprehensive evaluation of pension adequacy, while evaluation only of the discrepancies in poverty and income between the genders is practically unreliable because it does not allow for inference as to pension adequacy across the total population.
Adequacy Indicators Used in the Analysis
In order to prove our thesis, based on Eurostat/EU-SILCFootnote 2
data for the years 2007–2012, we consider a set of statistical indicators that measure pension adequacy. An important selection criterion for these indicators, apart from their information load related to a given dimension of adequacy, is their cross-country comparability represented by the EU attribute associated with a given indicator according to standards set by the European Commission (2006). EU attribute means that a given indicator is classified among the commonly agreed indicators that allow comparative analysis of the progress made by the respective European countries towards achieving the goals of the OMC. These indicators, unlike those with the NAT attribute, enable direct cross-country comparative analyses and have a clear normative interpretation. The criterion of cross-country comparability of the selected criteria is associated with the thesis posed in this paper. We are interested in evaluating the adequacy of a pension system not in one country but in several dozen countries.
In the case of the first dimension of pension adequacy—protecting against poverty—we assume the at-risk-of-poverty rate for pensioners (ARP) to be the most representative indicator. It relates to the group of people whose main activity status is ‘retired’ and it expresses the share of pensioners with an equivalised disposable income below the at-risk-of-poverty threshold. The poverty threshold is assumed to be 60 % of median equivalised income after social transfers. However, it is difficult to state definitely whether this is the actual poverty threshold, as it is a relative value and it is not associated with expenditure but rather with the income of a household. Thus, in one country, income at the level of 60 % of the median income may permit a living standard above the poverty threshold, while in other countries this may not be so. Besides, theoretically, in a country where all live in poverty, according to the methodology adopted by Eurostat, there are always those who earn an income above 60 % of the median income and “theoretically” live outside the poverty zone. Thus, a poverty threshold measured by income (e.g. median income) is imperfect. However, in the Eurostat database there is no better indicator to measure poverty in the pensioner population. Moreover, the at-risk-of-poverty rate for pensioners is an indicator the structure of which allows major independence from the aggregated replacement rate discussed below. According to Eurostat, the cross-country comparability of ARP is high, which means that “data across countries is comparable from 2005 onwards. EU-SILC is based on a common framework defined by harmonised lists of primary and secondary variables, common concepts, a recommended design, common requirements (such as imputation procedures, weighting, sampling error calculation) and classifications aiming at maximising comparability of the information produced”.Footnote 3
Concerning the other dimension of pension adequacy—consumption smoothing—there are two essential measures based on income: the replacement rate (based on income, not on expenditure), which is the most commonly used in studies discussing pension adequacy, as well as the relative median income ratio for the population aged 65+ . As previously mentioned, the replacement rate is defined in a number of ways, by applying different numerators and denominators. However, it is always some kind of relationship between income in the retirement period and income in the period of economic activity. Also, the replacement rate may be computed for the total pensioner population or for its respective cohorts. According to the methodology applied by Eurostat, the aggregated replacement ratio (ARR) is defined as the ratio of the median individual gross pensions of the 65–74 age category relative to median individual gross earningsFootnote 4 of the 50–59 age category, excluding other social benefits. It takes into account gross income, which can be regarded as a disadvantage, because social security contributions paid by pensioners are usually much lower than by working people. Moreover, using gross instead of net income in comparative analyses may be problematic and may distort conclusions relating to countries with a progressive income tax. In addition, the cross-country comparability of this indicator is restricted also by the use of different pension system constructions.Footnote 5 Since the replacement rate is calculated as the median income of persons whose earnings in the retirement period (and in the period of economic activity, too) differ substantially, it does not provide required information on the level of pensioner poverty. Of course, it can be expected that countries with a higher replacement rate have a lower ratio of pensioners living below the poverty threshold, however, this relationship is not at all an evident one. It could be much weaker in a country with strong income asymmetry in the pensioner group. The advantage of such a replacement rate is a broad age range of pensioners and working people whose income (or, more specifically, median income) is provided for in the indicator’s numerator and denominator, respectively. This means that information included in this indicator covers a higher percentage of the surveyed population, but consequently, it is less detailed, e.g. compared to replacement rates calculated separately for respective age cohorts. However, considering the fact that according to Eurostat this rate is meant to be used mainly in comparative cross-country analyses, which refer to the achievement of OMC objectives, such a construction is justified. Currently, it is the only replacement rate calculated for the purpose of monitoring OMC objectives. Another rate, the aggregate replacement ratio (including other social benefits), although it is provided for in the OMC set of indicators, is currently being developed.
Another method to measure pensioner income, alternative or complementary to the replacement rate in evaluating the second dimension of pension adequacy is the relative median income ratio for the population aged 65+ (RMI), which is defined as the ratio of the median equivalised disposable incomeFootnote 6 of persons aged 60 and over to the median equivalised disposable income of persons aged between 0 and 59. The structure of this indicator is fully consistent with the macro-scale definition of the pension system, and the denominator of this indicator provides for median disposable income of persons aged between 0 and 59. It should be noted that while ARR provides for gross income without any additional social benefits, the RMI provides for total income after tax divided by the number of the members of a household, according to OECD methodology, and as such it may be complementary to the replacement rate. However, the cross-country comparability of RMI is high (not restricted as in the case of ARR). Nevertheless, the complementarity of the two indicators must not be assumed a priori, since their similarities may be evaluated only through a statistical analysis of their data series. Their definitions and calculation methods suggest that rankings of the surveyed countries with respect to the two indicators may be to some extent divergent.
The last dimension of pension system adequacy—income disproportions between the genders—may be measured, according to Eurostat methodology, by absolute differences between the values of relevant indicators described herein above, but calculated separately for men and women. These are:
-
Gender differences in the at-risk-of-poverty rate of elderly people 65+ (ARP_GD),Footnote 7
-
Gender differences in the aggregate replacement ratio (ARR_GD),
-
Gender differences in the relative median income ratio of elderly people 65+ (RMI_GD).
These indicators are used as a consequence of the choice of indicators to measure the first two dimensions of pensions adequacy and their application does not require any further justification. Assessment of differences in the poverty and income levels between men and women is associated with the assumption that pension systems should function in such a way as to ensure the right level of adequacy regardless of gender.
Data and Methods
Our analysis is focused on cross-country correlations between the above indicators. The research employs a cross-sectional time series (panel data) in the years 2007–2012 drawn from the Eurostat statistical database. The analysis covers 29 European countries: Austria, Belgium, Bulgaria, Czech Rep., Cyprus, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, Switzerland and the United Kingdom.
In order to obtain more robust results, our correlation analysis is based on two independent study procedures. The first one is founded on Spearman’s rank correlation coefficient, which is the nonparametric version of the Pearson correlation coefficient and measures the strength of association between two ranked variables. This coefficient was used to analyse pairs of the above-mentioned indicators in order to verify the similarity of rankings of the surveyed countries with respect to these indicators, disregarding the analytical form of possible correlation. In the case of pairs of indicators that remain in a positive relationship, a Spearman’s rank correlation coefficient close to 1 means a large convergence of the rankings. In the case of negatively correlated pairs, a Spearman’s coefficient close to −1 denotes a large convergence of the rankings. A low absolute value of the Spearman’s rank correlation coefficient suggests lack of convergence between rankings. We calculate the correlation coefficient separately for each year (cross-sectional data) and jointly for all the countries (panel data). The results of the analysis are presented in Table 1. The other procedure aimed at correlation analysis is based on panel regression of the following dependent variables: ARR_total, ARR_male, ARR_female, RMI_total, RMI_male, RMI_female, ARP_total, ARP_male, ARP_female, ARR_GD, RMI_GD, and ATP_GD. In total, we estimate 12 models. For models based on the panel data the test for the variance of the intercept in groups, the Breusch–Pagan test, the Hausman test and the Wald test were applied in order to select the proper form of the model: with fixed effects (FE) or random effects (RE), and with or without time effects (Ajmani 2009). With regard to time effects, initially we assumed that they should be insignificant because intuition suggests that in the period of few years the changes in pension system, if even they occur, should be rather very weak. This results from the fact that pension systems usually evolve smoothly and reforms to them show their effects in the long, not the short term. The Wald test generally confirmed our assumptions, therefore we decided not to include time effects. The results of these estimations are presented in Table 2.
Table 1 Spearman’s rank correlation coefficient between variables characterising pension adequacy
Table 2 Panel regression models for variables characterising pension adequacy
Our analysis does not take into account causality between the variables, but only the correlation between them (in the sense of similarity in their changes). This is because causal dependencies analysis would require involving a number of other variables that, according to the theory and practice, could affect the level of pensioner income and at the same time the aggregated replacement ratio, relative median income and the at-risk-of-poverty rate of pensioners. However, this issue is not relevant to this paper.
Results
Our analysis was preceded by dispersion coefficients calculations for ARR, RMI and ARP indicators. These reveal discrepancies between the coefficients of variation for the three analysed ratios. The RMI indicator is the least variable as its coefficient of variation fluctuates at the level of 0.11–0.13 and is quite convergent across the surveyed groups (all pensioners, retired men, retired women). The aggregated replacement ratio is slightly more varied (0.14–0.22) across the surveyed groups than in the case of the ARR indicator. The ARP indicator is definitely the most variable in the cross-country analysis, incomparably more than RMI and ARR. Its coefficient of variation oscillates at the level of 0.47–0.70 and is the highest for men. The fact that the indicator representing the first dimension of pension adequacy is much more variable compared to the indicators representing its second dimension means that the surveyed countries differ more in terms of the level of pensioner poverty than the level of pensioner income, which is a measure of the rate of consumption smoothing in the life cycle.
The results presented in Tables 1 and 2 lead to a number of conclusions that support the thesis that the aggregated replacement ratio is not a sufficient or fully representative measure and that it does not fully characterise pension adequacy. Spearman’s correlation coefficients are statistically significant for all analysed pairs of variables representing the first and the second dimension of pension adequacy (ARR, RMI and ARP). However, their values prove only a moderate non-parametric correlation between the ARR and the RMI, and between the ARR and ARP, both for indicators relevant to the total population and for each gender. When comparing the correlation between ARR and ARP (total, male and female) with RMI and ARP (total, male and female) an important conclusion can be drawn. RMI is more correlated with ARP than with ARR. This means that RMI reflects the poverty among pensioners better than ARR, especially with respect to the female group.
The correlation is weak or it does not exist at all between disproportionate indicators of income and poverty between men and women (ARP_GD, ARR_GD, RMI_GD) and income and poverty indicators for the total pensioner population (ARP_total, ARR_total, RMI_total). This confirms that inference on the basis of indicators reflecting gender differences in pension adequacy across the total population is unjustified.
Additionally, an analysis of the results of the panel regression model estimation leads to the conclusion that the aggregated replacement ratio is quite well explained by the other variable representing the same dimension of adequacy, namely the relative median income, but it is not explained, in the same model, by the at-risk-of-poverty rate. RMI also explains well the variable of the at-risk-of-poverty rate of elderly people 65+. Interestingly, the statistically significant variables in the model for the RMI indicator are both ARR and ARP. The above results relate both to the models for the total population and to the models for the respective genders (male and female) and suggest that the relative median income is a variable that quite well reflects both dimensions of pension adequacy—protecting against poverty and consumption smoothing—better than the most commonly used indicator, i.e. the replacement rate. Moreover, the model for ARR_GD suggests that in countries with a higher relative median income (RMI_total) the difference between the aggregated replacement ratio between men and women is greater. Interestingly, ARR_total is a statistically insignificant variable in this particular model. On the other hand, in the model for RMI_GD, ARR_total is statistically significant and RMI_total is not. Thus, in countries with a higher replacement rate in the total pensioner population, the difference in the relative median income between men and women is also greater. Neither ARR_total, nor RMI_total explain the differences in poverty between men and women aged 65+.
The synthetic conclusion that may be drawn from the above statistical and econometric analysis of the main indicators of pension system adequacy, understood as a three-dimensional phenomenon, is somewhat surprising, as it suggests that the aggregated replacement ratio, at least as calculated by Eurostat based on EU SILC data, is not the best measure of pension adequacy, considering two of its dimensions: protecting against poverty and consumption smoothing. This confirms our thesis. A better measure is the relative median income, as it better explains, with respect to the first two dimensions of adequacy, the variables ARR_total and ARP_total, and at the same time is itself explained by the two variables. The first two dimensions of pension adequacy generally do not reflect the third dimension and thus ARR and RMI are not sufficient measures of poverty and income disproportions between genders in the population of pensioners.