In recent decades, there has been growing interest in understanding how taxes affect taxpayer behaviour. Since governments collect most of their revenues directly from individuals, much of this attention has been directed towards understanding the different ways in which personal taxes influence individuals' decisions.
Many researchers have focused on how taxes influence specific outcomes (e.g., labour supply, savings, entrepreneurship, investment in education or skills). Others have studied the impact of taxes on a comprehensive measure of individual behaviour, individual taxable income (i.e., total or gross income minus deductions). This second approach captures the impact of marginal tax rates on both real economic behaviour and efforts to reduce taxes (e.g., claiming deductions or receiving non-taxable forms of compensation).
The impact of changes in tax rates on taxable income is not only of academic interest. Governments in the United States and the United Kingdom, among others, integrate behavioural responses into their assessment of personal income tax changes. For example, when the U.K. increased its top personal income tax rate to 50% from 40% (effective 2010), it took into account that those affected by the measure would reduce their taxable income in response.
This study provides a summary of empirical analysis undertaken by the Department of Finance to examine the impact of recent reductions in marginal tax rates in Canada on individual taxable income. This analysis has several strengths which allow it to contribute to the understanding of the impact of taxes on individual behaviour.
The results of the analysis are broadly consistent with other Canadian studies, providing strong evidence that individuals, especially those with higher incomes, do respond to changes in tax rates. This evidence is relevant both for tax policy analysis and for estimates of the revenue impact of tax policy changes.
This study is organized as follows. First, it presents the concept used to measure tax-related behavioural responses (i.e., the elasticity of taxable income or ETI), a brief review of previous studies and challenges associated with analysis in this area. Second, it describes the estimation strategies in this empirical analysis. Third, it presents the results of this analysis. Technical annexes describe the literature, data and modelling techniques.
The ETI measures the overall response of taxable income to changes in marginal tax rates. An elasticity describes the sensitivity of one economic variable to changes in another. In this context, the ETI refers to the percentage change in taxable income expected to result from a 1% change in the after-tax value of a marginal dollar of taxable income (with this latter measure referred to as the “net of tax rate”).
The response measured by the ETI results from two broad types of decisions made by individuals. Individuals can alter their real economic behaviour and/or adjust their efforts to reduce taxable income.
Real economic behaviour. Tax-induced changes in real economic behaviour can occur when changes in marginal tax rates affect the value of consumption relative to leisure. Individuals can respond through a number of channels to make new choices with respect to how much they work and consume across time. In particular, they can respond by adjusting the hours that they work, their work effort, their choice between paid employment and self-employment/entrepreneurship, their level of savings and their investment in skills/human capital.
Efforts to reduce taxable income. Taxpayers can also make efforts to reduce their taxable income with the aim of minimizing their tax payments, and changes in tax rates may influence this behaviour. Examples include the choice of form of remuneration, such as taxable earnings versus non-taxable fringe benefits, or the use of stock options and their associated deduction; and the use of other deductions, such as flow-through shares and the deduction of investment expenses.
As Chart 1 shows, many researchers have already analyzed the impact of tax rates on individual taxable income, mainly in the U.S. but also in Canada and other countries. Annex 1 reviews specific studies from this literature.
The key results from this literature are as follows:
There are two important challenges that must be addressed in producing credible estimates of the ETI. First, changes over time in income inequality have the potential to bias estimates of the ETI when changes in income inequality resulting from non-tax factors occur in the same time frame as a tax change. For example, if demand for highly skilled labour has increased significantly relative to demand for the labour of other workers, and related increases in the income of these workers coincides in time with tax rate changes, these income increases will be interpreted as resulting from changes in tax rates. A second key challenge results from transitory shifts from year to year in taxpayer income, rather than broad patterns across the income distribution. This occurs when an individual's income shifts significantly following a transitory shock (e.g., the receipt of performance pay or an unemployment spell) and eventually returns to its expected lifetime income profile. If not controlled for, the regression will interpret these non-tax-related shifts in taxable income in the year of the tax change as a tax-related effect, and can therefore lead to serious bias in the ETI estimates. Although early studies in the U.S. literature did not include controls to overcome these two challenges, subsequent studies, including the analysis by the Department of Finance summarized in this study, have paid particular attention to the issue in their estimation strategies.
Another factor to consider in the development of ETI estimates is the effect of income tax base shifting. In Canada, professionals and small business owners can typically choose to incorporate, and can thus be subject to both corporate income tax and personal income tax. Taxable income can shift between the personal income tax base and the corporate income tax base when the relative tax price of the two regimes changes (e.g., owner-operators are more likely to become incorporated when corporate tax rates become relatively more favourable and/or when incorporating makes it possible to defer tax). The U.S. studies examining tax base shifting report a significant amount of shifting following changes in the relative tax price of the personal income tax and corporate income tax regimes. However, most studies in the ETI literature focus on the personal income tax base and do not extend the analysis to include the effects from/on the corporate tax base. This suggests a potential bias in the current ETI estimates if one is interested in the impact of a tax measure on overall government revenues. The analysis prepared by the Department of Finance attenuates this bias by adding controls for when individuals move from one tax regime to another and by conducting sensitivity analysis specifically for those unlikely to shift income between the corporate and personal bases (see Annex 2 for details).
A final key consideration associated with producing credible ETI estimates includes accounting for the effects of changes in external factors on taxable income. These can be exogenous economic shocks affecting labour demand or investment income, but can also be linked to changes to institutional factors, such as changes in efforts by authorities to enforce tax rules and encourage tax compliance. Not accounting for external factors would affect the reliability of the estimates when the effects of changes in external factors on taxable income coincide with tax changes. The analysis prepared by the Department of Finance addresses this issue in two ways. First, it uses a comprehensive list of control variables (see Annex 2 for details). Second, it relies on interprovincial variation in tax rate changes to identify the ETI. Since the Canada Revenue Agency administers federal taxes for all Canadians (and provincial taxes for those living outside Quebec), enforcement is unlikely to vary significantly across provinces.
The empirical component of this study examines how individuals responded to tax policy changes implemented in the late 1990s and early 2000s, when personal income tax rates were reduced substantially at both the federal and provincial levels. The observed differences in the timing and the scope of the provincial tax cuts provide the source of variation permitting the estimation of a taxpayer response.
Focusing on the four largest provinces, Chart 2 shows that there were two periods in the late 1990s and early 2000s in which there was significant variation in changes in marginal tax rates for taxpayers in the top income decile.
First, there was a notable drop in effective marginal tax rates (EMTRs) in Ontario in the second half of the 1990s, while EMTRs in Quebec, Alberta and British Columbia remained relatively unchanged. During this time, the growth in taxable income (excluding capital gains) in Ontario outpaced that of Quebec and British Columbia, with only Alberta incomes matching the growth.
The second variation of interest is the sharp drop in EMTRs across Canada in 2001 and 2002. Although changes to federal tax rates and brackets reduced EMTRs in all provinces, significant reductions in provincial tax rates in British Columbia and Alberta in the same two years offer a significant source of interprovincial variation in EMTRs for possible identification of a taxpayer response. Chart 2 shows that following the tax cuts, taxable income growth in British Columbia and Alberta surpassed that in each of Ontario and Quebec.
For the econometric analysis, two estimation strategies are employed in this study to estimate the tax-related behavioural response in Canada. This is made possible by the availability of a comprehensive administrative panel dataset of individual tax records spanning the period from 1994 through 2006.
Applying these two models to the same data and tax reforms offers an opportunity to test both the robustness of each model and the efficiency of controls for exogenous events influencing taxpayer incomes. The two estimation strategies and issues of variable construction are described in Annex 2.
Because of the possibility that the ETI varies across income groups, separate regression analyses were conducted for selected income groups. Although the estimation was not limited only to these income groups, there are four income groups for which results under the two estimation strategies are comparable. They are the top 10% (real taxable income of more than $60,000), the top 5% (real taxable income of more than $80,000), the top 2% (real taxable income of more than $110,000) and the top 1% (real taxable income of more than $150,000). The top 10% of tax filers accounted for about 60% of personal income taxes paid (in 2006) while the top 1% accounted for almost 25%.
Estimation results from the two methodologies are consistent for each income group (Table 1). Using a sample incorporating individuals with real taxable income of about $60,000 per year or more, the estimated elasticity is approximately 0.2. This result implies that a 10% increase/decrease in the after-tax value of the final dollar of taxable income will result in approximately a 2% increase/decrease in the taxable income reported by the taxpayer.
|Elasticity of Taxable Income|
|Panel Data||Aggregate Data|
|Top 10%||$60,000 and over||0.19||0.19|
|Top 5%||$80,000 and over||0.32||0.30|
|Top 2%||$110,000 and over||0.51||0.46|
|Top 1%||$150,000 and over||0.72||0.62|
|All estimates are significant at the 1% level.|
The elasticity of 0.2 is the average degree of responsiveness for a relatively broad sample of taxpayers. This does not mean that the elasticity is 0.2 for all individuals with taxable income in the top 10% of the income distribution. In fact, estimates of the ETI increase substantially further up the taxable income distribution. For example, the responses by individuals in the top 5% of the taxable income distribution (with an elasticity of about 0.3) and the top 1% of the distribution (with an elasticity of 0.62 to 0.72, depending on the estimation strategy used) are substantially greater than the base estimate for the top 10%. These upper-income ETI estimates are consistent with those in the existing literature.
As discussed above, the ETI is a composite measure which reflects adjustments in real economic behaviour and changes in efforts to reduce taxable income. While there is no perfect way to decompose this estimate into the two broad types of individual responses, it is possible to explore the issue by focusing the analysis on specific subgroups that might be expected to differ in the opportunities available to them to adjust their taxable income.
Sensitivity analysis was undertaken, for example, with respect to the primary source of income of taxpayers. It might be expected that those with business or investment income would have greater opportunities to adjust their income than other individuals (e.g., salaried individuals without access to performance pay, stock options, etc.). When business owners and taxpayers who derive the largest share of income from investments, including those with stock options, are removed from the sample, the ETI for the broad income group drops to 0.12 from about 0.2. This result is consistent with the fact that business owners and investors generally have more flexibility to adjust their real economic activities and more opportunities to shelter income from personal income taxes than do most taxpayers.
It should be noted, however, that the estimated response for individuals in the top 1% remains significant even when business owners, investors and those with stock options are removed from the sample. The ETI falls to 0.68 for this restricted sample, using the individual-level panel approach, compared to the 0.72 estimated with the original sample. Further investigation reveals that the primary source of the response for these individuals may be from changes in employment income. The response of these individuals measured by the elasticity of gross employment income relative to the net of tax rate remains strong at about 0.65. While some of this may reflect a shift from non-taxable forms of compensation (e.g., fringe benefits) to taxable compensation, the magnitude of the response suggests that increases in the quantity and quality of labour supplied are playing an important role.
Understanding how changes in tax rates influence the behaviour of taxpayers in both generating and reporting taxable income is important for the evaluation of tax policy. This study has summarized an analysis of how individuals responded to reductions in federal and provincial tax rates in the late 1990s and early 2000s. The ETI for the top 10% of Canadian tax filers is estimated by this study to be about 0.2. This response is in line with estimates in other Canadian studies and is lower than the consensus elasticity of 0.4 in the international literature for a comparable, broad taxpayer group.
Estimates of the behavioural response in the Canadian data, however, increase substantially as the analysis is restricted to higher-income groups. This pattern is consistent with the pattern of estimates for high-income earners in the international literature. The estimated ETI for the top 5% of the taxable income distribution in this study is approximately 0.3; for the top 1% the estimated ETI ranges between approximately 0.6 and 0.7.
The results of the analysis are broadly consistent with other studies, providing strong evidence that Canadians, especially those with higher incomes, do respond to changes in tax rates. This evidence is relevant both for tax policy analysis and for estimates of the revenue impact of tax policy changes.
Chart 1 in the body of the study presents the ETI estimates deriving from major groupings within the ETI literature. The variability in the estimates on view in the chart is attributable to a wide range of elements including (i) modelling assumptions (e.g., the absence of adequate controls for transitory income shocks in early studies such as Lindsey (1987) and Feldstein (1995)), (ii) country-specific factors such as access to tax deductions and enforcement, (iii) the nature of the tax reform and which group of taxpayers were affected, and (iv) the class of taxpayers examined by a study (e.g., studies of chief executive officers only (Eissa and Giertz (2006) produce much higher estimates than those focusing on a broader group of taxpayers).
It is also important to note that the convention in these studies is to remove capital gains from income given their typically volatile realization and their different tax treatment compared to other types of income. Several studies have, however, specifically examined the relationship between capital gains realization and taxes. Auerbach (1988) and Saez, Slemrod and Giertz (forthcoming) provide a detailed review of this literature.
As noted in Chart 1, the U.S. literature has been the most productive since the inception of this line of inquiry, with researchers gradually refining the estimation methodologies employed and the associated estimates. Lindsey (1987) was the first to measure the responsiveness of taxpayers, following the broad-based tax cuts introduced by the Economic Recovery Tax Act of 1981. Using an Internal Revenue Service tax return dataset (1980–1984), he found evidence of a very strong response following the reform, with the ETI varying from 1.6 to 1.8. The methodology that he employed has some serious shortcomings, however, that were relatively common in empirical studies of the time. In particular, the estimation framework treats the reform as a natural experiment, without controlling for other exogenous shocks and income mobility.
Feldstein (1995) and Auten and Carroll (1995) responded by studying the 1986 Tax Reform Act, which significantly reduced marginal tax rates for high-income individuals in the U.S. (from 50% to 28%) and included base-broadening changes. Their approaches advanced the literature by adopting a more robust methodology to control for non-tax factors and by making use of panel data to observe taxpayers across time. Feldstein (1995) estimates the average ETI to be between 1.04 and 1.48 and the ETI for high-income filers to be 3.0, but recognizes that the small number of high-income taxpayers in his sample may not accurately reflect the population of high-income taxpayers and therefore may limit the reliability of his estimates. At the same time, his estimates are consistent with those for high-income earners in Auten and Carroll (1995), who use a similar approach but more data and a more sophisticated model for estimation. In addition to replicating Feldstein's estimates, they examine the robustness of the estimates to alternate modelling specifications. Their preferred specification generates an ETI of 0.74, but they note a sensitivity of their estimates to model selection, as well as likely sample selection issues. The estimates in both of these studies were further compromised by a lack of sufficient controls to handle the aforementioned problem of transitory income shocks.
Auten and Carroll (1999) revisited the impact of the 1986 Tax Reform Act and paid particular attention to modelling individual characteristics and transitory income shocks, further improving upon their earlier work by using a more representative sample. These improvements lowered the ETI estimate significantly—reducing it from 1.1 using an approach similar to Feldstein (1995) to 0.6 in their preferred specification. Carroll (1998) also offers evidence of a lower ETI in his study of the effects of the tax increases of the 1990 and 1993 Omnibus Budget Reconciliation Acts on taxable income. The methodology used is similar to that applied by Auten and Carroll (1999), but makes use of panel data extending from 1989 through 1995 instead of only a short panel straddling the year of the reform. The preferred estimate deriving from this study is 0.4.
Gruber and Saez (2002) also present an estimate of 0.4 by relying on an even larger panel dataset (1979–1990) to study the series of reforms enacted in the 1980s. Their model specification resembles that in the preceding literature except that it allows for additional controls to capture the effects of transitory income shocks. A recent study by Auten, Carroll and Gee (2008) focusing on the 2001 and 2003 tax rate reductions introduced by the Bush administration also estimated an ETI of 0.4. The authors employ a methodology similar to that used by Auten and Carroll (1999).
Saez (2004) introduces an aggregate income share analysis (1960–2000), as an alternative to a panel data model, to estimate the response of the share of income of taxpayers at the high end of the income distribution (the top 1%) to changes in tax rates. The share of income is regressed against the change in the marginal tax rates and a series of time trends. An aggregated approach has the advantage of avoiding estimation problems associated with transitory shocks to individual income, but it loses the advantages associated with some of the socio-demographic detail in individual-level data. Saez's preferred aggregate shares model estimates the ETI for his high-income sample to be approximately 0.6-0.7. Brewer, Saez and Shephard (2008) apply the methodology to the U.K. and find an ETI for high-income earners of between 0.46 and 0.73 for the years from 1978 through 2003.
Sillamaa and Veall (2001) is the only published Canadian study examining the responsiveness of Canadian taxpayers to the 1987–88 reform using panel data. The 1987–88 reform affected marginal tax rates across most of the income spectrum by reducing the number of brackets from 11 to 4 and by significantly broadening the taxable income base (as a result of the elimination or conversion of several deduction items to non-refundable credits). The authors follow the Auten and Carroll (1999) methodology and use Statistics Canada's Longitudinal Administrative Databank, a panel dataset tracking 20% of all tax filers. The analysis compares the 1986 and 1989 taxable incomes of tax filers. The ETI estimate resulting from this study, at 0.25 for working-age individuals, is lower than the broad estimates in the U.S. studies previously cited.
Saez and Veall (2005) apply the aggregate share methodology for estimating the ETI proposed by Saez (2004) to Canadian data for the period from 1920 to 2000. The broader goal of their study is to analyze a surge in the income shares of top income groups in Canada in recent decades, while considering hypotheses as to why the surge in incomes has been almost as large in Canada as in the United States despite a more modest drop in effective marginal tax rates in Canada. Controlling for U.S. income growth, their preferred estimate for the ETI associated with the top 1% of Canadian earners is 0.17 using data from 1972 to 2000.
More than a dozen additional studies were considered in preparing this review, including seven examining taxpayer responses to major tax reforms in Scandinavian countries and the remaining studies examining taxpayer behaviour in France, Hungary, Germany, Japan and the U.K. These studies generally show a behavioural response of around 0.4, with the lower estimates attributed to the more egalitarian European countries (i.e. France and the Scandinavian countries).
This study uses a micro-level taxpayer database made available by the Canada Revenue Agency. The longitudinal panel was created using a representative sample of the universe data that is equivalent to 10% of the universe database. The panel data span from 1994 to 2006 and contain both tax and socioeconomic information pertaining to taxpayers (comprising 20 million observations over 13 years).
The EMTRs are calculated for individuals in the sample using the Department of Finance's T1 microsimulation model. These rates include federal and provincial income tax parameters, as well as parameters for family benefits (such as the Canada Child Tax Benefit and the National Child Benefit supplement) and for employee contributions to Employment Insurance (EI) and the Canada/Quebec Pension Plan (CPP/QPP). Allowable spousal deduction/credit transfers are also accounted for in the EMTRs. Employers' contributions to EI and the CPP/QPP as well as provincial payroll taxes are, however, not included. Although a case could be made to include these premiums in our analysis, as they may be shifted by employers to low- to average-skilled employees through wage reductions, data limitations prevent us from including them at this time.
Two adjustments were made to the standard definition of taxable income. First, the definition of taxable income used in this study follows the convention in the international literature by excluding capital gains from income. Realized capital gains are subject to special tax treatment and their volatility amplifies the empirical challenges associated with transitory income shifts. The second adjustment relates to the definition of taxable income. It is standard practice to adopt a constant definition of taxable income across years to estimate the ETI as, in the absence of the adjustment, the effects of legislative changes to deductions or exemptions on taxable income would contaminate the estimates of responses to changes in marginal tax rates. The adjustment applied in this study consists of using the 1996 definition of taxable income, with the main items adjusted being Registered Retirement Savings Plan limits, child care deduction limits, the dividend gross-up rate and related credit levels, the deduction for workers in Quebec, and the inclusion percentage associated with the deduction for stock options.
The first estimation approach applied in this study uses panel data and the benchmark model in the literature (similar to the Gruber and Saez (2002) model described in the literature review) to estimate the ETI from variation in EMTRs across provinces and over time. It is defined as follows:
|log TI i,t / TI i,t-1 = α + ε log [(1 - τ i,t) / (1 - τ i,t-1)] +||N||M|
|βmZmi,t + ui,t||(1)|
Subscripts i and t represent the individual and the period in which the new tax parameters applied. TI reprents taxable income, t is the EMTR, R encompasses a number of controls for transitory income shifts in taxable income, and Z represents a set of other determinants of the change in taxable income. The constant is denoted by a and ui,t is the error term. Regressions are weighted by taxable income to ensure proper representation of individuals in the overall tax base.
Two types of controls for transitory income shifts have been retained. The first asserts a non-linear relationship between the change in taxable income and the log of lagged income. It is meant to capture the general pattern of transitory income shifts in the data. For instance, individuals with abnormally high income are more likely to revert back to their life-cycle income path in the following year, and conversely, those with very low income have a greater likelihood of moving up the income distribution. The non-linearity is modelled by a 10-piece spline function, with knots placed at the decile thresholds of the log of lagged income. The 10-piece spline function has become a relatively standard control for transitory income shifts in the ETI literature.
Modelling the general relationship between the change in income and prior-year income using a 10-piece spline may not, however, capture very large shifts in income in the data. Such atypical changes in income will contaminate estimates of tax-related behavioural response if the timing of these shifts in income coincides with a tax change. The associated bias would also most likely be amplified by the fact that some of these individuals at the top of the income distribution have large enough weights in the sample to significantly influence the overall estimates. To avoid the problem, the model also includes dummy variables to identify substantial transitory deviations in income from more predictable paths. We define these deviations as the ratio of the individual's income in year t to the individual's median real income, calculated over the years in which the individual is in the sample (hereafter defined as TIt/M). Six groups are earmarked as experiencing large deviations in income according to this measure. They represent fractiles of the top and bottom 5% of the distribution of TIt/M. The top 5% (less the top 1%), 1% (less the top 0.5%), and the top 0.5% are groups that include individuals withTIt/Mgreater than 1.5, 2.2 and 3.5 respectively, and the lowest 5% (less the bottom 1%), 1% (less the bottom 0.5%), and the bottom 0.5% include individuals withTIt/Mless than 0.72, 0.50 and 0.45 respectively. In other words, as an example, a dummy variable for the top 5% (less the top 1%) is used to identify individuals with income in year t that is greater than 1.5 times their median income and less than 2.2 times their median income.
The character Z in the model represents a number of other determinants of the change in taxable income. These include gender, age, the presence of children under 18 in the home, marital status, filing method, means of producing income (defined as the major source of income), and year-specific provincial effects.
Contrary to the usual approach to measuring the effect of marital status, in which a dichotomous variable indicates whether an individual has a spouse, a number of dummy variables identifying both a change in marital status and the base state (whether or not married) are included in the model. The rationale for this addition is that a change in marital status may significantly alter taxable income for some individuals. A newly separated or widowed individual, for example, might be likely to increase his or her hours of work or to join the labour force if he or she was previously a homemaker. A newly separated individual could also adjust his or her work decisions to help finance spousal or child support payments, although at the same time such support payments could reduce taxable income (as they may be deductible). To capture some of these effects, included in the model are variables identifying individuals living with a spouse for more than a year, individuals who were married or who began living in a common-law situation during the last year, and new singles (separated, divorced or widowed).
Dummy variables representing an individual's means of generating income identify pensioners, investors, and unincorporated and incorporated business owners. The first three types are identified by the income category providing more of their income than any other type. Incorporated business owners are identified using data on EI contributions. That is, business owners with more than 40% of the interest in a company do not pay EI premiums on their remuneration. Three dummies are constructed for incorporated business owners: those who have been incorporated for more than two years; owners of newly incorporated businesses; and owners moving out of incorporation for any reason (selling of the business, bankruptcy, choosing unincorporated business activities, etc.). The dummy variables identifying primary means of generating income are mutually exclusive.
Year-specific provincial effects were also included in the model if a province's aggregate taxable income deviated significantly from the national trend. These province-year dummies signal taxable income greater than the average national growth rate plus or minus twice the average absolute deviation of the provincial growth rates. A dummy was also added for the year 2000 to capture the sharp increase in taxable income as a result of the high-tech bubble.
In a progressive tax system, a positive income shock will push an individual's income into a higher tax bracket, resulting in a positive correlation between the EMTR (t) and taxable income (TI). This endogeneity of the tax price is dealt with using an instrumental variable technique (two-stage least squares), in which the instrument is constructed by calculating a synthetic EMTR in year t. This is done by growing taxable income in year t-1 by the overall average growth rate of taxable income from t-1 to t, and then computing the EMTR the individual would have faced on that imputed income. The instrument then becomes the difference (in logs) of the synthetic tax price in year t and the actual tax price in year t-1.
The second estimation strategy employed uses an aggregate income share model that exploits the same variation in EMTRs—both longitudinal and across provincial jurisdictions—to estimate a tax-related behavioural response.
The response for each specific income group is estimated using the following equation:
|log S p,t = ω + ε log (1 - τ p,t)+||X||R|
|σrpZrp,t + up,t|
The dependent variable, s, represents the total taxable income of a group as a share of total taxable income in year t and province p, and the explanatory variables include the (average, income-weighted) log of the net of tax rate (1-t), as well as a number of national and province-specific macroeconomic controls (v and Z). Finally, ω is the constant and up,t represents the error term.
The modelling approach adopted is similar to that employed by Slemrod (1996) and Saez (2004) and discussed in Saez, Slemrod and Giertz (forthcoming). Cross-sectional, time-series shares of taxable income and net of tax rates for segments of the income distribution comprising moderate, high and very high income earners are pooled to form the analysis dataset for each regression. The major divergence in the modelling approach of this study from the approaches taken in the aforementioned literature is that the richness of the available administrative data permits, as proposed by Saez, Slemrod and Giertz (forthcoming), the extension of the model to subnational jurisdictions. The sample sizes in the panel data being used are sufficient to increase the degrees of freedom in the analysis and to exploit provincial variation in the timing of combined federal and provincial tax changes. The addition of this further source of variation also permits the incorporation of controls for non-tax-related outcomes, such as the collapse of the high-tech bubble and increasing income inequality, without destroying identification.
The approach employed in this literature has been to run the regressions as the log of the income share of the fractile in question on its (average, income-weighted) net of tax rate. Such regressions control automatically for average income growth, with the addition of time trends controlling for a divergence in growth in income from the average for a specific fractile. This study therefore follows the convention of regressing the share of income accruing to a particular income fractile on the (average, income-weighted) net of tax rate for that group, with income shares and net of tax rates calculated by province for each fractile. The regressions are weighted by total fractile income, by province.
A number of variations were considered in making the final selection of independent variables in the regressions, with an effort made to provide for comparability across fractiles and to aim for parsimony in model selection. The initial analysis undertaken used federal-level taxable income thresholds to define fractile income shares by province. For the broader groups such as the top decile, or those with moderate or higher income, the dispersion of income growth across provinces in the later years in the series meant that a large proportion of the samples for provinces with slower earnings growth were being dropped. As a result, the preferred model using this data saw shares defined using provincial thresholds of taxable income per fractile. General time dummies to control for broad changes in income by fractile across time were also included in the final regression models. An oil price dummy was added to these regressions for Alberta, specifically, to control for explosive growth in real income in that province in the latter years of the data.
A further note on time trends and dummies is required here. A range of model specifications, such as examples inclusive of aggregate quadratic and cubic time trends, were considered at various stages in the analysis. The analysis was also extended to include specifications with provincial linear, quadratic and cubic time trends. The results were broadly robust to a wide range of specifications, permitting the selection of the preferred model comprising aggregate time dummies, an Alberta oil price dummy, province dummy variables to control for province-specific effects, and a macroeconomic variable specific to each province (the province-specific employment rate). The inclusion of a real gross domestic product (GDP) variable in addition to the employment rate variable for each province was considered, but given the very high correlation between the two series there was no additional gain to including provincial real GDP in the final regressions.
|Top Marginal Rates (%)|
|Newfoundland and Labrador||53.3||53.3||53.3||53.7||51.3||48.6||48.6||48.6||48.6||48.6||48.6|
|Prince Edward Island||50.3||50.3||50.3||49.1||48.8||47.4||47.4||47.4||47.4||47.4||47.4|
|Year-to-Year Percentage Change|
|Newfoundland and Labrador||3.9||0.0||0.0||0.6||-4.4||-5.2||0.0||0.0||0.0||0.0||0.0|
|Prince Edward Island||0.0||0.0||0.0||-2.4||-0.6||-2.9||0.0||0.0||0.0||0.0||0.0|
|1 The federal-provincial top marginal rate in Quebec includes the impact of the Quebec Abatement.
2 The federal-provincial top marginal rate in Ontario excludes the impact of the Ontario Health Premium.
Aarbu, K. and T. Thoresen (2001). “Income Responses to Tax Changes: Evidence From the Norwegian Tax Reform.” National Tax Journal, 54 (2): 319-334.
Auerbach, A. J. (1988). “Capital Gains Taxation in the United States: Realizations, Revenue, and Rhetoric.” Brookings Papers on Economic Activity, 19 (2): 595-631.
Auten, G. and R. Carroll (1995). “Behavior of the Affluent and the 1986 Tax Reform Act.” Proceedings of the 87th Annual Conference on Taxation of the National Tax Association, Columbus, Ohio: 70-76.
Auten, G. and R. Carroll (1999). “The Effect of Income Taxes on Household Income.” The Review of Economics and Statistics, 81 (4): 681-693.
Auten, G., R. Carroll and G. Gee (2008). “The 2001 and 2003 Tax Rate Reductions: An Overview and Estimate of the Taxable Income Response.” National Tax Journal, 61 (3): 345-364.
Bakos, P., P. Benczúr and D. Benedek (2008). “The Elasticity of Taxable Income: Estimates and Flat Tax Predictions Using the Hungarian Tax Changes in 2005.” European University Institute Working Paper RSCAS 2008/32.
Bianchi, M., B. Gudmundsson and G. Zoega (2001). “Iceland's Natural Experiment in Supply-Side Economics.” American Economic Review, 91 (December): 1564-1579.
Blomquist, S. and H. Selin (2009). “Hourly Wage Rate and Taxable Labor Income Responsiveness to Changes in Marginal Tax Rates.” CESifo Working Paper No. 2644.
Brewer, M., E. Saez and A. Shephard (2008). “Means-Testing and Tax Rates on Earnings.” Institute for Fiscal Studies Working Paper, Prepared for the Report of a Commission on Reforming the Tax System for the 21st Century, Chaired by Sir James Mirrlees.
Brewer, M. and J. Browne (2009). “Can More Revenue Be Raised by Increasing Income Tax Rates for the Very Rich?” Institute for Fiscal Studies Briefing Note BN84.
Carroll, R. (1998). “Tax Rates, Taxpayer Behavior, and the 1993 Tax Act.” U.S. Department of the Treasury, Office of Tax Analysis Working Paper 79.
Eissa, N. and S. Giertz (2006). “Trends in High Incomes and Behavioral Responses to Taxation: Evidence From Executive Compensation and Statistics of Income Data.” U.S. Congressional Budget Office Working Paper 2006-14.
Feldstein, M. (1995). “The Effect of Marginal Tax Rates on Taxable Income: A Panel Study of the 1986 Tax Reform Act.” Journal of Political Economy, 103 (3): 551-572.
Gagné, R., J-F. Nadeau and F. Vaillancourt (2000). “Taxpayers' Response to Tax Rate Changes: A Canadian Panel Study.” CIRANO Working Paper 2000s-59.
Giertz, S. (2004). “Recent Literature on Taxable-Income Elasticities.” U.S. Congressional Budget Office Technical Paper 2004-16.
Giertz, S. (2008). “Taxable Income Responses to 1990s Tax Acts: Further Explorations.” U.S. Congressional Budget Office Working Paper 2008-08.
Goolsbee, A. (2000). “What Happens When You Tax the Rich? Evidence From Executive Compensation.” Journal of Political Economy, 108 (2): 352-378.
Gordon, R. and J. MacKie-Mason (1994). “Tax Distortions to the Choice of Organizational Form.” Journal of Public Economics, 55 (2): 279-306.
Gordon, R. and J. Slemrod (2000). “Are ‘Real' Responses to Taxes Simply Income Shifting Between Corporate and Personal Tax Bases?” J. Slemrod, ed., Does Atlas Shrug? The Economic Consequences of Taxing the Rich, Harvard University Press and the Russell Sage Foundation.
Gottfried, P. and H. Schellhorn (2004). “Empirical Evidence on the Effects of Marginal Tax Rates on Income—The German Case.” IAW-Diskussionspapiere 15.
Gruber, J. and E. Saez (2002). “The Elasticity of Taxable Income: Evidence and Implications.” Journal of Public Economics, 84: 1-32.
Hansson, A. (2007). “Taxpayers' Responsiveness to Tax Rate Changes and Implications for the Cost of Taxation in Sweden.” International Tax and Public Finance, 14 (5): 563-582.
Holmlund, B. and M. Söderström (2008). “Estimating Dynamic Income Responses to Tax Reforms: Swedish Evidence.” Institute for Labour Market Policy Evaluation Working Paper 2008:28.
Kopczuk, W. (2005). “Tax Bases, Tax Rates and the Elasticity of Reported Income.” Journal of Public Economics, 89 (11-12): 2093-2119.
Lindsey, L. (1987). “Individual Taxpayer Response to Tax Cuts: 1982-1984, With Implications for the Revenue Maximizing Tax Rate.” Journal of Public Economics, 33: 173-206.
Ljunge, M. and K. Ragan (2005). “Labor Supply and the Tax Reform of the Century.” University of Chicago Working Paper.
Long, J. (1999). “The Impact of Marginal Tax Rates on Taxable Income: Evidence From State Income Tax Differentials.” Southern Economic Journal, 65 (4): 855-869.
MacKie-Mason, J. and R. Gordon (1997). “How Much Do Taxes Discourage Incorporation?” The Journal of Finance, 52 (2): 477-505.
Meghir, C and D. Phillips (2008). “Labour Supply and Taxes.” Institute for Fiscal Studies Working Paper W08/04.
Moffitt, R. and M. Wilhelm (2000). “Taxation and the Labor Supply Decisions of the Affluent.” J. Slemrod, ed., Does Atlas Shrug? The Economic Consequences of Taxing the Rich, Harvard University Press and the Russell Sage Foundation.
Moriguchi, C. (2010). “Top Wage Incomes in Japan, 1951-2005.” Journal of the Japanese and International Economies, 24:301-333.
Piketty, T. (1999). “Les hauts revenus face aux modifications des taux marginaux supérieurs de l'impôt sur le revenu en France, 1970-1996.” Économie & Prévision, 138-139: 25-60.
Saez, E. (2003). “The Effect of Marginal Tax Rates on Income: A Panel Study of ‘Bracket Creep.'” Journal of Public Economics, 87: 1231-1258.
Saez, E. (2004). “Reported Incomes and Marginal Tax Rates, 1960-2000: Evidence and Policy Implications.” James Poterba, ed., Tax Policy and the Economy, 18: 117-174.
Saez, E. and M. Veall (2005). “The Evolution of High Incomes in Northern America: Lessons From Canadian Evidence.” The American Economic Review, 95 (3): 831-849
Saez, E., J. Slemrod and S. Giertz (forthcoming). “The Elasticity of Taxable Income With Respect to Marginal Tax Rates: A Critical Review.” Journal of Economic Literature.
Selén, J. (2002). “Taxable Income Responses to Tax Changes: A Panel Analysis of the 1990/91 Swedish Reform.” FIEF Working Paper No. 177.
Sillamaa, M. A. and M. Veall (2001). “The Effect of Marginal Tax Rates on Taxable Income: A Panel Study of the 1988 Tax Flattening in Canada.” Journal of Public Economics, 80 (3): 341-356.
Slemrod, J. (1996). “High-Income Families and the Tax Changes of the 1980s: The Anatomy of Behavioral Response.” M. Feldstein and J. Poterba, cds., Empirial Foundations of Houseold Taxation, University of Chicago Press and the National Bureau of Economic Research.
Slemrod, J. (1998). “Methodological Issues in Measuring and Interpreting Taxable Income Elasticities.” National Tax Journal, 51 (4): 773-788.
Triest, R. K. (1998). “Econometric Issues in Estimating the Behavioral Response to Taxation: A Nontechnical Introduction.” National Tax Journal, 1 (4): 761-772.
 See Brewer and Browne (2009) for background on this U.K. experience.
 In some instances, the ETI may not fully reflect the long-term impact of changes in real economic behaviour. For example, although lower marginal tax rates may encourage investment in human capital and generate higher earnings as a result, the short-term response may be a reduction in taxable income as individuals acquire new skills (e.g., leave the labour market to attend school). A similar analogy can also be made when an individual moves out of the paid workforce and faces adjustment costs associated with starting a business. These two examples would have the effect of underestimating the long-term impact of changes in real economic behaviour.
 Note, however, that efforts to reduce taxable income can also have real economic effects. For example, investments in flow-through shares can affect the allocation of capital across sectors of the economy. Thus, the two broad types of decisions affected by marginal tax rates should not be assumed to be completely independent of one another.
 Triest (1998), Slemrod (1998), Giertz (2004) and Saez, Slemrod and Giertz (forthcoming) provide interesting viewpoints on the empirical challenges surrounding the estimation of ETIs.
 Lindsey (1987) and Feldstein (1995) (the very high ETI values at the left of Chart 1).
 Annex 2 provides greater detail as to how the empirical analysis summarized in this study addresses these issues.
 See Gordon and MacKie-Mason (1994), MacKie-Mason and Gordon (1997) and Gordon and Slemrod (2000).
 The empirical analysis presented in the following section covers all 10 provinces; Table A2.1 in Annex 2 provides top marginal federal-provincial tax rates for all provinces during the survey period, as well as their year-to-year changes.
 Effective marginal tax rates include federal and provincial income tax parameters, as well as parameters for family benefits and payroll taxes. A detailed description of the calculation is provided in Annex 2.
 Chart 2 presents EMTRs and taxable income shares for the top 10% of the taxable income distributions in the four largest provinces for the period 1995–2006. Both EMTRs and taxable income shares are normalized, meaning that their values in 1995 are set equal to one. This normalization is done to facilitate the interpretation of the trends in each of these two variables.
 See Annex 1 for details on Gruber and Saez (2002).
 All reported real taxable income levels are in 2006 dollars. Taxable income refers to taxable income less capital gains in all of the estimation results.
 This estimate is, however, highly sensitive to assumptions about the variables included in the empirical model and the income groups included in the sample.
 A study of income mobility in the data was completed in the early stages of the current study, but the results are not published here.
 The estimates in this study were not sensitive to the inclusion of higher order splines in which additional knots were defined at the very top of the income distribution (e.g., the top 5%, 1%, 0.1%, etc.).
 The use of a general trend per fractile is supported by the argument that within these groups of relatively skilled workers, especially at very high levels of earnings, similar broad economic trends will have been influential on earnings.