Tax Effects on Portfolio Choice

The study's central focus is examining how taxation influences household portfolio choices. Existing theoretical models suggest that under differential tax systems, households will adjust their portfolios to minimize their tax burden, favoring less heavily taxed assets. However, a major challenge is isolating the pure effect of taxation from the confounding effects of income and wealth, which are often correlated with marginal tax rates. Previous research approaches, such as studying portfolio changes around tax reforms (a 'diff-in-diff' approach), have limitations, including the difficulty in identifying suitable control groups and the potential for delayed adjustments by households to tax changes. The paper proposes a novel approach, capitalizing on the existing differences in the effective marginal tax rates across Canadian households which have the same household income, in order to circumvent some of these issues. This innovative strategy aims to provide more robust estimates of the impact of tax rates on portfolio allocation decisions.

The research leverages a unique feature of Canada's individual income tax system. Unlike joint taxation systems (like that of the United States), Canada's individual tax system allows couples with identical household incomes to experience different effective tax rates on capital income depending on how their labor income is distributed. This difference stems from the ability to attribute capital income to the lower-earning spouse, thereby lowering the overall tax liability. This system thus provides a source of exogenous variation in marginal tax rates that can be used to estimate the effect of taxation on portfolio allocation while holding household income and wealth constant. The paper argues that this unique aspect of the Canadian tax system provides a robust setting to study the relationship between marginal tax rates and portfolio choices, without the confounding effect of differences in overall wealth or income.

The study acknowledges potential challenges to its approach. First, it relies on the assumption that households in individual taxation systems actively shift assets to minimize capital income tax. Evidence supporting this assumption is provided by examining the distribution of capital income before and after the 1988 Canadian tax reform, and by referencing Stephens and Ward-Batts (2004)'s UK study. Second, households with equally distributed labor income might differ in preferences or bargaining power compared to those with unequal income distribution. To address these concerns, a placebo test is implemented using US data (1998 Survey of Consumer Finances (SCF) and the 1999 Panel Study of Income Dynamics (PSID)). Since the US uses joint taxation, any observed correlation between labor income shares and portfolio allocation in US data would indicate preference or bargaining power differences, whereas the lack thereof would strengthen the Canadian study's methodology. The results of the placebo test are crucial for validating the study's assumptions and the causal link inferred between marginal tax rates and portfolio choices.

The paper reviews existing literature on taxation's impact on portfolio allocation, citing Poterba (2001) as a comprehensive overview. The literature indicates relatively few empirical studies focused on this topic, highlighting the challenge in finding substantial and plausibly exogenous variation in tax rates. The authors mention previous studies by Scholz (1994), Samwick (2000), and Poterba and Samwick (2003), who used a 'diff-in-diff' approach to study portfolio allocation around tax reforms. The 'diff-in-diff' approach, while useful, faces its challenges; the common trends assumption may be violated, and there is a tradeoff between the risk of missing delayed adjustments to tax changes and the risk of confounding tax effects with other time effects. This review emphasizes the need for alternative estimation approaches with distinct sources of variation to address the shortcomings of previous research methods and to increase confidence in the findings regarding the effect of taxation on portfolio allocation.

This section investigates whether Canadian couples adjust their asset allocation within the household to minimize tax liabilities. The analysis centers on the 1988 Canadian tax reform, which moved the system away from joint taxation towards individual taxation. The shift created an opportunity for tax savings by attributing capital income to the lower-earning spouse (typically the wife). This section explores this hypothesis by analyzing the distribution of capital income within households before and after the 1988 reform. The findings will determine whether couples exploited the new tax regime by shifting asset ownership to minimize their tax burdens, a critical point for supporting the main study's strategy which uses intra-household income distribution as a proxy for tax rates on capital income. The study highlights the differences between the Canadian and UK tax systems following similar reforms. While the UK allowed couples more flexibility, transfers of ownership in Canada might attract taxes. This aspect of Canadian tax law is important in the analysis of how tax minimization might occur.

To provide further evidence of intra-household asset reallocation for tax purposes, the study incorporates the findings of Crossley and Jeon (2007). Crossley and Jeon examined the impact of the 1988 Canadian tax reform on the labor supply of married women, specifically focusing on low-education women married to high-income husbands. They found a significant increase in labor force participation among this group, suggesting a response to the tax reform. This study expands on Crossley and Jeon's work by examining the effect of the 1988 reform on the reported capital income of low-education married women and their husbands. The expectation is that a decrease in the effective marginal tax rate for these wives (those married to high-earning husbands) would lead to a reallocation of asset ownership to take advantage of tax savings opportunities introduced by the reform. By analyzing data from the Canadian Survey of Consumer Finances (SCF) covering the years 1986 to 1991, the research seeks to find further confirmation of asset reallocation in response to tax incentives.

The results presented in Table 1 show that wives who experienced a significant decrease in their marginal tax rate after the 1988 reform were more likely to report capital income. Furthermore, the magnitude of the increase in reported capital income is substantial. These results support the proposition that Canadian couples, like their UK counterparts (Stephens and Ward-Batts, 2004), reallocate asset ownership to minimize tax liabilities. This finding is crucial because it underpins the core assumption of the main study: that the effective marginal tax rate on capital income often reflects the marginal tax rate of the spouse with lower labor earnings. This means that within couples having the same household labor income, marginal tax rates on capital income vary according to how labor income is distributed. This observed behavior directly supports the main analysis which uses this variation in marginal tax rates to assess the effect of taxation on portfolio choices. Therefore, this section provides compelling evidence for the validity of the study’s methodology.

The primary data source is the 1999 Canadian Survey of Financial Security (SFS), which includes a supplement of 2,000 high-income households. Sample weights are used to ensure the data represents the Canadian population. For comparative analysis and a placebo test, two major US datasets are employed: the 1998 Survey of Consumer Finances (SCF) and the 1999 wealth module of the Panel Study of Income Dynamics (PSID). The SCF, while considered the best source of information on US household finances, has a limitation: detailed income information is not available at the individual level. This necessitates constructing intra-household income shares using data only on wages and salaries. The PSID, a long-running panel survey, provides more comprehensive individual income information, though its portfolio allocation data is less detailed than the SCF. Both US datasets are used to conduct placebo tests under a joint taxation system, providing a crucial benchmark for comparison against the Canadian data analyzed under the individual taxation system.

Assets in the SFS, SCF, and PSID are categorized by their tax characteristics into three groups: heavily taxed assets (interest-bearing assets), moderately taxed assets (stocks and mutual funds), and tax-favored (deferred) assets (RRSPs in Canada, IRAs and Keogh accounts in the US). This categorization considers how different asset classes are treated under the tax systems being studied. The classification acknowledges the complexities of taxation, such as the difference in treatment of dividend payments versus capital gains within stocks. Individual income is defined consistently across datasets as the sum of various sources, while the household head is designated as the major income earner. Sample restrictions are implemented to exclude Quebec residents (due to differences in the Quebec tax system) and self-employed individuals (due to complexities in marginal tax rate calculation). This produces a working sample of 3710 households for the SFS, with comparable restricted samples created from the SCF and PSID data.

The analysis employs two-limit Tobit estimation to model portfolio shares. Tobit regression is used because portfolio shares are bounded between 0 and 1, and many households have shares of particular asset classes at these bounds. The income share of the minor earner is used as an instrumental variable (IV) for the marginal tax rate faced by the household, allowing researchers to address endogeneity concerns. The use of instrumental variables is justified by the argument that the income share of the minor earner is a significant determinant of the effective marginal tax rate on capital income but is not directly related to portfolio decisions other than through its effect on this tax rate. The use of an instrumental variable is necessary because it is believed that the marginal tax rate is endogenous. A placebo test, using similar samples drawn from the SCF and PSID, is conducted to assess the validity of this instrumental variable. This placebo test is essential to ensure that the observed relationships are indeed driven by the differential tax effects, and not other factors such as household preferences or bargaining dynamics.

The primary analysis reveals a statistically significant positive relationship between the intra-household distribution of labor income and the share of tax-favored assets in Canadian households' portfolios. Specifically, households with more equal income shares between spouses (and therefore higher marginal tax rates) tend to hold a larger proportion of their wealth in tax-advantaged investments. This finding supports the hypothesis that households respond to tax incentives by strategically allocating assets to minimize their tax burdens. The magnitude of this effect, while statistically significant, is economically modest: a ten percentage point increase in the marginal tax rate increases the mean portfolio share of tax-favored assets by only two percentage points. This suggests that while tax considerations play a role in household portfolio decisions, other factors like risk aversion or liquidity needs are also likely significant in shaping investment strategies. The observed effect holds even when controlling for household wealth and income.

To validate the findings and rule out alternative explanations, the study conducts placebo tests using US data from the 1998 Survey of Consumer Finances (SCF) and the 1999 Panel Study of Income Dynamics (PSID). These tests are critical because the US employs a joint taxation system, where the effective marginal tax rate on capital income is independent of the intra-household distribution of labor income. Therefore, if a similar relationship between labor income shares and portfolio allocation were observed in the US data, it would suggest that factors other than tax incentives (such as household preferences or bargaining power) are driving the observed portfolio effects in Canada. The results from the US data, as shown in Table 6, show that the coefficient on the income share of the minor earner is not statistically different from zero. This finding strengthens the confidence in the Canadian results, indicating that the observed effect in Canada is genuinely a response to taxation, rather than a reflection of underlying household heterogeneity.

The analysis addresses the potential endogeneity of the marginal tax rate by using the income share of the minor earner as an instrumental variable (IV). This approach is justified because the minor earner's income share affects the household's effective marginal tax rate without directly influencing portfolio choice through other channels. The IV estimates, reported in Table 7, confirm the positive and significant relationship between marginal tax rates and the holding of tax-favored assets in Canada, even after accounting for potential endogeneity. Further robustness checks are conducted by excluding self-employed households and Quebec residents from the sample, due to complexities in calculating marginal tax rates for these groups. The findings remain qualitatively similar, although slightly weaker, in the restricted sample. Additional analysis in the appendix shows that higher net wealth, higher income, and higher education are all associated with higher shares of tax-favored assets, and lower shares of heavily taxed assets. The results are robust and lend further confidence to the primary findings of the study.