UC San Diego

UC San Diego Electronic Theses and Dissertations

Title
Essays in Public Finance

Permalink
https://escholarship.org/uc/item/6p81w5bq

Author
Miyoshi, Yoshiyuki

Publication Date
2016

Peer reviewed|Thesis/dissertation

eScholarship.org Powered by the California Digital Library

University of California
 UNIVERSITY OF CALIFORNIA, SAN DIEGO

Essays in Public Finance

A dissertation submitted in partial satisfaction of the

requirements for the degree
Doctor of Philosophy

in

Economics

by

Yoshiyuki Miyoshi

Committee in charge:

Professor Roger Gordon, Chair

Professor Julie Cullen
Professor Joseph Engelberg
Professor Joshua Graff Zivin
Professor Natalia Ramondo
Professor Alexis Toda

2016
 Copyright
Yoshiyuki Miyoshi, 2016
All rights reserved.
The dissertation of Yoshiyuki Miyoshi is approved, and it is
acceptable in quality and form for publication on microfilm
and electronically:

Chair

University of California, San Diego

2016

iii
 TABLE OF CONTENTS

Signature Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

Abstract of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

Chapter 1 Does Sales Factor Apportionment Benefit the Welfare of State? . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Overview and discussions . . . . . . . . . . . . . . 6
1.2.2 Production technology and firms . . . . . . . . . . . 10
1.2.3 Workers . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.4 Governments . . . . . . . . . . . . . . . . . . . . . 14
1.3 State tax policy . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Optimal tax policy in autarky . . . . . . . . . . . . . 16
1.3.2 Effects of taxing apportionment factors . . . . . . . 18
1.4 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.4.1 Parameters . . . . . . . . . . . . . . . . . . . . . . 22
1.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . 24
1.5 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.7.1 Profit maximization problem for firms . . . . . . . . 32
1.7.2 Proof for Proposition 1 . . . . . . . . . . . . . . . . 34
1.7.3 Proof for Proposition 2 . . . . . . . . . . . . . . . . 35

Chapter 2 Risk-Taking and Growth under Taxation with Moral Hazard . . . 37

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3 Solving for equilibrium . . . . . . . . . . . . . . . . . . . . 49
2.3.1 Individual problem . . . . . . . . . . . . . . . . . . 50
2.3.2 Equilibrium conditions . . . . . . . . . . . . . . . . 52
2.3.3 Welfare . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3.4 Stationary distribution . . . . . . . . . . . . . . . . 57

iv
 2.4 Numerical example . . . . . . . . . . . . . . . . . . . . . . 60
2.4.1 Functional forms . . . . . . . . . . . . . . . . . . . 61
2.4.2 Calibration . . . . . . . . . . . . . . . . . . . . . . 62
2.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . 63
2.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . 69
2.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.6.1 Solution algorithm . . . . . . . . . . . . . . . . . . 69

Chapter 3 Rental Housing Investment and Tax Reforms: An Empirical Study

of Tax Clientele Model . . . . . . . . . . . . . . . . . . . . . . . 72
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2 A tax clientele model for rental housing investment . . . . . 76
3.3 Tax reforms in 1980s . . . . . . . . . . . . . . . . . . . . . 80
3.4 Data and methodology . . . . . . . . . . . . . . . . . . . . 82
3.4.1 Survey of Consumer Finances . . . . . . . . . . . . 82
3.4.2 Summary information on owners of rental housing . 83
3.4.3 Estimating marginal tax rates . . . . . . . . . . . . . 84
3.4.4 Empirical specification . . . . . . . . . . . . . . . . 85
3.5 Empirical findings . . . . . . . . . . . . . . . . . . . . . . . 86
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 90

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

v
 LIST OF FIGURES

Figure 1.1: The Effective Average Federal Personal Income Tax Rate . . . . . . 24
Figure 1.2: The Effects of Sales Factor Weights on Welfare and Aggregate
Variables in the Two-State Model . . . . . . . . . . . . . . . . . . 25
Figure 1.3: The Effects of Sales Factor Weights on Welfare and Aggregate
Variables in the 50-State Model . . . . . . . . . . . . . . . . . . . 28

Figure 2.1: Comparative statics with respect to the consumption tax rate (η = 0.05) 64
Figure 2.2: Comparative statics with respect to the consumption tax rate (η =
0.075) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

vi
 LIST OF TABLES

Table 1.1: Optimal State Tax Rates in Nash Equilibrium of the Two-State Model 25
Table 1.2: Optimal State Tax Rates in Nash Equilibrium of the 50-State Model . 27

Table 2.1: Parameter values . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Table 3.1: Probability of ownership (percent) of residential rental properties . . 83

Table 3.2: Probit estimates for the ownership of residential rental properties . . 87
Table 3.3: Average marginal effects of marginal tax rate on the probability of
ownership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

vii
 ACKNOWLEDGEMENTS

I would like to acknowledge Roger Gordon for his support as the chair of my
committee. Studying public finance under a master of the field has been an honor and a
privilege. His mentoring and guidance have always been invaluable, and he taught me
how to deepen my understanding of real world issues using the economic theory.
I would also like to thank Alexis Akira Toda for his mentoring and many hours of
discussion. Working on an ambitious research project together with him was very helpful
especially when I was struggling to make the transition from a student to a researcher.
Many additional thanks to Julie Cullen, Joseph Engelberg, Joshua Graff Zivin and
Natalia Ramondo. Their time, thoughts, and feedback helped me immensely throughout
this process.
Of course, the wonderful opportunity to spend four years following my intellectual
passions would never have been possible without the love and support of my family.
A special thanks to Natsuki Miyoshi, Mihoko Miyoshi, Shigeto Miyoshi and Kazuo
Miyoshi.
Chapter 2, in full, is currently being prepared for submission for publication of
the material. Miyoshi, Yoshiyuki; Toda, Alexis Akira. The dissertation author was the
primary investigator and author of this material.

viii
 VITA

2002 Bachelor of Arts, University of Tokyo, Japan

2008 Master of Arts, Tufts University

2016 Doctor of Philosophy, University of California, San Diego

ix
 ABSTRACT OF THE DISSERTATION

Essays in Public Finance

by

Yoshiyuki Miyoshi

Doctor of Philosophy in Economics

University of California, San Diego, 2016

Professor Roger Gordon, Chair

Researches in public finance have clarified the effects of tax policies that are
neither originally intended nor easily observable. This work contributes to the literature
of public finance by rigorously examining the effects of three particular tax policies using
quantitative models and econometric analysis.
Chapter 1 aims to answer the question as to what is the optimal apportionment
formula for the US state corporate income tax. States have raised sales apportionment
weight and lower payroll weight to stimulate local labor demand. However, the policy
discussions often ignore the negative effect of sales apportionment tax; the tax distorts
the sales allocation of firms across states and causes an increase in local price level in

x
the state. I construct a quantitative model that incorporates the effects of apportionment
formula both on local labor demand and on the price level. The calibration suggests that
the sales weight should be zero for the optimal tax policy of a single state because the
negative effect on price level outweighs the positive effect on local labor under a range of
plausible parameters.
Chapter 2 studies the effect of taxation on entrepreneurs’ risk-taking, portfolio
choice, and economic growth in the presence of a moral hazard problem in a dynamic
general equilibrium model. The moral hazard problem occurs because the return of
a type of entrepreneurial project depends on entrepreneurs’ effort, which is private
information. By collecting proportional consumption tax and redistributing the revenue,
the government offers an opportunity for risk sharing to encourage risk-taking and spur
economic growth. We show that when the moral hazard problem is absent, the full
insurance is optimal in terms of welfare, and the economy grows faster. But if the
moral hazard problem exists even in a slight degree, risk sharing through taxation cannot
improve welfare.
Chapter 3 examines the relationship between household marginal tax rates and
the probability of owning rental housing. I focus on the special provisions about passive
losses introduced by 1986 TRA to test the theoretical prediction about this relationship.
The empirical results based on the Surveys of Consumer Finances offer modest support
for the prediction.

xi
Chapter 1

Does Sales Factor Apportionment

Benefit the Welfare of State?

Abstract

States have raised the sales apportionment weight and lowered the payroll weight
to stimulate local labor demand. However, the policy discussions often ignore the negative
effect of sales apportionment tax; the tax distorts the sales allocation of firms across
states and causes an increase in the local price level in the state. This study examines the
optimal tax policy from the perspective of a single state and predicts Nash equilibrium
with a quantitative model that incorporates the effects of apportionment formula both on
local labor demand and on the price level. The calibration suggests that the sales weight
should be zero for the optimal state tax policy because the negative effect on the price
level outweighs the positive effect on the local labor demand under a range of plausible
parameters.

1
 2

1.1 Introduction

Formula apportionment constitutes a major difference in the systems of corporate

income tax between the federal and state governments in the U.S.1 The apportionment
formula determines what fraction of the taxable domestic profit of a multi-state firm is
subject to state corporate income tax in each state. Adjusting this formula can significantly
change the distribution of tax liability of a firm across states and then can affect the
firm’s decision about location, production and sales. The previous studies have found
apportionment formula has impacts on the welfare and aggregate economic variables,
including the prices and employment.
Most of the U.S. states use the three-factor apportionment formula, which defines
the tax liability of multi-state firm j for state n as follows:

 n n n 
n n W ( j) n K ( j) n S ( j)
T ( j) = tCn γW + γK + γS πT ( j), (1.1)
W ( j) K( j) S( j)

in which W n ( j), K n ( j), and Sn ( j) represent payroll, property, and sales of firm j in state
n, respectively; W ( j), K( j), and S( j) represent total domestic payroll, property, and sales
n , γn , and γn are the weights for payroll, property, and sales
of firm j respectively; γW K S

factors respectively set by state n and ∑h γnh = 1; πT ( j) is the taxable domestic profit for
firm j. In the U.S., states can choose these factor weights independently as long as the
weights sum up to one. From this formula, state corporate income tax can be viewed as a
combination of three separate taxes on payroll, property, and sales of firms, as McLure
(1981) points out.
Most states used to opt for the equally weighted formula, which puts the equal
weight on three factors, because the Multistate Tax Compact recommended the formula.
1 Formula apportionment is adopted in some of other developed countries: for example, Canada and

Japan. But, unlike in the U.S., the formula is uniformly set by the national government in those countries.
 3

After the Supreme Court upheld the right of states to use other formulas than the equal
weight formula in 1978, many states started to lessen the weights on payroll and property
factors and put more weight on the sales factor to stimulate the demand for local employ-
ment (Mazerov 2001). This trend in state tax policy still continues, if not accelerates,
these days. Now 19 states have even adopted the single sales factor apportionment, which
put the full weight on the sales factor. The conventional wisdom in policy discussions is
that larger sales weight benefits the state, especially through an increase in the demand
for local labor, but the policy is not desirable from the perspective of nation because the
competition of tax policy among state governments eventurelly leads to a “prisoner’s
dilemma” type equilibrium.
It is often overlooked, however, that taxing the sales factor has its own cost. If
a state sets a higher tax rate for sales factor (tCn γnS in equation 1.1) than the other states,
the tax liability of a firm increases as the firm sells more in the state.2 Thus the higher
tax rate for the sales factor distorts the sales of firms in the way that firms sell less in the
state. At the aggregate level, this leads to relatively less sales and higher prices in the
state compared to other states with a lower tax rate for the sales factor. In sum, a greater
sales weight will not only raise local real income, either through improved employment
or through a higher wage rate, but it will also reduce local real income through a higher
price level at the same time. As a whole, the total effect of change in the apportionment
formula is ambiguous.
The goal of this study is to derive the optimal corporate income tax policy for state
governments, including the optimal approtionment weights, with a quantitative model
that is calibrated plausibly. In particular, the model formalizes the trade-off between local
labor demand and price distortion caused by the apportionment formula. The literature
2 In this paper, I assume the sales in apportionment formula is defined as sales at destination rather than

at origin. Gordon and Wilson (1986) follows the same definition because it matches the provisions of tax
code of most U.S. states.
 4

on formula apportionment has successfully illuminated the former effect, especially

by Goolsbee and Maydew (2000). However, no theoretical studies have been done to
examine the optimal apportionment formula with a model that incorporates this trade-off
and is able to evaluate the combined effect.3
Another key feature of my model is that states use personal income tax as well
as corporate income tax as a source of revenue. If states only considered the effect of
corporate income tax on the local labor market and if personal income tax were available
to them with no cost, then the best tax policy would be choosing zero (or even negative)
corporate income tax and collect the necessary revenue by personal income tax. On the
contrary, it is a conspicuous observation from the actual state tax policy in the U.S. that
most states impose corporate income tax and, on average, the corporate income tax rate
is close to the top marginal tax rate of personal income: these rates are 7.33% and 7.58%
respectively in 2014, if weighted by state GDP.4 Since the revenue from personal income
tax acounts for much larger share of the total state tax revenue than corporate income tax,
it is important to examine states’ choice about personal income tax when a study wants
to derive implications for the optimal state corporate income tax. However, the previous
theoretical studies on the optimal apportionment formula do not allow states to use other
sources for revenue than corporate income tax.5
States need both personal and corporate income taxes in my model because of
income shifting behavior among workers. Workers can shift their income from labor
income to capital income if their effective tax rate for labor income is higher than the
3 As examples of studies regarding the effect of sales factor, Gordon and Wilson (1986) theoretically
analyze the effect of sales weight on output prices separately from the effects of property or payroll weight,
and derive the result of “cross-hauling” caused by uneven tax rates for sales factor. The empirical study
of Edmiston and Arze del Granado (2006) reports that the increase in sales weight led to a drop in local
sales of multi-state firms as well as a rise in payroll of them in State of Georgia. Fajgelbaum et al. (2015)
focuses on the improvement of welfare that can be attained by harmonizing state tax policy, including
apportionment formula, rather than the optimal tax policy from the perspective of a single state.
4 This relation is also found in federal corporate and personal income taxes.
5 For examples, see Anand and Sansing (2000), Pinto (2007), and Runkel and Schjelderup (2011).
 5

rate for capital income. Thus if a state sets personal and corporate income tax rates being
wildly different, the gap in the tax rates severely distorts such income shifting behavior
of workers. In short, state govenrments use corporate income tax as the “backstop” for
implementation of personal income tax system (Mirrlees 2011).
The model used in this study is a general equilibrium trade model in which
multiple states set their tax policy strategically in the setting of static game. Firms
respond to the state tax policy by choosing location in which to operate and the vector of
price and output for each state. Basically, the tax rate for the payroll factor affects the
location choice of firms while the tax rate for the sales factor does the prices and output
that firms choose. In the calibration exercises, I look for Nash equilibrium of the model.
Contrary to the popular policy recommendation and the implications from the existing
literature, the model suggests that zero sales weight should be chosen in equilibrium
under plausible paraemters because the negative effect of sales apportionment on the
price level outweighs its positive effect on the local labor demand. In addition, even if
the current equilbrium is distorted due to some exogenous restriction on apportionment
weights, say that the sales weight should equal one third, the best response of a state is
to reduce sales weight to zero once it starts to have the right to adjust the weights on its
own.
Section 2 develops the model and defines the objective of state governments.
Section 3 describes the specific features of the model: the optimal balance between
corporate and personal income taxes and the effects of apportioment weights on the
aggregate economic variables. Section 4 presents the results of calibration and simulation.
Section 5 discusses the validity of the model, including comparing the model’s prediction
with the existing empirical study. Section 6 concludes.
 6

1.2 Model

1.2.1 Overview and discussions

This study use a model to derive implications about the optimal tax policy from
the perspective of a single state government and Nash equilibrium of state tax policy
of multiple states. Therefore it is critical to define the objective of state government
plausibly. State governments in this model aim to maximize the social welfare in the
state, which is defined as the purely utilitarian welfare function, by using a restricted
set of policy instruments. A state government collects its revenue through personal and
corporate income taxes, and the schedules of those taxes are constrained to be linear. It
produces services that the private sector does not produce, and provides an equal amount
of the services (hereafter referred to as the state government services) to each of its
residents. The federal government also imposes personal and corporate income taxes.
Federal personal income tax has a progressive schedule while federal corporate income
tax a linear one. Another important assumption in my model is that workers supply their
endowment of labor inelastically regardless of the state and federal tax policies.
The key feature of my model is that state governments set personal and corporate
income tax rates at the same time and keep a balance between these two tax rates. The
mechanism to detemine the optimal balance for state govenrments is centered on income
shifting behavior of high income workers. In this model, workers can choose to receive
the payment for the labor they supply as capital income rather than as labor income
at no cost if they want. Capital income is subject to corporate income tax as part of
firm’s taxable profit but not subject to any tax at the personal level. In addition, workers
are heterogeneous in labor efficiency and then in income. For workers with high labor
efficiency, it may be advantageous to shift their income from labor income to capital
income since federal personal income tax has a progressive schedule while federal and
 7

state corporate income tax has a linear one. Thus given federal and state tax rates, there
is the cut-off value of income; workers whose income is above the cutoff value undertake
income shifting.
The objective function of state governments in this model is defined as it is not
sensitive to the allocation of income across workers. Although the utility function of
workers is a Cobb-Douglas function that is quasi-concave in the private goods and state
government services, it is defined to be linear in private consumption. Thus the social
welfare function that adds up the utility of workers with an equal weight is linear in the
aggregate private consumption, or equivalently in the sum of real after-tax income across
workers in the state.
Given this objective function, state governments do not care about the distribution
of effective tax rates across workers as the result of income shifting. State governments,
however, do care about the aggregate income of workers after federal taxes are collected.
A state can change the cutoff income for income shifting in the state by setting different
tax rates for personal and corporate income taxes, and, as the cutoff income changes,
federal tax collection also changes. As I argue in the next section in detail, if state
tax policy did not affect the economic variables, including the wage rate and the price
level, equal tax rates for state corporate and personal income taxes would minimize
federal tax collection. In other words, it would be best for states not to use uneven tax
rates for corporate and personal income taxes because such tax policy distorts income
shifting behaviour of workers. The combination of inelastic labor supply and the social
welfare function that is linear in private consumption enables the model to incorporate
the balance between corporate and personal income taxes as an endogenous choice of
state governments in a simple way.
Although my model makes states choose the optimal mix of corporate and
personal income taxes endogenously, the limited policy instruments for the government
 8

sector and inelastic labor supply of workers are restrictive and different from the standard
optimal taxation literature, in which the government searches for the flexible tax schedule
that maximizes the social welfare defined with a certain degree of redistributive taste
under the constraint of private information.6 However, the structure of my model makes
the focus of this study—the trade-off between local labor demand and the price level—
transparent. Moreover, it approximates the principal features of actual state government
finances to a considerable extent, and can readily offer quantitative policy implications
for state governments within the restriction of policy tools.
In practice, the progressivity of state personal income tax is moderate on average:
among the 43 states that impose personal income tax (including the District of Columbia),
10 states have a linear tax schedule in 2015.7 In addition, in the 18 states out of the 33
states that have progressive tax schedules, the top rate is reached at a household income
level below the national average, and, in most cases, below $20,000. The average tax rate
for the household with the national average income is 5.34%, if weighted by state GDP,
while the weighted average tax rate for the highest bracket is 7.33% in 2014. As for the
effective tax rates rather than the statutory rates, Gordon and Cullen (2012) report the
effective marginal tax rates of federal and state personal income tax that are calculated
based on the data including individual tax returns. The effective marginal tax rates of
state income tax range between 3.5% and 6.2% across the top four income groups out
of the five income groups, excluding the bottom group, while the federal rates range
between 15.0% and 31.2% across the same groups.8 As for corporate income tax, two
6 Notable studies with policy implications in the standard optimal taxation literature are Saez (2001)
and Diamond and Saez (2011).
7 The states that do not impose personal income tax collect a large fraction of their revenue from general

sales tax, except for Alaska and Wyoming. General sales tax with a single tax rate is considered to be
equivalent to a linear personal income tax in my static model, although the model does not accomodate the
option of sales tax as an endogeneous choice for state governments.
8 Gordon and Cullen (2012) report a very high marginal tax rate of state personal income tax for the

bottom income group. They attribute it to transfer income that the households in the group lose as their
income goes up.
 9

thirds of states have linear tax schedules. Therefore, it appears a plausible approximation
of the actual U.S. state tax policy to restrict the choice of state personal and corporate
income tax schedules to being linear.
On the spending side of state government finances, the largest share is spent on
education (35.6%). State governments are also involved in provision of large-scale public
infrastructure, most notably highway projects (6.7%). These facts motivate my model to
have state governments produce services that the private sector does not and provide the
services equally to each of their residents.9
The model is a general equilibrium trade model of multiple states. In the following
two sections, the economy is assumed to consist of two states, A and B, for the purpose of
presentation although it is relatively straightforward to extend the model to an economy
with more than two states. State governments are players of a static game who aim
to maximize their objective function taking account of general equilibrium outcome.
There are continuum of firms of fixed measure in the economy. Each of them produces
a differentiated good. Every firm is perfectly mobile and chooses one of the states to
operate in to maximize its profit, considering the wage rates, local productivity, and state
tax policies. On the contrary, workers are immobile in this model and consume a set of
the differentiated goods and the state government services. Only state governments can
produce the state government services, using the set of differentiated goods as inputs.
The differentiated goods can be transported between states with no cost, but the state
government services cannot be transported to the other state.
9 The treatment of state expenditure here is almost equivalent to Fajgelbaum et al. (2015), in which the
level of state government spending affects the level of local amenity that enters consumers’ utility equally,
although the level of state government spending also affects the productivity of firms in their model. On the
contrary, state governments also engage in income redistribution not through a highly progressive income
tax, but through the expenditure on public welfare programs, including Medicaid. That type of expenditure
accounts for 30.8% of total expenditure in 2013. Gordon and Cullen (2012) focus on this role of state
governments and provide a theoretical framework for the optimal taxation in which each of multi-level
governments tries to maximize thier own welfare function by nonlinear labor income tax.
 10

1.2.2 Production technology and firms

Firms of measure M produce the differentiated goods using labor as the only
input. The production function for firm j is

x( j) = zn ( j)l( j),

in which zn ( j) is the productivity of firm j if it operates in state n. A perfectly competitive

sector in each state produces the final good from the differentiated goods according to
the following technology:

ˆ  σ
σ−1 σ−1
X= x( j) σ dj , (1.2)
M

in which the substitution parameter σ > 1. The final good is either consumed by workers
in the state or used to produce the state government services. The price of the final good
in state n is expressed as

ˆ  1
1−σ
n n 1−σ
P = p ( j) dj ,
M

in which pn ( j) is the price that firm j sets for its good sold in state n.
Firms maximize their profits as they choose their production levels, the prices for
their goods, and the locations of production. A firm operates in a single state and is not
allowed to merge with another firm in this model. All the firms are subject to federal and
state corporate income tax and not allowed to choose any organizational forms to which
corporate income tax is not applied.
The state corporate income tax system follows a two-factor formula apportion-
ment.10 In addition to their corporate income tax rates, state governments set their
10 Production does not require capital as an input in this model. However, if production function is
 11

apportionment weights on the payroll and sales factors. Let γnS be the sales weight of
state n, and then its payroll weight equals 1 − γnS . If firm j chooses to locate in state n,
the tax liability for the firm is

Tn ( j) = tCF + t¯n ( j) πTn ( j),

in which tCF is federal corporate income tax rate; t¯n ( j) is the effective state corporate
income tax rate for firm j, which is calculated based on the apportionment formula; πTn ( j)
is the taxable profit for firm j.11 The apportionment formula defines t¯n ( j) as

tCA γAS pA ( j)xA ( j) + tCB γBS pB ( j)xB ( j)

t¯n ( j) = tCn (1 − γnS ) + , (1.3)
pA ( j)xA ( j) + pB ( j)xB ( j)

in which tCn is the statutory rate for corporate income tax of state n; xm ( j) is the amount
of firm j’s good that is sold in state m.
The payment to workers is the only cost of production for firms. In this model, a
worker can negotiate with the firm to receive the payment either as labor income or as
capital income. The payment of labor wage is deductible from the taxable profit of firms.
But if a firm pays capital income to a worker rather than labor wage, the payment adds to
the firm’s taxable profit and is subject to federal and state corporate income tax. Thus the
wage rate for capital income payment to the income-shifters at firm j that operates in
state n is determined as
w̃Cn ( j) = 1 − tCF − t¯n ( j) wnN ,


(1.4)

in which wnN is the prevailing market wage rate for labor income in state n. The tilde on a
homogeneous of degree one in labor and capital and if capital is also immobile (like land), the following
results remain true.
11 To make the model simple and transparent, the federal government in this model does not allow firms

to deduct the payment of state corporate income tax from their federal taxable income, unlike the U.S. tax
code.
 12

variable in this paper means that the variable is the after-tax one.
The after-tax profit of firm j, when it chooses to locate in state n, is represented
as

xnA ( j) + xnB ( j)
 
π̃n ( j) =(1 − tCF A A B B n
− t¯n ( j)) pn ( j)xn ( j) + pn ( j)xn ( j) − wN (1 − θn )
zn ( j)
xA ( j) + xnB ( j)
− w̃Cn θn n , (1.5)
zn ( j)

in which θ is the share of efficiency unit of labor that is provided by income-shifters; the
subscripts n on some of the variables repersent being conditional on operating in state n.
In this model, θn is not a choice variable of each firm, but every firm in a state is assumed
to employ the same fraction of income shifters. On the other hand, the effective state
tax rate t¯n ( j) is an endogenous variable that is determined by the firm’s decision on the
allocation of sales among the states. Nonetheless, all the firms in the same state choose
the same allocation of sales shares as the solution to its profit maximization problem,
even if they are heterogeneous in productivity. Refer to Section 1.7.1 for the proof.12
Since it implies all the firms in the same state are subject to the same effective state tax
rate t¯n , the wage rate for income-shifters w̃Cn is uniquely determined by equation 1.4. Note
that this makes all the firms in state n indifferent about θn from equation 1.5. Finally,
firm j chooses to locate in the state where the maximum of π̃n ( j) is the larger.
Since firms engage in monopolistic competition, firms earn positive profits after
paying out capital income to income-shifters. Those profits are pooled and distributed to
workers by a perfectly competitive financial sector. ρA fraction of the total after-tax profit
is distributed to workers in state A while the remaining, the fraction of ρB = 1 − ρA , to
12 Fajgelbaumet al. (2015) first prove this result of constant t¯n ( j) across all the firms in the same state
for the standard Dixit-Stiglitz model of monopolistic competition. The proof in Section 1.7.1 shows their
result can be extended to the model with income shifting with minor adjustments.
 13

state B.

1.2.3 Workers

There are N n workers in state n. Workers are heterogeneous in terms of their

endowment of efficiency unit of labor, l n (i) for worker i, and they supply labor inelasti-
cally. Ln stands for the mean efficiency unit of labor of workers in state n, and then the
aggregate labor supply in state n equals N n Ln . In addition to labor income, and capital
income for income-shifters, workers receive dividend from the financial sector. Workers
in state n own shares for dividend in proportion to their efficiency unit of labor.13 Thus
the dividend payment for worker i in state n is

ρn l(i)Π̃
d˜n (i) = ,
N n Ln

in which Π̃ is the aggregate profit of firms after the payments for income shifters and for
the federal and state corporate tax liabilities.
Labor income is subject to federal and state personal income taxes while capital
income for income-shifters and dividend income are not. Since federal personal income
tax has a progressive schedule, the liability of that tax is larger for high income workers.
Let tNF (i, n) be a shorthand notation for the average tax rate for worker i in state n if the
worker chooses not to be an income-shifter: tNF (i, n) = TNF (wnN l n (i))/(wnN l n (i)), in which
TNF () is the function for the schedule of federal personal income tax liability. Thus the
13 Imake this assumption regarding the distribution of financial income being motivated by the well-
known fact that the distribution of wealth is significantly skewed. However, the following results are not
affected by the assumption because, as I explain below, the state welfare function is linear in aggregate
income.
 14

total after-tax income of worker i in state n is


w̃Cn l n (i) + d˜n (i)

 if i is an income-shifter,
ỹn (i) = (1.6)
(1 − tNF (i, n) − tNn )wnN l n (i) + d˜n (i) otherwise.



in which tNn is state personal income tax rate in state n. Since the cost of income shifting
for workers is zero in this model, they choose the type of income that makes ỹn (i) the
larger. Thus equation 1.4 implies that worker i in state n becomes an income-shifter if
the personal income tax liability is larger than the incidnce of federal and state corporate
income tax, or tNF (i, n) + tNn > tCF + t¯n .
Workers consume the state government service as well as the private final good.
Worker i in state n has the following utility function:

ỹn (i) n αG
un (i) = (g ) , (1.7)
Pn

in which gn is the state government service provided to worker i by the state government
in n; αG > 0 is the preference parameter for the state government services. Note that
this utility function is a standard Cobb-Douglas function and exhibits the quasi-concave
property although it is linear in (real) income.

1.2.4 Governments

State governments maximize the social welfare of residents that is defined as the
purely utilitarian welfare function:

ˆ
n
V = un (i) f n (i)N n di, (1.8)
i∈n
 15

in which f n (i) is the probability density function for the distribution of heterogeneous
workers in state n. The budget constraint for state government is

N n gn ≤ znG (T n /Pn ), (1.9)

in which znG is the productivity for state government services in state n; T n is the aggregate
tax revenue for state n.14 State governments take into account only how the federal tax
collection affects the utility of their residents, but not how the changes in federal revenue
caused by the state’s fiscal policy affect the utility.15 The federal government takes the
final good from the economy as its tax collection.
Since the utility function defined as equation 1.7 is linear in real income for
worker i and gn is constant across all the workers in state n, the social welfare function
for state n is linear in the aggregate income of the state. Therefore state governments
in this model are not sensitive to the allocation of income among their residents. This
feature isolates the aggregate effects of formula apportionment on the welfare of state
from the issue of income redistribution, still keeping the heterogeneity across workers,
which is needed to incorporate the choice of state governments about the optimal mix of
personal and corporate income taxes.
14 Since the state welfare function preserves the property of Cobb-Douglas utility function, zn
is irrelevant
G
for the optimal tax policy.
15 This assumption can be justified, for example, when the federal government covers a lot more states

with which the two states have no economic interaction, and when the two states are too small to have a
significant effect on the total federal revenue.
 16

1.3 State tax policy

1.3.1 Optimal tax policy in autarky

I use a variant of my model, which turns the trading economy into the autarky
economy, to illustrate how the environment of this model determines the optimal mix
of state corporate and personal income taxes. In this case, state A has no economic
interaction with state B: there is no trade between the two states, and the profits of firms
are not pooled nationally, but are distriburted within the state. The federal government,
however, still collects tax from both states. Since the apportionment weights are irrelevant
in this case, the variables the state governments use are personal and corporate income
tax rates.
From equation 1.2, the demand for good j is:

Yn
x( j) = p( j)−σ , (1.10)
(Pn )1−σ

in which Y n is the before-tax aggregate income in state n. The nominal price level is
fixed in such a way that Y n is normalized to one. The optimization problem for firm j is:

wnN
 
max (1 − tCF − tCn ) p( j)x( j) − x( j) ,
p( j),x( j) z( j)

subject to equation 1.10. This objective function is simplified from equation 1.5 by
substituting equation 1.4. It can be shown that the equilibrium before-tax wage rate and
price level do not depend on either federal or state tax policy:

 
σ−1 1
wnN = ;
σ N n Ln

Pn = (z̄n )−1 ,
 17

in which
ˆ  1
σ−1
n n σ−1 n
z̄ = z ( j) g ( j)d j ,
M

in which gn ( j) is the probability density function for firms in state n. The implication
that state tax policy does not distort the production in the autarky economy makes the
problem of optimal state tax policy simple as Proposition 1 shows:16

Proposition 1. [The optimal state tax rates in autarky] In the case of autarky, the
optimal rates for state personal and corporate income tax are equal: tSn = tCn .

The proof for Proposition 1 is presented in Section 1.7.2. This result can be
interpreted as follows. If there were no state taxes, the federal personal and corporate
income taxes would completely determine the cutoff efficiency unit of labor l¯ in state
¯ = t F wn l.¯ Those who are more efficient than l¯ engage in income
n such that TNF (wnN l) C N

shifting while those who are less efficient do not. With this being said for the no state
tax case, if the state sets equal rates for both income taxes, the cutoff efficiency does
not change. However, if the state tax rates do not match, it changes the cutoff efficiency
and distort the behavior of workers that originally minimizes the federal tax burden.
Therefore uneven state tax rates increase the federal tax collection, which flows out of the
state economy. Since the welfare function for state government is linear in the aggregate
after-tax income and is not sensitive to the allocation of after-tax income among the
residents, the state government does not want to distort the cutoff efficiency.
The interpretation can be extended to a broader perspective. My model abstracts
from elastic labor supply and the benefits and costs regarding firms’ choice of organiza-
tional form. Although this feature makes the model tractable, its cost is that the optimal
federal personal and corporate income tax schedules should be taken as exogenous.17 If
16 Althoughmonopolistic competition is present in this model, the production activity is not distorted
because workers supply labor inelastically and all the profits of firms are distributed to workers.
17 Gordon and Mackie-Mason (1994) and Mackie-Mason and Gordon (1997) provide the empirical
 18

the federal government implements its tax policy, especially the arrangement of personal
and corporate income taxes, considering factors that my model cannot capture, Proposi-
tion 1 suggests state governments respect the choice of federal government except for a
uniform linear tax that does not distort firms’ choice of organizational form in the case
of autarky. As the following sections argue, however, when the two states interact, state
governments are faced with the trade-off between the opportunity for boosting local labor
demand and the distortion to income shifting behaviour. In addition, it complicates the
state governments’ problem that part of the tax burden can be exported through taxing
the profits of firms in the other state.

1.3.2 Effects of taxing apportionment factors

In the model of trading states, state corporate income tax affects the location and
production choices of firms since firms are perfectly mobile across the states. It requires
specifiying the distribution of productivity among firms to analyze the effect of state
corporte tax on firms’ location choice. I choose the assumption of homoegneous firms,
in which zn ( j) = z for all j and n in the rest of this section and the following calibration
exercises. Section 1.5 discussses the implications of this assumption.
The corpoarte tax rate affects firms’ choice, interacting with the apportionment
weights. If the effective tax rate on the payroll factor in state A is higher than in state
B, it reduces the demand for labor in state A and then lowers the before-tax wage rate
there. If the effective tax rate on the sales factor in state A is higher, it raises output prices
in the state since firms have an incentive to reduce the corporate income tax burden by
selling less in state A and more in state B. The rest of this section examines the effects of
state corporate income tax in the two polar cases regarding the apportionment weights
evidence that the difference between corporate and personal income tax distorts the choice of organizational
form. Piketty et al. (2014) propose a model for the optimal labor income tax in the presence of income
shifting.
 19

to illustrate how the choice of the weights affects firms’ behavior and the aggregate
variables.

Payroll factor

First assume the states put the full weight on payroll factor and zero on sales
factor. Since firms are homogeneous and mobile, they are indifferent about the location
in equilibrium:
π̃A ( j) = π̃B ( j) for all j, (1.11)

in which

 A B 
n xn ( j) + xn ( j)
π̃n ( j) = max (1 − tCF − tCn ) A A B B
pn ( j)xn ( j) + pn ( j)xn ( j) − wN .
pn ( j),xn ( j) z

The first order conditions imply that frims sell their goods at the same price in both states:

wnN
 
σ
pAn ( j) = pBn ( j) = .
σ−1 z

It is possible to derive the relation between the corporate income tax rates and the
equilibrium wage rates:
1
wAN 1 − tCF − tCA
  σ−1
= . (1.12)
wBN 1 − tCF − tCB

This implies the elasticity of wage rate with respect to net-of-tax rate, 1 − tCF − tCn , is
1/(σ − 1) if the tax rate in the other state is fixed.
Equation 1.12 means that state governments can increase the local wage rate by
reducing its corporate income tax rate, or its effective tax rate for the payroll factor more
precisely.18 This result fits the previous empirical studies: for example, Goolsbee and
18 Inthe limiting case, state corporate tax does not affect the local wage as σ goes to infinity. For the
market is perfectly competitive and firms earn no economic proft in this case. Corporate income tax
revenue only comes from income shifters in the state and the full incidence falls on those workers without
 20

Maydew (2000) and Edmiston and Arze del Granado (2006).

However, there are two considerations that offset this positive effect of lower
corporate income tax rate. First, if firms earn economic profits and state tax policy does
not affect the distribution of ownerships of firms, as this model presumes, the incidence
of state corporate income tax partly falls on the owners outside the state. This makes
taxing the profits of firms a less costly way to raise revenue. Second, if a state government
taxes firms lightly, the gap between personal and corporate tax rates changes the income
shifting behavior from one that federal tax policy intends and, as a result, it increases
the federal tax collection, which is taken away from the state economy. Therefore, the
quantitative implications for optimal state tax policy in this case depends not only on the
substitution parameter σ, but also on the distribution of firm ownerships and federal tax
policy.

Sales factor

Next assume the states put the full weight on the sales factor and zero on the
payroll factor. In this case, the effective corporate tax rate for firms is not affected
by the location of firms, but it is completely determined by the ratio of sales between
the two states.19 State corporprate income tax does not affect the production cost, and
homogeneous firms locate in whichever state that offers the lower production cost. Thus,
wAN = wBN = wN in equilibrium.
distortion.
19 In practice, the effective tax rate also depends on the location of firms because of the nexus rules in

the U.S. tax code. Section 1.5 discusses this issue further.
 21

The optimization problem for firm j is:

 A B 
n xn ( j) + xn ( j)
max (1 − tCF − t¯n ) A A B B
pn ( j)xn ( j) + pn ( j)xn ( j) − (1 − θn )wN
pn ( j),xn ( j) z
xnA ( j) + xnB ( j)
− θn wCn ,
z

in which
tCA pAn ( j)xnA ( j) + tCB pBn ( j)xnB ( j)
t¯n ( j) = .
pAn ( j)xnA ( j) + pBn ( j)xnB ( j)

The closed-form solution for equilium is not derived in this case, but the qualitative effect
of taxing sales factor on the price level can be shown as follows:

Proposition 2. [Distortion by unequal sales factor taxation] When states use the sin-
gle sales factor, namely γAS = γBS = 1 in equation 1.3, the following statements concerning
the prices of final good, PA and PB , hold.

1. If tCA = tCB , then PA = PB = P̄, in which

 
σ wN
P̄ = .
σ−1 z

2. If tCA > tCB , then PA > P̄ > PB .

Refer to Section 1.7.3 for the proof. Proposition 2 shows a greater weight on the
sales factor carries some cost to the state economy as the local price level rises. Note
that the structure for the optimal tax policy is similar to the one in the case of the full
payroll weight; a low corporate tax rate may improve the state’s welfare through a lower
price level, which leads to higher real income because wN is equal among states. But
there are two offsetting effects: the distortion caused by the gap between corporate and
personal income tax rates and the possible benefit of exporting tax burden through taxing
the corporate profits of firms in the other state.
 22

1.4 Calibration

The arguments in the previous section illustrate the effects of taxing the payroll
and sales factors separately by factors. It shows state governments are faced with the
trade-off when they consider apportionment weights: if a state raises its weight on sales,
then the local wage will go up, but so will the price of final good in the state at the same
time, which harms the state welfare by reducing the real income. Calibration exercises
are necessary to evaluate the total effect and find the optimal tax policy for states. I first
explain the parameters for the calibration, and then report the results.

1.4.1 Parameters

The income distribution among workers is important because it determines how

many workers undertake income shifting. Considering labor income is linear in efficiency
unit of labor and only the highest part of the distribution matters for income shifting
behavior in the model, I assume the distribution of efficiency unit of labor follows a
Pareto distribution.20 The density function of efficiency in state n is defined as

η(lmn )η
f n (l(i)) = ,
(l(i))η+1

in which lmn is the efficiency unit of the least efficient workers in state n and η > 1. I
set η = 5/3 following Jones (2015). Since the mean efficiency in state n is Ln , lmn =
Ln (η − 1)/η.
Federal personal income tax rate is the other determinant for income shifting
behaviour. I replicate its progressive schedule following Gouveia and Strauss (1994).21
20 In particular, the distribution of middle and low income workers does not affect the state welfare
because the welfare function is linear in aggregate income in this model. For the highest part of the
distribution, Piketty and Saez (2013) show the U.S. before-tax income closely follows a Pareto distribution
especially above the income of $400,000.
21 Conesa et al. (2009) and Conesa and Krueger (2006) also use this function to parametrize the optimal
 23

In their model, the effective average tax rate for federal personal income tax is expressed
as:
tNF (yN ) = b2 − b2 (b1 (yN )b0 + 1)−1/b0 , (1.13)

in which yN is labor income. They estimate the parameters in equation 1.13 and report
values of b0 = 0.768, b1 = 0.031, and b2 = 0.258 for year 1989. Note that the limit
of tNF equals b2 as labor income goes to infinity. Since the highest marginal tax rate is
lower in 1989 than today, I use value of b2 = 0.396, which is the highest marginal tax
rate in the current tax schedule, in the following calibration exercises. Since equation
1.13 is not linear in labor income, I have to adjust the nominal price level to replicate
the tax schedule properly. The nominal price level is set such that the mean before-tax
labor income in the model equals $76,000 when the two states are symmetric.22 The
effective tax rate for federal personal income tax by income level is presented in Figure
1.1. Combined with the Pareto distribution of labor efficiency, the share of shifted income
out of total labor income (θ) equals 0.137 when two identical states use a symmetric tax
policy.23
The elasticity of substitution σ in the final good production function is the key
parameter in this model since it affects the elasticity of wage rate with respect to tax
rates and the taxable profits of firms. I use the value of σ = 4 following Fajgelbaum et al.
(2015) as the benchmark. The preference for state government services αG is set at 0.116
so as to make the optimal state corporate income tax rate in the autarky case equal the
federal personal income tax schedule.
22 The mean household income in the U.S. is $75,738 in 2014.
23 As a related statistic, the share of state corporate income tax revenue out of the sum of state personal

and corprate income tax revenue is 12.7%. Thus the value of θ here does not seem extreme. Moreover, the
following result does not change much if I use the parametric function for the effective federal personal
income tax schedule that is adopted by Benabou (2002) and Heathcote et al. (2014):

yN − λ(yN )1−τ
tNF (yN ) = .
yN
 24

0.35

0.3

0.25

0.2

0.15

0.1

0.05

0
0 100 200 300 400 500 600
Labor income ($000)
Figure 1.1: The Effective Average Federal Personal Income Tax Rate

average tax rate of U.S. states weighted by state GDP, which is 7.58% in 2014.

1.4.2 Results

In the calibration exercises, I derive Nash equilibrium of state tax policy by

iteration. I focus on the case of symmetric states and homogeneous firms although my
model allows heterogeneity in the key variables, including the distribution of productivity
of firms across the states. I start with the model of two states, and then extend the model
to the economy of 50 states to calibrate the optimal tax policy for an average state in the
U.S.
The first column of Table 1.1 reports the optimal state tax policy in Nash equi-
librium of the model of two symmetric states. The model suggests the optimal sales
apportionment weight is zero: the state governments should use the full payroll weight.
To examine this result, Panel (a) of Figure 1.2 reports how state welfare, the
local wage rate, and the local price level change as the sales factor weight in one of the
states increases from zero through 100% while the tax policy in the other state is fixed.
 25

Table 1.1: Optimal State Tax Rates in Nash Equilibrium of the Two-State Model
σ 4 3 2 1.5
tCn 6.79% 6.50% 6.65% 7.01%
γnS 0 0 0.603 1
tNn 7.75% 7.77% 7.04% 5.79%

0.02 0.015

0.01
0.01

0.005
0
0
−0.01
−0.005

−0.02 Welfare
Wage rate −0.01
Inversed price level
−0.03 −0.015
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Sales factor weight Sales factor weight
(a) From the Equilibrium of Unrestricted γ (b) From the Equilibrium of γ = 1/3
Figure 1.2: The Effects of Sales Factor Weights on Welfare and Aggregate Variables in
the Two-State Model

The y-axis represents the rate of change in the variables. As expected, the local wage
rate increases as the state puts more weight on the sales factor. If the state adopts the
single sales factor, the local wage rate will go up by 1.9%. However, the local price level
increases at a faster pace than wage rate at the same time. For example, when the single
sales factor is used, the price level goes up by 2.2%. As a whole, the state welfare keeps
decreasing as the state raises its sales factor weight. The degree of decrease in welfare
is equivalent to a 0.93% decrease in private consumption for the change in sales weight
from zero to 100%.
The optimal state corporate income tax rate is lower than personal inecome tax
rate due to the negative effect of corporate income tax on local wage rate. However,
tCn does not deviate far from tNn since the distortion in income shifting behavior and the
resulting increase in federal tax payment discourage states from reducing tC much.
Equation 1.12 implies that the local wage rate is more sensitive to tCn , or the tax
 26

rate on the payroll factor tCn (1 − γnS ) in this context, when σ is lower. This suggests the
positive effect of sales apportionment on the local wage may outweigh its negative effect
on the price level in case of a lower σ. The second through fourth columns of Table 1.1
report the optimal state tax policy under a few different values of σ that are lower than 4.
While the optimal sales weight is still zero even if σ = 3, the optimal weight increases
rapidly to 0.6 and then to 1 as σ goes down to 2 and then to 1.5 respectively. Nonetheless,
to obtain a positive γnS as the optimum requires a considerably smaller value of σ than the
standard values in the trade literature.
Another observation for the various values of σ is that the relationship between
the tax rates for corporate and personal income taxes gets reversed as σ goes down.
The optimal corporate income tax rate is higher than the personal income tax rate when
σ = 1.5. When σ is close to one, the markup rate σ/(σ − 1) is very high, which leads to
large corporate profits of firms. In this case, a state can export a significant portion of
corporate tax liability to the owners of firms who live in the other state. Therefore when
σ is close to one, corporate income tax becomes an attractive tool for revenue for states.
Most of the states in the U.S. used to follow the Multistate Tax Compact that rec-
ommended states should adopt the equally weighted formula. Thus it is worth examining
how Nash equilibirum looks if γnS is fixed at one third rather than considered as one of
the choice variables of states. In this case, tCn = 7.16% and tNn = 7.34% in equilibrium. If
states start to be allowed to freely choose the value of γnS suddenly, for example due to
the ruling by the Court, adopting zero sales weight is the best response in this case too as
Panel (b) of Figure 1.2 shows. The panel presents the result of same simulation as Panel
(a) but starting from the Nash equilibrium of fixed γnS being equal to one third. Even
though the local wage rate will decrease if state lowers γnS from one third, the positive
effect of decrease in the price level outweighs the effect on wage rate.
Although the current model has described the economy of two states, it is readily
 27

extended to the economy of many states if they are symmetric. To simulate results for
an average state in the U.S., I also calibrate the Nash equilibrium of the model of 50
symmetric states. Table 1.2 reports the optimal tax policy in Nash equilibrium of the
50-state economy for various values of σ. When σ equals 4 or 3, the optimal sales weight
is zero again while tCn is lower and tNn is slightly higher than the two-state model. In this
economy, the optimal sales weight is zero even when σ equals as low as 2. If the sales
apportionment weight is exogenously fixed at one third, the two tax rates come closer as
in the two-state economy: tCn = 6.70% and tNn = 7.22%.

Table 1.2: Optimal State Tax Rates in Nash Equilibrium of the 50-State Model
σ 4 3 2 1.5
tCn 6.18% 5.82% 5.16% 7.82%
γnS 0 0 0 1
tNn 7.88% 8.02% 8.59% 3.22%

Figure 1.3 examines how state welfare, local wage rate, and local price level
change as sales weight deviates from the equilibrium value. The shape of curves in both
panels look very similar to those in Figure 1.2. However, the magnitude of changes is
almost doubled compared to the two-state case. For example, when the sales apportion-
ment is not restricted, raising the sales weight from zero to one increases wage rate by
3.6% and price level by 4.3%.

1.5 Discussions

Goolsbee and Maydew (2000) is one of the first rigorous empirical studies that
estimate the impact of payroll apportionment tax on the local labor demand. They use
extensive panel data to find the statistically significant impact of apportionment weights
on the local labor demand: if the sales weight is raised from one third to one half and the
payroll and property factors are lowered accordingly, manufacturing employment in the
 28

0.04 0.03

0.02
0.02

0.01
0
0
−0.02
−0.01

−0.04 Welfare
Wage rate −0.02
Inversed price level
−0.06 −0.03
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Sales factor weight Sales factor weight
(a) From the Equilibrium of Unrestricted γ (b) From the Equilibrium of γ = 1/3
Figure 1.3: The Effects of Sales Factor Weights on Welfare and Aggregate Variables in
the 50-State Model

state goes up by 1.1%.24 Their result provides some test of how relevant the model of
this study is to the U.S. economy. The dataset of Goolsbee and Maydew (2000) covers
the period from 1978 through 1994. Most states used the equally weighted formula at the
beginning of the period, and then more and more states started putting a greater weight
on the sales factor towards the end of the period. Thus I choose Nash equilibrium of the
model of 50 states in which all the states are required to set γnS = 1/3 and look at the
growth rate of local wage rate when one of the states deviates from the equilibrium by
raising γnS from one third to one half. My model predicts the growth rate equals 0.62%.
This value is not equal to 1.1%, the point estimate of Goolsbee and Maydew (2000), but
it is in the same order and within one standard error of their point estimate.25 Basically,
the prediction of my model is not far from their estimate even though the model is meant
to be parsimonious for transparency and tractability.
24 Goolsbee and Maydew (2000) calculate this result based on their estimate for the elasticity of
manufacturing employment with respect to state payroll tax burden (−1.92) and the mean corprate income
tax rate in their dataset (7.3%). Although the mean tax rate predicted in the equilibrium of my model is
slightly different (6.7%), the result of calculation replacing their mean tax rate with the predicted rate is
not significantly different from the value of 1.1%.
25 Moreover, Goolsbee and Maydew (2000) find that the estimate for elasticity of non-manufacturing

employment is smaller than manufacturing. If the average of those estimates are taken, their estimate will
be closer to the prediction of my model.
 29

The use of some continuous distribution to represent heterogeneity in locations

and firms is standard in the recent trade literature.26 Although the calibration in the
previous section assumes homogeneous productivity across states and firms, the model
can accommodate such distributions in productivity. Firms may have some idiosyncratic
preferences about their location as in Fajgelbaum et al. (2015) and Suárez Serrato and
Zidar (2014); for example, an oil company may be more profitable in Alaska, and some
kind of agricultural production more suitable in Iowa. If such heterogeneity is brought
into my model, the elasticity of the number of firms in a state with respect to its corporate
income tax rate will be lessened compared to the case of perfectly homogeneous firms.
This leads to a smaller positive effect of reducing the payroll factor tax rate on local labor
demand, and the positive effect is less likely to outweigh the negative effect of higher
price level. Therefore the assumption of homogeneous firms used in the calibration can be
considered as the most favorable for a positive sales weight. Since the calibration suggests
there is no point of using a positive sales weight under the assumption of homogeneous
firms, a positive sales weight will not be justified in the model of heterogeneous firms
either, at least under the baseline value of σ = 4.
Another restrictive assumption of the model is that workers cannot move across
state borders. But this assumption is also the most favorable for a positive sales weight.
The mobility of workers can be introduced in the model, for example by heterogeneous
preferences of workers regarding locations. The mobility, however, will reduce the gap
in local wage rates between states with different tax rates. Thus a decrease in payroll
apportionment tax will raise the wage rate less than the model of immobile workers.
Moreover, the assumption of immobile workers has another advantage; the model can
avoid the issue of how to define the state welfare with changing population.
The complexity of U.S. state tax rules that is beyond the scope of the current
26 The classic examples from this literature include Eaton and Kortum (2002) and Melitz (2003).
 30

study includes tax nexus and throwback rules. The U.S. state tax rules do not allow a
state to impose corporate income tax on a firm unless the firm establishes nexus in the
state; for exapmle, a firm does not establish nexus if it has no contact with a state except
for soliciting sales of tangible products in the state. In addition, the majority of U.S.
states have the throwback rule; under the rule, the sales of tangible goods are counted as
sales in the origin state if the seller does not establish nexus in the destination state.
If a perfectly competitive retail industry exists in the economy and if producing
firms do not have to perform any business activity in the destination states, then producing
firms sell all the goods to retail companies in a state that has zero tax rate for the sales
factor. The retail companies distribute the goods to the final consumers across states. If
this story is true, firms never report positive sales to states with a positive tax rate for the
sales factor as long as there exists a state that adopts zero rate for the sales factor. However,
Edmiston and Arze del Granado (2006) examine the data set of corporate income tax
returns filed to the State of Georgia by multistate firms and report the share of sales in
Georgia is 4.4% in 1992, when the corporate income tax rate and sales apportionment
weight in the state were 6% and 1/3 respectively. Moreover, they estimate the sales share
in Georgia decreased by 6.3% when the state raised the sales apportionment weight to
1/2. Thus firms do not seem to simply minimize the tax burden on the sales factor by
concentrating their nexus in a zero-rate state, but they have nexus in various states for
some reasons and respond to a change in apportionment formula of a state in a nontrivial
way.
In this context, my model can be interpreted in such a way that firms establish
nexus in all the states that they sell goods to, for example by setting up an office at an
infinitesimal cost. In reality, probably it is possible for firms in some industries to sell in
states without establishing nexus, for example online retail companies, but it is hard for
some industries to do so, for example, car makers. Therefore it is necessary to extend
 31

the model to incorporate nexus choice of firms and heterogeneity across industries in
terms of production locations if one wants to include these aspects of U.S. tax rules in
the analysis.

1.6 Conclusion

Recently, increasing the sales apportionment weight, including adopting the single
sales factor, is a popular policy choice among U.S. states, and its effect of stimulating the
local labor demand is well recognized in the literature. However, its possible negative
effect on the local price level is often overlooked in policy discussions. This paper
searches for the optimal state tax policy including apportionment weights and predicts
Nash equilibrium of multiple states with a model that incorporates these effects of
apportionment formula. Contrary to the popular argument, the model suggests the
optimal sales weight should be zero even from the perspective of a single state because a
positve sales weight increases the local price level more than the local wage rate. This
result is consistent under a wide range of plausible parameters.
Two explanations are possible to reconcile the model’s suggestion and the domi-
nant trend among U.S. states. First, the negative effect of sales apportionment tax may
actually be overlooked by state policy makers and constituents. The positive effect on
local labor demand is often visible; for example, some firms may open new plants or
cancel layoffs because of a change in tax policy. However, changes in the price level are
harder to detect. It requires a rigorous empirical study to find out the causality between
state tax policy and the local price level.
The second possibility is that this model may not capture how tax rules affect
firms’ decision about production and sales perfectly and overestimate the negative effect.
Since the tax rules of U.S. states are complex as discussed in the previous section, firms
 32

may be actively avoiding the tax burden on sales apportionment by adjusting nexus,
though not so perfectly as the simple story predicts. To verify these hypotheses, empirical
studies are needed to find out how firms allocate and report sales shares across states in
practice.

1.7 Appendix

1.7.1 Profit maximization problem for firms

This subsection proves that all the firms in the same state choose the same
allocation of sales shares among the states. The proof is a variant of the one by Fajgelbaum
et al. (2015). If firm j chooses to operate in state n, the profit maximization problem for
firm j is:

max π̃n ( j)
{pn ( j),xn ( j)}

−σ Ym
subject to xnm ( j) = pm
n ( j) for m = {A, B},
(Pm )1−σ

in which π̃n ( j) is defined in equation 1.5. Dividing the first order condition of π̃n ( j) with
respect to pAn ( j) by pAn ( j)−σ−1Y A (PA )σ−1 and using the constraint give:

wnN w̃n
 
(1 − tCF A
− t¯n ( j)) (1 − σ)pn ( j) + σ(1 − θn ) + σθn C
zn ( j) zn ( j)
h i pA ( j)
− (1 − σ) tSA − (tSA sAn ( j) + tSB sBn ( j)) n πTn ( j) = 0, (1.14)
Sn ( j)

in which tSm = tCm γm

S , the effective tax rate for the sales factor in state m; Sn ( j) is the total

sales of firm j; sm
n ( j) is the share of sales in state m out of Sn ( j). Solving equation 1.14
 33

for pAn ( j) by using w̃Cn = (1 − tCF − t¯n ( j))wnN gives:

1 σ wnN
pAn ( j) = , (1.15)
1 − t˜nA (πTn ( j)/Sn ( j)) σ − 1 zn ( j)

in which
tSA − (tSA sAn ( j) + tSB sBn ( j))
t˜nA = . (1.16)
1 − tCF − t¯n ( j)

The taxable profit can be expressed as:

wnN
 
πTn = Sn ( j) ∑ sm
n ( j) 1 − (1 − θn ) . (1.17)
m={A,B}
zn ( j)pmn ( j)

Substituting equation 1.15 into equation 1.17 gives πTn = Sn ( j)[1 + θn (σ − 1)]/σ. This
implies
σ σ wnN
pAn ( j) = . (1.18)
σ − t˜nA [1 + θn (σ − 1)] σ − 1 zn ( j)

Finally, note that the sales shares are independent of productivity, zn ( j):

pAn ( j)1−σȲ A
sAn ( j) = 1−σȲ m
∑m={A,B} pm n ( j)
{σ − t˜nA [1 + θn (σ − 1)]}σ−1Ȳ A
= , (1.19)
∑m={A,B} {σ − t˜nm [1 + θn (σ − 1)]}σ−1Ȳ m

in which Ȳ m = Y m (Pm )σ−1 . The symmetric equation holds for sBn ( j). Equations 1.16 and
1.19 and the corresponding equations for state B define a system for {t˜nm } and {sm
n ( j)}

whose solution is independent from zn ( j). Threfore sm m

n ( j) = sn and t¯n ( j) = t¯n for all the

firms in state n. This result can easily be extended to the case of more than two states.
 34

1.7.2 Proof for Proposition 1

From equations 1.7, 1.8 and 1.9, the problem for the government of state n is:

ˆ
ỹn (i) n αG n
max n
(g ) f (i)N n di
n n
tC ,tN i∈n P

subject to N n Pn gn = znG T n .

If the before-tax aggregate income is normalized to one, as in Section 1.3.1, it can be

shown Π = 1/σ and
l(i)
d˜n (i) = (1 − tCF − tCn ) n n , (1.20)
σN L

in which Π is the before-tax aggregate markup. By substituting equations 1.4, 1.6 and
1.20, dropping the variables that are exogeneous for the state government, and changing
the variable for integration from i to l(i), the state government’s problem can be rewritten:

"ˆ
∞
max
n n
(T n )αG (1 − tCF − tCn )wnN l fLn (l)dl
tC ,tN l¯
ˆ l¯
#
1 − tCF − tCn
+ (1 − tNF (l) − tNn )wnN l fLn (l)dl +
0 σN n

ˆ ∞ ˆ l¯
n 1
subject to T = tCn wnN N n l fLn (l)dl + tNn wnN N n l fLn (l)dl + tCn ,
l¯ 0 σ

in which fLn (l) is the probability density function of labor efficiency in state n; l¯ is the
¯ + tn =
cutoff labor efficiency for income shifting, which is defined implicitly as tNF (l) N

tCF + tCn . I assume the federal personal income tax schedule tNF (l) is differentiable and
∂tNF /∂l is strictly positive everewhere. Dividing the first order condition with respect to
 35

tNn by that with respect to tCn gives:

´ l¯ n n n
∂T n /∂tNn w N l f (l)dl
n n = ´ ∞ 0n Nn n L . (1.21)
∂T /∂tC l¯ wN N l f L (l)dl + 1/σ

The partial derivatives of T n with respect to tNn and tCn are:

ˆ l¯
∂T n ∂l¯
n = wnN N n l fLn (l)dl + (tNn − tCn ) n wn N n l¯ fLn (l);
¯ (1.22)
∂tN 0 ∂tN
ˆ ∞ ¯
∂T n n n n 1 n n ∂l n n ¯ n ¯
= w N N l f L (l)dl + + (tN − tC ) w N l fL (l). (1.23)
∂tCn l¯ σ ∂tCn

From equations 1.22 and 1.23, note that equation 1.21 holds when tNn = tCn . Therefore
the optimal rates for state personal and corporate income taxes are equal in the case of
autarky.

1.7.3 Proof for Proposition 2

Since the assumptions made in Proposition 2 constitute a special case of Section

1.7.1, equation 1.18 still holds:

σ σ wN
pAn ( j) = A
. (1.24)
˜
σ − tn [1 + θn (σ − 1)] σ − 1 z

From equation 1.16, note t˜nA = 0 if tCA = tCB . Then equation 1.24 implies:

 
σ wN
pAn ( j) = . (1.25)
σ−1 z

Since equation 1.25 holds for all j and does not depend on the state, the first statement of
Proposition 2 holds.
Equation 1.16 implies t˜nB < 0 < t˜nA if tCA > tCB . It is straightforward from equation
 36

1.24 to show
 
σ wN
pBn < < pAn .
σ−1 z

Therefore the second statement of Proposition 2 holds.

Chapter 2

Risk-Taking and Growth under

Taxation with Moral Hazard

Abstract

We study the effect of taxation on entrepreneurs’ risk-taking, portfolio choice,

and economic growth in the presence of a moral hazard problem in a dynamic general
equilibrium model. The model economy is populated by a continuum of agents who
invest capital in projects that are subject to aggregate and idiosyncratic investment risks.
The moral hazard problem occurs because the return of a type of project depends on
entrepreneurs’ effort, which is private information. By collecting proportional consump-
tion tax and redistributing the revenue, the government offers an opportunity for risk
sharing to encourage risk-taking and spur economic growth. When the moral hazard
problem is absent, the full insurance is optimal in terms of welfare, and the economy
grows faster. But if the moral hazard problem exists even in a slight degree, risk sharing
through taxation cannot improve welfare.

37
 38

2.1 Introduction

It has long been recognized that taxation can encourage risk-taking since the
seminal work of Domar and Musgrave (1944). This somewhat surprising view holds
because the government can share the risk of investment as a “partner” for private
entrepreneurs. Since entrepreneurial risk-taking is commonly considered as an important
driver of macroeconomic growth, it is a natural direction of research to embed this
intuition regarding taxation and risk-taking in the context of economic growth. Some of
the subsequent studies extended the literature by examining this intuition in stochastic
endogenous growth models (Eaton, 1981; Asea and Turnovsky, 1998; Kenc, 2004).
In the literature of corporate finance, the study of entrepreneurial activity stresses
the importance of incentive problems, including the moral hazard problem of entrepreneurs.
For example, if the performance (e.g., expected return) of entrepreneurial investment
depends on the degree of entrepreneur’s effort and if the effort is not publicly observ-
able, it will lead to distortion for socially optimal risk-taking. Although it is easy to
expect that the moral hazard problem will alter the established conclusions about taxation
and risk-taking, however, the literature has not formally incorporated this problem into
the analytical framework until recently, especially in the setting of dynamic general
equilibrium.1
Our study presents quantitative evaluations about how introducing a moral hazard
problem changes the policy implications regarding taxation and risk-taking using an
overlapping generations, incomplete-market dynamic general equilibrium model. In the
model, there is a continuum of ex ante identical agents (entrepreneurs) who are subject
1 There are some exceptions. The literature has argued for no loss offset rule, which is commonly
adopted in many developed countries, based on an incentive problem; the proposed rationale is that
entrepreneurs try to mingle personal consumption with investment losses for tax purposes if the full offset
is allowed in the tax code, while it is difficult for the government to detect this behaviour (Atkinson and
Stiglitz, 2015, Poterba, 2002). Atkinson and Stiglitz (2015), which is originally published in 1980, also
present a simple model in which there is information friction between “capitalists” and “managers” and
capital income taxation tends to reduce risk-taking (p. 97-99).
 39

to uninsurable idiosyncratic investment risks. Agents can invest their capital in two
technologies, one with only aggregate risk (interpreted as the stock market), and another
with both aggregate and idiosyncratic risks (interpreted as private equity) in the absence
of government intervention. The expected return on private equity investment depends
on entrepreneur’s effort, which is private information since the effort is not observable.
When the government intervenes in the allocation of risk, the choice of the effort level
becomes a source of moral hazard problem.
We consider a rather non-traditional policy tool instead of capital income tax,
which has drawn attention in the related literature. The government in our model imposes
a flat-rate consumption tax and supplies a financial instrument—which we simply refer to
as a “government bond”—which is the right for receiving an equal share of government
tax revenue in every period in the future. Agents are allowed to trade this government
bond at any point in time and allocate their wealth among the two real investment
technologies, the government bond, and current consumption. The combination of
consumption tax and government bond provides agents with an opportunity to insure
themselves against private idiosyncratic risk, for which there is no private insurance
market.
If the outcome of private investment did not depend on effort, in other words,
if there were no moral hazard problem, the welfare would be maximized by the full
insurance through the government fiscal policy. The optimal consumption tax rate would
be infinite (equivalent to 100% income tax), and the economy would grow at a higher rate.
That is because the risk sharing through government intervention perfectly eliminates
the idiosyncratic risk of private equity (but not its aggregate risk) to make the investment
less risky, and agents invest more on private equity, which is assumed to have a higher
expected return than the stock market. The government intervention improves the welfare
of agents by raising the expected utility in the future due to a faster economic growth as
 40

well as by the direct effect of risk sharing.

Once the moral hazard problem is considered, however, the conclusion dramat-
ically changes. With an arguably low elasticity of effort, which is translated as the
elasticity of the expected return of private equity with respect to net-of-tax rate to be 0.03
in equilibrium,2 our quantitative model implies that the optimal tax rate for consumption
tax should be 63%, which is far from the full insurance. Since no other tax is imposed in
our model, this tax rate can be regarded as being equivalent to 39% proportional income
tax. This result is brought about because risk sharing through taxation exacerbates the
moral hazard problem and makes agents exert less effort on private equity, even though
taxation still successfully encourages agents to allocate a larger share of their wealth to
private equity (the riskier asset) rather than the market asset. Moreover, with a slightly
higher elasticity (0.06), we find risk-sharing through taxation cannot lead to a Pareto
improvement among all the generations. While taxation at a low rate can encourage
risk-taking, spur economic growth, and improve the welfare of future generations, the in-
tratemporal wedge caused by the moral hazard problem hinders the welfare improvement
for the initial generation. This result sheds light on the important issue that previous
studies have not addressed: more risk-taking induced by taxation does not necessarily
coincide with welfare improvement. Therefore it is essential to take into accout of the
moral hazard problem explicitly when the effects of taxation on risk-taking are studied.
Another feature of our model is that it endogenously generates the cross-sectional
wealth distribution that is consistent with the stylized observation, as the stationary
distribution of dynamic general equilibrium model that accommodates heterogeneity.
Thus we can see how the combination of consumption tax and government bond affects
the stationary distribution of wealth. The cross-sectional wealth distribution generated
by our model is double Pareto—a distribution with two Pareto tails—and therefore the
2 For the difinition of net-of-tax rate in this study, see footnote 14 in subsection 2.4.3.
 41

inequality is captured by the Pareto exponent. Because the government intervention

enables more risk sharing, it reduces the idiosyncratic volatility in individual portfolios,
and therefore reduces inequality (increases the Pareto exponent).
Our paper adds new quantitative insights to the literature of taxation and risk-
taking. After Domar and Musgrave (1944) pioneered the research in this area, Mossin
(1968) and Stiglitz (1969) formalized the analysis in the expected utility framework.
These studies conclude capital income taxation with full loss offset encourages risk-
taking in the economy under certain types of utility functions including the Constant
Relative Risk Aversion utility function. In addition, Atkinson and Stiglitz (2015) show the
conclusion remains true in a broader environment when the tax revenue is redistributed to
households, eliminating idiosyncratic risk due to the law of large numbers, rather than is
used in a way that it affects households’ utility separately (or does not affect at all). While
these studies take a partial equilibrium approach,3 the next generation of studies in this
literature examine how taxation affects the allocation of risk between private and public
sectors in general equilibrium models, particularly considering that the government
budget constraint also becomes stochastic when the government collects revenue from a
tax on stochastic returns from investment. Gordon (1985) shows corporate income tax
leaves risk-taking in the economy unchanged in a general equilibrium model. He argues
that the government can be interpreted as collecting a fraction of expected return from
private investment and charging the market price for risk sharing that it offers to investors
in his model, in which the government is no better than private markets at spreading
risk. The AK type endogenous growth models of Eaton, 1981, Asea and Turnovsky,
1998, and Kenc, 2004 assume the government are required to spend a constant fraction
of macroeconomic output and imply the government risk sharing does not necessarily
increase aggregate risk-taking.4
3 Sandmo (1985) includes detailed review of this strand of literature.
4 While empirical examination about whether taxation encourages risk-taking is difficult because of
 42

Our study deviates from this line of the literature towards a new direction. In our
general equilibrium model, the government can definitely improve the welfare through
taxation if there is no moral hazard problem. Since we assume there is no private market
for insurance against idiosyncratic risk of private equity investment, the government has
a power to offer an insurance against the risk by means of taxation. Instead of allocation
of risk between government and private sectors, we focus on how the presence of moral
hazard problem undermines the power of government. Our results suggest the moral
hazard problem reduces the power of government significantly.
Our paper also provides support for the relevance of the recent literature called
New Dynamic Public Finance. The studies in the literature look for the conditions for
optimal taxation in the presence of information friction in dynamic settings. Although
most of the studies examine cases in which labor skills are private information (Golosov et
al., 2003; Kocherlakota, 2005), Albanesi (2011) addresses the optimal taxation problem
for entrepreneurs. In her model, entrepreneurs’ effort is private information while
entrepreneurs are ex ante identical, and it affects the expected return of investment, as in
our model. Admittedly our approach is different from that of the New Dynamic Public
Finance literature, including Albanesi (2011), notably because we restrict our attention
to a specific tax structure while the studies of New Dynamic Public Finance search
broader class of tax instruments only considering the constraint of private information.5
Nonetheless, we view our study has valuable implications in support for this literature.
The quantitative results from our model show the optimal tax policy varies substantially
according to the degree of moral hazard problem and underscore the importance of taking
into account the private information issues when we study the tax policy regarding risky
investment.
complicated incentives due to the provisions in the tax code, Cullen and Gordon (2007) provide convincing
evidence that tax provisions have large effects on the amount of risk-taking.
5 This point is well articulated by Golosov et al. (2007).
 43

This paper is also related to the small but growing literature on dynamic general
equilibrium models with heterogeneous agents that are analytically solvable and generate
heavy-tailed cross-sectional distributions, such as Benhabib et al. (2011) and Benhabib
et al. (2016), Toda (2014), and Aoki and Nirei (2014). In all of these papers, agents are
subject to uninsurable investment risks, and the upper tail of the cross-sectional wealth
distribution obeys the power law (consistent with empirical findings). In Benhabib et al.
(2011), agents live for fixed finite periods and are subject to both idiosyncratic capital
and labor income risks, and have a bequest motive. They study the effect of estate tax
and redistribution on inequality. In Benhabib et al. (2016), agents die at a constant
probability of death and receive means-tested subsidies financed by capital taxation. The
cross-sectional distribution is double Pareto, which has two Pareto tails. The double
Pareto distribution has been shown to robustly arise in a large class of dynamic general
equilibrium models (Toda, 2014) as well as fit the U.S. income and consumption data
well (Toda, 2012; Toda and Walsh, 2015). Aoki and Nirei (2014) jointly study the firm
size distribution and top income distribution and obtain the transitory dynamics of the
top income share when the marginal tax rate changes.
Our paper is closest to Benhabib et al. (2016) in that the model features a redis-
tributive fiscal policy and a stationary double Pareto distribution. There are important
differences, however. First, in our paper the fiscal tools are proportional comsumption
tax and tradable government bond as opposed to means-tested subsidies. Second, our
model features the moral hazard problem, aggregate shocks, and endogenous asset prices.
This point, although it may seem only technical at first glance, is important because
without aggregate shocks the government’s fiscal policy does not affect the asset span.
With aggregate shocks, the government bond will typically increase the asset span and
improve welfare. Finally, we study the interaction of moral hazard problem and risk
sharing through taxation affects growth, welfare, and inequality and derive the optimal
 44

tax rate.

2.2 Model

The model is a continuous-time stochastic growth model (AK model) with het-
erogeneous agents and a government.6 Time is continuous and is denoted by 0 ≤ t < ∞.
At t = 0, there is a continuum of ex ante identical agents with mass 1. Each agent dies at
a constant Poisson rate δ > 0. As soon as an agent dies, a new agent is born (Yaari, 1965;
Blanchard, 1985). The agents are not altruistically connected, and there is no bequest
motive.

Preference Preferences are represented by the continuous-time analog of the Epstein-

Zin constant relative risk aversion (CRRA), constant elasticity of intertemporal substi-
tution (CEIS) utility function with discount rate β > 0, relative risk aversion γ > 0, and
elasticity of intertemporal substitution ε > 0 with disutility from effort. More precisely,
the continuation value J satisfies

ˆ ∞
J(t, w(t)) = Et f (c(s)h(e(s)), J(s, w(s)))ds, (2.1)
t

where
1−1/ε
β + δ x1−1/ε − ((1 − γ)y) 1−γ
f (x, y) = (2.2)
1 − 1/ε γ−1/ε
((1 − γ)y) 1−γ

is the aggregator,7 c(t) is consumption, w(t) is wealth, and h(e(t)) is the disutility from
effort e(t).
6 Saito(1998) studies the asset pricing implication of idiosyncratic investment risks in a similar settings.
Toda (2014) studies a general AK model in discrete time without a government. Toda (2015) considers
the optimal taxation of physical and human capital in a Markov setting but without intergenerational
considerations or moral hazard issues.
7 See Duffie and Epstein (1992) for details.
 45

Technologies There are two types of technologies in which agents can invest their
capital. The first is the aggregate stock market, which is interpreted as a stochastic
constant-returns-to-scale (AK) technology. The capital invested in the stock market
evolves according to
dK/K = µm dt + σm dBm , (2.3)

where µm is the expected return, σm > 0 is volatility, and Bm is a standard Brownian

motion.
The second is the private equity, which is also an AK technology but requires
effort to operate. The capital invested in agent i’s private equity evolves according to

dK/K = µ p (e)dt + σ p dB p + vdBi , (2.4)

where µ p (e) is the expected return when the effort level is e, B p and Bi are independent
standard Brownian motions that represent the aggregate and idiosyncratic shock to capital,
and σ p , v > 0 are their volatilities. Each agent can only invest in his own private equity.
Bi is assumed to be i.i.d. across agents and is uninsurable. Note that only the expected
return µ p (e) depends on the effort level. This is because if the idiosyncratic volatility v
also depends on effort, since v is observable by calculating the quadratic variation of the
value of invested capital, it is possible to back-out the effort level. Thus it is necessary
that v is independent of e so that e remains unobservable.8
The stock market and the private equity may well be correlated: let

dBm dB p = ρdt,
8 There is an additional reason that we leave v independent of effort. Even if less effort were associated
with larger v for the purpose of representing the moral hazard problem in another dimension, it would not
harm the attractivenss of government bond as a financial asset; the government can perfectly eliminate the
fraction of variation in its tax revenue that is attributed to the idiosyncratic risk of private equity by the law
of large numbers because of the i.i.d. assumption.
 46

where −1 < ρ < 1 is the correlation coefficient between the stock market and the aggre-
gate component of private equity.
This specification of the technologies is not the most general one for which we
can obtain closed-form solutions. The model is equally tractable even if aggregate capital
and private equity jointly enters the production function nonlinearly, provided that it is
homogeneous of degree 1.9 The reason why we assume that the stock market and private
equity are each AK technologies is because with nonlinearities, the equilibrium becomes
constrained inefficient (Toda, 2014), and hence the government can improve welfare
even without introducing new assets. With AK technologies and no effort choice, the
equilibrium is constrained efficient (Toda, 2014), so any welfare improvement from the
government intervention will be entirely due to the expansion in the asset span.

Government bond To provide agents with an opportunity to insure against the idiosyn-
cratic investment risk, for which there is no private market, the government issues bond
and redistributes tax revenue to the owners of bond. Although the government bond is
handed out equally to each agent in the economy at the initial time point, agents can
trade government bond afterwards. While the simplest way of redistribution would be to
give an equal share of tax revenue to each agent in every period over lifetime (demogrant
or “basic income”), the tradable government bond allows us to focus on the pure effect
of extending the asset span by government fiscal policy on entrepreneurial risk-taking
without wealth effect.
In our model, the choice of tax strucute is also different from the standard ap-
proach in the literature on taxation and risk-taking. The government imposes proportional
consumption tax on every agent rather than capital income tax. The reason for this choice
is to make information friction as transparent as possible. The rules about loss offset has
9 See Angeletos (2007) and Toda (2014) for such models.
 47

been one of the most discussed topic in the literature since Domar and Musgrave (1944).
Although the studies mostly agree that allowing loss offset encourages risk-taking, at
least in a partial equilibrium setting, many developed countries adpot no offset rule. The
common rational for no offset rule proposed in the literature is that it is difficult for
government to distinguish between losses from risky investment and private consumption
(Atkinson and Stiglitz, 2015; Poterba, 2002). Thus capital income tax would cause
another type of moral hazard problem in which agents try to take advantage of loss offset.
On the contrary, consumption tax is immune to this type of problem and allows us to
focus on the moral hazard problem due to entrepreneurial effort. A consumption tax
(value-added tax) is also attractive from an administrative point of view (especially in
developing countries) since the invoice system discourages tax evasion at a low admin-
istrative cost, at least as long as the tax rate is not so high that people start to engage
in barter trades. It is a formidable task for the government to capture captial income
of individuals from all sources, espcially in the era of global integration. In fact, the
value-added tax is a major source of revenue in most developing and developed countries
(except in U.S., which does not have a nation-wide value added tax). Another advantage
of taxing consumption is that it is a better proxy of permanent income than just income.
Throughout this paper, we assume that the government taxes consumption at a
flat rate τ and distribute the entire tax revenue to the owners of government bond, whose
unit is normalized such that the aggregate supply equals unity at all time. Therefore if
aggregate consumption is C, the dividend of government bond is τC per unit.

Financial assets and transfers In addition to the two types of physical capital and
government bond, agents can trade a risk-free asset in zero net supply, whose risk-free
rate r is determined in equilibrium. If there is no aggregate risk (so σm = σ p = 0),
then the government bond will be risk-free and therefore will have the same rate as the
 48

risk-free asset.
As mentioned earlier, each agent dies at constant Poisson rate δ > 0 and gets
reborn. Assets are transfered between deceased and newborn agents as follows. First,
there is an annuity market for stock and risk-free asset: an agent receives an annuity δ
per asset holdings when alive and transfers the asset to the annuity company upon death.
This arrangement is optimal from agents’ point of view and the annuity company breaks
even. Second, private equity cannot be pledged for annuity, possibly because in reality
the value of private equity may be difficult to observe or evaluate, or the value of a private
business may be highly dependent on the ability of the entrepreneur. When an agent
dies, all of her capital in private equity is transfered to the government, which in turn
is split equally among newborn agents. This transfer might be broadly interpreted as
public goods such as mandatory elementary education. Third, the government bond is
not pledgeable for annuity either. Although the government bonds are tradable among
agents, all shares must be returned to the government upon death.10 The government
then gives 1 share of the bond to each newborn agent.

Equilibrium The definition of the competitive equilibrium is standard: given the

consumption tax rate τ, an equilibrium consists of individual choices (consumption,
effort, and portfolio) and asset prices (risk-free asset and government bond) such that (i)
individual choices are optimal, (ii) asset markets clear, and (iii) the government budget is
balanced.
10 This rule might be difficult to enforce in practice, since if agents have some private information that
predicts death (e.g., sickness), those agents will have the incentive to sell off the bond before death. A
practical solution would be to make the bonds nontradable (which is actually what is happening in reality
since social security benefits cannot be used as collateral), although it will make the theoretical analysis
difficult since we lose homogeneity. At least theoretically, making the bonds tradable and requiring to
return upon death cause no difficulty since the death rate δ is constant.
 49

2.3 Solving for equilibrium

This section explains how to solve for the equilibrium. Let C(t) denote the
aggregate consumption. From the structure of the model, C is some geometric Brownian
motion
dC/C = µc dt + σc dBc , (2.5)

where the drift µc and volatility σc are determined in equilibrium and Bc is some combi-
nation of the fundamental shocks Bm and B p . Let P(t) be the price of the government
bond and D(t) be the dividend. Since consumption is taxed at a flat rate τ and the tax
revenues are paid out as dividend to the government bond, we have

D(t) = τC(t).

From the structure of the model, the government bond price must be proportional to
aggregate consumption. Therefore P(t) follows the same geometric Brownian motion as
aggregate consumption (except for the difference in levels), (2.5). Since the government
bond pays out dividend, its instantaneous total return is

dP D
+ dt = (µc + d)dt + σc dBc , (2.6)
P P

where d = D/P is the dividend yield to be determined in equilibrium. Let µg = µc + d be

the expected total return on the government bond.
 50

2.3.1 Individual problem

By (2.3), (2.4), and (2.6), the budget constraint of agent i is

dwi /wi = ((µm + δ)θ1 + µ p (e)θ2 + µg θ3 )dt + (1 − θ1 − θ2 − θ3 )(r + δ)dt

+ θ1 σm dBm + θ2 (σ p dB p + vdBi ) + θ3 σc dBc − (1 + τ)mdt, (2.7)

where m is the propensity to consume out of wealth (m = ci /wi , where ci is the con-
sumption rate), wi is wealth, (θ1 , θ2 , θ3 ) are the fraction of wealth invested in the stock,
private equity, and government bond, r is the risk-free rate, δ is the insurance premium
on mortality risk, and τ is the consumption tax rate. Note that δ appears only in the terms
corresponding to the stock market and the risk-free asset because those are the only assets
insurable against the mortality risk. For notational simplicity, let
     
θ1  µm + δ  σm dBm 
     
θ=
θ2  , µ(e) =  µ p (e)  , dX = σ p dB p + vdBi  ,
    
     
θ3 µg σc dBc

and define the instantaneous variance matrix Σ by (dX)(dX)0 = Σdt. Then the budget
constraint (2.7) becomes

dwi /wi = (r + δ + θ0 (µ(e) − r − δ) − (1 + τ)m)dt + θ0 dX. (2.8)

Maximizing the recursive utility defined by (2.1) and (2.2) subject to the budget
constraint (2.8) is a standard Merton (1971)-type optimal consumption-portfolio problem
except for the presence of an effort choice, a consumption tax, and recursive preferences.11
Before solving the individual problem, we impose a few equilibrium conditions in order
11 Svensson (1989) solves the optimal portfolio problem with Epstein-Zin preferences.
 51

to simplify the computation of equilibrium. First, since the utility function as well as the
budget constraint (2.8) are homothetic in wealth, the optimal consumption rate, effort,
and portfolio choice are common across all agents. Since the risk-free asset is in zero
net supply, there will be no trade in the risk-free asset in equilibrium. Therefore it
must be ∑3j=1 θ j = 1. Furthermore, since technologies cannot be operated in reverse
and the government bond is in positive net supply, we have θ j ≥ 0 for all j. Let
n o
∆ = θ ∈ R3+ ∑3j=1 θ j = 1 be the set of admissible portfolios.
The Hamilton-Jacobi-Bellman (HJB) equation of the optimal consumption-effort-
portfolio problem is
0 = max [D J + f (mh(e)w, J)] , (2.9)
θ,m,e

where
1
D J = Jt + wJw (θ0 µ(e) − (1 + τ)m) + w2 Jww θ0 Σθ.
2

(Here the risk-free rate does not appear because θ ∈ ∆.) By the homotheticity of the
utility function, it should be clear that the value function takes the form

a1−γ 1−γ
J(w) = w
1−γ

for some a > 0. Substituting this functional form into (2.9) and retaining terms that
contain θ, the optimal portfolio is the maximizer of

γ
µ(e)0 θ − θ0 Σθ, (2.10)
2

which is quadratic in θ ∈ ∆.
 52

The first-order condition with respect to the consumption rate m is

β+δ
− (1 + τ)wJw + γ−1/ε
(mh(e)w)−1/ε hw = 0
((1 − γ)J) 1−γ
 ε  ε  1−ε
β+δ h(e)ε−1 β+δ a
⇐⇒ m =  = .
γ−1/ε ε

1+τ 1+τ h(e)
w Jw ((1 − γ)J) 1−γ

Substituting this optimal m into the HJB equation (2.9), after some algebra we obtain

1
1 h(e) h 
0 γ 0 i 1−ε
a = (β + δ) 1−1/ε ε(β + δ) + (1 − ε) µ(e) θ − θ Σθ . (2.11)
1+τ 2

The optimal consumption rate is then

1 h  γ i
m= ε(β + δ) + (1 − ε) µ(e)0 θ − θ0 Σθ . (2.12)
1+τ 2

Since J is increasing in a, the optimal effort is the maximizer of (2.11). There is a

trade-off here. Increasing e will lower the utility from leisure h(e), whereas increase the
expected return µ(e).

2.3.2 Equilibrium conditions

Substituting the optimal consumption rule (2.12) into the budget constraint (2.7),
individual wealth evolves according to

 γ 
dwi /wi = ε(µ(e)0 θ − β − δ) + (1 − ε) θ0 Σθ dt + θ0 dX. (2.13)
2

The aggregate wealth obeys a similar diffusion. The only difference is that individuals that
are alive receive annuity δ on the fraction of wealth 1 − θ2 − θ3 = θ1 invested in the stock
and the risk-free asset, but the annuity cancel out with the amount the annuity company
 53

collects from deceased agents. Therefore the aggregate wealth evolves according to

 γ 
dW /W = ε(µ(e)0 θ − β − δ) + (1 − ε) θ0 Σθ − δθ1 dt + θ0 dX̄, (2.14)
2

where dX̄ = (σm dBm , σ p dB p , σc dBc )0 is the vector of aggregate shocks. Note that (2.14)
differs from (2.13) in only two aspects: first, in (2.14) dX is replaced by dX̄, in which
the idiosyncratic risk disappears by the law of large numbers; second, (2.14) contains the
term −δθ1 , which accounts for the annuity payment on the stock and the risk-free asset.
Since consumption is proportional to wealth, individual consumption also obeys
the diffusion (2.13), and aggregate consumption obeys (2.14). Therefore (2.14) must be
consistent with (2.5). Comparing coefficients, it must be the case that

γ
µc = ε(µ(e)0 θ − β − δ) + (1 − ε) θ0 Σθ − δθ1 , (2.15a)
2
1
σc dBc = θ0 dX̄ ⇐⇒ σc dBc = (θ1 σm dBm + θ2 σ p dB p ). (2.15b)
1 − θ3

Finally, since in equilibrium fraction θ3 of wealth is in government bond and there is a

unit supply of the bond, its price must satisfy

P = θ3W.

Therefore the dividend yield is

D τmW τm
d= = = , (2.16)
P θ3W θ3

where m is the optimal consumption rate given by (2.12).

We can obtain the equilibrium as follows.

1. Let z = (z1 , z2 , z3 , z4 ) = (µg , e, θ1 , θ2 ) be the equilibrium objects, where µg is the

 54

expected total return on the government bond, e is the effort, and θ1 , θ2 are the
fraction of wealth invested in the stock market and private equity.

2. Let Σ(z) be the variance matrix of dX = (σm dBm , σ p dB p + vdBi , σc dBc )0 with
σc dBc replaced by (2.15b) with

θ = (θ1 , θ2 , 1 − θ1 − θ2 ) = (z3 , z4 , 1 − z3 − z4 ).

3. Let µ(z) = (µm + δ, µ p (e), µg )0 be the vector of expected returns (note that µg = z1
and e = z2 ), m(z) be the optimal consumption rate in (2.12) with z substituted,
and a(z) be right-hand side of (2.11) with µ(e), Σ replaced by µ(z) and Σ(z),
respectively. Define F : R4 → R4 by

γ τm(z)
F1 (z) = ε(µ(z)0 θ − β − δ) + (1 − ε) θ0 Σθ − δθ1 + , (2.17a)
2 θ3
F2 (z) = arg max a(z), (2.17b)
e=z2

F3 (z) = θ1 , (2.17c)

F4 (z) = θ2 , (2.17d)

where θ = (θ1 , θ2 , θ3 )0 is the arg max of (2.10) over θ ∈ ∆ with µ(e) = µ(z) and
Σ = Σ(z). F1 defines the expected total return on government bond, which is the
sum of aggregate consumption growth in (2.15a) and the dividend yield in (2.16).
F2 defines the optimal effort choice. F3 and F4 define the optimal portfolio.

4. The equilibrium is a fixed point of the mapping F : R4 → R4 .

The equation F1 (z) = z1 is linear and therefore it can be solved by hand. After some
 55

tedious algebra, the result is

 −1
τ
z1 = µg = 1 − εθ3 − (1 − ε) µ − δθ1
1+τ
  !
τ  γ 0 
− 1− ε(β + δ) + (1 − ε) µ − θ Σθ , (2.18)
(1 + τ)θ3 2

where µ = (µm + δ)θ1 + µ p (e)θ2 .

2.3.3 Welfare

Assume that agents are ex ante identical, so they all start with initial capital
normalized to 1 at t = 0. Letting P0 be the initial price of the government bond and
W0 be the initial aggregate wealth (physical capital plus the market capitalization of the
government bond), we have
P0 = θ3W0 .

Since the government grants 1 share of the bond to every agent, the initial aggregate
wealth is
W0 = 1 + P0 .

Combining these two equations, we get

1
W0 = . (2.19)
1 − θ3

1 1−γ ,
The value function of an agent with wealth w is J(w) = 1−γ (aw) where a is given by
(2.11). Since J(w) is a monotonic transformation of aw, the welfare of an agent at t = 0
is
a
V0 = aW0 = (2.20)
1 − θ3
 56

in consumption equivalent.
The welfare criterion for the future generations are more complicated. Let wt be
the initial wealth of an agent born at t, and Wt be the aggregate wealth. Since fraction
of wealth θ1 of a deceased agent goes to the annuity industry, we have wt = (1 − θ1 )Wt .
Therefore the certainty equivalent of the generation t welfare is

  1
1 1−γ
1−γ 1
Vt = E (1 − γ) (awt )1−γ = a(1 − θ1 ) E[Wt ] 1−γ . (2.21)
1−γ

Since Wt is a geometric Brownian motion with drift µc and volatility σc , by Itô’s lemma
we have
σ2c
 
d logW = µc − dt + σc dBc .
2

Hence logWt − logW0 ∼ N((µc − σ2c /2)t, σ2c t), so using the moment generating function
of the normal distribution, we have

σ2c σ2c t
  
1−γ 2
E[Wt ] = W0 exp µc − t(1 − γ) + (1 − γ) . (2.22)
2 2

Combining (2.19), (2.21), and (2.22), we obtain

γσ2c
  
a(1 − θ1 )
Vt = exp µc − t . (2.23)
1 − θ3 2

a
Thus there is a conflict of interest across generations. All generations care about 1−θ3 ,

which is essentially the coefficient of the value function, taking initial aggregate capital
as given. The initial generation cares only about this quantity. However, the future
generations also care about the proportion of wealth they inherit, 1 − θ1 , and the risk-
γσ2c
adjusted economic growth rate µc − 2 . The generations born right after t = 0 care only
a
about 1 − θ1 (on top of 1−θ3 ), while the generations in the far distant future care only
γσ2c
about µc − 2 because the growth dominates any difference in the initial endowment.
 57

Therefore if the government wishes to make a Pareto improvement, it must

simultaneously increase the three quantities

a a(1 − θ1 ) γσ2c
, , µc − . (2.24)
1 − θ3 1 − θ3 2

2.3.4 Stationary distribution

To compute the stationary distribution of the economy, using (2.5), rewrite (2.13)
and (2.14) as

dwi /wi = (µc + δθ1 )dt + σc dBc + θ2 vdBi ,

dW /W = µc dt + σc dBc .

By Itô’s formula and dBc dBi = 0, we obtain

σ2c + θ22 v2
 
d log wi = µc + δθ1 − dt + σc dBc + θ2 vdBi ,
2
σ2c
 
d logW = µc − dt + σc dBc .
2

Taking the difference, we obtain

θ22 v2
 
d log(wi /W ) = δθ1 − dt + θ2 vdBi . (2.25)
2

Therefore the logarithm of individual wealth relative to aggregate wealth evolves accord-
ing to a Brownian motion.
We can solve for the entire wealth dynamics following Toda (2014).12 For
12 Seealso Gabaix (2009) for a method to solve for the stationary distribution by using the Fokker-Planck
equation (Kolmogorov forward equation).
 58

notational simplicity, let

θ22 v2
µ = δθ1 − ,
2
σ = θ2 v,

m = log(θ2 + θ3 ).

Fix time t > 0. For agents with age t (i.e., those already born at t = 0), since agents
are ex ante identical, the initial wealth relative to aggregate wealth is wi (0)/W (0) = 1
(which is 0 in logs). Since the death rate is δ, there is a mass e−δt of such agents. For
agents with age s < t, since they receive wealth from private equity of deceased agents
and one share of the government bond, their initial wealth relative to aggregate wealth
must be θ2 + θ3 (which is log(θ2 + θ3 ) in logs), the fraction of wealth a typical agent
invests in the private equity and the government bond. Since log relative wealth obeys
the Brownian motion, the cross-sectional mean and variance increases linearly with age.
Therefore the cross-sectional density of log relative wealth at t is the normal mixture

2 ˆ t 2
1 − (x−µt)
−δt 1 − (x−m−µs))
f (x,t) = e √ e 2σ t +
2 δe−δs √ e 2σ2 s ds
σ πt 0 σ πs
1 − (x−µt)2
= e−δt √ e 2σ2t
σ πt
µ(x−m)
δe σ2 h − κ|x−m|  |x−m| √ κ|x−m|
 √ i
+ e σ Φ − σ√t + κ t − e σ Φ − |x−m| √ −κ t
σ t
,
κσ

p
where κ = 2δ + (µ/σ)2 and Φ is the cumulative distribution function of the standard
normal distribution.13 As t → ∞, all terms except the second converge to 0. Therefore
13 See Proposition 17 in Toda (2014) and its proof for the details of this derivation.
 59

the stationary distribution is

δ − µ(x−m) κ|x−m|
− σ
f (x) = lim f (x,t) = e σ2
t→∞ κσ

 αζ e−α|x−m| , (x > m)


α+ζ
= (2.26)
 αζ −ζ|x−m|


α+ζ
e , (x ≤ m)

where r
2δ µ2 µ
α, ζ = 2
+ 4 ∓ 2.
σ σ σ

Note that −α and ζ are solutions to the quadratic equation

σ2 2 θ22 v2 2 θ22 v2
 
ϕ(λ) := λ − µλ − δ = λ − δθ1 − λ − δ = 0.
2 2 2

Since ϕ(0) = −δ < 0 and ϕ(−1) = δ(θ1 − 1) < 0, we have

−α < −1 < 0 < ζ ⇐⇒ α > 1, ζ > 0.

The distribution (2.26) is known as Laplace (Kotz et al., 2001). Consequently,

the distribution of relative wealth wi /W , which is the exponential of log relative wealth,
is double Pareto (Reed, 2001; Toda, 2012) with power law exponents α, ζ and location
parameter (mode if ζ > 1) em = θ2 + θ3 . The precise density function of relative wealth
is 
αζ
(θ + θ3 )α x−α−1 ,



α+ζ 2
(x > θ2 + θ3 )
fw/W (x) =
αζ
(θ + θ3 )−ζ xζ−1 .



α+ζ 2
(x ≤ θ2 + θ3 )

Note that since α → 1 as δ → 0, we recover Zipf (1949)’s law when the death probability
gets smaller (agents live longer). The fact that the stationary cross-sectional distribution
becomes double Pareto is not surprising because Toda (2014) shows that it arises robustly
 60

in a large class of dynamic general equilibrium models. However, the magnitude of the
tail exponent α is model-dependent. Here we have α > 1 and α → 1 as δ → 0, consistent
with Zipf’s law. In Benhabib et al. (2016), where there are inheritance and means-tested
government subsidies, the tail exponent (in their notation β2 ) is > 2.
Toda (2012) calculates the Gini coefficient for the double Pareto distribution
explicitly. Using

2δ 2δ
αζ = = ,
σ2 σ2
2µ 2δθ1
ζ − α = 2 = 2 − 1,
σ σ s
r
2δ µ 2 4δ 4δ2 θ21
α+ζ = 2 + = 1 − (2 − θ 1 ) + ,
σ2 σ4 σ2 σ4

the result is

2α2 + 2αζ + 2ζ2 + ζ − α

Gw/W =
(α + ζ)(2α − 1)(2ζ + 1)
2(ζ − α)2 + 4αζ + (ζ − α)
=
(α + ζ)(4αζ − 2(ζ − α) − 1)
4δ2 θ2
1 + σδ2 (8 − 6θ1 ) + σ4 1
= q ,
4δ 4δ 4δ2 θ21
(1 + σ2 (2 − θ1 )) 1 + σ2 (2 − θ1 ) + σ4

where σ = θ2 v. The Gini coefficient approaches 1 (extreme inequality) as δ → 0.

2.4 Numerical example

Since in our model the government has an ability to offer an insurance through
taxation against the risk for which there is no private market (the idiosyncratic risk
of private equity), if the moral hazard problem did not occur, the welfare of agents
would be maximized by full insurance: the optimal consumption tax rate would be
 61

infinite, which is equivalent to 100% income tax. The main interest of current study
is quantitative evaluation for the implication of moral hazard problem: with a small
elasticity of entrepreneur’s effort, whether it is still possible to improve the welfare
by risk sharing through taxation and, if possible, how much the optimal tax rate is
affected. Our model is solvable and suitable for numerical simulations while the model is
elaborate enough to answer questions regarding welfare, portfolio allocation (risk-taking),
economic growth, and wealth distribution in the presence of moral hazard problem. The
results from numerical simulations of our model show the moral hazard problem has a
significant impact on the usefulness of risk sharing through fiscal policy.

2.4.1 Functional forms

We assume the effort takes values in 0 ≤ e ≤ 1, the disutility from effort is

h(e) = (1 − e)η with 0 < η < 1. If we interpret l = 1 − e as “leisure”, then the elasticity
of the utility from leisure with respect to leisure is η. The private equity return is

µ p (e) = µ + (µ̄ − µ)e

¯ ¯

with µ < µ̄. This specification means that the private equity return is µ̄ with maximal
¯
effort (e = 1) and µ with minimal effort (e = 0). Then the first-order condition for the
¯
maximization of (2.11) (after substituting z and log-differentiating) is

η (µ̄ − µ)θ2
− + ¯
= 0. (2.27)
1 − e ε(β + δ) + (1 − ε) µ(e)0 θ − 2γ θ0 Σθ
 62

This equation is linear in e and therefore can be solved by hand. After some tedious
algebra, the solution is

!
1 ε(β + δ) + (1 − ε) µ(0)0 θ − 2γ θ0 Σθ
e= 1−η , (2.28)
1 + η(1 − ε) (µ̄ − µ)θ2
¯

where µ(0) = (µm + δ, µ, µg )0 is the vector of expected returns with zero effort.
¯

2.4.2 Calibration

We calibrate the parameters at annual frequency. The relative risk aversion

coefficient γ = 4, which is relatively common in the macro literature. The elasticity of
intertemporal substitution is ε = 0.7. We are aware that the vast majority of the empirical
estimates of EIS is less than 1, and most macro papers assume so but most macro-finance
papers choose a value larger than 1 (Schmidt and Toda, 2015). The results are not
sensitive to this choice, and all the graphs below look very similar whether ε = 0.1 or
ε = 1.5. The effective discount rate is β + δ = 0.04, which is standard. We set the death
rate to δ = 0.01, which implies an average lifespan of 1/δ = 100 years. We choose
such a low rate because we interpret agents as dynasties, not necessarily as households.
The stock market return and volatility are µm = 0.04 and σm = 0.08, which are arguably
low. We choose these numbers because we interpret the first technology as a mature
industry. We set the maximum and minimum return on technology and the aggregate
and idiosyncratic volatility to be µ̄ = 0.1 (10%), µ = 0.02 (2%), σ p = 0.15 (15%), and
¯
v = 0.1 (10%), which seem reasonable. The correlation between the stock market and
private equity is ρ = 0.8. As for the most important parameter, the elasticity of effort, we
use two values of 0.05 and 0.075 for η because we want to see how much effect those
small elasticities have on the optimal tax policy. Table 2.1 summarizes the parameter
values. We then solve for the equilibrium by changing the consumption tax rate from 0
 63

to 100%.
Table 2.1: Parameter values
Description Symbol Value
Discount rate β 0.03
Risk aversion γ 4
Elasticity of substitution ε 0.7
Death rate δ 0.01
Elasticity of effort η 0.05; 0.075
Stock market return µm 0.04
Stock market volatility σm 0.08
Maximum investment return µ̄ 0.1
Minimum investment return µ 0.02
Aggregate volatility σ¯p 0.15
Correlation ρ 0.8
Idiosyncratic volatility v 0.1

2.4.3 Results

Figure 2.1 reports the effects of consumption tax on the variables of interest for
the case of η = 0.05. Figure 2.1a shows the effort exerted by entrepreneurs. Since the
risk sharing through taxation provides an insurance, it exacerbates the moral hazard
problem as expected. But the reduction in effort is moderate because of our choice of
small η; when consumption tax is newly introduced at the rate of 1%, the reduction in
effort makes the expected return of private equity go down from 9.50 to 9.49 percentage
point. If the elasticity of the expected return with respect to net-of-tax rate is calculated
based on these figures, the elasticity is quite small at τ = 0, equal to 0.03.14
Despite the lower expected return on private equity caused by moral hazard,
Figure 2.1b shows that the economy grows faster (the graph shows the expected aggregate
consumption growth rate, µc ). This is caused by three forces. First, higher tax allows
14 We define the net-of-tax rate τN as τN = 1/(1 + τ). The elasticity of the expected return ζ is defined
as ζ = (∆µ p (e)/∆τN )(τN /µ p (e)). ζ increases as the tax rate goes up, but not much. At the optimal tax rate
for the initial generation, 63% as shown below, ζ is still small and equal to 0.06.
 64

0.94 2.5

0.93

Economic Growth (%)

2.4

0.92

Effort
2.3
0.91

2.2
0.9

0.89 2.1
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(a) Effort (b) Economic growth

0.6 60

Composition of Real Investment (%)

Stock Stock
Private Equity Private Equity
0.5
Government Bond
55
Ratio in Portfolio

0.4

0.3 50

0.2
45
0.1

0 40
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(c) Portfolio (d) Fraction of real investment

0.04 0.0386

Generation 0
Welfare of Generation 0

0.03 Generation 0+
Generation ∞ 0.0385
Welfare Criterion

0.02
0.0384
0.01

0.0383
0

-0.01 0.0382
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(e) Welfare criterion of each generation (f) Welfare of initial generation

6.5 0.22
Upper tail
6 Lower tail
0.2
Power Law Exponents

5.5
Gini Coefficient

5 0.18

4.5 0.16
4
0.14
3.5

3 0.12
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(g) Power law exponents (h) Gini coefficient

Figure 2.1: Comparative statics with respect to the consumption tax rate (η = 0.05)
 65

better risk sharing and induces the agents to allocate more capital to the high-risk, high-
return private equity. Second, this reallocation of capital increases the total aggregate
risk in the economy (because the aggregate volatility of private equity is higher than that
of the stock market) and induces agents to save more. Third, due to better risk sharing,
the idiosyncratic risk declines, which induces the agents to save less. As discussed in
Schmidt and Toda (2015), precautionary savings occur if and only if ε < 1; the sign
is reversed if ε > 1. In either case, the capital reallocation effect dominates. In other
words, taxation encourages risk-taking, being consistent with the conclusion of traditional
literature, in this case of very small degree of moral hazard problem, and the economy
grows faster.
Figure 2.1c shows the portfolio share of the stock market (θ1 ), private equity
(θ2 ), and the government bond (θ3 ). Because increasing the tax rate raises the market
capitalization of the bond, θ3 monotonically increases. Both θ1 and θ2 declines, but
private equity decreases relatively slower, which means that agents allocate more capital
θ2
to private equity. In fact, within real investment, the fraction of private equity θ1 +θ2

increases (Figure 2.1d). (Of course, increasing the tax further causes too much moral
hazard and the private equity will collapse to zero eventually, but that does not happen
until a consumption tax of around 270%.)
Figure 2.1e shows the welfare criteria of each generation: for the initial generation
a a(1−θ1 )
(t = 0), 1−θ3 ; for generations born right after (t = 0+), 1−θ3 ; for generations in the far
γσ2c
distant future (t = ∞), µc − 2 ; see (2.24). Since the welfare of the initial generation is
fairly flat, Figure 2.1f magnifies the graph. According to these graphs, all generations gain
up to tax rate around 63% (39% tax burden out of income). For generations 0 and 0+, the
unit is consumption equivalent. At τ = 63%, the welfare gain of the initial generation is
0.89%, while that of the generation born right after is 68%. Future generations gain even
more since the risk-adjusted growth is higher. Therefore risk sharing through taxation
 66

can achieve a Pareto improvement in this case, and most benefit is enjoyed by the future
generations. The reason why the generation born right after t = 0 gains so much is
because they inherit a larger fraction of wealth since private equity investment increases,
which by assumption cannot be pledged for annuity. If some part of private equity can be
used for annuity, the conclusion might be different.
Figure 2.1g shows the power law exponent of the cross-sectional consumption
distribution. With no tax, the exponents are 4 for the upper tail and 3 for the lower tail,
which are roughly the values for the U.S. consumption distribution (Toda and Walsh,
2015). As the tax rate increases, so do the power law exponents because better risk
sharing reduces the idiosyncratic volatility, which determines inequality. In fact, the Gini
coefficient in (2.1h) monotonically decreases.
We show the result of the same exercise for the other parameter choice, η = 0.075,
in Figure 2.2. Since the elasticity of effort is higher in this case, entreprener’s effort
level goes down more rapidly than the previous case as the tax rate increases as shown
in Figure 2.2a. Because of this change in effort, the expected return of private equity
decreases from 9.17% to 9.16% when consumption tax is introduced by 1%, which
implies the elasticity of the expected return with respect to net-of-tax rate equals 0.06 at
τ = 0. Even though the elasticity is still quite low, just slightly higher than the previous
case, the remaining variables respond to consumption tax very differently.
While economic growth is spurred by taxation for low tax rates, the growth rate
turns to decrease as the tax rate goes up beyond 40% (Figure 2.2b). Figures 2.2c and 2.2d
suggest the main reason for this movement of economic growth rate. As in the previous
case, the portfolio share of government bond increases consistently since increasing
the tax rate raises the market capitalization of the bond. As for the two investment
technologies, agents raises the share of private equity as the tax rate increases when the
tax rate is low, but then they start to reduce its share once the tax rate goes beyond certain
 67

0.9 1.95

0.88 1.9

Economic Growth (%)

0.86 1.85

Effort
0.84 1.8

0.82 1.75

0.8 1.7

0.78 1.65
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(a) Effort (b) Economic growth

0.7 70

Composition of Real Investment (%)

Stock Stock
0.6 Private Equity Private Equity
Government Bond
60
0.5
Ratio in Portfolio

0.4
50
0.3

0.2
40
0.1

0 30
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(c) Portfolio (d) Fraction of real investment

0.04 0.036
Generation 0
Generation 0+
Welfare of Generation 0

0.03 Generation ∞ 0.0355

Welfare Criterion

0.02 0.035

0.01 0.0345

0 0.034

-0.01 0.0335
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(e) Welfare criterion of each generation (f) Welfare of initial generation

9 0.2
Upper tail
8 Lower tail 0.18
Power Law Exponents

Gini Coefficient

7 0.16

6 0.14

5 0.12

4 0.1

3 0.08
0 20 40 60 80 100 0 20 40 60 80 100
Consumption Tax (%) Consumption Tax (%)

(g) Power law exponents (h) Gini coefficient

Figure 2.2: Comparative statics with respect to the consumption tax rate (η = 0.075)
 68

level. This result is caused by the following mechanism: taxation alters the attractiveness
of private equity as a financial asset in two ways. First, a higher tax rate makes private
equitly more attractive by reducing idiosyncratic risk through risk sharing. Second, since
the moral hazard problem gets more severe as the tax rate increases, it makes private
equily less attractive through the decrease in expected return. Figure 2.2d shows the
former effect outweighs the latter at first until the tax rate of 47%, then the relation is
reversed. This effect of taxation on portfolio allocation is reflected in the hump-shaped
graph of economic growth rate.
Figure 2.2e shows it is possible to improve the welfare of future generations by a
low tax rate, but Figure 2.2f indicates the welfare of initial generation cannot be improved
by the fiscal policy tool we are considering. It may look puzzling that even the lowest
tax rate harms the welfare of initial generation at first glance; when the tax rate is low,
risk-taking is encouraged, economic growth is accelerated, and the effort level is reduced,
which contributes to a higher utility. In fact, consumption tax has its own negative effect
on utility through the wedge on the choice between consumption and effort (or leisure
consumption).15 Since the positive gain from risk sharing through taxation is small for
the initial generation, the loss from the intratemporal wedge exceeds the gain and reduces
the welfare in this case. This result sheds light on the important issue that the previous
studies have ignored. Even if taxation successfully encourages risk sharing, it does not
necessarily imply improvement of welfare when the performance of risky investment
depends on entrepreneur’s effort.
Figure 2.2g and 2.2h show a higher tax rate invariably contributes to mitigating
inequality in the economy better even when the responses of other variables to a increase
in the tax rate are very different between the cases.
15 This type of effect is not specific to our choice of consumption tax. For example, capital income tax

also brings the intratemporal wedge.

 69

2.5 Concluding remarks

Economists have argued that taxation may well encourage risk-taking. Although
the studies with that conclusion are still relevant, the implication of moral hazard problem
has been largely ignored in the literature. This theoretical study aims to answer how the
moral hazard problem affects the optimal consumption tax rate, employing a tractable
model that can describe welfare, risk-taking, economic growth, and inequality. The
results of our numerical simulations show that the effect of moral hazard problem is
significant. If the return of risky asset depends on entrepreneur’s effort, risk sharing
through taxation cannot improve the welfare even under an arguably low elasticity of
effort.
To derive relevant implications for policy debates, the key parameter is the degree
of moral hazard, for which we look at the elasticity of the expected return with respect to
net-of-tax rate in this study. Although it seems difficult to obtain reliable estimation for
this parameter by an empirical study, our results suggest understanding the moral hazard
problem is critical for the study of taxation and risk-taking.

2.6 Appendix

2.6.1 Solution algorithm

This section explains how to numerically compute the equilibrium. As explained

in the preceding sections, the equilibrium is a fixed point of a nonlinear mapping F :
R4 → R4 . Let z = (µg , e, θ1 , θ2 ) be the initial guess of the equilibrium objects. We use
the Matlab fsolve command, except that we use a few tricks to improve the numerical
stability.
 70

Updating the government bond return We use (2.17a) to update the government
bond return µg . One caveat is that since F1 (z) contains θ3 = 1 − θ1 − θ2 in the numerator,
which is zero when the tax rate is zero, it introduces a discontinuity. To deal with this, we
solve the autarky case (τ = 0) separately, and put a reasonable bound on µg (say [−1, 1])
so that the updated µg will stay in the bound even if F1 (z) is infinite.

Updating the effort We use (2.28) to update the effort. If e < 0, we replace by 0. If
e > 1, we replace by 1.

Updating the portfolio To update the portfolio, we first solve the optimal portfolio
problem with no shortselling constraints. For generality let J be the number of assets,
µ ∈ RJ be the vector of expected returns, and Σ be the J × J variance-covariance matrix.
Then the optimal portfolio problem is

γ
maximize µ0 θ − θ0 Σθ subject to θ ≥ 0, 10 θ = 1. (2.29)
2

Clearly, our portfolio problem (2.10) is a special case of (2.29) with J = 3. The gradient
of the objective function of (2.29) is µ − γΣθ. Hence it is straightforward to solve for the
optimal portfolio problem numerically using the fmincon command in Matlab.
However, numerical optimization introduces instability in the subsequent equilib-
rium computation. To deal with this issue, we note that most of the time the nonnegativity
constraint θ ≥ 0 does not bind. Then the problem becomes equality constrained, which
we can solve analytically. Let φ = (θ1 , . . . , θJ−1 )0 be the portfolio share of all assets
except the last. Using the accounting constraint 10 θ = 1, it follows that θ = a + Aφ, where

   
0J−1   IJ−1 
a =
|{z}   and A =  .
1
|{z}
−1 0
J×1 J×(J−1) J−1
 71

Here 0K , 1K , IK denote the K-vector of zeros, K-vector of ones, and K × K identity matrix.
Substituting θ = a + Aφ, the optimal portfolio problem (2.29) (without the nonnegativity
constraint) becomes

γ
maximize µ0 (a + Aφ) − (a + Aφ)0 Σ(a + Aφ).
2

By the first-order condition, the solution is

1
A0 (µ − γΣa) − γA0 ΣAφ = 0 ⇐⇒ φ = (A0 ΣA)−1 A0 (µ − γΣa).
γ

Using this expression, we can compute the optimal portfolio θ = a + Aφ. In practice,
we first compute this θ, and if it satisfies θ ≥ 0, then we know that it is the (unique)
solution. If some element of θ is negative, then we switch to the numerical solution using
fmincon.
One caveat is that when the tax rate is very high (so moral hazard is severe) and
the minimum return µ is low, zero investment in private equity (θ2 = 0) may be optimal.
¯
Then the government bond and stock market will be identical from individual’s point of
view, so the optimal portfolio is indeterminate. In this case we need to pin down θ1 and
θ3 = 1 − θ1 from (2.18) using the no-arbitrage condition µg = µm + δ.
Chapter 2, in full, is currently being prepared for submission for publication of
the material. Miyoshi, Yoshiyuki; Toda, Alexis Akira. The dissertation author was the
primary investigator and author of this material.
Chapter 3

Rental Housing Investment and Tax

Reforms: An Empirical Study of Tax
Clientele Model

Abstract

This paper examines the relationship between household marginal tax rates and
the probability of owning rental housing. A simple model of tax clienteles predicts a
positive association between the marginal tax rate and the ownership of rental housing
in general, but the passive loss provisions introduced by the tax reform in 1986 is
expected to limit this association to households with income less than the threshold
value. The empirical results based on 1983, 1989, 1992, 1995, and 1998 Surveys of
Consumer Finances show the marginal tax rate has a positive association with rental
housing ownership only for the group of households that can deduct passive losses on
rental housing investment, implying the tax clientel model is relevant for rental housing
investment.

72
 73

3.1 Introduction

The effects of taxation on housing investment has been one of the major topics in
public finance.1 In the literature, owner-occupied housing has drawn considerably more
interest than rental housing.2 Although the homeownership rate in the US is relatively
high compared to the other developed countries, one third of the households in the nation
live in rental housing. Thus expanding our understanding of how taxation affects rental
housing investment is important.
The interaction between progressive income tax and holding of rental properties
is an aspect of this subject that has not been studied extensively. It has been known since
Modigliani and Miller (1958) that investors are divided into “clienteles” of different types
of financial assets if the marginal tax rate varies across investors and the financial assets
are subject to different rules of taxation. For example, if stocks are taxed more lightly
than bonds, possibly due to a low capital gains tax rate, stocks attract investors with a
high marginal tax rate. The goal of this study is to provide empirical evidence about
whether the tax clientele model can be applied to the portfolio choice including rental
housing investment. Examining the relevance of tax clientele model for rental housing is
meaningful because the tax rate applied to (marginal) investors of rental housing affects
the cost of rental housing provision and eventually the market rent.3
The literature of emprical studies of taxation and portfolio choice is not large
because of data availability, complicated incentives caused by the tax code, and potential
endogeneity of tax rates. I take advantage of the provisions regarding passive losses that
was introduced as part of the Tax Reform Act (TRA) in 1986 to identify the effect of
1 Rosen (1985) surveys the early economic studies on this topic.
2 For example, the studies of taxation and owner-occupied housig include Glaeser and Shapiro (2002),
Hilber and Turner (2014), Poterba (1984) and Poterba and Sinai (2008), Rosen and Rosen (1980).
3 As a related figure, Poterba (1990) estimates TRA will increase the steady state rent by 10-15 percent

assuming the marginal investors remain in the top bracket. If the tax rate for the marginal investors declines
due to the passive loss provisions, the estimate for the rent increase should be larger.
 74

taxation on holding of rental housing. Before TRA, households could deduct passive
losses from their ordinary income. Since at that time it was common for rental housing
investment to generate tax losses especially for the first several years,4 the deduction of
passive losses was valuable for investors with high marginal tax rates. However, TRA
restricted households from deducting passive losses as a general rule in an attempt to
deal with tax sheltering activities. The interesting feature that accompanied this general
rule is that households whose adjusted gross income is below a certain level are allowed
to deduct tax losses for rental housing investment from the ordinary income. Therefore
the taxation may affect the two groups of housholds—households with income above or
below the threshold income—differently.
I first present a simple model of tax clienteles, in which a household with a high
marginal tax rate is more likely to own residential rental properties in its portfolio if
there is no restriction on passive losses. But the model implies that in the presence of the
restriction on passive losses, the marginal tax rate does not correlate with the ownership
of residental rental properties for households whose income is above the threshold while
the positive correlation still holds for the other group of households.
To test the prediction of the model, I use data from five years of the Surveys of
Consumer Finances (SCF), which span before and after TRA. I estimate the effect of
marginal tax rate on the probability of ownership of residential rental properties. The
econometric issue is the potential endogeneity of marginal income tax rate. The theory
predicts the effect of marginal tax rate on portfolio choice, but in turn portfolio choice
affects income and then the income tax rate. I use the first-dollar marginal tax rate, which
is a standard remedy in the empirical public finance literature, to address this issue.
The results of probit estimation are modestly consistent with the predictions of
the model although some of the results make the interpretation ambiguous. For three
4 See Follain et al. (1987) for a formal analysis.
 75

years out of four years after TRA, the estimates for the marginal tax rate are (marginally)
statistically significant for the group of households that are allowed deduction of passive
losses. On the contrary, the estimate for the other groups of households are never
statistically significant for the same years. Thus I conclude the tax clientele model is
relevant for rental housing investment as well as for the portfolio choice among financial
assets although the standard errors are high, possibly due to the idiosyncratic nature of
rental housing investment.
This study aims to contribute to two strands of literature. First, in the literature
on rental housing investment the interaction between the progressive tax schedule and
portfolio choice has been mostly ignored although there are a couple of exceptions.
Berkovec and Fullerton (1992) construct and estimate a general equilibrium model
in which the supply of rental housing is endogenously determined and households
choose their portfolio considering a progressive income tax schedule. While the model
of Berkovec and Fullerton (1992) is essentially static, Chambers et al. (2009) use a
dynamic model and solve it for steady state equilibrium. Both studies provide interesting
implication for policy proposals by simulations, but they presume household should
follow the solution to the standard maximization problem in the presence of progressive
taxation without empirical evidence.5 This study provides the first empirical evidence
that households actually follow the prediction of the tax clientele model.
Second, this study also contributes to the empirical literature on taxation and
portfolio choice. Among the previous empirical studies on this subject,6 the current
study is closest to Poterba and Samwick (2002) in that they use cross-section data from
the several years of SCF and try to remedy the endogeneity problem regarding the
marginal tax rate by the first dollar variable. The contribution of this study is the test of
5 In addition, neither Berkovec and Fullerton (1992) nor Chambers et al. (2009) take into accout the

restriction on passive losses.

6 Poterba (2002) provides an extensive survey of the literature.
 76

the tax clientele model for rental housing investment with an additional dimension of
identification that utilizes the passive loss provisions for rental housing.
The strucure of the rest of the paper is as follows: section 2 present a simple
model for taxation and portfolio choice including rental housing investment, in which the
passive loss provisions can be incorporated. Section 3 summarizes the changes in the
tax code in 1980s that are relevant to rental housing investment. Section 4 presents the
dataset and the empirical specification. Section 5 reports the estimation results. Section 6
is the brief conclusion.

3.2 A tax clientele model for rental housing investment

Although the previous studies have proposed various tax clientele models that
provide theoretical tools to analyze how taxation affects households’ portfolio choice
among financial assets, such as bonds, stocks, and tax-exempt and tax-deferred assets,
few studies include real assets in their scope.7 I provide a simple clientele model
to obtain some insights as to the relation between taxation and incentive for rental
housing investment. In the model, agents aim to maximize their return, and they are
heterogeneous in terms of income tax rates. There are three types of assets: taxable
bond, tax-exempt bond, and residential rental property. All the assets are riskless in this
economy; taxable and tax-exempt bond pay deteministic amounts to the owners, and
housing rent is determined outside the model without uncertainty. The asset prices and
returns are determined in equilibrium to clear the supply of assets. Therefore if there
were no rental housing, it would be the model of Miller (1977): in equilibrium, Agents
with high tax rates choose to own tax-exempt bond while agents with low tax rates own
7 The classical examples for clientele models include Auerbach and King (1983) and Miller (1977).
 77

taxable bond. Formally,

Ei = Wi , Bi = 0 if (1 − ti )rb < re ,

Ei = 0, Bi = Wi otherwise, (3.1)

in which Wi is the total wealth of household i; Ei and Bi are the values of tax-exempt
bond and taxable bond owned by household i, respectively; rb and re are the returns
to tax-exempt bond and taxable bond determined in the market, respectively; ti is the
income tax rate for household i.
In the model of this study, households can also invest in the residential rental
property, and they are allowed to borrow a loan to finance the investment up to the value
of property at the interest rate equal to rb . The model assumes an exgenous debt limit
on total balance for this type of loan for each household, Di . While rent income is taxed
at the same rate as income from taxable bond, households deduct depreciation of the
property and interest payment for the loan. The rate of tax depreciation d may not be
equal to the rate of economic depreciation δ, but both rates are assumed to be constant
over the lifetime of property. The asset price P for a unit of housing is also assumed to be
constant across time.8 Household i’s problem to maximize the net return is represented
as

max re Ei + (1 − ti )rb Bi + (1 − ti )(RHi − rb λi PHi ) − (δ − ti d)PHi (3.2)

E,B,H,λ

s.t. Wi = Ei + Bi + (1 − λi )PHi

in which Hi is the amount of rental housing; λi is the loan-to-value ratio of rental housing
investment.
8 This assumption is pluasible if the economy is in the steady state.
 78

Let t ∗ be the real number such that (1 − t ∗ )rb = re . Then by the same argument
regarding equation 3.1, households whose ti is greater than t ∗ make no investment in
taxable bond in this model as well. Moreover, a definite implication for loan-to-value
ration can be derived for this group of households: they choose the 100% loan-to-value
ratio because after-tax cost of property loan interest, (1 − ti )rb , is less than the return
from investment in tax-exempt bond, re , for this group. On the contrary, households with
tax rates less than t ∗ invest no wealth in tax-exempt bond. If a household in this group
wants to invest in rental housing, it is indifferent about how much to borrow to finance
the investment.
First, consider the rates of tax and economic depreciation are equal: d = δ. Then
equation 3.2 can be rewritten as

re Ei + (1 − ti )rb Bi + (1 − ti )(RHi − dPHi − rb PHi ).

In this case, the equilibrium gross return for rental housing is equal to the sum of
depreciation rate and market interest rate: R/P = d + rb . Households with any income
tax rates are indifferent about how much to invest in rental housing since there is no
excess return due to heterogeneity in tax rates.
Second, let the tax depreciation rate be greater than the economic depreciation.9
Then the return for rental housing investment, which is represented as follows, is an
9 This is a common assumption to analyze the tax incentive for housing investment, especially investment

behavior in early 80s, when tax sheltering activities that take advantage of residential rental properties
were rampant (Ozanne 2012, Poterba 1993). In practice, however, although the tax depreciation rate is
larger only in early years of lifetime of rental properties, the rate is smaller (usually zero) than economic
depreciation rate in later years. To incorporate this feature of tax code, the model should have a housing
asset price as a function of its vintage, and it makes the model significantly more complicated. Follain et al.
(1987) disucuss the relation between the “front-loading” of depreciation allowance and the passive loss
provision.
 79

increasing function of ti :

 
R b
(1 − ti ) − r − (δ − ti d).
P

In equilibrium, households invest in rental housing up to its debt limit Di if its tax rate
is greater than the cutoff tax rate t ∗∗ . The cutoff tax rate and housing asset price are
determined by the system of equations:

 
∗∗ R
(1 − t ) − r − (δ − t ∗∗ d) = 0,
b
P
ˆ
Di
di = H̄, (3.3)
ti >t ∗∗ P

in which H̄ is the aggregate demand for housing, which is determined outside the model.
The taxable income from rental housing investment is non-positive for any
households in equilibrium in this model. If a household owns Hi unit of rental hous-
ing and ti ≥ t ∗∗ , it can be confirmed that the taxable income from rental housing,
RHi − dPHi − rb PHi , is less than or equal to zero and that the equality holds only for the
marginal household whose tax rate equals t ∗∗ . Therefore the tax rules regarding losses on
rental housing investment are critical for the incentive for housing investment. Consider
the tax provision that does not allow households whose tax rate is greater than some rate
t¯ to offset their income from financial investment against losses on housing investment,
which is similar to the anti-shelter provision introduced by 1986 Tax Reform Act. Then
for any households that cannot deduct tax losses from rental housing from the other
income, the after-tax return from housing investment is negative:

R t ∗∗ (δ − d)
− rb − δ = < 0.
P 1 − t ∗∗

Thus households that are not qualified for deduction of losses on rental housing do not
 80

own any rental property. In addition, note that this expression for the return does not
depend on the individual tax rate for households, ti . The equilibrium under this rule of
tax losses is described as follows: there exists the cutoff tax rate t˜∗∗ , and households own
rental property to their debt limit if their tax rate satisfies t˜∗∗ ≤ ti ≤ t¯; equation 3.3, the
second equation for equilibrium conditions, is modified as

ˆ
Di
di = H̄.
t˜∗∗ ≤ti ≤t¯ P

This model is too simple and includes some bold assumptions, including no
consideration of uncertainty. In particular, if a strict borrowing constraint for low and
middle income households is present in reality, it may well severely distort the prediction
of the model, concentration of ownership of rental housing below the threshold. However,
it offers some intuitions about the effect of income taxation on households’ portfolio
choice among financial assets and rental housing. In general, households with higher
marginal tax rates are more likely to own rental properties if the rules of tax depreciation
is more favorable than economic depreciation. If the tax code does not allow some
households with high marginal tax rates to deduct losses on rental housing, the marginal
tax rate should affect the group of households that are qualified for deduction significantly
even after controlling the income and wealth effects. The empirical analysis to follow
examines the latter intuition using the data set of household portfolio before and after the
tax reform in 1986.

3.3 Tax reforms in 1980s

The US federal income tax system experienced two major changes during 1980s:
the Economic Recovery Tax Act (ERTA) in 1981 and TRA in 1986. These reforms had
very different effects on the incentives for rental housing investment. The provisions
 81

that seemed to affect rental housing investment include ones regarding depreciation,
capital gains, and passive losses. ERTA increased the tax advantage of rental housing
by shortening the tax lifetime for those property from 32 to 15 years. On the contrary,
TRA couteracted this effect by extending the liftime to 27.5 years. A similar pattern is
found for capital gains tax rates. While ERTA lowered the marginal tax rate on long-term
capital gains for the top income bracket from 28 percent to 20 percent, TRA put the top
tax rate back to 28 percent. Since there were no special provisions for captail gains from
rental property, unlike for owner-occupied properties, the changes in capital gains tax
rate should have had real impacts on the incentive of investors.
In addition to these changes, TRA had some provisions to prevent tax payers from
using rental housing as a tax shelter vehicle. It has widely been recognized that rental
housing was one of the most active shelter vehicles in the early 1980s partly because of
the generous depreciation rule introduced by ERTA (Follain et al. 1987, Ozanne 2012,
Poterba 1993). Investors could usually report taxable losses even though they did not
have any economic losses. TRA aimed to deter this practice by the provision regarding
passive losses. The provision made it no longer possible to offset ordinary taxable
income with passive losses, as which taxable losses on rental housing are usually counted,
but it allowed only passive income to be offset. However, TRA also had the special
exemption for this anti-shelter provision: if landlord’s modified adjusted gross income is
below $100,000, the landlord may deduct up to $25,000 in passive losses against other
income.10 The maximum amount of deduction starts to phase out beyond $100,000 of
modified adjusted gross income and becomes zero at $150,000. Therefore landlords
whose income satisfies the requirement of exemption could still have the benefit of
passive losses after TRA. Although there have been a considerable volume of studies for
10 The modified adjusted gross income is defined as adjusted gross income figured without taking into

accout the taxable amount of social security, the deductible contributions to individual retirement accounts
and some pension plans, any passive activity income and loss, and other income components that are
usually in small amounts.
 82

the effects of changes in depreciation rule and capital gains tax rate, no formal empirical
study has been conducted for the effect of the anti-shelter provision of TRA.11 The
econometric analysis in the next section is focused on that effect.

3.4 Data and methodology

3.4.1 Survey of Consumer Finances

I use data from SCF conducted in 1983, 1989, 1992, 1995, and 1998.12 SCF
is a triennial survey that started in 1983. Those surveys cover the time period that is
suitable to examine the short-term and long-term effects of ERTA and TRA. SCF has
great advantages for the purpose of this study; it reports the detailed composition of
financial and non-financial assets, including the information about rental real estate
properties as well as owner-occupied housing. One of the important features of SCF is
its oversampling of high-income households. Since the distribution of wealth is skewed
towards high-income households, the oversampling helps us obtain accurate information
about households’ portfolio choice. SCF also reports various income components of
households, which makes it possible to estimate their federal income tax liabilities
and marginal tax rates.13 Since this study examines how tax rates and rules affect
affect households’ incentive for rental housing investment, these types of information
is essential. One weekness of the data set of this study is that it does not include data
from 1986 SCF. 1986 SCF could be useful to look at households’ behavior regarding
rental housing just before the introduction of TRA, the period for which experts have
11 Poterba (1993) briefly argues that the anti-shelter provition of TRA worked citing the statistics of real
estate partnership sales.
12 Poterba and Samwick (2002) use SCF of same years and finds broadly consistent evidence with

the standard clientele model regarding the correlation between the marginal tax rate and the ownership
probability for financial assets.
13 Since the public data sets of SCF do not report the geographic information of households, their state

tax liabilities and marinal tax rates cannot be estimated.

 83

Table 3.1: Probability of ownership (percent) of residential rental properties

Income range 1998 1995 1992 1989 1983

($000, 1995 dollars)
0-15 1.91 2.64 2.18 2.04 1.63
15-25 4.86 5.11 5.03 3.41 3.82
25-50 8.25 6.75 7.52 9.61 6.05
50-75 13.54 10.07 12.40 10.79 11.56
75-100 16.80 13.69 16.64 15.13 17.55
100-250 24.62 23.24 27.66 29.45 19.94
250 + 39.98 27.27 43.98 34.50 35.32
Total 8.92 7.39 8.27 8.67 6.85
Notes: Househoselds are weighted by sample weights in each year.

said that tax sheltering activities with rental housing investment was rampant. Unlike
the surveys in the other years, however, 1986 SCF does not report the information about
income components, and it is impossible to estimate the marginal tax rates of households
accurately enough.

3.4.2 Summary information on owners of rental housing

Based on the information of SCF, I define owners of rental housing as households

that own any type of residential properties, including single family houses and apartments.
Households are defined as owners no matter if they own the properties independently,
jontly with others, or through partnerships.
Table 3.1 presents summary information on the ownership probabilities of rental
housing by income group. There is a strong positive correlation between income and
rental housing ownership. Moreover, this correlation is fairly constant over time, even
after TRA. Since the marginal income tax rate is correlated with income, it is not clear
whether this pattern is caused by the income effect or the effect of marginal tax rate on
portfolio choice.
 84

3.4.3 Estimating marginal tax rates

I obtain the estimates of marginal tax rates for the SCF households from the
National Bureau of Economic Research’s TAXSIM program. Although it is possible
to estimate the marginal tax rate by making full use of the information about income
components reported in SCF, there is an econometric issue when such estimates for
marginal tax rates are used to study the tax effects on portfolio choice as mentioned
above. The marginal tax rate is an endogenous variable for households because it is
affected by households’ portfolio choice; for example, if a household that receives high
financial income takes advantage of tax sheltering activities heavily, its resulting marginal
tax rate may be as low as middle income households. Exsiting empirical studies avoid
this issue by constructing the “first-dollar” marginal tax rate, which is the marginal tax
rate if income from financial assets is assumed to equal zero.14
I basically follow the procedure of Poterba and Samwick (2002) to construct the
first-dollar marginal tax rates. Let YBit be the vector of income components and deductions
for household i in time t when the household’s income from dividends, interest, and
capital gains is assumed to be zero. Then T (YBit ) is the federal income tax liability
calculated by TAXSIM based on YBit . The first dollar marginal tax rate is defined as
τit = [T (YBit + ∆) − T (YBit )]/∆, in which ∆ is an increment in ordinary income. The
increment is chosen as the maximum of 5% of the total value of household’s financial
assets and $100. Poterba and Samwick (2002) justify this choice for increment because
5% can be considered as the normal return for the financial assets.
14 Ina recent study, Kawano (2014) uses this type of variable to test the dividend clientele hypothesis
with data from SCF.
 85

3.4.4 Empirical specification

This study is focused on how the rules regarding passive losses that was introduced
by TRA affected households’ incentive for rental housing investment. I estimate probit
models for rental housing ownership as a function of the marginal tax rate and the
modified aggregate gross income (MAGI) year by year. I also control for a variety of
socioeconomic and demographic variables.15
The estimating equation is given by

y∗it = γL I(Ỹit ≤ C)τit + γH I(Ỹit > C)τit + ηI(Ỹit ≤ C) + xit β + εit ,

P(yit = 1|Ỹit , xit ) = Φ(y∗it ),

in which yit is the binary variable for ownership of residential rental properties; y∗it is the
latent variable; I() is the indicator function, Ỹit is MAGI and C is the cutoff value of
MAGI for the passive losses rule exemption, which is set as $100,000 for all the survey
years; xit is the vector of control variables including the constant term; γL , γH , η, and β
are the parameters to be estimated and εit is the error term.
The parameters of interest are γL and γH . η, the coefficient on the binary variable
that indicates MAGI is no more than the cutoff, is included for the technical reason: it
allows the effect of marginal tax rate to have different intercepts across the two MAGI
groups. If the tax code does not treat losses from retal housing investment differently
across income groups, households with a higher marginal tax rate has a stronger incentive
to participate in rental housing investment for the purpose of offsetting their positive
income. After the negative losses rule was put in effect in 1986, however, the argument
in section 3.2 predicts that γL is positively larger in magnitude than γH because the
15 Ihave also estimated tobit models for the ratio of loan balance collateralized by residential rental
properties to the total financial value for each household. The results are quite similar to the ones from
probit models, arguably reflecting very small probability of ownership for residential rental properties.
 86

high-income households can no longer offset their income with passive losses from rental
housing.
Since the portfolio choice is usually expected to be affected by income and
wealth effects, I include a set of binary variables that indicate the income and net worth
categories of households. Those categories are reported in thousands of 1995 dollars.
The control variables include demographic variables: educational attainment, age, gender
of household head, marital status, and the size of household. SCF asks households about
their subjective attitude towards taking risk. The variable “risk averse” takes unity if the
household is not willing to take any financial risks, and it takes zero otherwise.

3.5 Empirical findings

Table 3.2 reports the probit coefficients and standard errors for the ownership
of residential rental properties by years. For the years after 1986 except for 1992, the
coefficients for the marginal tax rate for households from the lower MAGI group are
marginally statistically significant: the figures fail to satisfy the 5% significance level,
but the p-values range between 0.053 and 0.060. On the contrary, the marginal tax rate
coefficient for the higher MAGI group are cosistently statistically insignificant, and it
is even negative for year 1998. The results provide modest support for the prediction
made for the effect of passive losses rule introduced by 1986 TRA for these three years.
However, the result for 1992 is an anomaly. The marginal tax rate coefficients for both
MAGI groups are negative although they are not statistically significant. I cannot provide
a plausible explanation for this anomaly. But there may be systematic inaccuracy in the
first-dollar marginal tax rate calculation for this year because the results from the probit
estimation for ownership of another asset category, tax-deferred accounts, also shows an
anomaly only for this year, as I will present in the following.
 87

Table 3.2: Probit estimates for the ownership of residential rental properties
Year 1998 1995 1992 1989 1983
Coeff. S.E. Coeff. S.E. Coeff. S.E. Coeff. S.E. Coeff. S.E.
Constant -1.313 0.363 -2.236 0.429 -2.184 0.451 -2.122 0.594 -2.408 0.814

MAGI<=100 -0.388 0.314 -0.312 0.328 -0.152 0.303 -0.135 0.438 -0.352 0.751

MTR
MAGI<=100 0.939 0.485 0.920 0.490 -0.313 0.503 0.959 0.511 0.813 0.582
MAGI>100 -0.762 0.798 0.554 0.834 -0.361 0.826 0.257 1.336 0.164 1.386

Income
15-25 -0.055 0.157 0.074 0.139 0.174 0.150 0.044 0.169 0.151 0.158
25-50 0.188 0.144 0.064 0.137 0.242 0.149 0.266 0.155 0.156 0.182
50-75 0.101 0.171 -0.047 0.165 0.319 0.178 0.264 0.187 0.276 0.238
75-100 0.137 0.187 -0.041 0.176 0.246 0.192 0.162 0.212 0.272 0.287
100-250 0.265 0.190 -0.212 0.185 0.226 0.200 0.527 0.207 0.227 0.320
250 + 0.418 0.202 -0.083 0.194 0.492 0.209 0.633 0.228 0.504 0.358

Net worth
50-100 0.501 0.138 0.271 0.144 0.384 0.136 0.342 0.146 0.224 0.122
100-250 0.850 0.119 0.347 0.124 0.781 0.114 0.611 0.125 0.730 0.108
250-1000 1.167 0.120 0.938 0.117 1.284 0.119 1.001 0.128 0.996 0.126
1000 + 1.554 0.137 1.340 0.128 1.603 0.135 1.304 0.154 1.002 0.176

Education
High school -0.112 0.114 0.046 0.113 -0.104 0.118 -0.061 0.114 -0.147 0.105
Some college 0.023 0.113 0.084 0.115 0.024 0.118 -0.108 0.123 -0.043 0.113
College degree -0.067 0.117 0.054 0.117 -0.095 0.119 -0.109 0.123 -0.112 0.126
Post college -0.259 0.120 0.180 0.119 -0.083 0.121 -0.033 0.126 -0.170 0.124

Age
25-34 -0.622 0.207 0.071 0.273 0.497 0.327 0.335 0.402 0.751 0.343
35-44 -0.407 0.194 0.241 0.268 0.538 0.324 0.346 0.400 0.685 0.347
45-54 -0.240 0.192 0.302 0.268 0.603 0.324 0.487 0.400 0.660 0.348
55-64 -0.091 0.195 0.593 0.270 0.495 0.327 0.446 0.401 0.798 0.347
65 + -0.200 0.196 0.506 0.270 0.362 0.327 0.269 0.402 0.665 0.351

Risk averse -0.341 0.078 -0.139 0.070 -0.150 0.068 -0.148 0.073 -0.103 0.077
Female -0.288 0.103 -0.041 0.106 -0.223 0.110 -0.035 0.128 -0.121 0.132
Married -0.049 0.085 0.176 0.088 -0.068 0.093 -0.058 0.110 -0.026 0.119
HH Size -0.022 0.027 0.002 0.025 -0.037 0.026 -0.065 0.028 -0.044 0.029
 88

How the marginal tax rate affected the incentive for rental housing investment
in the run-up to 1986 TRA is another topic of interest. The argument in section 3.2
predicts that the marginal tax rate will be associated with the ownership of rental housing
and the association will be particularly stronger for the high MAGI group because 1981
ERTA introduced the favorable depreciation rule, allowed households to report large tax
losses on rental housing investment. However, the marginal tax rate coefficients are not
statistically significant although both coefficients are positive. This result might be due
to the fact that the data is just one year after the enactment of ERTA since the income
variables of 1983 SCF is based on tax filing in 1982. Households might have taken some
time to take advantage of new provisions in the tax code. At the same time, this result
may well be caused by weak identification of variation in the marginal tax rate, especially
if the large standard error for the high MAGI group is considered. This hypothesis seems
plausible because the equivalent coefficient in the probit estimation for ownership of
tax-deffered accounts turns out to be statistically insignificant for this year.
The result for year 1983 makes it difficult to derive a definite conclusion about
another interesting topic: the adjustmet speed for households’ portfolio. Scholz (1994)
conducts the descriptive analysis of porfolio structure using the 1983 and 1989 Surveys
of Consumer Finances and finds basically no change in real estate investment between
these two years, which this study confirms in part by Table 3.1. Gordon (1994) suggests
a potential explanation that the portfolio adjustment may be slow because of the long life
of real property. If the marginal tax rate is positively associated with a high probability
of rental housing ownership just before 1986, which could be tested more clearly if a
data set of good quality for around 1985 were available, the combination of marginal
tax rate coefficients for 1989 would suggest households adjusted their portfolio quite
quickly in response to the tax reform, even including the real asset. If that hypothesis is
true, the cross-sectional pattern of ownership of rental housing across the income groups,
 89

Table 3.3: Average marginal effects of marginal tax rate on the probability of ownership

Year 1998 1995 1992 1989 1983

Residential rental properties
MAGI<=100 0.186* 0.182* -0.063 0.194* 0.109
(1.94) (1.88) (-0.62) (1.88) (1.40)
MAGI>100 -0.151 0.109 -0.073 0.052 0.022
(-0.96) (0.66) (-0.44) (0.19) (0.12)

Tax-deferred accounts
MAGI<=100 0.092 0.396** 0.283** 0.413** 0.361**
(0.89) (3.73) (2.65) (3.82) (3.72)
MAGI>100 0.590** 0.892** 0.256 0.850** 0.076
(2.83) (4.06) (1.20) (2.51) (0.28)
Notes: Z-statistic for the original probit estimation are in the parentheses. Asterisks indicate
statistical significance at the 5% (**) and 10% (*) significance levels. See text for further
discussion.

which is presented in Table 3.1, is caused by the income effect, implying the income
effect dominates the effect of progressive tax schedule based on the tax clientele model.
However, only with the current resutls, no definite conclusion about the speed of portfolio
adjustment can be drawn.
The coefficients for income and net worth variables are estimated as predicted in
most of the years. Especially, high net worth has a strong association with the probablity
of rental housing ownership while high income usually are associated with a higher
probability. As a whole, none of the demographic control variables have an explanatory
power. This result contrasts with Poterba and Samwick (2002), in which at least a
college-level degree are associated with a higher probability of any categories of financial
assets.
The anomalies found in the probit results may make someone doubt the reliability
of identification power of the first-dollar marginal tax variable used in this study. Although
Poterba and Samwick (2002) use the variable constructed by the same procedure and find
the predicted associations between the marginal tax rate and the ownership probability
 90

for some financial asset categories, they do not estimate the coefficient separately by
income groups as this study does. Thus, as a reference, I pick tax-deferred accounts, for
which the standard clientele model predicts positive association between the marginal tax
rate and the ownership probability and estimate the probit model for this financial asset
with the same variables as in Table 3.2.16 Table 3.3 presents the average marginal effects
calculated from the probit estimations for the ownership probabilities for residential
rental properties and tax-deferred accounts, along with the Z-statistics in the original
probit estimations. For 1989, 1995, and 1998, the coefficient for the marginal tax rate
for the high MAGI group, which is consistent with the theory, although, as I mentioned
earlier, the estimation fails to find statistically significant evidence for 1983 and 1992.
From this result, the first-dollar marginal tax rate appears to be a relevant variable for an
empirical study on the tax clientele model.

3.6 Conclusion

This study estimates the probit model for the probability of owning rental hous-
ing to test the prediction of the tax clientele model in the presence of the passive loss
provisions introduced by TRA in 1986. The results offer modest support for the predic-
tions although some of the results make interpretation difficult. The marginal tax rate is
positively associated with the probability of rental housing ownership for the group of
households to whom the exception for restriction on passive losses is applied for most of
the years after TRA. The marginal tax rate indicates no association with rental housing
for the other group, which is subject to the restriction.
The most diffecult part of this empirical study is the high correlation and endo-
geneity between the marginal tax rate and income. There are convincing arguments for
16 The tax-deferred accounts variable in this study takes unity if the household has an asset of positive
value in IRA account or Keogh account.
 91

us to expect the probability of rental housing ownership is correlated with income itself,
not necessarily through the marginal tax rate. Households with high income may be more
willing to invest in risky assets, including real estates. Since purchasing real estates is
commonly financed by debt to a considerable extent, the existence of credit constraint
for low and middle income households will hinder them from rental housing investment.
SCF provides detailed information needed to estimate tax liabilities and makes it possible
to identify the variation in the marginal tax rate after controlling income. However, more
detailed data on taxation would allow us to generate a more precise variable for the
marginal tax rate and reduce the standard errors in the estimation.
References

Albanesi, Stefania (2011). “Optimal Taxation of Entrepreneurial Capital with Private

Information”. In: Working Paper.
Anand, Bharat N. and Richard Sansing (2000). “The Weighting Game: Formula Ap-
portionment as an Instrument of Public Policy”. In: National Tax Journal 53.2,
pp. 183–199.
Angeletos, George-Marios (2007). “Uninsured idiosyncratic investment risk and aggre-
gate saving”. In: Review of Economic Dynamics 10.1, pp. 1–30.
Aoki, Shuhei and Makoto Nirei (2014). “Zipf’s Law, Pareto’s Law, and the Evolution of
Top Incomes in the U.S.”
Asea, Patrick K. and Stephen J. Turnovsky (1998). “Capital Income Taxation and Risk-
Taking in a Small Open Economy”. In: Journal of Public Economics 68, pp. 55–
90.
Atkinson, Anthony B. and Joseph E. Stiglitz (2015). Lectures on Public Economics.
Updated. Princeton University Press.
Auerbach, Alan J. and Mervyn A. King (1983). “Taxation, Portfolio Choice, and Debt-
Equity Ratios: A General Equilibrium Model”. In: Quarterly Journal of Eco-
nomics 98.4, pp. 587–610.
Benabou, Roland (2002). “Tax and Education Policy in a Heterogeneous Agent Economy:
What Levels of Redistribution Maximize Growth and Efficiency?” In: Economet-
rica 70.2, pp. 481–517.
Benhabib, Jess, Alberto Bisin, and Shenghao Zhu (2011). “The Distribution of Wealth
and Fiscal Policy in Economies with Finitely Lived Agents”. In: Econometrica
79.1, pp. 123–157.
— (2016). “The Distribution of Wealth in the Blanchard-Yaari Model”. In: Macroe-
conomic Dynamics 20, pp. 466–481.

92
 93

Berkovec, James and Don Fullerton (1992). “A General Equilibrium Model of Housing,
Taxes, and Portfolio Choice”. In: Journal of Political Economy 100.2, pp. 390–
429.
Blanchard, Olivier J. (1985). “Debt, Deficits, and Finite Horizons”. In: Journal of Political
Economy 93.2, pp. 223–247.
Chambers, Matthew, Carlos Garriga, and Don E. Schlagenhauf (2009). “Housing Policy
and the Progressivity of Income Taxation”. In: Journal of Monetary Economics
56.8, pp. 1116–1134.
Conesa, Juan Carlos and Dirk Krueger (2006). “On the optimal progressivity of the
income tax code”. In: Journal of Monetary Economics 53.7, pp. 1425–1450.
Conesa, Juan Carlos, Sagiri Kitao, and Dirk Krueger (2009). “Taxing Capital? Not a Bad
Idea After All!” In: American Economic Review 99.1, pp. 25–48.
Cullen, Julie Berry and Roger H. Gordon (2007). “Taxes and Entrepreneurial Risk-taking:
Theory and Evidence for the U.S.” In: Journal of Public Economics 91.7-8,
pp. 1479–1505.
Diamond, Peter A. and Emmanuel Saez (2011). “The Case for a Progressive Tax: From
Basic Research to Policy Recommendations”. In: Journal of Economic Perspec-
tives 25.4, pp. 165–190.
Domar, Evsey D. and Richard A. Musgrave (1944). “Proportional Income Taxation and
Risk-Taking”. In: Quarterly Journal of Economics 58.3, pp. 388–422.
Duffie, Darrell and Larry G. Epstein (1992). “Stochastic Differential Utility”. In: Econo-
metrica 60.2, pp. 353–394.
Eaton, Jonathan (1981). “Fiscal Policy, Inflation and the Accumulation of Risky Capital”.
In: Review of Economic Studies 48.3, pp. 435–445.
Eaton, Jonathan and Samuel Kortum (2002). “Technology, Geography, and Trade”. In:
Econometrica 70.5, pp. 1741–1779.
Edmiston, Kelly D. and F. Javier Arze del Granado (2006). “Economic Effects of Ap-
portionment Formula Changes: Results from a Panel of Corporate Income Tax
Returns”. In: Public Finance Review 34.5, pp. 483–504.
Fajgelbaum, Pablo D., Eduardo Morales, Juan Carlos Suárez Serrato, and Owen Zidar
(2015). “State Taxes and Spatial Misallocation”.
Follain, James R., Patric H. Hendershott, and David C. Ling (1987). “Understanding the
Real Estate Provisions of Tax Reform: Motivation and Impact”. In: National Tax
Journal 40.3, pp. 363–372.
 94

Gabaix, Xavier (2009). “Power Laws in Economics and Finance”. In: Annual Review of
Economics 1.1, pp. 255–293.
Glaeser, Edward L. and Jesse M. Shapiro (2002). “The Benefits of the Home Mortgage
Interest Deduction”. In: NBER Working Paper.
Golosov, Mikhail, Narayana Kocherlakota, and Aleh Tsyvinski (2003). “Optimal Indirect
and Capital Taxation”. In: Review of Economic Studies 70.3, pp. 569–587.
Golosov, Mikhail, Aleh Tsyvinski, and Iván Werning (2007). “New Dynamic Public
Finance: A User’s Guide”. In: NBER Macroeconomics Annual. Ed. by Daron
Acemoglu, Kenneth Rogoff, and Michael Woodford. Vol. 21. MIT Press. Chap. 5,
pp. 317–388.
Goolsbee, Austan and Edward L. Maydew (2000). “Coveting Thy Neighbor’s Manu-
facturing: the Dilemma of State Income Apportionment”. In: Journal of Public
Economics 75.1, pp. 125–143.
Gordon, Roger H. (1985). “Taxation of corporate capital income: Tax revenues versus
tax distortions”. In: Quarterly Journal of Economics 100.1, pp. 1–27.
— (1994). “Comments”. In: Tax Progressivity and Income Inequality. Ed. by Joel
Slemrod. MIT Press, Cambridge, MA, pp. 268–274.
Gordon, Roger H. and Julie Berry Cullen (2012). “Income Redistribution in a Federal
System of Governments”. In: Journal of Public Economics 96.11-12, pp. 1100–
1109.
Gordon, Roger H. and Jeffrey K. Mackie-Mason (1994). “Tax Distortions to the Choice
of Organizational Form”. In: Journal of Public Economics 55.2, pp. 279–306.
Gordon, Roger H. and John D. Wilson (1986). “An Examination of Multijurisdictional
Corporate Income Taxation under Formula Apportionment”. In: Econometrica
54.6, pp. 1357–1373.
Gouveia, Miguel and Robert P. Strauss (1994). “Effective Federal Individual Income Tax
Functions: An Exploratory Empirical Analysis”. In: National Tax Journal 47.2,
pp. 317–339.
Heathcote, Jonathan, Kjetil Storesletten, and Giovanni L. Violante (2014). “Consumption
and labor supply with partial insurance: An analytical framework”. In: American
Economic Review 104.7, pp. 2075–2126.
Hilber, Christian A. L. and Tracy M. Turner (2014). “The Mortgage Interest Deduction
and its Impact on Homeownership Decisions”. In: Review of Economics and
Statistics 96.4, pp. 618–637.
Jones, Charles I. (2015). “Pareto and Piketty: The Macroeconomics of Top Income and
Wealth Inequality”. In: Journal of Economic Perspectives 29.1, pp. 29–46.
 95

Kawano, Laura (2014). “The dividend clientele hypothesis: Evidence from the 2003 tax
act”. In: American Economic Journal: Economic Policy 6.1, pp. 114–136.
Kenc, Turalay (2004). “Taxation, Risk-Taking and Growth: A Continuous-Time Stochas-
tic General Equilibrium Analysis with Labor-Leisure Choice”. In: Journal of
Economic Dynamics and Control 28.8, pp. 1511–1539.
Kocherlakota, Narayana (2005). “Zero Expected Wealth Tax: A Mirrlees Approach to
Dynamic Optimal Taxation”. In: Econometrica 73.5, pp. 1587–1621.
Kotz, Samuel, Tomasz J. Kozubowski, and Krzysztof Podgórski (2001). The Laplace
Distribution and Generalizations. Boston: Birkhäuser.
Mackie-Mason, Jeffrey K. and Roger H. Gordon (1997). “How Much Do Taxes Discour-
age Incorporation?” In: Journal of Finance 52.2, pp. 477–505.
Mazerov, Michael (2001). The Single-Sales-Factor Formula: A Boon to Economic Devel-
opment or a Costly Giveaway? Tech. rep. Center on Budget and Policy Priorities.
McLure, Charles E. (1981). “The Elusive Incidence of the Corporate Income Tax: the
State Case”. In: Public Finance Review 9.4, pp. 395–413.
Melitz, Marc J. (2003). “The Impact of Trade on Intra-Industry Reallocations and Aggre-
gate Industry Productivity”. In: Econometrica 71.6, pp. 1695–1725.
Merton, Robert C. (1971). Optimum Consumption and Portfolio Rules in a Continuous-
Time Model.
Miller, Merton H. (1977). “Debt and Taxes”. In: Journal of Finance 32.2, pp. 261–275.
Mirrlees, James A., ed. (2011). Tax by Design. Institute for Fiscal Studies. Oxford.
Modigliani, Franco and Merton H. Miller (1958). “The Cost of Capital, Corporation
Finance and the Theory of Investment”. In: American Economic Review 48.3,
pp. 261–297. arXiv: arXiv:1011.1669v3.
Mossin, Jan (1968). “Taxation and Risk-Taking: An Expected Utility Approach”. In:
Economica 35.137, pp. 74–82.
Ozanne, Larry (2012). “Taxation of Owner-Occupied and Rental Housing”. In: Congres-
sional Budget Office Working Paper. Working Paper Series.
Piketty, Thomas and Emmanuel Saez (2013). “Optimal Labor Income Taxation”. In:
Handbook of Public Economics. Vol. 5. Elsevier B.V., pp. 392–474.
Piketty, Thomas, Emmanuel Saez, and Stefanie Stantcheva (2014). “Optimal Taxation
of Top Labor Incomes: A Tale of Three Elasticities”. In: American Economic
Journal: Economic Policy 6.1, pp. 230–271.
 96

Pinto, Santiago M. (2007). “Corporate Profit Tax, Capital Mobility, and Formula Appor-
tionment”. In: Journal of Urban Economics 62.1, pp. 76–102.
Poterba, James M. (1984). “Tax Subsidies to Owner-Occupied Housing: An Asset-Market
Approach”. In: Quarterly Journal of Economics 99.4, pp. 729–752.
— (1990). “Taxation and Housing Markets: Preliminary Evidence on the Effects of
Recent Tax Reforms”. In: NBER Working Paper. arXiv: arXiv:1011.1669v3.
— (1993). “Tax Reform and Housing Market in the Late 1980s: Who Knew What,
and When Did They Know It?” In: Real estate and the credit crunch: Proceedings
of a conference held at Melvin Village, New Hampshire, September 1992. Federal
Reserve Bank of Boston, pp. 230–251.
— (2002). “Taxation, Risk-Taking, and Household Portfolio Behavior”. In: Hand-
book of Public Economics. Ed. by Alan J. Auerbach and Martin Feldstein. Vol. 3.
Elsevier, pp. 1109–1171.
Poterba, James M. and Andrew A. Samwick (2002). “Taxation and Household Portfolio
Composition: US Evidence from the 1980s and 1990s”. In: Journal of Public
Economics 87.1, pp. 5–38.
Poterba, James M. and Todd Sinai (2008). “Tax expenditures for owner-occupied Housing:
Deductions for Property Taxes and mortgage interest and the exclusion of imputed
rental income”. In: American Economic Review 98.2, pp. 84–89.
Reed, William J. (2001). “The Pareto, Zipf and Other Power Laws”. In: Economics
Letters 74.1, pp. 15–19.
Rosen, Harvey S. (1985). “Housing subsidies: Effects on housing decisions, efficiency,
and equity”. In: Handbook of Public Economics. Ed. by Alan J. Auerbach and
Martin Feldstein. Vol. 1, pp. 375–420.
Rosen, Harvey S. and Kenneth T. Rosen (1980). “Federal taxes and homeownership:
Evidence from time series”. In: Journal of Political Economy 88.1, pp. 59–75.
Runkel, Marco and Guttorm Schjelderup (2011). “The Choice of Apportionment Factors
under Formula Apportionment”. In: International Economic Review 52.3, pp. 913–
934.
Saez, Emmanuel (2001). “Using Elasticities to Derive Optimal Income Tax Rates”. In:
Review of Economic Studies 68.1, pp. 205–229.
Saito, Makoto (1998). “A Simple Model of Incomplete Insurance the Case of Permanent
Shocks”. In: Journal of Economic Dynamics and Control 22.5, pp. 763–777.
Sandmo, Agnar (1985). “The Effects of Taxation on Savings and Risk Taking”. In:
Handbook of Public Economics. Ed. by Alan J. Auerbach and Martin Feldstein.
Vol. 1, pp. 265–311.
 97

Schmidt, Lawrence D. W. and Alexis Akira Toda (2015). “Do You Save More or Less in
Response to Bad News? A New Identification of the Elasticity of Intertemporal
Substitution”.
Scholz, John Karl (1994). “Tax Progressivity and Household Portfolios: Descriptive
Evidence from the Surveys of Consumer Finances”. In: Tax Progressivity and
Income Inequality. Ed. by Joel Slemrod. MIT Press, Cambridge, MA, pp. 219–
267.
Stiglitz, Joseph E. (1969). “The Effects of Income, Wealth, and Capital Gains Taxation
on Risk-Taking”. In: Quarterly Journal of Economics 83.2, pp. 263–283.
Suárez Serrato, Juan Carlos and Owen Zidar (2014). “Who Benefits from State Corporate
Tax Cuts? A Local Labor Markets Approach with Heterogeneous Firms”.
Svensson, Lars E. O. (1989). “Portfolio Choice with Non-Expected Utility in Continuous
Time”. In: Economics Letters 30.4, pp. 313–317.
Toda, Alexis Akira (2012). “The Double Power Law in Income Distribution: Explanations
and Evidence”. In: Journal of Economic Behavior and Organization 84.1, pp. 364–
381.
— (2014). “Incomplete Market Dynamics and Cross-Sectional Distributions”. In:
Journal of Economic Theory 154, pp. 310–348.
— (2015). “Asset Prices and Efficiency in a Krebs Economy”. In: Review of Eco-
nomic Dynamics 18.4, pp. 957–978.
Toda, Alexis Akira and Kieran Walsh (2015). “The Double Power Law in Consumption
and Implications for Testing Euler Equations”. In: Journal of Political Economy
123.5, pp. 1177–1200.
Yaari, Menahem E. (1965). “Uncertain Lifetime, Life Insurance, and the Theory of the
Consumer”. In: Review of Economics Studies 32.2, pp. 137–150.
Zipf, George K. (1949). Human Behavior and the Principle of Least Effort. Addison
Wesley, Cambridge, MA.