New evidence on taxes and portfolio choice
Introduction
The effect of taxation on household portfolio choice has long been an important question facing researchers and policy makers. Theoretical models predict that under a differential taxation system, households' portfolio allocation decisions are based not only on the risk–return relationship of assets but also on their tax characteristics. Given risk and return characteristics, households should minimize tax burden by tilting their portfolios toward less heavily taxed assets. Moreover, less taxed households face a smaller incentive to invest in tax-favored assets. For these theoretical predictions to be a useful guide to tax policy, empirical evidence is required that confirms the qualitative predictions of theory, and which provides quantitative estimates of the magnitudes of relevant effects.
The literature on tax avoidance (Slemrod and Yitzhaki, 2002) suggests that tax elasticities may be much larger for aspects of financial arrangements (for example, the timing of income) than for real variables (such as labor supply or savings levels.) Portfolio choice may be in the former category. Nevertheless, there have been relatively few empirical studies of the effect of taxation on portfolio allocation. The existing literature is well surveyed by Poterba (2002a). The key challenge in this literature is to find a substantial and plausibly exogenous source of variation in tax rates. The contribution of this paper is to derive new estimates of tax effects on portfolio allocation using a novel source of variation in tax rates.
Cross-sectional differences in the marginal tax rates (MTRs) generate variation across households in the relative after-tax returns of different assets classes. Examples of this kind of study include Feldstein, 1976, Hubbard, 1985, King and Leape, 1998 and Poterba and Samwick (2003). These studies typically find a strong effect of taxes on asset allocation. A possible problem with this approach however, is that household MTRs are highly correlated with income, and so it is difficult to disentangle a pure tax effect from income or wealth effects on portfolio allocation.
Another strategy, pursued first by Scholz (1994), and then Samwick (2000) is to study changes in portfolio allocation around tax reforms (using a “diff-in-diff” approach.) Scholz (1994) finds much less evidence of a tax effect on household portfolios. The great advantage of this approach is that the tax reform generates variation in MTRs within income groups. Moreover, if a suitable control group can be identified, it is possible to control for general time effects, under a “common trends” assumption. However, the common trends assumption may also be a weakness of this approach. It may be difficult to identify, with any confidence, control groups that were unaffected by the tax reform, but experience time effects that are similar to the time effects experienced by those that were affected by the reform. Moreover, the investigator faces a difficult tradeoff in deciding the interval over which the data should be “differenced”. Households may not rebalance portfolios instantaneously, because of transaction costs or other sources of inertia. Thus a strategy that compares periods just before and after the reform risks missing delayed adjustments to changes in taxation (Gordon, 1994, Poterba and Samwick, 2003.) On the other hand, a strategy of comparing data from long before a tax reform with data well after the tax reform rests more heavily on the common trends assumption and so suffers from a greater risk of confounding a tax effect with other time effects.
It is well known that the last few decades witnessed significant trends in household portfolio allocation. Prior to 1980s most households' financial wealth was held in simple forms (mostly in liquid and safe assets) in most industrialized countries. This has changed considerably since 1990s; now a large proportion of households in these countries hold significantly more sophisticated portfolios. Financial liberalization, declining information costs, attraction of employer-sponsored retirement accounts (such as 401(k)s in the U.S) and introduction of tax advantaged investment tools (such as registered education saving accounts in Canada) are among the explanations offered for this trend.2 The large size of time effects in portfolio allocations makes the common trends assumption particularly worrying.
In summary, the literature contains estimates based on cross-sectional variation in tax rates and estimates based on variation generated by tax reforms. The former suggest much more significant tax effects than the latter, but plausible concerns can be raised about both sources of variation. In light of this, it would be useful to have additional estimates of tax effects on portfolio allocation, based on alternative sources of variation in tax rates. The strategy proposed in this paper is to exploit variation in MTRs across households with the same total earnings, which arise in progressive income tax systems with individual taxation. In jurisdictions with individual taxation, such as Canada, two households with the same total earnings, but divided differently between the principal and secondary earner, may face a different MTR on the first dollar of household capital income. In particular, households in which most of the labor income is earned by one individual will face a lower MTR on the first dollar of capital income than a household with fairly equal income shares. This is because the former household can attribute capital income to the household member with lower labor earnings. This advantage does not exist in systems of joint taxation, as in the United States. Thus in Canada (and other countries with individual taxation) it is possible to study the effect of MTR on portfolio allocation in a cross-section, while holding constant household income and wealth. We use the 1999 Canadian Survey of Financial Security to implement this empirical strategy.
Of course, there are a number of important challenges to the validity of this strategy. We will address these empirically in our analysis. First, this strategy rests on the assumption that in systems with individual taxation, households shift financial assets to the secondary earner (or lower earning spouse) in order to minimize the taxation of capital income. Stephens and Ward-Batts (2004) report evidence in support of this proposition from a study of the 1990 change from joint to individual taxation in the U.K. Below we provide further evidence supporting this assumption, by studying the distribution of capital income within households before and after the Canadian tax reform of 1988. For many households, that tax reform had the effect of making the Canadian tax system less joint.
Second, households in which labor earnings are fairly equally contributed may differ in important ways (in addition to effective MTR) from households that have similar total labor income and wealth, but greater inequality in labor income shares. For example, households with two labor incomes of fifty thousand dollars annually may have different preferences (including risk tolerance) than a household with a single labor income of one hundred thousand dollars per year. Alternatively, intra-household decision making may proceed quite differently in these two households.3
To address this second set of concerns, we implement a “placebo test”. In particular, we study the relationship between labor income shares and portfolio allocation (holding wealth and total income constant) in two U.S. data sets: the 1998 Survey of Consumer of Finances (SCF) and the 1999 wealth module of the Panel Study of Income Dynamics (PSID). Because the U.S. has joint taxation, the first dollar MTR on capital income is unaffected by the distribution of labor income within the household. Thus a correlation between labor income shares and portfolio allocation in these data would suggest important heterogeneity in preferences or household bargaining, while the absence of such a correlation would support our empirical strategy for identifying tax effects.4
Veall (2001) employs Canadian micro data to estimate the effect of taxes on households' contributions to tax-favored retirement savings accounts (Registered Retirement Savings Plans, — RRSPs). Milligan (2002) studies the effect of taxes on RRSP participation, again with Canadian micro data. Note that RRSP contributions and participation reflect decisions both about the level of saving and about portfolio allocation. Neither author exploits the identification strategy that we propose but rather they employ a more traditional approach based on temporal and provincial variation in tax rates (following the work on U.S. tax reforms cited above.) Veall examines the Canadian tax reform of 1988. He finds a negative relationship (though not statistically different from zero in all specifications) between RRSP contributions and marginal tax rates — contradicting the prediction that less taxed households face a smaller incentive to invest in tax-favored assets. In contrast, Milligan, who looks at participation (rather than contributions conditional on participation) and uses a combination of temporal and cross-province variation in tax rates, finds that a 10 percentage point increase in the marginal tax rate increases the participation probability by eight percent. Milligan notes that a potential explanation for Veall's finding is that tax changes in the period he studied may be overwhelmed by general trends in RRSP behavior. This is one example of the kind of concern with the traditional identification strategy which we outlined above. In contrast, our proposed identification strategy does not employ temporal variation. To the best of our knowledge, our strategy has not been previously employed (neither in Canada nor other jurisdictions.)
A preview of our main results is as follows. We first find that Canadian households do shift capital income within the household to take advantage of the system of individual taxation. Thus means that households with the same total income but distributed differently between spouses will face different MTRs on capital income. Turning to the main part of our analysis, we find some evidence that Canadian households with more equal income shares (and hence a higher MTRs) tilt their portfolios towards taxed-favored assets (holding wealth and total income constant.) We only find this robustly for households in the top half of the income distribution. We find no evidence of this effect in the SCF or PSID, suggesting that the small effect we observe in the Canadian data is a true tax response, and not attributable to heterogeneity in preferences or intra-household bargaining. Our benchmark specification suggests that among more affluent households and at the means of the data, a 10 percentage point increase in marginal tax rates increases the portfolio share of taxed-favored assets by 1.7 percentage points, and decreases portfolio share of moderately taxed assets by 1.3 percentage points. These are small effects.
Our evidence that Canadian couples allocate financial assets among partners in order to minimize tax liabilities is presented in the next Section. We then turn to the effect of marginal tax rates on portfolio choice. Section 3 elaborates on our data and methods, and results are presented in Section 4. Section 5 concludes.
Section snippets
The allocation of investment income within households
Variation in the distribution of labor income within households with the same total income generates variation in the effective MTR on capital income if households allocate capital income across household members in order to minimize their tax liability. Typically this would mean having the partner with lower labor income hold financial assets which generate taxable income. Stephens and Ward-Batts (2004) report evidence that U.K. households follow such a strategy. Their study is based on the
Data
Our main estimates are based on master files from the Canadian Survey of Financial Security, SFS. This survey involved personal interviews in May and June of 1999. The sample includes a supplement of 2000 households selected from geographical areas with a larger concentration of high-income households. Sample weights provided by the survey are used to make the data representative of the Canadian population as a whole.
The SFS (1999) individual files contain information on labor income of all
Reduced form estimates
The left-hand column of Table 4 presents estimates of Eq. (1), for all three asset categories in the (Canadian) Survey of Financial Security. Coefficients on the income share (αk) are reported in the first row. There is some evidence of a reduction in holdings of moderately taxed assets for households in which income share of the minor earner is higher, though the effects on other tax categories are not precisely estimated. The results indicate that higher levels of wealth and income are
Conclusion
Identifying the effect of taxation on household portfolio allocation requires plausibly exogenous variation in marginal tax rates. In progressive tax systems, marginal tax rates vary with income levels, but income, or wealth, almost surely affects portfolio choice directly. In systems of individual taxation – like Canada's – couples with the same level of household income (and wealth) can face different effective tax rates on capital income if labor income is distributed differently within
Acknowledgments
We gratefully acknowledge the support of the Social and Economic Dimensions of an Aging Population (SEDAP) Research Program at McMaster University, the Social Sciences and Humanities Research Council of Canada (SSHRC, Grant # 410-2006-1296), the Australian Research Council (Grant # DP0772731) and the Centre for Financial Analysis and Policy (CFAP) at the University of Cambridge. For helpful comments and suggestions we thank Michael Veall, Kevin Milligan, seminar participants at the University
References (26)
Saving and the intra-household distribution of income: an empirical investigation
Richerche Economiche
(1995)- et al.
Wealth and portfolio composition: theory and evidence
Journal of Public Economics
(1998) Taxation, risk taking and household portfolio behavior
- et al.
Taxation and household portfolio composition: US evidence from the 1980s and 1990s
Journal of Public Economic
(2003) - et al.
Portfolio choice and health status
Journal of Financial Economics
(2004) - et al.
Tax avoidance, evasion and administration
- et al.
The impact of separate taxation on the intra-household allocation of assets: evidence from the UK
Journal of Public Economics
(2004) - et al.
Mostly Harmless Econometrics: An Empiricists' Companion
(2008) - et al.
Asset Allocation and Asset Location: Household Evidence from the Survey of Consumer Finances
Journal of Public Economics
(2004) Stockholding behavior of U.S. households: evidence from the 1983–1989 survey of consumer finances
The Review of Economics and Statistics
(1998)
The saving behaviour of a two-person household
Scandinavian Journal of Economics
Joint taxation and the labor supply of married women: evidence from the Canadian Tax Reform of 1988
Fiscal Studies
Cited by (26)
Individuals’ financial risk-taking and peer influence
2022, Quarterly Review of Economics and FinanceTaxation and the external wealth of nations: Evidence from bilateral portfolio holdings
2022, Journal of International Money and FinanceCitation Excerpt :The estimated tax elasticities of portfolio shares in our paper (which range from 0.114 to 0.155 for the entire sample – they are higher for the tax changes triggered by JGTRRA) are at the lower end of the range found in the literature. For example, in the survey of Poterba (2002) tax elasticities of 0.4 to 0.48 are reported based on cross-sectional variation, whereas Alan et al. (2010) estimate a tax elasticity of 0.17. The lower estimates may reflect that our results are net of clientele effects9 as well as that, due to frictions such as trading costs, foreign portfolio holdings are less sensitive to tax changes.
Tax planning of married couples and intra-household income inequality
2019, Journal of Public EconomicsCitation Excerpt :But they also note that further research is needed to explain why an overwhelming majority of households do not fully exploit tax-planning opportunities. Also Alan et al. (2010) provide support for effects of separate taxation on households' portfolio choice. Stoewhase (2011) explores German income tax data to find that more than 20% of couples leave tax-planning opportunities unexploited.
Financial incentives for mortgage prepayment behavior: Evidence from Dutch micro data
2018, Journal of Housing EconomicsTax measures and household financial behaviour: Evidence from France
2017, Quarterly Review of Economics and FinanceCitation Excerpt :These studies find that the variation in marginal tax rates helps explain differences in portfolio structure across households. This result has also been found in Canada (Alan, Atalay, Crossley, & Jeon, 2010). Studies differ in the data used and in how they disentangle the pure tax effect from the income effect on portfolio allocation.
Optimal dynamic tax evasion: A portfolio approach
2016, Journal of Economic Behavior and OrganizationCitation Excerpt :The effects of taxation on household portfolios have long been debated in the literature. Theoretical models predict that under different tax systems, the optimal portfolio allocation depends not only on the risk/return profile, but also on tax characteristics including compensating capital gains with capital losses and whether tax is due upon maturity or realization (Poterba, 2002; Alan et al., 2010). Surprisingly, tax evasion has not received the same attention despite its policy implications (Landskroner et al., 1990; Lee, 1995 present two of the few exceptions).
- 1
Author order is alphabetical.