Elsevier

Journal of Public Economics

Volume 136, April 2016, Pages 14-29
Journal of Public Economics

Optimal income taxation when asset taxation is limited

https://doi.org/10.1016/j.jpubeco.2016.02.003Get rights and content

  • We study optimal labor income taxes when capital income taxation is limited.

  • We use a dynamic moral hazard model with observable and nonobservable assets.

  • Limited observability of assets reduces the possibility to tax capital.

  • Optimal labor income taxes become less progressive when tax on capital is limited.

  • We illustrate the quantitative impact of capital taxation on labor tax progressivity.

Abstract

Several frictions restrict the government's ability to tax assets. First, it is very costly to monitor trades on international asset markets. Second, agents can resort to nonobservable low-return assets such as cash, gold or foreign currencies if taxes on observable assets become too high. This paper shows that limitations in asset taxation have important consequences for the taxation of labor income. We study a simple dynamic moral hazard model of social insurance with observable and nonobservable saving decisions. We find that optimal labor income taxes become less progressive when the ability to tax savings is limited.

Introduction

The existence of international asset markets implies that taxation authorities do not have perfect control over agents' wealth and consumption. This creates an important obstacle for tax policy:

“In a world of high and growing capital mobility there is a limit to the amount of tax that can be levied without inducing investors to hide their wealth in foreign tax havens.” (Mirrlees Review 2010, p. 916)

According to a study by the Tax Justice Network, in 2010 more than $21 trillion of global private financial wealth was invested in offshore accounts and not reported to the tax authorities. These concerns motivated the recent legal steps by the most developed economies to crack down on off-shore bank accounts and the traditional norms of bank secrecy. However, even when agents choose not to hide their wealth abroad, they have access to a number of nonobservable storage technologies at home. For example, agents can accumulate cash, gold, or durable goods. These assets bring lower returns, but nonetheless impose limits for the collection of taxes on assets that are more easily observed.

Motivated by these considerations, this paper explores optimal tax systems in a framework where assets are nonobservable. Our findings show that labor income taxes become less progressive when the government's ability to observe and tax assets is limited. Throughout the paper, we measure the progressivity of labor income taxes based on the concavity of the associated consumption allocation. Tax systems and allocations are closely related in our framework because constrained efficient allocations can be implemented with nonlinear taxes on labor income and linear taxes on assets. Because of this relationship, our measure of progressivity captures the slope of the marginal tax rate on labor income.1 We obtain our results by contrasting two stylized environments. In the first, consumption and assets are observable (and contractable) for the government. In the second environment, these choices are private information. We compare the constrained efficient allocations of the two scenarios. Assuming that absolute risk aversion is convex, we find that in the scenario with nonobservable assets, optimal consumption moves in a less concave way with labor income. In this sense, the optimal allocation becomes less progressive when assets are nonobservable.

We study a tractable dynamic model of social insurance. A continuum of ex ante identical agents influence their labor incomes by exerting effort. Labor income is subject to uncertainty and effort is private information. This creates a moral hazard problem. The social planner faces a trade-off between insuring agents against idiosyncratic income uncertainty and the associated disincentive effects. In addition, agents have access to a risk-free asset, which gives them a means for self-insurance.

In this model, the planner wants to manipulate agents' asset decisions, because asset accumulation provides an insurance against the idiosyncratic income uncertainty and thereby reduces the incentives to exert effort.2 When fully capable of doing so, the planner uses asset taxation to deter the agent from accumulating assets and labor income taxation to balance consumption insurance and effort incentives optimally. When asset taxation is limited, the planner is forced to use labor income taxation also to reduce the agent's incentive to save. Efficiency requires that, for each income state, the costs of increasing the agent's utility by a marginal unit equal the benefits of doing so. A marginal increase in utility in a state with consumption c reduces the agent's marginal return to savings in that state by Ra(c), where a(c) is the absolute risk aversion of the agent at consumption c in that state and R is the asset return.3 Hence, under limited asset taxation, there is an additional social return to allocating utility to a given state and this return is proportional to the level of absolute risk aversion of the agent. Therefore, unless absolute risk aversion is constant or linear, limits to asset taxation have direct implications for the curvature of optimal consumption.4 In particular, whenever absolute risk aversion is convex, the planner finds it optimal to generate an additional convexity (or reduced progressivity) of consumption when asset taxation is limited.

The paper also illustrates the quantitative impact of asset taxation on optimal labor tax progressivity. The calibration of the key parameters of our framework is not straightforward because the technology that determines how effort affects future income is not directly observable. We use consumption and income data from the PSID (Panel Study of Income Dynamics) as adapted by Blundell et al. (2008). In the calibration exercise, we assume that the data is generated by a tax system where labor income taxes are set optimally given the existing asset income tax rate.5 This calibration strategy has the advantage that all fundamental parameters can be identified using only one cross-sectional joint distribution of consumption and income.

Based on the calibrated parameters, we compare the optimal allocations of optimal asset taxation (observable assets) and limited asset taxation (hidden assets). Under optimal asset taxation, the progressivity of the optimal allocation increases sizeably. Our measure of the curvature of consumption increases by 4% for log utility, and between 24% and 49% for relative risk aversion levels of 2 and 3. The tax rates on asset income in the scenario with observable assets are high and exceed 90% for all specifications. Although our numerical simulations are only illustrative, they suggest that a limited capacity to tax assets can have a substantial impact on the optimal progressivity of income taxes.

The paper proceeds as follows. Section 2 surveys the related literature. Section 3 describes the setup of the model. Section 4 presents the theoretical results of the paper. We show that hidden asset accumulation leads to optimal consumption schemes that are less progressive. Section 5 illustrates the quantitative implications of our results. Section 6 discusses model limitations and viable extensions. The appendix collects all proofs that are omitted from the main text and provides further details on the calibration strategy for Section 5.

Section snippets

Related literature

To the best of our knowledge, this is the first paper that explores optimal income taxation in a framework where assets are nonobservable. Recent work on dynamic Mirrleesian economies analyzes optimal income taxes when assets are observable/taxable without frictions; see Farhi and Werning, 2013, Golosov et al., 2013. While the Mirrlees (1971) framework focuses on redistribution in a population with heterogeneous skills that are exogenously distributed, our approach highlights the social

Model

Consider a benevolent social planner (the principal) whose objective is to maximize the welfare of its citizens. The economy consists of a continuum of ex ante identical agents who live for two periods, t = 0,1, and can influence their period-1 labor income realizations by exerting effort. The planner designs an allocation to insure them against idiosyncratic risk and provide them with appropriate incentives to exert effort. The planner's budget must be (intertemporally) balanced.

Preferences. The

Theoretical results on progressivity

We are interested in the shape of second best and third best consumption schemes c(y). As we saw above, this shape is related one-to-one to the curvature of labor income taxes in the associated decentralization.

Definition 1

We say that an allocation (c0,c(⋅),e0) is progressive if c′(y) is decreasing in y. We call the allocation regressive if c′(y) is increasing in y.

Recall that τ(y) = c(y)  y denotes the agent's transfer in labor income state y; hence the negative of τ(y) represents the labor income tax.

Quantitative exploration

In this section, we parametrize our model to illustrate the quantitative effects of limited asset taxation. The quantitative exploration serves multiple purposes. First, we complement our theoretical results. For example, recall that the theoretical results on nonlinear likelihood ratios compare two allocations that implement the same effort level. In this section, we allow effort to change between the two scenarios. The second target of this exercise is to quantitatively explore how a limited

Discussion of model limitations and extensions

This paper analyzed how limitations to asset taxation change the optimal tax code on labor income. Assuming preferences with convex absolute risk aversion, we found that optimal consumption moves in a more convex way with labor income when asset accumulation cannot be perfectly controlled by the planner. In terms of our decentralization, this implies that taxes on labor income become less progressive when limitations to asset taxation are binding. We complemented our theoretical results with a

Acknowledgments

Earlier drafts of this paper were circulated under the title “Optimal Income Taxation with Asset Accumulation.” The paper has benefitted from helpful discussions at numerous seminars and conferences. We particularly thank Claus Thustrup Kreiner, two anonymous referees, Varadarajan V. Chari, Mikhail Golosov, Chris Sleet, and Oleg Tsyvinski for their insightful comments. Sebastian Koehne gratefully acknowledges the financial support from the Knut och Alice Wallenbergs Stiftelse (grant 2012.0315).

References (30)

  • ConesaJ.C. et al.

    Taxing capital? Not a bad idea after all!

    Am. Econ. Rev.

    (2009)
  • DiamondP.A. et al.

    The case for a progressive tax: from basic research to policy recommendations

    J. Econ. Perspect.

    (2011)
  • DomeijD. et al.

    On the distributional effects of reducing capital taxes

    Int. Econ. Rev.

    (2004)
  • EatonJ. et al.

    Optimal redistributive taxation and uncertainty

    Q. J. Econ.

    (1980)
  • FarhiE. et al.

    Insurance and taxation over the life cycle

    Rev. Econ. Stud.

    (2013)
  • View full text