Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/de-gruyter/does-the-early-release-of-retirement-savings-prolong-labor-market-RN0cIQqjxf
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- Publisher
- de Gruyter
- Copyright
- © 2022 Andrew Dudley Carter, published by Sciendo
- ISSN
- 2193-9004
- DOI
- 10.2478/izajolp-2022-0010
- Publisher site
- See Article on Publisher Site
Abstract
1IntroductionGovernments around the world are encouraging older citizens to remain in the labor market for longer to combat the economic and fiscal challenges posed by aging populations. Relatively static retirement ages combined with lengthening life expectancies have resulted in larger shares of the population utilizing publicly funded pensions, health, and aged care services. Governments have typically responded by increasing the qualifying ages for social security benefits and by offering targeted incentives to remain in the workforce.Increasing the qualifying ages of benefits has been found to be an effective way to increase labor supply rates for older workers,Mastrobuoni (2009), Staubli and Zweimller (2013), Hanel and Riphahn (2012), Vestad (2013), and Atalay and Barrett (2015) examine these effects in the United States, Austria, Switzerland, Norway, and Australia, respectively. while targeted incentives are often found to have little impact.See Disney et al. (2010), Ramnath (2013), Feng (2014), and Laun (2017). This paper contributes to the international literature by examining Australia's Transition to Retirement Income Streams (TRIS),Also known as Transition to Retirement (TTR) provisions. a program that offers a novel approach to incentivizing labor supply for older workers.TRIS was introduced on July 1, 2005, and continues to be available. TRIS intends to combat the concern of workers retiring prematurely just to access their retirement savings. Prior to TRIS, workers had to be aged ≥65 years to access their retirement savings while remaining in the labor market. TRIS offers limited early access to a worker's compulsory retirement savings (known as superannuation in Australia) for those in the 55- to 65-year age range. Policy makers envisaged that TRIS could be used as an income supplement for those who reduce their working hours as they “transition to retirement”. TRIS amounts are taxed at the individual's marginal income tax rate minus a 15% tax offset. From July 1, 2007, superannuation withdrawals, including TRIS, became tax free for workers aged >60 years. This change further incentivized program participation for this age group. An unintended consequence of the program is the opportunity for individuals to minimize tax without decreasing working hours. “Tax-effective” strategies can be devised by drawing a TRIS while, in the same year, cycling salary-sacrificed (pre-income tax) contributions back into their superannuation fund.See Section A.2 in Appendix for more information on „tax-effective” strategies. The Productivity Commission (2015) indicated that it is difficult to precisely ascertain the purpose for which people are using TRIS and, in particular, the extent to which TRIS incomes are encouraging people to remain in the labor market for longer versus simply as a mechanism to minimize tax. The key question this paper seeks to address is whether TRIS increases labor force participation of older workers. We also present some evidence on behavior that appears to be consistent with tax minimization, a response that seems at odds with the program's intent.We consider the program in the context of a simple life cycle model that serves as a framework to interpret the empirical results. Specifically, we use this model to understand the effect of the TRIS program on an individual's optimal retirement age. We find that the model has confounding effects, which leads to ambiguous optimal retirement age predictions. Despite this, our intuition suggests that the TRIS program is likely to increase the optimal retirement age.We empirically estimate the labor supply and earnings response using administrative data from the Australian Taxation Office (ATO). We use a difference-in-differences (D-i-D) design to measure an “intention-to-treat” effect by comparing the changes in participation and earnings in the program's initial years by exploiting the qualifying age threshold (55 years). Similarly, following a policy change from July 1, 2007, which introduced tax-free superannuation withdrawals for people aged >60 years, we repeat this analysis to examine the additional “tax free” effect for this age group. The causal estimates we report are local treatment effects for the “treatment” age groups only.We find no employment response in the program's first year, a small positive effect in the program's second year for males only (1.0%), and larger effects in subsequent years for both males (1.4%) and females (1.1%). The timing and magnitude of the effects appear consistent with program adoption rates, which were low initially. We suspect that the slow adoption of the program is partly explained by a lack of program awareness and other frictions that dampen adoption. Adopting TRIS requires action on the individual's behalf; including research, possible financial advice, and setting up a TRIS pension account with one's superannuation fund. Individuals on higher incomes are more likely to use and materially benefit from TRIS.Modest deviations from the common trends assumption weaken the causal interpretation of our D-i-D estimates. In these years, we provide explanations for other known fiscal, macroeconomic, and demographic changes. However, the consistency between the program adoption rates and the causal estimates provide some confidence that we are detecting the program effects.We contribute to the international literature on tax system design for older workers approaching retirement. We adapt a life cycle model to consider the effect of the program on retirement age and empirically estimate the labor supply response. In addition, we provide previously unavailable insight into TRIS program adoption.2Background2.1Superannuation in AustraliaAustralia's universal superannuation scheme was introduced in the early 1990s with the goal to provide income in retirement while reducing the reliance on the publicly funded age pension. For the majority of people, superannuation is held in industry and retail funds, which are regulated by the Australian Prudential Regulation Authority (APRA).There is, however, an increasing share of people who elect to use self-managed super-annuation funds (SMSFs) to allow greater flexibility and control over the management of their retirement income.The Australian Taxation Office (ATO) and the Australian Securities and Investment Commission (ASIC) work collaboratively to regulate SMSFs. The rules and governance arrangements are complex and have changed over time. Employers are generally required to make superannuation contributions on an employee's behalf at the Superannuation Guarantee contribution rate, though some employers voluntarily elect to pay more. The Superannuation Guarantee rate is currently 9.5% of an employee's ordinary time earnings. In general, employer superannuation contributions, along with superannuation earnings within the fund, are taxed concessionally at a flat rate of 15%.Currently, superannuation is only partially funding retirement for individuals leaving the labor market. This is because most retirees tend to have lower superannuation account balances due to the fact they have only received compulsory employer contributions for part of their working lives and at comparatively low rates throughout the Superannuation Guarantee's introductory years. This means that many retirees will continue to rely on the age pension for some time until the superannuation system matures.The concessional tax treatment of superannuation contributions, earnings, and income streams are designed to encourage and bolster retirement savings. These concessions, however, result in a high public cost in forgone tax revenue and the benefits disproportionally go to the wealthy.Department of the Treasury (2012) analysis shows that the share of superannuation tax concessions disproportionally goes to those on higher incomes. In 2012–13, e.g., the Treasury estimates that the top 5% of contributors received 20.3% of contribution concessions. Official estimates indicate that superannuation concessions on earnings and contributions are the second- and third-largest tax expenditures after the tax-free treatment on sales of owner-occupied housing. The Department of the Treasury (2018) estimates that the tax relief for superannuation earnings and contributions accounts for $19.3 billion and $16.9 billion, respectively, in 2016–17 alone. These figures emphasize the need to understand how the superannuation system is performing and its distributive effects. For further detail, the Productivity Commission (2015) published a review of Australia's retirement income system, which highlights design issues and incentives that are embedded in the system.2.2Transition to Retirement Income Streams (TRIS)TRIS is an Australian Government program that intends to enhance the labor supply of workers aged between 55 years and 65 years. The program was introduced on July 1, 2005, and continues to be available. TRIS offers limited early release to a worker's compulsory superannuation (retirement) savings. Prior to the introduction of TRIS, a superannuation fund member had to satisfy the “conditions of release” to access their savings. For most, this involved reaching the superannuation preservation ageAccess to superannuation is generally restricted to those who have reached their preservation age. The preservation age is based on an individual's date of birth. It is 55 years of age for individuals born before July 1, 1960, and gradually increases to 60 years in 1-year increments for individuals born after June 30, 1964. and retiring from the labor market. There was concern among policy makers that these conditions would lead to workers prematurely leaving the labor market just to access their savings. TRIS aims to mitigate this effect by enabling limited “early access” superannuation drawdowns for qualifying workers. Policy makers envisaged that TRIS could be used to supplement the salaries of those who reduce their working hours as they “transition to retirement”.TRIS offers up to 10 years early access to superannuation for a worker who remains in the labor market up to the age of ≥65 years. This represents a 10-year differential between the superannuation preservation age (55 years) and the age that an individual is free to access their superannuation irrespective of their working status (65 years). A person in the 60- to 65-year age range was required to leave the labor market to access his or her superannuation, while a person in the 55- to 60-year age range was additionally required to declare that he or she had no intention of returning to the labor market.TRIS conditionally relaxed these requirements by offering a capped noncommutable superannuation income stream (i.e., not a lump sum) for those who met the superannuation preservation age requirements and continued to work. The annual TRIS must be no <4% of an individual's account balance at the beginning of the financial year and no >10%. TRIS attracts the equivalent tax treatment as would apply for retired individuals. This includes a 15% tax offset on the annual amount of the income stream, along with tax-free earnings within the super-annuation fund.Superannuation fund earnings would otherwise attract a 15% tax rate while the fund is in an accumulation phase. This tax-free exemption for earnings on TRIS accounts was later repealed from July 1, 2017. From July 1, 2007, the Simplified Superannuation package brought a change that made superannuation drawdowns, including TRIS, tax free for people aged >60 years. This increased the financial incentive to adopt TRIS.TRIS was designed to supplement the incomes of older workers who decide to reduce their engagement in the labor market. This could include those who move from full- to part-time working arrangements or who otherwise reduce their work responsibilities with a corresponding reduction in remuneration. A potential failing of the TRIS design is that a work test was not implemented. Full-time workers are eligible to use TRIS even if they have no intention of transitioning to retirement. A work test was abandoned in the original design as it was argued that it would place an unreasonable compliance burden on superannuation funds. Hanegbi (2013) suggests that this burden should be shifted onto the taxpayer seeking to use TRIS through self-assessment.For more detail, Section A.1 in Appendix provides a sense of the program benefits through three simplified use-case examples. Section A.2 in Appendix provides details on the strategies that can be applied to reduce one's tax liability. We define a “tax-effective strategy” as a scenario whereby workers draw a TRIS pension while, within the same year, make salary-sacrificed (pre-income tax) contributions back into their superannuation fund.3Conceptual FrameworkWe adapt a simple life cycle framework to consider the effect of the TRIS program on an individual's choice to continue working or retire. Specifically, we seek to understand the effect of the program on an individual's optimal retirement age. Life cycle models have been widely used in the pensions and retirement behavior literature. Relevant examples we draw from are presented by Burbidge and Robb (1980) and, more recently, Atalay and Barrett (2015), who consider the effects of changes to pension plans on the retirement decisions of individuals. The models begin with a person who seeks to determine the optimal time to retire from the labor market. More time in the labor market results in higher savings in retirement. The trade-off of working longer is less leisure.The life cycle model assumes that an individual maximizes his or her lifetime utility subject to his or her lifetime budget constraint. Utility is defined as a function of consumption and leisure U(Ct,Lt), where marginal utility is held constant over the life cycle. For simplicity, we assume there are only two states in the life cycle: (i) the period that an individual works; and (ii) the period that an individual is retired. An individual begins work at time “0”, assumes he or she will live to a specific age T, and will spend time until age R working in the labor market. By construction, T minus R equals the time spent in retirement.The discounted value of lifetime utility V over time t is presented as follows:(1)V=∫0RU(Ct,0)e−δtdt+∫0RU(Ct,1)e−δtdt,V = \int_0^R {U\left( {{C_t},0} \right){e^{ - \delta t}}dt + } \int_0^R {U\left( {{C_t},1} \right){e^{ - \delta t}}dt,} where the time spent working is denoted as “0”, time in retirement is “1”, and δ is the discount rate per period of time t. An individual works full time unless he or she chooses to move to a part-time work arrangement by utilizing the TRIS program. Therefore, leisure can only be varied over the life cycle by retiring or, in this analysis, by participating in the TRIS program. TRIS income is captured in the first integral of Eq. (1) given participation in the program is conditional on working.The lifetime budget constraint (Eq. (2)) shows that the lifetime discounted value of consumption C equals the discounted value of income from work Y, income from TRIS TR, and the retirement income RI. A constraint on TR is provided in Eq (2) given TRIS must be no <4% of a worker's account balance at the beginning of the financial year and no >10%. To simplify the model, we define retirement income as income drawn from superannuation only. TRISq represents the qualifying age for the TRIS program. The first two integrals in Eq, (2) overlap from the TRIS qualifying age TRISq until the retirement age R, given participation in the TRIS program is conditional on working.(2)∫0TCte−rtdt=∫0R(1−α)Yte−rtdt+∫TRISqR(1−β)TRte−rtdt+∫RT(1−θ)RIie−rtdt, s.t. TRt={RIt⋅x|0.04≤x≤0.10}\matrix{{\int_0^T {{C_t}{e^{ - rt}}dt} } \hfill & = \hfill & {\int_0^R {\left( {1 - \alpha } \right){Y_t}{e^{ - rt}}dt + \int_{TRI{S_q}}^R {\left( {1 - \beta } \right)T{R_t}{e^{ - rt}}dt} } } \hfill \cr {} \hfill & {} \hfill & { + \int_R^T {\left( {1 - \theta } \right)R{I_i}{e^{ - rt}}dt,\,{\bf{s}}{\bf{.t}}{\bf{.}}\,T{R_t} = \left\{ {R{I_t} \cdot x|0.04 \le x \le 0.10} \right\}} } \hfill \cr } In Eq. (2), α, β, and θ represent the tax rates on earnings from work, TRIS, and retirement income, respectively. We account for the tax settings to emphasize the tax concessions that TRIS TR and retirement income RI attract, relative to income from work Y. Recall that for TRIS recipients, β will be zero for individuals aged ≥60 years from 2007–08 and will attract 15% tax offset on this income otherwise. From this point, to simplify the notation, we do not include the tax rate components or the TRt constraint in subsequent equations.The relationship between the income components in Eq. (2) is complicated and depends on several factors. In Eq. (3), we show that retirement income RI at time t is a function of the lifetime income from work Y and the lifetime income from TRIS TR. The derivative of income from work will be >0 (fY > 0) given that more time in the labor market will result in more retirement income. The derivative of TRIS income will be <0 (fTR < 0) given the program is providing early access to retirement income RI. The latter, however, assumes that TRt is greater than post-tax contributions and earnings within an account for a given year. This may not be the case for all workers, depending on the level of TRIS drawn in a given year, and it is particularly unlikely for those who devise tax-effective strategies.See Sections A.1 and A.2 in Appendix for further information on tax planning with TRIS.(3)RIt=f(Y,TR).R{I_t} = f\left( {{\bf{Y,TR}}} \right).The optimization problem with respect to C and R can be expressed as a Lagrangian function, using Eqs (1) and (2), and solved for any value of R, where R is constrained as a value >0 and <T. The individual seeks to maximize utility subject to the budget constraint with respect to C and R. For simplicity, we express U(C,0) = UCW and U(C,1) = UCR and let δ equal r.(4) ZR ZTL= UCWe−rtdt+ UCRe−rtdt0 R (ZT ZR ZR ZT)−λ Cte−rtdt− Yte−rtdt− TRte−rtdt− RIte−rtdt 0 0 TRISq R\matrix{{\;\;\;\;\;\;\;\;ZR\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;ZT} \hfill \cr {L = \;\;\;\;\;\;\;\;\;\;\;\;UCWe - rtdt + \;\;\;\;UCRe - rtdt} \hfill \cr {0\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;R} \hfill \cr {\;\;\;\;\;\;\;\left( {ZT\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;ZR\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;ZR\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;ZT} \right)} \hfill \cr { - \lambda \;\;\;\;\;\;\;\;\;\;\;\;\;Cte - rtdt - \;\;\;\;\;\;\;\;Yte - rtdt - \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;TRte - rtdt - \;\;\;\;\;\;\;RIte - rtdt} \hfill \cr {\;\;\;\;\;\;\;\;0\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;0\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;TRI{S_q}\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;R} \hfill \cr } The first-order conditions (excluding the budget constraint) state that the individual's marginal utility of consumption while both working and in retirement are equal. This equals the Lagrange multiplier λ constant (Eq. (5)) which, by definition, represents the marginal utility of wealth.(5)UCW=UCR=UC=λ;U_C^W = U_C^R = {U_C} = \lambda ;(6)UCR−UCW+λ{Yt+∫TRISqRdTRdRe−rt dt−RIt}=0.U_C^R - U_C^W + \lambda \left\{ {{Y_t} + \int_{TRI{S_q}}^R {{{dTR} \over {dR}}{e^{ - rt}}\,dt - R{I_t}} } \right\} = 0.The individual seeks to maximize utility subject to C and R. We set Eq. (6) to zero and, using Eq. (5), rearrange Eq. (6) to arrive at Eq. (7).(7)UCR−UCWUC=Yt+∫TRISqRdTRdRe−rt dt−RIt.{{U_C^R - U_C^W} \over {{U_C}}} = {Y_t} + \int_{TRI{S_q}}^R {{{dTR} \over {dR}}{e^{ - rt}}\,dt - R{I_t}} .The left-hand side of Eq. (7) shows the marginal rate of substitution between retirement and consumption. This is the marginal utility gained from an additional year in retirement divided by the marginal utility of an increase in consumption per year.The comparative statics reveal that TRIS influences an individual's optimal retirement age R* in two ways. We first consider the response to an increase in working income resulting from participation in the TRIS program. We define working income WI as follows:WI=∫0RYte−rt dt+∫TRISqRTRte−rt dt.WI = \int_0^R {{Y_t}{e^{ - rt}}\,dt + } \int_{TRI{S_q}}^R {T{R_t}{e^{ - rt}}\,dt} .Therefore, a positive “income effect” is introduced when TR > 0, along with a “price effect” given the concessional tax treatment β of this income (recall, β is shown in Eq. (2)). Together, the effects are similar to a wage increase or, equivalently, an increase to an individual's budget constraint. A wage increase has two competing responses in the life cycle model. First, a wage increase reduces the optimal retirement age. This is because income and savings goals can be achieved earlier than they would have otherwise in the absence of the program. Second, there is a substitution effect. A higher wage will, contrarily, increase lifetime consumption. People may, therefore, choose to substitute time in retirement for additional time in the labor market to increase income and savings. The theory on the dominant effect is ambiguous. Our intuition for the TRIS program, however, suggests that the substitution effect will be stronger. This is because the income effect may appear small in the context of total lifetime earnings, noting that TRIS program benefits are restricted to the latter years of one's working life. A dominant substitution effect will, therefore, increase the optimal retirement age R*.A second response arises due to the TRIS program's interaction with “savings on retirement”, defined as retirement income RI in our model. The effect on the optimal retirement age R* will depend on how TRIS usage changes an individual's retirement income. For TRIS program participants, accessing some of their retirement income early may result in less superannuation on retirement than would have otherwise been the case.Recall, RI=∫RTRIte−rt dtRI = \int_R^T {R{I_t}{e^{ - rt}}\,dt} so, if ∂RI <0, then it can be shown that ∂R*/∂RI <0. In this scenario, the life cycle model predicts that the optimal retirement age will increase in response to a decrease in retirement income, assuming that leisure is a normal good. The opposite is true in a scenario where retirement income increases. Using a tax-effective strategy to boost retirement income is a way in which an increase in savings can happen. It is, however, still possible for retirement savings to increase without using a tax-effective strategy. This can be achieved by drawing a modest TRIS, which is more than offset by ongoing contributions back into superannuation and fund earnings in the working years prior to retirement.Taken together, the net effects from the life cycle model's theoretical predictions on R* are ambiguous.Please see the study by Burbidge and Robb (1980) for further details on the mathematics and graphical analysis behind the comparative statics discussion in this section. Our intuition, however, suggests that the TRIS program is likely to increase the optimal retirement age. This is based on the following logic. First, the addition of TRIS TR to lifetime working income WI will likely increase the optimal retirement age, assuming a dominant substitution effect. Second, the retirement income RI response may be positive to the degree that workers are using TRIS as policy makers envisaged. That is, as an income supplement to smooth consumption for those who reduce working hours as they “transition to retirement”. This will result in a gradual reduction in retirement savings in each year that TRIS clauses are utilized. Ultimately, teasing out the magnitude of the two responses is a matter for empirical research.4Related LiteratureWe are not aware whether there are other countries that have piloted approaches such as TRIS to prolong participation. As such, the literature on the effects of similar programs appears scarce. Typically, early access to compulsory retirement savings are only provided in rare and exceptional circumstances, including severe financial hardship, on compassionate grounds, or for people with terminal medical conditions. The literature on retirement decisions in response to policy change usually focuses more on changes to policy parameters, such as increases to the pension or social security-qualifying ages. Examples include the papers by the following authors: Mastrobuoni (2009), Staubli and Zweimller (2013), Hanel and Riphahn (2012), Vestad (2013), and Atalay and Barrett (2015), who examine these effects in the United States, Austria, Switzerland, Norway, and Australia, respectively. These studies find that increase in the qualifying age is effective at prolonging time in the labor market, with larger responses identified for lower-educated workers. Meanwhile, Hanel (2010) examines a financial incentive that aimed to delay retirement following a pension reform in Germany. The reform introduced a change that was estimated to have reduced pension benefits for early retirees, which delayed retirement by about 10 months on average. Laun (2017) examines the effect of age-targeted credits on labor force participation of older workers in Sweden. The results show small positive extensive margin effects, which lead the author to conclude that tax incentives for older workers can be a viable tool for delaying retirement.Ramnath (2013) examines taxpayer responses to the Saver's Credit program in the United States. This is a tax incentive designed to encourage retirement savings among low- and middle-income earners. The author notes that, while the incentives are generous, take-up of the program is low due to its complexity and the nonrefundable nature of the credit; meaning a substantial share of the target group does not realize the benefits. Similarly, Feng (2014) examines the effect of tax incentives on salary-sacrifices (pretax) superannuation contribution participation in Australia. The author finds that participation in the program is relatively low, despite generous tax incentives. Various reasons are put forward to explain these results, including the lack of knowledge of the policy, competing vehicles for long-term saving, and a common belief that compulsory saving through Australia's superannuation guarantee will be “enough” to fund one's retirement. Similarly, Disney et al. (2010) examine the participation of a new private pension arrangement in the United Kingdom, which aimed to incentivize retirement savings. They find little or no impact on savings behavior. The authors also note that there is little agreement in the literature in terms of what policies are most effective at encouraging private savings.A lack of tax system engagement for some groups is a recurring theme in the literature, and this is thought to limit the effectiveness of targeted programs. Eissa and Liebman (1996) cite evidence from interviews with earned income tax credit (EITC) recipients, which revealed that many individuals had heard of the EITC program in the United States but did not understand how it related to their earnings. Chetty et al. (2009) find that the provision of EITC program information to EITC recipients at the right time can induce material labor supply responses. This study provides evidence to support to the hypothesis that people do not fully optimize behavior in response to government policy due to a lack of engagement; challenging what is often a core assumption in the public finance literature. For retirement savings incentives, Feng (2014) also cited a lack of knowledge as a factor that results in low participation. Worthington (2008) suggests that a policy response to increase knowledge could be to provide subsidized, or compulsory, retirement planning advice for people at particular work-life milestones.5Data and Variable ConstructionWe use ATO administrative data and Australian population estimates (Australian Bureau of Statistics, 2020), following the data construction methodology of Carter and Breunig (2019). We augment data from three sources to derive labor supply rates for the age groups of interest within the Australian resident population. This approach is required given that the ATO data alone do not account for the entire population of working and nonworking individuals. Table 1 illustrates how the different data sources are combined to construct labor supply rates.Table 1Data sourcesWorkedDid not workFiledIncome tax return dataIncome tax return dataDid not filePAYG payment summary data (for salary and wage payments only)Residual population calculated from ABS estimatesABS, Australian Bureau of Statistics; PAYG, Pay-As-You-Go.We use the ATO's income tax return (ITR) data to capture the share of the population that filed a tax return. Using these data, we classify those who worked and those who did not according to the specific sources of their income (this approach is discussed in Section 5.1). We use the ATO's salary and wage Pay-As-You-Go (PAYG) payment summary data to account for the small share of individuals who appear to have worked but did not file a return. We account for the remaining nonworking population by “topping up” our sample to match aggregate resident population estimates, published by the Australian Bureau of Statistics (2020). We exclude nonresident tax filers to align the data construction method more closely with the ABS definition. We also exclude a very small share of deceased tax filers who passed away before the beginning of a given lodgment year.Filing can occur beyond death in cases where taxable income continues to accrue until the estate of the deceased is dissolved.We draw on the ATO's superannuation Member Contribution Statement (MCS) and SMSF Annual Return data to observe contributions into superannuation accounts. We include employer contributions to broaden our “working” and “earnings” definitions to capture the small share of individuals who salary sacrifice their entire salary and wage payments into their superannuation account. Workers in this category would not have otherwise been counted by our “working” definitions.Using this approach, we construct a panel data set for a period that spans 16 financial years, from 2000–01 to 2015–16, for people who are aged 53–62 years by the end of a given financial year. The panel captures the 5 years before the TRIS was introduced, along with the first 11 years in which it was available. There are 40.2 million observations over this period, which, by construction, matches the ABS population estimates. The data were extracted from ATO systems on November 1, 2018.5.1Measures of labor supply and earningsThe ATO data do not directly record the working status of people. There are, however, various options to derive measures of labor supply from the data. For instance, it seems reasonable to assume that people who report salary and wage payments worked to earn this income. This simple inference can be broadened to include other “earned income” fields from the tax return. To do so, we start with the ATO definition that was used as a work test to administer the now-discontinued Mature Age Worker Tax Offset (MAWTO). This definition, known as “Net income from working” (NIFW), is presented in Table 2. NIFW is the sum of work-related income minus work-related expenses. For our study, we add “total employer superannuation contributions” (Component 12 in Table 2) as mentioned in the Section 5. This addition, however, does not have a material effect on our results.Table 2“Net income from working Indicator 2”: adjusted Australian Taxation Office definitionNet income from working=Total gross salary and wage payments(1)+Income from allowances, earnings, tips, director's fees, etc.(2)+Attributed personal services income(3)+Total reportable fringe benefits (RFB) amounts (if RFB ≥ RFB threshold)(4)–Work-related car expenses(5)–Work-related travel expenses(6)–Work-related clothing expenses(7)–Work-related self-education expenses(8)–Other work-related expenses(9)–Low-value pool deductiona(10)+Net income from working (Appendix section)b(11)+Total employer superannuation contributions(12)aLow-value pool deductions refer to “low-cost” and “low-value” assets used in the course of generating income. These are assets costing <$1,000, which can be depreciated over multiple tax lodgment years.bNIFW (Appendix section) refers to business and partnership income that is derived from working.NIFW was not recorded in ATO databases for workers below the MAWTO's qualifying age and in the years in which the MAWTO was not in force. Further, it cannot be perfectly recalculated for people with business and partnership income because they were required to complete a supplementary schedule in the years in which the MAWTO was in force. This schedule asked taxpayers to separate their business and partnership income that were derived from work (as opposed to passive income). These earned-income components are shown as Component 11 in Table 2.Given this issue, we recalculate NIFW using all components from Table 2. The intention is to provide a variable for our analysis that is consistent over time and for all age groups of interest. This approach slightly overstates Component 11 income given that we cannot separate the share that is attributed to work. It does not, however, have a material effect on labor supply rates for people in this group, given that most report work-related earnings from at least one other source.We examine three labor supply measures to provide insight into the sensitivity of the definitions of employment and corresponding earnings on our results:NIFW Indicator 2: This measure recalculates ATO's “NIFW indicator” measure by summing the tax return components shown in Table 2.NIFW Indicator 3: This measure removes Component 11 in Table 2 from the definition. This change reduces employment rates by around 12% for males and 7% for females.Salary and wage (SW) indicator: This is a simple measure that further underestimates the true labor supply rates. Salary and wage payments (Component 1) are, by far, the most commonly reported earning component in Table 2. This measure reduces employment rates by 14% for males and 8% for females, relative to NIFW Indicator 2.The differences in derived labor supply rates are shown graphically in Figure 1 for males and females aged 56 years by the end of a given financial year. For this analysis, we prefer NIFW Indicator 2 given that it captures a broader group of people with business and partnership income who are more likely to materially benefit from the TRIS program.Figure 1Derived employment rates for individuals aged 56 years.NIFW, net income from working; SW, salary and wage.In contrast to survey data, the ATO's administrative data are better suited to this study given that we wish to examine the specific effects of TRIS, which is administered through the tax and superannuation systems. The administrative data provide a richer source of relevant income components and flows to and from superannuation accounts, including data on those who receive tax offsets on qualifying superannuation income streams. The administrative data also enable more precise estimates, relative to survey data sets, given the additional statistical power obtained from larger samples.There are however drawbacks. The administrative data have fewer variables to control for individual characteristics. We also cannot ascertain when an individual worked within a given year, or for how long, given that a measure for hours worked is not available. Thus, we can only analyze extensive margin effects of the program. We do present results on the “earned income” effects, which could be considered an imperfect proxy for intensive margin effects.Another data limitation is that we are unable to observe spouses. This means that we are not able to gain insight into the joint retirement decisions of couples, or spouses, in response to the program. This would be an interesting question to revisit if the ATO were to make a household-level data product available.5.2Identifying TRIS recipientsTRIS recipients cannot be directly identified in the ATO data. While TRIS amounts are taxable income for all individuals in the program's initial years, there is no way to distinguish TRIS from other taxable superannuation income streams. This is with the exception of individuals who draw TRIS from SMSFs from 2008–09. From this year, there was a change to the SMSF income tax return form, which required direct identification of TRIS drawdowns. Fortunately, identifying TRIS recipients is not critical for the main analysis on the employment and earnings effects. We do, however, attempt to identify people with “TRIS-like-behavior” by deriving rules. These rules identify people as TRIS recipients if they worked (according to our three NIFW definitions) while drawing income from their superannuation in the same year. For taxpayers who meet this rule, we include an additional condition to ensure that recipients are receiving a tax offset of 15% of the superannuation income stream.We actually allow for a tax offset range between 14% and 16% to account for rounding effects, but we find that the results are not sensitive to this allowance. This additional condition filters out individuals who receive income streams from untaxed superannuation funds (generally government-defined benefits schemes), which attract a 10% offset. Further, given that the data are recorded on an annual basis, we remove individuals who observe “TRIS-like behavior” in a single year only. It is likely that the majority of workers in this category are ordinary retirees. That is, they simply worked part of the year before retiring and drawing a superannuation income stream in the same financial year. In Figure 2, we show the share of the population who exhibit “TRIS-like-behavior”, as identified by the NIFW Indicator 2 measure, for males and females. These charts reveal that the adoption of TRIS appears relatively subdued in the first 2 years in which the program was available, before increasing sharply and plateauing toward the end of the period.Figure 2“TRIS-like-behavior”, using NIFW Indicator 2 TRIS rule.NIFW, net income from working; TRIS, Transition to Retirement Income Streams.There are three points we wish to emphasize. First, the figures show that there is a relatively stable cohort of individuals who are classified as TRIS recipients before the policy was introduced (pre-2005–06). It is likely that these people are already receiving an annuity stream before TRIS is introduced. Unfortunately, our rule fails to separate this cohort from actual TRIS recipients. Second, there is also a very small and stable share of individuals who are aged 54 years, below the superannuation preservation age, who fit our “TRIS-like-behavior” definition. This cohort could be individuals with rare and exceptional circumstances who are receiving superannuation income streams before reaching the superannuation preservation age. Third, our rule fails to capture recipients aged >60 years from 2007–08. This is because TRIS incomes were no longer reported on the ITR given the tax-free status of this income.Data are available from 2007–08 for those drawing TRIS from SMSFs following a change to the SMSF annual return, which introduced a field to identify TRIS directly. Previously, TRIS incomes were pooled with other reportable superannuation income streams. The benefit of the SMSF data is that they provide insight into program adoption details for individuals aged ≥60 years. Taking these data at face value (i.e., not using our “TRIS-like-behavior” rule), we plot the share of the population who reported TRIS by age and income year in Figure 3. In this figure, we have pooled males and females with age on the x-axis to show the increased uptake for people in the age range of 60–64 years (who benefit from tax-free TRIS). The figure reveals three points of interest. First, we see that program adoption continues to increase with time and age. Second, we observe that the rate of change in uptake for people aged 60 years becomes more acute in the latter years. It therefore appears to imply a “learning effect” as program adoption matures. Third, we can also see the expected drop-off in TRIS participation from the age of 65 years given that this group is free to access their superannuation irrespective of their working status.Figure 3Reported SMSF TRIS, males and females pooled.SMSF, self-managed superannuation funds; TRIS, Transition to Retirement Income Streams.In Section A.3 of Appendix, we use APRA data as an independent cross-check for our derived “TRIS-like-behavior” rules. Despite some technical issues between the two data sources, we find that our method produces TRIS recipient numbers that are broadly comparable to the APRA data. The APRA statistics additionally show the strength in TRIS adoption for individuals aged >60 years, which we cannot observe in the ATO data for non-SMSFs. Adoption for the 60–64 year age grouping almost doubles relative to the 55–59 year age grouping for individuals with APRA regulated (non-SMSF) accounts.6Identification StrategyWe use a difference-in-differences (D-i-D) design to detect the labor supply and corresponding earnings response to the TRIS. This is an “intention-to-treat” effect given the voluntary nature of program participation. We compare the difference in labor supply rates of 54- (control) and 56-year-olds (treatment) in the TRIS's first year (2005–06) with the same difference in the year before it was introduced (2004–05). We do not focus on the difference in 55-year-olds given that there is a partial-treatment effect for this group. This arises due to the annual frequency of the data, meaning that only around half of the people who are aged 55 years by the end of a given financial year are eligible for TRIS.The first income year in which the TRIS was available, along with the minimum qualifying age, provides crisp boundaries to assign as “control” and “treatment” groups before and after TRIS's introduction. We compare individuals who are close in age as they are likely to be similar in other ways. We subsequently repeat this analysis for the corresponding measures of earnings. We estimate the effects for males and females separately, given that the labor supply rates differ by sex. Pooled results were estimated but are not presented given that they show a weighted average of the D-i-D coefficients by sex.Noting that the TRIS adoption profile appeared relatively flat in the first year, we then examine the effect on the D-i-D estimates of skipping the first year in which TRIS options were available. Under this estimation, the control group remains the same; however, 2006–07 is assigned as the treatment year. The intention is to see whether an increased effect is detected in the second year. We then again repeat this approach for the program's third year. These D-i-D estimates are discussed in Section “Empirical results”.Once we account for the different labor supply rates by age and sex, along with wage inflation between periods, our control and treatment groups appear similar in other ways. The summary statistics are presented in Tables A3 and A4 in Appendix. These tables include statistics on two “treatment” groups that we examine separately (2005–06 and 2006–07).We repeat this approach to estimate the additional “tax-free effect” for workers aged ≥60 years from 2007–08 (treatment), by comparing the labor supply rates in 2006–07 (control). In this case, we compare labor supply rates of individuals aged 59 years (control group) and with individuals aged 61 years (treatment group). As with the previous estimates, we skip reporting results for individuals aged 60 years given the partial-treatment effect.A key identifying assumption of D-i-D is that common trends hold in periods where there is no “treatment”. This assumption is required in order to isolate the treatment effect of TRIS with confidence. We graphically present these trends and discuss the factors that challenge this assumption in Section “Robustness checks”.The D-i-D design only attempts to detect the local treatment effects for the two “treatment” age groups mentioned above, namely, 56- and 61-year-old individuals. The D-i-D design does not attempt to identify effects for other age groups who could have different responses.A potential issue for the causal identification of program effects is the degree to which “anticipation effects” exist. For example, it could be the case that an individual intended to retire at 53 years, but who, in response to the program, continues to work until 57 years to utilize the benefits of the TRIS program in the latter 2 years before retirement. Similarly, a non-TRIS participant might have otherwise intended to retire at the age of 58 years, yet, in response to the program, continues working until 62 years of age to draw a tax-free TRIS for the latter 2 years. In these examples, “anticipation effects” are a problem to the degree to which the TRIS program is inducing a labor supply response for those in our “control group”. Unfortunately, not much can be done about this issue. We cannot observe individuals in this group nor pick up the effect. Intuitively, we think this effect is probably negligible. To the degree that “anticipation effects” exist, they would indicate that our main results should be interpreted as lower-bound estimates on the true program effects.7Model SpecificationWe estimate linear probability models to detect the effect of the TRIS on labor supply (see Eq. (9)). A binary dependent variable participationit takes the value of “1” if an individual is working in a given year. We define working as a “nonzero” amount of earned income according to our definitions (the “earned income” components were previously shown in Table 2). The explanatory variables are also binary. The first takes a value of “1” if the TRIS qualifying age was met (Di = 1 if aged 56 years). The second indicates whether TRIS was available in a given year (Tt = 1 if TRIS was available). Finally, the D-i-D estimator is the estimate of β3 from the interaction of Tt = 1 and Di = 1.(8)participationit=β0+β1Di+β2Tt+β3(Di⋅Tt)+ɛit.participatio{n_{it}} = {\beta _0} + {\beta _1}{D_i} + {\beta _2}{T_t} + {\beta _3}\left( {{D_i} \cdot {T_t}} \right) + {\varepsilon _{it}}.No additional controls are available for the labor supply estimation. This is because our data construction approach utilizes official population estimates to account for the assumed nonworking population that is not observed in the administrative data. Hence, no further information is available on this group other than their sex.To examine the effect on earnings, we substitute the binary dependent variable from Eq. (9) with the corresponding log dollar value of “earned income” (Eq. (10)). This is the sum of some or all of the income components in Table 2, depending on the measure. This estimation is conditional on working, so we exclude the nonworking individuals from the data, as defined by the NIFW measures. Noting that there is a small share of individuals who report negative business income (a loss), we convert their negative income values to $0.01 and assign an additional dummy variable as a control for this group.(9)ln(incomeit)=β0+β1Di+β2Tt+β3(Di⋅Tt)+β4 negativeit+ɛit.{ln}\left( {incom{e_{it}}} \right) = {\beta _0} + {\beta _1}{D_i} + {\beta _2}{T_t} + {\beta _3}\left( {{D_i} \cdot {T_t}} \right) + {\beta _4}\,negativ{e_{it}}\, + {\varepsilon _{it}}.8Empirical ResultsTable 3 presents the average employment rates for the control and treatment groups in the 12-month periods before and after the introduction of the TRIS. The employment rates in Columns 1 and 2 show that a higher share of people aged 54 years (control group) work, compared with people aged 56 years (treatment group). This is consistent with observed labor supply rates, which peak at around age 50 before beginning to trend down as workers gradually retire with age. The difference in employment rates between the two periods is shown in Column 3, and the D-i-D estimate is shown in Column 4. Labor supply rates range from 57.7% for females aged 56 years under the salary and wage measure, to 84.3% for males aged 54 years under the more comprehensive NIFW Indicator 2 measure. The D-i-D estimates for the treatment group (individuals aged 56 years) are presented in Column 4. These estimates imply that TRIS did not have an effect that was significantly different from zero in the first year in which it was available. These results appear consistent with the TRIS adoption profile (shown in Figure 2).Table 3Labor supply rates in periods before and after the introduction of TRISPre-TRIS 2004–05(1)TRIS 2005–06(2)Difference (2 – 1)(3)Difference-in-difference(4)MalesANIFW Indicator 2Treatment (age: 56 years) [249,728]0.7977 (0.0011)0.8080 (0.0011)0.0102*** (0.0016)Control (age: 54 years) [256,949]0.8317 (0.001)0.8432 (0.001)0.0115*** (0.0015)−0.0013 (0.0022)BNIFW Indicator 3Treatment (age: 56 years) [249,728]0.6729 (0.0013)0.6867 (0.0013)0.0138*** (0.0019)Control (age: 54 years) [256,949]0.7111 (0.0013)0.7234 (0.0012)0.0124*** (0.0018)0.0014 (0.0026)CSalary and wage indicatorTreatment (age: 56 years) [249,728]0.6557 (0.0014)0.6711 (0.0013)0.0154*** (0.0019)Control (age: 54 years) [256,949]0.6962 (0.0013)0.7089 (0.0013)0.0127*** (0.0018)0.0027 (0.0026)FemalesDNIFW Indicator 2Treatment (age: 56 years) [250,487]0.6603 (0.0013)0.6778 (0.0013)0.0175*** (0.0019)Control (age: 54 years) [258,953]0.7168 (0.0013)0.7362 (0.0012)0.0194*** (0.0018)−0.0019 (0.0026)ENIFW Indicator 3Treatment (age: 56 years) [250,487]0.5873 (0.0014)0.6060 (0.0014)0.0186*** (0.002)Control (age: 54 years) [258,953]0.6476 (0.0013)0.6687 (0.0013)0.0212*** (0.0019)−0.0025 (0.0027)FSalary and wage indicatorTreatment (age: 56 years) [250,487]0.5765 (0.0014)0.5959 (0.0014)0.0194*** (0.002)Control (age: 54 years) [258,953]0.6387 (0.0013)0.6597 (0.0013)0.0211*** (0.0019)−0.0017 (0.0027)Notes: Labor supply equals one if the specific NIFW measure has a dollar value that does not equal zero. Sample size is presented in square brackets, and robust standard errors are in parentheses.***, **, and * denote statistical significance at the 0.01, 0.05, and 0.1 levels, respectively.NIFW, net income from working; TRIS, Transition to Retirement Income Streams.Noting the delayed adoption of the TRIS program, we test the effect of skipping the first year (2005–06) to see whether we pick up a response if 2006–07 is assigned as the treatment year. In Table 4, we present two variations of the D-i-D coefficients. Column 1 shows the D-i-D estimators from the previous table (Table 3) and Column 2 shows the D-i-D estimators that pick up the second year's response. The results for males now become slightly positive and significant. The response ranges from half a percentage point for NIFW Indicator 2, to 1.0 percentage point for NIFW Indicator 3 and the SW indicator. The employment response for females remains statistically insignificant.Table 4TRIS effects on “labor supply” and “earnings” (D-i-D coefficients)2004–05 vs 2005–06(1)2004–05 vs 2006–07(2)ITR, PAYG, and ABS estimates54 years vs 56 years54 years vs 56 yearsLabor supplyMalesNIFW 2 D-i-D (TRIS · Age)−0.0013[0.0022]0.0052**[0.0021]NIFW 3 D-i-D (TRIS · Age)0.0014[0.0026]0.0102***[0.0026]Salary and wage D-i-D (TRIS · Age)0.0027[0.0026]0.0101***[0.0026]FemalesNIFW 2 D-i-D (TRIS · Age)−0.0019[0.0026]−0.0008[0.0025]NIFW 3 D-i-D (TRIS · Age)−0.0025[0.0027]0.0001[0.0027]Salary and wage D-i-D (TRIS · Age)−0.0017[0.0027]0.0003[0.0027]ITR and PAYG data54 years vs 56 years54 years vs 56 yearsEarned incomeMalesNIFW 2 D-i-D (TRIS · Age)0.0173**[0.0073]0.0244***[0.0073]NIFW 3 D-i-D (TRIS · Age)0.0018[0.0079]0.0161**[0.0079]Salary and wage D-i-D (TRIS · Age)0.0021[0.0080]0.0179**[0.0080]FemalesNIFW 2 D-i-D (TRIS · Age)−0.0080[0.0083]0.0012[0.0082]NIFW 3 D-i-D (TRIS · Age)−0.0175**[0.0084]−0.0099[0.0082]Salary and wage D-i-D (TRIS · Age)−0.0190**[0.0084]−0.0084[0.0083]2006–07 vs 2007–082006–07 vs 2008–09ITR, PAYG, and ABS estimates59 years vs 61 years59 years vs 61 yearsLabor supplyMalesNIFW 2 D-i-D (TRIS · Age)0.0247***[0.0026]0.0110***[0.0026]NIFW 3 D-i-D (TRIS · Age)0.0253***[0.0028]0.0136***[0.0028]Salary and wage D-i-D (TRIS · Age)0.0238***[0.0028]0.0115***[0.0029]FemalesNIFW 2 D-i-D (TRIS · Age)0.0134***[0.0029]0.0058**[0.0029]NIFW 3 D-i-D (TRIS · Age)0.0113***[0.0029]0.0058**[0.0029]Salary and wage D-i-D (TRIS · Age)0.0113***[0.0029]0.0050*[0.0029]ITR and PAYG data59 years vs 61 years59 years vs 61 yearsEarned incomeMalesNIFW 2 D-i-D (TRIS · Age)0.0024[0.0088]0.0020[0.0092]NIFW 3 D-i-D (TRIS · Age)−0.0149[0.0093]−0.0133[0.0097]Salary and wage D-i-D (TRIS · Age)−0.0166[0.0102]−0.0110[0.0099]FemalesNIFW 2 D-i-D (TRIS · Age)0.0074[0.0112]0.0137[0.0111]NIFW 3 D-i-D (TRIS · Age)−0.0229**[0.0113]−0.0197*[0.0111]Salary and wage D-i-D (TRIS · Age)−0.0263**[0.0114]−0.0217*[0.0111]Notes: Robust standard errors are presented in brackets.***, **, and * denote statistical significance at the 0.01, 0.05, and 0.1 levels, respectively.ABS, Australian Bureau of Statistics; D-i-D, difference-in-differences; ITR, income tax return; NIFW, net income from working; PAYG, Pay-As-You-Go; TRIS, Transition to Retirement Income Streams.We also examine the effect of skipping the second year by assigning 2007–08 as the treatment year given that TRIS adoption appeared to increase substantially from this year. These results now show stronger labor supply responses of up to 1.4 percentage points for males and up to 1.1 percentage points for females. As with the previous estimates, the response appears consistent with the observed trends in TRIS adoption. We note, however, that the identification is weakened under this estimation given that the reassigned treatment year coincides with the “Simplified Superannuation” reforms package, which may have influenced the employment rates for our control and treatment groups differently, in addition to moving further away from our control year (2004–05). Nevertheless, the larger response in 2007–08 seems consistent with the TRIS adoption profile.The lower half of Table 4 shows the responses for individuals aged >60 years from 2006–07. We find statistically significant results for both men and women under both scenarios irrespective of whether we skip the first year in which TRIS amounts became a tax-free income (2007–08). The response for men is as high as 2.5 percentage points in 2007–08 and up to 1.4 percentage points in 2008–09 when this year is assigned as the treatment year. The response for women is around half the response for men under all measures.Table 4 also reports the results for the corresponding “earned income” (conditional on working) measures. These results are mixed, which probably reflects the balancing effect of the different TRIS strategies that workers can use. The D-i-D estimators for people aged 56 years range from “not significantly different from zero” to 2.4% for NIFW Indicator 2 when 2006–07 is assigned as the treatment year. The inverse is true for females, wherein we find that the results range from insignificant to negative 1.9% when 2005–06 is assigned as the treatment year. Taken at face value, this may imply that males are more likely to use TRIS to boost income in the current year (e.g., continue working full time while drawing TRIS), while females reduced their attachment to the labor market; however, TRIS is not quite topping up their incomes to prior levels. Unfortunately, it is not possible to disentangle these effects, given that a measure for time spent working is not available in the data.The lower half of Table 4 shows the earned income D-i-D estimates for individuals aged 61 years in 2007–08 and 2008–09 (skipping the first year). Here, we find no effect for males and, consistent with the previous estimates, slightly negative results for females. Column 1 shows that the earnings effect for women is not statistically different from zero for NIFW Indicator 2, while both NIFW Indicator 3 and the salary and wage indicator are negative and statistically significant. The estimates are similar but slightly stronger when 2008–09 is assigned as the treatment year.8.1TRIS adoption – descriptive analysisWe regress the binary variable “TRIS-like-behavior” against a series of explanatory variables to gain insight into TRIS adoption over time. The intention is to undertake some descriptive analysis to provide context for the main results. The underlying population for this analysis includes all tax filers in the 54–59 year age range for the financial years 2004–05 to 2015–16. Unfortunately, individuals aged ≥60 years are excluded given that our “TRIS-like-behavior” rule fails for this age group. The following linear probability model is estimated separately for males and females:(10)trisit=β0+β1D2006t+β2D2007t+β3D2008t+…+β11D2016t+ɛit.tri{s_{it}} = {\beta _0} + {\beta _1}D{2006_t} + {\beta _2}D{2007_t} + {\beta _3}D{2008_t} + \ldots + {\beta _{11}}D{2016_t} + {\varepsilon _{it}}.In Eq. (11), trisit is an indicator if the individual was drawing a TRIS in a given financial year. This variable is equivalent to the series plotted in Figure 2. The right-hand side includes financial year dummies to reveal how the adoption of TRIS has changed over time. Robust standard errors in this estimation are clustered at the individual. The coefficients show the TRIS adoption rates relative to the year before the policy was introduced (2004–05), the benchmark year where no dummy was assigned. We then extend this basic model to include controls for age as at June 30 in a given year, before further extending the estimation a third time for additional controls that are available. These controls include the log of taxable income,Negative taxable income (a loss) is possible for a small share of individuals with business income. In this case, we convert negative taxable income to 0.01 and include a dummy variable to control for individuals with negative taxable income. and the following binary controls: self-prepared tax return (as opposed to using a tax agent); whether the individual reported a partner; indicators of geographical remoteness; and taxpayers who report any income in the following categories: Personal Services Income (PSI); business and partnership income; dividend income; and rental income. The remoteness indicators were created by mapping Australian residential postcodes to the Australian Statistical Geographic Standard (Australian Bureau of Statistics, 2011). The regression outputs for the three specifications are presented in Tables A5 and A6 in Appendix for males and females, respectively.The results show that the year and age coefficients (in Columns (1), (2), and (3)) remain stable and, as expected, confirm that adoption of TRIS was slow in the initial years before picking up noticeably from 2007–08. Adoption rates continue to increase over the course of the remaining years. A similar trend is observed with the age variable in Column (2), whereby program adoption increases as individuals age. We note that the “partial-treatment” effect, previously mentioned, is captured for 55 year olds given the lower response.The regression coefficients show that males and females exhibit relatively similar adoption patterns across time. A difference, however, is that a higher share of males appear to use TRIS relative to females. Perhaps unsurprisingly, adoption of TRIS is positively correlated with income and for those with diverse sources of income (e.g., business, dividend, and rental income). Having professional services income (PSI) has a slightly negative effect for males and is not significant for females. The effect of a reported partner is slightly positive for males. The effect for those who self-prepare tax returns is not statistically significant for males and is slightly negative for females. TRIS use is less likely for individuals who live in more remote areas of Australia.9Discussion and Robustness ChecksWe examine the sensitivity of the results by extending the control and treatment age groupings to 2 years either side of the introduction of TRIS. Table A7 in Appendix shows that the effects are stronger for both males and females relative to the single-year age group estimates. Here, we compare groups aged 53–54 years (control group) with individuals aged 56–57 years (treatment group). The labor supply D-i-D estimators are all statistically significant, with effects ranging from around 0.6% for females in 2005–06 (TRIS's first year) to almost 1.9% for the salary and wage indicator in 2006–07 (TRIS's second year) for males. It appears that the double-age groupings, in this instance – the addition of people aged 57 years, are picking up the effect that older individuals are more likely to use TRIS. This effect is also observed in the TRIS adoption profiles (Figure 2) and the supplementary regression results (Tables A5 and A6 in Appendix). Given that program adoption is still maturing in this period, it supports our argument that the main results should be considered lower-bound estimates.Similarly, we examine the effects of the extended age grouping for the group aged >60 years. With this check, we compare the groups aged 58–59 years (control group) with individuals aged 61–62 years (treatment group). The D-i-D estimators in 2007–08, shown in Table A7 in Appendix, are still positive and statistically significant for males but are much weaker compared with the single-year age estimates. The D-i-D estimates for females are also weaker. They move from being positive and statistically significant in the single-year specification to being not significantly different from zero under the extended-age grouping specification. Under the second specification that skips the first tax-free year (now assigning 2007–08 as the control), the estimates remain similar to the single-year age groupings; however, the effects are slightly stronger for both males and females.It is common practice for studies that use D-i-D approaches to undertake placebo tests in periods where there is no treatment. These tests aim to demonstrate that the D-i-D coefficients return effects that are not statistically different from zero in periods where there was no treatment. For this check, we reassign the control and treatment periods for each of the neighboring 2 years in our panel. For example, we compare 2001 (assigned as control) with 2002 (assigned as treatment), then separately for 2002 (assigned as control) with 2003 (assigned as treatment), and repeat this analysis for all subsequent years.We estimate these placebo checks for males and females separately. The coefficients are presented graphically in Figures 4 and 5. The figures confirm that there are modest nonzero movements in other years. As Carter and Breunig (2019) found when they examined the responses to Australia's MAWTO, there are other known factors influencing labor supply rates in other years. These checks are therefore not true placebo tests given other fiscal, macroeconomic, and demographic changes. The large apparent “treatment” effects detected in 2002–03 and 2003–04 show the unusual effects of demographic patterns that are well documented. The instability in the early period up to 2003–04 are accounted for by individuals who were born in 1947, which was the peak year of the post-World War II baby boom, as documented by the Australian Bureau of Statistics (2004). This period, with its unusual demographic patterns, highlights a key problem with using D-i-D when common trends fail to hold. Fortunately, TRIS was implemented after this period at a time where the trends appear to have reverted to a more parallel state.Figure 4Rolling D-i-D estimates – 54 years vs 56 years of age, males.D-i-D, difference-in-differences; NIFW, net income from working; SW, salary and wage.Figure 5Rolling D-i-D estimates – 54 years vs 56 years of age, females.D-i-D, difference-in-differences; NIFW, net income from working; SW, salary and wage.The MAWTO, an EITC for people older than the age of 55 years, was introduced in 2004–05 and was found to have small positive labor supply effects. TRIS was introduced in the following year, which is a period highlighted by the dotted lines in the figures, along with the subsequent effects that are in line with the increase in the TRIS adoption profile. Problems isolating specific effects arise from 2007–08 given that this period coincided with the “Simplified Superannuation” reforms and the subsequent onset of the global financial crisis. The global financial crisis, in particular, may have delayed retirement decisions given the value of retirement investment holdings were significantly reduced.The trends in employment rates for individuals aged 54 years (control group) and individuals aged 56 years (treatment group) are shown in Figure 6. This provides an alternative way to view the rolling D-i-D estimates presented in Figures 4 and 5. Figure 6 shows the obvious labor supply spikes in 2000–01 and 2002–03, which are explained by unusual demographic patterns. The two highlighted areas in the figure show the key TRIS periods of interest for this paper.Figure 6Trends in employment rates (NIFW Indicator 2).NIFW, net income from working.These checks provide important context for the main results. Above all, care should be taking in interpreting the D-i-D estimates as we cannot demonstrate null effects in periods where there is no TRIS treatment. Further, our placebo-like tests lend support to the hypothesis that the TRIS program may have had more of an effect on employment rates in subsequent years as program adoption rates increased.It appears that <10% of workers in the 56–59 years age grouping are using TRIS by 2014–15 (according to our NIFW Indicator 2 measure). This seems low relative to the share of eligible workers who would likely benefit from the program. There are several reasons that could explain low adoption rates. First, low adoption in the initial years could be due to a “learning effect”. It may have taken time for knowledge of the program to spread and for workers to understand the benefits. Second, it is a voluntary program, which means that workers need to set up TRIS pension with their superannuation fund. Third, it is likely that a proportion of workers remain unaware of the provisions. Fourth, individuals on lower incomes and/or with lower superannuation account balances have less to gain from TRIS. This is because the maximum 10% annual account balance drawdown may be too low to make entering into the arrangement worthwhile. In addition, workers with lower income may have little, or perhaps nothing, to gain from a tax savings perspective if their income is already subject to lower marginal income tax rates.For those individuals who meet the “TRIS-like-behavior” rule, we attempt to identify the subset that appears to be using a tax-effective strategy. Recall that we define a tax-effective strategy as a worker who draws a TRIS pension while, within the same year, makes voluntary salary-sacrificed contributions back into their superannuation fund.We use the “reportable superannuation contributions” field from the individual's income tax return form, which isolates salary-sacrificed superannuation contributions from employer Superannuation Guarantee contributions. Separating out reportable superannuation contributions prior to 2009–10 is not possible as the amounts are pooled into a single field in the data. From 2009–10 to 2014–15, the final year in which we can derive our “TRIS-like-behavior” rule in our data, it appears that at least half of the workers in the 56–59 year age grouping are using tax-effective strategies. Unfortunately, we cannot say anything for individuals >60 years of age.The introduction of concessional contributions caps (see Figure A2 in Appendix) lessened opportunities for higher income earners to use TRIS to minimize tax if their concessional contributions were already at, or above, the cap. For those with “TRIS-like-behavior” in 2014–15, it appears that around 1% had already exhausted their concessional contributions limit and were therefore not able to use a tax-effective strategy. To provide some context, in 2014–15, the cap was $35,000, which implies, for workers receiving the Superannuation Guarantee at 9.5%, a corresponding annual salary of around $370,000 or more.We note that it is common for certain occupations to receive more than the Superannuation Guarantee rate. For example, Australian public servants usually receive 15.4% of ordinary time earnings and academia is known to offer even higher rates. This cap was further reduced to $25,000 from 2017–18.10ConclusionThe TRIS program is a novel approach to incentivize the labor supply of older workers. The program allows early access to their concessionally taxed retirement savings. This paper considers the effect of the program in the context of a life cycle model and empirically examines the effects on labor supply and earnings. Consistent with the literature, and despite generous financial incentives, program adoption is low and is concentrated toward higher earners for whom the tax benefits are larger. TRIS had no detectable effect on labor supply in its first year, small positive effects of around 1.0% for males only in its second year, and larger effects in the program's third year of up to 1.4% for both males and females. For people aged >60 years, we find stronger responses of up to 2.5% from 2007–08, the period when superannuation income streams became tax free for this group.The program's effects on earnings are ambiguous. This is because individuals can adopt different TRIS strategies depending on their income and savings goals. Despite this, we report that the TRIS had nil or very small positive earnings effects for males, and nil or slightly negative effects for females. Taken at face value, these small average effects might imply that males are more likely to use TRIS as a way to boost incomes while continuing to work full time, and females are more likely to use the policy as it is intended to support their transition to retirement. This is, however, difficult to ascertain given the lack in the data of a measure for time spent working.We highlight that the delayed program adoption weakens the clean identification of program effects, along with modest deviations to the common trends assumption. It appears that it took time for knowledge of the program to spread and, hence, for program adoption to mature. This delay is consistent with the magnitude and timing of the estimated labor supply effects, which are nil or slightly positive in the initial years and increase in subsequent years. A related issue is that our ability to accurately detect the program effects becomes less reliable the further we move away from the program's initial year. This means that we cannot estimate the causal effects for all age groups as program adoption matured. For this reason, we argue that our results should be viewed as a lower bound on the true program effects.Higher earners have more to gain from TRIS, and our analysis confirms that this group is more likely to participate in the program. The lack of a work test has provided the opportunity to use TRIS as a tool to minimize tax. It appears that at least half of the TRIS recipients are using tax-effective strategies. Hanegbi (2013) suggests that the program could be better targeted by requiring taxpayer self-assessment against specific criteria to ensure that the program is used as intended. The situation of workers entering into TRIS arrangements with no intention of reducing hours or workplace responsibilities seems at odds with the policy's intent – and particularly for those employing tax-effective strategies.
Journal
IZA Journal of Labor Policy – de Gruyter
Published: Jan 1, 2022
Keywords: labor supply; impact evaluation; mature age workers; retirement savings; transition to retirement; J19; J26; H2; I38