Tax Shelters and Corporate Debt Policy
John R. Graham
Professor of Finance
Fuqua School of Business
Duke University
Alan L. Tucker
Associate Professor of Finance
Lubin School of Business
Pace University
September 12, 2005
Abstract:
We gather a unique sample of 44 tax shelter cases to investigate the magnitude of tax
shelter activity and whether participating in a shelter is related to corporate debt policy.
The average deduction produced by the shelters in our sample is very large, equaling
approximately nine percent of asset value. These deductions are more than three times as
large as interest deductions for comparable companies. The firms in our sample use less
debt when they engage in tax sheltering. Compared to companies with similar pre-shelter
debt ratios, the debt ratios of firms engaged in tax shelters fall by about 8%. The tax
shelter firms in our sample appear underlevered if shelters are ignored but do not appear
underlevered once shelters are considered.
Keywords: Taxes, tax shelters, debt, capital structure
JEL Classifications: G32, K34
We thank an anonymous referee, Joseph Bankman, Alon Brav, Stefano Della Vigna, Mihir Desai, Lily
Fang, Wayne Ferson, Vidhan Goyal, Tom Keller, Mark Leary, Bob McDonald, Guy McDonough, Avri
Ravid, Michael Roberts, Chip Ryan, Bill Schwert, and participants in seminars at Boston College, the
Chicago Fed, DePaul, Harvard Business School, Louisiana State, the 2005 NBER Corporate Finance
Summer Institute, the Securities and Financial Markets conference, the UNC tax conference, York
University, and the 2005 WFA annual conference for helpful comments. Brian An, Si Li, Bin Wei, and
Julia Wu provided excellent research assistance. All errors are our own. Corresponding author contact
information: john.graham@duke.edu, phone (919) 660-7857, fax (919) 660-8038, or John R. Graham,
Fuqua School of Business, Duke University, Durham NC 27708-1020.
Corporate tax shelters are our number one problem (in enforcing the tax laws), not just
because they cost money but because they breed disrespect for the tax system
- Then Treasury Secretary Lawrence Summers quoted in “Corporations’ Taxes are Falling Even as
Individuals’ Burden Rises,” The New York Times, February 20, 2000.
The tax department (of a corporation) is viewed … as a profit center and a place that has
… an obligation … to aggressively reduce the tax burden.
- Larry Langdon, commissioner of the IRS large and midsize business division, and former tax
director at Hewlett-Packard, quoted in the same New York Times article.
1. Introduction
During both 1991 and 1992 Compaq Computer Corporation reported taxable income that
averaged more than $170 million. According to public financial statements, during these
same years Compaq’s debt averaged 1.3 percent of total assets at a time when
comparable firm debt ratios averaged 25 percent. Based on its apparently low debt ratio,
Compaq appeared to leave money on the table in terms of paying more taxes than
necessary. If the company had levered up to a 25 percent debt-to-assets ratio, the
incremental debt would have produced interest deductions worth approximately $65
million annually (assuming a nine percent coupon rate, which was the average for newly
issued investment grade corporate debt in the early 1990s). If Compaq’s federal tax rate
was the maximum 35%, the firm could have saved $23 million annually in federal taxes
(not to mention state and local taxes). Apparent bypassing of such large tax benefits has
led some researchers to argue that many firms appear to be under-levered (e.g., Miller
1977, Graham 2000).
One potential problem with the above argument is that apparently under-levered firms
may have “off balance sheet” tax deductions that are not easily observable, and which are
therefore often ignored in empirical analyses. For example, Graham, Lang, and
Shackleford (2004) document that debt policy at S&P 100 and Nasdaq 100 firms appears
to be conservative if stock option deductions are ignored but is notably less conservative
once option effects are considered. Stefanescu (2005) finds a similar effect for firms that
use defined benefit pension plans. In this paper, we investigate the magnitude of
corporate tax shelters, and also whether the use of tax shelters is related to corporate debt
policy.
Compaq is one of the tax shelter firms in our sample. In the case of Compaq, the
government alleges that the firm used cross-border dividend capture and transfer pricing
(described in Section 2) to produce at least $115 million in annual tax deductions in 1991
and 1992. That is, Compaq used tax shelters to produce tax deductions that are nearly
twice as large as the interest deductions that the firm appeared to forgo by maintaining a
debt ratio lower than at comparable firms (i.e., the $65 million in interest deductions
mentioned above). Said differently, if one were to factor in all deductions, including both
legal deductions that appear on financial statements and tax shelter deductions, Compaq
does not appear to be underlevered.
1
Given the sharp decline in corporate tax payments in the past decade, the possibility
exists that tax sheltering is a widespread and growing problem. For example, S&P 500
firms paid federal taxes of 29 cents per dollar of reported profits in 1994, but this fell to
just 18 cents per dollar one decade later (see Fig. 1). A similar declining effective tax rate
pattern also holds for all publicly traded firms. (The main exceptions to the declining tax
rate are the increase that occurred in 1993 when the maximum statutory corporate tax rate
increased from 34% to 35%, and at the end of the sample period). Corporate tax
sheltering activity may play an important role in this reduction in tax collections.
Current estimates indicate that sheltering allows U.S. firms to avoid over $10 billion in
federal income taxes annually.1 Recently the IRS claimed that one firm alone,
GlaxoSmithKline P.L.C., owes $5.2 billion in back taxes and penalties related to a
transfer pricing strategy dating back to 1989.2 Tax sheltering activity has also allegedly
led to a significant reduction in state tax collections. The Multistate Tax Commission
reports that state corporate income tax revenue, which totaled $35.4 billion in 2001,
would have been one-third larger had tax sheltering not occurred.3
Information about tax shelters is notoriously hard to find. Companies do not publicize
their use of shelters and the Internal Revenue Service (IRS) treats tax investigations
confidentially. Based on an exhaustive search of Tax Court records and financial news
stories, we identify 44 tax-sheltering cases (involving 43 firms) between 1975 and 2000.
To the best of our knowledge, our sample of tax sheltering cases is the largest collected to
date. Because the firms in our sample were caught and/or filed suit regarding their shelter
activity, there are sample selection issues related to our sample (discussed more fully in
Section 3.2). One issue is that our sample selection procedure might identify primarily
large tax shelter cases. Another issue is that the firms in our sample might have taken
specific actions, such as reducing debt ratios or filing suit against the government, that
led to their shelter being detected by the government and by our sample selection process.
To the extent that this is true, it hinders our ability to conclude that the characteristics of
firms in our sample are representative of other shelter firms. These issues potentially
make our results difficult to directly generalize to the population of firms, and we
interpret the results accordingly.
The tax shelters in our sample are large. The median shelter produces, on an annual basis,
a deduction sufficient to shield income equal to approximately nine percent of asset
value. In comparison, a debt-to-assets ratio of 30 percent produces interest deductions
equal to only three percent of assets if coupon rates are 10 percent.
We also investigate whether companies appear to substitute between the tax shields
1
See Bankman (1999). Moreover, recently California offered a one-time tax amnesty that led to
corporations reporting and extinguishing $30 billion in shelters, suggesting that aggregate sheltering at the
national level is much larger still. See also Bankman and Simmons (2003).
2
See "IRS says Glaxo owes $5.2 billion in taxes, interest," Philadelphia Inquirer, January 8, 2004.
3
See "Study: Corporate Tax Sheltering Linked to as Much as $12.4 Billion in Lost State Tax Revenues,"
PRNewswire, Washington, July 15, 2003.
2
provided by shelters and debt interest. As predicted by DeAngelo and Masulis (1980), we
document that firms use less debt when their non-debt tax shields (in this case,
deductions from tax shelters) are large. In particular, in the year(s) that the tax shelters
are in use, sheltering firm debt ratios are more than 800 basis points lower than the debt
ratios of similar-sized same-industry firms. This is true even though several years before
the shelter is implemented, shelter firm debt ratios are indistinguishable from matchedfirm debt ratios.
We also investigate whether the existence of tax shelters affects incremental financing
decisions. We find that 60 percent of the shelter firms in our sample issue debt sometime
during the years preceding the inception of tax shelter activity. In contrast, 70 percent of
matched firms issue debt. Holding all else constant in a logistic regression, shelter firms
are significantly less likely to issue debt than are non-shelter firms.
Tax shelters potentially affect more than just debt ratios. Stock prices can also be
affected. McGill and Outslay (2004, p. 751) summarize Tyco International Ltd’s belief
that by reducing its tax rate by 700 basis points via offshore activity, it increased its EPS,
which in turn increased its market capitalization by nearly $5 billion. Moreover, if tax
shelters substitute for debt-induced tax deductions, sheltering may also increase financial
slack, reduce expected bankruptcy costs, enhance credit quality, reduce the risk of
covenant violation, and reduce the cost of debt. For example, in our sample in the years
leading up to the inception of the tax shelter, we find that shelter firm credit ratings
improve one notch relative to matched firms, most likely due to falling debt ratios. Of
course, all of this must be weighed against the risk, potential penalties, and other negative
effects of engaging in tax shelters.
Our paper is related to other research that investigates tax shelters. Desai (2003)
investigates whether the growing wedge between taxable income reported on financial
statements and tax return income can be explained by accelerated depreciation, stock
options, or earnings management. He concludes that these traditional vehicles explain
only a portion of the wedge. Desai argues (p. 1) that new "enhanced opportunities for
avoiding and evading taxes through cheaper, more sophisticated, and less transparent
mechanisms" (i.e., tax shelters) explain at least one-third of the book-tax income gap as
of 1998. Schallheim and Wells (2004) measure nondebt tax shields based on the
difference between taxes paid and financial statement tax expense in an attempt to
capture the effect of "off financial statement" deductions such as accelerated
depreciation, stock option deductions, tax shelters, and the like. In contrast to often-found
results that are based on traditional measures of nondebt tax shields (NDTS), Schallheim
and Wells find that tax spread is negatively related to debt usage.
One advantage our approach has over Desai’s (2003) and Schallheim and Wells’ (2004)
is that we know actual sheltering activity (as alleged or proven by the government) and
hence do not need to indirectly infer the propensity of sheltering. The accompanying
disadvantage to our approach is that our sample is relatively small due to the difficulty of
identifying firm-specific instances of sheltering.
3
Our analysis is also related to papers that study whether observed debt ratios are too low
(e.g., Miller (1977), Parrino and Weisbach (1999), and Graham (2000)).4 For example,
because the ratio of interest deductions to expected income is quite small at many firms,
Graham (2000) argues that the average magnitude of debt usage appears to be small
relative to the tax benefits of debt. Underleverage appears to be severe at some firms that
on the surface appear to have a low marginal cost of debt (e.g., Compaq, AHP,
Microsoft).
Our analysis complements these papers, and for the firms in our sample offers a partial
solution to both the magnitude and cross-sectional “under-leverage” issues raised by
Graham (2000). In terms of magnitude, the shelters in our sample are very large, at least
three times what would be expected to be generated by debt interest deductions. Crosssectionally, as shown in Table 1, many of the firms in our sample appear to be “low cost
of debt” firms (that appear to use too little debt when shelters are ignored). Therefore,
debt policy at these firms is not as conservative as appears on the surface, due to plentiful
tax shelter deductions that reduce the need for debt. We do not claim, however, that
incorporating the effects of tax shelters eliminates the possibility that some firms are
underlevered. For one thing, we investigate a select sample of primarily large firms, so
our results do not directly address debt policy at small firms. For another, our analysis
does not address whether firms are underlevered in the years before or after our sample
period.
Finally, our paper is related to other works that investigate tax shelters, but at the macro
level. Clausing (2003) and Bartelsman and Beetsma (2003) find substantial evidence of
tax-motivated transfer pricing, which is one of the most popular tax shelters in our
sample. Hines (1997) reviews literature that finds indirect evidence of transfer pricing.
Finally, Desai, Foley, and Hines (2004) find evidence that multinationals relocate income
to tax haven countries, in part to delay repatriation to high tax rate parents.
Bankman (1999) provocatively argues that, as of the end of the 1990s, the tax shelter
industry was “growing at breakneck speed” and was valued at least in the tens of billions
of dollars. He reasons that even though many shelters will be struck down by the courts if
litigated, the dollar benefit of participating in a shelter far outweighs the cost, once one
incorporates the low probability of detection. Bankman cites at least two factors that have
contributed to increased sheltering activity: 1) the relative ease with which shelter
promoters can obtain favorable opinion letters from attorneys, which generally protect
corporations from penalties if the shelter is ruled illegal, and 2) business norms that have
weakened in the face of the attractive economics of sheltering income. While our sample
represents only the tip of the shelter iceberg described by Bankman, we believe that our
analysis sheds light on tax shelter participation and its effect on corporate policies.
4
Another group of papers argues that observed debt ratios are in fact not too low (e.g., Green and Hollifield
(2003), Hennesey and Whited (2004), Ju, Parrino, Poteshman, and Weisbach (2004)).
4
The rest of the paper proceeds as follows. Section 2 provides details about tax shelters
including the legal doctrines the government uses to combat tax evasion. It also lays out
the debt substitution hypothesis. Section 3 describes our sample and presents summary
statistics. Section 4 investigates how tax shelters affect the corporate use of debt and
Section 5 concludes.
2. Classifying and combating corporate tax shelters5
The U.S. Congress (Joint Committee on Taxation, 1999) defines a tax shelter as an
endeavor principally designed to avoid taxation without exposure to economic risk or
loss. Bankman (2003) argues that tax shelters are tax-motivated vehicles that use a literal
interpretation of government statute or regulation to misstate economic income in a
manner that is inconsistent with the spirit or intent of the statute or regulation. While
Bankman focuses on vehicles that are promoted by outside parties such as investment
banks, we also investigate activity that is more likely to be self-motivated, such as nonarms length transfer pricing. (More details on sample selection are provided in Section 3.)
Section 2.1 provides the context for our debt analysis by describing the hypothesis that
deductions from tax shelters can substitute for debt interest deductions. Section 2.2
sketches background information about the tax shelters used by firms in our sample.
Numerous other tax shelter vehicles exist but are not discussed due to space constraints.
Section 2.3 describes the key legal doctrines that the government uses as it attempts to
detect and eliminate tax shelters.
2.1. Tax shelters and the use of corporate debt
The tax management department at most modern corporations has many tools at its
disposal to reduce tax obligations (Scholes et al. 2002). One common feature in most of
the tax shelters described in Section 2.2 is that they effectively produce deductions that
can be used to offset income or gains. Therefore, under the right circumstances, each tax
shelter can be thought of as a separate lever that a corporate tax planner can pull to
reduce tax obligations in a given year.
In the spirit of DeAngelo and Masulis (1980), deductions produced by tax shelters are
non-debt tax shields (NDTS). DeAngelo and Masulis show that NDTS substitute for
interest tax deductions. Each firm has an optimal amount of total deductions, and if a firm
uses more NDTS it will use fewer debt tax deductions. One focus of our paper is to
investigate empirically how tax shelter deductions fit into the corporate debt decisionmaking process. Our hypothesis is that tax shelters substitute for the use of debt.
5
Information for this section is principally obtained from: (1) Joint Committee on Taxation, "Background
and Present Law Relating to Tax Shelters (JCX-19-02)", March 19, 2002; (2) "The Hustling of X-Rated
Tax Shelters," Forbes, December 14, 1998; (3) "The Problem of Corporate Tax Shelters: Discussion,
Analysis and Legislative Proposals," Department of the Treasury, July 1999; and (4) expert witness
involvement by one of the authors.
5
We do not argue that firms engage in (potentially illegal) tax shelters for the sole purpose
of reducing their use of debt. Rather, our perspective is that if a firm engages in tax
sheltering for any reason, to the extent that the shelters reduce taxable income, the firm
will use less debt financing. We also acknowledge that, as in most corporate finance
research, it is difficult to unambiguously prove direction of causality or order of
sequential choice. That is, a firm might use less debt after having first established tax
shelters, or it might resort to sheltering after discovering that it is unable to issue much
debt (for whatever reason). Therefore, our results should be interpreted as documenting
correlations, not causality.
In the cleanest test of the substitution hypothesis, tax shelters would lead to deductions
that enter the tax return in the exact same manner as interest deductions, that is, by
leading to a deduction that reduces taxable income (or equivalently by reducing revenues,
which also leads to less taxable income). Many of the shelters described below lead to
interest-like deductions; however, some lead to capital losses. Even in the case of capital
losses, shelter deductions can be transformed into interest-like deductions (e.g., see
Bankman (1999) for a discussion of a section 988 transaction or Cetta (2002) for other
techniques to convert capital losses into ordinary losses). Nonetheless, to study the
cleanest possible sample at one point, in some of our empirical tests we delete shelters
that produce paper capital losses.
2.2. Tax shelters used by firms in our sample
This section provides thumbnail sketches of the tax shelters used by the firms in our
sample. We note whether a shelter leads to interest-like deductions or capital losses, and
whether the shelter implementation has a direct effect on the debt ratio. For interested
readers, the references listed in footnote 5 and in the text below provide more detail about
the mechanics of tax shelters.
2.2.1. Lease-in, lease out (LILO)
Lease-in, lease-out transactions were popular from 1995 to 1999. In a typical deal, a U.S.
corporation leases long-lived property (e.g., a power plant) from a tax-indifferent party
(often a foreign municipality) and immediately subleases the property back to the same
party. The initial lease calls for the U.S. corporation to prepay the bulk of its rental
obligations. The U.S. company enjoys accelerated deductions on its tax return, while
amortizing the cost over a much longer horizon in financial statements. The accelerated
deductions produce a "time value" benefit vis-à-vis the (often large) net income the U.S.
firm realizes in the final year of the deal, when the transaction is reversed. (In some
cases, taxpayers attempt to generate paper losses through other shelters to offset the final
year income.) LILO transactions shelter ordinary income and are therefore a substitute
for interest expense. Revenue Ruling 99-14 and related 1999 regulations eliminate LILO
tax benefits.6
6
For more on the maturation of LILO deals, and closely related sale-in, lease-out (SILO) deals, see
6
American Banker reports that the IRS has identified 56 corporations that engaged in
LILO transactions, including a few of the largest banks in the U.S.: AmSouth Bankcorp,
FleetBoston Financial Corp., and BB&T Corp.7 These three banks also disclose in their
SEC filings that the IRS sent them a Notice of Deficiency, and we include them in our
sample. (A Notice of Deficiency denies the tax treatment sought in these deals,
effectively stating the amount of additional taxes the IRS believes that the firm must pay.
After receiving the Notice, the taxpayer must sue the government if it wants to try to
preserve the tax treatment that it sought in its initial tax filing.) As described in the
sample selection section below, we do not include firms that are reputed to have
participated in a tax shelter if we can not find evidence of a Notice of Deficiency.
2.2.2. Transfer pricing (TP)
In a typical transfer pricing deal, a U.S. corporation produces an asset at a low-tax foreign
subsidiary and said asset is sold to the parent company at an above market price. The
transfer is not arms-length because the price does not reflect the asset's value. Tax
sheltering transfer pricing has the effect of subjecting most of the profit from the ultimate
sale of the asset to the relatively low subsidiary tax rate. Because the parent company’s
taxable income is lower than it would be with arms-length transfer pricing, the firm has
less need for deductions from debt income. Hence, TP deals are indirect substitutes for
interest expense because they serve to reduce ordinary income. We hypothesize that, all
else equal, a firm engaged in income shifting via transfer pricing will use less debt. For
analysis of optimal transfer pricing, see Baldenius, Melumad, and Reichelstein (2004).
2.2.3. Corporate-owned life insurance (COLI)
In a typical COLI deal, the corporation purchases cash value life insurance on its
employees and borrows to pay some or all of the premiums. As time passes and a given
policy approaches payout, the value of the policy increases, as does the amount that can
be borrowed on that value. The eventual payout from the policy is not taxable, while the
interest on the borrowings is fully deductible. Due to interest and fees, a COLI deal yields
little if any before-tax value but produces positive net present value once tax benefits are
considered. See Bankman (2003) for details.
COLIs shelter ordinary income by producing interest deductions; therefore, they
substitute for non-shelter-related interest deductions. We give special attention to COLI
shelters in our empirical analysis. COLIs could increase a firm’s debt usage if they
effectively lead to additional borrowing (i.e., above “normal” borrowing) to finance the
insurance premiums, and therefore could be positively correlated with a company’s debt
ratio. In contrast, if companies that use a given tax shelter (e.g., a COLI) are also likely to
Luitjens (2004). Many of the hard assets of the nation’s transit system have been ostensibly purchased by a
taxpayer and leased back to the transit system under SILO arrangements.
7
See "A New Tax Battle for Big Banks: LILO Deals," American Banker, Vol. 168, Issue 165, August 27,
2003.
7
use other shelters and non-debt tax shields, COLI use could be negatively correlated with
the use of debt.8 Or these two effects could offset each other. Due to this ambiguity, at
one point in our empirical analysis, we exclude COLI firms.
In 1993, Winn-Dixie entered a COLI program to insure approximately 36,000 employees.
Winn-Dixie purchased whole-life policies and was the sole beneficiary. The company
regularly borrowed against the policy values. The all-in borrowing costs exceeded the net
cash surrender value and benefits paid on the policies. Winn-Dixie lost money on a
pretax basis each year. However, the tax deductions on the interest and fees produced
several billion dollars of tax benefits. Winn-Dixie terminated its program in 1997, after
1996 tax law changes limited the deductibility of interest on COLI policy loans.9
2.2.4. Cross-border dividend capture (CBDC)
In this strategy, a taxable entity purchases American Depository Receipts (ADRs) cum
dividend, capturing the dividend and the foreign withholding tax credit. Through a cross
trade, the entity then quickly resells the ADRs at the ex-dividend and withholding tax
adjusted price. The loss on the cross trade offsets the dividend to be collected by the
taxable entity, ignoring trading costs and sponsor fees. The tax credit is used to reduce
tax obligations. The sale and repurchase of the ADRs are done through an intermediary.
But ultimately the real seller is a tax-exempt institution such as an international mutual or
pension fund that can not make use of foreign tax credits. They are compensated by short
interest rebates on their ADR lending activity. In short, this strategy creates a secondary
marketplace wherein tax exempt institutions sell their unusable tax credits in return for
interest payments. (See Tucker (2002) for more details.) Graham (2003) demonstrates
that, all else equal, the tax incentive to finance with debt decreases with foreign tax
credits such as those described in this deal.
The appendix contains a case study of a CBDC involving Compaq. In deposition
testimony in the Compaq case, a single sponsor of this strategy indicated that it had
conducted hundreds of these deals during the early and mid 1990s (Tucker, 2002).
Changes to the Internal Revenue Code (IRC) made in 1997 were intended to preclude this
type of transaction from being conducted.
2.2.5. Contingent-payment installment sales (CPIS)
In a typical deal, a U.S. corporation having a sizable capital gain establishes an offshore
partnership with a tax-exempt foreign entity. The foreign entity is the majority partner.
The partnership engages in a series of securities transactions involving the sale of
8
Some anecdotal evidence is consistent with this possibility. Internal KPMG documents advise partners to
market aggressive new tax shelters to firms that have engaged in risky tax shelters in the past (The Wall
Street Journal, June 16, 2004, “KPMG Shelter Shaved $1.7 Billion Off Taxes of 29 Large Companies”).
9
See "Health Insurance Portability and Accountability Act of 1996," Pub. L. 104-191, sec. 510, 110 Stat.
2090 (1996).
8
existing assets in exchange for contingent-payment financial securities. Under (then
existing) IRC regulations pertaining to the partnership tax treatment of contingent
payment assets, this action had the claimed effect of producing a sizeable paper capital
gain for the partnership. The majority of said gain went to the tax-exempt majority
partner. Then the U.S. partner would buy out most of the foreign partner, leaving the U.S.
partner with the vast majority of the partnership. Later, when the contingent payment
securities were sold, a back-end paper capital loss was created, which was subsumed into
the U.S. parent corporation's financials for tax calculation purposes. Thus the preexisting
real capital gain was offset by a paper capital loss.
Often these deals were managed such that the credit and market risks occasioned by the
securities trades were hedged, so that any losses suffered by the partnership were shared
primarily by the domestic partner/US taxpayer, leaving the accommodating foreign entity
with little down-side risk. The temporary regulation upon which this strategy was based
has been changed so this type of deal is now precluded. See Bankman (1999) for more on
CPIS deals, which he refers to as “high-basis low-value” tax shelters.
CPIS offset capital gains and therefore are not a direct substitute for interest expense (i.e.,
they do not directly offset ordinary income). However, Bankman (1999) and Cetta (2002)
explain how a capital loss can be transformed into an ordinary loss. The appendix
contains a case study of a CPIS involving Colgate Palmolive.
2.2.6. Liquidation, recontribution (LR)
In a LR, a foreign partnership is established involving tax-exempt foreign entities. This
partnership engages in a series of securities trades that involve an asset or liability whose
future value is uncertain, for example, an open short position in a Treasury bill. Soon
thereafter, a U.S. corporation buys the vast majority of the partnership, serving to
liquidate the partnership, and then immediately reconstitutes a new partnership. By
attempting to make use of partnership regulations that speak to the computation and
allocation of a partnership's bases in its assets and liabilities, this liquidation and recontribution strategy can produce a targeted capital gain or loss for the U.S. firm. Like
CPIS deals, LR deals address capital income, not ordinary income, and are therefore not a
direct substitute for interest expense. Again, though, a converting transaction can
transform the capital loss into an ordinary loss. The appendix contains an example of a
LR deal involving Florida Power and Light. For more on LR deals, see Bird and Tucker
(2002).
2.2.7. Offshore intellectual property havens (OIPH)
U.S. multinationals have incentive to house intellectual property abroad so as to shelter
income from overseas sales.10 For example, a multinational may transfer a patent to a
newly formed Bermuda subsidiary so that royalties from sales of products made outside
the U.S. flow to the subsidiary, where they accumulate tax-free. The subsidiary pays for
10
"A new twist in tax avoidance: Firms send best ideas abroad," The Wall Street Journal, June 24, 2002.
9
part of the patent but the price is allegedly quite subjective. While royalties collected by
the subsidiary need to be reported to the IRS, the payments transferred back to the parent,
and subject to U.S. taxation, are often artificially low.11 OIPH deals therefore are akin to
transfer pricing in that they reduce revenue, which in turn reduces the marginal value of
interest deductions.
The American Jobs Creation Act, signed into law October 2, 2004, reduces for about one
year the tax rate on repatriated foreign profits (for profits accumulated prior to June 30,
2003). The tax rate on repatriated profits is temporarily reduced to 5.25% from its
standard level of 35%. Such a cut permits profits that have accumulated in the foreign
subsidiaries via OIPHs to be repatriated at favorable tax rates.
2.2.8. Contested liability acceleration strategy (CLAS)
In this strategy a firm establishes a “contested liability trust” with itself as beneficiary.
The firm transfers non-cash assets (e.g., an intracompany note or IOU) into the trust
equal to approximately what the firm expects to pay to resolve claims it is still contesting.
These claims may be related to liabilities including medical malpractice, shareholder
lawsuits, personal injury, or environmental actions. The firm receives deductions equal to
the amount of assets placed in the trust, thereby reducing taxable income. (The IRS
contends that CLAS deductions are invalid because the firm maintains control of the trust
assets.) The firm benefits because it accelerates the deductions on claims that could take
years if not decades to resolve. (Normally deductions can not be taken until the claim is
actually paid.) Due to the negative correlation predicted in Section 2.1, in years in which
CLAS deductions are large, we expect interest write-offs to be small, all else equal.
The Wall Street Journal reports that KPMG marketed this transaction to 29 companies
during 1999-2001, saving these firms an estimated $1.7 billion in federal taxes.12 We
explicitly confirm that five U.S. firms have engaged in this strategy. Delta Airlines,
Whirlpool Corp., Clear Channel Communications Inc., WorldCom Inc. (now MCI Inc.),
and Tenet Healthcare Corp disclose in their financial statements that they have received a
Notice of Deficiency from the IRS. These five firms are in our sample.
2.3. Key judicial doctrines
The government has developed and invoked five non-mutually exclusive judicial
doctrines to curb corporate tax shelters. Each is described in turn.
2.3.1. Sham transaction doctrine
Courts disallow two types of sham transactions. Shams in fact are transactions that never
11
Appleby, Spurling, and Kempe, one of Bermuda's biggest law firms, promotes its expertise in such
transactions in a brochure titled "Holding Intellectual Property Offshore."
12
The Wall Street Journal, June 16, 2004, "KPMG Shelter Shaved $1.7 Billion Off Taxes of 29 Large
Companies."
10
occur. Shams in substance occur but, absent tax considerations, lack economic substance
and/or business purpose. (The economic substance and business purpose "prongs" of the
sham transaction doctrine are described in Sections 2.3.2 and 2.3.3, respectively.)
Beginning with Winn-Dixie Stores, Inc. v. Commissioner, the government has used the
sham transaction doctrine to battle company-owned life insurance (COLI) programs.13 In
this case, the Court essentially held that the program was a sham in substance (and
disallowed the deductions sought) because absent tax considerations, the company lost
money.
The government is not always successful in applying the sham transaction doctrine. For
instance, in United Parcel Service of America, Inc. ("UPS") v. Commissioner, UPS
engaged in a program with a related Bermuda corporation.14 UPS insured its customers'
packages up to $100 at no extra cost, and also offered customers the ability to purchase
additional coverage. Before the program was established, UPS self-insured the risk. With
the advent of the related Bermuda corporation, UPS continued to administer all aspects of
the insurance, but through a ceding arrangement with an unrelated insurance company,
UPS paid all premiums to the Bermuda firm that reinsured the risk. The Court ruled that
this assignment of income was a sham in substance, and shifted all of the premium
income back to UPS for tax purposes. However, the Eleventh Circuit reversed the Tax
Court, finding that the arrangement had economic substance, in part because UPS could
no longer use the income stream it had access to when it self-insured. Still, the Eleventh
Circuit remanded the case for determination under more specific statutory income
reallocation provisions of the Internal Revenue Code.
2.3.2. Economic substance doctrine
Tax law requires that transactions have economic substance (a “profit motive”) separate
and distinct from an economic benefit achieved solely from tax reduction. In short, a
transaction must change a taxpayer's economic position in a meaningful nontax way for
the IRS to recognize the transaction's tax treatment. This doctrine is two-pronged, having
both objective and subjective elements, with the latter being virtually identical to the
business purpose doctrine (discussed below).
Modern application of this doctrine can be found in a series of related cases beginning
with ACM Partnership v. Commissioner. Here Colgate-Palmolive Company ("Colgate")
entered an offshore partnership with affiliates of Merrill Lynch (the strategy sponsor) and
ABN-AMRO (a Dutch bank). The partnership engaged in a series of securities trades that
had the effect of producing for Colgate a substantial paper capital loss. The transactions
conducted were complicated contingent payment installment sales (CPIS) that exploited
the application of special ratable basis recovery rules under a temporary Treasury
regulation that was then part of IRC Section 453. The Court found these CPIS payments
to be meaningless economically. In particular, the purchase and resale of private
placement notes in return for cash and interest-only securities by the partnership placed it
13
14
Winn-Dixie, 113 T.C. 254 (1999), aff'd 254 F.3d 1313 (11th Cir. 2001).
254 F.3d 1014 (11th Cir. 2001), rev'g 78 T.C.M. (CCH) 262 (1999).
11
in the same economic position as if it had elected simply to buy the interest-only
securities directly. And without the purchase and resale of the notes, the ratable basis
recovery rules would not apply, thus eliminating Colgate's claimed capital loss. Merrill
Lynch and its competitors marketed a similar strategy to other firms, including
AlliedSignal, Inc., Borden Inc., Brunswick Corporation, and American Home Products
Corp., collectively producing well over $1 billion in losses.
As with its application of the sham transaction doctrine, the government is not always
successful in applying the economic substance doctrine. Two examples are Compaq
Computer Corp. v. Commissioner and IES Industries, Inc. v. United States. These cases
deal with cross-border dividend capture. In essence, both companies engaged in the
nearly simultaneous purchase and resale of millions of dollars of ADRs, allowing each
company to capture the foreign tax credit occasioned by the withholding tax on the
dividends paid by the foreign companies underlying the ADRs. The government argued
that the trading strategy lacked economic substance because it merely presented a means
of creating a secondary marketplace wherein tax-exempt owners of ADRs could sell their
unusable foreign tax credits to firms that valued them more highly. However, the tax
benefits claimed by the taxpayers ultimately were sustained in litigation.15
2.3.3. Business purpose doctrine
The business purpose doctrine speaks to the motivation of the taxpayer when entering the
transaction. This doctrine tests whether, when entering the transaction, the taxpayer was
motivated by a business purpose other than obtaining tax benefits. In short, a transaction's
tax treatment is not valid in the eyes of the IRS if the transaction does not have a non-tax
business purpose. Relative to the economic substance doctrine, the business purpose
doctrine is a more subjective inquiry into whether the taxpayer intends the transaction to
serve a truly useful non-tax purpose. Often a corporation will attempt to imbue a
transaction with an alleged business purpose, to satisfy the doctrine and preserve the
strategy's tax treatment. For example, in ACM v. Commissioner, Colgate unsuccessfully
attempted to argue that the offshore partnership and its transactions served to smooth the
company's debt maturity profile.
2.3.4. Substance over form doctrine
The substance over form doctrine is closely related to the aforementioned judicial
doctrines. Substance over form holds that two or more transactions (the "devious path")
that achieve the same underlying economic result that could have been achieved with
fewer transactions (the "straight path") should not be taxed differently than had the
straight path been followed. The substance over form doctrine therefore permits the
government to distinguish between economic form and formalistic, legal tax form.
This doctrine allows the government to deny tax benefits occasioned by a taxpayer
15
Section 901(k) of the Internal Revenue Code, enacted in 1997 after the Compaq and IES transactions,
disallows the foreign tax credits claimed in any similar transactions conducted after 1997.
12
undertaking economically ancillary or circular steps, even if those steps are themselves
properly treated for tax purposes. For example, in ASA Investerings Partnership v.
Commissioner the government argued that certain transactions, key to producing
AlliedSignal's claimed capital loss, were circular and therefore should not be recognized
for tax purposes. And in Zeelandia Investerings Partnership v. Commissioner the
government contended that the transactions conducted by the offshore partnership were
simply reversed-engineered by Borden Inc. and its other partners through their trading of
off-balance sheet swaps. As such, the transactions conducted by the partnership had no
substance and should not be recognized in form for tax purposes.16
2.3.5. Step transaction doctrine
The step transaction doctrine is closely related to the substance over form doctrine.
Basically, the former holds that each separate transaction in a series of related
transactions ("steps") must have independent economic purpose, else the transactions can
be "stepped together" for tax purposes.
3. Data and summary statistics
3.1. Tax shelter and matched firm sample formation
We form a sample of companies that have been involved in tax sheltering cases against
the government, and/or have been served a Notice of Deficiency related to an alleged tax
shelter. Effectively, we include a firm in our sample if we can confirm that the
government has accused it of sheltering. A majority of the firms that litigated ultimately
lost their cases. Just two have definitively won, and one case has been remanded.
We use Lexis-Nexis to conduct exhaustive electronic searches for sheltering firms, using
two primary sources. First, we search the dockets of the various Tax Courts and other
courts for litigation involving public corporations and the use of an alleged tax shelter.
Second, we search the popular press for firms identified as having received a Notice of
Deficiency from the IRS stemming from an alleged tax shelter. In particular, we search
for word strings such as "tax shelter," "transfer pricing," "sham transaction doctrine,"
“Notice of Deficiency,” etc. When a firm is identified through the latter search process,
its SEC filings are checked to confirm that it indeed received a Notice. In addition, each
firm's SEC filings are searched to identify the tax years at issue. If the relevant tax years
can not be ascertained, the firm is not included in the sample.
This process identifies 43 publicly traded corporations (involving 44 total instances of
sheltering) that have been alleged or proven by the government to have illegally sheltered
taxable income between 1975 and 2000. In 29 of these 44 cases the taxpayer brought
litigation against either the Commissioner or the United States in an attempt to preserve
the tax treatment sought. While a few court cases involved tax years dating back two
16
The Borden case never went to trial because the company withdrew the litigation.
13
decades or more, except the five CLAS deals, all the cases were docketed in the 1990s.
Table 1 contains information about 44 separate tax shelter instances, 16 of which involve
transfer pricing, 11 COLI transactions, five contingent payment installment sales, five
contested liability accelerations, three LILO deals, two cross border dividend capture
cases, one liquidation/re-contribution, and one intellectual property haven. One firm,
Compaq, was accused of both cross-border dividend capture and transfer pricing for the
same tax years. The average shelter was active approximately five years. (By "active" we
mean the years that the government alleges that the firm used this particular tax shelter.)
The 44 cases involve tax shelters that were active a total of 152 firm-years during which
the firms were accused of sheltering.
Many firms in our sample allegedly participated in a given tax shelter for several
consecutive tax years. In part of our empirical analysis we study corporate activity a
given number of years before (or after) a shelter was conducted. Therefore, we collapse
the years that a firm allegedly sheltered down into a single “event year” in much of the
analysis that follows. We do this by averaging the corporate data for a given firm over all
the years the shelter occurred into a single “year t” observation. By doing this we can
easily compare across firms because t-1 represents the year before the shelter began and
t+1 is the year after it ended, regardless of the length of the sheltering activity. A
statistical advantage of collapsing the observations into one per shelter is that it helps
address concerns about lack of independence of multiple observations for the same firm.
Without a doubt, many other firms shelter in varying degrees but are not in our sample.
For example, trial testimony in ACM v. Commissioner indicates that Merrill Lynch
marketed (and likely facilitated) its contingent payment installment sale strategy to firms
not in our sample. However, due to private settlements and the like we can not identify
these firms. In other situations, once discovered by the government, a tax shelter is
terminated via legislation or regulation without ever denying the tax treatment sought or
occasioning litigation. At other times the government denies the tax treatment sought and
the taxpayer either acquiesces or settles out of court. In these cases, no public record of
sheltering exists.
We mention these caveats about identifying sheltering firms because we form a matched
sample in part of our empirical analysis. (The matching procedure is described in the next
paragraph.) It is likely that some of the match firms in fact use tax shelters, which should
work against our ability to identify differences in debt policy between sheltering and nonsheltering firms.
To form the matched sample, we examine the universe of Compustat firms that are not in
our tax shelter sample. We identify firms that are in the same 2-digit SIC industry as a
given sheltering firm in year t-1 (the year before the sheltering activity began). Among
same industry firms, matched firms are those with assets (return on assets) within +/- 25
percent (+/- 50 percent) of the sheltering firm's assets (return on assets) in year t-1.17 This
17
We alter the match criteria in three instances. For Exxon and Tenet, we require an asset match of +/-
14
procedure produces many matches in some cases and very few in others. Therefore, for
much of our analysis, for a given shelter firm we collapse all the matched firms down
into a single match; that is, one match per shelter observation. By doing this, our
empirical analysis is not dominated by cases for which our matching procedure happens
to produce a large number of matches. As described next, this procedure produces a
matched sample that is very similar in most dimensions to the shelter sample.
Table 2 provides summary statistics for the sheltering and matched firms. All financial
statement data are from Compustat. By construction the sheltering and matched firms are
close in asset value, with both samples averaging per firm total assets of more than ten
billion dollars. The average sales revenue at both sets of firms is about $7 billion
annually. Likewise, the firms have similar asset collateralization, with inventory plus
plant, property, and equipment divided by assets averaging about 50 percent in both
samples. Sheltering firms on average pay slightly less in federal income taxes (as a
percentage of taxable income), even though they are equally as profitable as the matched
firms. Sheltering firms have higher market-to-book ratios than do the matched firms. Out
of all these characteristics, only the difference in market/book is statistically significant
based on a comparison of means.
3.2. How big are tax shelters?
We are able to identify the dollar value of the tax deficiency for 24 of the firms in our
sample. For these firms, the median tax deficiency is more than $350 million per year, or
a median tax deficiency ratio of 3.1 percent of asset value. That is, the IRS claims that
due to tax sheltering, the firm underpaid taxes equal to about three percent of asset value.
Grossed up by a 35% corporate tax rate, this implies that the typical deduction associated
with these tax shelters is more than $1 billion per firm per year, or about nine percent of
total assets. If coupon rates were 10 percent, the typical firm would need to maintain a
debt ratio of 90 percent of asset value to produce tax deductions this large. The use of tax
shelters by the firms in our sample, therefore, is economically very important.
We do not claim that every company that shelters uses a tax shelter that produces
deductions of this magnitude. It is plausible that the government detects and pursues large
tax shelter instances, and therefore our sample contains tax shelters that are larger than
those that go undetected. Nonetheless, our sample indicates that tax shelters are
sometimes huge and research cited in the introduction indicates that the economy-wide
magnitude of sheltering is very large. It is also noteworthy that some of the firms in our
sample (e.g., Microsoft) are frequently cited as examples of companies that are
underlevered. The fact that we identify tax shelter activity at these companies indicates
that it is not a simple task to identify firms that are underlevered.
It is interesting that in the year of the alleged sheltering activity, the average debt ratio of
50%, rather than 25%. For another firm, we match on year t rather than t-1 because Compusat data are not
available for this firm in year t-1. Excluding these three shelter firms does not affect our results. Also, note
that our results do not change if we define return on assets based on pre-tax income rather than net income.
15
sheltering firms is 19 percent versus 27.4 percent for the matched firms (see Table 2).
This difference is statistically different at a 1% level. Observing smaller debt ratios
among sheltering firms is consistent with these firms using less debt because they obtain
their tax deductions from non-debt sources, namely from tax shelter vehicles. However, it
is possible that sheltering firms are fundamentally different from the matched firms, and
perhaps they have lower debt ratios for other reasons.
To further investigate the differences in debt ratios, we graph the mean debt ratio for both
sets of firms in Fig. 2. Debt is total debt (i.e, short-term plus long-term) divided by total
assets. The top panel in the figure indicates that shelter and matched firms have
approximately the same debt ratios seven or eight years before the sheltering activity took
place. (The shelter and matched firm debt ratios are not statistically different through year
t-4.) By the year of the alleged shelter, however, the debt-to-asset ratio of sheltering firms
is more than 800 basis points lower than for the matched firms, and the difference is near
its maximum in the year of the shelter. The year after the shelter ends, the difference
narrows between the debt ratios of the two sets of firms.18 This same basic pattern occurs
in the lower panel, which plots debt-to-market value (where market value is measured as
market equity plus book debt).
While we have no direct evidence explaining why debt ratios fall off gradually leading up
to the shelter year and then rebound, we note that this time-series pattern is consistent
with 1) the group of shelter firms sheltering in increasing amounts in the years leading up
to t=0, and 2) the government identifying only the years with the largest sheltering. Or, a
firm may play down its debt usage in anticipation of upcoming tax shelter activity. In
either scenario, it is not surprising to observe shelter firms having relatively low debt
ratios on either side of the shelter year. One implication of these scenarios is that one can
not interpret a sheltering firm’s debt ratio in the years just before or just after an alleged
shelter period as being "normal."
In the next section we perform multivariate regressions to investigate whether firms that
use tax shelters have low debt ratios, controlling for other characteristics. That analysis,
however, cannot eliminate the possibility that the negative relation between debt and
sheltering is at least partially caused by sample selection. For example, if the government
were to focus on firms with low and recently declining debt ratios as possible tax shelter
cases, then firms with these debt ratio patterns would end up in our sample. In this case,
the pattern in Fig. 2 might occur for reasons other than shelters causing low debt ratios.
While the possibility of such a spurious relation suggests caution in generalizing our
results to the full population of firms, there are three reasons why we do not think that
this issue is a major problem for our central hypothesis. First, the government often does
not detect tax shelter cases until several years after the shelter activity has ended;
therefore, the spurious relation story would need to also explain the recovery in debt
ratios that occurs after the shelter ends (but before detection). While this is possible, it
puts additional demands on the story. Second, we talked to several IRS agents who have
18
It is worth noting that the IRS typically does not contact a company about tax sheltering until t+2 or t+3.
16
expertise in the shelters undertaken by our sample firms. These agents say that having a
debt ratio that declined for many years and then recovered would not make a firm more
likely to end up under IRS scrutiny for possible tax sheltering, especially during our
sample period. Historically, a firm was investigated as a possible sheltering case when an
IRS field agent detected an unusually large (ordinary or capital) gain or loss. The IRS
traditionally has not employed statistical detection models for corporate sheltering
activity – models that might include factors such as debt ratios.19 Finally, even if the
government were to focus on companies with debt ratios that decline and then recover
(like the shelter firms in Fig. 2), this would not preclude the possibility that these firms
were in fact substituting between tax sheltering and the use of debt. The government
might just happen to choose to focus on firms that optimize in the manner hypothesized
by DeAngelo and Masulis (1980).
4. Do tax shelter deductions substitute for interest deductions?
In this section, we perform multivariate regressions to determine whether sheltering
activity is correlated with debt ratios, all else equal. We control for other factors that are
known to affect corporate debt policy by including right-hand side variables that have
been used in previous studies to explain debt policy. We start with the right-hand side
variables identified by Frank and Goyal (2004) as the most significant regressors in
empirical studies of debt policy. These variables are also used by Graham, Lemmon, and
Schallheim (1998) in their study of capital structure.
4.1 Regression analysis of the debt policy of tax shelter firms
The main regressions are purely cross-sectional, with two observations for each tax
shelter (one for the shelter firm and one for the matched firm). If the matching procedure
identifies more than one match firm, a single observation is created by averaging across
all the matches. We use this approach so that the relative number of matches for any
particular shelter case does not unduly affect our results. In a robustness check discussed
in Section 4.2, we alternatively use all the matched observations for all firms, thereby
greatly increasing sample size. The dependent variable in the main regressions is shortterm plus long-term debt, the quantity divided by total assets. Ordinary least squares
regressions are used (rather than tobit) because every observation in the main regressions
has nonzero debt.
The key right-hand side variable is a dummy variable equal to one if the firm has been
identified as having an active tax shelter and equal to zero otherwise. The other
regressors are intended to control for factors known to affect debt policy. Firm size (sales
revenue) is included to capture any economies of scale that exist in the issuance or use of
debt. The market-to-book ratio is intended to control for differences in investment
19
The IRS could move toward a statistical detection model in the future because corporations now file
electronic tax returns, and the new M-3 filing contains information that would be helpful in developing a
prediction model (see Boynton and Mills, 2004).
17
opportunities. A dividend dummy variable is included to control for differences in
informational asymmetry. Dividend-paying firms are hypothesized to be subject to less
information asymmetry and hence have more debt capacity. One of the most pervasive
facts about debt policy is that profitable firms use less debt (Myers, 1993), which could
reflect pecking order behavior in which internal funds (if available) are used before
external funding sources such as debt. Therefore, we include ROA as a control variable.
Another pervasive fact about debt policy is that firms with highly collateralizable assets
(e.g., inventory, property, plant, and equipment) use more debt, so we include the
proportion of assets that are collateralizable.
Finally, because recent research (e.g., Leary and Roberts, 2005) shows that companies
have a tendency to work towards target debt ratios, we include a lagged debt ratio as a
control variable. We choose a five-year lag to strike a balance between a lag of seven
years, at which time we observe nearly equal debt ratios between sheltering and matched
firms (in Fig. 2), and the more traditional one-year lag. A one-year lag seems
inappropriate in our setting because we include lagged debt to capture some element of
"normal" debt policy, and it seems likely that the shelters were already affecting the firms
in our sample by t-1. (Note also that the mean 5-year lagged debt ratios for matched and
shelter firms are not significantly different from each other.) All of the independent
variables (except the shelter dummy and 5-year lagged debt ratio) are lagged one period
because their year-t values are potentially determined jointly with debt policy.
The estimated coefficients for many of the right-hand side variables have the expected
signs (see Table 3). Dividend-paying firms use more debt than non-dividend firms. The
estimated coefficient indicates that dividend-paying firms have debt ratios that are 650
basis points higher than for non-dividend firms. The coefficients also indicate that
profitable firms use less debt, and that firms with collateralizable assets use more debt,
both as predicted. Firm size is not significant in this regression (and profitability only at a
10% level), perhaps because of the matching procedure that we used to form the sample.
The market-to-book ratio is positive, surprisingly, indicating that all else equal high
market-to-book firms in our sample use more debt.
The most important variable for the purposes of this paper is the tax shelter dummy. The
estimated coefficient indicates that firms that use tax shelters have debt ratios that are 550
basis points lower than debt ratios for the matched firms. Therefore, controlling for other
factors, tax shelter firms use less debt than do non-shelter firms, which is consistent with
shelters producing non-debt tax shields that substitute for the use of corporate debt. In
this main specification the adjusted R-squared is 54.2%, indicating a good fit. There are
76 observations used in the main specification, consisting of 38 shelter firms and 38
matched firms that have sufficient non-missing data to perform the regression.
Recent anecdotal evidence (e.g., Enron) suggests that firms that use tax shelters might
also use other forms of off-balance sheet financing. If this is true, then the negative
relation between debt ratios and tax shelter activity might not indicate that shelters are the
sole factor correlated with reduced use of debt; instead, participating in a shelter might be
18
correlated with the use of other NDTS that also lead to substitution away from debt. Mills
and Newberry (2004), however, show that firms with high debt ratios are more likely to
use off-balance sheet and hybrid financing instruments. Thus, the low debt ratios for our
tax shelter firms imply less non-shelter off-balance sheet financing. Therefore, contrary
to what the Enron anecdote might suggest, it seems unlikely that non-shelter off-balance
sheet NDTS are behind the negative relation that we document between debt ratios and
tax shelter activity.
4.2. Robustness of debt policy analysis
We perform several robustness checks to the main specification. Column 2 of Table 3
adds a variable equal to the calendar year that the shelter was allegedly being used. This
should control for time-trends in debt ratios that are not captured by the other variables.
While the trend variable is significant, the tax sheltering dummy variable remains
negative and significant, so our central finding holds. Column 3 repeats the main
regression (i.e., the specification summarized in Column 1) but drops the two variables
for which the estimated coefficients have unexpected signs in the main specification: size
and market-to-book. The estimated coefficients for the remaining factors are qualitatively
unchanged. Finally, Column 4 drops the five year lagged debt ratio from the
specification, which leads to the inclusion of eight additional observations in the analysis
(four shelters and four matches). In this case, the negative coefficient on the tax shelter
dummy variable is even larger and again indicates that sheltering firms have lower debt
ratios.
Although we use lagged explanatory variables in an attempt to circumvent possible
endogeneity, we recognize that this might be an imperfect fix. To further address this
issue, we perform 2-stage analyses in which we use "predicted sheltering" as an
explanatory variable (in place of the tax shelter dummy). Predicted sheltering is created
using the estimated coefficients from a regression with 0/1 sheltering as the dependent
variable. In addition to the tax shelter observations, the dataset for the first stage
regression includes 68 randomly chosen Compustat observations for firms that, as far as
we know, do not use tax shelters. The first-stage regression indicates that firms that are
large, profitable, and engage in intense R&D are relatively likely to shelter.20 The secondstage regression indicates that predicted tax sheltering leads to lower debt ratios,
corroborating our main result.
In an alternative 2-stage approach, we follow Lee (1978), Heckman (1979), and Fang
(2005) and use an endogenous switching model, which recognizes that the tax shelter
dummy might be endogenous to debt policy. The results of this approach indicate that if
tax shelter firms were to hypothetically not use shelters, their debt ratio would nearly
double, increasing from the observed mean of 19% to 36.6%. If nonshelter firms were to
hypothetically begin using tax shelters, their debt ratios would fall from the observed
mean of 27.4% to 15.3%. Both of these differences are significant (t-scores greater than
10). These analyses quantify that, controlling for endogeneity, the effect of tax sheltering
20
Desai, Foley, and Hines (2005) find that a similar set of variables explains tax haven activity.
19
activity on debt usage is economically very important in our sample.
Additional robustness checks are presented in Table 4. The specification in Column 1 of
Table 4 mimics the specification in Column 1 of Table 3, except the dependent variable is
debt-to-value. The estimated coefficient indicates that debt-to-value is 530 basis points
lower for tax sheltering firms, relative to matched firms, all else equal. In Column 2 and
the remaining columns, we return to using debt-to-assets as the dependent variable. In
Column 2 of Table 4, we delete the CPIS and LR observations because these shelters
address capital losses, which are not directly substitutable for interest deductions. We
also delete COLI observations because their effect on the debt ratio is ambiguous (see
Section 2.2.) These deletions produce a “clean sample” but do not affect the overall
inference.21
In Columns 4 and 5, we use a panel of data from t-8 to t+6. In this panel, rather than
collapsing all the matches down into one observation per shelter firm, we include all the
match firms in the regression. Also, each firm-year in which a tax shelter is active
receives its own observation (i.e., we do not collapse all active shelter years down into a
single observation like we do in Table 3 and the other columns of Table 4). Like the main
specification in Column 1 of Table 3, there are 38 sheltering firms in this specification;
however, there are now 1140 observations.
The main result holds in the panel regressions shown in Column 4 of Table 4: Firms that
use tax shelters have lower debt ratios. This specification is performed using OLS, to
enhance comparability with the other regressions. However, a couple dozen of the 1,140
observations are censored, so a tobit regression is also preformed. The estimated tobit
coefficients are nearly identical to those reported in Table 4.
In column 5, we add a new variable, shelter firm dummy, to address whether the lower
shelter-firm debt ratio is a time-series or a cross-sectional effect. That is, we investigate
whether the lower debt ratio for sheltering firms emanates from a debt ratio that falls
through time for shelter firms and/or results from cross-sectionally lower debt ratios for
shelter firms relative to matched firms. The new shelter firm dummy equals one for a
shelter firm in any year and equals zero for matched firms. This new variable captures
cross-sectional differences between shelter and matched firms. The shelter active dummy
captures differences in debt ratios for shelter firms in years for which a tax shelter is
being used (relative to debt ratios of shelter firms when the shelter is not being used). The
negative coefficients on both variables indicate that shelter firms use less debt than
matched firms in the years the shelter is not active, and also that shelter firms redefine
their debt ratios when the shelter is active. In this sense, therefore, the tax shelter effect is
both time-series and cross-sectional.
In two regressions that do not appear in the table, we more closely control for market-tobook and for sample composition. In the first, we require matching firms to have market21
In untabulated regressions, we find that the main results hold for just the firms that used transfer pricing
shelters, as well as the firms that did not use transfer pricing.
20
to-book that is at least half as large as the sheltering firm’s market-to-book, and no larger
than twice as big. Regressions based on this extra matching requirement produce
coefficients the magnitude and significance of which are similar to what is reported for
the main specification. In the other untabulated regression, we only examine sheltering
and matched firms that exist at least three years prior to and three years after the year of
the alleged shelter, reducing the sample to 36 shelters and 36 matches. The qualitative
results are similar to those previously reported, with sheltering firms using less debt than
matched firms.
Next, we address the possibility that asset value or profits increased more for shelter
firms (relative to match firms) during our sample period. This could explain why debt-toassets falls for shelter firms. First, we note that actual asset growth for shelter firms is
nearly identical to growth for match firms in the years leading up to the inception of the
shelter activity (not shown in a table). Similarly, cumulative profit from t-5 to t-1 for
shelter firms is statistically indistinguishable from cumulative profit for matched firms.
Second, we include a variable in the main specification that equals cumulative ROA from
year t-5 to t-1. We do this to better control for the effect of cumulative profitability on
debt ratios than can be done using the usual control of one-period lagged ROA. While the
cumulative ROA variable is significant at a 5% level, the shelter dummy still indicates
that shelter firms have lower debt ratios than do matched firms, all else equal (Column 3
of Table 4). These results taken together indicate that the lower debt-to-assets ratio for
the shelter firms is not a denominator effect.
Finally, we examine incremental debt issuance decisions from t-8 to t=0. By focusing on
the numerator of the debt ratio, we avoid the possibility that denominator effects drive
our results. If tax shelters create deductions that substitute for interest deductions, we
expect to see sheltering firms issue less debt. Unconditionally, 60 percent of sheltering
firms issue debt in the years leading up to tax shelter activity, in comparison to 70 percent
of non-sheltering firms. In a logistic regression that holds all else equal, sheltering firms
are significantly less likely to issue debt than are non-sheltering matched firms (see Table
5). The estimated coefficient for the tax shelter dummy indicates that tax shelter firms are
13.7% less likely to issue debt, all else equal.
5. Summary and conclusions
We investigate the use of 44 corporate tax shelters at 43 firms from 1975 to 2000. The
shelters in our sample are very large economically, producing annual deductions that
average about nine percent of asset value. The tax savings produced by these shelters are
much larger than interest tax deductions for comparable firms that we do not identify as
using tax shelters.
Not only are tax shelters large economically, they appear to interact with corporate debt
policy in interesting ways. Firms that use tax shelters use less debt on average than do
non-shelter firms. Regression coefficients indicate that, everything else equal, tax
21
sheltering firms’ debt-to-asset ratios are more than five percentage points lower than
leverage for non-shelter firms. These results are consistent with tax shelters being a nondebt tax shield that substitutes for the use of interest tax deductions (DeAngelo and
Masulis, 1980).
Overall, our results are consistent with the belief that tax sheltering activity is important
economically (Bankman (2003) and Desai (2003)). Given the recent dramatic reduction
in corporate tax receipts, learning more about the magnitude of tax shelters, and their
effect on other corporate policies, is an intriguing area for future research. Currently, the
confidentiality accorded to the identities of tax shelter firms hinders such research. Future
investigations that creatively obviate this lack of information have the potential to make
important contributions to our understanding of tax shelter activity. Some help might
come from the expanded disclosure requirements of the new tax form Schedule 10-3 (see
Boynton and Mills (2004) for details).
Appendix – Abridged Case Studies
Case study of Compaq’s cross-border dividend capture shelter (CBDC)
As an illustration of a cross-border dividend capture shelter, consider the transactions conducted
by Compaq Computer Corporation. On September 16, 1992, Compaq bought from and then
immediately sold to the same party 10 million American Depository Receipts (ADRs) of the
Royal Dutch Petroleum Company (RDP) of the Netherlands. The transactions were sponsored
and structured by Twenty First Securities Corporation (TFSC).22 Because TFSC was a small
brokerage firm and not a member of the New York Stock Exchange (NYSE), TFSC engaged Bear
Stearns, a major brokerage firm and member of the Exchange, to clear the ADR transactions.
Specifically, Bear, Stearns Securities Corporation acted as TFSC’s clearing broker, searched the
market, located and borrowed the ADRs from institutional investors (such as pension funds or
other tax-exempt institutions23) and banks and other brokerage firms (who held ADRs for
unidentified investors believed to be tax-exempt) for short selling to Compaq. Arthur J. Gallagher
& Co., an insurance broker and client of TFSC, was engaged by TFSC as the short-seller.
In accordance with prior arrangements by TFSC, in a series of 23 cross-trades completed within a
1-hour time span, Gallagher sold short the ADRs to Compaq at "prevailing" market prices plus
the expected net dividend. The net dividend is defined as the gross dividend adjusted for the both
the 15% foreign withholding tax and the guilder-dollar exchange rate. In each case, Compaq then
immediately sold the securities back to Gallagher at the same price less the net dividend.
Each of the 23 purchases resulted in an average ADR price that equaled the "prevailing" (at time
22
According to deposition testimony of TFSC officers, the firm had also arranged hundreds of similar
transactions for other clients during the 1990s.
23
As discussed shortly, the whole transaction was occasioned by the fact that ADRs could be borrowed
from a tax-exempt institution that had no use for the foreign tax credit but would still receive dividends that
were net of the foreign tax. TFSC’s marketing documents for the transactions noted that the strategy was
facilitated by the fact that pension funds and other tax exempt entities, that had significant holdings of
ADRs as a result of increased global investing, could not utilize the foreign tax credit. These tax-exempt
institutions participated - lending their ADRs and accommodating TFSC's clients - in return for standard
short interest rebates earned by the lending of securities.
22
of trade) ex-dividend price of the ADRs plus the net dividend of $1.9165. (This $1.9165 net
figure equals the gross dividend declared by RDP, adjusted for the 15% dividend withholding tax,
and also adjusted for the spot exchange rate (dollar-guilder) at which the declared dividend net of
the foreign withholding tax was repatriated into dollars.) This unique average ADR price (which
is a four-decimal-place price at a time when NYSE stocks were quoted in eighths) was
accomplished by trading different blocks of ADRs at the same time but at different prices.
The resale of the ADRs from Compaq to Gallagher occurred at the prevailing ex-dividend price,
implying a loss to Compaq equivalent to the dollar value of the net dividend, ignoring trading
costs. Compaq’s purchases totaled $887,577,130 (before commissions and margin interest) while
the sales amounted to $868,412,130 (before commissions and fees). The $19,165,000 difference
is equivalent to the $1.9165 per share net dividend.
The specific time line follows. As of the trading date (September 16, 1992), Royal Dutch had
already declared a dividend on its ADRs, to be paid on October 2. The ADR went ex-dividend on
September 14, and the dividend record date was September 18. To enable Compaq to capture the
declared dividend, the company purchased the ADRs with a special settlement date of September
17, 1992. Since this settlement date preceded the record date, Compaq was entitled to the
declared dividend. The ADRs, however, were already selling in the market at the ex-dividend
price. Hence, as structured by TFSC with a special settlement date, Compaq paid the prevailing
market price plus the net dividend that would be due the holder of the ADR.24 Compaq then sold
back the ADRs at the same price less the net dividend (which in this case was the "prevailing" exdividend price) with a regular settlement date, i.e., settlement in five days after the trading date or
September 21, 1992. Compaq agreed to pay commission to TFSC of $0.05 per ADR traded, i.e.,
$1,000,000 for the 20 million ADRs traded (purchases and sales). Compaq also paid margin
interest (at 4.75%) totaling $457,845.73 for the ADR purchase, and SEC fees of $28,947 on its
sale of the ADRs.
On October 2, 1992, Compaq’s reported gross dividend from RDP’s ADRs was $22,545,800. Of
this amount, $3,381,870, i.e., the 15% Netherlands withholding tax on the dividends, was
withheld by RDP. Thus, Compaq received a net dividend payment of $19,163,930. Compaq
claimed a foreign tax credit of $3,382,050 for 1992 in connection with the taxes withheld. In
general terms, Compaq paid $1.5 million to obtain a $3.3 million tax credit (note: a tax credit, not
a deduction). Graham (2003) demonstrates that foreign tax credits decrease the incentive to issue
domestic debt, as long as a firm has a positive probability of operating in “excess foreign tax
credit” status. This occurs because interest allocation rules can reduce allowable domestic interest
deductions for firms with foreign operations when the firm is excess credit.
A summary of Compaq’s transactions is provided below. This schedule shows that the ADR
strategy designed by TFSC had the economic effect of enabling Compaq to purchase a 34-cent
per share tax credit at a net cost of 21.5 cents per share.
Summary of Compaq Transactions
Purchase of ADRs (before commissions and margin interest)
24
Total Value
Ave. Price
per share
$887,577,130
$88.7577
NYSE regulations allowed transactions to be undertaken in this manner.
23
Sale of ADRs (before commissions and fees)
868,412,130
-----------($19,165,000)
86.8412
--------($1.9165)
($ 1,000,000)
(
457,846)
(
28,947)
------------($ 1,486,793)
($0.1000)
(0.0458)
( 0.0029)
--------( 0.1487)
Net Loss on Purchase/Sale (A)
($20,651,793)
=============
($2.0652)
=========
Gross Dividends Declared
Less: Dutch Taxes Withheld
$22,545,800
3,381,870
------------$19,163,930
=============
$2.2546
0.3382
--------$1.9164
=========
($ 1,487,893)
($
643,954)
------------($ 2,131,847)
$ 3,382,050
------------$ 1,250,203
=============
($0.1488)
($0.0664)
-------($0.2152)
$0.3382
-------$0.1230
========
Loss
Transactions Costs
Commissions on ADRs
Margin Interest
SEC Fees
Total Transactions Costs
Net Dividend Received (B)
Net Cash Flow (before U.S. Taxes) (A+B):
U.S. Income Tax
Net Cash Flow after U.S. Income Tax
Foreign Tax Credit
Net Cash Flow after U.S. Taxes
Case study of Colgate’s contingent-payment installment sale (CPIS)
Colgate Palmolive Company (Colgate) undertook a CPIS with the assistance of Algemene Bank
Nederland N.V. (ABN). Colgate reported a $105 million gain for 1988 (occasioned by a division
sale), and Merrill Lynch & Co. (Merrill) sponsored a plan to shelter said gain. Through newly
formed entities known as Southampton (for Colgate), Merrill Lynch MLCS Inc. (for Merrill), and
Kannex (for ABN), the three firms created, in October of 1989, a Curacao-based, Delawareincorporated partnership called ACM. ACM served as a special purpose vehicle to create a large
capital loss for Colgate, with corresponding gain allocated to Kannex, which was not subject to
U.S. tax. ACM was initially capitalized with $205 million, and the initial partnership interests
were Colgate/Southampton 17.1%, Kannex/ABN 82.6%, and MLCS/Merrill 0.3%. Upon its
formation, ACM purchased $205 million face value of private placement notes issued by
Citicorp. Three weeks later, in November 1989, ACM sold $175MM face value of these notes to
two buyers - BFCE (a French bank) and BOT (a Japanese bank) - for approximately $140 million
in cash and approximately $35 million worth of floating-rate, interest-only securities (issued by
said buyers). ACM reported the transaction under the contingent-payment installment sale
provisions of temporary IRC section 453 (known as the "ratable basis recovery rule"). ACM
reported a $110.7 million capital gain in the year of the sale (1989), mostly allocated to Kannex.
In a later tax year, after ACM had redeemed Kannex's interest, ACM sold the interest-only
securities and reported an $85 million capital loss that was allocated to Colgate. Colgate carried
the loss back to offset part of its 1988 gain.
More specifically, because ACM was to receive part of the consideration for the sale of the
Citicorp notes "after the close of the taxable year in which the disposition occurs" pursuant to
IRC section 453(b)(1), on its partnership return for the tax year ending November 30, 1989 ACM
treated the sale of the Citicorp notes as an installment sale under Temporary Treasury Regulation
Section 15a.453-1(c), whose ratable basis recovery rule provides that the taxpayer's basis "shall
be allocated to the taxable years in which payment may be received under the agreement in equal
24
annual amounts." So, ACM divided its $175,504,564 basis in the Citicorp notes ($175 million in
principal and $504,564 in accrued interest) equally among the six years over which payments
were to be received in exchange for those notes (reflecting the maturity of the interest-only
securities or "IOs"), and thus recovered one-sixth of that basis, or $29,250,761, during 1989.
Subtracting this basis from the $140 million in cash proceeds, ACM reported a 1989 gain of
$110,749,239, which it allocated among its partners according to their partnership shares. This
resulted in a gain of about $91.5MM to Kannex (which was not subject to U.S. income tax),
$18.9MM to Southampton, and $324k to MLCS.
The tax basis to be recovered over the remaining five years became $146,253,803. Of this
amount, $41,786,801 was attributable to the BFCE-issued IOs (whose actual cost was
$10,144,161). The remaining $104,467,002 in unrecovered tax basis was attributable to the
BOT-issued IOs (whose actual cost was $25,630,403). ACM distributed the BFCE IOs to
Southampton in early December 1989 (as a return of part of Southampton's contributed capital),
and Colgate/Southampton sold the BFCE IOs in late December 1989 for $9,406,180.
On its 1989 return, Southampton reported its $18,908,407 share of the capital gain from the
$140MM in cash received for the sale of the Citicorp notes, and reported a $32,429,839 capital
loss from its sale of the BFCE IOs. Because these capital losses completely offset the capital
gains, Southampton reported a net 1989 capital loss of $13,521,432 and did not report any net tax
liability on its share of ACM's gain from the disposition of the Citicorp notes.
In June 1991 Colgate/Southampton acquired a 38.31% share in ACM from ABN/Kannex for
$85,897,203, and Colgate/Southampton acquired an additional 6.69% share from Kannex for
$15,000,000, giving Colgate/Southampton a majority interest in ACM. This permitted Colgate to
consolidate ACM's holdings with its own on its books. ACM exercised a put option embedded in
the remaining $30MM face value of Citicorp notes, thus selling the remaining notes back to
Citicorp. In November of 1991, ACM redeemed the rest of ABN/Kannex's interest for
$100,775,915, leaving Southampton/Colgate with 99.7% of ACM. Then ACM sold the BOT IOs
(to BFCE) for $10,961,581. So, for the tax year 1991 ACM reported a capital loss of $84,997,111
from its sale of the BOT IOs (reflecting the selling price and the remaining $95,958,692 basis in
these IOs). Of this approximately $85MM loss, $5.8MM reflected a decline in value due to
lowered interest rates while $79.2MM resulted from the application of the ratable basis recovery
rule (which effectively added to the tax basis of the IOs five-sixths of the $140MM value of the
Citicorp notes which had been sold for cash in 1989).
Colgate claimed 99.7% of this $85MM capital loss on its 1991 return. The company then filed an
amended 1988 return reporting this loss as a carryback pursuant to IRC section 1212 to offset a
portion of its 1988 capital gain. So, the carryback of about $85MM plus the aforementioned
reported net loss in 1989 of about $13.5MM together provided a shelter of nearly $100MM for
Colgate. Given its marginal federal income tax rate, this shelter provided Colgate with a real
economic benefit of nearly $40MM, before transaction costs. The costs totaled only about
$15MM for this shelter.
Case study of FPL’s liquidation/re-contribution deal
In 1992, Florida Power and Light (FPL, currently the largest utility in the U.S.) allegedly wanted
to refresh an expiring loss carry forward of about $337 million. Goldman Sachs devised a plan in
which a newly created offshore partnership, called Salina, engaged in a series of securities
25
transactions. Originally Salina consisted of two partners that were also newly created and, in a
complex way, related to and managed by a large foreign bank - ABN AMRO Holdings NV.
After its formation in mid-December 1992, Salina immediately engaged in a series of securities
transactions, most notably the short sale of $344.4 million worth of 6-month Treasury bills and
the purchase of $140.3 million worth of 2-year Treasury notes. Salina also engaged in a reverse
repurchase agreement with ABN's New York office, that is, it loaned ABN NY $343.9 million.
Salina borrowed, via a repurchase agreement, from Goldman Sachs $70.1 million, and the two
original partners collectively posted $75.4 million in capital.
Following these transactions that became effective December 18, 1992, Salina's balance sheet
was as follows:
ASSETS
Time Deposits
2-Year Treasury Notes
Reverse Repo
LIABILITIES
6-month Treasury Bills Sold Short
Repo
PARTNERS' CAPITAL
$(millions)
5.1
140.3
343.9
344.4
70.1
75.4
Just ten days later, on December 28, 1992, FPL purchased a 98% partnership interest in Salina.
Under then existing partnership accounting rules,25 this act occasioned a liquidation of the
partnership whose assets and liabilities were then immediately re-contributed to an allegedly new
partnership. This new partnership kept the same name and same tax identification number as the
liquidated partnership. The liquidation/re-contribution occasioned the desired gain of $337
million as follows: The short bill position was valued by FPL at zero, leaving approximately $145
million (the "outside basis") in liabilities and equity for the new partnership (approximately $70
million on the Goldman repo and $75 million of equity).26,27 The approximately $5 million of
time deposits (cash) was then subtracted from the outside basis to give an "inside basis" of $140
million. This amount was then proportionally allocated to the remaining assets of approximately
$483 million: the approximately $343 million loan to ABN on the reverse repo plus the
approximately $140 million worth of Treasury notes. The loan represented about 71% of the
remaining assets, the notes the other 29%. Thus the inside basis of $140 million was allocated as
follows: $99.4 million to the loan (71% of $140 million) and $40.6 million to the notes (29% of
$140 million). Under existing partnership accounting rules previously cited, this allocation would
therefore lead to a paper gain on the loan, to be triggered whenever the loan receivable is
collected, of $243.6 million ($343MM - $99.4MM). This allocation would also lead to a paper
gain on the notes, to be triggered whenever the notes are sold, of $99.4MM ($140MM 25
IRC section 708(b)(1)(B) states that a sale of 50% or more of a partnership within a twelve-month period
constitutes a termination of said partnership.
26
FPL contended that the Treasury bill short sale obligation was not a "liability" as governed by IRC
section 752, but instead was governed by IRC section 1233 and regulation 1.1233-(1)(a), which states that a
short sale is treated as an "open transaction" for income tax purposes. Here a short seller can defer
recognition of income until closing the transactions by replacing the borrowed securities. Therefore, any
adjustments to FPL's basis must be deferred until the short sale transaction is complete. Remarkably,
neither subchapter K partnership provisions of the IRC nor relevant regulations interpreting IRC section
752 clearly define “liability.”
27
IRC section 732(b) discusses the pertinent basis allocations resulting from the termination of a
partnership.
26
$40.6MM). Adding these two amounts results in a gain of $343 million.
By the end of December 1992, and therefore after FPL became a 98% partner in Salina, the loan
receivable was collected and the notes were sold. Under existing partnership accounting rules,
FPL thereby obtained 98% of the $343 million gain, which matches its desired gain of $337
million.28 December of 1992 also marked the end of FPL's fiscal tax year as well as the expiration
of its loss carry forward.
Thus, FPL attempted to refresh its expiring capital loss carry forward by the engineering of a
paper gain in this LR transaction. It is the contention of the IRS that FPL hoped to generate real
gains beyond 1992 that could be offset by the refreshed capital loss carry forward. And that FPL
intended to engineer a new paper capital loss (via another tax shelter) that would serve to offset
the paper gain required to refresh the original but about-to-expire capital loss carry forward. If
this or similar strategies were applied repeatedly, the firm could conceivably delay the payment
of gains taxes indefinitely.
28
If a corporation/parent (FPL) owns more than 50% of a partnership, the partnership's financial statements
are integrated into those of the corporation/parent. In other words, the $337 million gain was reflected in
FPL's financials for 1992.
27
References
Baldenius, Tim, Melumad, Nahum, and Stefan Reichelstein, 2004, Integrating managerial and tax
objectives in transfer pricing, Accounting Review 79, 591-615.
Bankman, Joseph, 1999, The new market in corporate tax shelters, Tax Notes 83, 1775.
Bankman, Joseph, 2003, The tax shelter battle, in Crises and Tax Administration, ed. H. Aaron and J.
Slemrod, Brookings Institution.
Bankman, Joseph, and Daniel Simmons, 2003, Terminating tax shelters: Has California broken the
legislative logjam?, Tax Notes.
Bartelsman, Eric J., and Roel M. W. J. Beetsma, 2003, Why pay more? Corporate tax avoidance through
transfer pricing in OECD countries, Journal of Public Economics 87, 2225-2252.
Bird, Robert, and Alan Tucker, 2002, Tax sham or prudent investment?: A case study of Salina Partnership
v. Commissioner of Internal Revenue, Virginia Tax Review 22, 231-272.
Bonyton, Charles, and Lillian Mills, 2004, The evolving schedule M-3: A new era of corporate show and
tell? National Tax Journal LVII, 757-769.
Cetta, Anthony J., 2002, Tax efficiency in corporate cash management, Derivatives Report.
Clausing, Kimberly A., 2003, Tax-motivated transfer pricing and US intrafirm trade prices, Journal of
Public Economics 87, 2207-2223.
DeAngelo, Harry and Ronald W. Masulis, 1980, Optimal capital structure under corporate and personal
taxation, Journal of Financial Economics 8, 3-29.
Desai, Mihir A., 2003, The divergence between book and tax income, Tax Policy and the Economy 17, ed.
J. Poterba, MIT Press, 169-206.
Desai, Mihir A., C. Fritz Foley, and James R. Hines, Jr., 2004, Economic effects of regional tax havens,
Working Paper, Harvard University.
Desai, Mihir A., C. Fritz Foley, and James R. Hines, Jr., 2005, The demand for tax haven operations,
Working Paper, Harvard University.
Fang, Lily, 2005, Investment bank reputation and the price and quality of underwriting services, Journal of
Finance forthcoming.
Frank, Murray, and Vidhan Goyal, 2004, Capital structure decisions: Which factors are reliably important?
Working Paper, University of British Columbia.
Graham, John R., 2000, How big are the tax benefits of debt? Journal of Finance, 55, 1901-1941.
Graham, John R., 2003, Taxes and corporate finance: A review, Review of Financial Studies, 16, 10741128.
Graham, John R., Mark Lang, and Doug Shackelford, 2004, Employee stock options,
corporate taxes and debt policy, Journal of Finance 59, 1585-1618.
Graham, John R., Michael Lemmon, and James Schallheim, 1998, "Debt, leases, taxes, and the endogeneity
of corporate tax status,” Journal of Finance 53, 131-162.
28
Green, Richard C. and Burton Hollifield, 2003, The personal-tax advantages of equity, Journal of Financial
Economics, 67, 175-216.
Heckman, J. James, 1979, Sample selection bias as a specification error, Econometrica 47, 153-161.
Hennessy, Christopher A., and Toni M. Whited, 2004, Debt dynamics, Journal of Finance forthcoming.
Hines, James, 1997, Tax Policy and the activities of multinational corporations. Ed. Alan Auerbach, Fiscal
Policy: Lessons from Economic Research. MIT Press, Cambridge, 401-445.
Joint Committee on Taxation, 1999, Study of present-law penalty and interest provisions as required by
Section 3801 of the Internal Revenue Service Restructuring and Reform Act of 1998 (Including Provisions
Relating to Corporate Tax Shleters). JCS 3-99, Washington, D.C.: U.S. Government Printing Office.
Ju, Nengjiu, Robert Parrino, Allen M. Poteshman, and Michael S. Weisbach, 2004, Horses and rabbits?
Optimal dynamic capital structure from shareholder and manager perspectives, Journal of Financial and
Quantitative Analysis forthcoming.
Leary, Mark, and Michael Roberts, 2005, Do firms rebalance their capital structures?, Journal of Finance
forthcoming.
Lee, Lung-Fei, 1978, Unionism and wage rates: A simultaneous equations model with qualitative and
limited dependent variables, International Economic Review 19, 415-433.
Luitjens, Marlyce, 2004, Sale in-lease out (SILO) transactions, FSLG Newsletter (June).
McGill, Gary A., and Edmund Outslay, 2004, Lost in translation: Detecting tax shelter activity in financial
statements, National Tax Journal 57, 739-756.
Miller, Merton H., 1977, Debt and taxes, Journal of Finance, 32, 261-275.
Mills, Lillian, and Kaye Newberry, 2004, Firms' off-balance sheet and hybrid debt financing: Evidence
from their book-tax reporting differences, Journal of Accounting Research forthcoming.
Myers, Stewart C., 1993, Still searching for the optimal capital structure, Journal of Applied Corporate
Finance, 6, 4-14.
Parrino, Robert, and Michael Weisbach, 1999, Measuring investment distortions arising from stockholderbondholder conflicts, Journal of Financial Economics 53, 3-42.
Plesko, George, 2004, Corporate tax avoidance and the properties of corporate earnings, National Tax
Journal 57, 727-737,
Schallheim, James, and Kyle Wells, 2004, Are firms really under-levered?, Working Paper, University of
Utah.
Scholes, Myron., Mark Wolfson, Merle Erickson, Edward Maydew, and Terry Shevlin, 2002, Taxes and
business strategy, 2nd edition, Prentice-Hall: Englewood Cliffs, NJ.
Stefanescu, Irina, 2005, Capital structure decisions and corporate pension plans, Working paper, University
of North Carolina
Tucker, Alan, 2002, Trafficking in foreign tax credits, Global Finance Journal 13, No. 2, 1-15.
29
Figure 1. Mean effective tax rate for U.S. corporations.
The effective tax rate is defined as taxes paid from the statement of cash flows divided by
pretax net income. The tax rate is averaged over all firms in the S&P 500 or all firms on
NYSE/AMEX/Nasdaq (excluding REITs, ADRs, closed-end mutual funds, preferred
stocks, foreign stocks, unit investment trusts, and Americus Trust). The statutory
marginal tax rate for firms in the highest income tax bracket was 34% until 1992 and
35% thereafter.
30
25
20
15
03
04
20
02
01
00
99
98
97
NYSE/AMEX/Nasdaq
20
20
20
20
19
19
96
19
95
S&P500
19
93
94
19
19
91
90
89
92
19
19
19
19
19
88
10
19
Effective Tax Rate (%)
35
Figure 2. Time Series of debt ratios for tax shelter firms versus matched firms
Panel A plots debt-to-assets ratios for shelter firms and matched firms. Matched firms are in the
same industry as the shelter firm and have book assets, profitability, and market/book ratios that
are similar to shelter firm ratios. Panel B is similar but is based on debt-to-market value (where
market value is market equity plus book debt). The shelter firms allegedly participate in tax
shelters in year 0. For multi-year shelters, the debt ratio is averaged for all years the shelter is
active and presented as year 0. Year –1 is the year before the shelter begins and +1 is the year
after the shelter ends.
Panel A: Debt/Assets
0.3
0.25
0.2
0.15
0.1
-8
-7
-6
-5
-4
-3
-2
Match firms
-1
0
1
2
3
2
3
Shelter firms
Panel B: Debt/Market Value
0.3
0.25
0.2
0.15
0.1
-8
-7
-6
-5
-4
-3
-2
Match firms
-1
0
1
Shelter firms
31
Table 1. Sample of tax shelter firms
This table identifies the 43 publicly traded corporations, involving 44 total shelters, that
are identified as tax shelter users in our sample. One firm, Compaq Computer
Corporation, was alleged to use two separate shelters (transfer pricing and cross border
dividend capture). The total number of shelter years is 152. Case citations are provided
when the corporation has litigated against the government. Each firm received a Notice of
Deficiency from the government.
________________________________________________________________________________________________
Firm
Shelter type
Years
Case Citation (if applicable)
Colgate-Palmolive
CPIS (a)
1988
73 T.C.M. (CCH) 2189, 2215
AlliedSignal
CPIS
1990-1992
76 T.C.M. (CCH) 325
American Home Products
CPIS
1990-1993
167 F. Supp. 2d 298
Brunswick
CPIS
1990-1991
78 T.C.M. (CCH) 684
Borden Inc.
CPIS
1995
12927-95 (b)
Compaq Computer
CBDC/TP (c)
1991-1992
277 F.3d 778
IES Industries
CBDC
1991-1992
253 F.3d 350
Florida Power and Light
LR (d)
1992-1994
80 T.C.M. (CCH) 686
Seagate Tech Inc.
TP
1981-1987
102 T.C. No. 9
St. Jude Medical
TP
1981-1983
34 F.3d 1394
Microsoft Corp.
TP
1987-1989, 1991 75 T.C.M. (CCH) 1747
Bausch & Lomb
TP
1983-1987
T.C. Memo 1996-57
National Semiconductor
TP
1976-1981
67 T.C.M. (CCH) 2849
Exxon Corp.
TP
1980-1982
752 F.2d 650
Intel Corp.
TP
1978-1980
76 F.3d 976
Boeing Co.
TP
1979-1987
258 F.3d 958
Archer-Daniels-Midland Co. TP
1975-1978
37 F.3d 321
Phillips Petroleum
TP
1979-1982
104 T.C. 256
Perkin -Elmer Corp.
TP
1975-1981
66 T.C.M. (CCH) 634
Computervision Corp.
TP
1981
96 T.C. 652
Sunstrand Corp.
TP
1977-1978
96 T.C. 226
Brown-Forman Corp.
TP
1981, 1983
94 T.C. 919
Chevron Corp.
TP
1977-1978
104 T.C. No. 35
American Electric Power
COLI
1990-1996
136 F. Supp. 2d 762
Winn-Dixie
COLI
1993
254 F.3d 1313
Dow Chemical Co.
COLI
1989-1991
250 F. Supp. 2d 748
CM Holdings Inc.
COLI
1991-1994
254 B.R. 578
Bmc Software Inc.
OIPH (f)
1993
73 F. Supp. 2d 751
W.R. Grace & Co.
COLI
1989-1998
Hershey Foods Corp.
COLI
1989-1998
Western Resources Co.
COLI
1992-1993
Hillenbrand Industries Inc. COLI
1996-1998
Donnelly RR and Son Inc. COLI
1990-1998
Ruddick Corp.
COLI
1993-1998
National City Corp.
COLI
1990-1995
AmSouth Bancorp
LILO (g)
1998-1999
FleetBoston Financial Corp. LILO
1995-1997
BB&T Corp.
LILO
1996-1998
Delta Air Lines
CLAS (h)
2000
Whirlpool Corp.
CLAS
2000
Clear Channel Com. Inc.
CLAS
2000
WorldCom Inc.
CLAS
2000
Tenet Healthcare Corp.
CLAS
2000
-----------------------------------------------------------------------------------------------------------------------------------------(a) CPIS: Contingent payment installment sale.
(e) COLI: Company owned life insurance.
(b) Tax Court Doc. No. only as the Borden case never went to trial. (f) OIPH: Offshore intellectual property haven.
(c) CBDC: Cross border dividend capture. TP: Transfer pricing.
(g) LILO: Lease in/lease out.
(d) LR: Liquidation/re-contribution.
(h) CLAS: Contested liability acceleration strat.
32
Table 2. Summary statistics
Summary statistics are presented for the years that the shelter is active. If the shelter
lasted multiple years, the statistics are averaged across all of the years that the shelter was
allegedly active. Matched firms are in the same industry as the shelter firm and have book
assets within +/- 25 percent and profitability within +/- 50% of the shelter firm’s ratios in
the same year. All numbers are means except for tax deficiency/assets. The market/book
ratio is market equity plus book debt, the sum divided by book assets. Collateralization is
measured as plant, property, and equipment divided by book assets. ROA is the return on
assets, measured by net income divided by assets. Debt/Assets is total debt divided by
total assets. Deficiency is the dollar amount of taxes (divided by asset value) that the
Internal Revenue Service claims that the taxpayer is deficient, due to the tax shelter, in its
actual tax payments. Therefore, a $1 Deficiency would be produced by a tax shelter
deduction equal to $1/(corporate tax rate).
Assets (millions $)
Sales (millions $)
Shelter firms
12,440
7,687
Match firms
10,728
6,581
Difference
1712
1106
Mkt/Book
(PPE+inventory)/Assets
ROA
FedTaxPaid/Pre-tax inc.
2.99
0.5
0.068
0.234
2.26
0.518
0.068
0.272
0.73 *
-0.018
0
-0.038
Debt/Assets
Debt/Mkt value
0.19
0.138
0.274
0.23
-0.084 ***
-0.092 ***
Deficiency/Assets (median)
0.031
n.a.
*** difference is statistically significantly different from zero (based on a differences in
means test) at a 1% confidence level, ** at 5% confidence level, * at 10% confidence
level.
33
Table 3. Regression analysis investigating whether shelter firms use less debt
OLS regressions are performed to identify the factors that affect a given company's debtto-assets ratio. There are two observations for each tax shelter firm in the sample: one for
the tax shelter firm, and one for the matched firm, where the matching identifies sameindustry firms in the same years with assets within +/- 25% of shelter firm t=-1 assets and
+/- 50% of shelter firm profits. If there is more than one match that meets these criteria,
the mean of the values for a given variable is used. If the shelter is active more than one
year, the mean value for all years that the shelter is active is used in the regression. The
dependent variable is debt-to-assets, which is book debt divided by book assets. Shelter
dummy equals one if the firm has an active tax shelter and zero if the firm is a match
firm. A negative shelter coefficient indicates that tax shelter firms have lower debt ratios,
all else equal. Lag5(debt) is the debt ratio from five years before the shelter became
active. Mkt/book is the market value of the firm (market equity plus book debt) divided
by book assets. Div-pay dummy equals one if the firm pays dividends. ROA is net
income divided by book assets. Collateral/assets equals inventory, plant, property, and
equipment divided by book assets. Time trend equals the calendar year. The right-hand
side variables are all lagged one year, except for Lag5(debt) and the time trend. Robust t
statistics adjusted for year clustering and heteroskedasticity are in parentheses.
Intercept
Shelter active
dummy
Lag5(debt)
Sales
(x100,000)
Mkt/book
(x10)
Div-pay
dummy
ROA
Collateral
/assets
Time Trend
Adj-R2
N
Main
specification
(1)
0.082**
(2.313)
-0.055***
(-3.523)
0.357***
(4.794)
-0.034
(-0.351)
0.059***
(2.697)
0.065**
(2.340)
-0.385*
(-1.879)
0.101***
(2.958)
54.2%
76
(2)
-9.353***
(-4.872)
-0.052***
(-3.791)
0.327***
(4.504)
-0.046
(-1.031)
0.041*
(1.805)
0.073**
(2.511)
-0.150
(-0.749)
0.147***
(4.687)
0.005***
(4.936)
63.4%
76
(3)
0.102***
(2.805)
-0.055***
(-3.423)
0.387***
(4.724)
(4)
0.171***
(3.260)
-0.075***
(-4.203)
0.052**
(1.964)
-0.293*
(-1.765)
0.087**
(2.397)
-0.057
(-0.482)
0.047
(1.045 )
0.050
(1.246)
-0.594***
(-2.656)
0.165***
(4.775)
53.2%
78
38.7%
84
*** means statistically significantly different from zero at a 1% confidence level, ** at 5% confidence
level, * at 10% confidence level
34
Table 4. Robustness analysis of the effect of tax shelters on corporate debt policy
OLS regressions are performed to determine which factors affect a given company's debt
ratio. The caption to Table 3 applies with the following exceptions. The dependent
variable is debt/assets except in column (1), where debt/market value is used. Market
value is measured as book debt plus market equity. In column (2), the shelters that lead to
capital losses (CPIS and LR) and the shelter that might lead to an increased use of debt
ratio (COLI) are deleted, leaving only the shelters that most likely lead to interest-like
deductions. In column (3), one year-lagged ROA is replaced with the cumulative ROA
from t=-5 to t=-1. In columns (4) and (5), rather than averaging all years in which a
shelter is active into a single observation (as is done in column (1), (2), and (3), and in
Table 3), a full panel of data is used, with each firm-year being represented by its own
observation. Also, in columns (4) and (5), rather than averaging all matched firms into a
single observation (as is done in column (1), (2), and (3), and in Table 3), all match
observations are used, so each matched firm-year is represented by its own observation.
In column (5), the Shelter firm dummy equals one for shelter firms in all years (even
when the shelter is not active). Robust t statistics adjusted for year clustering and
heteroskedasticity are in parentheses.
Intercept
Shelter active
dummy
Shelter firm
Dummy
Lag5(debt)
Sales
(x100,000)
Mkt/book
(x10)
Div-pay
dummy
ROA
ROA from
t=-5 to t=-1
Collateral
/Assets
Adj-R2
N
Dep. Var=
debt/mktval
(1)
0.041
(1.225)
-0.053***
(-2.840)
No CPIS,
LR, COLI
shelters (2)
0.155***
(3.816)
-0.074***
(-3.648)
0.351***
(5.144)
0.032
(0.385)
0.018
(0.432)
0.074***
(3.232)
-0.248
(-1.142)
0.308***
(2.884)
-0.074
(-1.347)
0.087***
(4.908)
0.038
(1.224)
-0.721***
(-5.128)
0.073*
(1.704)
42.8%
76
(3)
0.091**
(2.341)
-0.058***
(-3.691)
0.341***
(4.290)
-0.034
(-0.355)
0.062***
(2.752)
0.059*
(1.881)
-0.051**
(-2.217)
0.096***
(3.058)
53.0%
76
0.076
(1.534)
63.3%
50
Multiple
match obs.
(4)
0.090***
(6.342)
-0.031***
(-3.900)
0.525***
(18.114)
-0.059***
(-3.158)
0.019
(0.939)
-0.006
(-0.470)
-0.283***
(-5.092)
Multiple
match obs.
(5)
0.104***
(6.753)
-0.020**
(-2.418)
-0.021***
(-3.009)
0.525***
(18.129)
-0.056***
(-3.134)
0.021
(1.015)
-0.010
(-0.757)
-0.308***
(-5.579)
0.093***
(6.768)
40.3%
1140
0.094***
(6.771)
40.7%
1140
*** means statistically significantly different from zero at a 1% confidence level, ** at 5% confidence
level, * at 10% confidence level
35
Table 5. Debt issuance decision
A logistic regression is performed to determine the factors that affect debt issuance. The
observations used in this analysis are from both shelter firms and matched firms. The
dependent variable equals one if a firm issues debt and equals zero otherwise. Data are
included from eight years before the shelter is active until the last year the shelter is
active (i.e., from t=-8 to t=0). Shelter dummy equals one for shelter firms and equals zero
if the firm is a match firm, so a negative coefficient means that shelter firms are less
likely to issue debt. Lag5(debt) is the debt ratio from five years before the shelter became
active. Mkt/book is the market value of the firm (market equity plus book debt) divided
by book assets. Div-pay dummy equals one if the firm pays a dividend. ROA is net
income divided by book assets. Collateral/assets equals inventory, plant, property, and
equipment divided by book assets. Marginal effects (slopes) for independent variables
(except Intercept) are reported. Robust t statistics adjusted for year clustering and
heteroskedasticity are in parentheses.
Intercept
Shelter dummy
Lag5(debt)
Log(Sales)
Mkt/book
Div-pay dummy
ROA
Collateral/Assets
N
Model AIC
p-value
Correct
predictions
Issue Debt
in t=0
-5.763***
(-8.331)
-0.137***
(-4.122)
1.017***
(7.467)
0.063***
(6.284)
0.029**
(2.046)
-0.026
(-0.436)
0.284
(0.923)
0.578***
(8.430)
778
0.0001
78%
*** statistically significantly different from zero at a 1% confidence level, ** at 5% confidence level, * at
10% confidence level
36