The effects of pension systems on the savings rate : a cross country analysis on 42 countries over the years 1981, 2002 and 2013

(1)

The effects of pension systems on the savings rate

A cross country analysis on 42 countries over the years 1981, 2002

and 2013.

Written by

Jan Jaap Peters

10452125

Supervised by

Nicoleta Ciurila

(2)

2

Abstract

This study investigated the effect of pension funds on the savings rate in comparison to the PAYG system. Data is obtained from 42 countries over the years 1981, 2002 and 2003. Three panel data analyses are made using the replacement rates provided by the PAYG system and pension funds, based on a study by Bloom, Canning, Mansfield and Moore (2007). The results showed no significant effects of both replacement rates on the savings rate. What this study did find, is that countries with a high old age dependency ratio had a lower savings rate and that high income per capita had a positive effect on the savings rate. To find the effects of both pension systems on the savings rate, more variables should be added in further research to distinguish for their effects in the presence of borrowing constraints, universal coverage, differences in contribution base and retirement incentives.

Statement of originality

This document is written by Jan Jaap Peters who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the

(3)

3

Introduction

Pension systems have received increased attention the past years. Most countries traditionally used a pay as you go (PAYG) system to provide for pension benefits. In Europe, the PAYG system provides for eighty percent of retirement benefits (Bruil, Schmitz, Gebraad & Bhageloe-Datadin, 2015). In this system, taxes are levied from the working population and directly paid out to current retirees. It is part of the social security system in several countries. However, in relation to the effects of ageing on demographics, there is increased criticism on this form of providing retirement benefits. Critics argue that it is financially unsustainable, and that in the long run it can pose an unbearable burden on an ever shrinking working population. In the graph below fertility rates across the world are showed for the past fifty years, based on a five year interval.

Figure 1: Fertility rates across the world 1960-2010 (World Bank, 2016).

In the long run, lower fertility would lead to an ageing population with a demographic emphasis on elderly people, and thus more retirees. An increasing amount of retirees would put increased pressure on pension systems in place. There are two types of pension systems through which retirement benefits can be provided: the PAYG system or fully funded pension funds. Both systems will receive further clarification. In this study, we are interested in the macro-economic effects of both systems, especially the effects on the savings rates in different countries. Over the past years, a number of countries have reformed pension

(5)

5

systems to funded systems. This is because pension funds are associated with higher savings rates (Aguila, 2011).

The effects on the savings rate deserve attention, because multiple studies have pointed out the relation between savings rates and future economic growth. Alders and Broer (2005) have investigated the effect of lower fertility on economic growth. Despite the fact that lower fertility in most developed countries has been accompanied with higher labor participation rates, they state that due to higher taxation from increased social security savings rates will drop. The decrease in savings rates also decreases the capital stock, and thereby affects lending and investments. This effect would eventually depress future growth. Gyarfas and Marquardt (2001) reach the same conclusion on the effect of lower fertility on economic growth.

A conclusion on the effect of pension systems on the savings rate can help

governments in formulating efficient policies in order to promote future economic growth in a time of decreasing fertility. It is a problem of public importance, because individuals are dependent in later life on their retirement benefits. If pension systems prove to be insufficient, or are causing economic problems, then this could affect the well-being of future retirees.

In this study, I researched if funded pension systems do contribute to a higher savings rate in comparison to the PAYG systems. Panel data analyses were used to determine the effects of both pension systems on the savings rate in 42 different countries over the years 1981, 2002, and 2013. Bloom et al. (2007) collected data on pension systems in 1981 and 2002 to determine the effects of pension systems on the savings rate under different circumstances. The analysis is done based upon their research, except for the year 2013. Three panel data analyses are made to see if earlier findings on the effects of pension systems would hold, when the year 2013 is added. No significant results on the effects of pension funds on the savings rate in comparison to the PAYG system were found. Yet I have looked for possible improvements for further research on the effect of pension systems on the savings rate.

In section one, I will analyze the effects of pension system on the savings rate as discussed so far in literature. This literature review is split in four parts. First there is a description of the working of both pension systems, second the theory on their effects on the savings rate is discussed, third the empirical findings are analyzed and fourth I will discuss other determinants of the savings rate. In section two, we will look at the methodology, data and the results of the panel data analyses. The results are discussed in section three and here

(6)

6

we will also look at the limitations to this study. From this a suggestion for further research is made. In section four, the conclusion of this study is presented.

1 Literature review

In this section, the literature on the effect of pension systems on the savings rate is discussed. First the two types of pension systems are described. Then the theoretical effects of pension systems on the savings rate are examined. Thereafter the empirical findings regarding these effects are displayed. In the latter case, I will present the research methods employed in these studies. Finally, the control variables used by Bloom et al. (2007) are reviewed in order to determine their expected effect on the savings rate. The review also serves purpose as the basis for the collection of data and the selected methodology. 1.1 Funded pension plans and the PAYG system

There are two types of pension systems through which retirement benefits are provided. First there is the PAYG system. In this system, taxes are levied from the current working population which are redistributed to present retirees. The second type of retirement funding is to use funded pension plans. In such a system, individuals contribute to pension funds during their working lives. Upon retirement, these pension funds pay retirement benefits to their former contributors.

Both systems can be based on defined contribution (DC) or defined benefits (DB), either partly or fully. With DC it is the future retiree who bears the investment risk, since he is dependent on the final value of his contributions for his retirement benefits. In the other case, the provider or sponsor of the funds has the obligation to provide the contributor a pre-determined benefit, thereby bearing the risk. Differences in contribution base will not be examined in this study, because these variables are not included in the analysis by Bloom et al. (2007).

1.2 Theoretical background of pension systems’ effects on the savings rate

Samwick (2000) elaborates on pension funds and the PAYG system, and their potential effects on the savings rate. First we look at the effects of the PAYG system. This system transfers wealth from the current working population to present retirees. The latter group has a higher propensity to consume, thus the transfer lowers the savings rate. Next to this,

(7)

7

progressive benefits formulas in a PAYG system also redistribute wealth from households with unexpected high incomes to lower income households. Again, the latter group has a higher propensity to consume, thereby lowering the savings rate. Next to this, the formula also protects households from future income insecurity, if future earnings turn out to be low. This would have a negative effects on households’ decision making on precautionary savings. Progressive benefits formulas in PAYG systems are mostly seen in developed countries.

Progressive benefits formulas are less common in countries that switched to funded pension systems, due to high transition costs of implementing these formulas. Less

progressivity in the redistribution of retirement benefits would then positively affect the savings rate. After a transition to a funded system, households face lower taxes, thereby obtaining higher disposable income during their working lives. A larger part of the income increase would be saved for retirement, and this increases the savings rate. Further on, since pension funds tend to be more transparent in contributions and future benefits, a ‘recognition effect’ could grow among households. This would point them at the importance of saving for retirement.

Using the life cycle hypothesis (LCH), Aguila (2011) predicts the effects of pension funds on low and high income households and their savings behavior. According to the LCH, household assets and wealth from social security are imperfect substitutes. An increase in social security wealth would negatively affect households their desired savings rate during working life. In the model, pension funds would increase social security wealth for low income households of all age, due to the indexation to inflation of minimum retirement benefits. This would not be the case with the PAYG system, at least in the Mexican case. Pension funds would not increase wealth from social security for high income households. In conclusion, increased wealth from social security with pension funds could thereby crowd out private savings for low income households.

Bailliu and Reisen (1997) use the overlapping generations model (OLG) to predict the effect of pension funds on the savings rate in comparison to the PAYG system. The OLG is the common model to predict macro-economic effect of pension systems, and is also used in the study by Samwick (2000). As with the LCH, in this model households save during

working life to provide for their future retirement. The model assumes that household savings determine the capital stock, and that therefore aggregate savings are mainly determined by households. Their model predicts that pension funds would not increase the savings rate, because it crowds out voluntary savings. Therefore, if households their desired savings rate is

(8)

8

exceeded by their contributions to pension funds, they would make up by borrowing against future retirement benefits. This would leave the savings rate of households unchanged. A solution to this effect would be to implement borrowing constraints in order to increase households’ savings rates. They explain that most of the increase of the savings rate would result from low income households with a low propensity to save, who are then forced to save by mandatory contributions.

Next to this, the rate of return of pension funds would also affect the savings rate. If the rate of return of pension funds would be higher than that of private savings because of tax exemptions, then the desired savings rate decreases due to the future income effect. It would also imply that fertility rates affects the savings rate through pension systems, since lower fertility decreases the rate of return of the PAYG system. Dependent on the volatility of financial assets in pension funds or the credibility of the funded system, households can either lower or increase precautionary savings. In comparison to the PAYG system pension funds can enable early retirement. The opportunity of early retirement could induce high income households to increase their savings rate in order to facilitate for this (Steurer, 2009).

Gyarfas and Marquardt (2001), using the OLG model, have predicted that a savings subsidy for retirement funding can be used to induce households to increase their saving rates in comparison to the PAYG system. According to them, this measure would increase the total capital stock available. The crowding out effect on private savings from the higher rate of return due to tax exemption is less important in their conclusion. Their predictions are supported by a more recent study by Roberts (2013).

In conclusion, pension funds can be expected to increase the savings rate in comparison to the PAYG system. However, this would be dependent on borrowing

constraints to prevent households from decreasing savings. We can assume that in most cases these constraints are present, since retirement benefits are generally not accepted as collateral for loans (Bailliu & Reisen, 1997). Yet the increase can still be hampered by crowding out effects on private savings. Most of the increase in savings rates from pension funds can be expected from low income households by mandatory contributions. A recognition effect would also increase the savings rate. Progressive redistribution formulas negatively affect the savings rate, and are present more often in the PAYG system. Both pension systems can decrease precautionary savings, dependent on the rate of return and the reliability of the pension system. In the next section, the empirical results on these predictions are discussed.

(9)

9

1.3 Empirical results on the effects of pension systems on the savings rate

Bailliu and Reisen (1997) conduct their research based on data from eleven different countries over the period 1982-1993. They find that an increase in retirement benefits from the PAYG system negatively affects the savings rate. Their results also showed that a funded pension system contributes to higher savings rates provided that it is mandatory, borrowing constraints exist, and tax exemptions on pension funds’ returns are limited. These results are supportive to our expectations. Samwick (2000) uses data from both developing and

developed countries, over a similar period as the forenamed study. He finds that a reform the PAYG system to pension funds can increase the savings rate. Further on, a recognition effect is mostly applicable to developing countries, but is difficult to estimate.

In the Mexican case, the increase in the savings rate turns out to be twenty-five percent lower than expected after a switch to a funded pension system. The hampered effect was caused by the crowding out effect on private savings. Therefore governments should promote voluntary pension contributions in order to increase the savings rate (Aguila, 2011). Bloom et al. (2007) found that the positive effect on the savings rate of pension funds

depends on the lifetime expectancy of the population and the presence of a retirement incentive and universal coverage. As before, demographics can be linked to the effect of pension systems on the savings rate. This positive effects disappears, when the pension system is funded for more than fifty percent by the PAYG system.

Loayza, Schmidt-Hebbel and Servéén (2000) have investigated the determinants of private savings rates. In other studies, they found that PAYG systems seem to have a negative or insignificant effective on the private savings rate, whereas pension funds have an expected positive effect on the private savings rate. However, the lack of reliable annual pension data prevented them from including these two variables in their study.

Ponds, Severinson and Yermo (2011) state that the problem of the PAYG system is that future liabilities on tax payers are unclear, and can impose unimagined costs. Unaware of this, individuals could therefore save too little. However, a switch to a funded system might not foster that recognition effect on the savings rate as suggested before. Pension reforms to funded systems can impose severe transition costs, and can even decrease transparency of future retirement benefits and liabilities (Disney, 2000). On the other hand, this effect could potentially stimulate individuals to increase precautionary savings, and would therefore be ambiguous on the savings rate.

(10)

10

the savings rate. However, the studies discussed are all based on data obtained until 2002. As Samwick (2000) suggested, adding data over a longer period could be an addition to earlier studies. In the next section, the control variables in the study by Bloom et al. (2007) will be reviewed, as they will be also used in this study on the effect of pension systems on the savings rate.

1.4 Other determinants of the savings rate

The following variables are reviewed to determine their expected effects on the savings rate: income per capita, annual growth of GDP per capita, lifetime expectancy at birth, and the old age dependency ratio. Theoretical predictions of these variables on the effects on the savings rate are discussed, and then compared to empirical results.

1.4.1 Income per capita

High income can facilitate the possibility of early retirement. As discussed before, this induces individuals to save more during working life in order to provide for this, thereby increasing their savings rates (Bloom, Canning & Moore, 2004). Samwick (2000) also points out the lower propensity to consume for high income households. Therefore high income is expected to have a positive effect on the savings rate. Loayza, Schmidt-Hebbel and Servéén (2000) found that higher income per capita has a positive effect on private savings rates. The study by Bloom et al. (2007) shows that high income increases savings rates. Yet Samwick (2000) found that income per capita has no significant impact on the savings rate.

1.4.2 Income growth per capita

Bloom et al. (2007) predict that an increase in income would lead to an increase in the savings rate, because the relative increase of income outweighs the increase of consumption. Using the LCH, Modigliani and Cao (2004) have predicted in the case of China that the growth rate of GDP is closely related to the savings rate over time. In relation to what is said in the section before, higher growth rates are expected to increase income, which in turn increases savings rates. They have demonstrated that despite low income per capita, China had one of the highest savings rates worldwide. This was partly due to the high growth rates of GDP. The same goes for Loayza, Schmidt-Hebbel and Servéén (2000), who found that a one percent increase in GDP leads to an equal increase in the savings rate. Bloom et al.

(11)

11

(2007) did not find a significant impact of the growth rate of GDP per capita on the savings rate.

1.4.3 Lifetime expectancy at birth

Lifetime expectancy at birth gives an indication of the expectation of individuals on the duration of their own life. Bloom et al. (2007) consider different scenarios for which increased lifetime expectancy can affect savings rates. They argue that after increased longevity and an unchanged retirement age in a funded pension system, individuals would increase savings to provide for longer retirement. However, if there are no retirement

incentives at a certain age, then individuals could prolong their working lives and leave their savings rate unchanged. The positive effect on the savings rate of increased longevity even disappears under the PAYG system due to social security. Therefore, the effect on the savings rate would depend on the type of pension system. Their empirical results are supportive of these predictions and state that the effect of increased longevity is dependent on the pension system. Samwick (2000) did not find any conclusive results on this variable.

1.4.4 The old age dependency ratio

The old age dependency ratio is the number of people older than 64, as a percentage of the working population. Modigliani and Cao (2004) have supported the use of this variable in determining savings rates, since it indicates demographic structure better than population growth rates. According to the LCH, people consume their savings at old age. This is

consistent with the working of pension funds, which are used at retirement age to provide for retirement benefits. Therefore the expected sign of the effect on the savings rate is negative. Both Bloom et al. (2007) and Loayza, Schmidt-Hebbel and Servéén (2000) have found evidence supportive of this, as did Modigliani and Cao (2004).

2 Data analysis

In this part of the study, the methodology and the data are discussed. At the end, the results of the panel data analyses are included. In the following part of the study, these results will be discussed.

(12)

12

2.1 Methodology

This paper is an extension to earlier research done on the effect of pension systems on the savings rate. The year 2013 is added to see whether earlier results on the effects on

pension systems would hold. Because reliable data on pension systems over a large time span is difficult to obtain, I have selected the paper by Bloom et al. (2007) to form the basis of this research.

Some variables from the model employed by Bloom et al. (2007) have been omitted. For example, data on the labor participation rate was unavailable for most countries before 2002. Because they also did not find any significant results on this variable, this was excluded from the analysis. Data on retirement incentives and universal coverage was unavaible for the year 2013 in the sources that were used, so these variables have been dropped too. The year 1961 is excluded, because of a lack of data available at the World Bank and the OECD. African countries are omitted from this study, also because no reliable data for pension systems is available for 2013. The definitions of the variables can differ from that of Bloom et al. (2007). For example they divided lifetime expectancy by hundred,

instead of using the log. Next to this, panel data analyses are used in this study, instead of an OLS-regression. An overview of all variables is given in the following section.

2.2 Data

Time series are used to determine the effects of the PAYG system and pension funds on the domestic savings rate. The analysis is based on 42 different countries, shown in Table 1:

Table 1: Countries included in the analysis

Argentina Chile Greece Jamaica Panama Sweden

Australia Colombia Hong Kong Japan Peru Switzerland

Austria Denmark India Malaysia Philippines Turkey

Belgium Dominican R. Indonesia Mexico Portugal United Kingdom

Bolivia Ecuador Ireland Netherlands Singapore United States

Brazil Finland Israel New Zealand Spain Uruguay

(13)

13

There are 126 annual observations divided over the years 1981, 2002 and 2013. The gross domestics savings rate is used as a dependent variable, denoted by the term Sct. 1 It is

calculated by substracting final consumption expenditure from GDP. The following independent variables have been used:

 RFct: The replacement rate provided by pension funds of country c at time t. The

replacement rate is the size of retirement benefits as a percentage of the last earned wage of a retiree. Data on this variable is obtained from the paper by Bloom et al. (2007) for 1981 and 2002. For 2013 information is collected from the Pensions At a Glance series by the OECD (2014) on OECD members, countries from Asia and the Pacific, and Latin America and the Caribbean. Bloom et al. (2007) define a system funded, when contributions to a fund can be freely invested in financial assets. Both provident funds or private funds belong to the funded system. Their guidelines were followed by checking for changes in pension laws and pension contributions. I also included voluntary contributions to pension funds. My reason to do so, is that Bloom et al. (2007) do not explicitly distinct between mandatory and voluntary contributions, and that countries like The Netherlands provide fiscal benefits to these contributions. To determine the part of the replacement rate provided by pension funds in 2013, I calculated the relative amount of retirement retirement funding that was contributed to pension funds. This percentage was then multiplied by the replacement rate in order to determine this variable.

 RPct: The replacement rate provided by the PAYG system of country c at time t. Data

on this variable is obtained from the same sources as before. Bloom et al. (2007) classify the PAYG system as a system where contributions to a social security fund can only be invested in government bonds, thereby making it an indirect contribution to government spending more than pension funding, or where contributions are directly distributed as retirement benefits to current retirees. Social security contributions, public contributions and provident funds that are limited to investing in domestic government debt, are all considered to belong to the PAYG system. For the calculation of this variable, the same method as before is applied.

 Ict: GDP per capita of country c at time t. Gross domestic product is defined as the

total value added by producers in the country in addition to product taxes. Subsidies are

(14)

14

substracted. Depreciation, depletion and other losses are not accounted for. Values are denoted in current U.S. dollars. GDP is divided by midyear population. Data is derived from the World Bank and complemented by additions from the OECD and the United Nations, for respectively New Zealand and Venezuela.

 Gct: GDP growth per capita of country c at time t. It measures the annual growth rate

of GDP per capita, expressed in local currency. The different notation of currency is not significant, since we are interested in the growth of annual income per person. Yet there is no adjustment for inflation. Data is obtained at the World Bank.

 logLct: The log of lifetime expectancy at birth in years (total population) of country c

at time t. Lifetime expectancy indicates the average amount of years that a newborn could expect to live, provided that mortality rates remain equal. Data is obtained at the World Bank.

 Act: The old age dependency ratio is the percentage of people older than 64 of the

working-age population (people aged between 15-64) of country c at time t. It indicates the numbers of dependents on the working age population, and gives an insight in demographic structure. Data is obtained at the World Bank.

These variables give the following formula for the first panel data analysis:

Sct = α+ β1 RFct + β2 RPct + β3 Ict + β4 Gct + β5 logLct + β6 Act + ɛ (1)

2.3 Hypotheses

For each independent variable the expected sign of the coefficient is given in the table below. These signs are based on the theoretical predictions and the empirical results

discussed in the literature review.

Table 2: Expected signs of the coefficients

Variables Sign of the coefficient

RFct + RPct - Ict + Gct + logLct + Act -

(15)

15

2.4 Analysis

Here, the tests that were used to determine for fixed or random effects and for homoscedasticity are discussed. The results of these tests are included with the results of the panel data analyses.

2.4.1 Hausman test

There are two options to do a panel data analysis. The first option is to use fixed effects. Using fixed effects is based on two assumptions. The first assumption is that a country has time-invariant characteristics that influence either the dependent or the independent variable. The second assumption is that these characteristics are unique per country. The second option is to conduct a panel data analysis with random effects. Here we assume that variation over the countries is random and also has no correlation with the dependent and independent variables.

To see whether fixed effects or random effects are preferable the Hausman test is used. Two panel data analyses with fixed and random effects are made. This test analyzes if there is correlation between the unique error terms and the independent and dependent variables, using the results from these analyses. The null hypothesis states that the estimators of the analyses with fixed and random effects are both consistent, but that the latter is most efficient. If the null hypothesis is rejected, then only the estimators with fixed effects are

consistent. The formula for the Hausman test is shown below (Princeton University, 2007) 2:

Chi2 = (b0 – b1) ' [(Var(b0) – Var(b1)) -1] (b1 – b0) (2)

2.4.2 Breusch-Pagan test

Next, we will check if heteroscedasticity is present. In that case the variation from the residuals is correlated with the independent variables, which can increase p-values. To test this the Breusch-Pagan test is used. Here, a regression is made with the squared residuals as the dependent variable. The null hypothesis states that the variance in the model is constant, and thus independent of the values of the explanatory variables. In that case homoscedasticity is present, and if the null hypothesis is rejected than there is heteroscedasticity. The formula for this test is shown here:

2_b

(16)

16

Ȗ2_{= ά+ γ1 RFct + γ2 RPct + γ3 Ict + γ4 Gct + γ5 logLct + γ6 Act + υ} ₍₃₎

2.4.3 Results of the panel data analyses

In addition to the panel data analysis made with all the variables included, two other tests are done. The second panel data analysis excludes all the variables that were found insignificant in the first analysis, except for the replacement rates provided by the PAYG system and pension funds. In this way we can see if the results for the other two variables hold, and if the significance levels and the coefficients of both replacement rates change. In the third test both replacement rates are replaced by the total replacement rate (denoted by

RRct) and the relative dependency on the PAYG system (denoted by RPAYGct). Here we

distinguish for the effect of the replacement rate itself on the savings rate. We can also see if countries with a high dependency rate on the PAYG system have a lower savings rate. The formulas for both tests are shown here:

Sct = α+ β1 RFct + β2 RPct + β3 Ict + β4 Act + ɛ (4)

Sct = α+ β1 RRct + β2 RPAYGct + β3 Ict + β4 Gct + β5 logLct + β6 Act + ɛ (5)

For all panel data analyses the Hausman and the Breusch-Pagan tests are repeated. In all cases fixed effects are used and robust standard errors were not necessary, following the results of these tests. The results of the panel data analyses and the Hausman and Breusch-Pagan tests are shown in Table 3.

(17)

17

Table 3: Results panel data analyses using fixed effects

Variables (1) Sct (2) Sct (3) Sct RFct -0.01943 (-0.75) -0.0194 (-0.76) RPct 0.027282 (0.89) 0.025404 (0.85) Ict 1.16E-06*** (2.16) 1.24E-06** (2.41) 1.66E-06*** (3.18) Gct -0.01124 (-0.04) 0.13822 (0.59) logLct 0.07722 (0.58) 0.095579 (0.81) Act -0.69908*** (-2.69) -.6396754*** (-2.72) -0.41576** (-2.52) RRct 0.030956 (1.5) RPAYGct -0.00669 (-0.34) Observations 126 126 126 Hausman test: Chi2-value 33.56 37.60 61.80

Breusch Pagan test: p-value 0.4441 0.3546 0.2639 T-values in parentheses *** p<0.01, ** p<0.05, * p<0.1

(18)

18

The results of the first panel data analysis in column 1 of Table 3 show that the effects of the replacement rates provided by pension funds and the PAYG system on the savings rate are both insignificant. The signs of their coefficients are contrary to what the theory suggests. The replacement rate funded by pension funds has a negative effect on the savings rate, whereas the other replacement rate has a positive effect. Further on, only income per capita and the old age dependency ratio have a significant effect on the savings rate, having respectively a positive and a negative effect. The first finding can point at a possible

difference in the savings rate between developing and developed countries, in which the latter would have higher savings rates. The second finding is consistent with the LCH and the working of pension systems. The sign of the coefficient of the log of lifetime expectancy is consistent with our expectations, but the effect of GDP growth per capita is surprisingly negative.

Column 2 shows that there are no remarkable changes in the signs of the coefficients and the p-values, when insignificant variables are excluded. The effects of both replacement rates remains insignificant, and the signs of their coefficients have not changed. A possible explanation for the negative effect of the replacement rate funded by pension funds might be that the usage of pension funds is associated with high replacement rates. A higher

replacement rate provides individuals with increased retirement benefits, and lowers the necessity of additional saving for retirement. In that case the sign of the replacement rate funded by pension funds could become negative. In column 3 the results are presented, where both replacement rates have been replaced by the total replacement rate and the dependency rate on the PAYG system to distinguish for both effects. The effect of the total replacement rate is positive and insignificant. The relative dependency on the PAYG has a negative effect on the savings rate, which is consistent with our expectations. GDP growth per capita now also has a positive sign, as was expected earlier. Yet the effects of both variables on the savings rate remain insignificant. There are no other remarkable changes.

3 Discussion

Next, we will analyze in more detail-certain aspects of our analysis: multicollinearity, the sample size, exclusion of variables and the results of the replacement rates in the year 2013. Thereafter, a suggestion for further research is made.

(19)

19

3.1 Multicollinearity

A cause for the high p-values in the results can be multicollinearity. This is the case when two or more independent variables are strongly correlated to each other. The model would be more sensitive to changes in the values of these variables. Another consequence is that the variance in the model can be increased, which leads to larger standard errors. It is also possible that the coefficients switch signs. In Table 4 the results of the correlation matrix are presented.

Table 4: Correlation matrix

RFct RPct Ict Gct logLct Act RFct 1.0000 RPct 0.3167 1.0000 Ict -0.0898 -0.0964 1.0000 Gct -0.0198 -0.1390 0.0004 1.0000 logLct 0.0012 0.1050 -0.2520 -0.0801 1.0000 Act 0.0708 0.1807 -0.4626 0.1458 -0.3905 1.0000

No correlation between variables larger than 0.4 is reported. The strongest correlation is shown between the replacement rates provided by pension funds and the PAYG system, and between the old age dependency ratio and the log of the lifetime expectancy. The correlation between both replacement rates is positive. This is surprising, because we would expect to see a substitution expect between both systems that would result in a negative coefficient. It might be that pension funding is equally distributed between the two systems, so that when the replacement rate provided for by the PAYG system is high, the replacement rate funded by pension funds is likely to be high too. The negative sign of the correlation between the old age dependency ratio and the log of lifetime expectancy is also unexpected. A higher lifetime expectancy would mean that more people live to old age, which eventually increases the ratio of people living that are 65 years and older.

3.2 Sample size

The sample size is not large. Nine African countries are excluded from this analysis, because of a lack data on the pension systems. Hence, my analysis has 27 observations less

(20)

20

than Bloom et al (2007). With only 126 observations, problems regarding the measure of significance are likely.

3.3 Omission of variables

This study excluded different variables from the study by Bloom et al (2007). The variables retirement incentive and universal coverage were not available for 2013, but were important in the study by Bloom et al (2007). In theory, an increase in lifetime expectancy with fully funded pension funds does not have to increase the savings rate if individuals prolong their working lives, unless a retirement incentive and universal coverage are in place. Universal coverage means that all working individuals are covered by the pension system. In that case all individuals are strongly motivated to retire at the common retirement age.

Further on, there is no variable used in this study that indicates borrowing constraints. Yet the positive effect of mandatory pension fund contributions on the savings rate depends on the presence of borrowing constraints.

3.4 Estimation of the replacement rates

The estimation of the replacement rate is not straightforward. Future pension wealth is dependent on fiscal regimes, changes in policy and the rates of return of different pension funds. Whitehouse (2012) uses the rates of return of pension systems and fiscal information to determine future pension entitlements. However, parameters influencing the rate of return per system can change over time. Because pension wealth is difficult to determine, it also becomes hard to calculate replacement rates. It is interesting to see if replacement rates and the dependency on the PAYG system have changed over the years 1981, 2002 and 2013. Below, the table shows the average replacement rates and the dependency rate on the PAYG system.

Table 5: Replacement rates and the dependency on the PAYG system

Years Average replacement rate Dependency rate on the PAYG system

1981 0,661428571 0,877933057

2002 0,716904762 0,678720884

2013 0,604021718 0,678704493

(21)

21

The average replacement rate is lower in 2013 in comparison to previous years. This could be due to the financial crisis in 2008 and the economic slowdown afterwards. This led to a decrease in the worth of financial assets managed by pension funds (World Bank, 2008). Another possibility is the rise of retirement ages after 2002, which negatively affects the worth of future pension wealth (Whitehouse, 2012). Further on, the dependency rate on the PAYG system has dropped in the period from 1981 until 2002, but seems to have not changed on average after 2002. Perhaps these changes in the replacement rates can have biased the outcome of the analyses.

3.5 Recommendations for further research

In this research, I studied the effect of pension systems on the savings rate. Some variables that were found influential on the effect of pension systems on the savings rate are excluded in this study, because of a lack of availability. These variables are the presence of retirement incentives, universal coverage and borrowing constraints. Data on borrowing constraints is better available for recent years at the World Bank. An indicator of this variable would be domestic credit provided by the financial sector as a percentage of GDP. The higher this rate is, the more likely it is that there are no tight borrowing constraints. Next to this, the contribution base can be added to the analysis by investigating government sources. If a pension system uses a DC basis, then the future retiree bears the risk of his investments. Because of future income uncertainty, he can increase his precautionary savings and so his savings rate. All these variables provide a valuable addition for further research, because it would be possible to distinguish for the effects of pension systems on the savings rate under different circumstances. In that case, earlier findings by for example Samwick (2000) and Bloom et al. (2007) can be complemented by more extensive research on recent data.

4 Conclusion

This study investigated the effect of the PAYG system and pension funds on the savings rate in 42 countries over the years 1981, 2002 and 2013. There are different reasons to assume that pension funds contribute to a higher savings rate, namely the recognition effect, mandatory contributions, and less usage of progressive redistribution formulas that transfer wealth to lower income households than under the PAYG system. The expected effects on precautionary savings of both pension systems are ambiguous. The positive effect

(22)

22

of pension funds on the savings rate would finally depend on the crowding out effect on private savings.

In earlier studies, results showed that pension funds have a positive effect on the savings rate in comparison to the PAYG system. Still most of these studies only used data until 2002. The contribution this study has made, is that recent data is added to see if earlier results would hold. Unfortunately, this study did not find any results that were supportive to earlier findings. The effects of the replacement rates provided by the PAYG system and pension funds on the savings rate were contrary to what was expected, and were also insignificant. This could be due to the small sample size and the exclusion of explanatory variables that were used in other studies. To confirm if pension funds contribute to a higher savings rate in comparison to the PAYG system, the sample size can be expanded by adding more recent years than only 2013. Data on this period is better available than before 2000, so a variable indicating borrowing constraints and more observations can be added.

Governmental sources can be investigated on retirement incentives, universal coverage and contribution base in pension systems. In that case, the research could then distinguish for the effects of pension systems under different circumstances. This would give a more complete answer on the positive effects of pension funds on the savings rate compared to the PAYG system.

(23)

23

Appendix

Appendix 1: All data on which the study is based

Countries Y _S_ct _RF_ct _RP_ct _I_ct _G_ct _L_ct _A_ct _RR_ct % PAYGct Argentina 1981 0,23 0 0,7 2756,4 -0,05 69,79 0,14 0,7 1 Australia 1981 0,27 0 0,19 11826,05 0,04 74,67 0,15 0,19 1 Austria 1981 0,25 0 0,8 9362,64 0 72,82 0,24 0,8 1 Belgium 1981 0,21 0 0,6 10679,68 0 73,63 0,22 0,6 1 Bolivia 1981 0,17 0 0,76 1031,52 0,01 50,49 0,08 0,76 1 Brazil 1981 0,23 0 0,95 2106,68 -0,04 62,31 0,07 0,95 1 Canada 1981 0,27 0 0,38 12279,07 0,04 75,48 0,15 0,38 1 Chile 1981 0,13 0,57 0 2863,47 0,05 68,43 0,1 0,57 0 Colombia 1981 0,18 0 0,85 1282,37 0,03 66,03 0,07 0,85 1 Denmark 1981 0,19 0 0,25 11984,02 0 74,24 0,23 0,25 1 Dominican Republic 1981 0,2 0 0,7 1222,67 0,05 63,44 0,06 0,7 1 Ecuador 1981 0,26 0 0,85 2665,32 0,06 63,69 0,08 0,85 1 Finland 1981 0,3 0 0,81 10934,57 0,02 73,75 0,18 0,81 1 France 1981 0,22 0 0,25 11110,56 0,02 74,3 0,22 0,25 1 Greece 1981 0,21 0 1 5380,27 -0,01 73,97 0,21 1 1 Hong Kong 1981 0,34 0 0,05 5991,33 0,1 75,33 0,09 0,05 1 India 1981 0,2 0,78 0 275,92 0,07 54,33 0,07 0,78 0 Indonesia 1981 0,32 0,14 0 612,5 0,09 60,04 0,07 0,14 0 Ireland 1981 0,14 0 0,15 5992,88 0,04 72,78 0,19 0,15 1 Israel 1981 0,04 0 0,66 5863,59 0,06 74,27 0,15 0,66 1 Italy 1981 0,23 0 0,8 7600,46 0,01 74,36 0,21 0,8 1 Jamaica 1981 0,12 0 0,13 1377,72 0,03 70,69 0,13 0,13 1 Japan 1981 0,32 0 0,71 10212,38 0,05 76,42 0,14 0,71 1 Malaysia 1981 0,26 0,59 0 1763,36 0,07 68,37 0,07 0,59 0 Mexico 1981 0,25 0 0,88 3524,75 0,09 67,04 0,08 0,88 1 Netherlands 1981 0,26 0 0,68 11373,42 0 75,94 0,18 0,68 1 New Zealand 1981 0,23 0 0,28 7813,81 0,05 73,63 0,16 0,28 1 Norway 1981 0,35 0 0,51 15512,51 0,02 75,87 0,24 0,51 1 Panama 1981 0,35 0 1 2128,62 0,1 70,62 0,08 1 1 Peru 1981 0,51 0 1 1184,44 0,06 60,66 0,07 1 1 Philippines 1981 0,25 0 1 731,73 0,04 62,43 0,06 1 1 Portugal 1981 0,21 0 0,7 3246,3 0,02 71,62 0,19 0,7 1 Singapore 1981 0,42 2,55 0 5595,3 0,11 72,59 0,08 2,55 0

(24)

24 Spain 1981 0,2 0 1 5359,15 0 75,53 0,18 1 1 Sri Lanka 1981 0,12 0 0,9 297,43 0,06 68,68 0,08 0,9 1 Sweden 1981 0,25 0 0,66 15366,67 0,01 76,03 0,26 0,66 1 Switzerland 1981 0,3 0 0,32 17110,45 0,02 75,7 0,21 0,32 1 Turkey 1981 0,14 0 0,6 1580,89 0,05 59,33 0,09 0,6 1 United Kingdom 1981 0,22 0,01 0,22 9599,31 0 74,03 0,24 0,23 0,96 United States 1981 0,24 0,04 0,44 13993,17 0,03 74,01 0,18 0,48 0,92 Uruguay 1981 0,18 0 0,7 3764,3 0,02 70,56 0,18 0,7 1 Venezuela 1981 0,3 0 0,62 4789,99 0 68,43 0,06 0,62 1 Argentina 2002 0,27 0 0,92 2579,21 -0,1 74,16 0,17 0,92 1 Australia 2002 0,25 0,46 0,2 20059,46 0,04 79,94 0,19 0,66 0,31 Austria 2002 0,28 0 0,8 26351,38 0,02 78,68 0,23 0,8 1 Belgium 2002 0,27 0 0,6 25052,34 0,02 78,08 0,27 0,6 1 Bolivia 2002 0,11 0,57 0 913,58 0,03 61,83 0,1 0,57 0 Brazil 2002 0,19 0 1 2810,24 0,04 70,85 0,09 1 1 Canada 2002 0,25 0 0,37 23995,02 0,03 79,6 0,19 0,37 1 Chile 2002 0,23 0,57 0 4566,53 0,03 77,57 0,13 0,57 0 Colombia 2002 0,14 0 0,85 2355,73 0,03 71,53 0,08 0,85 1 Denmark 2002 0,29 0,68 0,24 33228,7 0,01 76,9 0,23 0,92 0,27 Dominican Republic 2002 0,14 0,4 0 3008,47 0,06 71,06 0,09 0,4 0 Ecuador 2002 0,18 0 0,85 2183,97 0,05 73,5 0,09 0,85 1 Finland 2002 0,31 0,83 0,83 26834,03 0,02 78,12 0,23 1,66 0,5 France 2002 0,23 0 0,5 24275,25 0,02 79,27 0,26 0,5 1 Greece 2002 0,15 0 1 14005,36 0,04 78,65 0,27 1 1 Hong Kong 2002 0,32 0,57 0 24665,89 0,02 81,48 0,16 0,57 0 India 2002 0,25 0,96 0 480,63 0,04 63,4 0,08 0,96 0 Indonesia 2002 0,28 0,32 0 900,14 0,05 66,61 0,08 0,32 0 Ireland 2002 0,42 0 0,11 32537,96 0,06 77,64 0,16 0,11 1 Israel 2002 0,18 0 0,66 18431,16 0 79,46 0,17 0,66 1 Italy 2002 0,23 0 0,68 22205,85 0,01 80,23 0,28 0,68 1 Jamaica 2002 0,12 0 0,21 3706,78 0,02 70,78 0,14 0,21 1 Japan 2002 0,24 0 0,51 31235,59 0,01 81,57 0,27 0,51 1 Malaysia 2002 0,43 1,04 0 4132,67 0,06 73,21 0,07 1,04 0 Mexico 2002 0,19 0,27 0 7023,79 0,01 74,75 0,09 0,27 0 Netherlands 2002 0,29 0 0,58 28817,33 0,01 78,3 0,21 0,58 1 New Zealand 2002 0,26 0 0,26 16895,46 0,05 78,85 0,18 0,26 1

(25)

25 Norway 2002 0,34 0 0,42 43061,16 0,02 78,99 0,23 0,42 1 Panama 2002 0,21 0 1 3902,54 0,03 75,47 0,1 1 1 Peru 2002 0,17 0,46 0 2046,38 0,06 71,39 0,09 0,46 0 Philippines 2002 0,16 0 0,91 1000,78 0,04 66,93 0,06 0,91 1 Portugal 2002 0,18 0 0,8 12882,29 0,01 77,07 0,25 0,8 1 Singapore 2002 0,44 1,62 0 22016,57 0,05 78,56 0,11 1,62 0 Spain 2002 0,25 0 1 17019,54 0,03 79,57 0,25 1 1 Sri Lanka 2002 0,17 0 0,9 903,9 0,04 72,58 0,1 0,9 1 Sweden 2002 0,29 0 1,43 29571,71 0,03 79,85 0,27 1,43 1 Switzerland 2002 0,31 0,63 0,18 41336,73 0,01 80,39 0,23 0,81 0,23 Turkey 2002 0,2 0 0,79 3570,55 0,07 71,08 0,1 0,79 1 United Kingdom 2002 0,17 0,02 0,41 28301,21 0,03 78,15 0,25 0,43 0,96 United States 2002 0,18 0,05 0,45 38166,04 0,02 76,84 0,19 0,5 0,9 Uruguay 2002 0,15 0,38 0,22 4088,78 -0,07 75,1 0,22 0,6 0,37 Venezuela 2002 0,34 0 0,6 3657,2 -0,08 72,78 0,08 0,6 1 Argentina 2013 0,19 0 0,78 14443,07 0,03 75,99 0,17 0,78 1 Australia 2013 0,28 0,39 0,14 67627,78 0,03 82,2 0,22 0,53 0,26 Austria 2013 0,27 0 0,77 50557,81 0,01 80,9 0,28 0,77 1 Belgium 2013 0,24 0,16 0,42 46625,32 0,01 80,39 0,28 0,57 0,74 Bolivia 2013 0,26 0 0,69 2948,04 0,07 67,92 0,11 0,69 1 Brazil 2013 0,19 0 0,59 11711,01 0,03 74,13 0,11 0,59 1 Canada 2013 0,23 0,34 0,4 52305,26 0,03 81,41 0,23 0,74 0,54 Chile 2013 0,24 0,38 0,05 15741,72 0,05 81,2 0,16 0,42 0,12 Colombia 2013 0,22 0 0,71 8027,98 0,05 73,81 0,1 0,71 1 Denmark 2013 0,25 0,48 0,31 59818,64 0 80,31 0,28 0,79 0,39 Dominican Republic 2013 0,17 0 0,23 5968,67 0,05 73,32 0,11 0,23 1 Ecuador 2013 0,27 0 0,95 6051,62 0,05 75,65 0,11 0,95 1 Finland 2013 0,21 0 0,55 49492,83 -0,01 80,84 0,3 0,55 1 France 2013 0,21 0 0,59 42627,66 0,01 81,97 0,29 0,59 1 Greece 2013 0,09 0 0,54 21719,23 -0,03 80,64 0,32 0,54 1 Hong Kong SAR. China 2013 0,25 0,34 0 38364,2 0,04 83,84 0,19 0,34 0 India 2013 0,3 0,28 0,28 1455,11 0,07 67,67 0,09 0,55 0,5 Indonesia 2013 0,34 0,14 0 3623,54 0,06 68,71 0,08 0,14 0 Ireland 2013 0,38 0,44 0,37 51814,86 0,02 81,05 0,19 0,8 0,47 Israel 2013 0,22 0,52 0,23 36281,2 0,04 82,06 0,18 0,74 0,31 Italy 2013 0,2 0 0,72 35420,88 -0,01 82,3 0,34 0,72 1

(26)

26 Jamaica 2013 0 0 0,73 5226,1 0,01 73,47 0,14 0,73 1 Japan 2013 0,19 0 0,36 38633,71 0,02 83,34 0,41 0,36 1 Malaysia 2013 0,35 0,34 0 10973,66 0,05 74,57 0,08 0,34 0 Mexico 2013 0,21 0,22 0,09 10172,73 0,02 76,54 0,1 0,3 0,28 Netherlands 2013 0,29 0,62 0,3 51425,08 0 81,11 0,26 0,91 0,33 New Zealand 2013 0,07 0,15 0,41 34989 0,03 81,41 0,22 0,55 0,75 Norway 2013 0,39 0,19 0,46 102832,26 0,01 81,46 0,24 0,64 0,72 Panama 2013 0,18 0 0,79 11206,43 0,09 77,42 0,12 0,79 1 Peru 2013 0,28 0 1,05 6603,81 0,06 74,29 0,11 1,05 1 Philippines 2013 0,16 0 0,38 2786,96 0,08 68,14 0,07 0,38 1 Portugal 2013 0,16 0 0,55 21618,74 -0,01 80,38 0,31 0,55 1 Singapore 2013 0,53 0,37 0 55979,76 0,05 82,35 0,15 0,37 0 Spain 2013 0,23 0 0,74 29370,67 -0,01 82,43 0,27 0,74 1 Sri Lanka 2013 0,27 0,42 0 3628,27 0,04 74,25 0,13 0,42 0

1) The log of the life expectancy is not shown here. It is created in Stata, after collecting the data.

2) Bloom et al. (2007) calculate the replacement rate for workers of average income, therefore only replacement rates of average income workers are used in these calculations.

3) All public contributions in countries without provident funds were regarded as PAYG contributions. Mandatory and voluntary private contributions are specified as funded contributions.

4) Malaysia, Singapore, India, Hong Kong and Indonesia use provident funds. Only in India was a significant change in pension law, because since 2004 the New Pension system (NPS) is in force. It is since then possible to invest up to fifty percent of all financial assets in equity. The other part can be invested in government bonds (PFRDA, 2016). Fifty percent of the replacement rate will be considered to be provided for by the PAYG system and the other fifty percent as funded by pension funds, consistent with the definitions by Bloom et al. (2007). The other countries are marked as fully funded, as they were in 2002.

(27)

27

Appendix 2: Descriptive statistics

Variables Mean Standard deviation Minimum Maximum

Sct 0.2328646 0.0847499 -0.0085467 0.5221477 RFct 0.1696321 0.3440517 0 2.55 RPct 0.4911529 0.3396449 0 1.43 Ict 17486.98 19190.01 275.9167 102832.3 Gct 0.0239215 0.0330258 -0.1089448 0.1068384 Lct 74.47876 6.261076 50.48215 83.83171 Act 0.1634948 0.0782571 0.0564475 0.4036528 RRct 0.660785 0.3300975 0.05 2.55 RPAYGct 0.7451195 0.3910157 0 1

(28)

28

References

Literature

Alders, P. & Broer, P. (2005). Ageing, fertility and growth. Journal of Public Economics, 89(5-6), 1075-1095.

Aguila, E. (2011). Personal retirement accounts and saving. American Economic Journal: Economic policy 3(4), 1-24.

Bailliu, J. & Reisen, H. (1997). Do funded pensions contribute to higher aggregate savings?: A cross-country analysis. OECD Development Centre, Technical Papers 130.

Bloom, D. E., Canning, D., Mansfield, R., & Moore, M. (2007). Demographic change, social security systems, and savings. Journal of Monetary Economics 54, 92-114. Bloom, D.E., Canning, D., & Moore, M., (2004). The effects of improvements in health and

longevity on optimal retirement and saving. NBER Working Paper No. 10919. Bruil, A., Schmitz, C., Gebraad, J., & Bhageloe-Datadin, R. (2015). Totale

pensioenaanspraken van Nederland in beeld. Centraal Bureau voor de Statistiek. http://www.cbs.nl/NR/ rdonlyres/857C6E4C-312E-4C7E-AEFA-5CB882D B7218/0/ 2015totalepensioenaan sprakenvannederlandinbeeld.pdf. Visited on 18 December 2015. Disney, R. (2000). Crises in public pension programmes in OECD: What are the reform

options? The Economic Journal, 110(461). Gyárfas, G. & Marquardt, M. (2001) Pareto improving transition from a pay-as-you-go to a fully funded pension system in a model of endogenous growth. Journal of Population Economics 14(3). 445-453. Loayza, N., Schmidt-Hebbel, K., & Servéén, L. (2000). What drives private savings across

the world? The Review of Economics and Statistics 82(2). Modigliani F. & Cao, S. L. (2004). The Chinese saving puzzle and the life-cycle hypothesis.

Journal of Economic Literature, 42(1), 145-170. Ponds, E., Severinson, C., & Yermo, J. (2011). Funding in public sector pension plans:

international evidence. OECD Working Papers on Finance, Insurance and Private

Pensions, 8. http://dx.doi.org/10.1787/5kgcfnm8rgmp-en. Visited on 5 January 2016. Princeton University. (2007). Panel data analysis fixed and random effects using Stata.

https://www.princeton.edu/~otorres/Panel101.pdf. Visited on 15 January 2016. Roberts, M. A. (2013). Pareto-improving pension reform through technological

(29)

29

Samwick, A. A. (2000). Is pension reform conducive to higher saving? Review of Economics and Statistics, 82(2), 264-272.

Steurer, M. (2009). Extending the Aaron Condition for alternative Pay-as-You-Go pension systems. School of Economics, the University of New South Wales.

Whitehouse, E. (2012). Adequacy of Pension entitlements, replacement rates and pension wealth. http://documents.worldbank.org/curated/en/2012/01/16453588/adequacy-1-pension- entitlements-replacement-rates-http://documents.worldbank.org/curated/en/2012/01/16453588/adequacy-1-pension-wealth. Visited on 10 January 2016. World Bank. (2008). The financial crisis and mandatory pension systems in developing

countries. http://siteresources.worldbank.org/INTPENSIONS/Resources/395443-1121194657824/PRPNote-Financial_Crisis_12-10-2008.pdf. Visited on 21 January 2016.

Data

Bloom, D. E., Canning, D., Mansfield, R., & Moore, M. (2007). Demographic change, social security systems, and savings. Journal of Monetary Economics 54, 92-114.

IMF. (2015). http://www.imf.org/en/Data. Visited at 15 January 2016.

OECD. (2014). Pensions at a glance 2013 (Asia/Pacific). http://www.oecd-ilibrary.org/ doc server/download/8113211e.pdf?expires=1454163281&id=id&accname=guest&check sum=08489A298A252208BF5C5E70BF85E6CB. Visited on 10 January.

OECD. (2014). Pensions at a glance 2013 (Latin America and the Caribbean). http://www. oecd-ilibrary.org/docserver/download/8114181e.pdf?expires=1454163019&id=id& accname=guest&checksum=CE7FBE3D8D67910B61CFA59893AA4053. Visited on 12 January.

OECD. (2014). Pensions at a glance 2013 (OECD and G20 indicators). http://www.oecd.org/ pensions/public-pensions/OECDPensionsAtAGlance2013.pdf. Visited on 12 January. PFRDA. (2016). http://www.pfrda.org.in/searchnew.cshtml?Key=Investment. Visited on 18

January 2016.

United Nations. (2016). http://unstats.un.org/unsd/databases.htm. Visited on 18 January 2016. World Bank. (2015). World development indicators. http://data.worldbank.org/data-catalog

The effects of pension systems on the savings rate : a cross country analysis on 42 countries over the years 1981, 2002 and 2013