Chapter 4. Are financial incentives effective tools to increase participation in and contributions to retirement savings plans?

Effect of personal income tax rates and tax brackets

The empirical evidence reviewed here indicates that the design of the personal income tax system, in terms of tax rates and tax brackets, can encourage participation in and contributions to retirement savings plans for which contributions are deductible from taxable income. The evidence is less conclusive when contributions are taxed at a reduced fixed rate. In most countries, the personal income tax system is progressive, meaning that marginal tax rates increase with taxable income. Therefore, tax relief on contributions increases when taxable income jumps from one tax bracket to the next. The following studies examine how the design of the personal income tax system and reforms affecting tax rates influence participation in and contribution levels to such tax-favoured plans. In all these studies, the authors include controls for the level of income, so that the tax effect is not distorted by the effect of income.

Milligan (2002[2]) finds that marginal tax rates affect moderately the decision to contribute to Registered Retirement Savings Accounts (RRSPs) in Canada. RRSPs are personal pension plans for which contributions are deductible from income, investment earnings are exempt from taxation and withdrawals are taxed as income (“EET”). Using repeated cross-sections of the Family Expenditure Survey between 1982 and 1996, the author determines the effect of marginal tax rates, which vary over time and across provinces, on RRSP participation. He finds that marginal tax rates influence households’ decisions to participate in an RRSP, but this influence is relatively small. A 10 percentage point increase in the marginal tax rate is estimated to increase the probability of participation by 8%. This tax effect only explains 5.1% of the trend in RRSP participation between 1982 and 1996.

Focusing on self-employed individuals in the United States, Power and Rider (2002[3]) argue that taxes have a substantial effect on both the decision to contribute and the amount contributed to tax-favoured retirement savings plans. Self-employed individuals in the United States have access to Keogh plans and Simplified Employee Pension (SEP) plans. Under such plans, the self-employed person is treated as an employee of himself and can make tax-deductible contributions, up to certain limits. Using a panel of sole proprietors constructed from the Statistics of Income individual income tax files for 1985, 1989 and 1993 the authors determine the effect of changes in tax rates due to several tax reforms on participation in and contributions to Keogh and SEP plans. They distinguish the effect on low- and high-income earners, who were affected differently by the reforms. They calculate the tax price of contributions as (1-µ_C)/(1-µ_W), where µ_C and µ_W represent respectively the marginal tax rate on contributions and the marginal tax rate on withdrawals, so that the tax price increases when the tax incentive is reduced. They find that the tax price of contributions has a substantial effect on both the decision to contribute and the amount contributed to Keogh and SEP plans by the self-employed. The estimates suggest that a 1% increase in the tax price results in a 0.25 percentage point decrease in the probability of contributing and in a 2% decrease in contributions.

Crawford, Disney and Emmerson (2012[4]) provide empirical evidence that individuals facing higher marginal tax rates respond to tax incentives by increasing their participation in tax-favoured retirement savings plans. The authors examine whether participation in and contributions to private pension plans in the United Kingdom increase at the higher rate threshold for income tax. Since tax relief is given at the marginal tax rate through tax-deductible contributions, participating in a private pension plan should be more attractive to individuals just above the higher rate threshold than to those just below it in terms of immediate tax relief on contributions. The test is not simply to assess whether higher-income earners are more likely to participate in a pension plan but whether there is a discontinuity in participation at the higher rate threshold. Using repeated cross-sections of the Family Resources Survey between 2000-01 and 2008-09, the authors find evidence that individuals above the higher rate threshold of income tax are more likely to be members of a pension scheme. The authors however warn that this result should be interpreted with caution, as this could be either because the tax relief on contributions increase at that point or because, more generally, becoming a higher rate taxpayer has an effect on the individual’s perception of the need for tax planning. In addition, they find no evidence of an increase in the contribution rate at the higher rate threshold.

Carbonnier, Direr and Slimani Houti (2014[5]) find that high-income earners in France respond to the presence of tax incentives by contributing more to retirement savings accounts that mandate annuitisation at retirement (e.g. PERP plans). The authors use a large micro-file tax return dataset that includes incomes earned between 2006 and 2009. They first calculate the distance between the taxable income (i.e. household’s income net of all contributions in tax-deferred savings accounts) and the income tax thresholds so as to check whether individuals save in tax-favoured pension plans in order to declare a taxable income that would fall into a lower tax bracket. They then compare the contribution level to retirement savings accounts of two groups of individuals: i) those whose taxable income is higher than but close to a given tax threshold; and ii) those whose taxable income is just below the same threshold. They conduct this comparison at different tax thresholds and for different age groups.3 The authors find no evidence of individuals saving in order to reduce taxable income and fall into a lower tax bracket. However, they find that the tax incentive is effective in increasing contributions to PERP plans for high-income earners whose marginal tax rate is the highest, especially for those aged 45 and above.

Unfortunately, among low-income earners, the combination of a progressive tax system and tax-deductible contributions may not be effective in encouraging savings for retirement. Carbonnier, Direr and Slimani Houti (2014[5]) find that low-income earners aged 45 to 55 do not increase their contribution level when their marginal tax rate increases. This suggests that the structure of the income tax system is not the main factor for low-income earners in deciding what their contribution level should be.

Harju (2013[6]) shows that the 2005 tax reform in Finland affected individuals’ participation in and contributions to voluntary personal pension plans. Before the reform, contributions to personal pension plans were deducted from labour income, returns on investment were exempt from taxation and retirement benefits were taxed as labour income. Since the reform, contributions are deducted from capital income and benefits are taxed as capital income (returns are still tax exempt). The tax schedule of personal pension plans therefore changed from progressive to proportional, as capital income is taxed at a fixed rate. In other words, tax relief on contributions increased for low-income earners (their contributions are deducted at a higher rate since the reform) and decreased for high-income earners (their contributions are deducted at a lower rate since the reform). The author compares the change in personal pension participation and contributions of individuals affected by the 2005 reform and individuals not affected. He uses panel data covering the period from 2000 to 2007. The reform altered saving behaviour by reducing personal pension plan participation among high-income earners by 4 percentage points and increasing it among low-income earners by 1 to 2 percentage points. The reform also reduced annual contributions among high-income earners by over 20%. The estimated effects result entirely from the changed savings behaviour of men.

Finally, Feng (2014[7]) fails to demonstrate a positive impact of a reduced tax rate on contributions on participation in salary sacrifice arrangements in Australia. A salary sacrifice arrangement is where employees agree contractually and voluntarily to give up part of the remuneration that they would otherwise receive as salary or wages, in return for their employer providing contributions to a superannuation fund of the same amount. Salary sacrifice contributions attract a favourable tax treatment, being taxed at a fixed rate of 15% (same as for mandatory employer contributions), rather than at the marginal income tax rate. Using data from wave 10 of the Household Income and Labour Dynamic in Australia survey, the author determines the effect of higher tax relief on contributions on the participation in salary sacrifice arrangements at two marginal tax rate jump points of the personal income tax system, from 15% to 30% and from 30% to 38%. The results indicate that tax rate relief leads to a small but insignificant increase in participation in salary sacrifice arrangements. The author explains that this result is likely due to the complexity of the incentive schemes and competing demands for long-term savings in Australia.

Effect of changes in caps on tax-relieved contributions

Another group of studies provides evidence that individuals react to changes in caps on tax-relieved contributions. When contributions to tax-favoured retirement savings plans are deductible from income, there is usually a cap on the amount of contributions that attract tax relief. Reducing or increasing this cap influences the level of contributions, and also potentially the participation in tax-favoured retirement savings plans.

Milligan (2003[8]) finds that future changes in contribution limits have an impact on current contributions to RRSPs in Canada. Contribution limits for RRSPs underwent reform in 1990 and 1991. Using panel data of tax filers between 1987 and 1991, the author looks at the effect of a future change in the contribution limit on the change in current contributions for individuals who were affected differently by the reforms. He finds a negative relationship between future contribution limits and current contributions. The intuition behind this result is the following. When future contributions to the tax-preferred savings account are constrained by a contribution limit, current contributions increase in order to make use of the contribution room while it is available. The author calls it the “use-it-or-lose-it” motivation to contribute. By contrast, when the limit increases, a larger contribution may be made in the future. This diminishes the need of the taxpayer to contribute today.

Disney, Emmerson and Wakefiled (2007[9]) show that the change in the tax regime in 2001 in the United Kingdom affected participation in and contributions to personal pension accounts. Together with the introduction of Stakeholder Pensions in April 2001, the government changed the structure of tax relief which implied a significant increase in contribution limits to private pensions for low-income earners, especially for younger age groups. Using repeated cross-sections from the Family Resources Survey between 1999-2000 and 2002-03, the authors compare private pension participation before and after the change in the tax regime for those who were potentially affected by the contribution limit increase and for those who were not. They find that the change in the contribution limits affected both participation rates and contributions to private pensions among low-income earners, especially women and people leaving in couples. Private pension participation increased by around three percentage points more among individuals who were affected by the reform than among those who were not affected. Evidence also suggests an increase in contributions to private pensions among those affected by the reform: around GBP 0.8 per week for singles and GBP 4.3 per week per individual in couples.

Finally, Rutledge, Wu and Vitagliano (2014[10]) argue that the “catch-up” provision is effective in increasing contributions to 401(k) pension plans in the United States. The catch-up provision was enacted in 2001 and extends the 401(k) contribution limit for workers older than 50. In particular, the higher limit provides an incentive for workers previously constrained by the limit to increase their contributions, but does not change the incentive for workers unconstrained by the limit. The authors use longitudinal U.S. Social Security Administration data on tax-deferred earnings linked to the Survey of Income and Program Participation for the period from 1999 to 2005. They compare the change in 401(k) contributions for maximum contributors just after the age of 50 to similar participants just under 50, in the years around the policy change. The study finds that, compared with similar workers under 50, contributions increased by USD 540 more among individuals aged 50 and older who had approached the 401(k) contribution limits prior to turning 50. This suggests that older workers previously constrained by the contribution limit respond to the expanded contribution limit. The authors estimate that a one-dollar increase in the contribution limit leads to a 49-cent increase in 401(k) contributions.

Effect of tax credits

Tax incentives can also take the form of tax credits. The only available studies on tax credits refer to the Saver’s Credit in the United States, which supplements “EET” tax incentives for low and middle-income earners. The Saver’s Credit is a tax credit introduced in 2002 and designed to encourage retirement savings among low and middle income households. These individuals can claim a non-refundable tax credit on retirement contributions of up to USD 2 000 made to private pension plans, with credit rates equal to 10%, 20% or 50% at different income thresholds.

The take-up rate for this programme is very low. According to the Internal Revenue Service, in 2015, only 5.4% of all tax filers received a tax credit and the average credit was USD 178, far below the maximum USD 1 000. This low participation number is partially explained by the fact that the statistics consider all tax filers while only low and middle-income earners are eligible for the credit. Brown and John (2017[11]) provide several potential explanations for the limited impact of the programme. First, many low and middle-income earners do not contribute to private pension plans, mostly because they do not have access to occupational plans at their job.4 Second, many do not apply for the credit because of a general lack of awareness of the credit and the complexity of the claim form. Third, low-income earners may not have sufficient tax liability to receive the tax credit because it is non-refundable.

In addition, there are no studies showing that a higher credit rate increases contributions to private pensions. Ramnath (2013[12]) examines whether taxpayers underreport their income in order to receive a higher credit rate and analyses whether households that receive a higher credit rate contribute more. She uses Public Use Tax Files between 2001 and 2006 from the Internal Revenue Service and finds that taxpayers had a reported income just below the thresholds created by the Saver’s Credit more often after than before the introduction of the Saver’s Credit. However, when looking at the impact of changes in the credit rate on changes in average retirement contributions among taxpayers claiming the credit and having a reported income around the income thresholds, she finds no statistically significant evidence that receiving a higher credit rate increases individual contributions to private pensions.

Evidence for an increase in national savings

A first group of studies, mostly in the United States, concurs with Poterba, Venti and Wise (1996[16]) that, on aggregate, tax incentives increase retirement savings through an increase, at least moderate, in national savings.

Benjamin (2003[18]) estimates that, on average, about half of 401(k) balances represent new private savings and about one quarter represents new national savings. Using cross-section data from the 1990 Survey of Income and Program Participation, the author compares the wealth of 401(k) eligible and ineligible households within subgroups constructed according to a method that ensures that households in the same subgroup are comparable on many characteristics.5 The author estimates that, on average, about one quarter of aggregate 401(k) assets represents new national savings, one quarter represents savings done due to the tax break on contributions (i.e. forgone tax revenues), and one quarter results from conversions from pre-existing defined contribution (DC) plans or reduced defined benefit (DB) promises. The last quarter presumably represents reallocation from other asset classes, in particular home equity. In addition, he finds that responses to eligibility vary considerably across households, with households who normally save the most (homeowners, IRA participants, the better educated) largely reallocating other savings into 401(k) plans and not increasing their overall savings.

Ayuso, Jimeno and Villanueva (2007[19]) find a small but significant impact of tax incentives on savings, especially for high-income earners aged 46 to 55. The authors examine the impact of the introduction of tax-favoured private pension funds in Spain in 1988 on contributions to pension funds and on savings. Using a panel of income tax returns, the authors first observe that individuals in the top income quartile contribute more to private pension plans. The second part of their analysis therefore focuses on these individuals. Using the 1985-1991 waves of the Household Expenditure Survey they compare the consumption growth of individuals aged 36 to 65 to that of individuals younger than 35, before and after the introduction of private pensions. They find small consumption drops among individuals close to retirement age (aged 56 to 65 years old), the group that most actively contributed to the plan. They find a larger decrease in consumption expenditures for individuals aged 46 to 55. One way to interpret the results could be that households close to retirement consider pension funds and other savings forms as strong substitutes, and tend to contribute up to the tax-deductibility limit by reallocating their savings. By contrast, younger households, with plausibly less accumulated wealth and for whom contribution limits are less likely to be binding, need to increase their savings to take advantage of the tax incentives. They estimate that the overall amount of new savings is at most 25 cents per euro contributed on average.

Guariglia and Markose (2000[20]) and Rossi (2009[21]) find that contributions to personal pension schemes do not reduce ordinary savings in the United Kingdom. Both studies use several waves of the British Household Panel Survey (between 1992-1998 and 1994-2003, respectively) to analyse the displacement effect on savings due to the introduction of personal pension plans. Guariglia and Markose (2000[20]) examine the determinants of savings in personal pension plans and compare with those for conventional savings. They conclude that the two types of savings are accumulated for different purposes: voluntary contributions to personal pension plans are made essentially for retirement purposes, while conventional saving is undertaken for precautionary motives. In addition, they find no evidence of a negative relationship between the two types of savings, concluding that they are unlikely to offset each other completely. Rossi (2009[21]) determines the effect of private pension contributions on ordinary savings and models simultaneously the two decisions of saving, in ordinary forms and in personal pensions. She concludes that an additional contribution put aside in a personal pension plan would have no detrimental effect on ordinary savings, but would actually increase them. The two savings products are therefore complementary and not substitutes, as those respondents who save in pension forms also have higher savings.

Gelber (2011[22]) compares two groups of individuals eligible for 401(k) plans and shows that becoming eligible increases 401(k) balances, without reducing savings in other financial assets. Using longitudinal data on households’ savings from the 1996 Survey of Income and Program Participation, the author uses the fact that some employers impose a vesting period in their 401(k) plan to identify employees who change eligibility status. The author therefore compares the change in household savings of individuals who become eligible for a 401(k) plan in their second year in a job and individuals who remain eligible over the same period. Unsurprisingly, he finds that 401(k) eligibility raises 401(k) balances. The author did not find statistically significant evidence that 401(k) eligibility impacts savings in other financial assets, although the confidence intervals are so large that the author cannot completely rule out a negative impact. He also finds that, in response to 401(k) eligibility, IRA assets increase and the value of individuals’ cars decreases (the car value could be considered as a proxy for consumer durables as a whole), suggesting an increase in national savings.

Finally, using the introduction of a Roth option on 401(k) contributions in 2006, Beshears et al. (2017[23]) argue that upfront taxation increases savings. Since 1 January 2006, employers have had the option to include a Roth contribution alternative in their 401(k) retirement savings plan. Roth 401(k) contributions are not tax-deductible in the contribution year, but investment earnings and withdrawals in retirement are tax free (upfront taxation, “TEE”). This contrasts with traditional 401(k) accounts, for which contributions are tax-deductible in the contribution year, investment earnings are tax free and withdrawals are taxed (“EET”). With a Roth 401(k) option, individuals have the possibility of reducing their contributions to their 401(k) plan to keep the level of future, after-tax, retirement income constant. Using administrative data from 11 companies that added a Roth contribution option to their existing 401(k) plan between 2006 and 2010, the authors find no evidence that total 401(k) contribution rates differ between employees hired before and after introducing a Roth option, after controlling for differences in the socio-economic characteristics of the two cohorts of employees. In order to explain this result, the authors ran an online survey experiment on 7 000 DC plan participants. Respondents were asked to make a recommendation about a 401(k) contribution rate for an imaginary couple. Consistent with the previous findings, the authors find evidence that individuals are confused about and neglect the tax properties of traditional and Roth 401(k) accounts. The experimental results also suggest that “partition dependence” helps keep total contribution rates constant when a Roth option is introduced. Partition dependence results from a bias towards allocating an equal amount to every discrete option available in a choice set and predicts an increase in total savings when the choice set is larger, for example {current consumption, traditional 401(k) savings, Roth savings} instead of {current consumption, traditional 401(k) savings}.

Evidence for savings reallocation

Another body of the literature tends to argue that the increase in retirement savings following the introduction of tax incentives is mostly the result of savings reallocation, rather than new savings.

Attanasio and DeLeire (2002[24]) and Attanasio, Banks and Wakefield (2004[25]), argue that both in the United Kingdom and the United States, only relatively small fractions of assets in tax-favoured accounts can be considered to be new savings. For the United States, the authors compare changes in consumption, saving rates and non-IRA financial assets between two groups of individuals: “new IRA contributors” (households that just opened an IRA account) and “continuing IRA contributors” (households that previously had made IRA contributions). They use panel data from the Consumer Expenditure Survey between 1982 and 1990. The authors find no evidence that households financed their IRA contributions from reductions in consumption, at least in the first nine months following their initial participation in an IRA plan. They find evidence that households financed their IRA contributions from existing savings or from savings that would have been done anyway. Their results indicate that, at most, 9% of IRA contributions represented net additions to national savings.

For the United Kingdom, Attanasio, Banks and Wakefield (2004[25]) look at some descriptive statistics following the introduction of two specific savings vehicles with upfront taxation: Tax Exempt Special Savings Accounts (TESSAs) and Individual Savings Accounts (ISAs). TESSAs were introduced in 1991 and were subsequently replaced by ISAs in 1999. For both accounts, contributions are paid from net income and there is no subsequent tax on returns and withdrawals (“TEE”). The total amount that can be contributed to both accounts each year is limited. Looking at the average size of TESSA balances between 1991 and 1999, the authors observe a regular pattern of large contributions at the beginning of the year, followed by no further contributions for the rest of the fiscal year. They claim that this pattern would not typically be associated with reduced consumption patterns, but would rather signal a reallocation of assets. Using data from the Family Resources Survey between 1998 and 2002, the authors also show that the increase in participation rates of tax-favoured savings products following the introduction of ISAs was not concurrent with an increase in ownership of assets more generally, suggesting that ISAs did not encourage new savings.

Pence (2002[26]) finds that eligibility for 401(k) plans has a small but not significant effect on savings in the United States. The author compares changes in wealth of 401(k) eligible and ineligible workers between 1989 and 1998, using repeated cross-sections of the Survey of Consumer Finances. Although real median 401(k) balances grew substantially, from USD 4 000 in 1989 to USD 11 000 in 1998, the wealth of eligible households did not grow relative to the wealth of ineligible households. This result holds across income groups and when non-401(k) retirement assets are added to the wealth measure. The author concludes that differences in saving taste and initial wealth are likely to better explain the large account balances that eligible households accumulated over the 1990s, than the tax incentives embedded in 401(k) plans.

Contrary to Ayuso, Jimeno and Villanueva (2007[19]), Antón, Muñoz De Bustillo and Fernández-Macías (2014[14]) argue that tax incentives for retirement savings in Spain do not increase national savings. The authors assess the effect of participation (enrolment and contributions) to a pension plan on three different dependent variables: consumption, non-pension assets and total wealth. They use longitudinal data from the Spanish Survey of Household Finances of 2002 and 2005. The results suggest that contributions to pension funds do not increase national savings since they are not associated with a reduction in consumption. However, the evidence is inconclusive on the link between pension fund contributions and households’ non-pension assets and total wealth, meaning that these contributions may not come from a reallocation of other household assets or saving that would have been done anyway. In that case, the tax incentive would lead to an increase in private household savings but not in national savings, as the additional savings would come from the higher disposable income allowed by the existence of the tax relief.

Using panel data on savings covering the population of Denmark, Chetty et al. (2014[27]) show that tax incentives primarily induce individuals to reallocate assets from taxable accounts to tax-favoured retirement savings plans. The authors use a panel dataset that registers the wealth and savings of over 4 million individuals from 1994 to 2009. They assess the effect of tax incentives on total household savings by examining how individuals responded to a tax reform in 1999. That year, the government reduced the deduction that high-income earners could make for contributions to “capital” pension accounts, which are accounts that provide lump-sum payments at retirement. That reform, however, left unchanged the tax treatment of contributions to accounts that provide annuity payments.6 The authors compare the change in pension and non-pension savings for individuals with incomes above and below the income threshold for the top tax bracket. Individuals in the top income tax bracket significantly reduced their contributions to capital pensions following the reform. However, this reduction was driven by only 19% of the high-income earners affected by the reform, who changed their saving behaviour by reducing or stopping contributions to capital pensions. These “active savers” redirected nearly all that saving to annuity pension plans and other savings accounts, leaving their overall savings mostly unchanged. The vast majority of the workers affected by the reform were “passive savers” who continued to contribute the same amounts to both types of retirement plan and to other savings accounts. The authors conclude that the entire increase in retirement savings attributable to the previous level of the tax incentive had come from money that would have been saved, and was now being saved, in taxable savings accounts.

Finally, Paiella and Tiseno (2014[32]) find that the pension fund legislation passed in Italy in 1993-1995 had a negligible effect on saving rates. This reform was issued with the explicit goal of raising private long-term savings to offset the reduction in public pension benefits. It encouraged the creation of a proper private pension system to supplement public pensions by establishing pension funds as separate legal entities subject to a set of specific rules. It was aimed at increasing participation rates and ensured a favourable tax treatment for contributions (tax-deductible up to a limit) and withdrawals (taxed at favourable rates). To assess whether tax incentives of pension fund contributions have a positive effect on household savings, the authors compare changes in saving rates of pension fund contributors and non-contributors before and after the reform, using repeated cross-sections of the Survey on Household Income and Wealth between 1989 and 2006. The results show that the reform had a negligible effect on saving rates but led to a significant reduction in liquid financial assets (bank accounts and government bonds), suggesting a reallocation of savings.

Why is there such a disparity of results?

This section discusses the reasons behind such a disparity of results. It summarises some of the constraints and difficulties making the empirical measurement of whether tax incentives for retirement savings plans lead to an increase in national savings a complex task. There are a number of methodological issues that the different studies try to address in different ways. Without getting into sophisticated econometric and statistical explanations, some of the main issues are described below.

The first issue is that the effect of tax-favoured retirement plans on savings is difficult to measure accurately with the data at hand. Ideally, one would need to run an experiment where individuals would be split randomly into two groups. The individuals in the two groups would share the same characteristics overall and would be treated exactly in the same way, except that one group would be allowed to participate in a tax-favoured plan (“treatment group”), while withholding this possibility from the other group (“control group”). In this way, access to the tax-favoured plan is independent from individuals’ characteristics (and in particular their preferences to save) and this ensures that, when comparing total savings in the two groups, any difference can be attributed to the effect of the tax incentive only. Unfortunately, such random experiments are not easy to run and econometricians need to mimic the ideal experiment using data at hand.

In the case of private pension plans in general, what is observable in the data is the participation in these plans rather than the access to them. However, private pension contributors usually consist of households with stronger preferences to save, which means that they also save in other vehicles. The comparison of savings between contributors and non-contributors therefore combines effects of both the preference to save and the tax incentive.

In the case of the United States, it is possible to observe individuals’ eligibility for 401(k) pension plans. Some authors see this eligibility status as independent from the preferences to save, as only individuals whose employer establishes a 401(k) plans are eligible to contribute to such a plan (Benjamin, 2003[18]; Poterba, Venti and Wise, 1996[16]). They consider that, as eligibility is determined by employers, eligible and ineligible households are likely to have comparable preferences to save. However, other authors (Bernheim, 2002[33]; Engen, Gale and Scholz, 1996[17]) criticize this, claiming that employees with strong preferences to save probably tend to choose jobs that provide access to 401(k) pension plan, and that employers may consider employees’ preferences when deciding to set up a 401(k) plan. To address the possibility that eligibility correlates across individuals with their unobserved preferences to save, Gelber (2011[22]) examines a change in eligibility status, using the fact that some employers impose a vesting period in their 401(k) plan. He therefore compares two groups of eligible workers who plausibly have comparable preferences to save.

Another issue is that there is a trade-off between controlling for many variables that may interact with households’ preferences to save and the precision of the estimates. It is necessary to control for factors that can influence savings decisions in order to isolate the effect of tax incentives in the models. However, savings (and consumption) can be influenced by many factors. For example, age, gender, marital status, education, income and the number of children are some of the demographic and socioeconomic characteristics that influence savings decisions. Reforms affecting the generosity of public pensions also interact with savings decisions. Many studies document the effect of public pension wealth on savings and report a decline in savings for each unit of increase in public pension wealth (Alessie, Angelini and Van Santen, 2013[34]; Disney, 2006[35]; Hurd, Michaud and Rohwedder, 2012[36]; Lachowska and Myck, 2015[37]). There is therefore a trade-off between adding control variables and the precision of the estimates. Benjamin (2003[18]) overcomes this problem by classifying 401(k) eligible and ineligible households within subgroups defined according to the propensity score method. This method allows for controlling for a large number of household characteristics without reducing the precision of the estimates.

The choice of the dependent variable is important to determine whether tax incentives may have an impact on private savings only or more broadly on national savings.7 While most studies analyse the effect of tax-favoured retirement savings plans on households’ savings or wealth (Engelhardt, 2000[28]; Pence, 2002[26]), some studies analyse the effect of these plans on households’ consumption (Ayuso, Jimeno and Villanueva, 2007[19]) and other studies look at the effect on both consumption and savings (Antón, Muñoz De Bustillo and Fernández-Macías, 2014[14]; Attanasio, Banks and Wakefield, 2004[25]). As explained before, an increase in private savings does not necessarily translate into an increase in national savings, in the case when individuals’ contributions come from the higher disposable income allowed by the existence of the tax relief. Therefore, studies looking at the impact of tax incentives on consumption are more appropriate to conclude on the effect on national savings.8 In addition, looking at the impact of tax-favoured plans on different components of wealth can help our understanding of from which accounts households may reallocate their assets into tax-favoured ones.

Studies differ in their definition of household wealth and this may have implications for whether or not tax incentives may be found effective in raising savings. These definitions may be constrained by the data used. For example, Rossi (2009[21]) only considers ordinary savings, i.e. amounts saved in a bank, building society, or Post Office account. In Poterba, Venti and Wise (1996[16]) the wealth measure includes all financial assets. They argue that financial assets are the closest substitute for pension assets. Other studies consider that a broader definition of wealth should be used. For example, in Benjamin (2003[18]) total wealth equals net financial assets, including IRA and 401(k) assets, plus housing equity plus the value of business, property, and motor vehicles. Including housing wealth is motivated by several considerations. First, private pensions and housing are both illiquid, tax-favoured assets that are often held for long periods. Second, there are many ways to substitute between private pensions and housing equity, for example, by not accelerating mortgage payments or by not buying a bigger house. Engelhardt (2000[28]) has an even broader definition of wealth, which includes 401(k) assets, other pension assets, social security wealth, and financial and housing wealth. His view is that it is important to consider all assets that finance retirement consumption. In particular, individuals who are not eligible to join a 401(k) plan may have access to other types of pension plan (i.e. DB and non-401(k) DC plans). Not accounting for these assets in other pension plans may bias the estimated effect of tax incentives on household wealth.

Measurement errors in survey data can weaken the precision of estimates of the effect of tax incentives on household savings. Most studies use survey data to examine the effectiveness of tax incentives. These studies give access to rich information about households’ demographic and socioeconomic characteristics, which are necessary to control for factors that may influence saving decisions. However, individuals may not give precise answers regarding their eligibility for private pension plans, their income and the amount of assets and wealth that they hold in various products. For example, by comparing self- and firm-reported pension data, Engelhardt (2000[28]) finds significant measurement error in 401(k) eligibility in the self-reported data, and the error varies with household income. Low-to-middle income households understate eligibility, while high-income households overstate eligibility. In addition, the measurement error is highly correlated with saving behaviour. This leads to an upward bias of the effect of 401(k) eligibility on savings. Combining survey data with administrative registers (to have accurate measures of income, wealth and pension plan characteristics) can reduce measurement errors and improve the precision of estimates.

Longitudinal studies are better suited than cross-sectional ones to assess the impact of tax incentives for retirement savings plans, as it is necessary to observe the same individual or household over long periods of time. These long-term and repeated observations allow the analysis of the correlation between savings and participation in tax-favoured plans. Repeated cross-sections are imperfect substitutes for panel studies as they do not guarantee that households surveyed in different years are similar on all their characteristics (observable and non-observable). The study by Chetty et al. (2014[27]) is striking in this respect in the sense that the authors use administrative data (with a higher quality level and lower measurement error than self-reported information) for a large panel of individuals. They merge data from several administrative registers (the income tax register, the population register, and the Danish Integrated Database for Labour Market Research) to obtain annual information for the Danish population from 1995 to 2009. Their analysis relies on a panel of approximately 41 million observations for 4 million individuals and therefore leads to very robust results.

Finally, it may be difficult to generalise results when the effect is measured over a short period of time and/or for a specific population subgroup. For example, Attanasio and DeLeire (2002[24]) examine the consumption and saving behaviour in the first nine months after a household starts making contributions to an IRA. If the change in behaviour happens after nine months, their model would miss it. In the same vein, Engelhardt and Kumar (2011[31]) analyse the extent to which pension wealth crowds out non-pension wealth, but their sample is limited to older households from the Health and Retirement Study (i.e. individuals aged 51 to 61 year olds and their spouses). The authors acknowledge that their results may not fully characterize the saving response of younger workers to changes in pension wealth.

Low-income households are more likely to respond to tax incentives by increasing overall savings

The last group of studies analysed in this section distinguishes the effect of tax incentives by earnings level and shows that low-to-middle income earners are more likely to respond to tax incentives by increasing their overall savings, while high-income earners tend to reallocate their savings. All these empirical studies refer to the United States. They argue that, as low-income earners have little wealth, when they decide to contribute to a tax-favoured retirement savings plan, their contributions essentially represent new savings, as they have few other assets to reallocate. By contrast, high-income earners are more likely to finance contributions to tax-favoured accounts by shifting assets from other taxable accounts, rather than reducing their consumption.

Engelhardt (2000[28]) shows that 401(k) eligibility significantly increases household savings for low-to-middle income households, but that the overall impact on savings is not significant. The author combines self-reported and firm-reported pension information in order to account for measurement errors in both types of data. He uses a broad definition of household savings and uses detailed data on 401(k) pension plans, other pensions, Social Security, and private wealth from the 1992 Health and Retirement Study. The study shows that 401(k) eligibility significantly increases household savings for low-to-middle income households. This effect decreases with income and is not statistically different from zero for middle-to-upper income households. When all households are considered in aggregate, the study shows that 401(k) eligibility does not have a significant impact on household savings. In addition, as 401(k) eligibility does not seem to induce a substitution within financial assets and is inversely correlated with other pension assets, the results imply significant firm-level substitution of 401(k) for other pension assets.

Engen and Gale (2000[29]) find that the impact of 401(k) eligibility on households’ wealth varies significantly by earnings level. The authors use repeated cross-sections of the Survey of Income and Program Participation in 1987 and 1991. They compare the growth of households’ wealth within six different earnings groups for eligible and ineligible households. On the one hand, they find that for groups with low earnings, the impact of 401(k) eligibility on savings is significant and as large as the increase in 401(k) balances for eligible households in this group, suggesting that they finance all of their 401(k) contributions by increasing their savings. On the other hand, 401(k) eligibility has no significant impact on wealth accumulation for groups with high earnings. Overall, because high-earnings groups hold most of 401(k) assets, the authors estimate that only between 0% and 30% of 401(k) balances represented net additions to private savings between 1987 and 1991.

Chernozhukov and Hansen (2004[30]) examine the distributional effect of participation in 401(k) pension plans on wealth. Conditional on eligibility for a 401(k) plan, the authors determine the effect of 401(k) participation (defined as having a positive 401(k) balance) on net non-401(k) financial assets, net financial assets and total wealth. They use data from a sample of households from the 1991 Survey of Income and Program Participation and run a quantile regression which decomposes the effects for people in different deciles of the wealth distribution. They find that the effects of 401(k) participation on net financial assets and on total wealth are positive and significant over the entire range of the asset distribution, while the effect on non-401(k) financial assets is not significantly different from 0. For low-wealth households, the increase in net financial assets seems to translate entirely into an increase in total wealth. However, for high-wealth households, there is significant evidence of substitution from other (non-financial) forms of wealth, as the increase in total wealth due to 401(k) participation is much smaller than the increase in net financial assets.

Finally, Engelhardt and Kumar (2011[31]) show that pension wealth raises total household wealth and that most of the reallocation effect is concentrated in higher-wealth households. The authors aim at identifying the extent to which an increase in pension wealth is associated with a decline in non-pension wealth (so-called “crowding-out” effect). Rather than focusing on the effect of one particular type of plan, such as 401(k) plans, the authors include all types of occupational pension plan plus social security pensions to determine the effect of pension wealth on non-pension wealth at different points in the non-pension wealth distribution.9 They use both detailed administrative datasets and survey information from the 1992 wave of the Health and Retirement Study to get accurate data on pensions and lifetime earnings for older workers. They find that each dollar of pension wealth is associated with a 53 to 67 cent decline in non-pension wealth, meaning that pension wealth also raises total household wealth. In addition, most of the crowding-out effect is concentrated in higher-wealth households, with a decline in non-pension wealth of 30 to 75 cents per dollar of pension wealth. By contrast, the authors argue that policies targeted at increasing pension wealth for lower-wealth households will raise overall household wealth accumulation essentially dollar-for-dollar (no crowding-out effect).

To sum up, a reasonable estimate for the share of new savings in total retirement savings in tax-favoured plans would be between a quarter and a third. The studies presented earlier estimate that between 9% and 100% of retirement savings represent new savings. As suggested by Hubbard and Skinner (1996[15]), the actual impact of tax incentives probably lies between those two extremes. Low-to-middle income earners are more likely to respond to tax incentives by increasing their overall savings, but they hold disproportionately less of private pension assets. By contrast, high-income earners hold a large share of total private pension assets but tend to reallocate their savings. The overall share of new savings is therefore likely to be well under 50%. Following Ayuso, Jimeno and Villanueva (2007[19]), Benjamin (2003[18]), Engelhardt and Kumar (2011[31]), Engen and Gale (2000[29]) and Hubbard and Skinner (1996[15]), a reasonable estimate could be that new savings represent between a quarter and a third of retirement savings in tax-favoured plans.

Impact of matching contributions on participation in and contributions to retirement savings plans

Evidence gathered from the United States and Australia suggests that matching contributions increase participation in retirement savings plans. When combined with automatic enrolment, the additional effect of an employer match is modest. The evidence is mixed regarding the impact of matching on total contributions. None of the studies reviewed assess the impact of matching contributions on national savings.

Empirical evidence from a large body of the literature in the United States suggests that employer matching contributions increase participation in occupational pension plans. Choi (2015[41]) and Madrian (2013[42]) review the rich literature in the United States regarding the effect of matching contributions on participation in and contributions to private pension plans. Both conclude that the presence of a matching contribution increases participation and that the higher the match rate, the higher the impact on participation. As occupational pension plans already benefit from a favourable tax treatment (“EET”), employer matching contributions improve on the effect of tax incentives on pension plan participation.

Madrian (2013[42]) however qualifies the link between the match rate and the participation rate as modest. This statement relies on a study by Engelhardt and Kumar (2007[43]). Using a nationally representative random sample of individuals aged 51 to 61 and their spouses from the Health and Retirement Survey, they find that increasing the match rate by 25 percentage points (for example, from 25 cents per dollar of employee contribution to 50 cents) raises 401(k) participation by 5 percentage points. Choi (2015[41]) reports that an increase of 10 percentage points in the match rate increases the participation rate in a range from 2.5 percentage points to 6.3 percentage points, depending on the study.

By contrast, evidence in Australia suggests that low-to-middle income earners may be sensitive to the match rate and maximum entitlement. The government encourages low-to-middle income earners to make voluntary savings through the super co-contribution programme.10 In July 2012, the government reduced by half both the match rate (to 50%) and the maximum entitlement (to AUD 500).11 Between 2011-12 and 2012-13 the number of beneficiaries dropped by 40% and the payments dropped by 60%. Clearly, many people felt that the incentive was not high enough to continue making voluntary contributions to the superannuation system.12

Beshears et al. (2007[44]) look at the impact of matching contributions on participation in the context of automatic enrolment and find a limited effect. Many companies in the United States combine both automatic enrolment and employer matching contributions. The authors analyse the additional effect of an employer matching contribution on participation in automatic enrolment plans. A priori, the effect is not obvious. On the one hand, the opt-out rate could be much higher without a match, since the incentives to participate are reduced. On the other hand, if employee inertia drives the automatic enrolment participation effect, high participation rates would still be expected even without a matching contribution. The authors use the variation in the employer match structure and generosity within and across nine firms with automatic enrolment between 2002 and 2005 to assess the effect of the maximum employer match on plan participation. They find that a decrease in the maximum employer match by 1% of pay reduces the plan participation rate at six months of eligibility by 1.8 to 3.8 percentage points. Therefore, the success of automatic enrolment at increasing participation in occupational pension plans does not appear to rely much on having an employer match.

The impact of matching on total contributions to retirement savings plans is not clear. Choi (2015[41]) and Madrian (2013[42]) both show that the empirical evidence in the United States is mixed. Some studies find a positive relationship between the match rate and contribution levels, others find no relationship and yet other studies find a negative relationship.

Choi et al. (2002[45]) show, however, that the rate of employee contribution up to which the employer offers the match (i.e. the match threshold) has an impact on employee contributions. They study a company with a 50% match rate that increased its match threshold from 5% to 7% of income for union workers and from 6% to 8% of income for management employees. They observe an immediate change in the distribution of employee contribution rates towards an increase in the proportion of participants contributing between 7% and 8%.

In addition, in Australia, government matching contributions may be effective at raising contribution rates for low-income earners. OECD (2012[46]) shows that, despite the co-contribution programme, low-income earners are less likely to make voluntary contributions to the superannuation system than other income groups. However, those who do contribute voluntarily tend to have much higher contribution rates than other income groups, as they need to make a larger contribution effort to get the maximum entitlement.

Impact of fixed nominal subsidies on participation and on national savings

Fixed nominal subsidies paid by the government into pension accounts can be found in Chile (bonus per child), Germany (Riester plans), Lithuania (second pillar pensions), Mexico (social quota) and Turkey (automatic enrolment scheme).13 They are designed to attract low-income individuals, as the subsidy represents a higher share of their income.

Evidence related to the introduction of the Riester pension scheme in Germany shows that subsidies are effective in attracting families with children and low-income earners.14 Indeed, despite the fact that participation rates of Riester pensions increase with income, Riester pensions are more equally distributed by income than occupational pensions and unsubsidised private pension plans. The different studies also suggest that Riester pensions have a positive impact on overall savings, although the evidence is mixed for low-income earners.

Börsch-Supan, Coppola and Reil-Held (2012[47]) show that Riester pensions achieve higher coverage rates among low-income households than other private pension plans. Using data from the SAVE survey between 2001 and 2010, the authors show that the proportion of households holding any type of private pension plans (Riester plans, occupational plans and other private pension plans) increases with disposable household income. In 2009, only around 19% of households in the lowest income bracket had taken up Riester pension plans, while almost a half of those in the two upper income brackets participated in the scheme. However, the distribution of participation is even more skewed towards high-income earners for occupational pension plans and other private pension plans, which cover only around 5% of households in the lowest income bracket. Therefore, among low-income households, Riester pensions are much more common than any other form of private pension provision. As a result, Riester pensions are more equally distributed by income than occupational pensions or unsubsidised private pension schemes. In addition, in contrast with occupational pensions, the uptake of Riester plans has been quite dynamic for all income groups between 2003 and 2010, including in the bottom income quintile. The authors therefore conclude that the Riester scheme has not completely failed in attracting low-income earners.

Börsch-Supan, Coppola and Reil-Held (2012[47]) also find no evidence that Riester pensions have reduced savings in other savings vehicles. Using individuals’ declaration about changes in their saving behaviour analysed by Coppola and Reil-Held (2010[48]), they show that only a minority of households say that they saved less in total since starting to be enrolled in a Riester pension plan, while most households respond that they have saved more. This trend holds across all income groups. In addition, at a macro level, the increase of participation in Riester pensions was not accompanied by a decline in participation in other types of private pension plans.

The other studies described below do not measure participation in Reister pensions but look at the impact of the introduction of Riester plans on households’ savings, in particular for low-income households. Their results suggest that low-income households reallocate existing savings from non-subsidised contracts to subsidised savings contracts, somehow contradicting individuals’ declarations as reported in Coppola and Reil-Held (2010[48]). However, this may be because very low earners only need to contribute EUR 60 to get the full subsidy. This amount would be easy to reallocate from other savings accounts.

Corneo, Keese and Schröder (2008[49]) argue that the introduction of the Riester pension scheme had no impact on the probability to save or on the saving rate of low-income households. The authors use data from the German Socio-Economic Panel between 2000 and 2006 to assess the impact of the Riester scheme on households’ saving propensities. They use income to distinguish between people with different incentives to save in a Riester plan, as they do not observe which individuals actually participate in a Riester plan in each year in the data. They calculate the “subsidy ratio” as the public subsidy (in the form of fixed nominal subsidies or tax deductions) divided by the total savings amount for additional old-age provision. They observe that the subsidy ratio is the lowest for middle-income households, as the fixed nominal subsidies represent a larger share of income for low-income households, while high-income households benefit from tax deductions. Therefore, the authors compare the probability to save and the saving rates before and after the introduction of Riester plans for low-income households (with high subsidy ratio) and middle-income households (with low subsidy ratio). The findings suggest that the probability to save and the saving rates of low-income households did not increase more after the introduction of the Riester scheme despite them having higher incentives to save in such a plan. The authors therefore conclude that, rather than generating new savings, Riester plans seem to induce people to reallocate existing savings from non-subsidised contracts to subsidised savings contracts.

Using the same database, Kolerus, Koske and Hüfner (2012[50]) find that the introduction of the Riester pension scheme in 2002 was associated with a general increase in the household savings rate, except for low-income households, thus confirming the findings of Corneo, Keese and Schröder (2008[49]). The authors analyse the driving factors of changes in the German household savings rate over the period 1991 to 2008 and in particular the impact of the introduction of the Riester scheme in 2002 on households’ savings rate. They find that the introduction of the Riester pension scheme in 2002 had a positive effect on the household savings rate, both for households that signed up for a Riester plan and for those that did not (suggesting that the intensive debate about the importance of private retirement savings that was triggered by the reform also had an impact). However, the interaction term between being a low-income household and having a Riester plan is significant and negative, suggesting that low-income households do not save more after signing up for a Riester plan but rather reduce other forms of savings in exchange for subsidised savings under the Riester scheme.

Pfarr and Schneider (2013[51]) concur with the two previous studies that low-income earners do not increase their old-age provision because of the Riester scheme. They find however a positive effect of the child subsidy on participation. Using panel data from the SAVE survey between 2005 and 2009, the authors examine simultaneously the determinants of participation in Riester plans and in private pension insurance. They find that the decisions to participate in a Riester plan and in private pension insurance are related. The impact of the number of children is positive and significant on Riester participation, while it is insignificant or even negative on private pension insurance participation. This shows that child subsidies have an impact on the take up of Riester plans. However, low-income earners are less likely to participate in a Riester plan and in private pension insurance, suggesting that the incentives in the Riester scheme are not leading to higher old-age provisions for the low-income earners. In addition, alternative investments (building loan contracts, cash-value life insurance contracts, accumulation plans, stocks and investment bonds) positively affect the uptake of both forms of provisions, suggesting that individuals use various forms of assets to make provisions for old age.

Finally, it is interesting to note that the design of the subsidies does not encourage individuals to contribute above the 4% threshold. OECD (2012[46]) shows that average contribution rates are similar across all income groups, around the 4% minimum required by legislation to obtain the full subsidy. This threshold may act as an anchor in people’s mind and does not provide any incentive to go beyond it.

Impact of a mix of policies on national savings

KiwiSaver is a voluntary retirement savings scheme introduced in 2007 in New Zealand with the aim of increasing the retirement wealth of the population. It offers an interesting case study as it uses automatic enrolment and offers both matching contributions and fixed nominal subsidies (up to 2015). Its launch was also accompanied by effective communication and education campaigns (OECD, 2014[52]). In terms of participation, KiwiSaver is a real success with over 2.6 million people enrolled as at June 2016. In addition, KiwiSaver members’ distribution by income is similar to the one of the eligible population, meaning that low-income earners are not disproportionately left out of the scheme (Inland Revenue, 2015[53]).

Two studies assess the impact of the KiwiSaver scheme on national savings and find little to no effect. However, they only cover the first 3.5 years of the scheme, which may not be sufficient to reach any final conclusion. The two papers assess the KiwiSaver scheme as a whole and do not disaggregate the effect by policy. Non-tax incentives are however an important factor as most members of the scheme to date have joined the plan voluntarily, not through automatic enrolment.

Law, Meehan and Scobie (2011[54]) assess individuals’ saving behaviour after the introduction of the KiwiSaver scheme and find little to no effect on national savings. The authors use a national survey conducted in 2010 that asked respondents how the contributions they were making at the time to KiwiSaver would have been used in the absence of the scheme. On average, KiwiSaver members reported that they would have used 64% of the money to save in other vehicles or to reduce their debt had they not joined KiwiSaver. In other words, only 36% of their KiwiSaver contributions represent additional savings over and above those that would have been made anyway. This share is reduced to 29% after accounting for the fact that high-income earners contribute a disproportionate share of total savings. This share of new private savings may not be enough to offset the state contributions (through borrowing), such that, in the long run, KiwiSaver’s effect on national saving is found to be negligible.

The analysis in Law and Scobie (2014[55]) provides consistent results. The authors use linked data from the Survey of Family, Income and Employment and from Inland Revenue Department administrative data on KiwiSaver membership, covering the period 2002 to 2010. They analyse the extent to which membership of KiwiSaver is associated with greater accumulations of net wealth. Their results show that KiwiSaver membership has no significant positive effect on net wealth accumulation, suggesting that KiwiSaver contributions come primarily from a reallocation of other savings.

There is an important caveat for the two previous papers though, as they assess the impact of KiwiSaver up to 2010. This means that the analysis only considers the first 3.5 years of the scheme. In addition, the impact of the global financial crisis was still large in 2010. As the analysis in Law and Scobie (2014[55]) does not distinguish the changes in net wealth due to financial asset price movements from those due to individuals modifying their saving behaviour, the New Zealand Institute of Economic Research (2015[56]) argues that the above results should not be generalised to the future.