Taxation of Pensions in a Country-Calibrated OLG Model: The Case of Australia

Transcription

1 Taxation of Pensions in a Country-Calibrated OLG Model: The Case of Australia George Kudrna

2 Taxation of Pensions in a Country-Calibrated OLG Model: The Case of Australia George Kudrna y September 2015 Abstract This paper provides a quantitative analysis of the most recent blueprint for reform of the taxation of superannuation (Australia s term for private pensions) as recommended in the Australia s Future Tax System [AFTS] (2010). The AFTS proposed a switch from the current comprehensive income tax approach to a prepaid expenditure tax approach with a at rate of contribution rebate. Using an overlapping-generations (OLG) model stylised to the Australian economy, we nd that the proposed reform improves vertical equity and reduces income inequality, as indicated by larger welfare gains and income improvements experienced by lower income households and a reduced Gini coe cient. The simulation results also show positive e ects on total assets and reduced public pension expenditures, but the reform yields a small e ciency loss. Keywords: Compulsory saving; pension and tax reforms; dynamic OLG model JEL Classi cation: H55; E21; C68 This research was conducted by the Australian Research Council Centre of Excellence in Population Ageing Research (project number CE ). The views expressed herein are those of the author and are not necessarily those of the Australian Research Council. y Mail address: George Kudrna, Centre of Excellence in Population Ageing Research (CEPAR), Australian School of Business, UNSW, Sydney 2052 Australia; g.kudrna@unsw.edu.au; Telephone:

3 1 Introduction Private pensions around the world receive concessional tax treatments that aim to increase private retirement incomes and household savings. As shown in Table 1, most countries tax private pensions under the expenditure tax "Exempt-Exempt-Taxed" (EET) regime where contributions and fund income are exempt from any taxation but bene ts are treated as ordinary income and taxed progressively. An alternative approach is the pre-paid expenditure tax "Taxed-Exempt-Exempt" (TEE) regime, which allows no deductions of contributions from gross income but then applies no further tax. By contrast, Australia s superannuation (Australia s term for private pensions) applies a comprehensive income tax (T*T*E) approach, which sees at tax rates on contributions and fund income, with bene ts being generally tax free. Although the system is concessional for most income earners in comparison with the personal income taxation, the distribution of tax concessions is inequitable and skewed to higher income earners. 1 According to Financial System Inquiry (2014), almost 60 percent of superannuation tax concessions are paid to the top quintile of income earners, whereas households in the bottom quintile receive little or no income tax bene t from superannuation. Furthermore, Treasury (2012) shows that the largest recipients of total government support for retirement incomes (including Australia s age pension) are in the top one percent of income earners. Table 1: Taxation of private pensions in selected countries EET TTE TEE ETT EET* [a] EET T*TE T*T*E France Canada Hungary Denmark New Zealand Australia Germany Finland Luxembourg Italy Ireland Greece United States Sweden Japan Iceland (Roth IRA) Korea Netherlands Slovakia Norway Spain Poland Turkey Switzerland United Kingdom United States Belgium Austria Portugal Brazil Chile Notes: E=exempt; T=taxed under personal income tax; T*=concessional tax or partial exemption; [a] Partial exemptions apply mainly to lump sums, with income streams often taxed as ordinary income. Source : Yoo and de Serres (2004), OECD (2011). 1 The statutory rates of the contribution and fund earnings taxes are 15 percent, while the Australian personal income tax system is progressive with ve tax brackets, currently ranging from 0% to the top marginal tax rate of 49% (including the Medicare levy and the Temporary Budget Repair Levy). 2

4 This paper provides a quantitative analysis of the most recent blueprint for reform of superannuation taxation as recommended in the Australia s Future Tax System [AFTS] (2010) - that is a switch from the current comprehensive income tax (T*T*E) approach to a pre-paid expenditure tax (TEE) approach with a at rate of contribution rebate. Under the reform, the existing at tax on contributions that is currently paid by superannuation funds is abolished, and the contributions are treated as ordinary income and taxed in the hands of households progressively in their marginal tax rates. Rather than the complete removal of the fund earnings tax, the AFTS proposed a 50 percent reduction in its statutory rate that would also apply to the (currently tax free) earnings from assets supporting superannuation income streams. We also examine the e ects of feasible alternatives to the AFTS proposed reform, including a shift to a standard TEE regime without and with the contribution rebate. The theoretical basis for the analysis of a superannuation tax policy change is provided by Creedy and Guest (2008a), while Creedy and Guest (2008b) employ a computable overlapping generations (OLG) model to investigate the e ects of di erent tax treatments of Australia s superannuation that is motivated by the removal of the taxation of superannuation bene ts in July There is also a large body of international literature that uses OLG models to examine the economic e ects of voluntary tax-preferred retirement saving accounts (see, for example, Imrohoroglu et al. (1998), Fehr et al. (2008) and Nishiyama, 2011). In general, these studies nd positive e ects on national wealth, capital stock and long run welfare, although these e ects vary greatly due mainly to di erent assumptions about government budget balancing policy instruments. Note that Fehr et al. (2008) assess the e ects of these accounts taxed under both the post-paid expenditure EET regime (front-loaded accounts) and pre-paid expenditure TEE regime (back-loaded accounts), showing signi cantly greater wealth accumulation with front-loaded accounts. The quantitative analysis undertaken in this paper also builds on computable OLG models. We use an extended version of the general equilibrium OLG model in Kudrna and Woodland (2011, 2013) that incorporates intra-generational heterogeneity among households disaggregated into income quintiles based on the ABS (2007) data and that allows for gradual withdrawals of superannuation savings. Note that while Creedy and Guest (2008b) examine the e ects of eliminating the concessional tax rates on either contributions or fund earnings or bene ts, we investigate the systemic reform to su- 3

5 perannuation taxation with the contributions taxed progressively as ordinary income. Moreover, our model embodies a detailed model-equivalent representation of Australia s superannuation, income taxation and age pension policy settings to allow for interactions between household behaviour and these policy settings, which are essential for measuring the impacts of such reform. Further note that in contrast to the analysis of introducing voluntary tax-preferred retirement accounts by Fehr et al. (2008) and others, we analyse the tax reforms of already-established superannuation. Our simulation results indicate that the AFTS proposed reform to superannuation taxation improves vertical equity and reduces income inequality upon impact, as well as in the long run, as shown by greater relative gains in welfare and net income shares for lower income individuals and a reduced Gini coe cient. Speci cally, the Gini coe cient measured in net income falls by 1:74 percent in the long run and the long run welfare gain (loss) to the lowest (highest) income quintile is 0:31 percent (0:4 percent) in initial resources. We also show positive e ects on domestic assets and reduced age pension expenditures. However, the reform yields an aggregate e ciency loss of 0:35 percent in initial resources. The e ciency loss is signi cantly smaller (but the equity implications are not as favourable) if the less distortive, consumption tax (rather than progressive income taxes) is used to maintain a balanced government budget. 2 In the next section, we provide a technical description of our economic model. Section 3 reports on the calibration of the model and compares the benchmark steady state equilibrium solutions with Australian data. Section 4 presents the simulation results for the examined reforms to superannuation taxation, concentrating on equity and e ciency implications of the AFTS proposed reform. Finally, Section 5 o ers some concluding remarks and suggestions for future research. 2 The economic model The model is a small open economy variant of Auerbach and Kotliko s (1987) OLG model that is stylised to the Australian economy. It builds on Kudrna and Woodland (2011, 2013) by including (i) a detailed disaggregation of households into income quintiles 2 We simulate the e ects of the AFTS proposed reform to superannuation taxation, assuming budgetequilibrating changes in either the progressive income taxation or the consumption tax rate. Each case requires a tax hike to nance the contribution rebate - a component of the reform. 4

6 based on ABS (2007) data and (ii) main aspects of mandatory superannuation, means tested age pension and progressive income tax policy settings. The interaction between household behaviour and the retirement income and tax policy settings is essential for the analysis of the AFTS reform to superannuation tax policy. Distinguishing households by age and income type is equally important because it allows us to examine both the inter- and intra-generational welfare (or equity) implications. One important extension is that this model allows households to draw down their superannuation assets gradually, compared to the lump sum payouts of superannuation savings assumed by Kudrna and Woodland (2013). In the present framework, households after reaching age 60 years can draw down their superannuation assets, with their withdrawals subjected only to the legislated minimum withdrawal limits. A detailed description of the model is provided below. 2.1 Households We consider a model economy that is populated by sequences of cohorts distinguished by age a and income type i. In particular, there are 70 generations aged from 21 to 90 years at any time t, with each generation consisting of the lowest, second, third, fourth and highest quintiles of households. Every year, a new generation aged 21 years enters the model structure and faces random survival with a maximum lifespan of 70 years, while the oldest generation aged 90 years dies. Random survival is given by the conditional survival probabilities denoted by s a. We use a stationary demographic setup with a constant population growth rate, n, which together with survival probabilities, gives time-invariant cohort shares, a = [s a (1 + n)] a 1 : Each i-type household who begins her economic life at time t is assumed to optimally choose consumption, c, and leisure, l, at each age to maximise the expected lifetime utility function given by 1 max fc i t+a 21 ;li t+a 21 g 1 1= X90 a=21 S a a 21 u(c i t+a 21; l i t+a 21) 1 1= ; (1) 5

7 subject to the within period budget constraint written as A i a;t = (1 + r)a i a 1;t 1 + LE i a;t + AP i a65;t + SB i a60;t + ST i a<65 + B i 45<a<65;t T (y i a;t) (1 + c ) c i a;t; (2) where the annual utility, u(c; l); being discounted by the subjective discount factor, ; and the unconditional survival probability, S a = Q a j=21 s j 1: In the per period budget constraint (2), A i a;t denotes the stock of ordinary private assets for type i household at the end of age a and time t, which equals the assets at the beginning of the period, plus the sum of interest income, ra i a 1;t 1, labour earnings, LE i a;t, age pension, AP i a65;t, superannuation pension, SB i a60;t; social transfer payment, ST i a<65; and bequest receipts, B i 45<a<65;t; minus the sum of income taxes paid, T (y i a;t), and consumption expenditures, (1 + c t) c i a;t. 3 Labour earnings are the product of labour supply, 1 l i a;t; and the hourly wage, w t e i a, where w t is the market wage rate for a person with unit e ciency and e i a is the age- and income-speci c earnings ability variable. The labour supply is required to be non-negative, 1 l i a;t 0. 4 The taxable income, y i a;t = LE i a;t +ra i a 1;t 1 +AP i a65;t; comprises labour earnings, investment income and the age pension. The accidental bequests, B i 45<a<65;t, are assumed to be equally redistributed to all surviving i-type households aged between 45 and 65 years. Households are assumed to be born with no wealth and to exhaust all accumulated wealth at the maximum age of 90 years, so that A i 20;t = A i 90;t+70 = 0. 5 We also impose borrowing constraints by requiring A i a;t 0 to prevent younger households from borrowing against their superannuation payouts, as the superannuation guarantee legislation prohibits such borrowing. 3 The social transfers, ST i a<65; are assumed to be paid to households in the lowest to the fourth income quintile aged younger than 65 years (see Appendix for the details). 4 Notice that when the household chooses to allocate all time endowment to leisure, l i a;t = 1; the household is fully retired from workforce. Further note that retirement is not irreversible in our model, as households may return to work after an episode of not working. 5 The assumption of no planned bequests is based on Gokhale et al. (2001) who document a number of studies that found strong empirical evidence against inter-generational altruism, supporting the view that bequests are largely accidental. 6

8 2.2 Private and public pensions Australia has a three pillar retirement income system. The rst is a mandatory, publiclymanaged "safety net" pillar comprising the age pension. The second is also mandatory, but is a privately-managed and fully-funded superannuation guarantee scheme. The third pillar includes other long term private savings such as voluntary superannuation. Our model incorporates the main features of the two publicly stipulated pillars - age pension and mandatory superannuation. We begin with compulsory superannuation as superannuation assets and/or incomes from these assets a ect the age pension through its means testing. The superannuation guarantee mandates employers to contribute a given percentage of gross wages into the employee s superannuation fund. We assume that mandatory contributions are made on behalf of all working households at the contribution rate, cr, from their gross labour earnings, LEa;t. i These contributions net of the contribution tax, s cr; are added to the stock of superannuation assets, SA i a;t; that earns fund income at the after-tax interest rate, (1 r a<60) r. 6 Superannuation assets are assumed to be preserved in the fund until households reach age 60. The households aged 60 years and over can draw down their superannuation savings as pensions, SBa60;t, i which are included in the per-period budget constraint (2). The stock of superannuation assets accumulates in the fund according to SA i a;t = [1 + (1 r a<60) r] SA i a 1;t 1 + (1 s ) cr LE i a;t SB i a60;t; (3) where r a<60 is the e ective earnings tax rate and s denotes the contribution tax rate. Note that households aged 60 and over can decide on the amount of their superannuation pensions, SB i a60;t, but these pensions must satisfy the minimum withdrawal limits. 7 The age pension, AP i a65;t; is paid to households aged 65 years and over provided that they satisfy the means test. The means test comprises the income test, AP i i a65;t; and the asset test, AP a i a65;t, with the test that results in lower age pension payments (i.e., 6 Following the existing legislation that exempts any earnings generated by superannuation assets supporting a pension (or other income stream), we make a simplifying assumption that only the fund earnings of households younger than 60 are taxed. 7 The minimum limits are aged based. These are 4 percent of the balance for households younger than 65 years, 5 percent for those aged 65-74, 6 percent for years old, 7 percent for years old, 9 percent for years old and 11 percent for those 90 years old. 7

9 binding test) applied. The means testing of the age pension can be expressed as AP i a65;t = min AP i i a65;t; AP a i a65;t AP i i a65;t = max min p; p by i a65;t IT ; 0 AP a i a65;t = max min p; p (A i a65;t + SA i a65;t) AT ; 0 ; (4) where p is the single rate of the maximum age pension, is the income taper rate, represents the annual asset taper rate, IT denotes the income free threshold and AT is the asset free treshold. The income assessed under the pension income test, by a65;t i = r(a i a 1;t 1 + SA i a 1;t 1) + 0:5 LEa;t; i consists of interest earnings generated from superannuation and non-superannuation assets and half of labour earnings. 8;9 The AFTS proposed reform in the aforementioned algebraic description involves the complete removal of s and a 50 percent reduction in r, with the reduced fund earnings rate of r also applied to the earnings from superannuation assets of households aged 60 years and over. In addition, the superannuation contributions are taxed progressively as other private income in hands of households, with the progressive income tax, T (ya;t); i in per-period budget constraint (2) applied to the taxable income that becames ya;t i = (1 + cr) LEa;t+rA i i a 1;t 1+APa65;t: i Furthermore, the right-hand side of (2) is extended to include superannuation contribution o set, crlea;t, i where denotes the contribution o set rate. 2.3 The rest of the model The model is a general equilibrium model which, in addition to the household and pension sectors, includes the production, government and foreign sectors. The production sector comprises a single producer that represents a large number of perfectly competitive rms. This representative producer maximises the present value of 8 According to the current policy, only half of labour earnings up to $13,000 per year is assessed under the income test. 9 We follow the means testing rules for superannuation as closely as possible (see FaCSIA, 2009). The superannuation assets are assessed in full under the asset test and the interest income generated by superannuation assets is subject to the income test, re ecting the means testing applied to shortterm income streams. In fact, the deeming approach is applied to these pensions and other nancial investments with the pre-speci ed rates of return on these assets. 8

10 all future pro ts discounted at the world interest rate, r, speci ed by 1X max D t 1 f (F (K t ; L t ) C(I t ; K t ) I t (1 + cr)w t L t ) ; (5) t=0 by choosing capital, K t ; labour input, L t ; and net investment, I t ; subject to the (per capita) capital accumulation equation of the form (1 + n)k t+1 = I t + (1 ) K t ; (6) where D t = (1+n) t =(1+r) t accounts for discounting and population growth, f stands for the corporation tax rate, F (K t ; L t ) represents the production of gross output, C(I t ; K t ) gives the adjustment cost function and the term (1 + cr)w t L t denotes the total wage bill, which also includes mandatory contributions. The government is assumed to maintain a balanced budget, which includes the pension expenditures, AP t ; social transfers, ST; and public consumption, G on the expenditure side and the tax revenues from household income, T Rt Y ; consumption, T Rt C ; and superannuation, T Rt S ; and rm s pro ts, T Rt F on the income side. The per capita pension expenditures and tax receipts from households and rms in period t are given by AP t = P 5 i=1! P 90 i a=65 a APa;t i T R Y t = P 5 i=1! i T R C t = P 5 i=1! i T R S t = P 5 i=1! i P 90 P 90 a=21 a T (y i a;t) a=21 a c t c i a;t (7) s cr LEa;t i + r a<60 rsa i a 1;t 1 P 90 a=21 a T R F t = f (Y t q t K t (1 + cr)w t L t ); which are the weighted averages of each component across households, with weights given by the intra-generational shares,! i ; and cohort shares, a. 10 In the per capita corporation tax revenue, Y t is the output net of adjustment costs and q t K t represents depreciation of the value of the capital stock. The government budget is assumed to be balanced in every t by adjusting (i) the progressive income tax schedule, T (y i a;t); (proportionally rising/reducing average and marginal tax rates) or (ii) the consumption tax rate, c t: 10 Note that government consumption and social transfers are assumed to be constant, with the per capita social transfers equal to ST = P 4 i=1! P 64 i a=21 a STa: i 9

11 The foreign sector is represented by the international budget constraint. The constraint equates capital ows with the current account and can be written in per capita terms as (1 + n)f D t+1 F D t = T B t rf D t ; (8) where F D t to be the per capita net foreign debt at the beginning of time t; T B t is the trade balance (or net export) and rf D t represents the interest payments on net foreign debt. The domestic interest rate, r, is exogenous in this small open economy model and equal to the world interest rate. The endogenous variables of the model are determined such that all agents optimise their objective functions subject to any constraints and such that all markets clear in every period. The clearing conditions for labour, capital and output markets are L t = P 5 i=1! P 90 i a=21 ei a;t(1 la;t) i a q t K t = P 5 i=1! i Y t = P 5 i=1! i P 90 a=21 Ai a;t + SAa;t i a F D t (9) P 90 a=21 ci a;t a + I t + G t + T B t ; where q t is the price of capital (i.e., Tobin s q) that is obtained by solving the rm s pro t maximisation problem. 3 Calibrating the model The model is calibrated to the key Australian aggregates averaged over the ve-year period ending in June We assume a stationary demographic environment with the constant population growth rate, n; set to current 1:8 percent per year. The population growth rate together with the male survival probabilities, s a ; taken from the life tables (ABS, 2010a) generates the existing old aged dependency ratio of 0:2. The intra-generational shares,! i ; are equal to 0:2 for each income class because of the income quintiles used by ABS (2007). Below we discus intra-generational di erences among households, present the per-period utility and production functions and report the values for the model parameters. We also provide a comparison of the model generated solutions with Australian data for some variables. 10

12 3.1 Income heterogeneity among households We consider ve income types of households in each generation that di er by their exogenously given earnings ability and social transfer payments (excluding the age pension). The earnings ability is the potential wage earned with all time endowment allocated to work. Using the estimated lifetime wage function for males with completed high school education from Reilly et al. (2005) and the income distribution shift parameter, i ; the earnings ability variable, e i a; is constructed as e i a = i e 2:235+0:04(a 17) 0:00067(a 17)2 ; where i is set to 0:26 for the lowest quintile, 0:55 for the second quintile, one for the third quintile, 1:52 for the fourth quintile and 2:63 for the highest quintile. These values are the ratios of the private incomes of lower and higher quintiles to the private income of the third quintile, calculated from ABS (2007) - Table 7, p.22. Hence, the earnings ability pro le for middle income households (i.e., those in the third quintile) is taken from Reilly et al. (2005) and the pro les for lower and higher income quintiles are shifted down and up to approximate the private income distribution in Australia. 11 In order to match not only private income but also gross total income for each income quintile, we assume that households receive social transfers, denoted by ST i a in equation (2). These payments, which are assumed to be constant and received by households (except for those in the highest quintile) aged younger than 65 years, are calculated as follows. First, we use the ABS (2007) data to derive the share of social transfers in gross total income for each eligible quintile. These shares are 0:44 for the lowest quintile, 0:3 for the second quintile, 0:15 for the third quintile and 0:06 for the fourth quintile. Then, we calculate the value of social transfers for eligible households in the benchmark steady state such that these payments together with the endogenous age pension yield the aforementioned shares in their lifetime gross income. 11 We also assume the earnings ability for each income class after age 65 to decline at a constant rate to reach zero at age 90 as Reilly et al. considered only workers aged years. 11

13 3.2 Preferences and technology Our choices of the annual utility and production functions and of the parameter values are standard in the literature. The per-period utility function takes the constant elasticity of substitution (CES) form u(c; l) = c (1 1=) + l (1 1=) 1=(1 1=) ; where the intra-temporal elasticity of substitution, ; is set to 0:9 and the value for the leisure distribution parameter, ; is 1:4. The remaining parameters in the lifetime utility (1) are the inter-temporal elasticity of substitution, = 0:4; and the discount factor, = 0:99, with its value chosen to generate the capital output ratio of 3 (ABS, 2010b). The technology is described by the standard CES production function h (1 1=) F (K t ; L t ) = "K t + (1 ")L (1 1=) t i [1=(1 1=)] ; where the technology constant, = 0:88; is calibrated to reproduce the market wage rate that is normalised to one in the benchmark steady state equilibrium. The elasticity of substitution in production, = 0:87; and the capital intensity parameter, " = 0:45; are calibrated via the producer s rst order conditions to match the interest rate and national account data for factor shares. The exogenous interest rate is set to 4 percent, which is the same rate as in Creedy and Guest (2008b). Following Fehr (2000), the adjustment cost function is assumed to be quadratic in net investment and given by C(I t ; K t ) = 0:5 (I t =K t (n + )) 2 K t ; where the value for the capital depreciation rate, ; that is set to target the investment capital ratio of 0.09 (ABS, 2010b) is 7:2 percent and the adjustment cost parameter of = 10 is taken from Auerbach and Kotliko (1987). We also target the ratio of net foreign debt to capital stock of 0:195, re ecting net foreign ownership of about 19:5 per cent of Australia s capital stock (ABS, 2010b). 12

14 3.3 Policy parameters Table 2 reports the values for taxation and retirement income policy parameters. The values for the age pension and superannuation parameters are those applicable in September The age pension eligibility age is 65 years. The consumption tax rates is set to the statutory GST rate of 10 percent. We then compute the adjustment parameter to replicate the average ratio of this tax revenues to GDP, which was 0:0389 over the ve-year period ending in June 2010 (Commonwealth of Australia, 2011). The product of the statutory tax rate and the computed adjustment parameter give the e ective rates on consumption, c = 6:94%: The corporation tax rate is set to the statutory rate of 30 percent and the government budget is assumed to be balanced with no government debt. Table 2: Values of policy parameters in benchmark model Description Value Source Age pension Maximum age pension p.a. (in $100,000) Data Income test threshold p.a. (in $100,000) Data Asset test threshold (in $100,000) 3.07 Data Income reduction rate 0.5 Data Asset reduction rate (annual) Data Superannuation Mandatory contribution rate 0.09 Data Contribution tax rate 0.15 Data Effective earnings tax rate Data Tax policy Statutory consumption tax rate [GST] 0.1 Data Statutory corporation tax rate 0.3 Data Income tax function Estimated [d] Notes: [a] The Income tax function is estimated, using the income tax schedule. The model incorporates the di erentiable approximation function of the Australian progressive personal income tax schedule in The approximation income tax, T (y); is a function of taxable income, and it takes the following form: M P 1 T (y) = t 5 (y) t 5 (yt 1 ) exp z=1 t 5 (y) = m 5 (y yt 5 ) + tax 5 ; (0:1) z z yz ; z = 1; :::; M 1; z where z = ( 1 ; 2 ; 3 ; 4 ) is a parameter vector, M denotes the number of tax brackets (M = 5), yt 1 and yt 5 are the lowest and highest tax thresholds (yt 1 = 0 and yt 5 = 180, 13

15 expressed in $1,000), m 5 is the top marginal tax rate (m 5 = 0:45) and tax 5 is the tax payable at the highest threshold (t = 54:55, expressed in $1,000). The parameter vector z = ( 1 ; 2 ; 3 ; 4 ) is estimated by nonlinear least squares using the Stata software. We construct a grid of equally spaced incomes in the range [0, 200.5] and the corresponding income taxes payable based on the Australian tax schedule, with both variables expressed in units of $1,000. The obtained parameter estimates are z = ( , , , ). 3.4 Computation and benchmark solution After specifying the parameter values, we compute the solution to the benchmark steady state equilibrium, using the GAMS software. 12 Our algorithm applies the iterative Gauss- Seidel computational method that was suggested by Auerbach and Kotliko (1987). The steps carried out to solve for the steady states and the transition paths are listed in Kudrna and Woodland (2011). In this subsection, we outline the way of dealing with the non-convexity of the household budget set that is caused by the age pension means test. We follow Altig et al. (2001) to handle the kinked households budget constraints and identify households that choose to locate at the kinks in particular periods by evaluating their income assessable under the pension income test. If the assessable incomes are close (rounded to 6 decimal places) to the income threshold of the pension income test, we set these incomes exactly to that threshold. By doing that we put such households exactly at kinks in each period in which being at a kink is optimal. Table 3 reports on the benchmark steady state solution for the key macroeconomic ratios and household net income variables and provides a comparison with the Australian data reported as averages over the ve-year period ending in June As shown, we match the distribution of net incomes across the income quintiles of households and the Gini coe cient measured in net income very closely with the actual data. 12 We use GAMS software also to compute the transition paths of the superannuation taxation reforms. 13 We have also undertaken a comparison of selected life-cycle pro les from the benchmark model with data drawn from HILDA (Wooden et al., 2002) data sets. The age pro les for labour supply and labour earnings t comfortably with the HILDA data, while total assets at advanced ages generated by the model are smaller than those observed in the data. This is because our model abstracts from any bequest motives and thus all households are required to exhaust their assets at the maximum age of 90. Details of this comparison can be obtained from the authors. 14

16 Table 3: Benchmark steady state solution and Australian data Benchmark Australia Variable model Expenditures on GDP (% of GDP) Private consumption Investment Government consumption Trade balance Government indicators (% of GDP) Pension expenditure Social transfers (a) Personal income tax revenue Corporation tax revenue Consumption tax revenue (GST) Superannuation tax revenue Net income shares Lowest quintile Second quintile Third quintile Fourth quintile Highest quintile Gini coefficient Notes: The data for Australia are taken from ABS (2010b, 2011) and are presented as averages during ; [a] These include disability pension and family benefits. Table 3 also indicates that the expenditures on GDP (or components of aggregate demand) expressed in per cent of GDP are very close to their actual values, except for the trade balance, which is positive and implied by the calibration target for the net foreign debt to capital ratio. Importantly, the model does a good job in matching government indicators with the actual data. Notice that several government indicators such as the pension expenditure in percent of GDP are replicated exactly, using adjustment parameters to target the underlying scal ratios (see the Appendix for more details). The di erence between the model and actual revenues from superannuation taxes is due to the assumed 40 years of superannuation accumulations in the model, whereas the superannuation guarantee was introduced only in 1992 with 3 per cent minimum contributions initially. The revenue from personal income taxes is also somewhat higher in the model as our approximation function of the Australian income tax schedule abstracts from any tax o sets. 15

17 4 Simulation results We now employ the model speci ed above to investigate the economic e ects of the AFTS proposed reform. Recall that the AFTS recommended e ectively a shift from the current comprehensive income tax T*T*E regime with concessional tax rates on employer contributions to a pre-paid expenditure tax (TEE) regime. Our model-based simulations account for the following changes of the AFTS proposed reform: (i) The removal of the existing concessional 15 percent tax rate, with mandatory superannuation contributions taxed progressively as other income; 14 (ii) Superannuation contribution rebate (or o set) of 15 percent; 15 and (iii) 50 percent reduction in the fund (statutory and e ective) earnings rate imposed also on the earnings from assets supporting superannuation pensions. 16 The AFTS proposed reform is assumed to be implemented at once in 2010, with the existing generations in 2010 and all future born generations optimally responding to these superannuation tax changes by adjusting their consumption/saving and leisure/labour supply choices over their remaining lifetimes. 17 Speci cally, there will be direct effects of the reform on the lifecycle behaviour of households through changes in their tax payments and lifetime net incomes, as well as indirect e ects due predominantly to budget-equilibrating tax adjustments. As pointed out earlier, we consider two budgetequilibrating tax adjustments, with the changes in either the income tax schedule or the consumption tax rate assumed to balance the government budget. 14 The model only includes mandatory employer contributions, and so we are not able to capture the e ects of the proposed removal of the government co-contribution on some personal contributions and of the spouse contribution tax o set. However, these changes would only a ect a small number of people relative to the other components of the reform. 15 The AFTS provides some micro-simulation analyses to, for example, calculate the e ects of the reform on replacement rates. In their calculations, the rate of the contribution o set is actually 20 percent. This is to keep the e ective tax rate on contributions at 15 percent for the majority of income earners in the new income tax schedule with 0, 35 and 45 percent marginal tax rates, which is also proposed in the report. We set the o set at 15 percent, because we use the income tax schedule with the majority of tax payers facing the 30 percent marginal tax rate. 16 The AFTS pointed out that by keeping in place imputation credits for dividend income, the e ective rate of fund earnings tax on average fund would be close to zero. In our reform scenario, the e ective rate of the fund earnings tax is 3.75 percent. 17 Notice that we only consider the AFTS superannuation tax changes and abstract from any other pension policy changes, including the recently legislated increases in the age pension access age and mandatory superannuation contribution rate that are to be implemented gradually in the near future. Hence, throughout the transition path, the age pension access age and the mandatory contribution rate are kept at 65 years and 9 percent, respectively (as they were in 2010). 16

18 The simulation results for the AFTS proposed reform are discussed in terms of the macroeconomic, equity and e ciency e ects. The discussion involves the e ects of the reform upon the impact in 2010, during the transition as well as in the long term. We rst provide an overview of the key results for the AFTS reform and then we discuss the results in more detail. We conclude this section by discussing the e ects of alternative reforms, including a switch to the standard TEE regime with and without the contribution rebate. 4.1 Overview The AFTS reform, which adopts a variant of the TEE regime with a contribution o set basically consists of three parts. The rst part is to abolish the concessional 15 percent tax on mandatory contributions and to partially eliminate the fund earnings tax. These changes lead to signi cantly greater accumulations of superannuation assets for all households. Although larger superannuation assets partly o set ordinary non-superannuation assets, total domestic assets (or national wealth) increase as a result of the reform. The reform also improves self-funding in retirement as increased savings and interest incomes at older ages mean that the means tests become more binding for potential age pension recipients, resulting in lower government expenditures on the age pension. The second and third parts of the reform are to treat the contributions as ordinary taxable income and to refund 15 percent of the contributions back to households. Whereas the second policy increases taxable income (i.e., income tax base) for all households, only higher income households e ectively pay larger income taxes as lower income types are more than compensated via the contribution o set. Nevertheless, the government revenue from progressive income taxes increases, but the fall in superannuation tax revenue requires a tax hike to the government budget to be balanced. As expected, the required increases in progressive income taxes are more distortive for the lifecycle behaviour of households, leading to less favourable outcomes for most macroeconomic aggregates in comparison with the alternative, budget-equilibrating adjustments in the consumption tax rate. Of a particular interest of this analysis are the e ects of the AFTS proposed reform on equity that we measure by changes in welfare, net income shares among the income quintiles and the Gini coe cient. The main nding is that the reform improves in vertical 17

19 or intra-generational equity and reduces income inequality upon the impact and in the long term, as the welfare and net income shares of lower income households increase (compared to lower welfare of higher income types) and the Gini coe cient declines. These equity e ects also depend on the choice of the budget-balancing tax instrument, with larger equity improvements found when the income tax adjustments are assumed to balance the budget. This is because the required increases in progressive income taxes a ect negatively especially higher income households, while the increases in the regressive, consumption tax have greater negative e ects on welfare of lower income types. The reform is also shown to improve inter-generational equity, increasing average welfare for young and future born generations, while older generations attain lower welfare due to the tax applied to superannuation fund income earned in the decumulation or pension phase. In contrast to these positive equity e ects, the reform yields an aggregate e ciency loss of 0.35 percent in initial resources, which is signi cantly smaller if the less distortive consumption tax (rather than progressive income taxes) is adjusted to keep the government budget in balance. 4.2 Macroeconomic implications The AFTS policy recommendations with the progressive taxation of contributions and the contribution o set represent a systemic reform of the superannuation tax policy with potentially important macroeconomic implications for the Australian economy. The model-based results for the macroeconomic e ects of the AFTS reform are displayed in Table 4 as percentage changes in the selected per capita variables in the selected years of the transition from the benchmark steady state solution. Recall that the reform is implemented in 2010, with the results for that year depicting the impact e ect. The results are also reported for the two transitional years of 2015 and 2030 and in the long run. In the discussion below, we focus on the implications for the key macroeconomic variables such as domestic assets, labour supply and consumption as well as for the main government indicators. 18

20 Table 4: Macroeconomic implications of adopting the AFTS proposed reform (Percentage changes in the selected variables from the initial (benchmark) steady state solution) Variable (i) Budget equilibrating income tax (ii) Budget equilibrating consumption Initial SS adjustments tax adjustments solution Long Long [a] run run Domestic assets Capital stock Foreign debt Labour supply Output (GDP) National product (GNP) Consumption Age pension expenditure Income tax revenue Superannuation revenue Tax rate [b] Notes: [a] The monetary values of the initial steady state solutions are expressed in units of $100,000 and per capita; [b] These are percentage changes either in progressive income taxation or in the consumption tax rate. The removal of the superannuation contribution tax rate, combined with the partial elimination of fund earning tax leads to larger superannuation assets, which generate the increases in total domestic assets reported in Table 4. The increases in total domestic assets are somewhat mitigated by reduced ordinary, non-superannuation assets, which decline due to higher progressive income taxes on younger generation of especially wello households. 18 While the total wealth (or domestic assets) increases, the e ect of the AFTS reform on the capital stock is negative. In the short and medium term, the decreases in the capital stock are caused by lower capital prices (not displayed) but the long run decrease is entirely driven by lower labour input. Thus, the increases in overall household savings are exported abroad, leading to substantial reductions in net foreign debt. The e ects of the reform on per capita labour supply are mostly negative as working age generations of higher income types facing increased progressive income taxes reduce their working hours. Moreover, the contribution o set represents the pure income e ect on labour supply of low income households that e ectively face no or negative tax on their superannuation contributions. Further note that assuming the budget-equilibrating adjustments in more distortive, progressive income taxation reduces per capita labour 18 As mentioned, we prevent all households from borrowing against their future superannuation payouts by imposing borrowing constrains. Removing these constraints would generate larger reductions in private non-superannuation assets and may in fact reduce total wealth. Note that Creedy and Guest (2008b), who allow for borrowing in their model, nd lower savings rate generated by their simulation of the removal of the concessional tax rate on superannuation contributions. 19

21 supply by almost 1 percent, compared to a 0:67 percent long run decline in per capita labour supply under the budget-equilibrating consumption tax changes. 19 The output (or GDP) is produced, using the capital stock and labour supply, and so the negative e ects of the reform on output follow the changes in these two inputs to production. The e ects on national income (or GNP) that equals the output less the interest payments on foreign debt are more favourable because of lower foreign debt. The largest expenditure on GDP is consumption, with per capita consumption declining in the long run by 0:98 percent with the income tax adjustments and by 0:32 percent with the consumption tax adjustments. This is an outcome of reduced net consumption of older retired households (paying the tax on superannuation fund earnings in the decumulation phase) and working age generations of mainly well-o households (paying higher marginal income tax rates). 20 The results for the government expenditure on the age pension indicate improved selffunding in retirement as reliance of the elderly on this public pension declines. Given the means testing of larger superannuation assets and earnings from these assets, the long run decrease in age pension expenditures is 1:19 percent with the income tax adjustments or 1:58 percent with the consumption tax adjustments. As for the government revenues, we only present the e ects on the income tax (net of the contribution o set) and superannuation tax revenues, which are directly a ected by the reform. As expected, the reform increases the revenue from personal income taxes that now apply also to superannuation contributions, whereas the superannuation tax revenues collected only from the reduced fund earnings tax fall signi cantly. Upon the impact in 2010, the changes in the two tax revenues roughly o set each other, with no signi cant adjustment in either of the two budget-equilibrating tax instruments required. During the transition, the improved superannuation tax revenues (due to large superannuation accumulations) are not enough to o set a falling income tax base, calling for tax hikes to maintain a balanced government 19 The wage rate, which is not displayed, is unchanged in the long term in this small open economy model but during the transition it moves in the opposite direction to the changes in average labour supply, determined by the marginal product of labour. 20 The e ects on the other components of aggregate demand are not displayed. In brief, government consumption is kept constant over the transition and the policy e ects on investment demand are similar to those on the capital stock. The net export, which balances output supply with domestic demand, decreases in the long run as the output falls more than domestic demand. 20

22 budget. 21 In the long run, the budget-equilibrating increase in the average income tax rate is 3:5 percent, while the increase of almost 8 percent in the consumption tax rate is required for the government budget to be in balance. The income taxes collect more than three times more in revenues than the consumption tax rate (i.e., the GST rate applied to about 70 percent of consumption goods), which partly explains the relative di erence in these two tax hikes. However, the income tax hikes have far more detrimental impacts on most macroeconomic aggregates such as assets, labour supply and consumption, relative to the displayed increases in the consumption tax rate. 4.3 Equity and e ciency implications One of the main objectives of our quantitative analysis is to examine whether and to what extent the AFTS proposed changes to the current superannuation tax policy improve equity. We rst calculate the policy e ects on net income shares for the ve income quintiles and on the Gini coe cient to examine the implications for vertical equity. We then discuss the distribution welfare implications for di erent generations (measuring the e ects on inter-generational equity) and for di erent income quintiles (measuring the e ects on vertical equity). Table 5 reports the equity e ects of the reform that are measured as the percentage changes in net income shares (for selected income quintiles) and in the Gini coe cient in net income from their values in the benchmark steady state with the existing superannuation tax arrangements. 21 Note that all components of the income tax base (superannuation contributions, labour earnings, income from non-superannuation private assets and the age pension) decrease over the transition, but the main reason for a declining income tax revenue is lower labour earnings. The labour earnings decline because of the income e ect of larger superannuation assets on labour supply and disincentives to work from increased progressive income taxes. 21

23 Table 5: Equity implications of adopting the AFTS proposed reform (Percentage changes in net income shares from the initial steady state solution) Budget balancing tax instrument (i) Progressive income taxation (ii) Consumption tax rate Notes: [a] Measured in net income. Period Selected income type Lowest Third Highest quintile quintile quintile Gini coefficient [a] Long run Long run The results in Table 5 show that the reform improve vertical equity and reduce income inequality upon impact, during the transition and in the long term as the net income share for the lowest (highest) income quintile increases (declines) and the Gini coe cient decreases. These equity e ects are robust to the choice of a tax policy balancing the government budget, but there are important di erences between the examined budgetequilibrating tax instruments. First, the improvements in vertical equity are greater with the income tax adjustments. For example, the long run decrease in the Gini coe cient is 1:74 percent with the income tax adjustments, compared to a 0:9 percent long run decrease in the Gini coe cient with the consumption tax adjustments. The main reason for this di erence is that higher progressive income tax rates have negative impact especially on more a uent households, while the burden of higher regressive consumption taxes is born more heavily by lower income households. In addition, the calculations of net income shares and the Gini coe cient directly include the budget-equilibrating increases in income tax rates through higher income tax payments, whereas the required increases in the consumption tax rate do not directly a ect net incomes. Second, relative to the impact e ect, the implications for vertical equity with the income tax adjustments improve during the transition and in the long term, but the opposite is true for the consumption tax adjustments. This can be observed in Table 5 by comparing the short run and long run e ects on net income shares of the lowest and highest income quintiles and on the Gini coe cient. As already mentioned, the increases in progressive income tax rates are distortive especially for higher income households, resulting in further declines in their net incomes and a decreasing Gini coe cient over 22