Embed Size (px)
Citation preview
Tuck School of Business at Dartmouth
Working Paper No. 2005-24
Using Bilateral Advance Pricing Agreements to Resolve Tax Transfer Pricing
Disputes
ANJA DE WAEGENAERE Tilburg University, Center for Economic Research (CentER)
RICHARD C. SANSING
Tuck School of Business at Dartmouth
J. WIELHOUWER Tilburg University, CentER
July 21, 2005
This paper can be downloaded from the Social Science Research Network Electronic Paper Collection:
http://ssrn.com/abstract=766044
Using Bilateral Advance Pricing Agreements to Resolve Tax Transfer Pricing Disputes
ABSTRACT: We investigate the use of bilateral advance pricing agreements (BAPAs)
to resolve transfer pricing disputes between a taxpayer and two tax authorities. BAPAs
are designed to protect firms from double taxation while reducing expected compliance
costs (audits costs plus BAPA implementation costs). We identify settings in which we
expect BAPAs to arise and investigate the effect of the program on compliance costs. We
show that a reduction in expected compliance costs is a necessary but not sufficient
condition for an agreement to be reached. Agreements are more likely to arise when the
amount of income potentially subject to double taxation is low and the difference in tax
rates between the two countries is high. We also show that the existence of the BAPA
program can increase tax compliance costs in settings in which the absence of a BAPA
conveys information to the tax authorities, inducing them to audit a taxpayer without a
BAPA more intensively than they would have had the BAPA program not existed.
Although this outcome is socially inefficient, it increases the expected net payoff (taxes
collected less tax compliance costs) of the country with the higher tax rate.
Keywords: Bilateral advance pricing agreements, transfer pricing, tax compliance, game theor
1
I. INTRODUCTION
Income earned by a multinational enterprise can be subject to double taxation if
multiple tax authorities assert the right to tax the income. In particular, double taxation
arises to the extent that governments use different transfer pricing rules to allocate
income between countries. In the 1997, 1999, and 2001 Ernst & Young LLP Transfer
Pricing Global Surveys, over 80% of multinationals identified “double tax relief” as an
important international tax issue because transfer pricing adjustments typically result in
double taxation (Ackerman and Hobster 2001).
Governments also have severe difficulties with transfer pricing disputes. Transfer
pricing cases are factual intensive investigations featuring both legal and economic
analysis and involving both domestic and foreign entities. Billions of dollars of tax
revenue are at stake; for example, the multinational pharmaceutical firm
GlaxoSmithKline is fighting a $2.7 billion U.S. tax deficiency notice arising from
transfer pricing disputes involving its 1989-1996 tax years (Sullivan 2004). The rapidly
growing volume of international transactions between related entities has put the
traditional system of international tax compliance under considerable strain.
In response to this mutually unsatisfactory situation, the U.S. has developed and
many countries have implemented a new model of tax compliance: the bilateral advance
pricing agreement (BAPA). In a BAPA, the taxpayer approaches the governments prior
to engaging in a transaction with a foreign related party to negotiate a mutually
acceptable transfer pricing method. In a BAPA, the taxpayer provides detailed
information regarding the proposed transaction to the governments. In return, the
2
governments commit to a transfer pricing method that ensures that the taxpayer will not
be subject to double taxation.
Williams (1996) describes the APA program, the procedures involved, and the
costs and benefits of the program. Ring (2000) also describes the program and evaluates
it in terms of legal theories of administrative law. Ackerman (2001) examines how the
program has evolved into a more standardized format over time.
In this paper, we investigate the use of BAPAs using a game-theoretic approach to
tax compliance. We examine two research questions. First, under what circumstances do
we expect firms and governments to enter into a BAPA, and under what circumstances do
we expect firms to follow the traditional approach involving auditing and litigation?
Second, what are the effects of the BAPA program on aggregate tax compliance costs,
which we define to be the sum of expected audit costs and the cost of negotiating and
implementing the BAPA?
With respect to our first research question, we first find that a reduction in
aggregate tax compliance costs is a necessary but not sufficient condition for a BAPA to
occur. The fact that by definition a BAPA requires the parties to agree upon a single
transfer price makes an agreement infeasible even though aggregate tax compliance costs
would be lower with an agreement. Second, we find that a BAPA is more likely to be
used when the amount of income subject to double taxation because of asymmetric
transfer pricing rules is low. Although the rationale for the BAPA program is to provide a
mechanism by which the firm can avoid double taxation, the ability of the firm and the
countries to find a mutually acceptable transfer price decreases as the amount of income
potentially subject to double taxation increases. Third, we find that BAPAs occur less
3
frequently when the tax rates of the two countries are similar. Expected audit costs are
lower when the tax rates are similar because the country with the higher tax rate does not
have to audit as frequently to deter the taxpayer from shifting its taxable income toward
the lower tax rate country via its choice of transfer price. When expected audit costs are
lower, there is less to gain from a BAPA and so it is used less frequently.
With respect to our second research question, we find that the effect of the BAPA
program on compliance costs depends on whether the absence of a BAPA reveals
information about the taxpayer that influences the governments’ audit decisions. Even
though the tax auditors are not able to use disclosures made by the taxpayer in the BAPA
process in other tax compliance proceedings the absence of a BAPA may reveal taxpayer
information.1 If the absence of a BAPA reveals no taxpayer information, the BAPA
program reduces aggregate audit costs; otherwise, the parties would not have an incentive
to enter into the BAPA in the first place. However, if the absence of a BAPA does reveal
taxpayer information, the existence of the program can increase aggregate tax compliance
costs by inducing the countries to increase the frequency with which they audit the
taxpayer, compared to what they would have done if the BAPA program had not existed.
Despite the increase in expected compliance costs, the BAPA program increases the
expected payoff (taxes collected minus compliance costs) of the country with the higher
tax rate. Therefore, we expect that high tax countries will most aggressively promote the
use of BAPAs.
In section 2 we review the literature that relates to our paper. Section 3
characterizes the equilibrium behavior of the taxpayer and the two tax authorities when a
BAPA does not occur. Section 4 characterizes the equilibrium behavior of the taxpayer
4
and the two tax authorities when the parties have an opportunity to enter a BAPA. In
section 5 we discuss the empirical implications of our results. We also examine the
effects of the BAPA program on tax compliance costs. Section 6 concludes.
II. RELATED LITERATURE
Tomohara (2004) is the only other paper to formally investigate the economic
effects of BAPAs. The focus of that paper is on the firm’s production decisions in the
presence of a BAPA, whereas our paper focuses on whether the taxpayer and
governments will enter into a BAPA and how the presence of the BAPA program affects
their reporting and auditing decisions.
Our paper relates to several topics that have been thoroughly explored in the tax
literature. Many papers have modeled the interaction between the taxpayer and the tax
authority as a game between a wealth-maximizing taxpayer and a tax authority trying to
maximize government revenues net of audit costs (Graetz, Reinganum and Wilde 1986;
Reinganum and Wilde 1986; Beck and Jung 1989; Sansing 1993; Rhoades 1997;
Rhoades 1999; Mills and Sansing 2000; Feltham and Paquette 2002.) De Waegenaere,
Sansing, and Wielhouwer (2005) extend the standard tax compliance model to a
multinational setting in which the taxpayer interacts with two tax authorities enforcing
the transfer pricing rules of countries with different tax rates. We use a similar approach
to determine the equilibrium behaviors of the taxpayer and the tax authorities in case a
BAPA is not entered into.
Our study of the BAPA process relates to research on alternatives to litigation as a
means of resolving legal disputes. Samuelson (1991) investigates the use of final-offer
arbitration under asymmetric information. Sansing (1997) applies a variation of
5
Samuelson’s model to a tax litigation setting. He finds settings in which the taxpayer and
the tax authority will only agree to arbitration for certain realizations of their private
information. Our results are similar in that whether the parties enter into a BAPA depends
on the taxpayer’s private information.
As BAPAs are used to resolve tax transfer pricing disputes, our paper relates to
that literature as well. In particular, the fact that the two countries have asymmetric tax
transfer pricing rules that would result in double taxation in absence of a BAPA gives the
taxpayer the incentive to enter into a BAPA. It also provides the conditions under which
the BAPA process will break down. Other papers that have investigated the
distributional and efficiency effects of countries employing asymmetric tax transfer
pricing rules include Halperin and Srinidhi (1987) and Elitzur and Mintz (1996). There is
a much larger literature on distributional and efficiency effects of tax transfer pricing
rules when both countries use the same transfer pricing rule; Baldenius, Melumad, and
Reichelstein (2004) review this literature in their paper.
III. THE REPORTING GAME
A firm F earns income generated in a transaction between related parties
operating in two different countries, H and L. We normalize the amount of income to one,
without loss of generality. The two countries impose tax rates of tH and tL, respectively,
LHtt ! . The tax laws in the two countries are different with respect to the measurement
of taxable income associated with the transaction. Under H’s law, the income taxed by H
is yH, 10 !! Hy , and the income taxed by L is 1−yH. Under L’s law, the income taxed by
L is yL, 10 !! Ly and the income taxed by H is 1−yL. We assume that each country
measures income in a manner that is favorable to itself, so 1!+ LH yy . We assume that
6
yH and yL are realizations from a probability distribution f(yH , yL). We let x represent the
amount of income which would be subject to double taxation if each country were to
apply its own law, so 1!+= LH yyx . We assume that x has strictly positive probability
density on the interval [0,1]. We let µ=][xE .
The firm privately observes the realization (yH , yL) and files a pair of internally
consistent tax reports (r,1−r), where r is the amount of income from the related party
transaction reported to H and 1−r is the amount of income reported to L. We let
!
r ="yH + (1#" )(1# yL ), where α is the weight that F puts on country H’s law and
!
1"#
is the weight placed on L’s law. We do not constrain α to be between zero and one,
although in equilibrium α is never less than zero nor greater than one. We illustrate the
firm’s reporting problem in Figure 1.
[INSERT FIGURE 1 HERE]
Each country has the option to audit the taxpayer’s report. The cost of an audit
is
!
K > 0 . We assume that HtK µ< , which ensures that auditing is not so costly as to deter
H from auditing even if F reports in accordance with L’s law. The audit detects the
difference between the reported income and the income measured under that country’s
tax law. Country H (L) conducts an audit of the taxpayer’s related party transaction with
probability δH (δL). We assume that r and 1−r are contained in a much more extensive
report that aggregates the taxable income from all of the firm’s transactions. The variance
of the aggregated report is sufficiently large that the report contains essentially no
information regarding the realization of (yH , yL). Therefore, the decision to audit the
taxpayer’s related party transaction does not depend on the report. All aspects of the
model are common knowledge except for the realization (yH , yL).
7
The payoff to country H given a particular realization (yH , yL) and report r is:
])([ KrytrtP HHHHH !!+= " (1)
The payoff to country L given a particular realization (yH , yL) and report 1−r is:
]))1(([)1( KrytrtP lLLLL !!!+!= " (2)
The payoff to the firm given a particular realization (yH , yL) and reports (r,1−r) is:
)]}1([)1{()]([ ryrtryrtP LLLHHHF !!+!!!+!= "" (3)
As an alternative to dealing with the conflict in the countries’ tax laws via the
audit and litigation process, the firm can deal with the conflict in the countries’ tax laws
cooperatively via a bilateral advance pricing agreement (BAPA).
We first characterize the behavior of the players in case they do not enter into a
BAPA. We define a Nash equilibrium for this reporting game as a reporting strategy
),( LH yyr that maximizes the firm’s expected payoff given the equilibrium audit
strategies of H and L; an audit strategy H
! for country H that maximizes its expected
payoff given the firm’s reporting strategy; and an audit strategy L
! for country L that
maximizes its expected payoff given the firm’s reporting strategy.
Proposition 1 establishes the optimal strategies of the players in the reporting
game. Because the reporting game only occurs if the players have not entered into a
BAPA, the countries may be able to infer information regarding the value of the amount
of income potentially subject to double taxation, x, from the fact that the firm is filing a
tax return without a BAPA. We let µ̂ denote the conditional expectation of x, given that
a BAPA is not entered into. We assume in the following proposition that µµ !ˆ ; we will
8
show in the proof of Proposition 2 that this assumption indeed holds in this game because
in equilibrium µ̂ is always weakly greater than
!
µ.
There are two equilibria depending on K, tax rates, and µ̂ . If the cost of auditing
(K) is sufficiently low compared to the tax rates and the expected amount of income that
is subject to double taxation ( µ̂ ), both H and L audit all reports and F reports part of its
income according to L’s rules and part according to H’s rules. We refer to this as the
heavy audit equilibrium. If the cost of auditing is sufficiently high, country L never audits
and country H uses a mixed audit strategy that makes F indifferent regarding its report.
The firm reports part of its income according to L’s rules and part according to H’s rules
so as to make H indifferent between auditing and not auditing the report. We refer to this
as the light audit equilibrium.
Proposition 1 Assume that .ˆ µµ !
(a) If LH
LH
tt
ttK
+!µ̂ , then the following strategies constitute a Nash equilibrium:
(i) F reports r to country H and 1−r to country L, where !"
#$%
& '(
H
H
Lt
Kt
t
K
µ
µ
µ)
ˆ
ˆ,
ˆ.
(ii) H always audits, i.e. 1=H
! .
(iii) L always audits, i.e. 1=L
! .
(b) If LH
LH
tt
ttK
+!µ̂ , then the following strategies constitute a Nash equilibrium:
(i) F reports r to country H and 1−r to country L, where H
H
t
Kt
µ
µ!
ˆ
ˆ "= .
(ii) H audits with probability H
LH
H
t
tt !=" .
(iii) L never audits, i.e. 0=L
! .
9
In the heavy audit equilibrium, the cost of auditing is so low relative to the
countries’ tax rates and the expected amount of income subject to double taxation that
one of the countries will have an optimal strategy of always auditing. To deter the firm
from reporting all of its income according to the rules of that country, the other country
should also audit all reports. Given that both countries will audit the report, the only
equilibrium reporting strategy for the firm is to report its income in a way that provides
both countries with an incentive to audit.
In the light audit equilibrium, the cost of auditing is sufficiently high that country
L has an optimal strategy of never auditing. Country H audits just enough to deter the
firm from reporting its income according to country L’s law. This audit frequency is
decreasing in the tax rate ratio H
L
t
t . The firm in turn reports just enough of its income in
accordance with country H’s law to deter it from auditing all reports. The fraction of the
income reported in accordance with country H’s law is increasing in tH and µ̂ and
decreasing in the audit cost K.
IV. BILATERAL ADVANCE PRICING AGREEMENTS
In this section, we characterize the actions of the players when they have the
opportunity to enter a BAPA. In a BAPA, the firm approaches countries H and L
regarding the possibility of agreeing to an allocation of the income in which z is allocated
to H and 1−z is allocated to L. This allocation ensures that double taxation does not occur.
The firm reveals the realization (yH , yL) to H and L for those realizations of (yH , yL) for
which there exists an allocation z for which all parties would be willing to enter into a
BAPA. The cost to all parties of negotiating and implementing a BAPA is C≥ 0.
10
We define a feasible BAPA to be one in which the allocation of income (z,1−z)
and a division of the total costs C are such that each player receives a (weakly) higher
payoff under the BAPA than without the BAPA. We allow any nonnegative allocation of
the cost among the three players. The players’ payoffs before the allocation of C are:
LHFtzztA )1( !!!= , (4)
HHztA = , (5)
LLtzA )1( != . (6)
Feasibility of a BAPA requires three conditions to be satisfied for a BAPA to be
feasible: AF≥ PF, AH≥ PH, and AL≥ PL. Furthermore, we require AF+AH+AL-C≥
PF+PH+PL; we refer to a BAPA as desirable if this latter condition is satisfied. Because
tax payments are zero-sum wealth transfers between the three players, this condition is
satisfied if the BAPA cost C is less than the expected of 2K under the heavy audit
equilibrium and H
LH
t
ttK )( ! under the light audit equilibrium.
The interaction among the players when the BAPA program exists proceeds as
follows. First, the firm observes (yH , yL). We assume that the players will agree to a
BAPA when one is both feasible and desirable. F then files its tax returns in a manner
consistent with the BAPA, and H and L accept the returns as filed. If a BAPA is either
infeasible or undesirable, the firm files its tax returns without a BAPA, then the players
play the reporting game described in section 2.2
The feasibility of a BAPA will sometimes depend on x, the level of income
potentially subject to double taxation. The desirability of a BAPA depends on the costs of
implementing a BAPA (C) and on whether both governments would audit the report in
11
absence of a BAPA (the heavy audit equilibrium) or whether H would audit with some
probability and L would never audit (the light audit equilibrium). That decision in turn
depends on whether LH
LH
tt
ttK
+!µ̂ or
LH
LH
tt
ttK
+!µ̂ , as shown in Proposition 1. We let
!
"( x) be an indicator variable that equals one if a feasible and desirable BAPA exists and
zero otherwise.
Three types of outcomes are possible. If the cost of the BAPA exceeds the
countries’ expected aggregate audit costs, then a BAPA would not be desirable, so that
0)x( =! for all x, and the players would not enter a BAPA for any value of x. When
BAPAs never occur, the tax authorities cannot infer any information about x from the
absence of a BAPA. In that case, the reporting game is played as described in Proposition
1 with µµ =ˆ .
Second, if the players would choose their heavy audit strategies in the reporting
game (i.e., when LH
LH
tt
ttˆK
+!µ ) and the cost of a BAPA is less than the aggregate audit
costs (C<2K), a BAPA is desirable. However, if the amount of income that is potentially
subject to double taxation ( )1!+= LH yyx is too high, there does not exist a transfer
price leading to an income allocation (z, 1−z) that makes both H and L better off under a
BAPA than they would be under the heavy audit equilibrium, even if F were to bear all of
the cost of the BAPA. The largest value of x for which a BAPA is feasible, which we
denote x*, is LH
LH
tt
ttKx
)(* += . If *
xx ! , F realizes that a BAPA is both feasible and
desirable, so it proposes a BAPA. If *xx ! , F realizes that a BAPA is infeasible and so
the players will choose their heavy audit strategies in the reporting game. Therefore,
12
!
"( x) = 1 if *xx ! and
!
"( x) = 0 if
!
x " x*. It is in this situation that the governments
update their beliefs regarding x given the absence of a BAPA. Because LH
LH
tt
ttKx
)(* +=
and
!
ˆ µ = E[ x | x " x*] ,
LH
LH
tt
ttK
+!µ̂
is equivalent to LH
LH
tt
ttK
+! .
Third, if the players would choose their light audit strategies in the reporting game
(i.e., when LH
LH
tt
ttˆK
+!µ ) and the cost of a BAPA is less than H’s expected audit costs
!
C " K#H
=K(t
H$ t
L)
tH
%
& '
(
) * , a BAPA is desirable and feasible for all values of x, so that
!
"( x) = 1 for all x; therefore, the players will negotiate a BAPA for all realizations of x.
The reporting game is never played in this case. This equilibrium is supported by the off-
equilibrium path beliefs of the tax authorities of H and L that .1]0)x(|x[Eˆ === !µ
Proposition 2 characterizes the indicator functions )x(! that are consistent with
the decision rule of the parties to enter a BAPA if and only if one is both feasible and
desirable. For the parameter values LH
LH
LH
LH
tt
ttK
tt
tt
+!!
+
µ and KCt
ttK
H
LH 2)(
!!" ,
two different functions )(x! are both consistent with that decision rule. This occurs
because whether a BAPA is feasible and desirable (i.e. whether 1)( =x! or 0)( =x! )
depends on whether the parties would choose their heavy audit or light audit strategies in
the reporting game, which in turn depends on µ̂ . However, µ̂ itself depends on whether
BAPAs are feasible and desirable (i.e. the function )(x! ).
13
Proposition 2
(a) If KC 2! , then
!
"( x) = 0 for all x.
(b) If LH
LH
tt
ttK
+! and KC 2! , then
!
"( x) = 1 if *xx ! and
!
"( x) = 0 if
!
x " x*.
(c) If
!
K "µt
HtL
tH
+ tL
and KCt
ttK
H
LH 2)(
!!" , then
!
"( x) = 0 for all x.
(d) If LH
LH
tt
ttK
+! and
H
LH
t
ttKC
)( !" , then
!
"( x) = 1 for all x.
Propositions 3-5 characterize the Nash equilibria that arise when the players have
the opportunity to enter into a BAPA instead of playing the reporting game. Proposition 3
characterizes behavior when C is sufficiently high. When K2C ! , the cost of a BAPA is
so high that a BAPA is not desirable, even if the players would choose their heavy audit
strategies in the reporting game. Since BAPAs are never desirable, the tax authorities
cannot infer any information about x from the absence of a BAPA. The equilibrium then
depends on K, as described in Proposition 1 with µµ =ˆ .
Proposition 3 If
!
C " 2K , then the Nash equilibria are:
(a) if
!
K "µt
HtL
tH
+ tL
, then
!
"( x) = 0 for all x and the players choose their heavy audit
strategies in the reporting game.
(b) if ,
LH
LH
tt
ttK
+!µ then
!
"( x) = 0 for all x and the players choose their light audit
strategies in the reporting game.
14
Proposition 4 characterizes the equilibria in which C takes on intermediate values
!
K(tH " tL )
tH# C # 2K
$
% &
'
( ) . For these values of C, a BAPA is undesirable when the players
would choose their light audit strategies but is desirable when the players would choose
their heavy audit strategies. However, in the latter case, a BAPA is only feasible for
values of *xx ! . If the players enter a BAPA for *
xx ! , then heavy auditing is the
optimal response in the absence of a BAPA.
Proposition 4 If
!
K(tH " tL )
tH# C # 2K , then the Nash equilibria are:
(a) if
!
K "tHtL
tH
+ tL
, then
!
"( x) = 1 if *xx ! and
!
"( x) = 0 if
!
x " x*, and the players
choose their heavy audit strategies in the reporting game;.
(b) if
!
K "µt
HtL
tH
+ tL
, then
!
"( x) = 0 for all x and the players choose their light audit
strategies in the reporting game.
An important implication of Proposition 4 is that either type of equilibrium can
arise when
!
µtHtL
tH
+ tL
" K "tHtL
tH
+ tL
and
!
K(tH " tL )
tH# C # 2K . If the firm anticipates that
the countries will choose their heavy audit strategies, then it should seek a BAPA when
*xx ! , in which case heavy audit strategies are appropriate. On the other hand, if the
firm anticipates that the countries will choose their light audit strategies, it should never
seek a BAPA, in which case light audit strategies are appropriate. While we offer no
prediction regarding which equilibrium will be played, aggregate payoffs are higher in
the “no BAPA/light audit” equilibrium while country H’s payoff is greater in the “some
15
BAPA/heavy audit” equilibrium. This occurs because country H has an expected payoff
equal to
!
tH yH "K in either the heavy or light audit equilibrium and has a minimum
payoff of
!
tH yH "K in a BAPA. We predict that high-tax countries will aggressively
encourage the use of BAPAs to resolve tax transfer pricing disputes.
Proposition 5 characterizes the Nash equilibria when C is low (H
LH
t
ttKC
)( !" ).
A BAPA is always desirable in that case. When
!
K "tHtL
tH
+ tL
, a BAPA is only feasible if
*xx ! , in which case the players would choose their heavy audit strategies in the absence
of a BAPA. When
!
K "tHtL
tH
+ tL
, a BAPA is feasible for all x.
Proposition 5 If H
LH
t
ttKC
)( !" , then the Nash equilibria are:
(a) if
!
K "tHtL
tH
+ tL
, then
!
"( x) = 1 if *xx ! and
!
"( x) = 0 if
!
x " x*, and the players
choose their heavy audit strategies in the reporting game;
(b) if
!
K "tHtL
tH
+ tL
, then
!
"( x) = 1 for all x and the off-equilibrium path beliefs are that
!
ˆ µ = 1.
Figure 2 illustrates the equilibrium outcomes for
!
tH
= 40%, tL
= 20% , and
5.0=µ .
[INSERT FIGURE 2 HERE]
16
V. DISCUSSION AND ANALYSIS
Empirical implications
We first consider the empirical implications of our analysis. We identify three
implications of our results. First, a reduction in expected tax compliance costs is a
necessary but not sufficient condition for a BAPA to occur. This occurs because the
desirability of a BAPA depends on the difference between the cost of a BAPA and the
aggregate expected audit costs of the countries, whereas the feasibility of a BAPA
sometimes requires that the amount of income potentially subject to double taxation be
low. All feasible BAPAs are desirable, but the converse does not hold.
Second, a BAPA is more likely to be entered into when the amount of income
subject to double taxation because of asymmetric transfer pricing rules is low, i.e. *xx ! .
Although the rationale for the BAPA program is to provide a mechanism by which the
firm can avoid double taxation, the fact that by definition a BAPA requires the parties to
agree upon a single transfer price makes the procedure infeasible when the countries’ tax
laws imply large differences in their respective transfer prices.
Third, when cost of auditing is sufficiently high, BAPAs occur less frequently
when the tax rates of the two countries are similar. When K is high and the tax rates are
similar, the low-tax country does not audit (because of the high audit cost) and the high-
tax country audits infrequently (because little auditing is required to deter the firm from
reporting using the low-tax country’s transfer price.) Therefore, a BAPA is only desirable
when the cost C of negotiating and implementing the BAPA is very low. Therefore
BAPAs are more likely to be desirable when the firm has a stronger incentive to shift
17
income to the lower-tax country due to large differences in tax rates, and less likely to be
desirable when the two countries have similar tax rates.
Efficiency implications of the BAPA program
We complete our analysis by comparing the cost of tax compliance (tax auditing
plus BAPA implementation costs) when the BAPA program exists to what the cost would
have been in absence of the program. We consider two cases. First, when either K is very
low !!"
#$$%
&
+'
LH
LH
tt
ttK
µ , very high !!"
#$$%
&
+'
LH
LH
tt
ttK , or BAPA costs are very high (
!
C " 2K ),
the reporting game is unaffected by the BAPA program. When BAPAs are entered into,
all the players’ payoffs weakly improve, because all feasible BAPAs are desirable.
However, for intermediate values of K !!"
#$$%
&
+''
+LH
LH
LH
LH
tt
ttK
tt
ttµ and
!
C " 2K , the
very existence of the BAPA program can change the audit strategies that occur in the
reporting game. When
!
K(tH " tL )
tH# C # 2K
$
% &
'
( ) , Proposition 4(a) shows that an equilibrium
exists in which the players choose their heavy audit strategies when a BAPA does not
occur but would have chosen their light audit strategies if the BAPA program did not
exist. This occurs because the tax audit authorities of countries L and H infer that the
potential level of double taxation is high from the absence of a BAPA. They infer
important taxpayer information from the absence of a BAPA even though they learn
nothing directly about the taxpayer from the BAPA process itself.
As Proposition 4(b) shows, this inefficiency need not arise, because it is still an
equilibrium for the firm to not seek a BAPA and for the players to choose their light audit
18
strategies in the reporting game. Because the “no BAPA and light audit” equilibrium
features lower aggregate compliance costs than the “some BAPA and heavy audit”
equilibrium, at least one player will prefer the equilibrium without BAPAs. However,
country H’s payoffs are higher in the equilibrium in which a BAPA arises for sufficiently
low x. We expect high tax rate countries to be the most aggressive in promoting BAPAs
for this reason; whether the BAPA program creates an inefficiency depends on whether
the high-tax country succeeds in getting the players to play its preferred equilibrium.
When BAPA costs are very low !!"
#$$%
& '(
H
LH
t
ttKC
)( , the efficiency implications of
the BAPA program are mixed. When the potential for double taxation is low ( )*xx ! , the
players enter into a BAPA, which is the efficient outcome. When the potential for double
taxation is high, however, a BAPA is not feasible and players choose their heavy audit
strategies, but would have chosen their light audit strategies if the BAPA program did not
exist. Therefore, total tax compliance costs increase.
Proposition 6 summarizes the implications of the existence of a BAPA program
on total tax compliance costs.
Proposition 6 The existence of a BAPA program
(a) decreases compliance costs for *xx ! if
LH
LH
tt
ttK
+!µ and KC 2! ;
(b) decreases compliance costs ifLH
LH
tt
ttK
+! and
H
LH
t
ttKC
)( !" ;
(c) increases compliance costs if LH
LH
LH
LH
tt
ttK
tt
tt
+!!
+
µ and
!
K(tH " tL )
tH# C # 2K
when the players choose the equilibrium strategies described in Proposition 4(a)
19
and has no effect on compliance costs when the players choose the equilibrium
strategies described in Proposition 4(b);.
(d) decreases compliance costs for *xx ! and increases compliance costs for *
xx !
if LH
LH
LH
LH
tt
ttK
tt
tt
+!!
+
µ and H
LH
t
ttKC
)( !" .
For combinations of parameter values not mentioned in Proposition 6, the BAPA
program has no effect on tax compliance costs. We illustrate the possibilities described in
Proposition 6 in Figure 3.
[INSERT FIGURE 3 ABOUT HERE]
VI. CONCLUSIONS
We have modeled the resolution of tax transfer pricing disputes as a strategic game
between the firm and two governments. BAPAs have emerged as an alternative means of
resolving these disputes. BAPAs are designed to protect firms from double taxation and
reduce tax compliance costs.
Our analysis yields three insights into when such agreements will arise. First, a
reduction in expected tax compliance costs is a necessary but not sufficient condition for
a BAPA to occur. Second, BAPAs are less likely to occur when the amount of income
potentially subject to double taxation because of asymmetric transfer pricing rules is high.
Third, BAPAs occur less frequently when the tax rates of the two countries are similar.
We also use investigate the effect of the BAPA program on tax compliance costs.
We find that the program can increase tax compliance costs. This occurs when the
absence of a BAPA conveys information to the tax authorities, inducing them to increase
the level of auditing over what is would have been in the absence of a BAPA program.
20
Although this is a socially inefficient outcome, it increases the expected payoff of the
high-tax country. Therefore, we expect that countries with the highest tax rates will most
aggressively promote the use of BAPAs.
21
REFERENCES
Ackerman, R. 2001. Will U.S. standardization erode key flexibility in APAs? Journal of
International Taxation 12 (February): 12
Ackerman, R. and J. Hobster. 2001. Transfer pricing practices, perspectives, and trends in
22 countries. Tax Notes International 24 (December 17): 1151-1158.
Baldenius, T., N. Melumad, and S. Reichelstein. 2004. Integrating managerial and tax
objectives in transfer pricing. The Accounting Review 79 (July): 591-615.
Beck, P., and W. Jung. 1989. Taxpayers’ reporting decisions and auditing under
information asymmetry. The Accounting Review 64 (July): 468-487.
De Waegenaere, A., R. Sansing, and J. Wielhouwer. 2005. Who benefits from
inconsistent multinational tax transfer pricing rules? Contemporary
Accounting Research (forthcoming).
Elitzur, R. and J. Mintz. Transfer pricing rules and corporate tax competition. Journal of
Public Economics 60 (June): 401-422.
Feltham, G. and S. Paquette. 2002. The interrelationship between estimated tax payments
and taxpayer compliance. Journal of the American Taxation Association 24
(Supplement): 27-45.
Graetz, M., J. Reinganum, and L. Wilde. 1986. The tax compliance game: Toward an
interactive theory of law enforcement. Journal of Law, Economics, and
Organization 2 (Spring): 1-32.
Halperin, R., and B. Srinidhi. 1987. The effects of U.S. income tax regulations’ transfer
pricing rules on allocative efficiency. The Accounting Review 62 (October):
686-706.
22
Mills, L., and R. Sansing. 2000. Strategic tax and financial reporting decisions: Theory
and evidence. Contemporary Accounting Research 17 (Spring): 85-106.
Reinganum, J., and L. Wilde. 1986. Equilibrium verification and reporting policies in a
model of tax compliance. International Economic Review 27 (October): 739-
760.
Rhoades, S. 1997. Costly false detection errors and taxpayer rights legislation:
Implications for tax compliance, audit policy, and revenue collections.
Journal of the American Taxation Association 64 (Supplement): 27-47.
Rhoades, S. 1999. The impact of multiple component reporting on tax compliance and
audit strategies. The Accounting Review 74 (January): 63-85.
Ring, D. 2000. On the frontier of procedural innovation: Advance pricing agreements and
the struggle to allocate income for cross border taxation. Michigan Journal of
International Law 21 (Winter): 143-234.
Samuelson, W. 1991. Final-offer arbitration under incomplete information. Management
Science 37 (October): 1234-1247.
Sansing, R. 1993. Information acquisition in a tax compliance game. The Accounting
Review 68 (October): 874-884.
Sansing, R. 1997. Voluntary binding arbitration as an alternative to Tax Court litigation.
National Tax Journal 50 (June): 279-296.
Sullivan, M. 2004. With billions at stake, Glaxo puts U.S. APA program on trial. Tax
Notes International 34 (May 3): 456-463.
23
Tomohara, A. 2004. Inefficiencies of bilateral advanced pricing agreements (BAPA) in
taxing multinational companies. National Tax Journal 57 (December): 863-
873.
Williams, E. 1996. Basics of the U.S. advance pricing agreement program. Tax Notes
International 13 (August 26): 723-731.
24
The firm’s reporting problem
Figure 1
Note: The bold line is the range of reports )r1,r( ! for which ].1,0[!"
25
Nash equilibrium outcomes for
!
tH
= 40%, tL
= 20%, µ = 0.5.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Always BAPA Never BAPA, light audit
BAPA if x<x*,
heavy audit Never BAPA, heavy audit
BAPA if x<x*, heavy audit
OR
Never BAPA, light audit
C
K
Figure 2
26
Effects of BAPA program on tax compliance for
!
tH
= 40%, tL
= 20%, µ = 0.5.
0 0.05 0.1 0.15 0.2 0.25 0.3
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
D
I
U
U
D if x<x*
I if x>x*
I or U
D if x<x*
U if x>x*
U
C
K
Figure 3
Note: D indicates that compliance costs are decreasing; I indicates that compliance costs are increasing; and U indicates that compliance costs are unaffected.
27
APPENDIX
Proof of Proposition 1
(a)LH
LH
tt
ttK
+!µ̂
(i) Given the audit strategies 1=
H! and 1=
L! , the firm is indifferent over all
possible values of α. Substituting )1)(1( LH yyr !!+= "" into equation (3) yields:
xtytytP HLLLHF !!!!= )1(
which does not depend on α.
(ii) Given the firm’s equilibrium reporting strategy !!"
#$$%
& '(
H
H
Lt
Kt
t
K
µ
µ
µ)
ˆ
ˆ,
ˆ,
it is optimal for H to always audit because the expected incremental payoff from
auditing is:
KyytEKrytE LHHHH !!!!=!! ))]1()(1([])([ "
KxtEH
!!= ])1([ "
which is positive because .ˆ
ˆ
H
H
t
Kt
µ
µ!
"#
(iii) Given the firm’s equilibrium reporting strategy !"
#$%
& '(
H
H
Lt
Kt
t
K
µ
µ
µ)
ˆ
ˆ,
ˆ, it is
optimal for L to always audit because the expected incremental payoff from
auditing is:
KyytEKrytE HLLLL !!!=!!! ))]1(([]))1(([ "
KxtEL
!= ][ "
28
which is positive because Lt
K
µ!
ˆ" . The assumption that
LH
LH
tt
ttK
+!µ̂ ensures that
an α exists in the interval ].1,0[ˆ
ˆ,
ˆ!"
#
$%&
' ()
H
H
Lt
Kt
t
K
µ
µ
µ*
(b)LH
LH
tt
ttK
+!µ̂ :
(i) H is indifferent between auditing and not auditing the report. H′s incremental
payoff from auditing is:
0))]1()(1([])([ =!!!!=!! KyytEKrytE LHHHH "
because H
H
t
Kt
µ
µ!
ˆ
ˆ "= . The assumptions that
HtK µ! and µµ !ˆ ensures that
.10 !!"
(ii) L never audits because its incremental payoff from auditing is:
KyytEKrytE HLLLL !!!=!!! ))]1(([]))1(([ "
KxtEL
!= ][ "
which is negative because H
H
t
Kt
µ
µ!
ˆ
ˆ "= and
LH
LH
tt
ttK
+!µ̂ .
(iii) Given H
LH
H
t
tt !=" and 0=
L! , F is indifferent among its reports, because
substituting )1)(1( LH yyr !!+= "" into equation (3) yields:
)1( HLHHF ytytP !!!=
which is independent of α. QED
29
Proof of Proposition 2:
(a) When ,2KC ! a BAPA is never desirable because the BAPA costs will be greater
than audit costs and thus 0)x( =! for all x.
(b) When LH
LH
tt
ttK
+!µ̂ , the players will choose the heavy auditing strategies in the
reporting game and incur audit costs of K2 . Because KC 2! , a BAPA is desirable.
A BAPA is feasible if for the realization (yH, yL) if there is a z that weakly improves the
payoffs of H, L, and F, i.e.:
Kytzt HHH !" (A1)
Kytzt LLL !"! )1( (A2)
LLHHLH ytytztzt +!"+ )1( . (A3)
A z exists that satisfies equations (A1)−(A3) if and only if:
L
L
H
Ht
Ky
t
Ky +!"! 1
and
LH
LLHH
H
Htt
ytyt
t
Ky
!
!!"!
)1( .
These two conditions are satisfied if and only if .)(
1 *
LH
LH
LHtt
ttKxyyx
+=!"+=
Therefore,
!
"( x) = 1 if *xx ! and
!
"( x) = 0 if
!
x " x*. This implies that
!
ˆ µ = E[ x | x " x*] . Because 1ˆ
*!! µx , there exists a )1,0[!p such that
)1(]|[ˆ **ppxxxxE !+="=µ . Substituting this expression for µ̂ into the
30
condition LH
LH
tt
ttK
+!µ̂ and using the fact that
LH
LH
tt
ttKx
)(* += implies that
LH
LH
tt
ttK
+!µ̂ is equivalent to
LH
LH
tt
ttK
+! . Therefore,
!
"( x) = 1 if *xx ! and
!
"( x) = 0
if
!
x " x* if
LH
LH
tt
ttK
+! .
(c) WhenLH
LH
tt
ttK
+!µ̂ , the players would choose the light audit strategies in the
reporting game. Because H
LH
t
ttKC
)( !" , a BAPA is undesirable. If 0)x( =! for all
x, µ̂ µ= , so that LH
LH
tt
ttK
+!µ̂ is equivalent to
LH
LH
tt
ttK
+!µ . So 0)x( =! for all x if
!
K "µt
HtL
tH
+ tL
and H
LH
t
ttKC
)( !" .
(d) When LH
LH
tt
ttK
+!µ̂ , the players would choose the light audit strategies in the
reporting game and incur audit costs of H
LH
H
t
ttKK
)( !=" . Because
H
LH
t
ttKC
)( !" ,
a BAPA is desirable. A BAPA is feasible if for the realization (yH, yL) there is a z that
weakly improves the expected payoffs of H, L, and F. The payoffs of H, L, and F in
the light audit equilibrium are:
))1(())1(( Kxtxyt HHHH !!+!! "#"
!"
#$%
&+'
H
HLt
Kxyt
µ̂1 , where [ ] 10)x(|xEˆ === !µ given the off-
equilibrium path beliefs of H and L.
31
).1( HLHH ytyt !!!
Therefore, the conditions under which a feasible BAPA exists are:
)1(ˆ
H
L
H
L
HHHt
tK
t
tKxytzt !!!"
µ (A4)
!"
#$%
&+'('
H
HLLt
Kxytzt
µ̂1)1( (A5)
)1()1( HLHHLH ytytztzt !+"!+ . (A6)
A z exists that satisfies equations (A4)−(A6) if and only if:
H
H
H
L
HH
L
H
Ht
Kxy
t
t
t
K
t
t
t
Kxy
µµ ˆ)1(
ˆ!"!!! ,
which is true for all x because 1ˆ =µ and 1!x . Therefore, a BAPA is feasible for all x
and thus 1)x( =! for all x.
The off-equilibrium path belief that 1ˆ =µ is the only belief that is consistent with
the conjectured equilibrium. For any other belief, a BAPA would not be feasible for
realizations of µ̂x ! and so F would choose 0)x( =! when µ̂x ! . However, the
only solution to µµ ˆ]ˆ|[ =!xxE is 1ˆ =µ . This implies that the condition LH
LH
tt
ttK
+!µ̂
is equivalent to LH
LH
tt
ttK
+! . Therefore, 1)x( =! for all x if
LH
LH
tt
ttK
+! and
H
LH
t
ttKC
)( !" . QED
32
Proof of Proposition 3
(a) Because KC 2! , Proposition 2(a) shows that 0)x( =! for all x, which implies
µµ =ˆ . Proposition 1(a) implies that the players choose their heavy audit strategies in
the reporting game because
!
K "µt
HtL
tH
+ tL
.
(b) Because KC 2! , Proposition 2(a) shows that 0)x( =! for all x, which implies
µµ =ˆ . Proposition 1(b) implies that the players choose their light audit strategies
in the reporting game because
!
K "µt
HtL
tH
+ tL
. QED
Proof of Proposition 4
(a) Because LH
LH
tt
ttK
+! and KC 2! , we know from Proposition 2(b) that
!
"( x) = 1
if *xx ! and
!
"( x) = 0 if
!
x " x*, and that
LH
LH
tt
ttK
+!µ̂ is equivalent to
LH
LH
tt
ttK
+! . Proposition 1(a) therefore implies that the players will choose their
heavy audit strategies in the reporting game.
(b) Because
!
K "µt
HtL
tH
+ tL
and
!
K(tH " tL )
tH# C # 2K , we know from Proposition 2(c)
that 0)x( =! for all x, which implies that
!
ˆ µ = µ. if, and that LH
LH
tt
ttK
+!µ̂ is
equivalent to LH
LH
tt
ttK
+! . Proposition 1(b) implies that the players will choose
their light audit strategies in the reporting game. QED
33
Proof of Proposition 5
(a) Because
!
K "tHtL
tH
+ tL
and
!
C "K(t
H# t
L)
tH
, Proposition 2(b) shows that
!
"( x) = 1 if
*xx ! and
!
"( x) = 0 if
!
x " x*, and that
LH
LH
tt
ttK
+!µ̂ is equivalent to
LH
LH
tt
ttK
+! .
Proposition 1(a) implies that the players will choose their heavy audit strategies in
the reporting game.
(b) Because
!
K "tHtL
tH
+ tL
and
!
C "K(t
H# t
L)
tH
, Proposition 2(d) shows that
!
"( x) = 1
for all x and that
!
ˆ µ = 1. QED
Proof of Proposition 6 In the absence of a BAPA program, aggregate audit costs would
be 2K if LH
LH
tt
ttK
+!µ and
H
LH
t
ttK )( ! if LH
LH
tt
ttK
+!µ (Proposition 1 with µµ =ˆ .) BAPA
costs are always C.
(a) From Proposition 4(a), it follows that a BAPA will occur when *
xx ! and the
players will choose their heavy audit strategies when .*xx ! In the absence of the
BAPA program, the players would choose their heavy audit strategies for all x because
!
K "µt
HtL
tH
+ tL
. Because KC 2! , the existence of the BAPA program decreases
compliance costs when *xx ! and has no effect on compliance costs when .
*xx !
(b) From Proposition 5(b), it follows that a BAPA will occur for all x. In the absence
of the BAPA program, the players would choose their light audit strategies for all x.
34
Because H
LH
t
ttKC
)( !" , the existence of the BAPA program decreases compliance
costs for all x.
(c) From Proposition 4(a), it is an equilibrium for a BAPA to occur when *
xx ! and
for the players will choose their heavy audit strategies when .*xx ! In the absence of
the BAPA program, the players would choose their light audit strategies for all x
because
!
K "µt
HtL
tH
+ tL
. Because H
LH
t
ttKC
)( !" , the existence of the BAPA program
increases compliance costs for all x when that equilibrium occurs. However, from
Proposition 4(b) it is also an equilibrium for the players to not enter into a BAPA and
choose their light audit strategies. In that case, compliance costs would not change.
(d) From Proposition 5(a), it follows that a BAPA will occur when *
xx ! and the
players will choose their heavy audit strategies when
!
x " x* because
LH
LH
tt
ttK
+! . In
the absence of the BAPA program, the players would choose their light audit strategies
for all x because
!
K "µt
HtL
tH
+ tL
. Because H
LH
t
ttKC
)( !" , the existence of the BAPA
program decreases compliance costs when *xx ! and increases compliance costs
!
K "µt
HtL
tH
+ tL
when .*xx ! QED
35
ENDNOTES
1 An important practical issue in this process is whether the governments have the option
of acting in bad faith by engaging in BAPA negotiations solely to obtain information
about the taxpayer, which it then uses in an adversarial audit process. Taxpayers of
course would not be willing to negotiations under such circumstances. To guard against
this outcome, Sections 9.03 and 9.04 of Rev. Proc. 91-22 [1991 C.B. 526] bars the
Internal Revenue Service from using taxpayer disclosures in any proceeding not covered
by an advance pricing agreement.
2 Given that firms and the governments behave in this manner in general, deviating in any
particular instance (e.g., by having the firm not propose a BAPA that is both feasible and
desirable) would decrease the deviating player’s payoff . However, other sets of
behaviors also have this property. For example, if firms could collectively boycott the
BAPA process for certain parameter values, doing so could alter the governments’
optimal audit strategies in the reporting game.
