Analytical Studies Branch Research Paper Series
Market Behaviour Versus Tax Planning Responses to Changes in Marginal Income Tax Rates Among Older Couples

Abstract

The standard model of labour supply predicts that individuals’ taxable incomes will bunch at convex kink points in the budget set created by government tax and transfer systems. However, empirical evidence of such behaviour is scarce. Using administrative tax data from Canada, this paper estimates bunching responses in taxable income among older individuals at discontinuities in marginal tax rates created by federal and provincial and territorial systems, from 2001 to 2012. The results show that bunching is prevalent across the income distribution, but this is mostly driven by married individuals from 2007 onward. Further analysis reveals this occurs predominantly because pensioners split some of their pension income with their spouses who tend to have lower marginal tax rates. Pensioners do this pursuant to the introduction of this tax planning program in the 2007 federal budget. These findings offer credible evidence of significant intra-household tax planning behaviour that depends on the availability of deductions.

Keywords: pension income; elasticity of taxable income; tax planning; income splitting; bunching; empirical density design

Executive summary

This paper investigates the extent to which older Canadian taxfilers, aged 60 to 69, respond to predictable changes in marginal tax rates created by the tax and transfer system by exhibiting sorting behaviour in taxable income. More precisely, the prediction from the standard model of labour supply that individuals’ taxable incomes bunch at convex kink points in their budget sets is tested, in an empirical density design.

Using administrative tax data for the years from 2001 to 2012, the analysis assesses how individuals respond to changes in marginal tax rates created at the lower bounds of the second, third and fourth federal tax brackets; the lower bounds of the second and third provincial and territorial tax brackets; and the thresholds at which the Old Age Security and Employment Insurance benefits start being clawed back through recovery taxes. The results indicate the following.

1. Taxable income is responsive to the changes in tax rates at the thresholds analyzed. Bunching is observed at every threshold. Hence, individuals are sorting in their taxable incomes in response to the tax and transfer system, as expected.
2. Bunching appears the most prevalent among married (either legally married or in a common-law relationship) individuals from 2007 onward.

However, a closer inspection shows the following:

3. Bunching is driven primarily by individuals who are married and who used the pension income splitting provision of the tax code, pursuant to the introduction of this program in the 2007 federal budget.
4. In contrast, the bunching responses to changes in income tax rates appear to be negligible among
1. unmarried individuals
2. married individuals before 2007
3. married couples who do not have at least one spouse receiving private pension income and are ineligible for pension income splitting.

These findings offer credible evidence of intra-household tax planning behaviour that depends on the availability of deductions. Effectively, couples are not only using the pension income splitting program to reduce their own tax liabilities but also taking into account the marginal tax rates of their spouses to minimize household tax liabilities, as expected.

1 Introduction

Income taxes are often found to have small effects on individuals’ labour supply decisions and other economic behaviour (e.g., investment decisions) and large effects on individuals’ tax planning responses (Saez, Slemrod and Giertz 2012). Individuals can delay or avoid paying higher income taxes in many ways, such as by making tax-deductible charitable donations or contributions to retirement savings vehicles (Stiglitz 1985; Auten, Sieg and Clotfelter 2002; Bakija and Heim 2008). Shifting income from the personal to the corporate tax base is also a potential tax planning channel among business owners (Goolsbee 2000; Gordon and Slemrod 2000; Kreiner, Leth-Petersen and Skov 2014). Evidence suggests that the elasticity of taxable income (ETI) with respect to the marginal tax rate—a measure once viewed as a sufficient statistic for the efficiency cost of taxation (Feldstein 1995, 1999; Slemrod and Yitzhaki 2002)—depends on the availability of deductions and, therefore, can be manipulated by policy makers (Kopczuk 2005).

In general, the ETI is an incomplete measure of the deadweight loss from taxation when some burden of sheltering income is incurred as a resource or transfer cost (Chetty 2009). If the unit of taxation is the individual, sheltering can occur within households when family members split income—provided that resources are not shared, as in the unitary model of the household (Kornhauser 1993; Cantillon and Nolan 1998; Freiler, Stairs and Kitchen 2001; Gelber 2014). Consequently, a better understanding of the extent to which older couples reduce personal income tax liabilities through splitting warrants investigation.

The goal of this study is to assess the extent to which income taxes induce changes in market behaviour (sometimes called “real” responses) versus changes in tax planning strategies. The paper focuses particularly on income splitting between spouses as the channel through which tax avoidance may occur. Market response refers to a change in actual economic activity by individuals—such as a change in employment income through an adjustment in the number of hours worked. Tax planning response refers to the use of allowances, credits and deductions to reduce tax liabilities while holding constant economic behaviour.

The paper makes two key contributions. The first contribution is to use administrative data on a 20% nationally representative sample of Canadian taxfilers, for the years from 2001 to 2012, to estimate, in an empirical density design, the effects of changes in marginal tax rates on the taxable incomes of older individuals aged 60 to 69 (Saez 2010; Chetty et al. 2011). The standard model of labour supply predicts that individuals will bunch at convex kink points in their budget sets created by government tax and transfer systems, but evidence of such behaviour is scarce (Saez, Slemrod and Giertz 2012). This study finds significant bunching at various convex kink points along the income distribution associated with the lower bounds of the federal and provincial and territorial tax brackets, where marginal tax rates increase discontinuously as a result of tax progressivity. In addition, bunching is observed at two convex kink points where effective marginal tax rates change, namely the income thresholds where Employment Insurance (EI) and Old Age Security (OAS) benefits start being clawed back through recovery taxes. In these cases, responses are observed only among EI recipients and OAS-eligible individuals.

The second contribution of this paper is to estimate how much of this bunching is the result of intra-household tax planning. In theory, the magnitude of tax avoidance or evasion should be proportional to the expected costs of such behaviour, including costs from administration, effort or detection (Allingham and Sandmo 1972; Stiglitz 1985; Slemrod and Yitzhaki 2002). Pension income splitting in Canada is both legal and notional in the sense that it does not require a transfer of income or division of asset ownership. Pension income splitting is a no-cost tax avoidance channel for older couples; consequently, take-up is predicted to be high.^{Note 1} To separately identify the effect of splitting from other types of market behaviour or tax planning responses, the analysis exploits exogenous variation in individuals’ eligibility to split pension income by marital status and around the introduction of this program.^{Note 2} Moreover, the data provide information on whether individuals and their spouses are pension income recipients and the actual amounts of income they elect to split. This facilitates a direct comparison of bunching between couples who could have used or did use this tax provision and those who did not use it.

The results of this analysis show that bunching is only weakly observed before pension income splitting was introduced, as well as among unmarried individuals and couples who were ineligible for splitting after the reform was introduced. In contrast, the magnitude of bunching, after pension income splitting was introduced, among individuals collecting a private pension income or whose spouses do so is substantial. Further analysis conditional on whether splitting actually occurs shows that the bunching stems mostly from the use of this program. Within couples, bunching occurs from one of two possible behaviours:

1. an individual, who is a pensioner, sends enough income to a spouse to reduce taxable income to a lower tax bracket; or
2. an individual, whose spouse is a pensioner, receives income only up to the point where taxable income would otherwise enter a higher tax bracket.

Hence, pensioners are not only using this tax planning tool to lower their own tax liabilities but also to lower household tax liabilities, taking into account the marginal tax rates of their spouses. These findings offer credible evidence of intra-household tax planning behaviour that depends on the availability of deductions.

This paper contributes to a growing literature that seeks to disentangle market behaviour from tax planning responses to personal income taxation. Prior research shows that income shifting, both intertemporally and across tax bases, is substantial among high-income earners and the self-employed who have the most access to tax planning technologies (Gordon and Slemrod 2000; Goolsbee 2000; Chetty et al. 2011; le Maire and Schjerning 2013; Kreiner, Leth-Petersen and Skov 2014, 2016; Harju and Matikka 2016). However, few studies consider the extent to which families coordinate to reduce combined tax liabilities. This is likely because many studies that estimate the ETI are carried out using data from the United States, where the unit of taxation is the household. Other studies look at countries where the unit of taxation is the individual, but where household-level tax return data or the policy variation needed for proper identification are not available. A notable exception is Stephens and Ward-Batts (2004), who showed that the 1990 reform from joint to individual taxation in the United Kingdom led to an increase in the sharing of assets claimed by spouses with lower marginal tax rates. Wolfson and Legree (2015) used data from Canada on business owners and their families who work for the business and live in the same household. They posited that business owners make labour and dividend payments strategically to reduce household tax liabilities.^{Note 3} This paper is among the first to assess how, following the introduction of a mechanism that was explicitly designed to facilitate tax planning at the household level, taxable income responds to marginal tax rates within couples. It also relates more broadly to the work of Gugl (2009), who analyzed intra-household equity considerations of pension income splitting.

Since this paper centres on the responsiveness of older couples to income taxation, it relates to a large literature on the labour supply decisions of the elderly. Many studies investigate how pension receipt or retirement decisions are influenced by the design of public pension plans (Baker and Benjamin 1999; Feldstein and Liebman 2002; Baker, Gruber and Milligan 2003; French and Jones 2012). However, few studies consider the extent to which the tax code is a viable policy instrument for influencing the labour supply decisions of older workers (Schmidt and Sevak 2009; Alpert and Powell 2014; Messacar 2017). Lastly, this paper adds to a developing literature that estimates income responses to taxation using the empirical density design (Saez 2010; Chetty et al. 2011; le Maire and Schjerning 2013; Bastani and Selin 2014; Harju and Matikka 2015).

This paper proceeds as follows. The next section, Section 2, gives an overview of the standard model of labour supply and derives the bunching prediction; it includes a brief discussion of the empirical density approach for estimating excess mass. Section 3 summarizes features of the Canadian income tax system of relevance to the empirical analysis. Section 4 discusses the data and sample selection used in this study. Section 5 presents the baseline results, the analysis of the pension income splitting reform, two placebo checks, and tests of heterogeneous responses. Section 6 concludes.

2 Background

This section presents an overview of the theory of labour supply prediction that individuals bunch at convex kink points in their budget sets. Then, the method for estimating this bunching behaviour is briefly discussed. In particular, the theoretical and methodological contributions of Saez (2010) and Chetty et al. (2011), which are of direct relevance herein, are provided. The reader is encouraged to refer to those papers directly, as well as to the wide-spanning ETI literature discussed in Saez, Slemrod and Giertz (2012), for more information.

2.1 Theory

Saez (2010) considered the standard model of labour supply, in which an agent $i$ ’s utility depends positively on after-tax income, $c_i$, and negatively on before-tax income, $z_i$. The agent prefers to consume more but is hindered by the effort cost of supplying labour to earn more. The utility function is given by $u_i\left(c_i,z_i\right)$ and satisfies several conditions: (1) $\partial u_i/\partial c_i>0$ ; (2) $\partial u^2/\partial c^2<0$ ; (3) $\partial u/\partial z<0$ ; and (4) $\partial u^2/\partial z^2<0$.

To begin, suppose an agent faces a linear budget set with a constant marginal tax rate $t$, so that after-tax income is $c=z\left(1-t\right)$. A graphical depiction of the solution to this optimization problem is shown in Panel A of Figure 1. The agent will provide effort and earn income up to the point where the marginal benefit of consumption equals the marginal disutility of effort. For agent $i$, this results in the optimal levels of before-tax income, $z_i^{\star }$, and after-tax income, $c_i^{\star }$. The assumption that agents are heterogeneous in before-tax income because of underlying differences in preferences or ability is captured by heterogeneity in the utility function. The figure shows how the solution to this optimization problem varies across agents.

That the response in taxable income occurs through the labour supply adjustment is referred to as a market behaviour response. In other words, the actual behaviour of economic agents is being affected by income taxation in a standard manner as predicted by economic theory. To understand how workers are expected to respond, suppose that a convex kink were introduced in the budget set at income $\tilde{z}$ by increasing the marginal tax rate from $t$ to $t+dt$ on income earned above this tax threshold. Panel B of Figure 1 shows that, in this case, workers with earnings below the tax threshold (agents $i$ and $j$ in the figure) remain unaffected and continue to consume their initial amounts. However, an agent earning from $\tilde{z}$ to $\tilde{z}+d\tilde{z}$ (agent $k$ in the figure), who initially consumed $c_k^{\star }$ under the scenario of a constant tax rate, now finds it optimal to consume $c_i^{\star }$. Hence, the convex kink in the budget set results in an excess of agents who optimally bunch at the convex kink point in before-tax income. Panel C of Figure 1 shows how the density distribution of before-tax (taxable) income is expected to look—within the context of the theory—under the scenarios of a constant or progressive piecewise-linear tax system. In contrast, a “tax planning” response refers to the use of allowances, credits or deductions to modify taxable income so that bunching occurs but without any change in actual economic behaviour.

The standard prediction from the economic model assumes labour income (or another type of income that requires effort) is the primary income source. Especially among older workers, adjustments may be prevalent—compared with young and middle-aged workers—through transitions into flex-time, part-time (e.g., partial retirement) or bridge employment. If an individual’s main source of income requires no effort to collect, such as pensions, the size of the response is unclear. The response may be smaller because collecting income beyond a particular tax threshold has no effort cost. However, pensioners face an intertemporal trade-off between receiving an extra dollar at a higher marginal tax rate and delaying the receipt of that dollar until the following year, at a lower marginal tax rate. Bunching may be especially prevalent because the amount of pension income received can be manipulated. This issue is especially important in the context of this study, because only approximately one-third of taxfilers have employment earnings.

In addition, labour supply may not respond as predicted by the stylized model for many reasons. For example, many workers may have strict preferences to work a fixed number of hours per week. Similarly, employers may offer contracts with hours constraints. Search costs of changing jobs can significantly reduce the extent to which bunching at discontinuities in the marginal tax schedule occurs—this issue is explored in detail by Chetty et al. (2011). On balance, common labour frictions that are not always well represented in canonical models can significantly affect actual responses. The findings from this study, which centre on the introduction of pension income splitting, serve to credibly identify the effect of this tax planning tool on the ETI, which is analogous to that of a difference-in-differences approach.

2.2 Estimation

The approach developed by Saez (2010) and extended by Chetty et al. (2011) for estimating bunching in an empirical density design is used. To that end, the “bunch_count” Stata module (an “.ado” file) written by Olsen (2011) is employed in the analysis. This empirical procedure is briefly described here, with a discussion of how to interpret the results.

The approach begins by normalizing taxable income relative to the convex kink point analyzed; for example, having relative taxable income equal to zero in the reference year means taxable income was exactly equal to the tax threshold. Using this normalized taxable income variable, the study groups individuals with incomes within $10,000 on either side of the tax threshold into bins of width $250 (e.g., from -$10,000 to -$9,751, from -$9,750 to -$9,501, from -$9,500 to -$9,251), and bin counts (frequencies) are computed.

Then, a counterfactual density distribution is estimated. This gives a prediction of what the distribution would have looked like under a constant tax rate over the relevant range of taxable income. The counterfactual is estimated by fitting a polynomial to the bin counts, excluding data near the kink, where actual bunching would influence these results. The procedure adjusts for the fact that individuals at the convex kink point must come from other points to the right of this kink. The estimate of bunching is the excess mass around the convex kink point relative to the average density of the counterfactual earnings distribution within a range of income close to the convex kink point.^{Note 4}

For a bunching estimate $b$ obtained from this procedure, the point estimate indicates that the excess mass around the convex kink point is $b$ % of the height of the counterfactual distribution within the range of income close to the convex kink point. A parametric bootstrap procedure is used to compute standard errors for $b$.

3 Institutional details

Personal income tax in Canada is based on a measure of taxable income minus any permitted deductions; then credits are applied to determine the net amount payable. The unit of taxation in Canada is the individual. However the income tax system recognizes that individuals may have reduced abilities to pay income taxes when they have dependent spouses or relatives. This is provided for in the form of additional credits and of transfers of certain dependents’ unused credits to the taxpayer.

Income taxes are determined at both federal and provincial and territorial levels. Each government applies its own tax rates to a uniform measure of taxable income and offers its own credits to determine the net amount of income taxes owed. In 2012, federal taxable income was divided into four brackets. The tax rate was 15% for the first bracket ($42,707 or less), 22% for the second bracket (from $42,708 to $85,414), 26% for the third bracket (from $85,415 to $132,406), and 29% for the fourth bracket ($132,407 or more). This tax progressivity creates the convex kink points in individuals’ budget sets that are expected to induce bunching responses. The federal basic exemption in 2012 was $10,822, such that the marginal tax rate applied to taxable income below this threshold was 0%.^{Note 5}

At the provincial and territorial levels, the income tax structures and rates are significantly heterogeneous. For example, Alberta imposed a flat tax rate of 10% in 2012 on taxable income above a basic personal exemption of $17,282. The provinces that had only two tax brackets in 2012 were Newfoundland and Labrador, Prince Edward Island, Quebec, Ontario, Manitoba and Saskatchewan. The provinces and territories with three tax brackets in 2012 were New Brunswick, Northwest Territories, Yukon and Nunavut. The provinces with four tax brackets in 2012 were Nova Scotia and British Columbia. In general, provinces and territories set thresholds for their tax brackets that fall within the federal tax brackets. For example, in 2012, the lower bounds of the second and third brackets were the following: $32,893 and $65,785 in Newfoundland and Labrador; $39,020 and $78,043 in Ontario; and $37,013 and $74,028 in British Columbia. The differences between federal and provincial and territorial brackets are useful for the empirical analysis to separately identify bunching at these various convex kink points.

In addition to estimating bunching at these federal and provincial and territorial tax thresholds, this study also considers the extent to which individuals respond to clawback provisions in the OAS and EI programs. First, the OAS pension is a demogrant for individuals starting at age 65 (Baker, Gruber and Milligan 2003) based on Canadian legal status and residence requirements; the maximum annual OAS pension was $6,540 in 2012. However, if annual net world income exceeds a pre-specified amount, recipients must repay part or all of their pension benefits. In 2012, the threshold was $69,562, and the recovery tax rate was 15%.

Second, EI is a public insurance program providing temporary financial benefits to unemployed individuals, including regular benefits for those who lose their jobs through no control of their own and sickness benefits for people unable to work because of sickness or injury. Individuals who received regular benefits with net incomes exceeding a pre-specified threshold, which was $57,375 in 2012, may have been required to repay 30% of the lesser of net income in excess of the threshold or the total regular benefits paid in the tax year.

On January 1, 2007, the federal government implemented pension income splitting, which permits individuals to split income drawn from private pensions with their spouses. The pension recipients who are sending income (“pensioners”) may allocate up to 50% of eligible pension income to their spouse (“transferees”) to reduce joint tax liabilities. For those aged 65 years and older, eligible income includes annuity and registered retirement income fund payments, registered retirement savings plan annuity payments and other types of pension and retirement compensation. However, for those under age 65, eligible income is limited to registered pension plan (RPP) payments and certain benefits received as a result of the death of a spouse. As mentioned earlier, this transfer is notional in the sense that no direct transfer of income or division of asset ownership takes place. To declare the amount of pension income to be split each year, couples simply fill out and submit a form with their tax returns.

4 Data and sample selection

This study is based on an analysis of the Longitudinal Administrative Databank (LAD), a panel dataset comprising a 20% nationally representative sample of Canadian taxfilers. The sample was derived from T1 Family File (T1FF) income tax data from the Canada Revenue Agency. The T1FF is a yearly cross-sectional dataset of taxfilers and their census families. Individuals file tax returns independently in Canada, but census families were created in the T1FF according to the spousal social insurance number listed on each individual’s tax form or by matching the name, address, age, sex and marital status. As a result, the LAD contains a wide set of variables about demographics, earnings, income, taxes, transfers, credits, allowances and savings for the individuals represented and their spouse. The LAD is updated annually to ensure accurate cross-sectional representation. The tax data do not provide a direct measure of individuals’ marginal tax rates. To overcome this issue, these rates are calculated using the Canadian Tax and Credit Simulator (CTaCS) of Milligan (2016).

Table 1 presents the descriptive statistics for the sample used in this analysis. On average, individuals are 64.1 years old, of whom 52.0% are women and 72.3% are married (i.e., legally married or in common-law relationships; see Footnote 2). In addition, the table shows that a significant fraction of individuals have labour earnings from employment (34.7%) or self-employment (9.1%), investment income (44.6%), OAS income (42.8%), and private pension income (38.9%). The mean value of taxable income, conditional on these values being strictly positive, is $39,550, and the median value is $28,300 (expressed in 2012 constant dollars). Using CTaCS, the mean value of the predicted marginal income tax rate is 22.1%, and the median value is 23.6%.

The analysis centres on all Canadian taxfilers in the LAD, for the years from 2001 to 2012, who were 60 to 69 years of age in the reference year. This timeframe was chosen to provide a wide interval of observations for a comparative analysis over time, while centring the tax code reform from 2007 that introduced pension income splitting. This age group was chosen to focus on individuals who are both old enough to be collecting a pension, since this is a prerequisite for pension income splitting, and young enough to be employed, so that labour income responses to taxation are plausible. More precisely, the earliest age at which individuals can start to collect public pension income is 60, which serves as the benchmark for the lower bound of this target sample.^{Note 6} Individuals must start to receive income from their employer-sponsored RPP by the end of the year in which they turn age 69, which serves as the upper bound for this target sample.

5 Results

This section is organized as follows. First, an assessment is provided of how much individuals’ effective marginal tax rates actually change at the thresholds analyzed. The analysis centres on the lower bounds of the second, third and fourth federal income tax brackets, the lower bounds of the second and third provincial and territorial tax brackets, and the income thresholds at which OAS and EI benefits start being clawed back through recovery taxes.^{Note 7} The results are presented for the full period analyzed and separately for the periods preceding and following the introduction of pension income splitting. The analysis of bunching at the OAS recovery tax threshold is conditional on individuals aged 65 to 69 as a result of a benefit eligibility requirement for this program, and the analysis of bunching at the EI recovery tax threshold is conditional on individuals who are observed as receiving such benefits.^{Note 8}

Second, the analysis of bunching responses to the convex kink points is presented, both as a graphical inspection and in an empirical density design. Third, how individuals’ bunching responses to taxes changed as a result of the pension income splitting reform is assessed. Fourth, the results from placebo tests are presented that consider how individuals known to be unaffected by the OAS and EI recovery taxes behave at these thresholds. The section concludes with a heterogeneity analysis.

5.1 Changes in effective marginal tax rates

The bunching analysis exploits convex kink points in individuals’ budget sets. As a result of tax progressivity, individuals’ marginal tax rates should increase discontinuously at the thresholds analyzed. As a result, consideration of how much variation is actually observed in individuals’ marginal tax rates at these thresholds is warranted. For example, Milligan (2009) showed that income tax schedules in Canada are complex functions of age, province or territory of residence, income sources, spousal characteristics, and other factors that may interact to attenuate the effects of changes in direct taxes on effective marginal tax rates.

To investigate this issue, Charts 1-1 and 1-2 show how individuals’ effective marginal tax rates change at the thresholds analyzed. Chart 1-1 looks at the pre-reform period, and Chart 1-2 looks at the post-reform period. The tax rates (shown on the vertical axes) are calculated using CTaCS (Milligan 2016), and the running variable (shown on the horizontal axes) is taxable income relative to the threshold, as calculated by CTaCS. For example, in Panel A, an individual with 0 relative income has taxable income equal to the lower bound of the second federal tax bracket. Each panel shows the average values of the effective marginal tax rates within $250 bins over the range of taxable income spanning $10,000 on either side of the threshold. Each panel also shows the predicted values from an ordinary least squares (OLS) regression of marginal tax rates on a sextic polynomial in taxable income relative to the threshold. Specifically, in each case, the following equation is estimated using OLS.

$$MT{R}_{it}=\alpha +{\displaystyle \sum }_{k=1}^6{\beta }_k{\left(Y_{it}-T_t\right)}^k+\gamma 1\left(Y_{it}-T_t\ge 0\right)+{\epsilon }_{i\mathrm{t,}} \left(1\right)MathType@MTEF@5@5@+=\; feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn\; hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqr1ngB\; PrgifHhDYfgasaacH8YjY=vipgYlh9vqqj=hEeeu0xXdbba9frFj0=\; OqFfea0dXdd9vqai=hGuQ8kuc9pgc9q8qqaq=dir=f0=yqaiVgFr0x\; fr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaa\; aaa8qacaWGnbGaamivaiaadkfapaWaaSbaaSqaa8qacaWGPbGaamiD\; aaWdaeqaaOWdbiabg2da9iabeg7aHjabgUcaRmaawahabeWcpaqaa8\; qacaWGRbGaeyypa0JaaGymaaWdaeaapeGaaGOnaaqdpaqaa8qacqGH\; ris5aaGccqaHYoGypaWaaSbaaSqaa8qacaWGRbaapaqabaGcpeWaae\; Waa8aabaWdbiaadMfapaWaaSbaaSqaa8qacaWGPbGaamiDaaWdaeqa\; aOWdbiabgkHiTiaadsfapaWaaSbaaSqaa8qacaWG0baapaqabaaak8\; qacaGLOaGaayzkaaWdamaaCaaaleqabaWdbiaadUgaaaGccqGHRaWk\; cqaHZoWzcaaIXaWaaeWaa8aabaWdbiaadMfapaWaaSbaaSqaa8qaca\; WGPbGaamiDaaWdaeqaaOWdbiabgkHiTiaadsfapaWaaSbaaSqaa8qa\; caWG0baapaqabaGcpeGaeyyzImRaaGimaaGaayjkaiaawMcaaiabgU\; caRiabew7aL9aadaWgaaWcbaWdbiaadMgacaWG0baapaqabaaaaa@6548@$$

where $MT{R}_{it}$ is the effective marginal tax rate of each individual $i$ in year $t$ calculated using the CTaCS program; $Y_{it}$ is the corresponding level of taxable income; $T_t$ is the threshold for the particular convex kink point under analysis; $1(\cdot )$ is an indicator function that takes the value of “1” if its argument is true and “0” otherwise; and ${\epsilon }_{it}$ is a statistical residual. The parameter $\gamma$ captures the increase in the marginal tax rate at the tax threshold. Using OLS to estimate Equation (1) for each threshold is equivalent to estimating the average changes in effective marginal tax rates in a parametric regression discontinuity (RD) design (Imbens and Lemieux 2007; Lee and Lemieux 2010). As expected, the effective marginal tax rates tend to increase meaningfully at each threshold, even after potential mitigating factors from other dimensions of the tax system have been taken into account.

The charts also show that effective marginal tax rates increase from the second to the third tax bracket and from the third to the fourth tax bracket as a result of tax progressivity. This is consistent with expectations. The rate increases at the recovery tax thresholds for OAS, and EI benefits are comparatively large, since these programs claw back benefits at high rates of up to 15% (OAS) and 30% (EI). Lastly, as expected, a comparison of the two charts suggests that the rate changes are similar in each period.

The OLS regression estimates of the parameter $\gamma$ from Equation (1) are presented in Table 2, by period and separately for each year; the standard errors in this analysis are clustered by individual. The size of each estimate varies over time in the yearly analysis. This likely stems from the relatively small sample sizes in these cases. On balance, the estimates match expectations. For example, marginal tax rates are known to have increased from one bracket to the next by 6% to 7% at the lower bounds of the second federal tax bracket, 4% at the lower bounds of the third and 3% at the lower bounds of the fourth. The RD estimates of the changes in effective marginal tax rates are 6.097% for the second bracket, 3.893% for the third and 2.652% for the fourth.

5.2 Bunching analysis

Graphical inspections of bunching at the tax thresholds are shown in Charts 2-1 and 2-2. Chart 2-1 looks at data from the pre-reform period, and Chart 2-2 looks at the post-reform period. Each panel shows the distribution of taxable income relative to the relevant tax threshold over an interval spanning $10,000 on either side of that threshold. Each dot shows the percentage of individuals in the sample with taxable incomes within a bin of width $250.^{Note 9} In addition, each panel shows the estimated counterfactual distribution based on the procedure of Chetty et al. (2011).^{Note 10} This analysis appears to show that individuals are somewhat responsive to the discontinuous increases in their marginal tax rates in both periods, where spikes in the distributions of taxable income are observed at most tax thresholds.

The distributions of taxable income appear noisier as sample sizes decrease. For example, compare the results at the third federal tax bracket to those at the fourth bracket: the sample size is larger in the former than the latter. The distributions around the provincial and territorial tax thresholds are confounded by the fact that these thresholds vary by province and territory and are fairly close to other convex kink points that could also induce bunching. For this reason, the results of bunching at provincial and territorial tax thresholds should be interpreted a bit more cautiously than the other results. However, the bunching responses are substantial in all cases, despite the potential for these factors to have confounded the analysis.

A comparison of the results across periods reveals that the majority of bunching occurs in the post-reform period. From 2001 to 2006, before pension income splitting was introduced, only a small amount of bunching is observed at most convex kink points. This greatly contrasts with the extent of bunching observed from 2007 to 2012, where the spikes in the distributions are very apparent.

To better quantify the extent of bunching observed in these charts, Table 3 uses the empirical density estimator from Chetty et al. (2011) (see Footnote 8) to estimate the magnitude of excess mass at each tax threshold. These results confirm that bunching is more prevalent from 2007 to 2012 than from 2001 to 2006. For example, the amount of bunching following the reform is 15.8 times larger than it was before pension income splitting was introduced at the lower bound of the second federal tax bracket, 6.4 times larger for the third and 2.4 times larger for the fourth. The results by year are slightly noisier given that the sample sizes are smaller. This results in taxable income distributions that are estimated less precisely. Nevertheless, in these cases, bunching continues to be more prevalent in the post-reform years. This is especially apparent when the results are compared over time at the second and third federal tax brackets, as well as the OAS recovery tax threshold, where sample sizes are still comparatively large.

5.3 The effect of pension income splitting

The previous results show that income responses to taxation changed fundamentally from 2006 to 2007. This coincides with the introduction of pension income splitting. To further assess the extent to which the results are driven by splitting, this section delineates the analysis along several margins of variation that affected individuals’ eligibility to split pension income.

Table 4 repeats the previous analysis across individuals by marital status and whether they, or their spouses if married, are collecting private pension income. The effects of the convex kink points on effective marginal tax rates are reported in Panel A, and the subsequent bunching responses are reported in Panels B. The results show, first, that the effective marginal tax rate changes at each tax threshold are quite uniform across groups. This is not surprising given that the unit of taxation is the individual. The tax thresholds apply to everyone equally irrespective of marital status. Moreover, the tax code treats most types of income identically, so that whether the income derives from private pensions or other sources should not affect these results.^{Note 11}

Despite the uniformity of treatment, the bunching responses vary significantly across groups. Among unmarried individuals, the responses are all small in magnitude and generally insignificant except in a few cases, notably at the OAS and EI thresholds and the lower bound of the third provincial or territorial tax bracket. Similarly, married individuals without a pensioner in the couple, who are unable to split income, exhibit bunching at most thresholds, but the magnitudes of the responses are comparatively small. These results could differ from those of the unmarried non-pensioners for several reasons, including differences in market behaviour responses to taxation or access to other tax planning technologies related to marital status. In addition, the results may simply be more precisely estimated among married individuals given that 72.3% of the individuals in the sample are legally married or in common-law relationships, as shown in Table 1. This analysis also shows that the bunching responses of married individuals in a couple with at least one person receiving private pension income are both relatively large in magnitude and statistically significant at every tax threshold. Being a pensioner appears to be a precursor to responding actively to the changes in marginal tax rates in a manner predicted by the standard model, described earlier. This is consistent with expectations regarding the role of pension income splitting in driving the results.

To further explore the role of splitting, Table 5 conditions on married individuals in the post-reform period and on whether and how the pension income splitting program is actually used. On balance, the effects of the tax rate increases on effective marginal tax rates are quite uniform across groups, as shown in Panel A. The only notable exceptions are among transferees, who receive pension income from their spouses, at the lower bounds of the third provincial and territorial brackets and the fourth federal tax bracket threshold. However, most higher-income earners tend to send income to their spouses rather than receive income from them, and the imprecision of these estimates is partly the result of small sample sizes. However, as Panel B shows, the bunching responses at every tax threshold are almost entirely driven by individuals who utilize the pension income splitting program.