taxation and financial decision making€¦ · taxation and financial decision making andrew bird...

Taxation and Financial Decision Making

by

Andrew Bird

A thesis submitted in conformity with the requirementsfor the degree of Doctor of PhilosophyGraduate Department of Economics

University of Toronto

c⃝ Copyright 2013 by Andrew Bird

Abstract

Taxation and Financial Decision Making

Andrew Bird

Doctor of Philosophy

Graduate Department of Economics

University of Toronto

2013

Understanding the effects of taxes on the financial decision making process sheds light

on the process itself and has important ramifications for policymakers. In this thesis,

I study these effects in three different contexts: international acquisitions, executive

compensation, and dividend payout.

In Chapter 1, I investigate the possibility that tax rather than productivity differences

are driving international acquisition decisions. A theoretical model of this process yields

two testable implications of tax-induced sorting: that, relative to high-tax domestic

bidders, low-tax foreign bidders will specialize in both high profitability target firms and

those with low levels of tax deductions. I find support for both of these effects in the

U.S. acquisition market using cross-sectional variation in target profitability and industry-

level variation in deductions from the tax reform of bonus depreciation. Counterfactual

simulations show that this reform induced a large drop in foreign acquisitions, leading to

a significant loss of aggregate wealth.

In Chapter 2, I study a recent increase in the tax rate on stock options for a subset of

firms to learn about the effects of taxation on executive compensation. Using novel exec-

utive compensation data, I find that this tax increase resulted in an immediate reduction

in both option grants and the share of options in total compensation. There appears

to have been limited, if any, substitution towards other components of compensation.

Hence, the burden of the tax increase must have been substantially borne by the affected

executives.

ii

In Chapter 3, I use a 2006 tax cut in Canada to study the effects of dividend taxes

on corporate payout. Analysis of discrete dividend events suggests little effect from the

reform, in stark contrast with recent evidence from the United States. Difference-in-

differences estimates using control groups comprised of firms which were exogenously

unlikely to be affected by the reform suggest a small positive effect on net dividend initi-

ations. Finally, fixed effect models of regular dividends reveal a small increase around the

reform. However, the type of firms responding casts serious doubt on taxes as the cause.

Overall, these results are consistent with the small and open nature of the Canadian

economy.

iii

Acknowledgements

I would like to thank Michael Smart, Robert McMillan, Laurence Booth and Alex Ed-

wards for their patient guidance and support over the years.

Thanks also to Dwayne Benjamin, Gustavo Bobonis, Branko Boskovic, Kory Kroft,

Joshua Lewis, Nicholas Li, Giorgia Maffini, Peter Morrow, Aloysius Siow, and semi-

nar participants at Brown, Carleton, Carnegie Mellon, Guelph, HEC Montreal, Oxford,

Ryerson, Toronto, UBC and Victoria for their helpful comments.

I gratefully acknowledge financial support from the SSHRC CGS Doctoral Fellowship,

the Dorothy J. Powell Graduate Scholarship in International Economics and the Royal

Bank Graduate Fellowship in Public and Economic Policy.

Lastly, thanks to my parents and Sherrill – my first supervisory committee.

iv

To Jackie

...

for putting up with me

v

Contents

1 The Effects of Taxes on the Market for Corporate Control 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Prior Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.1.2 A Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3.2 Empirical Issues: Y . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.4 Results: Profitability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.4.1 Majority vs. Minority Transaction Comparison . . . . . . . . . . . 16

1.4.2 Tax Haven Acquirer Comparison Results . . . . . . . . . . . . . . 16

1.4.3 Extensions and Robustness . . . . . . . . . . . . . . . . . . . . . . 17

1.5 Full Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.5.1 Empirical Strategy: z . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.5.3 Capital Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1.5.4 World Wealth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1.5.5 Empirical Implementation . . . . . . . . . . . . . . . . . . . . . . 24

1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.A Sample Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Taxation and Executive Compensation: Evidence from Stock Options 41

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.2 Stock Option Taxation and the 2010 Reform . . . . . . . . . . . . . . . . 44

2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.4 Theoretical Executive Compensation Framework . . . . . . . . . . . . . . 50

2.5 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.5.1 Which firms use TSARs? . . . . . . . . . . . . . . . . . . . . . . . 51

vi

2.5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.A Model of Executive Compensation . . . . . . . . . . . . . . . . . . . . . . 61

3 Dividends and Shareholder Taxation: Evidence from Canada 74

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.1.1 Empirical Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.2 Dividend Taxation in Canada . . . . . . . . . . . . . . . . . . . . . . . . 78

3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3.3.1 Corporate Payout . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3.3.2 Control variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.3.3 Income Trusts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.4 Analysis of regular dividend events . . . . . . . . . . . . . . . . . . . . . 84

3.4.1 Difference-in-differences . . . . . . . . . . . . . . . . . . . . . . . 86

3.5 Analysis of regular dividend levels . . . . . . . . . . . . . . . . . . . . . . 89

3.5.1 Difference-in-differences . . . . . . . . . . . . . . . . . . . . . . . 91

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

3.A Variable Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Bibliography 108

vii

List of Tables

1.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.2 Profitability Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

1.3 Profitability Results: Tax Havens . . . . . . . . . . . . . . . . . . . . . . 38

1.4 Profitability Results: Robustness . . . . . . . . . . . . . . . . . . . . . . 39

1.5 Full Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2.1 Tax Treatment of Options . . . . . . . . . . . . . . . . . . . . . . . . . . 67


2.3 Comparison of TSAR and non-TSAR Companies . . . . . . . . . . . . . 69

2.4 TSAR Company List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

2.5 Results: Option Compensation and Option Share of Total Compensation 71

2.6 Results: Non-Option Compensation and Total Compensation . . . . . . . 72

2.7 Results by Year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73


3.2 Results: Discrete Event Models . . . . . . . . . . . . . . . . . . . . . . . 102

3.3 Results: Discrete Event Models (Short Sample) . . . . . . . . . . . . . . 103

3.4 Summary Statistics By Group . . . . . . . . . . . . . . . . . . . . . . . . 104

3.5 Results: Discrete Event Difference-in-Differences . . . . . . . . . . . . . . 105

3.6 Results: Payout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

3.7 Results: Payout Difference-in-Differences . . . . . . . . . . . . . . . . . . 107

viii

List of Figures

1.1 Fraction Foreign vs. Profitability by Industry . . . . . . . . . . . . . . . 30

1.2 Fraction Foreign vs. Profitability by Industry: Merger - Stake . . . . . . 31

1.3 Bonus Depreciation Variation per $ of Investment . . . . . . . . . . . . . 32

1.4 Post-Bonus Depreciation Ownership Changes . . . . . . . . . . . . . . . . 33

1.5 Welfare Loss from Bonus Depreciation . . . . . . . . . . . . . . . . . . . 34

1.6 Tax Wedges and Welfare Loss . . . . . . . . . . . . . . . . . . . . . . . . 35

2.1 Option Compensation by Option Type . . . . . . . . . . . . . . . . . . . 64

2.2 Non-Option Compensation by Option Type . . . . . . . . . . . . . . . . 65

2.3 Monthly Volatility by Option Type . . . . . . . . . . . . . . . . . . . . . 66

3.1 Corporate Payout in Canada: 1995-2010 . . . . . . . . . . . . . . . . . . 95

3.2 Corporate Payers in Canada: 1995-2010 . . . . . . . . . . . . . . . . . . 96

3.3 Regular Dividend Events . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

3.4 Dividend Events: Ineligible Shareholders as a Control Group . . . . . . . 98

3.5 Dividend Events: Inter-listed Firms as a Control Group . . . . . . . . . . 99

3.6 Dividend Levels: Treatment and Control Groups . . . . . . . . . . . . . . 100

ix

Chapter 1

The Effects of Taxes on the Market

for Corporate Control

1.1 Introduction

Cross-border mergers and acquisitions are a major component of foreign direct invest-

ment, making up more than three quarters of net global foreign direct investment flows

in 2006 (UNCTAD). Given the sheer scale of these flows, an understanding of how taxes

affect equilibrium in the market for corporate control, which gives rise to mergers and

acquisitions (M&A), is a key input to optimal policymaking. Furthermore, since the

ownership of an asset or firm is an important determinant of its productivity,1 there

could be significant consequences for aggregate wealth arising from any tax distortions

of the equilibrium in this market. Specifically, if some potential acquirers have a purely

tax-derived comparative advantage in acquiring certain assets, they may be able to out-

bid other potential acquirers that could make more productive use of the assets. Since an

acquirer’s post-deal tax savings are completely offset by government revenue losses at the

global level, such a situation represents a clear deadweight loss, as the real productivity

of the stock of assets is not maximized.

To investigate this issue, I develop a model that focuses on the competition among

potential acquirers to buy a specific target firm. The model is especially concerned with

how the tax rates of the potential acquirers, which are assumed to vary due to differential

abilities to shift income to lower tax jurisdictions, interact with the characteristics of the

target firm and the domestic tax system. The assumption that international acquirers

1For example, Becher et al. [2012] find that productivity gains are the main source of excess returnsfrom utility mergers, while Chen [2011] finds significant dispersion in labour productivity gains for thoseemployed at the targets of foreign direct investment (FDI), depending on the source of the FDI.

1

Chapter 1. The Effects of Taxes on the Market for Corporate Control 2

have tax advantages relative to domestic acquirers is consistent with evidence presented

by Markle and Shackelford [2012], who document significant differences between the ef-

fective tax rates of multinationals resident in different countries, and particularly high

rates for U.S. firms. Given such tax rate differences, my model gives rise to two testable

implications: that low-tax foreign bidders are more likely to acquire more profitable tar-

get firms than are domestic bidders, and that increases in available tax deductions lead

to decreases in the probability of foreign acquisition.2

These predicted dimensions of sorting show how the effects of tax rate differences can

be tested even in the absence of observable company-level tax rates. This is particularly

useful as companies will generally try to obscure their tax planning practices as much

as possible so as to avoid attracting the attention of national revenue authorities. Fur-

thermore, even if raw tax rates can be observed, what matters for corporate behaviour

is the effective tax rate, which must include the transactions costs associated with tax

planning, and these are inherently very difficult to discern from accounting disclosures.

Using data on acquisitions of U.S. public companies from 1990-2010, I test these

two theoretical implications and find strong evidence in support of the existence of tax

clienteles consistent with the theory. In the first test, using cross-sectional variation

over target firms, I find that a one standard-deviation increase in the profitability of the

target leads to a 16% increase in the probability that the acquirer will be foreign. The

main empirical difficulty is that this sorting may be due to non-tax differences between

foreign and domestic bidders. To address this issue, I use two distinct strategies, beyond

controlling for a variety of observable target characteristics, including industry and time

effects. First, I use minority transactions, wherein the bidder acquires less than 50% of

the target, as a control group to account for non-tax motivations for equity investments.

It seems reasonable to assume that majority and minority transactions are driven by

similar non-tax motivations, such as geographic diversification or technology transfer,

but that income-shifting and the lower tax rate it brings is only possible for majority

owners – those who make the financial and operating decisions. It turns out that minority

foreign transactions actually target less profitable targets than do domestic minority

transactions. Hence the effect of profitability on the probability of foreign majority

acquisitions is actually higher using this control group. The second strategy employed

is to split the foreign winners into tax-haven residents and non-tax haven residents. In

the comparison of tax haven vs. domestic bidders, the effect of profitability is much

2The model works equally well for the alternative scenario where it is domestic bidders that have thetax advantage. In this case, the comparative statics would flip signs. Then the two tests detailed in thetext can be thought of as also testing for the sign of the tax difference between domestic and foreignbidders. The data support my assumption of an advantage for foreign bidders.


stronger than in the non-tax haven vs. domestic comparison. This also provides strong

evidence that taxes are the economically relevant difference between bidders, given the

likely primacy of tax considerations in the decision to locate or incorporate in a tax

haven.

To test the second key implication of the theoretical model – that foreign bidders

have a comparative disadvantage in acquiring firms with high levels of tax deductions

– I implement a difference-in-differences strategy using industry-level variation in the

generosity of investment allowances due to the bonus depreciation tax reform after 2001.

In particular, a one standard deviation increase in depreciation allowances (relative to the

distribution of changes induced by the reform) yields an 18% decrease in relative foreign

acquisitions. The reduction was largest for industries with high levels of investment in

equipment, such as transportation, and minimal for industries like real estate, which

invest mostly in structures.

The theory delivers an expression for the probability of foreign takeover in equilibrium

that can be readily taken directly to the data, which allows me to go beyond the com-

parative statics in several interesting ways. First, I use the implications of the model to

identify how bidders’ discount rates vary with their tax rates and income-shifting oppor-

tunities. This extension shows that these discount rates reflect almost the full difference

in relative tax rates, which has important implications for optimal policy. I also use

the model to conduct a counterfactual experiment, which shows that ownership patterns

were significantly changed by the institution of bonus depreciation in 2001. In particular,

foreign acquirers were disadvantaged by the reform, leading to a probability of foreign

takeover that was 5.3 percentage points less than it would otherwise have been. The

model also allows for the calculation of the loss in wealth due to this change in foreign

takeovers. Conservatively, the reform costs on the order of $36 billion per year, or 5% of

the total assets traded in the M&A market.

Overall, these results draw attention to a nontrivial tax distortion in the U.S. ac-

quisition market, whereby the ultimate owner of a domestic firm may be determined

by skill in avoiding taxes rather than skill in making productive use of the assets. As

these two identified tax effects influence foreign acquisitions in different directions, the

aggregate effect of income-shifting on inbound merger activity is theoretically ambiguous.

However, regardless of the net effect, foreign firms are specializing in high-profit targets

which have relatively few available tax deductions. So even if the aggregate probability

effect were negligible, the set of firms that is targeted by international acquirers is not

the productivity-maximizing one. This violation of production efficiency decreases aggre-

gate wealth through a reduction in the productivity of assets. Furthermore, the theory


and empirical evidence show that these clienteles are shaped by domestic tax rates and

rules, and so offer important guidance for domestic policymaking. For instance, base-

broadening reforms intended to increase tax revenue by limiting allowable tax deductions

may have the unanticipated effect of encouraging foreign acquisitions.

1.1.1 Prior Literature

An extensive literature in corporate finance has investigated the importance of tax ben-

efits in driving merger and acquisition activity in the domestic context. Kaplan [1989]

finds that increased interest deductions (along with other tax effects) can account for

anywhere between 21% and 143% of the premium paid in management buyouts of public

U.S. firms. Hayn [1989] reports further evidence which suggests that tax considerations

motivated acquisitions in the 1980s, while Erickson [1998] finds that these same consider-

ations are a key determinant of the deal structure. Devos et al. [2009] investigate a small

sample of large mergers and find that tax-related synergies are positive and can account

for about 16% of the combined equity gain between the target and the acquirer following

the transaction; tax savings appear to be a more important factor in diversifying mergers.

A more recent literature has begun to address similar questions in an international

context by extending optimal tax models to settings where cross-border capital flows

take the form of transfers of ownership of existing assets. Desai and Hines [2003] propose

the welfare benchmark of capital ownership neutrality, whereby the world tax system

should ensure that different potential acquirers face similar relative tax burdens, so that

the pattern of asset ownership is not determined by tax considerations. These ideas

are formalized and investigated by Becker and Fuest [2010], who build a model of a

multinational corporation embarking on acquisitions both in its home market and a

foreign market. They derive repatriation tax systems under which the multinational’s

private decisions are nationally or globally optimal. My model differs from theirs by

taking the tax system as given (subject to income-shifting) and showing how these tax

provisions interact with target firm heterogeneity.

There are several recent empirical papers that address related international tax issues

using data on mergers and acquisitions. Huizinga and Voget [2009] provide an empirical

investigation of the importance of potential repatriation tax burdens after a cross-border

merger. They find an economically and statistically significant discouraging effect of

the potential repatriation tax burdens on the headquarters location after the merger.

These estimates are conditional on the specific target and acquirer and so do not address

possible distortions in real ownership patterns since the parties to the deal are taken as


given.

Arulampalam et al. [2012] also use firm-level merger data to investigate whether taxes

in host country i affect the probability that a multinational corporation resident in home

country j will choose to make an acquisition in country i. Their theoretical starting

point is the decision of a single parent company choosing which host countries to make an

acquisition in.3 They find that higher host country taxes discourage inbound acquisitions

in that country. My approach is similar in spirit to theirs but takes the perspective of a

single target firm and multiple potential acquirers, which is necessary in order to study

competition among bidders in the merger market.

Of particular relevance to my study, Swenson [1994] uses a number of U.S. tax re-

forms from the 1980s to study the general equilibrium tax mechanism suggested by

Scholes and Wolfson [1990]. They emphasize the distinction between explicit and im-

plicit taxes, where the latter arise from changes to pre-tax asset returns. In the context

of FDI, investors from countries with worldwide tax systems should prefer to buy assets

with high explicit taxes and low implicit taxes, since they would receive a tax credit for

any explicit taxes paid. Swenson finds empirical confirmation for this relative preference

using differences in FDI flows across countries following tax reforms which changed the

explicit/implicit tax mix. Hines [1996] also finds evidence for this mechanism by exploit-

ing state-level tax changes and consequent changes in the investment shares of investors

from countries with worldwide tax systems.

The issue of foreign-controlled domestic corporations paying lower taxes than com-

parable domestic corporations has also been an important issue in the economics and

accounting literatures for some time. Grubert et al. [1993] first documented this issue

using confidential U.S. corporate tax returns from 1980-1987. They found that foreign-

controlled domestic corporations tended to report relatively low levels of taxable income,

which fluctuated around zero on average. This is consistent with the use of strategic

transfer pricing to lower tax burdens. A number of papers followed, some confirming the

original observation and some refuting it; the main issue has been how to control for the

endogenous selection of ownership – my study addresses this directly. A recent example

is the case study of Blouin et al. [2005], which looked at post-merger tax returns for a

small sample of 31 comparable domestic and foreign targets and found no discernible

differences in taxable income reporting. Overall, this remains an unresolved question, to

which my study provides new insight.

3This focus on the acquirer is shared by well known models in the international trade literature, such asthat of Head and Ries [2008] which models cross-border acquisitions as trading off the benefits of controlwith the costs of monitoring by the acquirer, and the heterogeneous firms model of Nocke and Yeaple[2007], which focuses on the acquirer’s mode of entry.


The potential erosion of the U.S. corporate tax base and the implied consequences for

the competitiveness of U.S. firms remain important current policy issues. In particular,

a 2007 report from the United States Treasury Department (Report to the Congress on

Earnings Stripping, Transfer Pricing and U.S. Income Tax Treaties) was commissioned

by Congress to investigate “the potential for exploitation of inappropriate income-shifting

opportunities to erode the U.S. corporate tax base.” It was specifically concerned with

foreign-controlled domestic corporations using earnings stripping through debt or transfer

pricing of intangibles and finds evidence consistent with the use of these techniques. The

strongest evidence they find for lower tax liabilities for foreign-owned corporations is

from the case of so-called ‘corporate inversions’ – a type of transaction where a domestic

corporation rearranges its ownership structure so that it becomes headquartered in a

tax haven (for example, Bermuda, which levies no corporate tax), with the old domestic

parent now a subsidiary. This is a purely tax-motivated transaction and may involve tax

savings on the foreign earnings of the multinational, since the United States taxes the

worldwide earnings of its companies while Bermuda does not. In addition, taxes may

be reduced on domestic earnings, as these can to some extent be shifted away from the

U.S. to the new headquarters country. Desai and Hines [2002] find that market reactions

to corporate inversions imply that market participants expect the transaction to result

in both foreign and domestic tax savings. Albeit on a small sample, Seida and Wempe

[2004] find direct evidence of tax savings on the order of a third of pre-inversion effective

tax rates, mostly explained by domestic U.S. tax savings. Importantly, these tax savings

were legally accomplished, predominantly through intragroup debt, despite provisions

of the U.S. tax code, such as anti-earnings stripping, that were specifically designed to

protect the domestic tax base.4 To the extent that corporate inversions and foreign

takeovers, especially by tax haven residents, lead to similar opportunities to avoid U.S.

taxes on both foreign and domestic earnings, this evidence is directly related to the key

assumption in my study, regarding U.S. vs. foreign effective tax rate differentials.

1.1.2 A Case Study

The takeover battle for the U.S. electronics manufacturer AMP in 1998 illustrates the

potential for tax considerations to affect ownership pivotally, in a way that is directly

related to my research design of predicting whether the successful acquirer of a particular

target will be foreign.

Tyco and Allied Signal were the putative bidders, and were very similar on most

4Inverted corporations appeared to save a very significant amount of U.S. tax while staying underthe 1.5:1 safe harbour debt ratio.


margins, such as assets, sales and specific industry. However, though both companies

had been long-time U.S. residents, Tyco had inverted in 1997 to become a Bermuda

resident.5 In the end, Tyco’s winning margin was approximately $1B (or 10%), which

is of the same order of magnitude as the potential tax savings from applying Tyco’s tax

rate to AMP’s earnings, rather than Allied Signal’s. The reason that this estimate is so

large is that AMP was among the most profitable firms in its industry, yielding a large

amount of profit that could be shifted out of the United States. This simple estimate

of Tyco’s tax advantage approximately matches the size of the projected tax benefits of

inverting reported in public filings by Cooper Industries in 2001 and Stanley Works in

2002. Tyco’s aggressive tax strategies had certainly been noticed in the business press:

CEO Dennis Kozlowski . . .moved Tyco to Bermuda (in 1997), then set up

an elaborate machine to finance his empire, in which most debt was issued

by a Tyco subsidiary based in Luxembourg. It was an intricate but legal

scheme to shave Tyco’s tax bills to an absolute minimum. In fact, this tax-

avoidance mechanism continues to be one of Tyco’s most powerful competitive

advantages (Business Week, 2006).

The model and empirics in this chapter explore the general ownership implications of

multinational tax avoidance strategies.

The rest of the chapter proceeds as follows: Section 1.2 develops a simple theory

of the market for corporate control, leading to two key testable implications, Section

1.3 describes the empirical strategy and the data, Section 1.4 presents the results for

profitability sorting, and Section 1.5 shows the results from estimation of the full model

as well as counterfactuals and aggregate wealth calculations. Section 1.6 concludes.

1.2 Theoretical Model

The objective of the model I develop in this section is to show how target firm charac-

teristics and tax considerations interact in the market for corporate control to determine

the ownership of that target firm. To that end, the focus is on bidders’ valuations of

the target firm, as these will determine the winning bidder in any efficient bargaining

process, taking as given that the reservation price of the original owners will be met.

Consider a potential acquisition target, with pre-tax income consisting of profit Y

and available tax deductions z (such as depreciation allowances), so that the target has

5This type of ‘endogenous’ location was associated with significant transaction costs and so wasnever common; furthermore, in 2004, future inversions were effectively shut down by the American JobsCreation Act.


taxable income Y T ≡ Y −z. There are two potential acquirers: a representative domestic

bidder and a representative foreign bidder, indexed by subscripts d and f , respectively.

They are each characterized by a nontaxable, idiosyncratic benefit of control, θi+ϵi where

θi is a fixed component and ϵi is a stochastic component, and a discount rate, ri. Note

that this characterization allows for differential fixed costs of acquisition for the different

bidders through differences in the θi. Furthermore, it is assumed that the foreign bidder

has access to an income-shifting technology (Gordon and Hines [2002]).

The technology works as follows: if the foreign bidder acquires the target firm, it

can shift some profit from the home country, with tax rate τd, to a low-tax jurisdiction,

which has a corporate tax rate of τh < τd. This could be accomplished using intragroup

debt or by manipulating transfer prices of intangible assets, like patents or trademarks.6

However, the firm faces non-deductible compliance costs to shift ω of income. The cost is

convex and decreasing in existing taxable income (because of higher probability of audit

for low reported taxable income, or because of cashflow constraints), given by γ2ω2

Y T . Then

the optimal amount of profit to shift is a constant fraction of original pre-tax income.

The effective tax rate for the foreign acquirer can be shown to be τf = τd− (τd−τh)22γ

< τd.

Hence the income-shifting technology leads the foreign bidder to face a lower effective

tax rate on the income of the target, so that τd − τf > 0.7 Then the valuation of the

target firm by bidder i is:

Vi =(1− τi)Y + τiz

ri+ θi + ϵi

This valuation is composed of three parts: the after-tax profit, the value of available tax

shields and the nonpecuniary benefits of control.

An equilibrium in the market for control consists of an allocation, which is a proba-

bility of foreign ownership conditional on target and bidder characteristics, and a price

function, which dictates how any surplus in the deal is shared between the target and

the acquirer. However, as long as the allocation awards the target to the firm with the

higher (after-tax) valuation, the price function can be ignored in deriving the results that

follow.

6According to Kleinbard [2011], intangibles are not generally subject to normal transfer pricing rules;furthermore, OECD Guidelines take the position that a business opportunity, which may be connectedwith exploitation of an intangible asset, is not a tax-cognizable asset to which transfer pricing rulesapply.

7This ordering of the tax rates is the key output of the income-shifting technology and could bedelivered using different technological assumptions. For instance, both bidders could have the ability toshift income, with the foreign bidder able to do so at relatively low cost, γf < γd.


So, assuming only that the bargaining process is efficient8 in the sense that acquirer f

obtains the firm if and only if Vf −Vd ≥ 0, we can write the probability that the acquirer

will be foreign as:

Pforeign = P

(ϵf − ϵd > −

[1− τfrf

− 1− τdrd

]Y −

[τfrf

− τdrd

]z − θ

)with θ ≡ θf − θd.

This expression reveals two possible channels for taxes to affect ownership – either

from the direct effect of taxes on cashflows, or via tax-induced differences in the discount

rates. After-tax cashflow is composed of (1 − τi)Y , which is clearly decreasing in the

tax rate, and also the value of the tax shield from z dollars of deductions, τiz, which is

increasing in the value of z and increasing in the tax rate.

In general, we would expect that the tax advantage of the foreign bidder would lead

to a relatively higher discount rate or cost of capital, reflecting a higher opportunity cost

(since the foreign bidder can take advantage of its low tax rate on alternative investments

as well). To proceed further, we need to make an assumption about just how much

discount rates are affected by the differing tax rates of the two bidders. A mild but

sufficient restriction on this relationship for what follows is:

1 ≤ rf (τf )

rd(τd)≤ (1− τf )

(1− τd)

This just means that tax differences are partially shifted back to capital suppliers, so that

discount rates are decreasing in tax rates. At one extreme – perhaps because of perfect

capital markets – both bidders face identical discount rates, despite their differing tax

rates. The other extreme, which would arise with segmented, symmetric capital markets

where capital is in fixed supply, is that savers capture all the benefits of reduced tax

rates. In between these extremes, the elasticity of capital supply is positive and finite.

Given this mild assumption, which basically just rules out overshifting, there exist ϕ and

ψ, both greater than zero, such that:

Pforeign = H(ϕY − ψz + θ) (1.1)

where H(·) is the cumulative distribution of ϵd− ϵf . Then we have the following two key

comparative static implications of the model:9

8This is unlikely to be an exact description of reality, given the empirical success of behaviouralmodels of takeovers such as Shleifer and Vishny [2003]; a necessary condition for the results that followis just that the probability of the foreign bidder winning is increasing in its real valuation advantage.


1. An increase in target profitability (Y ) increases the probability that the acquirer

will be foreign, except for the extreme case of full backward shifting of taxes onto

capital suppliers.

2. An increase in the availability of tax shields (z) decreases the probability that the

acquirer will be foreign.

It is these two predictions of the model that will be tested empirically in Sections 1.4

and 1.5. The intuition for the first case is that for fixed profitability Y , post-tax cashflow

will be higher for the low-tax bidder except in the limiting case where this advantage

is fully offset by a higher discount rate. This effect is stronger the closer are the two

bidders’ costs of capital. The second result reflects the fact that the tax-deductibility of

z means that its value is just τiz, which is obviously increasing in the bidder’s tax rate.

Since the domestic bidder also has a cost of capital no higher than the foreign bidder,

it also discounts these higher tax savings at a lower rate than the foreign bidder, which

reinforces the direct effect of the tax savings.10

To understand what is going on in the model, it is helpful to examine the two extreme

cases for the discount rates:

1. Discount rates are identical, rd = rf , then ϕ = −ψ > 0; the effects of Y and z on

probability foreign will be equal in magnitude but opposite in sign.

2. Discount rates fully reflect differences in tax rates, rd1−τd

=rf

1−τfthen ϕ = 0 and

ψ < 0; only tax shields will affect the equilibrium probability.

The first case embodies the idea (as in Scholes and Wolfson [1990]) that investors

facing relatively low tax rates will have a comparative advantage in acquiring assets that

face relatively high explicit taxes. Since tax payments are increasing in pre-tax income,

Y − z, this intuition suggests that foreign investors, facing a lower tax rate, will have

an advantage in acquiring high-profit firms. It is also clear that profitability and tax

shields have a symmetric effect on the foreign probability, as increasing either by a dollar

9Note that both of these results flip signs if in fact it is the domestic bidders which have the taxadvantage. In this sense, the signs of the empirical estimates of ϕ and ψ can be thought of as jointlytesting the sign of τd − τf , rather than relying on the assumption of a foreign tax advantage.

10Since both effects go in the same direction, the second result is robust to an alternative income-shifting technology whereby both bidders deduct z at the same effective tax rate (despite differences inthe taxation of Y ). This would eliminate the difference in the actual cash savings from foregone tax, butwould leave the effect of different discount rates intact. An alternative rationale would be the well-knownmodel of DeAngelo and Masulis [1980], based on a higher likelihood of tax exhaustion with a lower taxrate, so that an additional dollar of deductions would be less valuable.


directly changes the valuation difference between the bidders by the difference in their

tax rates.

The second case is that envisioned by Desai and Hines [2003] with the idea of capital

ownership neutrality. The advantage of a lower tax rate is completely offset by a higher

discount rate. Loosely, the intuition is that though such a bidder would indeed derive

higher after-tax cashflows from the same before-tax cashflows as a bidder with a higher

tax rate, it could get the same relative tax benefit from acquiring any other asset, all else

equal. Then there is no direct comparative tax advantage. However, as discussed above,

the difference in discount rates leads to different valuations of tax shields, thus giving

the advantage to domestic bidders in the case of firms with high levels of tax shields.

Examining the relationships between ϕ and ψ in the two extreme cases suggests that

the ratio ϕψreveals information about the relative discount rates. If this ratio is unity,

then we have the case of equal discount rates; as the ratio decreases towards zero, we get

closer and closer to full backward shifting, as envisioned in the second extreme case.

1.3 Empirical Strategy

The estimating equation is exactly the empirical counterpart of equation (1.1):

P (foreign i) = H(ϕYi − ψzi + ηXi + ui) (1.2)

assuming a normal distribution for the difference in idiosyncratic productivities, and writ-

ing the fixed component θi ≡ ηXi + ui, which can be thought of as the non-tax related

valuation difference between the two bidders. In other words, there are observable and

unobservable components of this difference, which will be the focus of the empirical strat-

egy. As is typically the case in discrete choice settings, the above model is characterized

by scale invariance, so that rather than estimating the actual parameters of interest, I will

be estimating the parameters normalized by the variance of the productivity difference.

This issue is irrelevant in terms of testing the statistical significance of the model or for

estimating the magnitude and direction of tax-induced sorting, but will be important in

calculating changes in world wealth in Section 1.5.

1.3.1 Data

Thomson SDC Platinum is a comprehensive database of cross-border and domestic busi-

ness transactions. I take all majority transactions (where the acquirer ends up with >

50% of the company) and minority stake purchases (acquirer ends up with < 50%) that


involved a publicly-traded U.S. target from 1990-2010. Given a transaction from SDC,

the target company is matched to Compustat to get the necessary accounting variables.

Most cases without a successful match are due to the fact that though the target is

public, it is not listed on an exchange covered by Compustat.

For a transaction to make it into the main estimation sample, the target company

must have a match in Compustat with nonmissing total assets, earnings, debt and intan-

gibles. This last requirement is the one that shrinks the sample the most. Furthermore,

deals that are valued at less than one million dollars or that target companies with

less than ten million dollars in total assets are dropped. Further details related to the

construction of the estimation sample are discussed in Appendix 1.A.

The general approach is to take the set of target firms as given, and then predict

whether the successful acquirer will be foreign using characteristics of the target. Hence,

the focus is on the probability foreign, conditional on the target being successfully taken

over.

The dependent variable in the empirical work, foreigni, is a dummy variable that is

equal to one if the acquirer in the deal was foreign, and zero if the acquirer was a domestic

taxable entity. This means that deals with acquirers that were domestic but effectively

nontaxable (or at least face a much lower rate than the domestic statutory tax rate),

such as government-related entities, pension funds and private equity, are excluded from

the analysis. The key point is that the group of acquirers with foreigni = 1 is assumed

to face a lower tax rate than those with foreigni = 0. Given these criteria, 15.9% of the

majority sample has a foreign acquirer; in the full sample, which includes both majority

and minority transactions, the mean is 16.4%.

The main measure of profitability is earnings before interest, taxes, depreciation and

amortization (EBITDA), divided by total assets. This is a very broad measure that

should not be affected by tax planning techniques (which come into play when trans-

forming EBITDA into taxable income). The other accounting controls which are used

are intangible assets and long-term debt, both normalized by total assets, log total as-

sets and a dummy variable equal to one if profitability is negative, as a proxy for loss

carryforwards.

The main profitability measurement issue that must be confronted is that only pre-

takeover profitability is observed (at t− 1), since the target firm is almost always taken

private following the deal, which occurs at time t, ending the obligation to report public

results. Based on the theoretical model, what we would like is profitability at the time

that the takeover decision is made, which could be up to a year after the last publicly

available accounting disclosure. To deal with this issue, I use lagged accounting variables


and year/industry dummies to construct a very simple forecasting model for future prof-

itability. Specifically, I regress the first lag of profitability (the most recent available)

on the second lag of profitability and other accounting variables and dummies. This

produces a model of profitability in period t−1 in terms of information available at time

t−2. I then use period t−1 covariates to predict the unobserved profitability in period t,

at the time of the merger decision. Using further lags of profitability yields very similar

predictions, and is not done in the base case since this cuts the estimation sample. This

procedure is actually quite similar to a measurement error methodology, wherein each

lag of profitability is viewed as a measure of future profitability plus some independent

error.11

Table 1.1 presents summary statistics for the main sample of target firms as well as

the universe of firms in Compustat over the same period. The takeover sample is actually

quite similar to the population of public firms in the United States.

The empirical estimation will proceed as follows. To begin, the focus is on estimating

the profitability effect correctly (controlling for the tax shields effect with industry dum-

mies) using several techniques to deal with omitted variable bias. After presenting the

profitability results, I then discuss the difference-in-differences strategy for estimating ψ.

Finally, I estimate equation (1.1) in one step and use it to do a counterfactual policy and

wealth simulation.

1.3.2 Empirical Issues: Y

The main empirical complication in estimating ϕ is the possibility that profitability,

Y , itself may belong in the set of X variables, describing non-tax valuation differences

between the two types of bidders. This may be the case, for example, because of asym-

metric information between domestic and foreign acquirers, of the kind investigated by

Gordon and Bovenberg [1996].12 In particular, one might expect it to be easier for a

domestic acquirer to pick out targets with low current profitability but good future

prospects, using their superior knowledge of local market conditions. Or there could

be differences in the ‘multinational’ composition of the two acquirer groups, domestic

and foreign. This could be concerning given the relatively high productivity of multina-

tionals and the possibility of a complementarity between acquirer productivity and real

11Variations on this forecasting method, including the simplest method of using lagged profitabilitydirectly, or using further lags of profitability as instruments to correct measurement error yield verysimilar results throughout the rest of the chapter, as can be seen in Table 1.4.

12By building a model of cross-border investment with endogenous information acquisition,van Nieuwerburgh and Veldkamp [2009] show that this information ‘home bias’ persists in equilibrium.


transaction-related synergies.13 In general, the concern is that bidders would sort on

target profitability for reasons other than tax differences, so that we would observe such

sorting even if all potential bidders faced the same tax rate on average.

To deal with confounding issues of this nature, it is helpful to use minority purchases,

defined as ownership changes where the acquirer ends up with less than 50% of the target

after the transaction, as a control group. A minority, or stake, purchase provides many

of the same benefits in terms of acquiring ownership of part of the income stream as a

majority transaction without involving actual control of the target. Importantly, without

control, the acquirer cannot use income-shifting strategies since these require changing

financial and even operational decisions of the firm. Hence, such transactions could be

used as a control for other motives for cross-border transactions14 and so help to identify

any tax-specific effects more precisely. Specifically, if non-tax sorting works in the same

way for both majority and minority transactions, then observed sorting on profitability

that is unique to majority purchases must be due to the tax difference. Intuitively, this

strategy can be thought of as one of difference-in-differences using majority and minority

transactions as the two groups.

A potential remaining issue is that the documented profitability differences across

types of acquirers are not due to tax differences. To address this concern, it is useful to

employ a comparison between different types of acquirers where the tax differences are

starker and more likely to be of first order importance. Specifically, consider the case of

tax haven-resident acquirers. Such firms face very low or non-existent taxes levied by

their home countries, which is typically the key motivation to locate in such a country,

given that tax havens themselves typically have small populations and markets.

Hence, define haveni = foreigni, but exclude any deals where the foreign acquirer

was not resident in a tax haven,15 as the relevant indicator to be explained by target

firm characteristics. In this case, the statutory tax difference between the two groups is

approximately 35%, the U.S. corporate tax rate, notwithstanding transaction costs.

13Rhodes-Kropf and Robinson [2008] find that mergers pair together firms with similar market tobook ratios, which they attribute to complementarity interacting with search frictions in the market forcorporate control.

14This distinction could be weakened if stake purchases are generally preludes to acquisition of fullcontrol – a so-called ‘toehold’ transaction. In this case, stake purchases should be targeting similartargets as mergers. However, there is an empirical literature analyzing the toehold phenomenon whichsuggests that this is not a concern, and in any case would bias the results against finding a difference.According to Betton et al. [2009], using data on public company transactions from 1973-2002, 13% of allbids for control had any toehold, with only 3% having been acquired within six months of the takeoverbid announcement.

15The tax haven characterization is taken from Hines and Rice [1994], although is mostly driven byacquirers from Bermuda and Switzerland, which would be on any reasonable list of tax havens.


This larger tax rate difference implies a larger direct cashflow benefit for haven ac-

quirers relative to domestic acquirers, and so a comparison of domestic versus haven

acquirers should yield stronger profitability sorting than domestic vs. non-haven foreign

acquirers.

1.4 Results: Profitability

Table 1.2 shows the results from estimating equation (1.2) using cross-sectional target firm

variation to examine the effect of target profitability on the probability of the acquirer

being foreign.

Looking at the first row of column (1), which includes the accounting controls and

year dummies, the semi-elasticity of probability foreign with respect to profitability is

2.20 (standard error: .49). For a one standard deviation increase in profitability, all else

equal, this semi-elasticity corresponds to an increase in the chance of foreign acquisition of

4.9 percentage points. This positive effect of profitability on probability foreign matches

the prediction of the theoretical model.

A key possible confounding concern is the possibility that cross-country differences in

industrial composition or differences in regulation across industries may mean that foreign

acquirers on the whole have an affinity for takeovers in certain industries,16 which may

just happen to have higher profitability. However, it would also be perfectly reasonable

for the tax effect to manifest itself in terms of both inter- and intra-industry sorting. The

former can be seen in Figure 1.1, where there is clearly a positive relationship between

median industry profitability and mean probability of foreign takeover. How much of

this sorting one is willing to attribute to taxes dictates how much weight to put on

the decrease in the estimated effect in column (2), which includes industry dummies.

The estimated effect is still positive and significant, corresponding to an increase in the

probability foreign of 2.8 percentage points for a one standard deviation in profitability.

A comparison of the results in columns (1) and (2) confirms that foreign acquirers both

preferentially sort into more profitable industries as well as to more profitable firms within

those industries. In the same vein, controlling for differential industry time trends or even

interacting industry and year effects yields similar results.

16Harford [2005] and Gorton et al. [2009], among many others, highlight the importance of industry-level variation in explaining merger activity.


1.4.1 Majority vs. Minority Transaction Comparison

Columns (3) and (4) of Table 1.2 implement the difference-in-differences style majority-

minority transaction comparison. Specifically, each of the independent variables in the

model is also included as an interaction with a dummy for a majority transaction. If

the assumption about similar non-tax motivations for both types of transactions is valid,

then the coefficient on the majority interacted profitability variable corresponds directly

to ϕ from the model, which captures the extent of profitability sorting that is driven

by tax differences between bidders. The coefficient on the non-interacted profitability

variable then describes non-tax motivations for sorting on profitability.

Looking at the first row of column (3), we can see that profitability sorting is stronger

than in the baseline model of column (1), with the semi-elasticity increasing from 2.20

to 3.00 (standard error: .78). This is directly related to the non-interacted profitability

semi-elasticity of the second row, which is negative. Hence, it appears that in the absence

of tax differences, foreign acquirers would actually prefer lower profitability targets. This

is consistent with the result in Kotter and Lel [2011] that sovereign wealth funds, a group

of investors typically facing no home country taxation, tend to target poorly performing

firms facing financial difficulties for their portfolio investments. The pattern of effects by

industry also provides some suggestive evidence that this difference is most pronounced

for high-tech firms, which may be explained by a particularly strong technology transfer

motivation for deals by foreign acquirers.

Going across the table to column (4), which adds industry dummies, the estimate

again drops somewhat, though remaining positive and significant, reflecting similar sort-

ing both within and across industries. Figure 1.2 gives a graphical representation of

inter-industry sorting for this comparison.

1.4.2 Tax Haven Acquirer Comparison Results

Table 1.3 splits the observed profitability sorting from the main results into comparisons

between domestic acquisitions and two mutually exhaustive groups of foreign acquirers.

The first row of results shows the semi-elasticity of probability foreign with respect to

profitability where the sample excludes tax haven acquirers, while the second row shows

the same quantity excluding non-tax haven acquirers.

With or without industry controls, and using the baseline sample or the majority-

minority comparison, profitability more strongly predicts the probability of foreign takeover

for the set of tax haven acquirers than for foreign non-tax haven acquirers. Specifically,

the tax haven group shows about twice as strong a preference for more profitable tar-


gets than does the latter. Had both groups exhibited similar magnitudes of sorting, the

concern would have been that the observed effect was driven by some other difference

between foreign and domestic bidders.

Overall, this table provides significant additional evidence that the observed prof-

itability sorting is due to tax differences between the bidders, since the relative tax rates

of haven and non-haven acquirers are quite clear.

1.4.3 Extensions and Robustness

Table 1.4 presents results from a number of extensions and robustness checks to the prof-

itability sorting result. Row (1) shows the baseline profitability estimates with accounting

and year controls, as in the first column of Table 1.2.

An important potential barrier for an acquirer attempting to shift income out of a

target company is the presence of minority oppression rules in the United States. These

dictate that a majority shareholder cannot enter into transactions that directly disadvan-

tage minority shareholders, at least without offering compensation. This would definitely

be a hurdle for a transaction which shifted income from one company to another company

owned by the majority shareholder, since this transfers income away from the minority

shareholder. For this reason, one would imagine that an income-shifting motivation

would lead to purchases of the whole target company (and thus buying out any existing

minority shareholders).17 This suggests looking at an alternative sample of deals, con-

sisting of only purchases of 100% of the target company. In such a sample, the tax effects

should be magnified, and row (2) shows this to be the case. This is not surprising, since

including transactions where income-shifting was not possible or was more costly should

bias the result downwards.

A possible concern is that different size acquirers have differential preferences over

target firm types, and, in turn, foreign and domestic acquirers vary in size, perhaps

because higher fixed costs preclude smaller foreign firms from making acquisitions in the

United States. To check this, row (3) includes a control for the log of acquirer total assets.

The coefficient estimates are similar to the baseline case. However, due to relatively poor

availability of this variable (the sample size drops from 5355 to 3814), mainly due to

non-publicly traded acquirers, it is otherwise not included in the models considered in

this study.

One important difference between foreign and domestic acquirers is in the type of

consideration used: foreign acquirers are more likely to pay cash for the target (49%

17This idea is consistent with Mintz and Weichenrieder [2005], who find that the leverage of Germanmultinational subsidiaries is sensitive to host country tax rates, but only for wholly-owned subsidiaries.


of takeovers) than are domestic acquirers (29% of takeovers), which is consistent with

Faccio and Masulis [2005]. To the extent that this difference is correlated with target

profitability, perhaps because relative bargaining strengths dictate that the bidder has to

use cash to pay for the highest quality targets, one might be concerned that it is driving

my results. In row (4), I include a dummy variable for cash-only deals, and find that it

decreases the profitability effect slightly, though it remains large and significant.

The sign and significance of the profitability effect is also preserved by using different

measures of profitability, such as pre-tax income, in row (5). Alternatively, in row (6),

rather than use the forecasting method described in Section 1.4, the profitability measure

is just the lagged ratio of earnings before interest, taxes, depreciation and amortization

to total assets, and a similar effect is estimated. The somewhat smaller magnitude is

also expected, given the likely presence of attenuation bias due to measurement error.

Row (7) proceeds in the opposite direction, by adding a second lag of profitability to the

future profitability forecast – the results are essentially unchanged.

An interesting observation is that dropping the smallest target firms from the sample,

in rows (8) and (9) substantially increases the estimated profitability effect. This provides

some compelling evidence against the asymmetric information story discussed earlier. It

seems reasonable that the larger the target firm, the more information about the firm

and its prospects would be available because of greater media and analyst coverage. In

other words, asymmetric information would seem to be most important for the smallest

target firms. Rows (8) and (9) show that the effect of profitability on the probability of

foreign acquisition is actually much stronger for larger targets.

In non-linear models, such as the probit, heteroskedasticity in the errors can lead

to inconsistency of the coefficient estimates. However, in this case, after accounting

for such heteroskedasticity (in the accounting controls) in row (10), the profitability

estimates actually get slightly larger.18 This is also encouraging in the sense that any

complementarity between existing profitability and the real takeover surplus would likely

manifest itself as heteroskedasticity in profitability.19 It would be theoretically possible

for such a complementarity to drive sorting even in the absence of tax differences, but

row (10) shows that this is not the case.

Overall, the profitability effects are positive and significant across a wide variety of

18The theoretical model can be extended in a straightforward way to allow for heteroskedasticity inthe idiosyncratic productivities of the two bidders. This result suggests that any such heteroskedasticityis actually working against the hypothesized results, and so strengthens the original conclusions.

19For instance, if the idiosyncratic productivity is multiplicative in the productivity of the target, thenwe would observe larger ‘errors’ for more extreme profitabilities. What matters is whether the surpluscaptured by the acquirer in a takeover is greater for high or low profitability firms, i.e. complementarityvs. ‘corporate turnarounds’.


specifications. Hence, the original cross-sectional estimates appear to be quite robust.

1.5 Full Results

1.5.1 Empirical Strategy: z

I now return to the strategy for estimating the tax shields term, using bonus depreciation,

a recently common (2001-2004, 2008-2010) feature of the U.S. tax code. It allows firms

to write off, for tax purposes, an additional 30% or 50% of the cost of new equipment

investment in the first year. Because different industries use different types of assets,

bonus depreciation affects industries differently, depending on the type of equipment

used and the division of investment between equipment and structures, which were not

eligible for the bonus treatment.

The general approach is to compare pre-BD (1990-2001Q3) with all post-BD (2001Q4-

2010) transactions, given potentially strong anticipation effects from 2005-2007.20 This

suggests a clear difference-in-differences empirical strategy. Recall that the theory says

that industries which got a relatively large increase in tax shields from bonus depreciation

should experience relative decreases in the probability of foreign acquisition following the

reform.

Construction of tax shields measure

The construction of the bonus depreciation measure is based on Edgerton [2010] and

works as follows.

Let j denote an industry and k an asset type, then:

αPREj = Σkwjk,1997PVk,1997

and

αPOSTj = Σkwjk,1997(0.5 + 0.5(PVk,1997))

are the present values of depreciation allowances per dollar of investment pre- and post-

20House and Shapiro [2008] report a survey from the National Association of Business Economics takenin January of 2004 which found that 62% of business economists anticipated that bonus depreciationwould be extended past 2004.

20Note that we would expect such effects independently of whether the reform had large effects onmarginal investment due to substantial inframarginal cashflow benefits.


reform, respectively.

The asset weights, wjk, for each industry are from the detailed 1997 Capital Flows

table from the Bureau of Economic Analysis, and so would not be influenced by bonus

depreciation. From the expression, it is clear that the most affected assets are those with

the lowest pre-bonus depreciation present value of allowances, which tend to be those

with the longest depreciable lives.

This measure of the value of tax writeoffs varies across industry (based on the types

of assets in use) and over time only due to bonus depreciation. It varies from less than

.01 for Oil & Gas Extraction or Real Estate and Accommodation to greater than .05 for

Broadcasting and Telecommunications, Forestry and Fishing, Air Transportation, Water

Transportation or Paper Products. The distribution of the change in α for post-bonus

depreciation targets is shown in Figure 1.3. Importantly, the cross-industry structure of

the reform was determined directly by pre-existing levels of depreciation allowances, as

demonstrated by the expression above, and so was quite plausibly exogenous to takeover

activity.

To get the total value of future yearly depreciation allowances per dollar of assets

for a given firm rather than the value per dollar of investment, embodied in α, I need

a measure of investment. Specifically, I use investment rates by industry from 1997 (Ij)

to match the investment by asset data used to construct α, and to avoid endogeneity of

investment with respect to the reform. Multiplying this investment rate, which is just

investment divided by total assets, by α yields the desired measure of future tax shields

per year: zj ≡ αjIj.

1.5.2 Results

To estimate the full model, including possible sorting along the dimensions of both prof-

itability and tax shields, I implement a difference-in-differences framework, which is de-

rived directly from the theoretical model. The estimating equation is:

P (foreign i) = H(ϕYi − ψzPREi − ψPOST (zPOSTi − zPREi ) + θi) (1.3)

This is precisely as in equation (1.1), except that, notationally, I explicitly allow z to

vary around the reform. Table 1.5 presents the results.

The second row of results are all consistent with the theoretical prediction that in-

dustries with the highest increases in depreciation allowances should experience relative

declines in the probability of foreign acquisitions. In particular, in column (1), the semi-

elasticity of probability foreign with respect to the tax shields measure is -35.41 (standard


error: 20.75), which, for a one standard deviation increase in zPOST − zPRE, amounts

to a decrease of 2.2 percentage points in the probability of foreign acquisition. Note

that the model of this column is not a full difference-in-differences model, as it includes

zPRE as a regressor rather than industry dummies to control for pre-reform differences in

probability foreign for different levels of tax shields. Once industry controls are added in

column (2), the semi-elasticity actually increases in magnitude to -44.44 (se: 20.71) and

is now strongly significant. A possible concern is that this change in probability foreign

is driven by industry trends surrounding the reform rather than the reform itself. To

that end, column (3) includes 20 industry-specific time trends, and the estimate actually

increases significantly, which suggests that secular industry trends in foreign takeovers

are actually working against finding an effect from the reform.

Given that bonus depreciation was enacted in 2001 (and was made retroactive to

September 11, 2001), one might be concerned that the effect of the reform on foreign

takeovers is confounded with heightened regulatory sensitivity to the security implications

of such takeovers. To account for such changes, I collect data published by the Committee

on Foreign Investment in the United States (CFIUS), which is tasked by Congress with

assessing the national security implications of foreign takeovers in the U.S. and potentially

blocking them, either overtly or by dissuading the potential acquirer. They reveal, at the

three or four digit NAICS level, the distribution of target firms for which a notice was

filed. I use this disclosure to encode a dummy variable which is equal to unity if CFIUS

reported a covered transaction in that target’s industry in any year since 2005 (the start

of public availability of the data). This is the case for about 58% of the post-reform

transactions in my sample. I then include this dummy as well as its interaction with the

post-reform dummy as additional controls in θi in equation (1.3) in column (4) of Table

1.5. The tax shield semi-elasticity is actually somewhat larger than the baseline case of

column (2), and neither the additional security dummy nor its post-reform interaction is

large or statistically significant.

Additional specifications, where the national security dummy variable is unity only

if CFIUS covered transactions are above some minimum level relative to the number

of transactions in my sample,21 provide a very similar story, suggesting that changes in

concerns about national security are not driving the observed sorting around the bonus

depreciation reform.

Examining the first row of Table 1.5 reveals that the profitability semi-elasticities

21The idea is to count only industries with serious security concerns – for example, restricting toindustries with at least half as many CFIUS notices as transactions in my sample covers about 25% ofpost-reform transactions, and yields similar results.


estimated from the full model are very similar to those of the previous section. This is to

be expected, as profitability and the change in tax shields from bonus depreciation are

basically uncorrelated, conditional on the basic set of accounting controls.

The tax shields estimates can be applied to directly calculate the extensive margin

effect of the tax distortions – that is, how much would the fraction of foreign acqui-

sitions change in aggregate from 2001 onwards if bonus depreciation had never been

implemented. This involves comparing the actual probability of foreign takeover for each

target with the counterfactual probability in the absence of bonus depreciation. This is

easily accomplished by setting zPOST = zPRE in equation (1.3) and calculating the new

probability, then averaging over industries.

The ownership effects of bonus depreciation, broken up by industry, are shown in

Figure 1.4. The dark bar shows the estimated probability of foreign takeover for each

industry, while the addition of the light bar indicates how much higher this probability

would have been in the absence of the reform. The difference goes in the same direc-

tion for all industries (since the reform always increased z) and is largest for industries

with high investment rates and large benefits from the reform, such as the construction

and transportation sectors. On the other hand, an industry like real estate, where the

prevalence of structures limits the relevance of bonus depreciation on equipment, and

which makes relatively little investment per dollar of assets, was not much affected by

the reform.

The estimated aggregate effect was to decrease foreign ownership following a takeover

by 5.3 percentage points in the post-reform period. Specifically, I find a counterfactual

aggregate foreign ownership probability of 24.3%, relative to an estimated 19.0% in the

presence of the reform. This roughly corresponds to a change in post-transaction owner-

ship from foreign to domestic of $190B worth of firms (measured by total assets), which

constitutes a striking side effect of a tax reform that ostensibly had nothing to do with

asset ownership decisions.22

1.5.3 Capital Markets

The preceding sections have examined and verified the two key predictions of the the-

oretical model in Section 1.2. However, much more can be learned by comparing the

magnitudes of these two effects. In particular, their ratio sheds light on the nature of

22To put these magnitudes into context, they can be compared to the ownership variation estimated byHuizinga and Voget [2009]. Their counterfactual experiment envisions a U.S. tax reform moving from aworldwide to a territorial system of international taxation and they find that the fraction of cross-borderdeals involving a U.S. company that end up with a U.S. headquarters would increase from 48% to 56%.


discount rate differences between bidders, determined in capital markets. Intuitively,

this is because both effects can be decomposed into a direct cashflow effect (i.e. more

after-tax cash remaining from given profitability for a foreign acquirer or more after-tax

cash remaining from given tax shields for a domestic acquirer) and a cost of capital ef-

fect. Hence, this ratio can be used to test for the two extreme cases for cost of capital

differences. First of all, if this ratio is unity, then the cost of capital is the same for the

two bidders; if it is zero, then costs of capital fully reflect tax differences between the

bidders.

The third row of results in Table 1.5 shows this key ratio as well as its standard

error. The estimates are all near zero, and we can always reject the hypothesis that the

true value is unity with a high degree of statistical significance, which would be the case

of identical discount rates for the two bidders. Typically, the hypothesis that the true

value is, in fact, zero, cannot be rejected, which means that discount rates approximately

reflect the full tax differences across bidders. This means that taxes are fully shifted

back to capital suppliers, so that the bidder facing the relatively lower tax rate faces a

commensurately higher cost of capital. Another way of expressing this point, which will

be important in the wealth calculation that follows, is that the pre-tax cost of capital,

r∗i ≡ ri1−τi is the same in each country. Hence, for equal tax rates and real productivities,

a given level of profitability makes the same contribution to world wealth regardless of

the owner of the asset, even though after-tax costs of capital are not equalized across

bidders.

1.5.4 World Wealth

The striking extensive margin effects from bonus depreciation illustrated in Figure 1.4

lead naturally to the question of the importance of this channel to shareholder wealth

and tax revenues. Given the multinational focus of the model, the natural benchmark

is world wealth. The goal is to find an empirically implementable expression for the

change in world wealth from a change in the generosity of tax shields. Let s be the share

of tax revenue going to the foreign country in case of a foreign acquisition (since some

tax revenue would leave the domestic country and possibly accrue to the foreign country

through post-merger income-shifting).23 Then we can write world wealth as the value

of the firm plus tax revenues, discounted by the relevant country-specific pre-tax rate of

23Effective tax payments by the acquirer are actually composed of payments to both governmentsplus transaction costs related to income-shifting. I assume that these extra costs can be thought of aslump-sum transfers to other agents in one of the two countries, so that a fraction s of effective taxes go(lump sum) to foreign agents and 1− s to domestic agents.


return:

WW = If

[1− τfr∗f

Y +τfr∗fz + ϵf + θf + (

sτfr∗f

+(1− s)τf

r∗d)(Y − z)

]

+ (1− If )

[1− τdr∗d

Y +τdr∗dz + ϵd + θd +

τdr∗d(Y − z)

]= If

[(1

r∗f− 1

r∗d)Y + ϵf − ϵd + θ − (1− s)τf (

1

r∗f− 1

r∗d)(Y − z)

]+Y

r∗d

= If [ϵf − ϵd + θ] +Y

r∗d

where If is an indicator variable for a foreign takeover and the last line uses r∗f = r∗d, as

found in the previous subsection.

The final line makes clear that the optimal decision rule (If ) is to grant the target to

the foreign acquirer if ϵf+θ > ϵd; that is, to let the winner be the bidder with the highest

real productivity, which can only be the case when taxes do not affect ownership. This

would be the case if either tax differences were eliminated or the two effects happened to

be exactly offsetting.

Figure 1.6 graphically illustrates the change in world wealth from an increase in the

generosity of tax shields.

1.5.5 Empirical Implementation

The preceding results are all independent of the scale parameter, which arises, as in

all discrete choice models, because of scale invariance. That is, one could multiply each

valuation by some constant and not change any of the results on the extensive margin. In

the probit models which have been used to this point, there is an implicit normalization

of the error variance to unity.

We want the change in wealth going from zPRE to zPOST (due to bonus depreciation)

expressed in terms of identified parameters. Integrating over ϵ ≡ ϵf − ϵd:

∆WW =

∫ ϕY−ψzPRE+θ

ϕY−ψzPOST+θ

[ϵ+ θ]dF (ϵ)


The issue is the normalization of the errors: ϵ ≡ ϵ/σ which implies ϕ ≡ ϕ/σ etc. where

the ‘hat’ parameters are what is produced by the estimation. Substituting yields:

= σ

∫ ϕY−ψzPOST+θ

ϕY−ψzPRE+θ

[ϵ+ θ]f(ϵ)dϵ (1.4)

This is the world wealth change, per dollar of target assets, from the ownership effects

of the reform.24 To get the aggregate change, this expression is multiplied by the total

assets of the target and summed over all targets in the market in the post-reform period.

There are several important assumptions underlying this expression. First of all, I must

assume that the costs of capital are not themselves affected by the reform, though this

is consistent with the finding in Desai and Goolsbee [2004] that bonus depreciation led

to investment increases of only one to two percent. A related point is that this is the

change in wealth from the ownership margin only, and so does not include the potential

effect of these induced changes in investment levels.25

Regardless of which set of estimates is used to calculate this wealth effect, the result

is always a negative number times the (positive) unknown scale parameter. This is not

surprising, since the tax shields effect, which discourages foreign acquisitions, outweighs

the positive profitability effect in the empirical results, so that the estimated tax wedges

always discourage foreign ownership on net. Then, since world wealth is falling in the

magnitude of the tax wedge, and bonus depreciation increases the size of this wedge, the

net effect is negative.

However, for comparative purposes, it is very useful to have an actual dollar measure

of the change in wealth, beyond just identifying the direction of the change. For this,

an estimate of the scale parameter, σ, is necessary, as the estimation procedure above

cannot identify it. Hence, further data are necessary.

Estimation of scale

Intuitively, to transform the estimated quantity distortion into a dollar value, it is nec-

essary to know something about the valuation of the runner-up bidder. Then, given the

already estimated tax wedge, one could calculate how much real value was potentially

lost by the less productive bidder acquiring the target. With ideal data, it would be

24The cutoff productivities are not affected since each component is normalized, so that the cutoffis scale invariant. This is exactly what allows calculation of the counterfactual probabilities withoutworrying about the scale.

25To the extent that the reform actually caused increased investment, the valuation difference betweenbidders with different tax rates would actually increase, exacerbating the distortion.


possible to estimate a model with an additional equation describing the difference be-

tween the two highest bids, which would allow the scale to be identified. Unfortunately,

the runner-up’s valuation is usually not observed and so this multiple equation approach

is not feasible. However, I have collected a dataset of losing bids from SDC and from

media descriptions of merger fights. This is sufficient to recover a rough estimate of the

necessary parameter.

Exhaustive search yielded a dataset of 300 cases in the original sample with an iden-

tifiable losing bidder and associated bid. Of these, 48 have a foreign winning bidder

and domestic losing bidder or vice versa. However, in some specifications, I include the

remaining cases with matching winner and loser to increase the sample size. This should

bias my estimate downward, because in cases where the winner and loser are both do-

mestic, the difference in their bids should be strictly lower than that between the winner

and the unobserved highest foreign bid.

The estimating equation is as follows, where the quantity of interest is the standard

deviation of the residual, ei:

Pdi − Pfi = βXi + ei (1.5)

where Pdi is the price offered by the top domestic bidder, Pfi is the price offered by the top

foreign bidder, Xi is a broad set of target level controls, including both tax and non-tax

variables, and i indexes the target firm. Both prices are normalized by the total assets of

the target. This formulation parallels the valuation difference from the theoretical model.

To the extent that some of the surplus in the acquisition is captured by the acquirer, the

estimated standard deviation will understate the true variation.26

Including various sets of controls, paralleling earlier sections, yields a root mean

squared error of approximately 0.38; this estimate is not much changed by the inclusion

of deals where both bidders are either domestic or foreign. In the context of the model,

this parameter is the standard deviation of the difference in idiosyncratic productivities

between domestic and foreign bidders per dollar of assets.27

As can be seen from equation (1.4), the unitless estimates described above must be

multiplied by this scale to get a dollar value for the wealth change. This procedure yields

26Andrade et al. [2001] survey the literature and perform their own updated empirical analysis to findthat approximately all gains from a merger accrue to target firm shareholders, though this is an area ofconsiderable recent debate. For example, Netter et al. [2011] find that the gain to acquirers is usuallypositive in a very broad sample of takeovers; Savor and Lu [2009] use exogenous takeover failures to showthat stock mergers create value for the acquirer’s shareholders; Ahern [2012] reports that the averagegains to the target and the acquirer in a merger are approximately equal.

27I also consider an alternative specification wherein I treat the observed market value prior to thetakeover bid as an estimate of the next-best valuation – this allows for a much larger sample size of 3910.In this case, the estimated standard deviation is about 0.5.


a wealth loss of approximately $360 billion from 2001 to 2010, relative to $7,325 billion

worth of assets traded in the M&A market in my sample. This corresponds to a novel

welfare effect from this reform of $36 billion per year. Alternatively, the aggregate loss is

worth about 5% of the total assets of target companies in the bonus depreciation period.

It is important to note that the estimated effect comes from taking the set of acquired

firms as given, and so does not include changes driven by reform-induced selection into

or out of this sample. To the extent that tax effects do not seem to be important to

selection in the first place, this channel appears to be relatively less important.

Figure 1.5 shows how this wealth loss varies across specifications from Table 1.5 and

for different values of the scale parameter, in terms of both percentage of assets and in

dollar terms. Across all specifications and for a wide set of scale parameters, the implied

distortion is large, especially in the context of the magnitude and goal of the reform.

This wealth change is made up of two parts: tax revenues and shareholder wealth.

Since the reform led to a decrease in foreign acquisitions, which are assumed subject to

a lower tax rate, tax revenues must have actually increased.28 Hence, shareholder wealth

fell by more than the $360B figure. How this loss was distributed between foreign and

domestic shareholders depends on how the takeover price distributes the deal surplus be-

tween target and acquirer shareholders. If, for example, target firm shareholders receive

approximately the whole surplus, as suggested by Andrade et al. [2001], then the full

effect of the shareholder wealth loss accrues to those shareholders through lower trans-

action values. In this empirically plausible case, domestic wealth falls, highlighting the

importance of this channel to domestic policymakers.

1.6 Conclusion

This chapter presents a model of cross-border mergers and acquisitions which gives a

set of empirical predictions about the nature of tax clienteles. The empirical results

show that foreign acquirers systematically target more profitable firms for acquisitions.

As would be expected if this observation is driven by tax differences, the results are

strikingly larger for tax haven-resident acquirers. Furthermore, an exogenous increase

in the value of tax shields for firms in particular industries leads to relative decreases in

foreign acquisitions in those industries most affected by the reform. These results are all

consistent with the theoretical model.

The model also implies that the relationship between the magnitudes of these two

dimensions of sorting can be used to test for cost of capital differences between the two

28This abstracts from the direct effect of the reform on tax revenues, as discussed above.


types of bidders. In particular, the empirical results imply the interesting finding that

the tax differences between foreign and domestic bidders are strongly reflected in their

discount rates. This has significant implications for optimal tax policy and, in particu-

lar, highlights the importance of differential valuation of tax shields (or tax bases more

broadly) in determining ownership of assets, which has been an under-appreciated point

in previous literature. Increasing the availability of tax shields, perhaps by increasing

the generosity of depreciation allowances for given investment, appears to be a powerful

way of influencing the market for corporate control to the advantage of domestic acquir-

ers. However, simulations using variation in these allowances from bonus depreciation

suggest that the induced shift in ownership towards domestic companies actually has

a large, negative effect on world wealth. Whether such a reform is nonetheless good

for the domestic economy depends on whether the existing level of foreign ownership is

too high or too low, and how transaction gains are shared between target and acquirer

shareholders.

Overall, this chapter provides a variety of evidence for the importance of tax factors

in the market for corporate control, which significantly affect the pattern of foreign

ownership both within and across industries. Several different policies could be pursued

to address this distortion. Increasing barriers to income-shifting, either through stronger

enforcement or stricter transfer pricing and earnings stripping rules, would address the

problem to the extent that differences in discount rates are only due to income-shifting.

However, such an approach has already proven difficult, as evidenced by my results, and

would have its own costs, in terms of distorting real cross-border production, research

and financing decisions. Notwithstanding any associated net revenue losses, a decrease in

the statutory corporate income tax rate would directly decrease the incentive for income-

shifting, which would decrease the valuation wedge between foreign and domestic bidders

and so lead to a more efficient ownership pattern and higher aggregate wealth.


1.A Sample Construction

A small complication is caused by some cases of multiple transactions associated with one

announcement date. To deal with these, all the transactions from the same announcement

date, for the same target and acquirer, are aggregated by adding up the transaction

values and fraction of shares acquired to yield a single transaction that is included in the

estimation sample.

N Value ($B)

All mergers with U.S. target 145,619 15,298

...target is public 9,970 8,735

...match in Compustat 7,565 8,341

...meet size restrictions 6,809 8,286

...necessary accounting controls 5,939 7,120

...necessary acquirer type 5,383 6,461

Most unmatched public companies are due to being listed on exchanges that are not

covered by Compustat. The remaining missing matches are due to changes in CUSIPs

and company names in the early 1990s, before SEC EDGAR data were available to aid in

the matching. Though the number of transactions declines a fair amount after imposing

necessary restrictions, the estimation sample still contains a significant fraction of the

relevant deals by transaction value.


Mining

Utilities

Construction

Manufacturing

Wholesale Trade

Retail Trade

Transportation

Information

Finance & InsuranceReal Estate

Professional, Scientific & Technical Services

Admin. Support, Waste Man. & Remediation Services

Health Care & Social AssistanceAccommodation & Food Services

.05

.1.1

5.2

.25

Fra

ctio

n F

orei

gn

0 .05 .1 .15Median Profitability

Fraction Foreign vs. Profitability

Figure 1.1: Within 20 NAICS-defined industries, this is a scatter plot of the fraction oftargets which were acquired by a foreign bidder against the median profitability in thatindustry. The size of the datapoint is a qualitative indicator of the number of transactionsobserved in that industry.


Mining

Utilities

Construction ManufacturingWholesale Trade

Retail Trade TransportationInformation

Finance & Insurance

Real Estate

Professional, Scientific & Technical ServicesAdmin. Support, Waste Man. & Remediation Services

Health Care & Social AssistanceAccomodation & Food Services

−.1

−.0

50

.05

.1F

ract

ion

For

eign

−.02 0 .02 .04 .06Median Profitability

Fraction Foreign vs. Profitability: Merger − Stake

Figure 1.2: For each of 20 NAICS-defined industries, the y-variable is the differencebetween the probability of foreign acquisition and the probability of a foreign stakepurchase; the x-variable is the relative difference in profitability between majority andminority acquisition targets. The size of the datapoint is a qualitative indicator of thenumber of transactions observed in that industry.


020

040

060

080

0F

requ

ency

0 .02 .04 .06 .08Alpha(PostBD) − Alpha(PreBD)

Bonus Depreciation Variation per $ of Investment

Figure 1.3: This is a histogram of the bonus depreciation-induced change in the presentvalue of depreciation allowances for all post-reform acquisition targets. It shows thechanges in α, the present value of depreciation allowances per dollar of assets, inducedby the reform.


0 .1 .2 .3Fraction Foreign

Construction

Utilities

Real Estate

Accomm. & Food Services

Health Care & Soc. Assistance

Transportation

Wholesale Trade

Support Services

Retail Trade

Mining

Professional Services

Information

Finance & Insurance

Manufacturing

Post−BD Ownership Changes

Figure 1.4: This shows the ownership changes caused by bonus depreciation – the darkline shows estimated foreign probability with the reform and the light line shows thecounterfactual effect of removing the reform.


200

400

600

800

1000

Bill

ions

of $

05

1015

% o

f Ass

ets

.2 .3 .4 .5 .6Scale (SD)

Base Model Industry Trends No Industry

Welfare Loss from Bonus Depreciation

Figure 1.5: This shows the loss in world wealth from the ownership changes caused bybonus depreciation (for each of the specifications from Table 1.5 and different values forthe σ parameter.


ϵϵ∗

ϵPRE

ϵPOST

Ω

θ+ϵ

θ+W

PRE+ϵ

θ+W

POST+ϵ

0

Figure

1.6:Theshad

edarea

isthewealthchan

gecausedbyachan

gein

thetaxwedge

from

WPRE

≡ϕY

−ψzP

RE

toW

POST,withproductivitydifference

ϵ≡ϵ f

−ϵ d

and

theoverbarsdenotingthecutoffvalueforeach

taxwedge.Theillustratedcase

show

sa

negativeinitialtax

wedge

(asisfoundem

pirically),whichdiscouragesforeignacquisitions,

andso

lead

sto

acutoffproductivitydifference

(foreign

less

dom

estic)

that

ishigher

than

thefirstbest.

Theincrease

intaxshieldsmakes

thetaxwedge

morenegativean

dso

causesaloss

inworld

wealth.


Table 1.1: N = 5383 for main estimation sample. The ‘All Firms’ category includes allfirms in Compustat from 1990-2010 with greater than $10M in assets and non-missingvalues for all accounting controls. Standard errors are in parentheses.

Majority Sample All Compustat FirmsMedian Mean Mean

Total Assets ($M) 221 2,324 6,222(21,179) (58,681)

Profitability (%) 7.0 5.0 5.8(14.0) (22.6)

I(Prof. < 0) (%) - 20.2 18.8Intangibles (%) 1.9 11.0 10.3

(17.1) (16.7)Debt (%) 8.2 17.5 18.9

(24.8) (28.5)Foreign (%) - 16.0 -Haven (%) - 2.0 -

N 5,383 5,383 142,739


P (foreign i) = H(ϕYi − ψzi + ηXi + ui)

Table 1.2: Probits, all containing accounting controls and year dummies. Values aresemi-elasticities of probability foreign with respect to profitability. *, **, *** denotesignificance at 10%, 5% and 1%, respectively. Mean probability foreign is 0.159, and thestandard deviation of profitability is 0.140. Standard errors are bootstrapped over 100repetitions to account for variability in the construction of the profitability measure.

Majority Majority - MinorityProfitability * Majority 2.204*** 1.265** 2.996*** 2.253***

(.489) (.509) (.775) (.807)Profitability - - -0.797 -0.994*

(.563) (.594)Total Assets * Majority .0285 .0849*** .0204 .04938*

(.0205) (.0221) (.0251) (.0286)Total Assets - - .00795 0.0352

(.0301) (.0346)Debt Ratio * Majority -.557* -.414* -.656 -.511

(.316) (.234) (.496) (.449)Debt Ratio - - .101 .0983

(.326) (.358)Intangibles * Majority -.0678 .0236 .118 -.00983

(.573) (.419) (.698) (.547)Intangibles - - -.186 .0334

(.315) (.358)Loss Dummy * Majority .691*** .455** .222 .0287

(.122) (.132) (.219) (.234)Loss Dummy - - .467*** .424***

(.152) (.158)Industry N Y N YN 5383 8715


Table 1.3: Probits, all containing accounting controls (log total assets, intangiblesratio, debt ratio and dummy for negative earnings) and year dummies. Values aresemi-elasticities of probability foreign with respect to profitability. Standard errors arebootstrapped with 100 repetitions to account for variability in the construction of theprofitability measure.


Majority Majority - Minority(1) (2) (3) (4)

Profitability 2.078*** 1.040* 2.794*** 2.068**foreign i = non-tax haven foreign acquirer (.560) (0.597) (0.894) (0.924)

Profitability 4.087*** 3.653** 5.468** 4.785**foreign i = tax haven acquirer (1.522) (1.624) (1.799) (2.023)

Industry N Y N YN 5277/4432 8480/7028


Table 1.4: *, **, *** denote significance at 10%, 5% and 1%, respectively. Standarderrors are in parentheses following the coefficient estimates. Each row includes accountingcontrols (log total assets, intangibles ratio, debt ratio and dummy for negative earnings)and year dummies. Standard errors are bootstrapped with 100 repetitions to account forvariability in the construction of the profitability measure.


Dep. var.: foreign acquirer = 1, domestic taxable acquirer = 0Profitability N

(1) Baseline 2.204*** (.489) 5383(2) Only Full Control Transactions 2.612*** (.586) 4738(3) Control for Acquirer Assets 3.100*** (.687) 3814(4) Control for cash deals 2.062*** (.515) 5383(5) Y = pre-tax income / assets 1.249*** (.389) 5383(6) Y = lagged EBITDA / assets 1.705*** (.373) 5383(7) Two profitability lags 2.231*** (.491) 5383(8) Total Assets > $25M 3.006*** (.536) 4879(9) Total Assets > $100M 3.976*** (.859) 3417(10) Allowing for heteroskedasticity in profitability 2.483*** (.533) 5199


Table 1.5: *, **, *** denote significance at 10%, 5% and 1%, respectively. All probitsinclude accounting controls (log total assets, intangibles ratio, debt ratio and dummyfor negative earnings) and year dummies. Standard errors are bootstrapped with 100repetitions and clustered at the industry level to account for variability in the constructionof the profitability measure. Column (4) also includes a national security dummy as wellas its interaction with the post-reform dummy (unreported, not statistically significant).

P (foreign i) = H(ϕYi − ψzPREi − ψPOST (zPOSTi − zPREi ) + θi)

(1) (2) (3) (4)Profitability (ϕ) 2.068*** 1.191** 1.192** 1.186**

(.759) (.470) (.484) (.496)Tax shields (ψ) 35.41* 44.44** 58.02** 52.16**

(20.75) (20.71) (25.97) (26.11)

ϕ/ψ .058 .027 .021 .022(.047) (.027) (.015) (.019)

Industry N Y Y NIndustry Trend N N Y NN 5366 5366 5366 5366

Chapter 2

Taxation and Executive

Compensation: Evidence from Stock

Options

2.1 Introduction

A clear understanding of how taxes affect executive compensation can provide valuable

insight into fundamental questions in both corporate and public finance. For example,

the nature of the process determining executive pay is an area of much debate in the

literature on corporate governance. Proponents of the board capture theory, such as

Bebchuk and Fried [2003], argue that managers wield substantial influence in bargaining

with boards of directors over their own pay. The magnitude of the compensation response

to a change in tax rates yields useful information about the extent of this bargaining

power. Furthermore, in light of rising income inequality and concern over government

budget deficits, there has been growing interest on the part of policymakers and the

general public in increasing top income tax rates. The possibility and desirability of

such a policy hinges critically on how high income earners, such as corporate executives,

respond to increases in their tax burden.

A recent tax reform in Canada, which greatly increased the effective tax rate on

stock option compensation for a subset of firms, provides an excellent opportunity to

study these issues. From mandatory public filings, I collect a novel panel dataset of

compensation for the top five executives at 600 firms for the four years from 2008 to 2011,

which includes a breakdown of the components of pay – the most important of which

are salary, cash bonus, share-based compensation (restricted or deferred share units) and

41

Chapter 2. Taxation and Executive Compensation: Evidence from Stock Options 42

stock option grants. There is considerable cross-firm and within-firm heterogeneity in

the fractions of compensation derived from these four sources.

Difference-in-differences results, using executives at firms unaffected by the reform as

a control group, suggest that this policy-induced tax increase resulted in a reduction in

both stock option grants and the fraction of total compensation made up of stock options

in the two years immediately following the reform. The point estimate suggests that

option compensation fell by approximately the full value of the lost corporate deduction.

The natural related question is the extent to which compensation was substituted towards

other types of payment whose tax treatment did not change with the reform, such as cash

bonuses. There is little evidence of any such substitution response. Hence, the burden

of the tax increase appears to have been substantially borne by the affected executives.

Overall, these findings are in contrast to much of the existing literature, discussed below,

which typically has not found differences in tax incentives to be important determinants

of executive compensation.

These results are useful inputs to models of executive compensation bargaining, and

the effects of possible policy responses to increasing income inequality, as in Piketty et al.

[2011]. In particular, they imply a nontrivial taxable income elasticity. Whether or not

this high elasticity has negative efficiency consequences depends crucially on whether

compensation reflects bargaining over rents or the outcome of a competitive market.

The observed decreases in compensation are consistent with the idea that executives are

in a strong bargaining position with respect to the board of directors. Furthermore,

the significant impact of the reform highlights the importance of considering the inter-

play between firm and personal-level tax incentives as a key determinant of executive

compensation.

There is a long literature on the determinants of executive compensation mainly cover-

ing firms in the United States, which is well surveyed by Murphy [1999] and Frydman and Jenter

[2010]. There is rather less work using Canadian data, despite a similar institutional en-

vironment and a variety of regulatory and tax policy changes. However, there are a few

papers in this area written using data made available by the Ontario Securities Commis-

sion for the 1993-1995 period. Zhou [2000] documents in these data that pay is sensitive to

performance and rises with firm size and Zhou [1999] further finds that pay-performance

sensitivity is lower in Canada than in the U.S., but that this disparity diminishes as firm

size increases.

In the area of tax consequences, Mawani [2003] and Klassen and Mawani [2000] use

the same data source to show the importance of tax and financial reporting incentives


as determinants of stock option grants and exercises.1 At the time, accounting rules for

stock options meant that such compensation needed only to be mentioned in a footnote

of the financial statements, rather than being expensed like other forms of compensation.

Hence, the tradeoffs were quite different than under the current accounting regime, where

the value of option compensation must be expensed, and so reduces accounting income.2

A number of other papers have looked specifically at the effects of taxation and reg-

ulation on executive pay in the United States. Goolsbee [2000] studies the top personal

income tax rate increase of 1993 and finds a significant decrease in taxable income of cor-

porate executives in response. However, it turns out that this effect can be almost entirely

attributed to changes in the timing of stock option exercise, with little long run effect.

Rose and Wolfram [2000] and Rose and Wolfram [2002] investigate another component

of the 1993 legislation, which limited corporate tax deductibility of pay to one million

dollars per year, unless it qualified as performance-based.3 Their conclusion is that this

regulation had little effect on salaries or total compensation, which they argue suggests

that executive pay is insulated from this kind of policy intervention. Hall and Liebman

[2000] do not find any evidence that US tax reforms in the 1980s influenced the exercise

of stock options. In a similar vein, Frydman and Molloy [2011] find little evidence for

taxes as an important determinant of executive compensation using a series of tax re-

forms in a long panel of the compensation of US executives from 1946-2005. However, the

methodology underlying most extant studies makes it difficult to identify effects of taxes

since cross-sectional variation in tax incentives for otherwise similar taxpayers is rare in

US tax reforms. This leaves open the possibility that countervailing secular trends in

compensation or the endogeneity of tax reforms is hiding the causal effects of tax policy.

On the other hand, on the broader question of taxable income elasticities, there

is evidence from outside of the executive compensation literature that the taxable in-

come of high income taxpayers does indeed respond to changes in tax rates. Specifically,

Gruber and Saez [2002], using US panel data, and Sillamaa and Veall [2001], using Cana-

dian data to investigate a 1988 rate reform, find the highest taxable income elasticities

for high income taxpayers. To some extent, my study bridges the gap between findings

such as these and a general lack of evidence for tax effects on executive compensation.

The remainder of the chapter proceeds as follows: Section 2.2 discusses how stock

1There is a similar literature on determinants of option compensation in the US, including Matsunaga[1995], Matsunaga et al. [1992] and Yermack [1995].

2Expensing of stock option compensation was required in Canada as of January 1, 2004, and in theUS for fiscal years beginning after June 15, 2005.

3The structure of the reform studied in this chapter is similar, though focused on performance-basedpay, rather than salaries.


options are taxed in Canada, both before and after the reform, Section 2.3 describes

the executive compensation data, Section 2.4 describes the theoretical considerations

underlying the determination of executive compensation and what they predict about

the effects of the reform, Section 2.5 presents the results of regressions attempting to

uncover the causal effect of the reform on compensation and Section 2.6 concludes.

2.2 Stock Option Taxation and the 2010 Reform

A typical employee stock option is a right granted to the employee to acquire a share

for a particular price (the exercise price) until a particular date (the expiration date).

The option vests, or becomes exercisable, at the vesting date, typically some time after

the grant date. The exercise price is typically set at the market price of the stock on

the grant date of the option.4 Stock options are a large and important component of

executive compensation plans, being granted to about 70% of executives at publicly listed

Canadian companies with market capitalization of more than one billion dollars. Stock

options make up about a third of compensation for those executives that receive them.

Prior to the 2010 Budget, taxation of stock options in Canada worked as follows: as

long as the exercise price of the option was at or above the market price at the grant

date, no tax is immediately due. When the option is exercised by the executive, she must

pay tax on the difference between the market price and the exercise price; however, the

‘stock option deduction’ provides preferential capital gains treatment on the income.5

Hence, the typical case is that the firm never gets a tax deduction, while the executive

pays tax at half her personal marginal income tax rate. This tax treatment is illustrated

in column (1) of Table 2.1.

However, the firm and executive can agree to ‘cash-out’ the option rather than have

it exercised conventionally, whereby the firm makes a cash payment to the executive to

cancel the option. The result is a deduction for the corporation, due to the cash outlay,

and regular employment income for the executive, as shown in column (2) of Table 2.1.6

Essentially, the status quo on exercise is beneficial tax treatment for the exercisee. With

a cash-out, the beneficial tax treatment goes to the firm. Of course, the amount of the

4In fact, this is always the case for publicly traded companies in Canada since Toronto Stock Exchangerules prohibit the granting of ‘in-the-money’ options, that is, those with an exercise price below themarket price at grant. Likewise, ‘out-of-the-money’ grants are very rare – Hall and Murphy [2002] notethat 94% of option grants to chief executive officers in the S&P 500 in 1998 were ‘at-the-money’.

5This parallels the tax treatment of incentive stock options (ISOs) in the U.S., though these are notcommonly used in large publicly traded companies.

6This approximates the treatment of non-qualified stock options (NQOs), the predominant form ofemployee stock options in the United States.


cash payment can be adjusted to share the tax savings with the employee in situations

where the potential corporate tax benefit is larger. This type of bargain is studied by

Mawani [2003], who finds that cash-outs are predicted by high effective corporate tax

rates and strong financial reporting incentives.

The interesting wrinkle to this dichotomy, which forms the basis of my study, is that

there is a particular type of stock option which was, before 2010, eligible for beneficial

treatment, simultaneously, for both the firm and the employee, as in column (3) of Table

2.1. This type of stock option combines a regular option with a stock appreciation right,

together called a tandem stock appreciation right (TSAR) and sometimes referred to as

an option with a cash settlement feature. When an employee exercised a TSAR, she

could elect to receive cash from the firm equal to the excess of the market price over the

exercise price, while retaining the stock option deduction (because, ostensibly, a stock

option was in fact being exercised). The firm could then deduct this cash payment. The

main requirement for a TSAR to receive this beneficial treatment was that the choice to

receive cash had to be at the discretion of the employee.7 For the purposes of the empirical

strategy employed in Section 2.5, it is important to note that the decision to use TSARs,

conditional on having a stock option plan, is basically a one time decision, after which

the provisions of the stock option plan are modified to include a cash settlement feature.

These types of modifications are very infrequent, so that switching back to stand-alone

options prior to 2010 appears to have occurred in only a few cases. Furthermore, it is

very rare that a firm will issue a mix of TSARs and regular options to executives in the

same year – firms choose one type or the other.8

However, a significant change to the ‘double deduction’ tax treatment for TSARs was

announced as part of the federal budget on March 4, 2010. Essentially, both existing and

new TSARs would now be treated identically to regular options for tax purposes. The

employee stock option deduction is only available if the employer foregoes its deduction

(or vice versa). Because the plan must have been set up so that the choice to take cash is

up to the employee, the effect of the reform was to remove the corporate deduction, which

was worth about 30% of the gain on exercise, depending on the combined federal and

provincial statutory corporate tax rate in 2010, and the taxable status of the corporation.

Additional evidence that it was in fact the corporate deduction that was lost after the

reform, rather than the employee’s stock option deduction, comes from investigation

7In addition, the option cannot have been granted in the money, though as discussed above, thisrequirement is always satisfied for publicly traded firms.

8Only four companies in the sample granted a mix of both types of options in 2009. In each of thesecases, TSARs appear to make up the vast majority of option grants, with the minority of regular optionscoming either from a corporate subsidiary or a legacy plan.


of financial statements of firms using TSARs prior to the reform. For example, Rogers

Communications, a large telecommunications conglomerate, reported a $40M tax increase

for 2010 from the stock option tax change. This is approximately equal to their reported

marginal tax rate for the year multiplied by their existing stock option compensation

liability, indicating that they expected not to be able to deduct this future liability in

computing their taxable income. Other companies making use of TSARs, such as Encana

and Canadian Pacific, reported similar tax increases with specific reference to the TSAR

provision of the 2010 Budget.

Several other aspects of stock option taxation were changed at the same time as the

reform of TSAR treatment. Previously, an employee could defer paying tax otherwise

due at option exercise until actually disposing of the shares for up to $100K worth of

shares per year; this was eliminated for publicly traded companies.9 Other changes re-

lated to options included strengthening of employer withholding requirements, which had

previously benefited from an administrative waiver as long as no cash was paid. Lastly,

changes were made to alleviate the tax burden on ‘phantom income’. This situation

arose, largely because of options granted by technology firms prior to the 2001 recession,

when employees had exercised options at high prices and then taken advantage of deferral

to delay paying tax until many years later. Unfortunately for these employees, the stock

prices of some technology bubble firms never recovered, leaving them with a tax liability

much larger than the value of their stock. The budget addressed this issue by limiting

the tax liability on the deferred benefit of previously exercised options to the proceeds of

the sale.

The effective date for these reforms in the legislation that was eventually enacted was

identical to that envisioned in the original budget proposal: the announcement date of

March 4, 2010. Overall, this package of reforms increased the effective tax rate on stock

options; however, the incremental increase in tax from the TSAR provision was itself

quite substantial.10

Changes to the taxation of stock options may have been expected prior to the 2010

Budget, in particular to address the issue of option-related ‘phantom income’, as this

9This provision would be of greater importance for option grants to employees at smaller firms orthose below the executive level, since the upper limit on deferral was well below the average option grantin my sample. Hence, the marginal incentive change from this provision should not have been too large.Furthermore, notwithstanding the benefits of deferral, Heath et al. [1999] find that employees at a largecompany in the U.S. almost always used ‘cashless exercise’ whereby shares are immediately sold withthe employee receiving cash for the excess of the market over the exercise price.

10The budget documents forecasted incremental revenue from the TSAR portion of the reform ofabout $300M per year over the following five years. As a rough point of comparison, federal governmenttax expenditure estimates imply that aggregate income from all stock option exercise is about $4-5B peryear (with aggregate stock option deductions at half that amount).


problem had been discussed in the media in the preceding years. In fact, a pressure group

called ‘Canadians for Fair and Equitable Taxation’ had arisen to lobby the government

on precisely this issue. However, the provision related to TSARs appears to have been

a genuine surprise to firms and executives. Perhaps the strongest evidence of this fact

is that the board of directors of ShawCor, a multinational energy services company

headquartered in Toronto, actually introduced TSARs as of March 3, 2010 – one day

prior to the announcement of the tax change. On March 31, 2011, the board eliminated

the TSAR feature of their stock option plan.11 Hence, it seems quite likely that it was

the beneficial tax treatment of TSARs that led to their introduction and the board

of directors must have believed the day prior to the reform that this treatment would

continue.

2.3 Data

For Canadian firms, executive compensation disclosure requirements are detailed in Form

51-102F6: Statement of Executive Compensation. They require disclosure on the chief

executive officer, the chief financial officer and the next three most highly compensated

officers of the company in the ‘Summary Compensation Table’ which is part of the

annually filed Management Information Circular. The requirements for this table were

substantially amended and widened12 starting in 2008. In particular, firms now must

provide a valuation for any stock options granted (typically following the method of

Black and Scholes [1973]), rather than just the number of options granted, and report an

incremental value for executive pension plans, to the extent that they differ from what

is available to non-executive employees.

To construct a panel of executive pay, information was collected from the Summary

Compensation Table for each year from 2008, the start of expanded compensation dis-

closure, to 2011 for the 600 largest publicly traded Canadian firms. These tables give the

executive’s name and position and provide detailed information about seven components

of their compensation. Three of the components are non-equity based – they result in

cash paid to the executive. These are salary, which typically does not depend on perfor-

mance, cash bonuses and long term incentive plans. The difference between the latter

two is that bonuses depend on only a single year’s performance, while long term incentive

plans apply to several years. These forms of compensation are taxed as ordinary income

11Note that this timing means that ShawCor was not actually able to grant any TSARs that wereeligible for the beneficial tax treatment.

12Disclosure requirements have essentially changed in parallel in Canada and the United States.


when received by the employee, with a contemporaneous deduction for the corporation.

Several equity-based compensation amounts are reported: the value of stock option

grants, discussed above, and the value of share-based awards. The latter category includes

restricted stock units and deferred share units. These are grants of stocks (as compared

with options to acquire stocks), which typically vest over time. Restricted stock usually

vests over three years, while deferred stock must be held until the executive leaves the

company. Tax is paid by the executive at ordinary rates, but these plans are usually

structured so that the tax is deferred until the stock is actually received by the executive,

some time after the compensation is actually earned.13 At such time, the corporation

would be able to deduct the value of the compensation.

Lastly, firms must report the change in the accrued value of the executive’s pension

plan, if any, and a residual category, which includes such items as payments for supple-

mentary health or dental insurance and travel and housing allowances. Payments into

registered pension plans are immediately deductible to the corporation with the executive

only paying tax when receiving the income in retirement (with tax-free compounding of

investment returns on the contributed funds). Fringe benefits are typically immediately

deductible to the firm and are not included in the employee’s income.

This data collection yields a panel with very similar compensation information to that

available for US executives in ExecuComp. The collected raw data cover about 600 firms

and 11,000 executive-years over the period 2008-2011. With this panel of executives and

their compensation in hand, I search the Annual Information Form, another mandatory

corporate filing, to collect information on the place of residence of the executive14 and

whether or not they were a member of the board of directors.

The main estimation sample was formed by applying the following criteria: the ex-

ecutive must be resident in Canada, to ensure that Canadian tax rules apply, must be

employed for the full year, as compensation for a part year worked is difficult to in-

terpret,15 and must have compensation data available in all four years from 2008-2011.

This last requirement ensures that executives were employed at their firm for at least

two years prior to the reform and two years after it. This mitigates the concern that

compensation for executives who started just prior to the reform or who left the firm just

13This puts rather strict requirements on the structure of these plans, as the tax code ordinarily doesnot allow the deferral of taxes on benefits earned through work in the current period, outside of pensionplans.

14This information is important since a non-trivial fraction of executives actually live in a differentcountry from the corporate headquarters.

15For example, when an executive is hired mid-year, some aspects of their compensation, such asbonuses, may be determined as though a full year was worked, whereas the salary is prorated. Forexecutives terminated during the year, this issue is similarly problematic.


after it is driven by their place in the career life cycle, rather than changes in performance

or business conditions. While this induces some survivor bias, it ensures that executives

in the sample are more similar to one another, which is important given the relatively

small sample size.

Table 2.2 displays summary statistics for the elements of executive compensation in

this sample. The statistics are presented split by whether any options were granted,

as this distinction is important in the regression models of Section 2.5. Compensation

variables are winsorized at the 5% level to minimize the influence of outliers, which tend

to be one-time special option grants or bonuses.16

An interesting supplement to the reported summary statistics is the fraction of total

pay derived from each individual component of compensation. For executives not receiv-

ing any options, salary makes up half of compensation with another 30% from bonuses

and 15% from share-based awards. For the rest of the sample, 30% of compensation

comes from options, at the approximately equal expense of the other three categories.

In both cases, pensions and other compensation provide the remaining 5%. There is

considerable variation around these fractions both within and across firms.

To identify which firms in the sample were using TSARs in the pre-reform period,

I searched management information circulars, annual information forms and financial

statements, either directly for descriptions of the firm’s stock option plans or details of

the accounting method used to determine option compensation expense (which indirectly

reveals the presence of tandem stock appreciation rights). Of the executive-years in my

main estimation sample, about 12% were at firms that used TSARs in the pre-reform

period. Without imposing any of the sample restrictions, such as on executive residence,

the proportion falls to about 9%.

A rough estimate of the gains on TSAR exercises at firms in my sample for 2009 is

$500 million, which is calculated for each firm as the number of options exercised times

the difference in the year-end price and the reported average exercise price. This value

represents about 10% of economy-wide option gains, as reported in the tax expenditure

analysis of the federal government. This number is conservative in the sense that the

necessary data on option exercises and prices are not available for all TSAR firms.17

To get firm-specific accounting information, the compensation data were merged with

Compustat and supplemented by share price data from the Canadian Financial Markets

16Ideally, one would directly identify these non-repeatable elements of compensation and deal withthem separately, but this is not feasible with the available data.

17Note that the source of the data for this calculation is not the executive compensation disclosurebut rather the financial statements, and so reflects options exercised by all employees at the relevantcompanies, so as to be comparable to the aggregate stock options tax expenditure.


Research Center.

2.4 Theoretical Executive Compensation Framework

The effects of the TSAR reform depend crucially on how executive compensation is

determined. This process can be thought of in two steps. First, firms and executives

bargain over total compensation – it is at this stage where board capture might be

relevant, or perhaps a competitive market determines this value. This involves somehow

splitting the surplus associated with the firm-executive match. Then firms can choose the

composition of the compensation package from among multiple forms of compensation,

each of which may be taxed in a different way with different non-tax costs and benefits

to both the firm and the employee. These non-tax differences could involve inherent

differences in risk or accounting treatment, for example. Appendix 2.A develops and

solves a simple model along these lines, but the results and intuition thereof, which will

inform the empirical analysis, are described in this section.

Applying this framework suggests that, for TSAR companies, option compensation is

determined by trading off the tax savings from the corporate deduction against the net

non-tax costs. Then, as the reform removes these additional tax savings and so lowers the

marginal benefit of using options,18 optimal option compensation must fall to regain the

equilibrium. This implies that prior to the reform, too many options were being granted

at these companies relative to the social optimum. Firms and executives were engaged

in tax arbitrage, choosing a level of options higher than they otherwise would have at

the expense of their silent partner, the government treasury. This causes higher non-tax

costs; for example, risk may not be optimally shared between firms and their executives.

At the same time as options are falling, the total surplus generated by the match

between firm and executive falls as the tax savings disappear. This will cause executive

compensation to fall as well, with the magnitude depending on the executive’s bargaining

power. If, for example, the executive had been earning the full amount of the surplus,

then she will bear the full brunt of the loss in tax savings. With less bargaining power, this

loss will be shared and compensation will not fall by as much, a perhaps counterintuitive

result.

Whether non-option compensation increases or decreases in response to this change

is a more subtle question, depending on the shape of the surplus function and the sub-

18Since only a minority of firms used TSARs and so were affected by the reform, it seems reasonableto assume that the reform did not meaningfully impact the outside option of the TSAR executives, sincetheir next best option was likely a non-TSAR firm.


stitutability of different forms of compensation. Intuitively, as long as there is some

substitutability in the executive’s utility function and the total surplus does not change

too quickly as options decrease, other compensation will increase to partly make up

for the decline in option compensation, with the option share of compensation unam-

biguously declining. The model in Appendix 2.A provides some clarification on these

conditions.

To summarize, the theoretical framework suggests that the TSAR reform will cause

option compensation and the option share of compensation to fall. The extent of substi-

tution towards non-option compensation and the magnitude of the fall in total compen-

sation depend crucially on the executive’s bargaining power.

2.5 Empirical Analysis

The basic empirical strategy used to test these predictions about how the tax change

affected compensation involves comparing the compensation of executives at TSAR and

non-TSAR firms in a difference-in-differences framework. For this to be a valid approach,

it is important to understand why some firms chose to use TSARs and some did not,

since this distinction is the basis of the identification strategy.

2.5.1 Which firms use TSARs?

Table 2.3 shows how firm performance and size vary by TSAR status, among firms that

used any options. Note that these statistics are implicitly weighted by the number of

executive-years from each firm which satisfy the sample selection criteria. Firms which

granted TSARs in the pre-reform years are larger and have exhibited better performance

as measured by shareholder return and return on assets. Executive turnover is a bit

lower at TSAR firms, while the incidence of promotions for existing employees is slightly

higher. While total compensation is about twice as high at TSAR firms, it is reassuring

that the option share of this total is similar in each group.

Table 2.4 is a list of TSAR firms in the estimation sample along with the industry

in which they operate. There does not appear to be any obvious industry clustering,

with representation from financial firms, manufacturing and natural resources of various

kinds.

There are a number of possible factors explaining variation in TSAR takeup: dif-

ferences in cashflow by option type, incomplete information about the tax treatment of

TSARs on the part of firms, differences in compensation transparency and differences


in accounting treatment. The obvious benefit of TSAR adoption is the newly available

corporate tax deduction, so the question is which of these possible countervailing forces

was limiting their use to only a minority of firms.

First of all, there would appear to be a cashflow disadvantage to TSARs since these

require the firm to make a cash payment to the executive. With a stand-alone option,

the cash payment actually goes in the opposite direction, as the executive must pay the

exercise price in cash to her employer. However, this apparent difference is misleading

since the firm, faced with the exercise of a tandem stock appreciation right, can always

sell a share to fund the cash payment to the executive at quite low cost.19 In fact, this

strategy provides a more appropriate comparison between the two choices, since it holds

the level of shares outstanding fixed, and dilution is an issue of central interest to existing

shareholders.

A second possibility is a lack of information about the tax benefits of TSARs. How-

ever, this too is unlikely, as the Canada Revenue Agency issued an interpretation bulletin

in 1996 clarifying the requirements for unlocking the beneficial tax treatment. Further-

more, very high profile companies such as the Bank of Nova Scotia, one of the largest

banks in Canada, and the oil refiner Suncor use tandem options and explicitly discuss

them in their financial statements, which makes it quite implausible that the rest of the

companies in the sample did not have sufficient information to take advantage of TSARs

if they so desired.20

Public disclosure requirements have been shown, for example by Murphy [1996], to

be an important determinant of executive compensation. This mechanism would be

relevant if one type of option compensation was more transparent to shareholders than

the other, as one could imagine cases where the board of directors and the executive

herself might prefer to obfuscate the true level of pay.21 However, the required disclosure

for stock option grant value in the Summary Compensation Table of the Management

Information Circular does not differ with the addition of a tandem stock appreciation

right - the valuation method does not adjust for any additional benefit to the executive

from the cash settlement feature.22

19As noted in Mawani [2003], in this case firms can sell shares to an investment dealer through aprivate placement, avoiding incremental disclosure requirements.

20Note that the auditors of TSAR firms are distributed roughly equally across the Big 4 audit firms,so any differential information about TSARs does not appear to have been coming from this source.

21Aboody et al. [2004] find that disclosed compensation was treated by investors as a value-relevantexpense of the firm even before recognition of this compensation was required. Rather, it was includedonly in a footnote to the financial statements.

22At least in a frictionless world, there is no reason to make any adjustment; if the executive preferscash, she can exercise the option and sell the share immediately, which is in fact quite common. Anyactual differences due to transaction costs associated with this strategy are unlikely to be large.


The remaining, and, it would appear, most important cost of using TSARs is the

difference in accounting treatment. Since 2003, the fair value method has been required

to account for option compensation which means that the expected value of an employee

stock option is measured at the grant date and then expensed against accounting income

in equal parts over the vesting period. For example, if an option vests over three years,

a third of the value of the option grant is subtracted from income in each year. On

the other hand, TSARs are accounted for in the same way as regular stock appreciation

rights, which requires use of the intrinsic value method. Rather than expensing the fair

value, at the end of each year, the company must take a compensation expense equal

to the current value of the TSAR as measured by the difference between the year-end

market price and the option’s exercise price – the ‘mark-to-market’ liability associated

with the option.23 Accordingly, a firm using TSARs will find that compensation expense

increases as its share price appreciates, a potentially important drag on earnings growth.

Of course, there is no associated fall in compensation expense when share prices decline

if the option is out of the money, so this treatment does not necessarily help to prop up

earnings when the firm performs poorly.24

Some compelling direct evidence that firms perceive this accounting difference in a

negative light comes from a letter written by the accounting firm PricewaterhouseCoopers

to the Minister of Finance in response to the option provisions of the 2010 Budget

([PricewaterhouseCoopers, 2010]). In arguing against the retroactive nature of the TSAR

aspect of the reform25 the letter describes how many firms have set up hedges against the

volatility in expenses imposed by this type of option. This revealed preference shows that

firms were willing to undertake costs to mitigate the adverse accounting consequences of

TSARs.

Additional evidence on the benefit side of the ledger – that it was actually the tax

benefits that were driving take-up of the tandem provision in the first place, comes from

the fact that many companies eliminated this provision in the months following the 2010

Budget. For example, CI Financial eliminated cash settlement of options on July 1, 2010,

specifically citing the tax reform. Imperial Metals made a similar change on May 19, 2010,

23Under IFRS 2, which was adopted in Canada starting in 2011, the compensation expense is therecalculated fair market value of the options at year end.

24Babenko and Tserlukevich [2009] discuss the tax benefits of NQOs, which give rise to a corporatededuction in the U.S. and argue that the positive correlation between stock option exercise and earningshas the beneficial side effect for firms of pushing tax deductions to high profitability, and so high taxrate, years.

25The argument was ultimately unsuccessful as the Budget provisions were enacted substantively asoriginally written. The reform was retroactive in the sense of applying to the exercise of even optionsthat had been granted under the original tax treatment.


while Suncor did so as of August 1, 2010. This suggests that the decision to use TSARs in

the pre-reform period involved trading off the tax benefit from the additional corporate

deduction against the accounting cost of increased earnings volatility. Furthermore, this

story is consistent with the fact that TSAR firms are larger and more successful, since

such companies face higher expected tax rates, all else equal. These firms would then

place a higher value on the corporate tax deduction associated with TSARs.

2.5.2 Results

The starting point of the empirical analysis is to investigate graphically the evolution of

option compensation around the reform for firms using TSARs relative to those using

regular options. Figure 2.1 illustrates this comparison, with average option compensation

per executive indexed to 100 for both groups in 2008. This allows an easier comparison

of compensation growth rates, since TSAR firms tend to grant more options on average.

The change in options from 2008 to 2009 appears to be similar across the two groups

and then diverges into 2010, with a relative decrease for TSAR firms in 2010 and 2011.

In total, option compensation falls by 6% for TSAR firms from 2008-2011, while it rises

by 23% at control firms. This is the expected direction of the effect, since the 2010

reform relatively increased the tax rate on option grants for TSAR firms. Interestingly,

the relative drop for these firms is 29%, which corresponds closely to the value of the lost

corporate deduction at the statutory corporate tax rate.

Figure 2.2 shows a similar scatter plot for non-option compensation covering the same

sample of executives. The issue here is whether or not some of the apparent decrease

in compensation seen in Figure 2.1 is actually compensation being shifted away from

the now more highly taxed options, rather than an actual loss in income for executives.

There appears to be some divergence around the reform, with the change in compensation

being somewhat larger for TSAR firms. Specifically, non-option compensation increases

by 45% at TSAR firms, compared with only 27% at control firms. However, this graphical

evidence is suggestive, rather than conclusive, since it could reflect other differences across

these two groups.

As the decision to adopt TSARs appears to have been quite permanent, absent a large

shock such as the 2010 reform, it makes sense to think about the tradeoffs inherent in the

TSAR adoption decision in terms of fixed or long-run firm characteristics. To address

these differences, in the regression models that follow, fixed effects are included at either

the firm or individual level. Even after controlling for fixed firm characteristics in this

way, one might be worried about time-varying firm characteristics – principally, measures


of firm performance that would quite plausibly affect various elements of compensation.

Following Hall and Liebman [2000], I include both the firm’s lagged yearly stock market

return and return on assets as measures of firm performance.26 It is also important

to control for firm size, as option use certainly increases with firm size,27 and firms in

the sample may be growing differentially over time. Specifically, the regression models

include quartile dummies for the latter three variables as a flexible way of allowing for

non-linearities in their effects.28

The second set of important control variables is a set of dummies for the role in the

company held by the executive, including chief executive officer, president, chief financial

officer and chief operating officer. Likewise, dummy variables are included for whether

the executive is either a member or the chairman of the board of directors. To the

extent that board capture is important, one might expect compensation to be higher for

executives with a seat at board meetings, though such a correlation might also reflect

the fact that only the best CEOs are also chosen as chairmen of the board. In addition,

a dummy variable is included which takes a value of one if the executive’s position at the

company, as described in the management information circular, has changed from the

previous year.29

Hence the empirical models in what follows are regressions of different measures of

executive compensation on firm (or executive) fixed effects, firm size and performance

quartile dummies, a set of executive position dummies and, most importantly, the in-

teraction of a post-reform dummy variable covering the 2010 and 2011 fiscal years with

a firm-level dummy variable indicating TSAR grants (along with level effects of these

dummies). Specifically, the following regression is estimated for each measure of com-

pensation, with i denoting an executive and t the year, and with the fixed effect ωi at

either the firm or individual level.

Compensationit = βXit+γ1TSARi X PostReformt+γ2TSARi+γ3PostReformt+ωi+ϵit

(2.1)

The coefficient on this interaction term represents the estimated causal effect of the

reform. This implements a difference-in-differences strategy whereby executives at TSAR

26This is particularly important given the financial crisis taking place near the start of the sample,which could have widened differences in firm performance.

27The raw correlation between the value of option grants and the firm’s market capitalization is 0.48.28Using polynomials of these three variables yields substantially similar conclusions.29Given the sample requirement that the executive be employed by the firm in both years, this typically

involves a promotion, which would be expected to result in increased compensation. However, this isrelatively rare in my data, occurring for only 3% of executive-years.


firms are the treatment group, and those at firms which grant regular options form the

control group. From the theoretical framework, this coefficient is expected to be negative

for different measures of option compensation and total compensation, and positive for

other forms of compensation if there is any substitution response. Standard errors are

always clustered at the firm level to account for the fact that compensation changes for

executives employed at the same firm are likely to be positively correlated.30

To start, we can see from the first column of Table 2.5 that the fraction of compen-

sation made up of option grants falls for TSAR firms relative to control firms around the

reform. Specifically, this fraction falls by a statistically significant 5.7 percentage points

from a base of 33%. Evidently, firms are discouraged from compensating employees with

stock options following this reform – strong evidence that compensation responds to

changes in its taxation.

The second column of Table 2.5 shows that option compensation fell by about 28%

in response to the reform, which is statistically significant at the 5% level. Despite the

fixed effects, executive controls and time-varying firm controls, this lines up rather closely

with the graphical evidence presented in Figure 1. The magnitude is consistent with firms

being able to maintain the same after-firm-tax stock option expense, ‘passing on’ the lost

deduction to employees in the form of lower stock option grants. Since the fraction of

compensation through options for executives receiving any option grants is about a third,

this decline in option compensation corresponds on average to a 9% drop in total pay for

affected executives, notwithstanding any substitution to other forms of compensation.

The control variables typically enter with the expected signs. In particular, larger

firms that yielded better shareholder returns tend to see higher growth in option pay both

in absolute terms and as a fraction of total pay. The chief executive officer and members

of the board of directors also tend to receive more option grants. The coefficient on the

post-reform dummy implies that though option compensation for the control group was

growing over time, it was doing so at a slower rate than the rest of compensation, as

the fraction of compensation from options actually fell over time. This could reflect the

impact of the other features of the 2010 Budget related to the taxation of options, which

should also have tended to decrease option use.

The third and fourth columns of Table 2.5 replace the firm fixed effects with 786 indi-

vidual executive fixed effects. The results are quite similar, notwithstanding an increase

in standard errors that is to be expected given the large increase in the number of fixed

30This is borne out in the data, as the clustered standard errors are much larger than regular robuststandard errors. Likewise, clustering instead at the executive rather than the firm level results in largedeclines in standard errors, as more variation is used.


effects relative to the sample size. The firm size and performance measures do not much

change, whereas the executive role dummy variables mostly lose their significance, since

their effects are now identified only from executives who change roles, and all position

changes in aggregate only add up to 3% of executive-years.

A concern in interpreting these results as causal is that we know from Table 2.3 that

TSAR firms are larger than non-TSAR firms. Hence, if for some non-tax reason, option

compensation is falling at large relative to small firms over this period, the estimated

treatment effect could be biased downwards. To address this potential bias, I construct

a dummy variable that is equal to unity if the executive’s firm was above the median

size31 in 2008. Then this dummy variable is interacted with the post-reform dummy

and included in the regression to allow for large firms to experience differential changes

in option compensation independent of any effects of the TSAR reform. This addition

leaves the estimate of the change in option compensation share almost identical at -5.3

percentage points with a similar standard error. The reform’s effect on option compen-

sation declines somewhat from 28% to 23% with the inclusion of this new variable, but

is still close to significant at the 5% level. The new interaction term itself is insignificant

in the fraction regression but negative and significant in the option regression, as would

be expected given the decline in the absolute value of the effect.

An implicit assumption in these models is that the firm is actually targeting the Black-

Scholes value of options when making its compensation choices, which seems a reasonable

starting point. If, however, the target is actually the number of options or some com-

posite measure of the executive’s ownership stake, then using the value, rather than the

unobserved target, could be introducing measurement error. A particular concern stems

from the fact that the Black-Scholes value of an option is increasing in the firm’s return

volatility. Hence, if treatment firms happened to experience decreased volatility relative

to control firms around the reform, their option grants would mechanically fall in value.

However, Figure 2.3, which shows how the volatility of monthly returns changed over

time for the two groups of firms, demonstrates a reassuring similarity in volatility so that

this issue does not appear to be driving the results.

The next step is to investigate the substitution response observed in Figure 2.2 more

rigorously. The first two columns of Table 2.6 present the results of regressions of the log-

arithm of non-option compensation and total compensation, respectively, for executives

also receiving positive option grants. This restriction helps to ensure that the control

group against which the treatment effect is measured is composed of executives receiving

31This is measured by total assets, rather than market capitalization, since the former is less volatile.Similar results obtain using the latter.


their compensation from similar sources with a similar risk profile. The coefficient on

the interaction term in the first column means that non-option pay increased by 3.3% for

treatment firms relative to control firms around the reform. However, with a standard

error of 5.8%, this increase is not close to statistical significance. Implicitly combining

this small increase with the significant decline in option compensation seen in Table 2.5

drives a decline of 5.3% in total compensation due to the reform, though the standard

error is again too large for a definitive conclusion. Including the large firm dummy and

allowing it to vary around the reform does not much change these estimates, except for

that of column (4) which increases in magnitude to a 9.6% drop in total compensation

for TSAR firms, and becomes significant at the 10% level. This is consistent with execu-

tives exercising enough bargaining power prior to the reform to capture most of the tax

savings from the corporate deduction. They then lose this part of the surplus after the

deduction is eliminated.

The control variables in these regressions have similar effects as in Table 2.5, except

that performance-sensitivity of non-option pay, at least as measured by its sensitivity to

shareholder returns, basically disappears. Evidently, any positive performance-sensitivity

of ex ante pay32 for Canadian executives is coming from changes in the value of option

grants. The positive and significant post-reform dummy implies, as suggested by Figure

2.2, that both non-option and total pay grew over the sample period, even conditional

on firm performance and growth.

The third and fourth columns add executives who did not receive any options to

the estimation sample. This increases the information available to estimate the path

of non-option and total compensation for the control group. The tradeoff is that these

executives might be more likely to differ from the treatment executives in unobservable

ways, which might increase bias. However, the estimates from this larger sample are very

similar: non-option compensation rises by several percent but the change is statistically

indistinguishable from zero. Likewise, total compensation falls, but the standard error

remains too large to conclude that this decline is statistically significant.

Table 2.7 splits the results from the previous two tables by year, rather than just pre-

and post-reform. This provides more insight into differences between the treatment and

control groups in the years prior to the reform and also shows the pattern of adjustment

following it. It seems reasonable to expect that any response to the reform should be

seen more strongly in 2011 than in 2010, given that firms would have had more time to

adjust their compensation decisions to reflect the new tax environment. In fact, this is

32However, ex post, the value of share-based awards and option holdings certainly rises with share-holder stock returns.


just what we see in the table. Both options and the option share of compensation start

to decline from 2009 to 2010 and continue to do so into 2011. These results drive the fact

that total compensation falls by more in 2011 than in 2010; this decline relative to 2008

is not statistically significant for the option-only sample but is significant at the 5% level

in the broader sample used in the fifth column. This effect is reinforced by the fact that

substitution towards non-option compensation actually declines in 2011, though again

by much less than the standard error.

Looking at the compensation difference between TSAR and non-TSAR executives

in 2009 is also informative. If the non-TSAR executives are to form an appropriate

control group, we would like to see similar changes in compensation in the two groups

of executives from 2008 to 2009. This is because the identifying assumption is that

compensation changes across these groups only because of the reform, which cannot

have affected 2009 compensation. It is reassuring that the growth in options and the

change in the option share are quite similar from 2008 to 2009. Though the difference is

not significant, the nontrivial decline in non-option compensation for the control group

casts some doubt on a causal interpretation of the year-by-year changes in this variable.

Note that the post-reform dummy in Table 2.6 captures the average difference, which

includes this pre-treatment decline, and this is why it shows a small increase in non-

option compensation after the reform. A similar caveat applies in interpreting the 2011

coefficient in the full sample of column (5) given the large, though insignificant, difference

in the groups from 2008 to 2009.

2.6 Conclusion

In this chapter, I construct a novel dataset of executive compensation in Canada to take

advantage of a tax reform which generated cross-sectional and time series variation in tax

incentives for firm stock option grants. Using a difference-in-differences strategy, I find

that option compensation fell substantially at affected firms after the reform. There is lit-

tle evidence that this fall in pay is mitigated by increases in other forms of compensation.

Therefore, it would appear that executives bear a substantial portion of the increased

tax burden. This has important consequences for optimal tax policy in the sense that

executive pay is indeed sensitive to policy intervention. TSAR firms and their executives

appear to have been using a suboptimally high level of option compensation because of

the potential tax savings. The observed change in compensation is also consistent with

the idea that executives have significant bargaining power, as it suggests that executives

had been capturing a large part of the surplus associated with their employment prior


to the reform.

These results and dataset also suggest opportunities for future research in this area,

for example, by using the jurisdiction of residence of the executives in combination with

other subnational and cross-country tax reforms to learn more about tax incidence and

the taxable income elasticity of high income taxpayers. Furthermore, the fact that option

compensation fell for some firms following the 2010 reform provides a useful instrumental

variable with which to assess the effects of executive ownership and performance-based

pay on risk-taking33 and agency conflicts. In particular, it may be the case that the

reduction in option grants identified in this chapter is associated with reduced risk-taking

behaviour by top executives or governance issues associated with reduced alignment

between the incentives of managers and shareholders.

33This question is addressed in a long literature, such as Kim and Lu [2011], which grapples withthe endogeneity of executive ownership and potentially opposing effects from incentive alignment andmanagerial entrenchment.


2.A Model of Executive Compensation

In this model, there is a firm and an executive bargaining over the executive’s compensa-

tion, which is split into two types, x1 and x2. These can be thought of loosely as salary

and TSAR options (expressed in present value terms), respectively. The differences be-

tween them are twofold: options earn an incremental corporate deduction, with value

τx2, but cause non-tax costs given by g(x2), which is increasing and convex.34 These

costs could come from dilution effects, increased risk or differential accounting costs, for

example.35 Then the firm’s profit is given by:

πF = [Y − x1 − x2] + τx2 − g(x2) (2.2)

where Y is the real surplus created by the firm-executive match, which is reduced by the

compensation paid to the executive. The fundamental tradeoff in the model is embodied

in the last two terms – the tax advantage of options against their non-tax net costs.

The executive’s utility is linear and she cares equally about both forms of compensa-

tion. Then her payoff is just:

πE = x1 + x2 (2.3)

It is slightly simpler to model the relative non-tax costs of x2 as falling on the firm,

although the conclusions would be similar if the employee bore some of these costs.

Compensation is determined using Nash bargaining, with the executive having a

bargaining weight of β ∈ [0, 1]. This means that compensation is chosen to maximize the

net surplus from the match, and then a fraction β accrues to the executive in the form

of x1 and x2. Net surplus is calculated by summing the payoffs of the two sides to get:

Π = Y + τx2 − g(x2) (2.4)

since the compensation paid is just a transfer from one side to the other. It is trivial

to show that the surplus is maximized when g′(x2) = t, given an interior solution with

both forms of compensation positive, so that the net non-tax marginal cost is just equal

to the marginal benefit of paying another dollar of options, the tax rate τ . Given the

34The assumption that taxes affect only option compensation is made simply to minimize notation.Both forms of compensation could face different tax rates at both the firm and executive level withoutchanging any conclusions because the comparative static of interest for thinking about TSARs involvesremoving or reducing the corporate deduction, leaving all other tax rates fixed.

35In fact, g(x2) can be negative for low x2, perhaps because of positive incentive effects, but the costturns positive once option compensation is large enough, for the reasons discussed above. Given that itis very rare for executives to receive only one form of compensation, this assumption seems reasonable.


assumptions on g(.), this means that the optimal option compensation, x∗2, is increasing

in the tax rate. The reform studied in this chapter involved a reduction in the value of

this tax deduction (to zero), so the predicted effect is a decline in the value of option

compensation.

The tax deduction causes option compensation to be inefficiently high. If the two

bargainers took into account the impact of their choices on government revenues, surplus

would be just Y − g(x2) so that the optimal choice x∗∗2 would satisfy g′(x∗∗2 ) = 0 and so

the socially optimal choice of options is lower than the privately optimal one. Essentially,

the firm and executive agree to take on higher non-tax costs as a form of tax arbitrage.

The next question is what happens to total compensation of the executive, which

is equal to βΠ. Hence, the relevant comparative static is β∂Π(x∗2)

∂τ. Using the envelope

theorem, this is just equal to βx∗2 > 0. Again, since the reform involves a reduction in

τ , this shows that total compensation must fall, and does so more strongly for higher

executive bargaining power. This is because the decline in surplus hurts the executive

more, the more of that surplus she had been capturing.

Since both total compensation and option compensation decline, the associated change

in x∗1 and the option fraction of compensation just depend on whether options decline

faster than total compensation or vice versa. Specifically, it can easily be shown that

∂x∗1∂τ

< 0 ⇐⇒ ∂Π(x∗2)

∂τ<∂x∗2∂τ

1

β(2.5)

and

∂F ∗

∂τ< 0 ⇐⇒ ∂Π(x∗2)

∂τ<∂x∗2∂τ

Π(x∗2)

x∗2(2.6)

where F ∗ ≡ x∗2x∗1+x

∗2is the option fraction of total compensation.

Since βΠ > x2 at an interior solution, the second condition is less strict than the first.

This means that there are some circumstances where the option fraction increases with

τ at the same time as x1 is increasing (just at a slower rate than x2). Whether these

conditions are satisfied depends on the nature of the g(.) function, and in a more general

context, on the substitutability between different forms of compensation. As long as the

total surplus function is not too sensitive to x2,36 then a decrease in τ , as with the TSAR

36If g(x2) =xγ2

γ , the isoelastic case, then it can be shown that these conditions hold as long as

γ < 1−τβτ + 1, which holds for any γ > 1 for sufficiently small τ or β. For the quadratic case, this

condition holds as long as τ < 0.5 regardless of the value of β.


reform, will result in an increase in other forms of compensation and a decrease in the

fraction of compensation paid in options. The increase in other compensation will be

lower, the higher the executive’s bargaining power. Intuitively, this is again because the

executive is hurt more by the decline in surplus from the lost tax savings when she had

been capturing a larger share.

A useful extension would involve more complicated non-tax cost functions which de-

pend on both forms of compensation, though the fundamental tradeoff of tax savings

versus real costs would still be present. Feedback from compensation choices to bargain-

ing power, perhaps because paying compensation that has the effect of increasing the

executive’s ownership stake increases her bargaining power, would also be interesting.


9010

011

012

013

0O

ptio

n C

ompe

nsat

ion

(200

8 =

100

)

2008 2009 2010 2011Year

Options (TSAR) Options (non−TSAR)

Mean Option Compensation by Option Type

Figure 2.1: This figure shows the evolution of average option compensation at TSARand non-TSAR firms over time, for executives in the estimation sample. Option com-pensation is set to 100 for each group in 2008 to facilitate the comparison, where theunderlying 2008 values are $641K per executive for TSAR firms and $351K per executivefor non-TSAR firms. The tax reform affects the TSAR group starting in 2010.


100

110

120

130

140

150

Non

−op

tion

Com

pens

atio

n (2

008

= 1

00)

2008 2009 2010 2011Year

Compensation (TSAR) Compensation (non−TSAR)

Mean Non−Option Compensation by Option Type

Figure 2.2: This figure shows the evolution of average non-option compensation atTSAR and non-TSAR firms over time, for executives in the estimation sample. Com-pensation is set to 100 for each group in 2008 to facilitate the comparison, where theunderlying 2008 values are $1,068K per executive for TSAR firms and $684K per ex-ecutive for non-TSAR firms. The tax reform affects the TSAR group starting in 2010.


810

1214

16M

onth

ly V

olat

ility

(%

)

2008 2009 2010 2011Year

Volatility (TSAR) Options (non−TSAR)

Monthly Return Volatility by Option Type

Figure 2.3: This figure shows the monthly stock return volatility by option type (TSARand non-TSAR) over time, for companies in the estimation sample. TSAR firms aresomewhat less volatile but volatility has changed in a similar way for both groups.


Table 2.1: This table shows the tax treatment at the corporate and personal level fora $100 gain from stock options at the point of exercise. A 30% corporate tax rate and46% top personal marginal tax rate are assumed (as in Ontario). The effective personaltax rate on stock options is then 23% due to the stock option deduction when it isavailable. (1) is for a regular stock option, (2) is for an option cash-out/cancellation orstock appreciation right and (3) is for a tandem stock appreciation right. Tax treatmentof a typical option in the US would be similar to (2) if tax rates were identical. Thechoice between (1) and (2) turns on the relevant tax rates in a specific situation and thechoice to cash-out the option can be negotiated at the time of exercise. From a purely taxperspective, (3) is strictly preferred for all positive tax rates. The 2010 reform changedthe default tax treatment of TSARs from (3) to (1).

(1) Stock Option (2) Cash-out/SAR (3) TSAR

Stock Option Gain 100 100 100

Personal Tax 23 46 23

Corporate Tax 0 -30 -30

Total Tax 23 16 -7

After-tax Option Value 77 84 107


Table 2.2: All dollar amounts are measured in thousands, shown conditional on apositive value, and winsorized at 5%. The TSAR dummy takes on a positive value forsome executives in the non-option sample because though they did not receive any optionsin that year, their firm grants TSARs when it grants options.

Option Firms Non-Option FirmsMean SD Mean SD

Total Compensation 1,392 (1,138) 867 (685)Salary 368 (183) 330 (164)Bonus 381 (371) 350 (365)Options 427 (433) 0 -Share-based 441 (412) 452 (396)Pension - amount 117 (96) 62 (76)Other - amount 42 (40) 32 (34)Salary - any .993 .987Bonus - any .854 .852Options - any 1 0Share-based - any .453 .433Pension - any .403 .453Other - any .640 .615CEO dummy .238 .242CFO dummy .219 .193Position change dummy .035 .034Director dummy .306 .329TSAR dummy .124 .042N 2,428 1,774


Table 2.3: This table shows firm characteristics by use of tandem stock appreciationrights, conditional on using options of some kind. Total Assets, market capitalization,total compensation and option compensation are measured in millions of dollars. Thevalues are weighted by the number of executive-years satisfying the sample restrictions foreach firm. The position change dummy is one if the executive’s position at the companychanged, conditional on still being in the sample. The executive turnover measure is thefraction of executive-years that otherwise satisfy the sample restrictions except that theexecutive does not show up in all four years of the sample.

TSAR (treatment) Non-TSAR (control)Median SD Median SD

Total Assets 6,498 (11,549) 1,179 (8,884)Market capitalization 3,630 (8,180) 975 (5,744)Lagged shareholder return .120 (.801) .088 (.890)Lagged return on assets .059 (.156) .020 (.070)Position change dummy .050 .032Executive turnover .357 .394Total compensation 1,946 (2,850) 918 (2,139)Option compensation 550 (1,366) 225 (912)N 302 2,126


Table 2.4: List of companies using tandem stock appreciation rights and their industries.A * denotes members of the S&P/TSX 60 Index.

Agrium Inc* Agricultural ChemicalsBank of Nova Scotia* Commercial BanksBMTC Group Inc Retail-Furniture StoresCanadian Natural Resources Limited* Crude Petroleum & Natural GasCE Franklin Ltd Wholesale-Industrial Machinery & EquipmentCenterra Gold Inc Gold and Silver OresCI Financial Corp Investment AdviceEncana Corp* Crude Petroleum & Natural GasEnsign Energy Services Inc Drilling Oil & Gas WellsHigh Liner Foods Inc Prepared Fish & SeafoodHusky Energy Inc* Petroleum RefiningImperial Metals Corp Metal MiningLinamar Corp Motor Vehicle Parts & AccessoriesLoblaw Companies Ltd* Retail-Grocery StoresNeo Material Technologies Miscellaneous Primary Metal ProductsNewAlta Corp Sanitary ServicesNexen Inc* Crude Petroleum & Natural GasParamount Resources Crude Petroleum & Natural GasPason Systems Inc Oil and Gas Field ServicesProgressive Waste Solutions Refuse SystemsQuebecor Inc Cable & Other Pay Television ServicesRogers Communications* Radiotelephone CommunicationsSherritt International Corp Metal MiningSuncor Energy Inc* Petroleum RefiningTalisman Energy* Crude Petroleum & Natural GasTVA Group Television Broadcasting StationsWest Fraser Timber Sawmills & Planing Mills


Table 2.5: The first two columns include 272 firm fixed effects, while the latter two use786 executive fixed effects. In each case, the TSAR dummy variable gets dropped sinceit does not vary over time for a particular firm or executive. The dependent variable inthe first and third columns is option compensation divided by total compensation, whilefor the second and fourth columns it is the logarithm of options. Standard errors areclustered at the firm level. *, **, *** denote significance at 10%, 5% and 1%, respectively.

Firm Fixed Effects Executive Fixed EffectsOptions/Total Log(Options) Options/Total Log(Options)

PostReform*TSAR -0.057** -0.281** -0.056** -0.278*(0.023) (0.135) (0.026) (0.147)

PostReform -0.033*** 0.049 -0.033** 0.074(0.013) (0.070) (0.015) (0.081)

Yearly Return: Quartile 2 0.037* 0.194** 0.036 0.184(0.020) (0.098) (0.023) (0.112)

Yearly Return: Quartile 3 0.052*** 0.257** 0.051*** 0.256**(0.017) (0.100) (0.020) (0.114)

Yearly Return: Quartile 4 0.057*** 0.347*** 0.056*** 0.329***(0.016) (0.085) (0.018) (0.096)

Return on Assets: Quartile 2 0.043** 0.238*** 0.041** 0.230**(0.017) (0.080) (0.019) (0.092)

Return on Assets: Quartile 3 0.016 0.170* 0.015 0.171(0.018) (0.092) (0.020) (0.104)

Return on Assets: Quartile 4 0.000 0.004 -0.005 -0.017(0.020) (0.098) (0.023) (0.113)

Market Capitalization: Quartile 2 0.068*** 0.493*** 0.073*** 0.530***(0.021) (0.139) (0.024) (0.153)



Position change dummy 0.016 0.114 0.020 0.192*(0.016) (0.092) (0.020) (0.108)

CEO dummy 0.013 0.503*** 0.036 0.346(0.016) (0.096) (0.042) (0.236)

President dummy 0.001 0.126* -0.024 -0.098(0.011) (0.071) (0.034) (0.171)

CFO dummy -0.002 0.065 -0.009 0.079(0.006) (0.045) (0.028) (0.174)

COO dummy -0.002 0.298*** -0.019 0.092(0.009) (0.058) (0.026) (0.174)

Director dummy 0.027** 0.378*** -0.019 0.039(0.013) (0.078) (0.053) (0.236)

Board Chair dummy 0.011 0.212 -0.026 -0.086(0.025) (0.136) (0.070) (0.390)

Adj. R-squared 0.646 0.712 0.618 0.741N 2,428 2,428 2,428 2,428


Table 2.6: All regressions include firm fixed effects, rising from 272 in the first twocolumns to 356 when firms not using options are added back to the sample in the third andfourth columns. The dependent variable in the first and third columns is the logarithmof non-option compensation, while for the second and fourth columns it is the logarithmof total compensation. Standard errors are clustered at the firm level. *, **, *** denotesignificance at 10%, 5% and 1%, respectively.

Option-only Sample Full SampleLog(NotOptions) Log(Total) Log(NotOptions) Log(Total)

PostReform*TSAR 0.033 -0.053 0.039 -0.063(0.058) (0.056) (0.058) (0.049)

PostReform 0.196*** 0.137*** 0.181*** 0.149***(0.027) (0.029) (0.023) (0.024)

Yearly Return: Quartile 2 -0.006 0.064 0.007 0.084**(0.033) (0.042) (0.033) (0.037)

Yearly Return: Quartile 3 -0.014 0.075* 0.021 0.110***(0.035) (0.040) (0.032) (0.036)

Yearly Return: Quartile 4 0.031 0.126*** 0.042 0.120***(0.037) (0.041) (0.033) (0.038)

Return on Assets: Quartile 2 -0.020 0.060 0.033 0.078(0.031) (0.037) (0.050) (0.053)

Return on Assets: Quartile 3 0.018 0.042 0.029 0.045(0.042) (0.041) (0.047) (0.049)

Return on Assets: Quartile 4 0.016 0.020 0.047 0.047(0.042) (0.040) (0.045) (0.047)

Market Capitalization: Quartile 2 0.085* 0.210*** 0.109*** 0.178***(0.049) (0.056) (0.036) (0.044)



Position change dummy 0.030 0.048 0.040 0.034(0.049) (0.049) (0.040) (0.046)

CEO dummy 0.417*** 0.406*** 0.282*** 0.271***(0.067) (0.060) (0.055) (0.055)

President dummy 0.115** 0.123*** 0.164*** 0.188***(0.051) (0.046) (0.046) (0.045)

CFO dummy 0.081*** 0.078*** 0.062** 0.062**(0.031) (0.030) (0.027) (0.027)

COO dummy 0.283*** 0.263*** 0.243*** 0.241***(0.040) (0.039) (0.035) (0.036)

Director dummy 0.204*** 0.261*** 0.307*** 0.333***(0.061) (0.054) (0.047) (0.044)

Board Chair dummy 0.150* 0.180** -0.058 -0.085(0.086) (0.078) (0.093) (0.095)

Adj. R-squared 0.843 0.824 0.746 0.729N 2,428 2,428 4,201 4,201


Table 2.7: All models include firm fixed effects, quartile dummies for market capi-talization, lagged shareholder return and lagged return on assets, and a set of executiveposition dummies, with standard errors clustered at the firm level. The first four columnsrequire positive option grants, while the final column relaxes this restriction. *, **, ***denote significance at 10%, 5% and 1%, respectively.

Options/Total Log(Options) Log(NotOptions) Log(Total) Log(Total)TSAR*2009 0.016 0.034 -0.059 -0.013 -0.084

(0.027) (0.148) (0.080) (0.074) (0.085)TSAR*2010 -0.039 -0.221 0.011 -0.033 -0.048

(0.028) (0.149) (0.087) (0.071) (0.064)TSAR*2011 -0.057** -0.307** -0.013 -0.092 -0.167**

(0.025) (0.140) (0.071) (0.065) (0.084)Adj. R-squared 0.648 0.713 0.845 0.826 0.731N 2,428 2,428 2,428 2,428 4,201

Chapter 3

Dividends and Shareholder

Taxation: Evidence from Canada

3.1 Introduction

The effect of shareholder taxation of dividends on corporate dividend payout is an open

question with important ramifications for efficiency and equity in tax policy. Since div-

idends are earned disproportionately by the wealthy, dividend taxes may have desirable

distributional implications. However, these may be accompanied by significant ineffi-

ciencies. The traditional source of inefficiency comes from the fact that income earned

by a corporation is already taxed at the corporate level, and so taxing it again at the

shareholder level implies double taxation of this income. Because of this double taxation,

a rational investor will require a higher rate of return from investing in the equity of a

corporation, which leads to a higher cost of capital for the firm, and so less investment.

The agency model of the firm identifies another possible issue associated with dividend

taxation. If such taxes discourage payout, free cash flow may be retained in firms where

it can cause, for example, wasteful empire building by managers (Jensen [1986]). Finally,

differential taxes at the investor level can have implications for optimal risk diversifica-

tion. Rather than allocating ownership according to the ability and desire to bear risk,

these tax differences may be driving ownership decisions.

I use a recent cut in shareholder dividend taxes in Canada to provide evidence on

the effect of dividend taxes on dividends, which in turn has important implications for

the effect of dividends on the cost of capital and investment. In particular, theoretical

models of the firm predict that investment will move in the same direction in the long

run as dividends. In line with this idea, the avowed aim of the tax cut was to “encourage

74

Chapter 3. Dividends and Shareholder Taxation: Evidence from Canada 75

savings, investment and economic growth (Budget 2006).” I find only limited evidence

for any successes of the policy along these margins. Decomposing the data to allow for

separate intensive and extensive effects of the policy yields treatment estimates that are

not economically significant relative to the existing flow of dividends or stock of dividend

payers. However, this lack of effect could be driven by confounding secular trends and so

suggest a difference-in-differences strategy to account for such time-varying factors. Two

control groups are used, both of which would be a priori unlikely to respond to the reform.

First, large shareholders who were not eligible for the reform due to being resident outside

of the country or because of the inter-corporate dividends received deduction might be

expected to influence the firms in which they invest against responding to the reform.

Second, firms inter-listed in the United States are likely catering to investors in that

country, who likewise did not enjoy the benefits of the reform. Evidence from these

control groups is mixed, with no differences seen between the inter-listed and not inter-

listed groups, but a small positive effect on net dividend initiations using the ineligible

shareholder classification. These findings are robust to changes in the sample, the set of

controls and the presence of group-level time trends.

For policy and revenue forecasting purposes, the aggregate level of regular dividends is

also important and may behave much differently than discrete dividend-changing events.

Hence the second part of the analysis focuses on the regular dividend payments them-

selves, which are dominated by a small number of large payers. I find some evidence,

using fixed effect models to capture the persistence of dividend payout, of an increase in

regular dividends around the reform. However, any increase seems to be driven primar-

ily by inter-listed firms, making it unlikely that the Canadian tax change is driving the

changes in their payout behaviour. Furthermore, these firms also increased their share

repurchases, the taxation of which did not change around the time of the reform. Over-

all, these findings are somewhat in tension with the extant empirical literature discussed

below.

These results also provide evidence in the debate between the new and old theories of

dividend payout, which are clearly described in Auerbach. The key distinction between

the two theories is the source of finance for marginal investments. The traditional view

suggests that new equity is the marginal source of finance. When an investor decides

whether to inject equity into a firm, he rationally considers the tax that will be due when

his investment pays dividends. Hence, a dividend tax will increase the required rate of

return, or cost of capital, for the firm. Firms will respond to this increased cost of capital

by decreasing investment. This decrease in investment will lead to a fall in dividend

payments in the long run. However, the old view actually predicts that no dividends will


be paid while the marginal source of finance is new equity, since any payout would require

an increase in new equity. To address this deficiency, the theory has been augmented to

include a direct contemporaneous benefit to paying dividends. For instance, dividends

can be used for signalling (Miller and Rock [1985]) or to mitigate agency problems due

to excess cash held in the firm. In these cases, a dividend tax will not only increase the

cost of capital but will also cause current dividend payments to be inefficiently low, since

these extra benefits of dividends cannot be accessed without triggering the tax.

On the other hand, the new view is built on the observation that most new equity

financing comes from retained earnings rather than new share issues. Once the marginal

decision becomes how to use the equity already contributed to the firm, the dividend tax

becomes irrelevant. This is because the equity inside the firm is trapped, in the sense

that it will trigger the dividend tax whenever it is distributed. Hence, the dividend tax is

a lump sum tax on corporate equity and does not affect the cost of capital or investment.

Auerbach and Hassett [2003] identify some circumstances in which dividend taxes could

still have an effect even within the new view paradigm. The most important case is

that of a temporary dividend tax cut, whereby firms would adjust the timing of their

dividends to take advantage of the lower rate, without changing their overall payout.

A different explanation for a limited effect of payout taxes in the context of this

study is that the marginal investor in the Canadian market may not have been affected

by the Canadian reform. This could be because the marginal investor, as in a Miller-

type clientele model (Miller [1977]) is tax-exempt or perhaps located in the United States.

However, the tax-adjusted capital asset pricing model (see Brennan [1970] or Bond et al.

[2007] for a related application) provides a different perspective on the issue. This model

incorporates a risk diversification motive and so yields an equilibrium where all investors

hold every security, even in the presence of differential taxation. The relevant tax rate

for each security is a weighted average of the marginal tax rates of different investors.

This view suggests that the tax cut ought to have had an effect in the Canadian market,

though this effect could be potentially very small, in proportion to the fraction of savings

held by taxable Canadian investors relative to all savings potentially available. In fact,

the prevailing view, since at least Booth [1987], is that the Canadian capital market is

segmented, where some investors and companies access international markets and some

do not.


3.1.1 Empirical Evidence

Much of the recent empirical evidence addressing the dividend theory debate comes from

studies of the Job Growth Taxpayer Relief Reconciliation Act of 2003 in the U.S., which

cut the maximum rate of personal tax on dividends from 35% to 15%. Chetty and Saez

[2005] find a significant positive effect on dividend payout from this reform throughout

the six quarters after the reform. They interpret this finding as supportive of the tra-

ditional view of dividends and in fact, Dhaliwal et al. [2007] find that the 2003 reform

did decrease the cost of equity capital. In line with this result, Campbell et al. [2011]

associate the reform with increases in corporate investment and Lin and Flannery [2013],

with decreases in firm leverage ratios. Blouin et al. [2011] jointly model the firm’s in-

vestor composition and payout choice, finding that directors and officers adjusted their

portfolios in response to the reform, at the same time as firms themselves reacted to the

change in tax incentives. On the other hand, recent research casts some doubt on the

striking findings of Chetty and Saez [2005] and the literature that followed; in particular,

Edgerton [2012] provides four distinct pieces of evidence mitigating the apparently large

response to the 2003 reform, including that dividends from real estate investment trusts,

which were not eligible for the reform, increased by a similar amount.

An alternative way to study this reform is the aggregate time series methodology of

Poterba [2004]. He models aggregate dividend payout as a function of the level and lags

of corporate profits and a parameter measuring the tax preference for dividends relative

to capital gains, finding a positive elasticity consistent with the old view. According to

the model, the effect should be seen only in the long run, and so is difficult to reconcile

with large, immediate effects.

There is also a fair amount of recent international evidence. Jacob and Jacob [2012] is

a large cross-country study of the effect of taxes on corporate payout for 6,035 firms from

25 countries1 over the period 1990-2008. They find that the tax penalty on dividends

over repurchases has important predictive power for that choice. Alstadsæter and Fjærli

[2009] study the introduction of a shareholder tax in Norway in 2006 which increased the

top marginal tax rate on dividends from 0% to 28% and find substantial intertemporal

shifting around the reform for non-listed firms.

Concerning the Canadian evidence in this debate, McKenzie and Thompson [1996] is

a useful survey. They note that most older studies using Canadian data find significant

effects of tax policy on equity prices, suggesting that regardless of the prevailing dividend

theory, Canadian tax policy can indeed affect the marginal investor in the Canadian mar-

1The sample includes 248 Canadian firms for an average of 5 years each.


ket. For example, McKenzie and Thompson [1995] study a tax reform from 1986, which

had the effect of increasing the effective tax rate on dividends. They find significant

negative effects on the prices of high dividend-yield stocks, which are consistent in mag-

nitude with a clientele consisting of top tax bracket individual investors. More recently,

Dutta et al. [2004] study ex-dividend day price drops and find evidence consistent with

a significant effect of dividend taxes on share prices. Perhaps most closely related to my

study, Kooli and L’Her [2010] look at the changing pattern of dividends and repurchases

from 1985-2003 in Canada which reveals a shift in payout from dividends to repurchases.

Their explanation for this phenomenon is the relative benefit of repurchases for financial

flexibility.2 Canadian evidence of a different type comes from a survey of managers un-

dertaken by Baker et al. [2012]. They report that taxation is no more than a second-order

determinant of dividend policy and that share repurchases are not used as a substitute

for dividends.

The remainder of the chapter proceeds as follows: Section 3.2 describes the structure

of dividend taxation in Canada and the 2006 reform, Section 3.3 presents the data,

while Sections 3.4 and 3.5 develop the empirical analysis of dividend events and levels,

respectively, and Section 3.6 concludes.

3.2 Dividend Taxation in Canada

The personal income tax system in Canada has, since 1972, partially integrated corporate

and personal income taxation through a ‘gross-up and credit’ system of dividend taxation.

The effective tax rate on dividends for an investor i, mi, generated by such a system is

as follows:

mi = τ + (1− τ)(1 +G)(ti − c) (3.1)

where τ is the corporate tax rate, G is the rate of the gross-up, ti is the investor’s marginal

tax rate (federal plus provincial) and c is the rate of the dividend tax credit (DTC). The

income is earned at the corporate level, where it is taxed at the corporate income tax

rate. What remains after this tax is paid is the amount of the dividend paid to the

investor. The gross-up is then applied to the dividend and this grossed up amount is

included in the investor’s taxable income. To get full integration, the gross-up should be

set to τ1−τ so that the full pre-tax amount of the dividend is included in personal income,

and the rate of the dividend tax credit should be set to τ to exactly offset the tax already

2They also find that the decrease in the capital gains inclusion rate in 2000 was associated with anincrease in the propensity to repurchase shares.


paid at the corporate level. If this were the case then the effective tax rate on dividends

for a particular investor would be the same as her marginal tax rate on ordinary income.

Prior to 2006, the rate of gross-up was 25% and a federal tax credit of 13.33% of

this amount was accorded. These specific rates led to approximately full integration

for dividend income flowing from small businesses, which are eligible for a lower rate of

corporate income tax.3 For large businesses, this particular gross-up and credit led to

underintegration, raising concern about double taxation of corporate income.

The dividend tax reform of 2006 (the so-called enhanced dividend tax credit) increased

the gross-up rate to 45% and the federal credit rate to 18.97% on dividends paid by large

corporations. This enhanced dividend tax credit yields roughly full integration for large

corporations, and preserves full integration for small businesses (by using the old gross-

up and credit rates for such firms). The effect of this reform was to decrease the total

effective tax rate on dividends for a top tax-bracket investor in Ontario from about

55% to 46%.4 Overall, the tax reform significantly decreased the taxation of dividends

for taxable Canadian investors, leaving it unchanged for tax-exempt Canadian investors

(including other corporations) and foreign investors.

Since the object of this study is the effect on corporate payout of the reform, it is

important to carefully consider the timing and predictability of the change. The enhanced

dividend tax credit was proposed by the Liberal government on November 23, 2005,

principally in response to the rise of income trusts. This business structure, under which

distributed income was taxed only at the shareholder level as ordinary income, had been

tax-advantaged relative to the corporate form because of underintegration. Corporate

conversions to income trusts were an increasing policy concern, mainly because of the

associated tax revenue losses.5 The enhanced dividend tax credit was partially meant

to level the playing field by lowering the effective tax burden on corporations without

changing the tax treatment of income trusts.6

3The special deduction for small businesses that yields this lower rate is completely phased out at$15M of taxable capital so is not relevant for the publicly traded companies analyzed in this study.

4This calculation uses equation (3.1) and follows the budget proposal’s assumptions of a notionalcorporate tax rate of 32% by 2010, with the combined federal and provincial credit rate increasing from18.46% to 32%, along with the gross-up going from 25% to 45%.

5Mintz and Richardson [2006] estimated an annual revenue loss from income trust conversions of$700M, consisting of a $1.8B decrease in corporate tax revenues partly offset by a $1.1B increase inpersonal tax revenue.

6McKenzie [2006] describes the economics and structure of income trusts in much more detail. Likelybecause the enhanced DTC did not effect foreign or tax exempt investors, this reform did not have thegovernment’s desired chilling effect on income trust conversions. Hence on October 31, 2006, a new taxregime for income trusts, the specialized investment flow-through regime, was announced, whereby theywould be effectively taxed at the same level as corporations starting in 2011. This led most incometrusts to convert back to corporations, as the tax advantage had been eliminated, while non-tax costs of


There is ample reason to believe that this original proposal was a surprise to investors.

In fact, a controversy ensued over whether the details of the proposed reform had been

leaked several hours prior to the announcement, due to a suspicious pattern and volume

of trading. This implies that market participants were surprised by the information

and so acted to adjust their portfolios. However, before the legislation could be passed,

the election in January of 2006 led to a change in government. That said, the two

parties with a reasonable chance to form the next government (the incumbent Liberal

Party and the Conservative Party) were both committed through campaign promises to

following through on the enhanced dividend tax credit proposal. The new Conservative

government indicated it would follow through immediately in the Budget 2006 speech

of May 2, 2006, and proposed draft implementation legislation in June of 2006, with

retroactive effect to the beginning of 2006. Throughout 2006, the provinces separately

indicated their intention to increase their own dividend tax credits in line with the federal

legislation, and ended up doing so.

For example, the province of Ontario detailed its enhanced DTC proposal on August

3, 2006. This plan followed the new federal rules for gross-up of eligible dividends and

increased the rate of the dividend tax credit from 5.13% to 6.5%, also retroactive to

the beginning of 2006, with plans to increase the credit rate to 7.7% by 2010. At such

rates, the Ontario dividend tax credit would provide for full integration with a provincial

corporate tax rate of 12%, which was the preferential rate applied to manufacturing and

several other industries.

Overall, it seems likely that the enhanced dividend tax credit was not anticipated

prior to the end of November of 2005, but that thereafter, corporations put a very high

degree of probability on a decrease in the effective tax rate on dividends. Furthermore,

the enhanced dividend credit appears now to be a permanent feature of the tax system,

as changes to the rate of gross-up and credit have already been announced to maintain

the present level of integration as the corporate tax has fallen through 2012.

Given the significant attention paid to the recent U.S. dividend tax cut, it is useful to

compare the impacts of the two reforms on effective tax rates. Prior to the U.S. reform,

dividends were first taxed at the corporate level and then again at the investor’s full

marginal tax rate. The tax reform of 2003 cut the tax at the investor level to a flat 15%

for high-income investors (or 5% for sufficiently low-income investors). This meant that

the weighted average tax rate on dividends (not including the corporate tax) fell from

29% to 17%, which corresponds to a 40% cut. For comparison, the Canadian reform

this form, such as increased agency costs from a less developed legal framework, remained.


cut the top marginal tax rate on dividends in Ontario from approximately 34% to 20%7

which is a drop of similar magnitude.

3.3 Data

3.3.1 Corporate Payout

Data on dividends come from the Canadian Financial Markets Research Center (CFMRC),

which covers all public companies listed on the Toronto Stock Exchange. The price ad-

justments file consists of daily records of dividend payments, among other events, such as

stock splits. To generate a dividend time series for each company, the dividend records

were aggregated up to the year level by stock symbol and usage code, which uniquely

identifies an issuer. This procedure allows for a company to have multiple classes of

common stock. Following the literature, only dividends on common shares, rather than

preferred shares, are included. The benefit of the CFMRC data over accounting data

from Compustat is the ability to distinguish between regular and special dividends. Data

on share repurchases come from Compustat and are calculated following the algorithm of

Grullon and Michaely [2002], which is described in Appendix 3.A, along with the rest of

the variable definitions. The firm-year is the unit of analysis throughout this study. All

dollar values are deflated to 2010 Canadian dollars using the consumer price index (and

the relevant Canada-U.S. exchange rate for companies which report in U.S. dollars).

The main sample displayed in the descriptive graphs and in the econometric anal-

ysis drops financial and utility companies, following Fama and French [2001] and the

subsequent literature.8 Furthermore, foreign-based firms are dropped, based on the clas-

sification in the CFMRC database, the necessary accounting control variables must be

available, and finally, firms which were ever income trusts are dropped. The latter two

restrictions are discussed in more detail below.

Figure 3.1 shows the time series of corporate payout in Canada from 1995-2010. It is

split into three components: regular dividends, special dividends and share repurchases.

7The tax rate immediately dropped by approximately five percentage points, with the further re-ductions to follow as provinces matched the federal government’s increase in dividend tax credit rates.Though this was the announced change when enacted in 2006, the actual shareholder dividend tax ratein Ontario did not end up falling by the amount originally envisaged by the federal government. Thishappened because Ontario did not continue to increase its credit rate after 2007; however, two otherlarge provinces (British Columbia and Alberta) both achieved approximately full integration by 2008.

8Specifically, firms with first digit SIC of ‘6’ (financials) and first two digits of ‘49’ (utilities) aredropped. The typical argument given in favour of this restriction is that these firms follow a fundamen-tally different payout strategy, particularly because of government regulation. Additionally, the sharerepurchases variable cannot be calculated for such companies, due to financial reporting differences.


First, the graph demonstrates that special dividends are not generally an important

component of payout.9 Regular dividends and share repurchases are of similar magnitude

and make up the bulk of corporate payout. Repurchases are shown mainly as a point

of comparison with dividends, but are not the main object of this study.10 As is well

known in the literature, regular dividends are much more stable and persistent than share

repurchases. In particular, the firm-level correlation between regular dividends and its

lagged value is .93; the equivalent value for repurchases is .60. The contrast between the

two is clear in 2009, as the worsening financial crisis appeared to cause a much larger

drop in repurchase activity than in dividends. The vertical line between 2005 and 2006

indicates when the enhanced dividend tax credit reform took place. The reform does

not appear to have had much effect on the aggregate time series, with regular dividends

continuing an upward trend through the reform.

The aggregate regular dividend time series is driven by the largest dividend payers,

given the extreme concentration of regular dividend payments.11 Hence, the payout

behaviour of the vast majority of firms is obscured in the aggregate data. A common

way to address this issue is to investigate the propensity to pay dividends at all. This

effectively weights all companies equally and so provides more information to learn about

the determinants of corporate payout. Figure 3.2 depicts the fraction of firms paying any

regular dividends, any special dividends, or making any share repurchases, respectively,

over time. Note that using the fraction of payers, rather than the raw number, accounts

for the fluctuating sample size over time due to firm entry and exit.

The time series of regular dividend payers reveals a rather different story. This fraction

is fairly stable in the years prior to the reform and appears to fall slightly from 2005-

2007, which implies that the run-up in the aggregate time series was driven by changes

on the intensive margin. Notably, a similar downward long-run trend in dividend payers

was documented by Fama and French [2001] in the U.S., where the fraction of dividend

paying firms fell from about 67% in 1978 to 21% in 1999.

The preceding graphs depict payout behaviour through 2010; however, in all of the

analysis that follows, the sample ends prior to 2008. This is done for several reasons.

The financial crisis could have involved a structural break in dividend behaviour, and

substantial federal corporate statutory income tax rate cuts began in 2008, causing the

9The small spike in 2005 is due to a special dividend paid on restructuring by one particular company(Sears).

10Note that repurchases are taxed as capital gains, the taxation of which did not change as part ofthe 2006 reform.

11The top decile of regular dividend payers covers 94% of the total in my sample while the top percentilecovers 44%.


rate to fall by about a third over the subsequent four years.

3.3.2 Control variables

To isolate the effect of the tax reform, it will be important to control for other time-

varying influences on dividend payout. The starting point is to account for firm-level

changes with the possibility of secular changes to dividend strategies discussed in a later

section. To that end, the CFMRC data are merged with Compustat using the company’s

6 digit CUSIP code. The required control variables are income before extraordinary

items, to measure profitability, market capitalization and total assets to control for firm

size, and the market-book ratio and asset growth rate from year t− 1 to t to account for

growth and growth opportunities, which might substitute for dividends. Furthermore,

cash holdings are used to control for changes in the availability of resources with which to

pay dividends to shareholders.12 Table 3.1 provides descriptive statistics for the payout

and control variables.

3.3.3 Income Trusts

As mentioned in the tax reform discussion, income trusts were a tax advantaged business

form of increasing importance in the mid-2000s13, characterized by their very high payout

to their investors. This phenomenon can have two kinds of effects on the analysis of

regular dividend behaviour around the 2006 reform. First, since firms which were ever

income trusts are dropped from the analysis, the sample of firms available to respond

to the reform is different than it would be in the absence of the income trust structure.

Hypothetically, the corporations which converted to income trusts may have responded

more or less to a dividend tax cut than those which did not. On one hand, converters

demonstrated the salience of taxes to their decision making and their high payouts imply

that they are certainly at the margin of increasing them. On the other hand, given the

necessity for income trusts to distribute essentially all their profits every year, income

trust adopters tended to be mature companies, which would finance themselves out of

retained earnings. This fits the new view paradigm of dividend taxation, so that these

firms might be expected not to change their behaviour in response to a dividend tax cut.

The net effect of these two arguments is unclear, but important to keep in mind when

12The control variables are winsorized at the 1% level to minimize the influence of outliers and mea-surement error. This yields a somewhat better model fit in general. However, the payout results are notmaterially affected by this adjustment.

13According to data from Compustat, the income trust share of total market capitalization on theToronto Stock Exchange peaked at roughly 15% in 2006.


interpreting the results in the following sections.

The second type of effect from the rise of income trusts is indirect. As a firm’s com-

petitors become income trusts and increase their payout to shareholders, it can respond

in two possible ways. The firm may increase dividends if it is trying to cater to the same

type of investors as income trusts. Alternatively, the market may become segmented,

with income trusts appealing to investors who want current income, and corporations

cutting or stabilizing dividends to attract investors looking for future growth. In such a

case, more competing income trusts could actually lead to less dividends paid by corpo-

rations. To address these indirect effects, the level and lag of new income trust inceptions

at the first digit SIC industry level in a particular year is included in the empirical model.

3.4 Analysis of regular dividend events

The purpose of the models in this section is to analyze the effect of the 2006 dividend

tax cut on dividend payout by Canadian firms. The two main theoretical models make

different predictions for this treatment effect, with the new view suggesting that dividend

payout will be unchanged, and the traditional view suggesting that the tax cut would

stimulate dividend payments.

As pointed out by Chetty and Saez [2005], it is difficult to estimate robust treatment

effects using regular dividends as the outcome variable. One problem is the obvious

persistence in this variable, but perhaps more importantly, aggregate dividend payments

are driven by a relatively small number of large payers, with most firms not paying

dividends at all. This makes the treatment effect very sensitive to small changes in the

specification and sensitive to outliers. However, for forecasting the aggregate effects of a

dividend tax cut, the effect on these large payers is of primary importance, and will be

investigated in Section 3.5.

To avoid this issue, it is helpful to start with the decomposition of dividend changes

into discrete changes on the extensive (dividend initiations or terminations) and inten-

sive (dividend increases or decreases) margins. Specifically, four dummy variables are

encoded14 which equal one when a firm initiates regular dividend payments, terminates

them, increases existing regular dividend payments by at least 20% or decreases them by

at least 20%, respectively.

Figure 3.3 shows the percentage of firms in the sample initiating, terminating, increas-

ing or decreasing regular dividends. These time series do not appear to follow a clear

pattern around the reform after 2005. The net entry into the dividend-paying sample,

14The definitions of these variables are in Appendix 3.A.


determined by initiations less terminations, is low throughout the sample period, imply-

ing that the long-run fall in the propensity to pay dividends in Canada is being driven

by the entry of non-dividend payers into the sample. The net number of initiations is

actually the highest in the several years prior to the reform. As can be verified in Table

3.1, dividend increases are the most common event by a factor of more than two, and,

not surprisingly, dividend decreases are very rare. The financial crisis seems to have been

associated with a large fall in dividend increases. There is also a noticeable up-turn in

initiations and increases in the last year pictured on the graph, 2010, suggesting the start

of a recovery though the main estimation sample ends in 2007 and so will not address

such behaviour.

The first step beyond the graphical analysis is to add in control variables to account for

firm level differences in the drivers of dividend payout over time. To that end, following

Chetty and Saez [2005], I estimate linear probability models15 of the form:

Eventi,t = βXi,t + γPostReformt + δt + ϵi,t (3.2)

where Xi,t includes the level and first lag of the firm’s profitability rate, logarithm of

total assets, market capitalization relative to total assets, market-book ratio and cash

holdings relative to total assets, and the firm’s total asset growth rate relative to the

prior year. Also included is a set of dummy variables for first-digit SIC to control for

fixed industry differences and a time trend. Eventi,t is a dummy variable that is equal

to unity if firm i undertook a regular dividend event in year t, where this refers to one

of initiations, terminations, increases or decreases depending on the model.

Table 3.2 presents the results of estimating Equation (3.2) for each kind of dividend

event. As expected given the graphical analysis, there is little effect of the reform on

dividend events, with the exception of a small and negative effect on regular dividend

initiations. The coefficient of -0.015 implies a fall of about 9 initiations per year, given the

sample size in the year preceding the reform of 604 (of which 149 paid regular dividends).

This estimate is about half of the initiation rate in 2005 of 2.8%. Dividend increases

appear to rise following the reform, though not by enough to attain statistical significance.

Still, this opposing evidence suggests caution in interpreting the fall in initiations as a

negative causal effect of the dividend tax cut. Terminations and decreases do not change

following the reform.

The only control variable which seems to have an important positive influence on the

dividend decision is the level and lag of income relative to total assets. As expected,

15Estimating probit models instead does not materially affect the conclusions.


increases in this variable increase the propensity for a firm to initiate or increase regular

dividends. The positive effect of income on decreases is difficult to interpret given the very

low propensity to decrease regular dividends. The other relatively clear pattern is that

falls in market capitalization, normalized by total assets, predict dividend terminations

and decreases. The trust inceptions variable has no effect in these models.

The last column of Table 3.2 applies the same model to a dummy variable equal to

one if the firm repurchased any shares, in order to check whether there is any obvious

substitution response. For instance, if share repurchases fell drastically while dividends

remained the same, the fraction of payout from dividends would have increased, which

could be interpreted as a positive treatment effect of the reform. However, this is not

the case, as the estimated effect is very small and statistically insignificant.

One possible source of concern in interpreting the preceding results is that the sample

size is fluctuating as firms enter and exit the data. For comparability of results with the

existing literature, it is useful to define a new estimation sample as the top X firms by

market capitalization in each year, where X is chosen as the number of firms available in

the year with the smallest sample. In addition, the sample is restricted to start in 2002

to keep the number of firms per year,16 and so the number of dividend events, at a high

enough level to detect changes. Specifically, the ‘short’ sample contains 587 firms per

year. This shorter sample also means that the time trend soaks up more secular variation

in dividend payout behaviour. Table 3.3 replicates Table 3.2 using this new sample. It

is immediately clear that the results are substantially the same, with initiations still

decreasing somewhat following the reform. The overall picture remains one of no effect

and suggests that changes in net firm entry and exit were not driving the small estimated

effects.

3.4.1 Difference-in-differences

The obvious problem with the results discussed above is that they may be driven entirely

by secular changes in dividend decisions, completely separate from the reform, such as

changes in dividend catering incentives.17 This could mean that a positive treatment

effect is counteracted by a negative secular trend. A potential solution to this problem

is to identify a group of firms which would not be expected to respond to the reform for

exogenous reasons but may otherwise employ similar payout strategies, conditional on

16Changing this starting year does not seem to make much difference to the results.17This idea was introduced by Baker and Wurgler [2004], whereby managers pay dividends when the

market rewards them for doing so. Preliminary investigations of these incentives in the Canadian market,for instance using the market-book premium of dividend payers relatively to non-payers, yield no obviouspattern.


firm characteristics. There are two possible candidate types of firms to be used as control

groups in a difference-in-differences strategy.

Presence of large corporate or foreign shareholder

In Canada, foreign investors pay a fixed withholding tax on dividends, based on the

relevant tax treaty with the investor’s country of residence, and are not eligible for the

dividend tax credit. Hence, the preferred payout policy of such investors should not

have been affected by the enhanced DTC. Similarly, pension funds are tax-exempt and

so are likewise not eligible for any dividend tax credit. Finally, corporations themselves

can deduct dividends received from other corporations for tax purposes, as long as they

own at least 10% of the dividend payer, and so do not get the credit. Hence, any firms

with large investors that are part of one of these groups should not be influenced by the

dividend tax cut.18

Therefore, I use data on large shareholders of the firms in the sample from the Fi-

nancial Post Corporate Database in 2008 to tabulate the fraction of each firm held by

an investor that is certain to be ineligible for the DTC.19 The first control group in the

results below consists of all firms which have at least one such shareholder owning more

than 10% of the firm. I define the treatment group conservatively, in the sense that

ownership data must also be available for such firms, and show no credit-ineligible share-

holders. For simplicity, this classification will be referred to as the ‘eligible/ineligible

shareholder’ classification, where ‘eligible shareholder’ just refers as shorthand to a firm

with no ineligible 10% shareholders.

The graphical difference-in-differences analysis of dividend events following this def-

inition is shown in Figure 3.4. This figure, and the remainder of the analysis in this

subsection, use the short sample to minimize the measurement error associated with us-

ing ownership data from a particular year, and to keep the sample size at a reasonable

level. This sample has 345 firms per year, after the requirement for availability of owner-

ship data. Considering the volatility of dividend events, the trends across the two groups

prior to the reform are not too dissimilar.20

18Barclay et al. [2009] study transactions where corporations, which have a tax preference for dividendsin the U.S. as well, acquire large blocks of shares in other firms – they find no evidence that dividendsincrease after these events nor that deals target dividend payers preferentially. On the contrary, usingCanadian data, Zeng [2011] finds that membership in a corporate group is associated with increases individends and decreases in share repurchases.

19Robustness checks using 2005 data from prior to the reform yield very similar results. There wasin fact little portfolio reallocation in response to the reform, at least at the level of 10% shareholdings,of the kind that might be expected based on the work of Desai and Dharmapala [2011] for the UnitedStates.

20One exception is a noticeable gap for initiations in 2004, which will be addressed in a robustness


Inter-listed firms

The second control group consists of Canadian corporations inter-listed in the United

States. Booth and Johnston [1984] found that such firms tend to be priced by investors

located in the United States. Since such investors were not affected by the enhanced

DTC reform, we would expect inter-listed firms to be much less likely to respond, even

though they face similar business conditions and institutions as other Canadian firms.

Figure 3.5 illustrates the graphical dividend event difference-in-differences using inter-

listed companies as a control group. The two groups track relatively closely both before

and after the reform for each type of event. This classification will be referred to as the

‘inter-listing’ classification.

Table 3.4 contains summary statistics split by the two treatment and control defini-

tions. This shows that the ineligible shareholder classification is well balanced in terms of

size, though the treatment firms may be more growth-oriented, with lower current prof-

itability but higher market capitalization and asset growth. The inter-listing comparison

yields groups of firms that are similar in terms of profitability but, as expected given

the high fixed costs of a stock market listing, inter-listed firms are much larger in terms

of both total assets and market capitalization. The final row of Table 3.4 is reassuring

in the sense that a very similar fraction of firms in each group pay regular dividends,

ranging from 28% to 31%.

Results

As before, and especially in light of the compositional differences in the groups, it is

important to control for firm level characteristics. This is accomplished using a linear

probability framework and controlling for a wide set of accounting variables:

Eventi,t = βXi,t + γ1Eligiblei X Postreformt + γ2NotInteri X PostReformt

+ γ3Eligiblei + γ4NotInteri + γ5PostReformt + δt + ϵi,t (3.3)

with X containing the same set of control variables as before. The coefficients γ1 and

γ2 yield the estimated treatment effects from the two control strategies described above.

Table 3.5 gives the results.

To start, for the eligible shareholder classification, the estimated treatment effect for

dividend initiations is positive and significant at the 5% level, and corresponds to 4.6

test.


more firms initiating dividends per year after the reform. For terminations, the effect

is negative. Combining these two results yields an estimated increase in initiations net

of terminations of approximately 7 per year. To put this in a larger context, it means

that the reform caused a rise in the number of dividend paying firms of about 10% in

total over the two subsequent years.21 As expected given the graphical evidence, the

effect on increases is positive, though not statistically significant. Since about half of the

firms in the sample have a large DTC-ineligible shareholder, any possible benefit from

cutting dividend taxes in this context is limited, provided this classification is descriptive

of actual behaviour.

In contrast, the second row of Table 3.5 shows that there are no significant differences

around the reform for any of the four dividend events using the inter-listing classifica-

tion.22

Reassuringly, adding group-specific time trends does not alter these results. Likewise,

estimating the model separately for each classification yields similar results. Inspection

of Figure 3.4 suggests concern that initiations for the two eligible and ineligible groups

diverge significantly in 2004, but are otherwise quite similar in the pre-reform period.

The direction of the difference mechanically increases the estimated effect of the reform.

Redoing the analysis with 2004 omitted causes a small decrease in the coefficient and

an increase in the standard error so that the effect is no longer statistically significant.

However, this omission does not change the estimated effect on regular dividend termi-

nations.

Overall, the difference-in-differences analysis of discrete dividend events provides some

evidence of a modest treatment effect from the reform among a subset of firms which did

not have large shareholders who could not take advantage of the reform. Since this effect

is not also seen for the inter-listing classification, the evidence is mixed and can support

at most a small positive effect of cutting dividend taxes on dividend payout.

3.5 Analysis of regular dividend levels

Analyzing discrete dividend events implicitly weights each firm equally. If the object is

to learn about firm behaviour, this is the correct approach to take, as it makes use of

all of the available data. However, regular dividend payout is highly concentrated, so for

21This calculation comes from the 14 net initiations (7 per year) relative to the 149 firms in the mainsample paying regular dividends in 2005.

22The ‘triple interaction’ combining the two classifications would be of interest but turns out not toreveal anything. This is likely due to the small number of dividend events in the overlap of the twotreatment groups.


assessing the aggregate outcomes of a tax cut, only the response of a small minority of

firms actually matters.

The first step is to control for observable influences on dividend payout in the firm-

level data making up the aggregate time series seen in Figure 3.1. This is done in the

following regression:

RegDivi,t = βXi,t + γPostReformt + ωi + δt + ϵi,t (3.4)

where X includes the level and first lag of the firm’s income, logarithm of total assets,

market capitalization, market-book ratio and cash holdings, and the firm’s total asset

growth rate relative to the prior year. Also included are firm fixed effects, to capture the

strong persistence in regular dividends,23 and a time trend. The results of estimating

this model are in Table 3.6.

The first column uses the full sample from 1995-2007 and shows an increase of $4.9M in

regular dividends per firm after the reform, which is quite large compared to the average

dividend payment of $10.5M in 2005. The second column reports a smaller, though

still statistically significant, estimate for the fixed number of firms sample (2002-2007).

Given that Table 3.2 showed only small changes in dividend initiations and increases

around the reform, this rise in dividend levels must be coming from heterogeneity in

those dividend events. To investigate this point further, the sample in the third column

of Table 3.6 excludes all firms which were in the top 10% of the size distribution by total

assets in 2002. This exclusion completely eliminates any effect from the reform, which

demonstrates the extreme concentration of changes in regular dividends in addition to the

aggregate concentration already documented. As a check for any substitution response

away from other types of corporate payout, the same model is run for share repurchases

in the fourth column, using the short sample. This shows no response to the reform.

The negative coefficient, which is significant at the 10% level, on same industry in-

come trust inceptions suggests that firms might be responding to the rise of high payout

competitors by refocusing on growth, rather than payout. The average number of trust

inceptions by industry is 2.5 in 2005, for example, so this negative effect on dividends is

approximately one million dollars for an average firm in that year. In general, however,

interpretation of the control variables is complicated by the concentration of the depen-

dent variable – since a small number of firms pays the bulk of aggregate dividends, a

small number of firms will be driving the effects of these control variables.

As Figure 3.1 shows regular dividends trending upwards possibly before the reform

23Estimating lagged dependent variable models using the Arellano-Bond method yields similar con-clusions.


in 2006, there is some concern that the post-reform dummy is picking up the effect of a

change that happened earlier. However, using ‘placebo’ reforms which include 2005 or

2004 in the post-reform period result in substantially lower estimates.

3.5.1 Difference-in-differences

A difference-in-differences strategy is also useful in the context of regular dividend levels

to investigate whether this apparent positive effect of the tax cut is causal. Figure

3.6 shows the time series in average regular dividends per firm, split by the two group

classifications discussed in Section 3.4: ineligible shareholders and firm inter-listing. The

top panel hints at a positive treatment effect as both groups track closely together until

the DTC-eligible group starts to increase in 2006. The bottom panel splits the time series

into three groups: majority inter-listing,24 minority inter-listing and no inter-listing. The

split within the inter-listed firms turns out to be important in the regression results, which

is to be expected given that the figure clearly shows that the majority group both pays

more regular dividends and saw a large relative increase in 2006 and 2007.

Given the importance of relatively few firms in aggregate payout, it is particularly

important to control for firm-level changes. Results from estimating equation 3.4 with

the two (or three) treatment group interactions used in Section 3.4 are in Table 3.7.

In line with the results for dividend initiations, the eligible shareholder classification

yields a positive treatment effect, though in this case it is not statistically significant.

The negative effect seen for the inter-listing classification is likewise not significant. To

investigate further the apparent increase in aggregate, the second column includes an

interaction of the post-reform dummy with a dummy for majority inter-listing, in addition

to interaction terms for the other two treatment groups. This model presents a much

different picture. Evidently, the firms which are increasing their regular dividends as of

2006 are those with a minority inter-listing, relative to both majority inter-listing and

no inter-listing at all. It does not seem reasonable to attribute this pattern of changes

to the dividend tax cut, especially since a similar pattern is observed looking only at the

top size decile of firms. This strategy significantly reduces the size differences across the

three groups, and so the possibility that such heterogeneity is driving the result. A tax

explanation would also be more compelling if the dividend event analysis had revealed a

similar pattern among inter-listed firms, which it did not.

Another piece of evidence in favour of this interpretation is seen in the third column of

Table 3.7, which uses all three group interactions to explain share repurchase behaviour.

24This includes companies for which more than 50% of their trading volume (by value) is done outsideof the Canadian market.


The pattern of results is almost identical to that for dividends. Hence, minority inter-

listed firms greatly increased both dividends and share repurchases, though the latter by

a much greater amount, even though their tax treatment did not change. This suggests

that, conditional on observable characteristics, this particular group of companies wanted

to increase their total payout for reasons independent of the enhanced DTC reform.

3.6 Conclusion

The effect of dividend taxes on corporate payout is an important question in both cor-

porate and public finance. I use a 2006 tax reform in Canada which substantially cut

the shareholder tax on dividends to provide new evidence to this debate. Making use of

control groups made up of firms which, a priori, would not be expected to respond to

the reform, I find evidence for at most a small, positive effect on net dividend initiations.

The aggregate change in regular dividends following the reform is also positive, but is

dominated by a small number of the largest firms. Since the increase is coming primarily

from inter-listed firms which also increased their share repurchases, the change in tax

incentives is very unlikely to have been the cause.

Overall, this study suggests that cutting domestic shareholder dividend taxes is not

an effective way to stimulate investment in a small, open economy, because the marginal

investor does not pay such taxes. Interestingly, tax changes in Canada since 2008 appear

to reflect this point. The combined dividend tax burden has been held approximately

constant but has been shifted increasingly from the corporate to the shareholder level.

This kind of reform seems to be aimed at the large set of investors (both foreign and tax

exempt) for whom entity level taxation is the primary concern.


3.A Variable Definitions

Regular dividend initiations

An initiation is defined to occur when a firm pays positive regular dividends in quarter

t and did not pay regular dividends in any of the previous four quarters. This measure

is aggregated up to the year level to indicate whether a firm initiated dividends in any

quarter of the year.

Regular dividend terminations

A termination is defined to occur when a firm paid positive regular dividends in quarter

t− 1 but does not pay regular dividends in any of the next four quarters. Further details

are necessary to code terminations at the end of a firm’s tenure in the sample since in

such cases we do not necessarily observe the proceeding four quarters. In this case, a

termination is coded as long as regular dividends are zero, with at least one more quarter

of zero dividends before the last time the firm is in the sample. Otherwise, the firm-

quarter would be coded as a dividend decrease. Again, this measure is aggregated up to

the year level.

Regular dividend increases

An increase is defined to occur when regular dividends exceed the maximum nominal

value of regular dividends in the preceding four quarters by at least 20% and the firm is

not initiating dividends in the current quarter. In aggregating this measure to the year

level, multiple increases are counted as one.

Regular dividend decreases

A decrease is defined in terms of yearly total dividends to avoid mistaking a change in

timing for a decrease. For such an event to be coded, the nominal value of dividends

must fall by at least 20% in a year where a termination is not also coded.

Share repurchases

Following Grullon and Michaely [2002], share repurchases are defined as total expenditure

on the purchase of common and preferred stock (prstkc) less any reduction in value of

the net number of outstanding preferred stock (pstkrv).


Market-book ratio

From Hoberg and Prabhala [2009], this ratio is total assets (at) less shareholder equity

(se) plus market equity (from CFMRC), all divided by total assets (at).


05

1015

Bill

ions

201

0$

1995 2000 2005 2010Year

Regular Dividends Special DividendsShare Repurchases

Corporate Payout in Canada: 1995−2010

Figure 3.1: The vertical line shows the beginning of the enhanced dividend tax creditreform. This figure shows regular and special dividends, collected from CFMRC priceadjustments data. Note that though there was a notable increase in regular dividends,the timing does not line up with the Canadian tax reform. Special dividends are usu-ally not important relative to regular dividends, except for a few particular quarterswith individual large payments. However, these payments generally correspond to largerestructurings or divestments and so do not appear to be tax-motivated.


0.1

.2.3

.4F

ract

ion

of S

ampl

e

1995 2000 2005 2010Year

Any Regular Any SpecialAny Repurchases

Corporate Payers in Canada: 1995−2010

Figure 3.2: The vertical line shows the beginning of the enhanced dividend tax creditreform. The solid line shows the percentage of dividend payers in the full sample, whilethe dashed line shows the same thing for the sample with all necessary controls available.There has been a significant decrease in the fraction of dividend payers over time.


0.0

2.0

4.0

6.0

8F

ract

ion

of S

ampl

e

1995 2000 2005 2010Year

Frac. Initiating Frac. IncreasingFrac. Decreasing Frac. Terminating

Regular Dividend Events: 1995−2010

Figure 3.3: This figure shows regular dividend initiations, terminations, increases anddecreases for the sample of non-financial, non-utility and non-income trust companieswith all necessary controls available. See Appendix 3.A for definitions of these variables.


.01

.02

.03

.04

.05

Fra

ctio

n of

Sam

ple

2002 2003 2004 2005 2006 2007Year

Initiations

0.0

5.1

.15

.2F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Increases

0.0

1.0

2.0

3F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Terminations

0.0

05.0

1.0

15.0

2F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Decreases

Dividend Events: Ineligible Shareholders as a Control Group

Eligible Ineligible

Figure 3.4: Each panel shows the time series of one type of regular dividend event, splitby whether the firm has any 10% DTC-ineligible shareholders.


0.0

1.0

2.0

3.0

4F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Initiations

.02

.06

.1.1

4F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Increases

0.0

1.0

2.0

3F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Terminations

0.0

05.0

1.0

15.0

2F

ract

ion

of S

ampl

e

2002 2003 2004 2005 2006 2007Year

Decreases

Dividend Events: Inter−listed Firms as a Control Group

Not Inter−listed Inter−listed

Figure 3.5: Each panel shows the time series of one type of regular dividend event, splitby whether the firm is inter-listed in the United States.


010

2030

4020

10 M

illio

ns$

2002 2003 2004 2005 2006 2007Year

Dividends (ineligible)

Dividends (eligible)

050

100

150

2010

Mill

ions

$

2002 2003 2004 2005 2006 2007Year

Dividends (majority inter−listed)

Dividends (minority inter−listed)

Dividends (not inter−listed)

Regular Dividends: Treatment vs. Control

Figure 3.6: The top panel splits average regular dividends per firm into those without(eligible) and with a 10% shareholder who is ineligible for the dividend tax credit. Thebottom panel splits regular dividends per firm into three groups, those with a majorityinter-listing, a minority inter-listing and no inter-listing. In each panel, the bottom-mostcategory in the legend is the ‘treatment’ group, which theoretically should have increasedthe most after the 2006 reform.


Table 3.1: Summary statistics for the full sample with all available controls, N=6965.The mean income/assets value in the first row would be positive and equal to .028weighted by total assets. The last four rows in the tables denote the probabilities ofundertaking the particular dividend event.

Mean (Std. Dev.)Income/Assets -0.084 (0.324)Income 39.8 (180)Log Assets 5.26 (1.93)Asset Growth Rate 0.209 (0.672)Market Cap./Assets 1.39 (1.70)Market Capitalization 901 (2620)Market-Book Ratio 2.17 (1.98)Cash/Assets 0.156 (0.207)Cash 76.9 (210)Industry Trust Inceptions 1.43 (2.08)Initiation 0.021 (0.143)Increase 0.054 (0.227)Termination 0.017 (0.13)Decrease 0.013 (0.112)


Table 3.2: This table displays coefficients from linear probability models of regulardividend events (and repurchases, in the last column) on a post-reform dummy andaccounting controls. Each column also includes a set of dummy variables for first digitSIC. The ‘L.’ denotes the first lag of variable which it precedes. The sample comprises allnon-financial, non-utility companies, which never became income trusts, from 1995-2007.*, **, *** denote significance at 10%, 5% and 1%, respectively.

Initiation Increase Termination Decrease Any RepurchasePost-Reform Dummy -0.015*** 0.011 0.001 0.004 -0.008

(0.006) (0.010) (0.005) (0.004) (0.016)Income/Assets 0.022*** 0.025*** 0.003 0.010*** 0.142***

(0.004) (0.006) (0.008) (0.004) (0.018)L.Income/Assets 0.015*** 0.016*** 0.006 -0.002 0.096***

(0.003) (0.005) (0.005) (0.003) (0.016)Log Assets -0.001 0.002 -0.022* -0.010* -0.048**

(0.005) (0.010) (0.012) (0.006) (0.024)L.Log Assets 0.003 0.021** 0.024* 0.015** 0.074***

(0.005) (0.010) (0.012) (0.006) (0.024)Asset Growth Rate -0.001 0.011 0.012* 0.005 -0.004

(0.004) (0.007) (0.007) (0.003) (0.015)Market Cap./Assets 0.000 0.002 -0.009** -0.004* -0.032**

(0.003) (0.005) (0.003) (0.002) (0.013)L.Market Cap./Assets -0.001 0.009 0.003 0.003 0.032*

(0.004) (0.007) (0.004) (0.003) (0.017)Market-Book Ratio 0.002 0.005 0.006* 0.003 0.021*

(0.002) (0.004) (0.003) (0.002) (0.011)L.Market-Book Ratio -0.000 -0.010* -0.002 -0.002 -0.027**

(0.003) (0.006) (0.003) (0.002) (0.014)Cash/Assets 0.015 -0.009 -0.017* 0.003 0.066*

(0.011) (0.015) (0.010) (0.008) (0.035)L.Cash/Assets -0.008 -0.001 0.008 0.004 0.041

(0.010) (0.015) (0.010) (0.007) (0.034)Industry Trust Inceptions -0.001 0.002 -0.001 -0.000 -0.011***

(0.001) (0.002) (0.001) (0.001) (0.002)L.Industry Trust Inceptions 0.000 0.002 0.000 -0.001 -0.015***

(0.001) (0.002) (0.001) (0.001) (0.003)Year 0.002* -0.000 -0.001 -0.001* 0.006**

(0.001) (0.001) (0.001) (0.001) (0.002)N 6,965 6,965 6,965 6,965 6,965


Table 3.3: This table displays coefficients from linear probability models of regulardividend events (and repurchases, in the last column) on a post-reform dummy andaccounting controls. Each column also includes a set of dummy variables for first digitSIC. The ‘L.’ denotes the first lag of variable which it precedes. The sample is comprisedof all non-financial, non-utility companies, which never became income trusts, and are inthe top 587 firms by market capitalization in the particular year from 2002-2007. *, **,*** denote significance at 10%, 5% and 1%, respectively.

Initiation Increase Termination Decrease Any RepurchasePost-Reform Dummy -0.019* -0.026 -0.008 0.005 0.001

(0.010) (0.016) (0.007) (0.005) (0.024)Income/Assets 0.023*** 0.026*** 0.010 0.004 0.093***

(0.007) (0.010) (0.007) (0.004) (0.027)L.Income/Assets 0.018*** 0.012 0.006** 0.003 0.100***

(0.005) (0.007) (0.003) (0.003) (0.025)Log Assets -0.003 -0.010 -0.024 0.001 -0.023

(0.008) (0.015) (0.019) (0.005) (0.032)L.Log Assets 0.005 0.040*** 0.025 0.003 0.055*

(0.008) (0.015) (0.019) (0.005) (0.032)Asset Growth Rate 0.001 0.024* 0.013 0.001 -0.014

(0.006) (0.012) (0.010) (0.003) (0.018)Market Cap./Assets 0.006 0.016 -0.011** -0.008* -0.012

(0.007) (0.011) (0.005) (0.005) (0.023)L.Market Cap./Assets -0.002 0.012 -0.004 -0.004 0.051**

(0.007) (0.011) (0.006) (0.004) (0.023)Market-Book Ratio -0.002 -0.003 0.008** 0.007* 0.008

(0.005) (0.009) (0.004) (0.004) (0.020)L.Market-Book Ratio 0.000 -0.012 0.004 0.002 -0.045**

(0.006) (0.009) (0.005) (0.004) (0.020)Cash/Assets 0.011 -0.022 -0.018* 0.015 0.028

(0.015) (0.019) (0.011) (0.012) (0.041)L.Cash/Assets -0.012 -0.015 0.013 0.004 0.037

(0.013) (0.017) (0.010) (0.009) (0.041)Industry Trust Inceptions -0.000 0.002 0.001 -0.000 -0.005

(0.001) (0.002) (0.001) (0.001) (0.003)L.Industry Trust Inceptions 0.001 -0.002 0.001 -0.001 -0.009**

(0.002) (0.003) (0.001) (0.001) (0.004)Year 0.002 0.012*** 0.003 -0.002 0.004

(0.003) (0.004) (0.002) (0.002) (0.008)N 3,522 3,522 3,522 3,522 3,522


Table 3.4: This table displays means (with standard deviations in parentheses) forthe sample of 345 firms from 2002-2007, split by whether or not they have an ineligibleshareholder and whether or not they are inter-listed. Initation, increase, termination anddecrease refer to regular dividend events; dividend payer is the fraction of firm-years inthe group which paid any regular dividends.

Eligible Ineligible Not Inter-listed Inter-listedIncome/Assets -0.075 -0.022 -0.045 -0.052

(0.309) (0.217) (0.272) (0.243)Log Assets 5.65 5.58 5.30 6.48

(2.14) (1.75) (1.76) (2.16)Asset Growth Rate 0.310 0.251 0.274 0.290

(0.813) (0.687) (0.757) (0.726)Market Cap./Assets 2.017 1.57 1.67 2.07

(2.07) (1.63) (1.83) (1.91)Market-Book Ratio 2.686 2.18 2.29 2.76

(2.31) (1.76) (2.03) (2.07)Cash/Assets 0.192 0.171 0.176 0.193

(0.217) (0.203) (0.203) (0.228)Initiation 0.028 0.027 0.031 0.017

(0.166) (0.162) (0.175) (0.128)Increase 0.104 0.060 0.074 0.097

(0.305) (0.238) (0.262) (0.297)Termination 0.010 0.010 0.010 0.009

(0.102) (0.099) (0.102) (0.096)Decrease 0.009 0.006 0.007 0.011

(0.097) (0.079) (0.081) (0.105)Dividend Payer 0.307 0.279 0.292 0.290

(0.462) (0.449) (0.455) (0.454)N 954 1,116 1,526 544


Table 3.5: Each linear probability model also includes the set of accounting controlsand dummy variables for first digit SIC. The estimation sample covers 2002-2007 andincludes 345 firms per year, of which 46% are in the DTC eligible group and 26% are notinter-listed. *, **, *** denote significance at 10%, 5% and 1%, respectively.

Initiation Increase Termination DecreasePost X DTC Eligible 0.029* 0.025 -0.015** 0.000

(0.015) (0.025) (0.008) (0.008)Post X Not Inter-listed -0.008 -0.030 0.005 -0.001

(0.015) (0.030) (0.007) (0.010)DTC Eligible -0.002 0.033** 0.009 0.004

(0.009) (0.013) (0.006) (0.004)Not Inter-listed 0.019** 0.033** -0.001 0.001

(0.010) (0.015) (0.007) (0.006)Post-Reform Dummy -0.016 -0.008 -0.008 -0.000

(0.017) (0.033) (0.012) (0.011)N 2,070 2,070 2,070 2,070


Table 3.6: Each regression includes company fixed effects. The first column uses thefull sample, while the second and fourth columns use the short sample of 587 firms from2002-2007. The third column starts with the short sample and drops all firms which werein the top 10% of the sample by total assets in 2002 to demonstrate that the increasearound the reform was limited to the largest companies. The dependent variable in thefirst three columns is regular dividends, and in the fourth is share repurchases. The ‘L.’denotes the first lag of variable which it precedes. *, **, *** denote significance at 10%,5% and 1%, respectively.

Dividends Dividends Dividends RepurchasesPost-Reform Dummy 4.863*** 3.943** 0.413 2.469

(1.400) (1.530) (0.457) (7.357)Income 0.012* -0.008 0.011 0.029

(0.007) (0.015) (0.008) (0.043)L.Income 0.041*** 0.042*** 0.013** 0.008

(0.008) (0.016) (0.007) (0.045)Log Assets -2.888*** -5.082** -1.003** 1.161

(0.809) (2.018) (0.392) (10.301)L.Log Assets 0.760 -1.105 0.504 -15.053

(0.942) (1.536) (0.546) (10.557)Asset Growth Rate 1.261** 1.704* 0.608* -7.073

(0.561) (0.914) (0.334) (8.869)Market Capitalization 0.000 -0.001 0.001 -0.008

(0.001) (0.002) (0.001) (0.012)L.Market Capitalization 0.007*** 0.009*** -0.001 0.021

(0.002) (0.002) (0.002) (0.013)Market-Book Ratio -0.064 -0.490* -0.245** -0.350

(0.189) (0.287) (0.096) (1.190)L.Market-Book Ratio -0.652*** -1.016*** -0.246 -0.492

(0.167) (0.281) (0.154) (1.233)Cash 0.020 0.049 -0.001 -0.017

(0.018) (0.036) (0.007) (0.055)L.Cash -0.009 0.005 0.023 0.066

(0.012) (0.015) (0.019) (0.106)Industry Trust Inceptions -0.433* -0.574* -0.055 -0.085

(0.255) (0.295) (0.072) (1.282)L.Industry Trust Inceptions -0.170 -0.363 0.010 -1.164

(0.246) (0.347) (0.086) (1.041)Year 0.160 0.684 0.328** 5.937***

(0.221) (0.538) (0.132) (2.103)N 6,965 3,522 3,213 3,522


Table 3.7: Each regression includes company fixed effects. The sample in each caseconsists of 345 firms in each year from 2002-2007. The dependent variable in the firsttwo columns is regular dividends and in the third is share repurchases. *, **, *** denotesignificance at 10%, 5% and 1%, respectively.

Regular Dividends Regular Dividends RepurchasesPost X DTC Eligible 9.524 2.836 8.223

(6.458) (4.619) (14.347)Post X Not Inter-listed -5.177 -31.835** -156.250***

(5.399) (14.365) (45.902)Post X Not Majority Inter-listed - 33.537*** 137.651***

(12.211) (45.382)Post-Reform Dummy -0.542 -0.161 9.293

(2.601) (2.554) (9.757)N 2,070 2,070 2,070

Bibliography

David Aboody, Mary E. Barth, and Ron Kasznik. SFAS No. 123 stock-based compensa-

tion expense and equity market values. The Accounting Review, 79(2):251–275, 2004.

Kenneth R. Ahern. Bargaining power and industry dependence in mergers. Journal of

Financial Economics, 103(3):530–550, 2012.

Annette Alstadsæter and Erik Fjærli. Neutral taxation of shareholder income? Corporate

responses to an announced dividend tax. International Tax and Public Finance, 16(4):

571–604, 2009.

Gregor Andrade, Mark L. Mitchell, and Erik Stafford. New evidence and perspectives

on mergers. Journal of Economic Perspectives, 15(2):103–120, 2001.

Wiji Arulampalam, Michael P. Devereux, and Federica Liberini. Taxes and the location

of targets. Working paper, 2012.

Alan Auerbach and Kevin Hassett. On the marginal source of investment funds. Journal

of Public Economics, 87(1):205–232, 2003.

Alan J. Auerbach. Taxation and corporate financial policy. In A. J. Auerbach and

M. Feldstein, editors, Handbook of Public Economics.

Ilona Babenko and Yuri Tserlukevich. Analyzing the tax benefits from employee stock

options. The Journal of Finance, 64(4):1797–1825, 2009.

H. Kent Baker, Bin Chang, Shantanu Dutta, and Samir Saadi. Why firms do not pay

dividends: The Canadian experience. Journal of Business Finance & Accounting, 39

(9-10):1330–1356, 2012.

Malcolm Baker and Jeffrey Wurgler. A catering theory of dividends. The Journal of

Finance, 59(3):1125–1165, 2004.

108

BIBLIOGRAPHY 109

Michael J. Barclay, Clifford G. Holderness, and Dennis P. Sheehan. Dividends and cor-

porate shareholders. Review of Financial Studies, 22(6):2423–2455, 2009.

Lucian A. Bebchuk and Jesse M. Fried. Executive compensation as an agency problem.

Journal of Economic Perspectives, 71:92, 2003.

David A. Becher, J. Harold Mulherin, and Ralph A. Walkling. Sources of gains in

corporate mergers: Refined tests from a neglected industry. Journal of Financial and

Quantitative Analysis, 47(1):57–89, 2012.

Johannes Becker and Clemens Fuest. Taxing foreign profits with international mergers

and acquisitions. International Economic Review, 51(1):171–186, 2010.

Sandra Betton, B. Espen Eckbo, and Karin S. Thorburn. Merger negotiations and the

toehold puzzle. Journal of Financial Economics, 91(2):158–178, 2009.

Fischer Black and Myron Scholes. The pricing of options and corporate liabilities. The

Journal of Political Economy, pages 637–654, 1973.

Jennifer L. Blouin, Julie H. Collins, and Douglas A. Shackelford. Does acquisition by

non-U.S. shareholders cause U.S. firms to pay less tax? Journal of the American

Taxation Association, 27(1):25–38, 2005.

Jennifer L. Blouin, Jana S. Raedy, and Douglas A. Shackelford. Dividends, share re-

purchases, and tax clienteles: Evidence from the 2003 reductions in shareholder taxes.

The Accounting Review, 86(3):887–914, 2011.

Stephen R. Bond, Michael P. Devereux, and Alexander Klemm. The effects of dividend

taxes on equity prices: A re-examination of the 1997 UK tax reform. Working Papers

0701, Oxford University Centre for Business Taxation, 2007.

Laurence Booth. The dividend tax credit and Canadian ownership objectives. Canadian

Journal of Economics, pages 321–339, 1987.

Laurence D. Booth and David J. Johnston. The ex-dividend day behavior of Canadian

stock prices: Tax changes and clientele effects. The Journal of Finance, 39(2):457–476,

1984.

Michael J. Brennan. Taxes, market valuation, and financial policy. National Tax Journal,

3(3):417–429, 1970.

BIBLIOGRAPHY 110

John L. Campbell, James A. Chyz, Dan S. Dhaliwal, and William C. Schwartz Jr. Did the

2003 Tax Act increase capital investments by corporations? Journal of the American

Taxation Association, 2011.

Wenjie Chen. The effect of investor origin on firm performance: Domestic and foreign

direct investment in the United States. Journal of International Economics, 83(2):

219–228, 2011.

Raj Chetty and Emmanuel Saez. Dividend taxes and corporate behavior: Evidence from

the 2003 dividend tax cut. Quarterly Journal of Economics, 120(3):791–833, 2005.

Harry DeAngelo and Ronald W. Masulis. Optimal capital structure under corporate and

personal taxation. Journal of Financial Economics, 8(1):3–29, 1980.

Mihir A Desai and Dhammika Dharmapala. Dividend taxes and international portfolio

choice. Review of Economics and Statistics, 93(1):266–284, 2011.

Mihir A. Desai and Austan Goolsbee. Investment, overhang and tax policy. Brookings

Papers on Economic Activity, 2004(2):285–355, 2004.

Mihir A. Desai and James R. Hines. Expectations and expatriations: Tracing the causes

and consequences of corporate inversions. National Tax Journal, 55:409–440, 2002.

Mihir A. Desai and James R. Hines. Evaluating international tax reform. National Tax

Journal, 56:487–502, 2003.

Erik Devos, Palani-Rajan Kadapakkam, and Srinivasan Krishnamurthy. How do mergers

create value? A comparison of taxes, market power, and efficiency improvements as

explanations for synergies. Review of Financial Studies, 22(3):1179–1211, 2009.

Dan Dhaliwal, Linda Krull, and Oliver Zhen Li. Did the 2003 Tax Act reduce the cost

of equity capital? Journal of Accounting and Economics, 43(1):121–150, 2007.

Shantanu Dutta, Vijay M. Jog, and Samir Saadi. Re-examination of the ex-dividend day

behaviour of Canadian stock prices. Working paper, 2004.

Jesse Edgerton. Investment incentives and corporate tax asymmetries. Journal of Public

Economics, 94(11-12):936–952, December 2010.

Jesse Edgerton. Four facts about dividend payouts and the 2003 tax cut. International

Tax and Public Finance, pages 1–16, 2012.

BIBLIOGRAPHY 111

Merle Erickson. The effect of taxes on the structure of corporate acquisitions. Journal

of Accounting Research, 36(2):279–298, 1998.

Mara Faccio and Ronald W. Masulis. The choice of payment method in European mergers

and acquisitions. The Journal of Finance, 60(3):1345–1388, June 2005.

Eugene F. Fama and Kenneth R. French. Disappearing dividends: changing firm char-

acteristics or lower propensity to pay? Journal of Financial Economics, 60(1):3–43,

2001.

Carola Frydman and Dirk Jenter. CEO compensation. Annual Review of Financial

Economics, 2(1):75–102, 2010.

Carola Frydman and Raven S. Molloy. Does tax policy affect executive compensation?

Evidence from postwar tax reforms. Journal of Public Economics, 95(11):1425–1437,

2011.

Austan Goolsbee. What happens when you tax the rich? Evidence from executive

compensation. The Journal of Political Economy, 108(2):352–378, 2000.

Roger H. Gordon and A. Lans Bovenberg. Why is capital so immobile internationally?

Possible explanations and implications for capital income taxation. American Eco-

nomic Review, 86(5):1057–1075, 1996.

Roger H. Gordon and James R. Hines. International taxation. In A. J. Auerbach and

M. Feldstein, editors, Handbook of Public Economics, volume 4 of Handbook of Public

Economics, chapter 28, pages 1935–1995. Elsevier, 2002.

Gary Gorton, Matthias Kahl, and Richard J. Rosen. Eat or be eaten: a theory of mergers

and firm size. The Journal of Finance, 64(3):1291–1344, 2009.

Jon Gruber and Emmanuel Saez. The elasticity of taxable income: Evidence and impli-

cations. Journal of Public Economics, 84(1):1–32, 2002.

Harry Grubert, Timothy Goodspeed, and Deborah L. Swenson. Explaining the low tax-

able income of foreign-controlled companies in the United States. In Alberto Giovan-

nini, R. Glen Hubbard, and Joel Slemrod, editors, Studies in International Taxation.

University of Chicago Press, 1993.

Gustavo Grullon and Roni Michaely. Dividends, share repurchases, and the substitution

hypothesis. The Journal of Finance, 57(4):1649–1684, 2002.

BIBLIOGRAPHY 112

Brian J. Hall and Jeffrey B. Liebman. The taxation of executive compensation. In Tax

Policy and the Economy, Volume 14, pages 1–44. MIT Press, 2000.

Brian J. Hall and Kevin J. Murphy. Stock options for undiversified executives. Journal

of Accounting and Economics, 33(1):3–42, 2002.

Jarrod Harford. What drives merger waves? Journal of Financial Economics, 77(3):

529–560, 2005.

Carla Hayn. Tax attributes as deteminants of shareholder gains in corporate acquisitions.

Journal of Financial Economics, 23(1):121–153, 1989.

Keith Head and John Ries. FDI as an outcome of the market for corporate control:

Theory and evidence. Journal of International Economics, 74(1):2–20, 2008.

Chip Heath, Steven Huddart, and Mark Lang. Psychological factors and stock option

exercise. Quarterly Journal of Economics, 114(2):601–627, 1999.

James R. Hines. Altered states: Taxes and the location of foreign direct investment in

America. American Economic Review, 86(5):1076–1094, 1996.

James R. Hines and Eric M. Rice. Fiscal paradise: Foreign tax havens and American

business. Quarterly Journal of Economics, 109(1):149–182, 1994.

Gerard Hoberg and Nagpurnanand R. Prabhala. Disappearing dividends, catering, and

risk. Review of Financial Studies, 22(1):79–116, 2009.

Christopher L. House and Matthew D. Shapiro. Temporary investment incentives: The-

ory with evidence from bonus depreciation. American Economic Review, 98(3):737–

768, 2008.

Harry P. Huizinga and Johannes Voget. International taxation and the direction and

volume of cross-border M&As. Journal of Finance, 64:1217–1249, 2009.

Marcus Jacob and Martin Jacob. Taxation, dividends and share repurchases: Taking

evidence global. Journal of Financial and Quantitative Analysis, Forthcoming, 2012.

Michael C. Jensen. Agency costs of free cash flow, corporate finance, and takeovers. The

American Economic Review, 76(2):323–329, 1986.

Steven Kaplan. Management buyouts: Evidence on taxes as a source of value. The

Journal of Finance, 44(3):611–632, July 1989.

BIBLIOGRAPHY 113

E. Han Kim and Yao Lu. CEO ownership, external governance, and risk-taking. Journal

of Financial Economics, 102(2):272–292, 2011.

Kenneth J. Klassen and Amin Mawani. The impact of financial and tax reporting incen-

tives on option grants to Canadian CEOs. Contemporary Accounting Research, 17(2):

227–262, 2000.

Maher Kooli and Jean-Francois L’Her. Dividends versus share repurchases evidence from

Canada: 1985–2003. Financial Review, 45(1):57–81, 2010.

Jason Kotter and Ugur Lel. Friends or foes? Target selection decision of sovereign wealth

funds and their consequences. Journal of Financial Economics, 101(2):360–381, 2011.

Leming Lin and Mark J. Flannery. Do personal taxes affect capital structure? Evidence

from the 2003 tax cut. Journal of Financial Economics, 2013.

Kevin S. Markle and Douglas A. Shackelford. Cross-country comparisons of corporate

income taxes. National Tax Journal, 65(3):493–528, 2012.

Steve Matsunaga, Terry Shevlin, and Dee Shores. Disqualifying dispositions of incentive

stock options: Tax benefits versus financial reporting costs. Journal of Accounting

Research, 30:37–68, 1992.

Steven R. Matsunaga. The effects of financial reporting costs on the use of employee

stock options. The Accounting Review, pages 1–26, 1995.

Amin Mawani. Cancellation of executive stock options: Tax and accounting income

considerations. Contemporary Accounting Research, 20(3):495–517, 2003.

Kenneth J. McKenzie. Income taxes, integration and income trusts. Canadian Tax

Journal, 54(3):633–656, 2006.

Kenneth J. McKenzie and Aileen J. Thompson. Dividend taxation and equity value: the

Canadian tax changes of 1986. Canadian Journal of Economics, 28(2):463–472, 1995.

Kenneth J. McKenzie and Aileen J. Thompson. The economic effects of dividend taxa-

tion. Working Paper 7, Technical Committee on Business Taxation, December 1996.

Merton H. Miller. Debt and taxes. The Journal of Finance, 32(32):261–275, 1977.

Merton H. Miller and Kevin Rock. Dividend policy under asymmetric information. The

Journal of Finance, 40(4):1031–1051, 1985.

BIBLIOGRAPHY 114

Jack Mintz and Stephen R. Richardson. Income trusts and integration of business and

investor taxes: A policy analysis and proposal. Canadian Tax Journal, 54(2):359–406,

2006.

Kevin J. Murphy. Reporting choice and the 1992 proxy disclosure rules. Journal of

Accounting, Auditing & Finance, 11(3):497–515, 1996.

Kevin J. Murphy. Executive compensation. Handbook of Labor Economics, 3:2485–2563,

1999.

Jeffry Netter, Mike Stegemoller, and M. Babajide Wintoki. Implications of data screens

on merger and acquisition analysis: A large sample study of mergers and acquisitions

from 1992 to 2009. Review of Financial Studies, 24(7):2316–2357, 2011.

Volker Nocke and Stephen Yeaple. Cross-border mergers and acquisitions vs. greenfield

foreign direct investment: The role of firm heterogeneity. Journal of International

Economics, 72(2):336–365, 2007.

Thomas Piketty, Emmanuel Saez, and Stefanie Stantcheva. Optimal taxation of top

labor incomes: A tale of three elasticities. Working Paper 17616, National Bureau of

Economic Research, November 2011.

James Poterba. Taxation and corporate payout policy. American Economic Review, 94

(2):171–175, 2004.

PricewaterhouseCoopers. RE: 2010 Federal Budget Employee Stock Option Proposals,

May 2010. URL http://pwc.to/1455lF2.

Matthew Rhodes-Kropf and David T. Robinson. The market for mergers and the bound-

aries of the firm. The Journal of Finance, 63(3):1169–1211, 2008.

Nancy L. Rose and Catherine Wolfram. Has the “million-dollar cap” affected CEO pay?

American Economic Review, 90(2):197–202, 2000.

Nancy L. Rose and Catherine Wolfram. Regulating executive pay: Using the tax code

to influence chief executive officer compensation. Journal of Labor Economics, 20(S2):

S138–S175, 2002.

Pavel G. Savor and Qi Lu. Do stock mergers create value for acquirers? The Journal of

Finance, 64(3):1061–1097, 2009.

BIBLIOGRAPHY 115

Myron S. Scholes and Mark A. Wolfson. The effects of changes in tax laws on corporate

reorganization activity. Journal of Business, 63 (Part 2):S141–S164, 1990.

Jim A. Seida and William F. Wempe. Effective tax rate changes and earnings stripping

following corporate inversion. National Tax Journal, 57(4):805–828, 2004.

Andrei Shleifer and Robert W. Vishny. Stock market driven acquisitions. Journal of


Mary-Anne Sillamaa and Michael R. Veall. The effect of marginal tax rates on tax-

able income: A panel study of the 1988 tax flattening in Canada. Journal of Public

Economics, 80(3):341–356, 2001.

Deborah L. Swenson. The impact of U.S. tax reform on foreign direct investment in the

United States. Journal of Public Economics, 54:243–266, 1994.

Stijn van Nieuwerburgh and Laura Veldkamp. Information immobility and the home bias

puzzle. The Journal of Finance, 64(3):1187–1215, 2009.

David Yermack. Do corporations award CEO stock options effectively? Journal of


Tao Zeng. Intercorporate ownership, taxes, and corporate payout choices. Accounting

Perspectives, 10(4):265–283, 2011.

Xianming Zhou. Executive compensation and managerial incentives: A comparison be-

tween Canada and the United States. Journal of Corporate Finance, 5(3):277–301,

1999.

Xianming Zhou. CEO pay, firm size, and corporate performance: Evidence from Canada.

Canadian Journal of Economics, 33(1):213–251, 2000.

taxation and financial decision making€¦ · taxation and financial decision making andrew bird...

Documents