1 Introduction
Tax relief on supplementary pensions represents one of the pillars of the so-called voluntary welfare (Barr, Reference Barr1992). The presence of tax incentives for encouraging private pension schemes is quite common in Organisation for Economic Development and Co-operation (OECD) economies, not only as a means of raising national saving but also as a way of attenuating future fiscal pressures on the public sector associated with public pensions, making compatible moderate increases in pension expenditure for the public sector with adequate pensions for the old-age population. In fact, voluntary pension plans are a piece of the World Bank and OECD's core recommendations for reforming pensions in Western countries; in particular, it is argued that these tax-favored plans should complement non-contributory pensions and mandatory contributory benefits.
Economic analysis of this policy has centered on its effect on national saving, as a positive by-product of the existence of private pension plans that further increase their attractiveness. If private pensions plans increase savings it can be argued that such a system of old-age provision will contribute to higher investment and future growth, making it easier to combine in the future higher pensions for pensioners and growing income for workers.Footnote 1 The evidence on the issue is still inconclusive, with most of the studies focusing on the USA and not reaching a consensus on its effectiveness. The present paper, which studies the effect of tax incentives for promoting supplementary pensions on saving in Spain, aims to enlarge the body of literature related to this topic. In particular, the research presented here is one of the few available for a country other than the USA and it profits from the use of a longitudinal survey that allows controlling for the effect of time-constant unobservable households heterogeneity using fixed-effects techniques. The results suggest that contributions to pension funds are not linked to higher national saving since they are not accompanied by falls in consumption. However, there is no clear evidence that pension funds contributions come from reshuffling other household assets or saving that would have been done anyway. Therefore, at most, it seems that this tax relief would increase private household saving but not national saving, as the additional saving would come from the higher disposable income allowed by the existence of the tax relief.
The rest of the paper unfolds as follows. The second section, which comes after this introduction, outlines the systems of tax incentives for encouraging retirement saving in Spain and summarizes the main findings of previous literature. The third and fourth sections describe the database and the methodology used in the study, respectively, while the fifth section is devoted to presenting and discussing the main results of the empirical analysis. The final section summarizes the paper's main conclusions.
2 Background and literature review
2.1 Supplementary pensions in Spain
Spanish authorities started to foster supplementary pension provision in 1988, when the first parametric reform of the public pay-as-you-go system was carried out. Regarding complementary pensions, for the first time, the government introduced some incentives for promoting private pension coverage on a voluntary basis. In particular, voluntary contributions to private pension funds became exempt from income tax (a progressive tax with several brackets) up to a certain limit, the returns to such investment were made tax-free and, finally, withdrawals were taxed (usually at a lower marginal tax rate because of declining incomes associated with old age).Footnote 2 These first tax reliefs were followed by exemptions from payroll taxes from 1995 and several tax incentives in corporate tax from 2001.Footnote 3 Nevertheless, the most beneficial tax treatment was in effect from 1999 to 2007, with not only higher general and specific contribution limits, but also a tax-exemption of 40% on lump-sum withdrawal payments. This special treatment on lump-sum payments was removed in 2007.Footnote 4
As might be expected from such tax policies, the number of contributors to pension funds grew exponentially from 1989 to 2009 (see Figure 1), reaching more than 10 million at the end of 2009. It is also relevant to point out that, according to the data from the Spanish Association of Investment and Pension Funds, about 81% of such pension plans were personal (<20% were occupational schemes or similar plans) and, according to the Spanish Directorate General for Insurance and Pension Funds (Directorio General de Seguros y Fondos de Pensiones, 2009), roughly 82.1% of total pension funds corresponded to defined contribution schemes, 0.7%, to defined benefit plans and the rest, to mixed systems. These figures are quite in line with international trends in pension systems design, which privilege personal defined contribution pensions.
Figure 1. Evolution of participation in supplementary pension schemes in Spain (1989–2009).
The growth in the number of contributors also translated into a rapid accumulation of funds: in barely two decades, the supplementary pension system has accumulated funds that accounted for more than 8% of Spanish GDP in 2009 (see Figure 2). Although these data put Spain in a position that is far from countries where either occupational or personal private pensions have a longer tradition, such as Canada, the USA or Chile, the relevance of these schemes is more pronounced than in Italy, Greece or France and very similar to other countries such as Poland or Hungary that have recently moved to mandatory personal accounts.
Figure 2. Pension fund assets in Spain. Upper panel: pension fund assets in OECD countries as a percentage of GDP (around 2009). Lower panel: pension fund assets in Spain as a percentage of GDP (1989–2009). Data from Belgium, France, Greece, Slovakia and Switzerland correspond to 2008; Japanese data are from 2005.
Unfortunately, information on the cost of the tax incentives for encouraging these benefits is remarkably limited. The data available are limited to information on income tax relief, which would account for approximately 0.2% of Spanish GDP in 2002 (Yoo and De Serres, Reference Yoo and De Serres2004; Antón, Reference Antón2007).
2.2 Supplementary pensions, tax incentives and saving
The effect of pension tax relief on saving is a highly controversial issue. From a theoretical point of view, assuming perfect capital markets and consumers with perfect foresight, the net effect of such a policy on private saving is ambiguous, as it results from the balance of a substitution effect associated with a higher rate of return on saving and an income effect linked to larger possibilities of consumption because of the tax break (Atkinson and Stiglitz, Reference Atkinson and Stiglitz1980; López, Reference López2000; Attanasio and DeLeire, Reference Attanasio and DeLeire2002; Bernheim, Reference Bernheim, Auerbach and Feldstein2002; Attanasio et al., Reference Attanasio, Banks and Wakefield2004).Footnote 5 Under less restrictive scenarios, using Behavioral Economics models contemplating problems of self-control or some type of bounded rationality, results are not straightforward either (Bernheim, Reference Bernheim, Auerbach and Feldstein2002). Bernheim points out that, in this theoretical framework, other simpler policies – such as an increase in consumption taxes – might be much more effective in raising household saving than tax-favoring voluntary pensions.
According to Attanasio and DeLeire (Reference Attanasio and DeLeire2002), the money put in pension funds can come from three different sources: first, money that otherwise would have been devoted to consumption; second, savings emanating from reshuffling assets or that would have been done anyway (in the absence of tax incentives); and third, savings associated with higher disposable income resulting from the tax break. Only in the first case do pension fund assets represent a net addition to national saving, whereas in the second case the effect on private saving is null and in the last case the higher private household saving is compensated by a lower public saving, resulting also in a null effect on national saving.Footnote 6
Most of the empirical evidence on this issue is taken from studies of the USA, particularly on individual retirement accounts (IRAs) and 401(k) plans, and is highly controversial. Surveys of this vast literature reveal very inconclusive and contradictory findings, from negative or null effects on national saving to very positive or even crowding-in effects (i.e., private saving would increase more than one monetary unity by each monetary unit contributed) (Engen et al., Reference Engen, Gale and Scholz1994; Bernheim, Reference Bernheim, Auerbach and Feldstein2002; Attanasio et al., Reference Attanasio, Banks and Wakefield2004; Börsch-Supan, Reference Börsch-Supan2004; Bosworth and Burtless, Reference Bosworth and Burtless2004; OECD, 2009; Attanasio and Wakefield, Reference Attanasio, Wakefield, Mirrlees, Adam, Besley, Blundell, Bond, Chote, Gammie, Johnson, Myles and Poterba2010; Attanasio and Weber, Reference Attanasio and Weber2010).Footnote 7 Studies based on the British case do not reach a consensus either (Guariglia and Markose, Reference Guariglia and Markose2000; Attanasio et al., Reference Attanasio, Banks and Wakefield2004; Rossi, Reference Rossi2009). Finally, two recent studies of Germany suggest that the introduction of tax-favored supplementary pensions (the so-called Riester reform) would not have contributed to increase saving (Börsch-Supan et al., Reference Börsch-Supan, Reil-Held and Schunk2007; Corneo et al., Reference Corneo, Keese and Schröder2010).
There is only one study of the Spanish case (Ayuso et al., Reference Ayuso, Jimeno and Villanueva2007), which explores the effects of the introduction of the tax breaks of the late eighties on consumption. They combine information from tax data and a household budget survey using a two-sample two-stage least-squares approach. They do not find any overall effect on saving, although positive and negative effects on particular age and income groups are reported.
This paper aims to contribute to the literature on this issue outside the USA. In order to do so, we profit from the use of fixed-effects techniques applied to a longitudinal household finance survey that includes information on wealth, pension funds and consumption. Because of reasons of availability, the use of panel databases has been very limited in previous work on this topic. Particularly, such type of literature is limited to the research work of Engen et al. (Reference Engen, Gale and Scholz1994) and Joines and Manegold (Reference Joines and Manegold1995), using the Internal Revenue Service-Michigan Tax Panel, and López-Murphy and Musalem (Reference López-Murphy and Musalem2004), who exploit a panel of countries. These three studies find that the contribution of voluntary pension funds to saving is very small.
3 Data
The database used in this analysis is the Spanish Survey of Household Finances (SSHF) of 2002 and 2005, the waves of the survey available at the moment of writing this paper, carried out by the Bank of Spain jointly with the National Statistics Institute (INE). The design of the survey was inspired by the American Survey of Consumer Finances and the Italian Survey of Household and Income Wealth and includes a multi-stage and stratified sampling, over-representing high-income households (Bover, Reference Bover2004, Reference Bover2008a). The survey contains detailed information on financial and non-financial wealth, income and durables and non-durables consumption of Spanish households. Furthermore, the two available waves of the SSHF allow the construction of a panel of 2,580 households (Bover, Reference Bover2008b), among which one third made contributions to supplementary pensions in 2002 or 2005.Footnote 8
One of the key issues in the survey was the treatment of missing values, whose presence is non-negligible in many variables related to income and, especially, wealth. After many efforts to minimize non-response, this issue was addressed by the designers of the SSHF using multiple imputation techniques, which were considered the most appropriate way of dealing with this problem (Barceló, Reference Barceló2006). Therefore, the Bank of Spain provided the researchers not only with original data but also with five sets of imputations to deal with the issue of missing values.
Both the descriptive and the multivariate analysis of the database were carried out using the five imputations included with the survey. All these calculations were performed using the software Stata 12.
4 Methodology
In order to explore the effects of pension contributions on national saving we follow the proposal of Attanasio and DeLeire (Reference Attanasio and DeLeire2002), who explore the impact of IRA tax deductions on saving in the USA. Following these authors, contributions to pension plans represent new national savings only when they result from lower levels of consumption. If participation in private pension schemes is not associated with lower consumption, then national savings does not increase, while, if the new saving exclusively comes from a higher level of disposable income because of tax relief, although private household saving is higher, the net effect on national saving is null because the increase in household saving is counteracted by a decrease in public saving linked to the lower tax revenue.
Our analysis entails a slight modification of the test proposed by Attanasio and DeLeire (Reference Attanasio and DeLeire2002). These authors do not use longitudinal econometric techniques but regress the change in consumption or saving in a short period of time on a set of household characteristics using a sample that contains households that contribute in some time to pension funds, assuming that this method allows for controlling for unobserved heterogeneity as long as contributing households should have similar preferences for saving and consumption. In the present paper, as our sample is smaller, we include both contributing and non-contributing households and estimate the regressions in levels but using fixed-effects techniques, which allow controlling for unobserved heterogeneity of all households.
In order to assess the impact of participation in pension plans on saving these authors suggest two kinds of tests. The first type of test is based on consumption and its main objective is to determine if enrolment in a pension plan leads to a lower level of household consumption, which, as mentioned earlier, is the only way by which these plans can boost national saving. In this strategy, we estimate the following expression:
$$C_{it} = \alpha + \beta X'_{it} + \gamma P_{it} + u_i + \varepsilon _{it} $$where C it denotes the consumption level of household i in time t, X it is a vector of household observable characteristics (detailed below), P it is the variable associated with participation in private pensions (either enrolment in a pension plan or yearly pension contributions depending on the specification), u i is a household-specific disturbance and ε it is a time-varying individual specific disturbance. The null hypothesis is that participation in pension plans does not affect consumption and, thus, it does not lead to increased national saving. This expression is estimated both in levels and logs.
The second test is based on household non-pension assets and consists in determining if pension plan contributions are made at the expense of existing assets or saving that would have been done anyway. The equation to be estimated unfolds as follows:
$$NPW_{it} = \alpha + \beta X'_{it} + \gamma P_{it} + u_i + \varepsilon _{it} $$where NPW it represents the non-pension wealth. The null hypothesis is now that pension saving does not negatively affect other types of saving, that is, that pension saving is not substituting other types of saving. We must have in mind that even if non-pension wealth is unaffected by pension saving, it does not mean that national saving increases as it requires a lower consumption level to be true. Furthermore, we investigate whether participation in pension plans affects the level of total household assets, that is, we estimate the equation
$$W_{it} = \alpha + \beta X'_{it} + \gamma P_{it} + u_i + \varepsilon _{it} $$where W it denotes the total household wealth.
In contrast to most previous empirical studies, we benefit from the use of panel data, which allows the removal of time-constant household unobserved heterogeneity using fixed-effects estimation (such unobserved heterogeneity is likely to play an important role in determining household saving decisions, as it has to reflect unobserved tastes for saving, attitudes to toward risk, ability, etc.). Therefore, the identification of the causal effect of pension contributions is achieved provided that the endogeneity of this variable is associated with time-varying observable characteristics and time-constant unobservable factors. One should bear in mind that, although we control for a wide range of observable characteristics, unobserved time-varying variables that are not independent of the dependent variable and contributions could lead to inconsistent estimators. In other words, our identification strategy assumes that there is no time-varying unobservable factor simultaneously affecting contributions and consumption or saving. In this respect, as mentioned above, using longitudinal data, this study goes a step further than most previous research.Footnote 9
As Attanasio and DeLeire (Reference Attanasio and DeLeire2002) suggest, the effects on national saving are more likely to appear when a household starts to contribute to a pension plan. Nevertheless, the effect should be observable in a longer period of time if there are not perfect capital markets or there are (tax deductible) contribution limits (as in the Spanish case). Therefore, we estimate the equations presented above using both a binary indicator of participation in pension plans and the annual level of contributions as key variables for determining the effect of the enrolment in pension plans on saving.
In Table 1, we list the variables included in the analysis, along with their definition. They comprise the outcome variables of the equations outlined above (consumption, total wealth and non-pension wealth), the policy variables of interest (being a contributor to a pension plan and total amount of contributions to pension plans) and the vector of observable characteristics used as control covariates: household head sex, age, education, marital status and employment status, household size, number of children <5 years old and between 5 and 15 years old, number of household members aged 65 years old and over, number of employed household members, number of household members with high educational attainment, household income and its squared and a time aggregate effect (a binary variable for the year 2005).
Table 1. Variables considered in the analysis
Source: Authors’ analysis from SSHF.
As is customary in policy evaluation and applied micro-econometrics (Angrist and Krueger, Reference Angrist, Krueger, Ashenfelter and Card1999; Duflo et al., Reference Duflo, Glennerster, Kremer, Schulz and Strauss2008; Angrist and Pischke, Reference Angrist and Pischke2009; Khandker et al., Reference Khandker, Koolwal and Samad2010), we focus on fixed-effects estimates irrespective of the correlation of time-constant unobserved heterogeneity with covariates. Random effects require much stronger assumptions for consistency, which, in practice, are difficult to fulfill. Particularly, consistency of random-effects estimates require that unobserved individual effects constant across time are not correlated with any of the observable covariates. In our particular case, this is not likely to be reasonable, as unobserved heterogeneity might be related to saving, attitudes to toward risk, ability, etc. and other factors that are constant over time and that are likely to be correlated with education, income, etc. Nevertheless, in the results section we return to this issue.
5 Results
The main descriptive statistics of the sample used in the analysis are presented in Table 2. The most relevant feature to be highlighted is the existence of significant differences in observable characteristics between contributing and non-contributing households: for instance, consumption, wealth, income or pension contributions are remarkably higher among contributing households.
Table 2. Main descriptive statistics of the sample
Note: Standard deviations have been computed using the first imputed dataset.
Source: Authors’ analysis from SSHF.
The results of the econometric analysis are displayed in Table 3.Footnote 10 First of all, they show that the null hypothesis asserting that contribution to voluntary pensions does not reduce consumption cannot be rejected in any of the proposed specifications, either using a binary variable for contributors or a continuous variable for contributed amounts. In other words, our results suggest that the contribution to private pensions does not reduce consumption, which would imply that, by definition, they cannot increase national savings. Both being a contributor and the amount of contribution seem to exert a positive rather than negative effect on consumption. When expressing consumption in logs, the impact of the binary variable remains positive but the effect of the continuous one vanishes. In order to check the robustness of the results, we repeated the analysis excluding durable goods from our consumption variable and the results hold. In addition, we performed the analysis restricting our sample to those households headed by individuals aged <65 years and, again, the results were basically the same as in the first set of estimates. As commented upon in the second section, one should keep in mind that both a null and a positive effect of tax incentives on consumption are perfectly coherent with standard Economic Theory. Apart from a substitution effect associated with a lower present consumption, other things being equal the tax break allow taxpayers to enjoy a higher disposable income – that is, it generates an income effect – inducing higher consumption. Which effect prevails is an empirical issue.
Table 3. Fixed-effects estimates of the impact of contributing to private pension plans on consumption and non-pension saving
Notes: Standard errors in parentheses.
***significant at 1%; **significant at 5%; *significant at 10%.
Control variables: an intercept, household head sex, household head, household head educational level, household head marital status, household head employment status, household size, no. of employed people in the household, no. of children aged <5 in the household, no. of children aged between 5 and 15 in the household, no. of people aged 65 and over in the household, no. of people with higher education in the household, household income, squared household income and a dummy for the year 2005.
Source: Authors’ analysis from SSHF.
In the second place, we test if being a contributor to a pension plan or the amount contributed is associated with lower non-pension wealth. In this case, the null hypothesis is that contributing to voluntary pensions is not compensated by lower non-pension assets, that is, there is no substitutability between both types of saving or pension saving would not have been generated otherwise. On the basis of the results obtained in this second test, we can reject the hypothesis of null substitutability when using the binary indicator of being part of a pension plan, whereas none of the estimated coefficients is statistically different from zero in the case of contributions. The results, therefore, are ambiguous and non-conclusive in this second case: private pensions would seem to substitute for other savings when using the binary indicator, but they would seem not to substitute when examining the actual amount contributed. The same results are found when we compute the impact of participation in complementary pensions on total household assets. Nevertheless, it should be highlighted that the standard error of the estimates is very large. It is worth mentioning that, according to the theoretical analysis of Attanasio and DeLeire (Reference Attanasio and DeLeire2002), in the absence of contribution limits to tax-favored contributions and capital-market imperfections, the eventual saving increase should be observed when contributions started rather than later. It is evident that liquidity constraints, capital market imperfections and contribution limits exist but this argument illustrates why the bulk of the eventual increase in household saving should be observed when a household start a pension scheme. This reasoning might help to make our results clearer, since the results for the effect of the binary variable (having a pension plan) are more robust than those obtained with the continuous one (amount of contributions).
In sum, these results suggest a null effect of contributions on national saving and an unclear effect on household private saving (the levels of significance are lower and the precision of estimates are larger). In the best of cases, the analysis depicts a situation where new household saving would be financed by the lower taxes paid by pension contributors. However, two cautionary notes should be kept in mind. Firstly, the estimated effect of contributions on household non-pension savings is extremely imprecise, probably because such an effect is too small to be clearly identified with a relatively small sample. In fact, pension wealth represents <4% of the total wealth of households. Secondly, missing values play a substantial role in the survey (an issue that, as mentioned in Section 3, is addressed using multiple imputation), especially regarding saving: while about 45% of the households surveyed had some imputed component of household saving in 2005, this proportion was <5% in the case of consumption. In principle, on the basis of this second feature, results based on consumption are less likely to be subject to measurement error.
Finally, it is worth recalling why we focus on fixed-effects estimates. As commented above, the use of random effects in the context of policy evaluation in applied micro-econometrics is rare, since consistency requires very strong assumptions and the efficiency gains associated with them if such assumptions hold tend to be small (Angrist and Pischke, Reference Angrist and Pischke2009). There is a range of Hausman-type tests described in the literature aimed at testing whether the absence of correlation between unobserved individual effects and covariates is reasonable. Basically, the idea behind these tests is to see if the difference between fixed-effects estimated coefficients and random-effect coefficients is statistically different from zero. However, as with any statistical test, there is the possibility of incurring a type II error, which consists in not rejecting the null when it is false (a false positive). Type II is related to the power of the test, which is lower the larger the variance of the coefficient to be estimated (in this case, related to the estimated variance of the fixed- and random-effect coefficients) and the smaller the sample size. In this case, a false negative might have very serious consequences: we would be leaving aside a consistent estimator (the fixed-effects estimator) and using an estimator with a smaller variance but that is inconsistent. In this paper, unfortunately, we are undoubtedly in a situation where the power of the test is small because we have a relatively small sample size and we have large standard errors not only because of the sample size but also because of the multiple imputed dataset. Therefore, irrespective of the results of any Hausman-type test one might carry out, there is a strong case for focusing on fixed-effects estimates. This particularly applies to the results obtained when the analysis is limited to households headed by people aged <65 years old, since the sample is somewhat smaller. This is a conservative position, since we are sacrificing efficiency for consistency.
Nevertheless, we implement a regression-based test proposed by Wooldridge (Reference Wooldridge2010), which is compatible with multiple imputed datasets like ours, which consists in, first, adding to the model the within-person means of time-varying variables (excluding aggregate time effects); second, estimating such augmented model using random-effects and, finally, testing whether the estimated coefficients for such within-person means are jointly statistically different from zero. Implementing this approach, we find that the null hypothesis (that coefficients for within-person means are zero, which has to do with the irrelevance of correlation between unobserved heterogeneity and covariates) is rejected at least at 5% confidence level in all cases but two: first, the model that explores the effect of having a private pension plan on total wealth in the sample that only includes household heads aged <65 and, second, when we study the effect of having a private pension plan on non-pension wealth in the same sample as the former. In the first case, the random-effects estimated coefficient for the dummy variable for having a private pension plan is negative but not statistically different from zero (−357,288.1, with a standard error of 445,741.8). In the second case, the random-effects estimated coefficient is negative and statistically significant at 10% level (−379,053.9, with a standard error of 226,412.4). Therefore, the results reported in the main text for these two cases are basically unchanged, suggesting no positive effect of voluntary private pensions on national saving.
6 Conclusions
Tax relief on supplementary private pensions, which have proliferated in recent years in OECD economies, has the promotion of saving as one of its primary objectives. The aim of this paper has been to assess to what extent this objective has been accomplished in the Spanish case. Using a longitudinal household finance survey and fixed-effect techniques, we have found that participation in this type of pension plan is not associated with a decrease in consumption. Therefore, a positive effect on national saving can be ruled out. This result is consistent with some macroeconomic evidence that suggests the lack of relevant effects of voluntary pension funds on national saving (López-Murphy and Musalem, Reference López-Murphy and Musalem2000). On the other hand, our empirical analysis has not provided conclusive evidence on the impact on household private saving. One should keep in mind the small size of the sample and the large standard errors associated with it and the multiple imputation procedure for missing variables. Therefore, although our research contributes to knowledge concerning the effects of voluntary private pensions on saving, it is far from being definitive and further research in this area (particularly research based on field experiments) is needed.
Overall, our findings cast additional doubts on the convenience of using these kinds of tax relief for fostering national saving. Furthermore, one has to keep in mind that policies that rely on tax breaks for accomplishing such aims have been evaluated as having a strongly regressive impact on income distribution (Burman et al., Reference Burman, Gale, Hall and Orszag2004; Hughes and Sinfield, Reference Hughes, Sinfield, Hughes and Stewart2004; Antón, Reference Antón2007). In this respect, it seems reasonable to explore other alternatives that, according to empirical evidence, are likely to be more equitable and effective policies for raising savings than tax credits or tax relief, such as matching incentives (Duflo et al., Reference Duflo, Gale, Liebman, Orszag and Saez2006, Reference Duflo, Gale, Liebman, Orszag and Saez2007) or the improvement of financial education (Lusardi, Reference Lusardi, Mitchell and Utkus2004; Lusardi et al., Reference Lusardi, Keller, Keller and Lusardi2008).
Annex
Table A1. Fixed-effects estimates of the impact of being a contributor to private pension plans on consumption and non-pension saving (detailed results, total sample)
Table A2. Fixed-effects estimates of the impact of being a contributor to private pension plans on consumption and non-pension saving (detailed results, total sample) (continued)
Table A3. Fixed-effects estimates of the impact of contributions to private pension plans on consumption and non-pension saving (detailed results, total sample)
Table A4. Fixed-effects estimates of the impact of contributions to private pension plans on consumption and non-pension saving (detailed results, total sample) (continued)
