SALAMA.FORMATTED.9.

DOC 4/26/2011 11:48 AM

THE FLAW OF AVERAGES IN U.S.

CORPORATE INCOME TAXES: AN
EVALUATION FROM THE TAXPAYER’S
PERSPECTIVE
*
By Michael H. Salama

TABLE OF CONTENTS

I. INTRODUCTION ............................................................................. 632

II. VARIABILITY OF INCOME: THE FLAW OF AVERAGES IN

ESTIMATED TAX ........................................................................... 635
A. Case Study: Forecasting the Tax Liability on a
Theatrical Film’s Performance ............................................ 635
1. A Limited Release Title ................................................. 635
2. Formulating Sounder Complex Models........................ 637
B. Tax Attribute Utilization and Progressive Effective
Tax Rates ............................................................................... 640
C. The Estimated Tax Payment Double-Whammy ................ 642

III. ACCOUNTING FOR TAX UNCERTAINTIES AND RELATED

PROVISION MATTERS................................................................... 644
A. Background on FASB Interpretation No. 48 (FIN 48 or
ASC 740-10) .......................................................................... 644
B. The Flaw of Extremes........................................................... 645
C. The Portfolio Effect ............................................................... 648

*
Michael H. Salama is the Vice President for Tax Administration and Senior
Tax Counsel for The Walt Disney Company. The thoughts and opinions expressed
herein are his own and not necessarily those of his employer. This paper was prepared
as part of a course of study at the Judge Business School at Cambridge University.
Sam Savage, a Fellow of the Judge Business School and Consulting Professor of
Management Science and Engineering at Stanford University, graciously served as
the advisor on this paper.

631
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

632 Virginia Tax Review [Vol. 30:631

D. Simpson’s Paradox in Risk Assessment............................... 649

1. State Income Tax’s Widely Known Administrative
Look-Back Policy ........................................................... 651
2. Examining Outside Counsel Opinions and
Experience....................................................................... 653
E. Scholtes Revenue Fallacy in Intercompany Relationships.. 656
1. Case Illustration............................................................... 656
2. Corollary: The Even-Steven........................................... 658

IV. CONCLUSIONS ............................................................................... 660

I. INTRODUCTION

In a market environment of intense competition for capital

investment and use of funds, well managed companies place a
significant emphasis on cash controls to optimize performance and
increase attractiveness to investors. In the United States one of the
most significant cash requirements is a company’s federal income tax
1
liability. For all but a few privileged firms, reliable estimation of cash
tax needs is crucial to operations. It is also hard to ignore that we live
in an age where company performance is measured by market
analysts on a quarter-to-quarter basis. This puts pressure on
companies: (1) to avoid surprises from the analyst consensus of
forecasted quarterly earnings and after-tax free cash flows; and (2) to
reduce volatility in effective tax rates which may be driven by income
tax return positions.
Significant unknowns exist for firms when they estimate their U.S.
income tax liability during a year for purposes of making quarterly
estimated tax payments. Corporations owing more than $500 in
income tax for their tax year must make estimated tax payments on a
quarterly basis to the Internal Revenue Service (Service); otherwise,
2
such corporations are subject to penalties. There is generally
variability in pre-tax book income. Macro and micro economic
conditions may, and often do, change during the year. These changes

1
For example, for FYE 1/31/10 Home Depot, Inc., a component of the Dow
Industrial Average 30, reported a tax provision of $1.362 billion on income from
continuing operations of $3.982 billion, an effective tax rate of 34.2%, just slightly
lower than the Federal statutory rate of 35%. The Home Depot, Inc., Annual Report
(Form 10-K) 20–21 (Mar. 22, 2010). The company reflected a tax payable in the
amount of $108 million. Id. at 33.
2
I.R.C. § 6655(f).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 633

cause fluctuation in consumer demand, the cost of inputs, and a host

of other business drivers that impact gross receipts. Moreover, certain
income streams may qualify for particular tax credits and/or benefits.
Incentives and tax attribute utilization may be dependent upon levels
of gross receipts. These items often impact the overall estimate of a
firm’s income tax liability. Uncertainty also exists in the income tax
provision context. The structure of the technical accounting rules
regarding tax contingencies and deferred tax assets may not yield
intuitive results in all cases.
Current estimated tax and provision models are typically based
upon simple point-estimates. Static data points fail to provide a range
of potential values and deny the user a meaningful way of making a
decision based upon one’s risk appetite. In addition, static models do
not account for non-linearities, which will cause the average expected
value to differ from the actual average. Sam Savage, a fellow at the
Judge Business School at Cambridge, has coined a term for this —
“the flaw of averages.” It states that: “Plans based on average
3
assumptions are wrong on average.” Savage provides a nice
illustration of this: “Consider a drunk staggering down the middle of a
busy highway. . . and assume that his average position is the
centerline. Then the state of the drunk at his average position is alive,
4
but on average he’s dead.” Savage’s take on this is an elegant way to
explain Jensen’s inequality. Jensen’s inequality provides that for the
convex (concave) function F of a random variable X, F(E(X)) ≤
E(F(X)) (or F(E(X)) ≥ E(F(X)) for concave functions) where E is the
5
expectation operator.
A survey of the literature reveals surprisingly little work in the tax
context in evaluating Jensen’s inequality given the broad population
impacted and the hundreds of millions of dollars at issue every year in
tax revenues and provision related items. Primary existing work in this
space addresses instances of the flaw of averages where the taxes are
6
contingent (e.g., sliding scale mineral royalties), consideration of

3
SAM L. SAVAGE, THE FLAW OF AVERAGES: WHY WE UNDERESTIMATE RISK
IN THE FACE OF UNCERTAINTY 11 (2009) (emphasis omitted).
4
Id. at 83.
5
There are in essence four cases of Jensen’s inequality where the graph of the
formula is: (a) a straight line and the formula’s average value is the formula as
evaluated at the average input; (b) a convex curve and the formula’s average value is
more than the formula assessed at the average input; (c) a concave curve and the
formula’s average value is less than the formula assessed at the average input; and (d)
something else — where “Jensen’s inequality keeps its mouth shut.” Id. at 91–92.
6
M.R. Samis, G.A. Davis & D.G. Laughton, Using Stochastic Discounted Cash
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

634 Virginia Tax Review [Vol. 30:631

7
optimal tax regime structuring by the government, the similar topic of
risk spreading properties of fixed fee, profit-sharing and other
8
common taxes, and the impact of differing national tax rates on
9
underwriting capacity and reinsurance markets. A thought piece was
published on the concept of expected values in financial reporting;
yet, this article includes little discussion of applications in the tax
context nor does it address methods or means for companies to
10
identify and proactively address the flaw of averages in tax. In the
current state, the academic literature has yet to identify key sources
and occurrences of the tax consequences of the flaws of averages.
Wide industry practice of using point estimates suggests a high
probability that companies are largely unaware of their very existence
in many contexts.
In Microsoft Excel, Jensen’s inequality and the flaws of averages
are often identifiable based on the occurrence of max, min, if, and
nested logic functions. The aim of this paper is to go beyond analysis
of commonly used Excel models, identify substantive key common tax
flaws of averages, provide readership awareness of them, and furnish
practical guidance on how to cope with them. Monte Carlo simulation
is employed to demonstrate how to most properly account for the
relevant uncertainties and influence decision making in light of
11
different risk appetites. The paper’s key finding is that probabilistic
modeling and analysis is critical to a constructive assessment of cash-
tax requirements and related income statement impacts associated

Flow and Real Option Monte Carlo Simulation to Analyse the Impacts of Contingent
Taxes on Mining Projects, June 19, 2007, http://inside.mines.edu/~gdavis/Papers/
AusIMM.pdf.
7
See, e.g., Helmuth Cremer & Firouz Gahvari, Uncertainty, Commitment, and
Optimal Taxation, 1 J. PUB. ECON. THEORY 51 (1999); Peter A. Diamond, Optimal
Income Taxation: An Example with a U-Shaped Pattern of Optimal Marginal Tax
Rates, 88 AM. ECON. REV. 83 (1998); J.A. Mirrlees, An Exploration in the Theory of
Optimum Income Taxation, 38 REV. ECON. STUD. 175 (1971); Jeff Strnad, Taxes and
Nonrenewable Resources: The Impact on Exploration and Development, 55 SMU L.
REV. 1683 (2002).
8
J.K. Sebenius & P.J.E. Stan, Risk-Spreading Properties of Common Tax and
Contract Instruments, 13 BELL J. ECON. 555 (1982).
9
James R. Garven & Henri Loubergé, Reinsurance, Taxes, and Efficiency: A
Contingent Claims Model of Insurance Market Equilibrium, 5 J. FIN. INTERMEDIATION
74 (1996).
10
L. Todd Johnson et al., Commentary, Expected Values in Financial Reporting,
7 ACCT. HORIZONS 77 (1993).
11
There are many software packages which may be used to conduct Monte
Carlo simulation within Microsoft Excel’s realm, such as @Risk, Crystal Ball®, and
XLSim®. The modeling in this paper relies on XLSim®.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 635

with uncertain tax positions. While this analysis should not supplant
managerial judgment and experience, it informs the analysis necessary
for making more reasoned conclusions about tax uncertainties.

II. VARIABILITY OF INCOME: THE FLAW OF AVERAGES

IN ESTIMATED TAX

A. Case Study: Forecasting the Tax Liability on a Theatrical

Film’s Performance

1. A Limited Release Title

Estimating one’s income tax requirements for the year and

making quarterly estimated tax payments, requires one to forecast
different aspects of business unit performance and the relative tax
burdens they carry. This paper starts with the simple example of a
limited release U.S. theatrical film and demonstrates why using
average values yields the wrong answer on average. Assume the price
per ticket is $12 and the exhibitor of the film receives 50% of gross
receipts. While ticket prices may vary from town to town, a static
figure is used to illustrate the main theme. The number of ticket sales
prior to the release of the film is unknown. The producer typically will
estimate a low case, most likely case, and best case scenario and use
the average for tax payment estimation purposes. For upper
management reporting purposes, the high and low will generally be
shared as well. Here, ticket sales estimations are: (1) low: 350,000; (2)
most likely: 550,000; and (3) high: 1.4 million. The average is 766,000.
However, as is common in the industry for limited initial releases,
should the film fare well in the opening weekend and first few weeks,
the producer will expand the number of screens upon which the film is
exhibited or extend the theatrical run. Assume the studio sets this
benchmark at 900,000 tickets. Applying a 35% federal statutory rate
to the average estimates would indicate a tax requirement of $1.6
million.

Number of Tickets Sold 766,666

Price Per Ticket $12
Exhibitor Share $4.6M
Total Producer Revenue $4.6M
Estimated Tax at 35% = $1.6M
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

636 Virginia Tax Review [Vol. 30:631

But wait, is this right? There is a non-linearity in this scenario. If

the film performance exceeds a certain level, there is a potential for
even more ticket sales. Using average values fails to take this
expansion option into account. Monte Carlo simulation was employed
to model the same film performance. While a time-series based
performance model might have been employed in this context, a more
simplified version was used for illustration purposes. Ticket sales were
estimated using the gen_normal distribution function in XLSim® =
max(0,gen_normal(550,000, 250,000)) (based on historical experience
for a similar type of title for illustration purposes). The max function
was used to eliminate negative values and its very use also introduces
another flaw of averages. The expansion option was given effect with
an if statement = if(max(0,gen_normal(550,000, 250,000)) > 900,000,
1,400,000, max(0,gen_normal(550,000, 250,000)). In other words, if on
a randomized basis, using the studio’s historic normal distribution, the
film exceeded 900,000 ticket sales it was assumed the film would be
held over and/or shown on more screens and achieve its high of 1.4
million in ticket sales; otherwise ticket sales were projected using the
gen_normal function. This model was run for 1,000 trials. The
statistics from the simulation reflect a very different story:

Estimated Tax
Probability* Profit Before Tax
at 35%

25% $2.33M $.82M

50% $3.34M $1.17M

75% $4.42M $1.55M

*Probability percentiles reflect the likelihood the amount is
equal to or less than the amount shown.

Here, using average revenue value does not produce the true average
estimated tax due. The static model would have suggested the firm
pay close to $400,000 more in estimated taxes then the probabilistic
model indicates.
Probabilistic modeling of course also provides a distribution of
potential outcomes so that one can decide upon the estimated tax
payment with more complete information. This is not to suggest that
probabilistic models necessarily produce “the right answers.”
However, they take into account uncertainty, cure the tax flaw of
averages, and allow the company to provide its own gloss on tax risk
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 637

sensitivity to the problem in a fashion otherwise unavailable with the

point estimate approach.

2. Formulating Sounder Complex Models

In the prior example the distribution for ticket demand was

simply given. From a practical perspective, businesses will want to
develop a library of distributions based upon operational results and
other available information. Although not a common practice in the
past, new technologies are facilitating the generation and maintenance
of such probabilistic libraries. Specifically, a new data type known as
the distribution string enables libraries of probability distributions
12
that may be shared within or across organizations. The taxpayer can
then make choices from that library for realistic distributions, beyond
mere management hunches, based upon appropriateness of fit and
similarities in traits to what is being estimated. For example, suppose
one is trying to estimate the total domestic box office dollars for a
broad theatrical release of an animation title prior to actual release.
One might gather historic information on its own prior releases as well
as publicly available information on other titles. Examining the shapes
for opening box office and cumulative box office for the top-200 post-
1980 releases, each appears to be lognormal (as does the ratio of one
to the other), corroborating a widely-held belief in modeling circles
that revenue streams will have a lognormal shape rather than a
13
normal one.

12
See ProbabilityManagement.org, http://probabilitymanagement.org.
13
See, e.g., LOGNORMAL DISTRIBUTIONS: THEORY AND APPLICATIONS 252–54
(Edwin L. Crow & Kunio Shimizu eds., 1988). Publicly available data sources should
not be ignored. For example, in this case much of this information is available online.
See, e.g., Box Office Mojo, Movie Box Office Results by Year, 1980-Present,
http://boxofficemojo.com/alltime/domestic.htm (last visited Feb. 23, 2011).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

638 Virginia Tax Review [Vol. 30:631

Depending on the nature and character of the soon-to-be released

title, it may make sense to select a subset of titles to form the total
top-200 feature animation distribution to estimate performance.
Feature animation titles might be sorted by relevant traits such as
computer-generated vs. hand-drawn, Summer vs. Winter release, 3D
vs. 2D, story line characteristics, voice-talent, etc. Consideration of
these traits helps to formulate an appropriate distribution shape
reflecting performance of titles similar to the one to be released.
Also, it is often useful to consider relationships within data sets
and operational results that might aid in estimation approaches. Here,
examining the top-200 post-1980 domestic box office feature
animation titles reveals a strong, R=.84, linear correlation between
opening box office and cumulative box office. This corroborates a
commonly held belief that the open has some relationship to how well
the film performs overall. Depicted below is the linear best fit for the
top fifty wide-release feature animation titles, demonstrating the
relationship between open and cumulative box office. The coefficient
2
of determination R of .7, reflects the responsive variable is largely
14
explained by the explanatory variable.

R2 = { ( 1 / N ) * Σ [ (xi - x) * (yi - y) ] / (σx * σy ) }2

14

N = the number of observations in the model, x = the mean of x values, y = the mean
of y values, σx = the standard deviation of x values, and σy = the standard deviation of
y values.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 639

One might employ this relationship to estimate the overall

performance of a title. A Monte Carlo simulation run on a
distribution of opening performances, with cumulative revenues
computed based upon historic multiples of opens, would provide one
with a probability curve. On a pre-opening basis, this approach is
superior to estimating the performance of the current film by simply
averaging the cumulative box offices of similar titles. This type of
model might be best employed after the opening weekend results are
in hand, given the relationship between the open and cumulative box
office results. Similarly, it can be refined as new information based
upon performance is in hand. The general approach of using events as
a basis for forecast future revenues or expenses may be incorporated
into the forecasting process for book-to-tax differences reflected in
Schedule M-1 Federal income tax return adjustments or book income
or expense trends too.
A common defensive theme respondents espouse to this
suggestion is that a proffered distribution curve does not accurately
capture their scenario and every potential variable that may affect
their business. Models can and should be improved upon, but they will
never be perfect. Sound modeling strives to capture the important
variables and not every potential variable regardless of significance.
The intelligent use of models is to inform decisions, not make
decisions. Opting for a point estimate approach over a well thought
out probabilistic model is like choosing to only eat fast food because
cooking at home is too much trouble. In this context, the new
technologies for storing probability distributions are akin to frozen
microwave gourmet meals and further strengthen the case for
probabilistic models.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

640 Virginia Tax Review [Vol. 30:631

B. Tax Attribute Utilization and Progressive Effective Tax Rates

The prior film example reflects the uncertainty faced in estimating

tax obligations based upon variability in operational results. The
existence and use of tax attributes adds further complexity and
additional opportunities for the flaw of averages to creep in. Suppose
a firm’s income averages $1,500x and that it has a net operating loss
(NOL) carryover that it must utilize in the next five years. The NOL’s
current year use is limited to the amount of positive income in the
current year. So for purposes of estimating and paying taxes for the
current year, on average the company would have a tax liability of
$490x ($1,500x average income less $100x NOL, times 35% corporate
tax rate). The company’s average effective income tax rate is 32.67%
($490x/$1,500x). But this model ignores the uncertainty associated
with the company’s income. Conducting a sensitivity analysis on the
income levels reveals the effective cash tax rate function is concave.
The effective rate is low as the income levels are low but approaches
the statutory rate as income increases and the NOLs have been fully
absorbed.

In order to forecast the company’s estimated tax liability, it

should develop a distribution for potential income levels and run a
Monte Carlo simulation to inform decision making around the levels
of NOLs likely to be utilized and the probabilities associated with
different levels of estimated taxes. For illustration, assume the
company is a start-up and has a great degree of income level
uncertainty such that it is equally likely the company’s performance
will range from a $1,000x loss to a $4,500x profit. Simulating this using
the gen_uniform function in XLSim® for 1,000 trials reflects an actual
average tax of $457x and a 30% chance no tax is due. Now assume the
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 641

company’s performance has less uncertainty, and its average revenue

is $1,500x with a standard deviation of $500x. Simulating this using a
max of (0,gen_normal(1500x,500x) for 1,000 trials reflects an actual
average of $483x. Non-linearities attributable to the tax attribute
15
utilization have an impact here. On average, the amount of tax due is
not the same amount as using mere averages to project the liability.
Just using the average results in paying more than the actual average
would have suggested. As with the feature animation film illustration,
one would want to develop a shape of uncertain numbers and then
build models based on them to produce a distribution of potential
results. Without doing so the impact of non-linearities will not be
given effect.
Now we know to look for the flaw of averages in estimating
income tax obligations. Beyond attempting to model the income and
resulting tax distributions, to arrive at an informed judgment one
should also consider what can be done from an operational
perspective. Many corporations engage in robust risk management
techniques to hedge variability in foreign exchange risk, the cost of
production inputs, and other variables. So, the question is, is there any
value to this from a tax perspective? Where the effective tax function
is linear, hedging generally will not impact the firm’s estimated tax
liability when one uses the average pre-tax income to compute
16
expected tax liability for the year. However, the existence of tax
attributes such as NOLs and foreign tax credits with limitations on
their use, research and development tax credits which may be capped
at certain levels, and the alternative minimum tax, may result in a
progressive corporate tax effective rate and a non-linear effective tax
17
rate curve, even though the statutory rate itself is fixed at 35%.
While this concept was touched upon in the literature as part of an
overall analysis of corporate risk management, it was not
quantitatively depicted. The case examples provided illustrate the
issues associated with concave effective tax rate functions where low
income levels carry low effective tax rates and high income levels bear
higher tax rates.
Where the effective tax function is non-linear there is a company
incentive to explore hedging techniques. Attempting to mitigate
variability of income may produce a result where the increase in tax in

15
Note that using the max function itself also introduces a non-linearity.
16
Clifford W. Smith, Jr., Corporate Risk Management: Theory and Practice, J.
DERIVATIVES, Summer 1995, at 21, 26.
17
Id. at 26.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

642 Virginia Tax Review [Vol. 30:631

poor years is less than the tax decrease in better years, driving the
18
expected tax liability down. However, this suggestion from the
literature does not appear to take into account that the penalty for an
upside surprise in income levels may be more tax liability. Incremental
post-tax profits are generally a welcome surprise, even where it means
accelerated utilization of tax attributes and a higher effective tax rate.
Hedging techniques will yield the most utility where they mitigate
potential losses that might cause tax attributes either to expire or have
their use delayed.

C. The Estimated Tax Payment Double-Whammy

At this juncture the paper has explored the impact of non-

linearities in revenue streams and tax attribute utilization on
estimated tax analysis. An obvious set of questions arise: “What
happens if we’re wrong? What if we over or under estimate our tax
liability?” The double whammy is yet another example of the flaw of
averages. It occurs when there is penalty on either side of missing the
estimate where the estimate is the average value. Savage provides the
19
example of a pharmaceutical firm distributing a perishable antibiotic.
The company must estimate the amount of antibiotic to stock and
decides upon using the average demand amount. However, if the
demand is less than the stocked amount, the firm suffers from a
$50/unit spoilage charge. If the demand is more than the stocked
amount it suffers from a penalty of $150/unit for expediting the
shipment of more units. This is like the example of the drunk in the
middle of the road; at the average position he is okay, but on average
he is dead. In the double-whammy, there is a financial penalty on
either side of the average position which is not negated. There are no
costs attributable to average demand but significant financial penalties
when average demand is not achieved.
The tax code provides a penalty for failure to pay estimated taxes.
Generally, corporations that underpay their estimated taxes have to
add to their tax an amount equal to the underpayment interest rate
times the amount of the underpayment for the period of
20
underpayment. Interest on underpayments are tied to the short-term
21
federal rates and compounded daily. The corporate underpayment

18
Id.
19
SAVAGE, supra note 3, at 14–16.
20
I.R.C. § 6655(a).
21
I.R.C. § 6621(a)(2).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 643

22
rate for the first quarter of 2011 is 3%. But there is also a penalty for
overpaying one’s estimated taxes — the company loses the use of the
funds. In most cases the company’s weighted average cost of capital
(WACC) will far exceed 3%. Where one pays the government too
much, overpayment interest is due at 3% generally, and 2% for
corporations who make overpayments under $10,000, with the portion
23
over $10,000 only receiving interest of 0.5%. This is unlikely to be
viewed as a good consolation prize to the management in one’s
corporate treasury function.
If one uses simple averages to estimate one’s tax obligation he’ll
typically end up with a penalty, either the Service demanding the 3%
underpayment penalty or the corporate function wanting to know why
funds were put on deposit with the government at 2% or 0.5% when
the company could have earned, say a 15% return. So, what should be
done? Neither the academic nor the management literature addresses
this concept from an estimated tax perspective. A prudent approach
would be to create a probabilistic model to estimate the impacts and
lean towards underpaying if the penalty for missing in that direction
(the 3% Service penalty) is less than missing in the other direction
(WACC less the 2% or 0.5% overpayment interest — or more likely
the short-term cost of capital which may look like the short-term risk-
free rate such as based on a one-year U.S. Treasury return — plus the
short-term return on a well performing equity). One would want to
create potential distributions for the uncertain numbers as in the prior
examples, layer on the cost impact of the penalties in either direction;
and then use a parametric simulation to determine what would likely
happen if the estimate was missed by various multiples. Where the
short-term opportunity cost of capital was lower than the full WACC
— which in most cases it certainly will be — or actually less than the
penalty for overpayment, this is still the framework to apply unless the
penalties on each side happen to be equal. This approach would lead
to a more informed use of funds. The company would, of course, need
to make payments on its ultimate actual tax liability (for example in
the fourth quarter, upon extension of the return filing, or with the
filing of the tax return). Since the under and overpayment amounts
are set by federal statute and tied to U.S. Treasuries, they are not
susceptible to substantial and immediate overhaul. Therefore, as of
the date of this paper this technique appears to be a stable method for
assessing cash tax requirements.

22
Rev. Rul. 2010-31, 2010-52 I.R.B. 898.
23
Id.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

644 Virginia Tax Review [Vol. 30:631

III. ACCOUNTING FOR TAX UNCERTAINTIES AND RELATED

PROVISION MATTERS

The first section of this paper focused on the flaw of averages in

the estimated tax liability context in light of the importance of cash
management strategies. Equally important to the firm is the
soundness of its financial statements. The second section of the paper
addresses several of the most pervasive flaws of averages in the tax
accounting context.

A. Background on FASB Interpretation No. 48

(FIN 48 or ASC 740-10)

Tax filing positions are sometimes not free from doubt. This may
occur for a variety of reasons including, for example, lack of definitive
applicable guidance, complete absence of guidance, or contradicting
legal authorities. Companies that prepare their financial statements in
accordance with U.S. Generally Accepted Accounting Principles (U.S.
GAAP) must apply the relevant Financial Accounting Standards
Board (FASB) guidance as codified. The FASB’s guidance on
accounting for uncertainties in income taxes contains a number of
flaws of averages. In some instances the critical learning is to merely
be aware of the occurrence as the technical accounting rules do not
allow for mitigation. However, in other instances one can and should
go beyond mere awareness to seek a better understanding of
underlying data sources to produce a more informed judgment. This
discussion provides practical recommendations where such are
feasible.
So why should we care about this? The guidance, and the flaws of
averages embedded within, reaches a broad array of firms. Generally
entities trading on major U.S. stock exchanges are required to prepare
24
their financial statements employing these standards. And, its scope
is similarly vast as it applies to positions taken, or anticipated to be
taken, on a tax return that directly or indirectly affect amounts
25
reported on financial statements. For example, the framework

24
ACCOUNTING STANDARDS CODIFICATION, § 740-10-05-6 (Fin. Accounting
Standards Bd. 2009). All corporations publicly traded in U.S. capital markets are
required to either prepare financial statements in accordance with U.S. GAAP or
provide a reconciliation to U.S. GAAP net equity. 17 C.F.R. § 229.10(e) (2008).
25
ACCOUNTING STANDARDS CODIFICATION, Topic 740 (Fin. Accounting
Standards Bd. 2009). The glossary defines the phrase “tax position” to include a
position in a previously filed return, or one expected to be taken on a future return,
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 645

applies to a company’s decision not to file a tax return for a particular

jurisdiction where there may be uncertainty as to whether it has a
nexus in a particular U.S. state, i.e., sufficient contacts with the state
to subject it to taxation therein, or a permanent establishment in a
foreign jurisdiction, which would subject the entity to potential
taxation therein.

B. The Flaw of Extremes

Under U.S. GAAP the starting point for assessing uncertain

income tax positions is defining the unit of account. A unit of account
is determined by reference to all relevant facts and circumstances of
26
the position in light of the relevant evidence. One must consider: (1)
the “manner in which the entity prepares and supports its income tax
27
return” and (2) how the taxing authority will examine the position.
Accounting and assurance firms generally take the view that a unit of
account, and its related assessment, ought to be based on its own
technical merits and facts. For example, PricewaterhouseCoopers
LLP’s manual provides: “The possibility of offset in the same or
another jurisdiction and the possibility that the position might be part
of a larger settlement should not affect the determination of the unit
28
of account.” Similarly, the technical merits of each position taken
under the unit of account standard must be evaluated separately —
“[e]ach tax position shall be evaluated without the consideration of
29
the possibility of offset or aggregation with other positions.”
Once one has defined the unit of account, one must now apply the
recognition threshold for uncertain tax positions. A firm shall
recognize the financial statement impact of its tax position when,

including but not limited to:

A. A decision not to file a return; B. An allocation or shift of income

between jurisdictions; C. The characterization of income or a decision to
exclude reporting taxable income in a tax return; D. A decision to classify a
transaction, entity, or other position in a tax return as tax exempt; E. An
entity’s status, including its status as a pass-through entity or a tax-exempt
not-for-profit entity.
Id. § 740-10-20.
26
Id. § 740-10-25-13.
27
Id.
28
PRICEWATERHOUSECOOPERS LLP, GUIDE TO ACCOUNTING FOR INCOME
TAXES § 16.3.1.2 (2010).
29
ACCOUNTING STANDARDS CODIFICATION § 740-10-25-7(c) (Fin. Accounting
Standards Bd. 2009).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

646 Virginia Tax Review [Vol. 30:631

based on the technical merits, the firm’s position is more-likely-than

not (MLTN) the position that will be sustained (through exam, related
30
appeal, or litigation). In this context, MLTN means greater than
31
50%. Where management is unable to reach this conclusion, no tax
benefit for the position can currently be reflected in the financial
statements. Take the unit of account worth $100x where based on the
technical merits the firm is unable to reach the MLTN standard.
However, suppose that based on its experience it reasonably believes
it has a 70% chance of sustaining $30x and a 30% chance of sustaining
$50x. While the weighted average of its distribution is $36x (also less
than the implied MLTN hurdle) under FIN 48 the company will book
no benefit. Moreover, the company will have to burden its income tax
provision with a reserve for the full $100x. This result is a flaw of
32
extremes and layers unnecessary “cushion” onto the taxpayer’s
financial statements.

When the tax matter is ultimately settled with the taxing

authorities, assuming the weighted average is achieved, the
accountants will reverse the reserve and credit the income statement
for the $36x. In many cases the MLTN source of the flaw of extremes
causes nothing more than a time shift in income statement items, with
an overly conservative starting point. It is unclear how this would be
helpful to the user of financial statements.
Two similar issues were considered in the literature regarding

30
Id. § 740-10-25-6.
31
Id. In applying this criteria: (1) it is presumed the tax position will be audited
and the taxing authority has full knowledge of the facts and relevant information, (2)
the determination is made based upon reference to sources of authorities in the tax
law or widely know administrative practice, and (3) each position is evaluated on its
own merits. Id. § 740-10-25-7.
32
See SAVAGE, supra note 3, at 133–36 (illustrating the same type of flaw for a
business unit budgeting scenario).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 647

33
accounting for bad debts and contingent legal claims. Interestingly,
prior to promulgation of FIN 48, uncertain income tax positions were
governed by FASB Statement No. 5. That statement provides that
“estimated loss from a contingency be accrued by a change to income
if it is probable that a liability has been incurred and the amount of
the loss can be reasonably estimated.” FASB’s standards do not
supply a number or range of values on its likelihood continuum. In
practice the term “probable” has been interpreted to mean 70%–75%
34
and some texts reflect a slightly higher percentage of 78%. Under
FASB Statement No. 5 a requirement for potential losses due to legal
claims would not be booked if the likelihood of prevailing was only
35%. The byproduct of applying this standard is the reverse of the
FIN 48 example above, which arguably understates the potential
liability.
So where do we go from here? One approach is for companies to
support the U.S. Securities and Exchange Commission’s (SEC)
consideration of convergence to International Financial Reporting
35
Standards (IFRS). Under that strategy firms might seek to influence
the rulemaking process to reduce/eliminate the flaw of extremes
embodied in FIN 48’s recognition threshold. From a tax perspective
many U.S. manufacturing firms have been concerned about
convergence to IFRS since International Accounting Standard (IAS)
2, as revised in 2003, eliminated the use of last-in-first-out (LIFO) for
determining the cost of inventories on the basis that it is not a reliable
reflection of actual inventory changes. Under the U.S. tax rules
taxpayers may only use LIFO if they use LIFO to report financial
36
results to partners, stockholders, etc. and for credit purposes. This
can result in material changes in financial statement position for some
firms. For example, based upon their filings with the SEC, Caterpillar
Inc.’s inventory value on December 31, 2009 would be $9,363 (in
millions) under first-in-first-out (FIFO) inventory accounting as

33
Sam Savage & Marc Van Allen, Accounting for Uncertainty: When the Only
Thing Certain Is the Date, J. PORTFOLIO MGMT., Fall 2002, at 31; Sam Savage, Some
Gratuitous Inflammatory Remarks on the Accounting Industry, 4 J. FORENSIC ACCT.
351 (2003) (addressing the bad debt issue).
34
WANDA A. WALLACE, AUDITING 955 (Thomson South-Western College
Publishing, 3d ed. 1995).
35
Information on the SEC’s roadmap for convergence and related information
can be located on their website. See Archived Information: Global Accounting
Standards, U.S. SECURITIES AND EXCHANGE COMMISSION, http://www.sec.gov/
spotlight/ifrsroadmap.htm (last visited Mar. 17, 2011).
36
I.R.C. § 472(c).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

648 Virginia Tax Review [Vol. 30:631

opposed to $6,360 (in millions) under LIFO.

IFRS treatment of tax uncertainties may produce a result that is
more reflective of reality. It remains an open question of whether that
will entice support for convergence. To date it has not. One method of
addressing uncertain tax positions would be to follow IAS 37R, which
allows one to look to the amount the entity would rationally pay at the
measurement date to relieve itself of the liability. Under this approach
one would accrue the lowest of: (1) the NPV of resources need to
satisfy the obligation; (2) the amount paid to extinguish the obligation;
and (3) the amount paid to transfer the obligation. The present value
in this context may be the probability weighted average of anticipated
cash payments for all expected outcomes. To the extent convergence
does not become a reality, firms need to consider methods to
demonstrate to the FASB the current rules are ineffective.

C. The Portfolio Effect

“When uncertain numbers are averaged together, the distribution

of the average goes up in the middle and down on the ends, becoming
more centralized. This centralizing of the distribution is the primary
37
manifestation of diversification.”
A similar effect is seen where the shape of uncertain numbers is
added together to form a portfolio of positions, including income tax
positions. Since FIN 48 requires each unit of account be considered
and evaluated separately and the MLTN recognition standard applied
for each unit of account, the tax accounting rules negate the ability for
one to recognize a benefit associated with the portfolio effects
inherent in managing multiple positions. A similar issue exits under
38
FASB Statement No. 5. Based on this construct limitation, this paper
does not address the manner in which a taxpayer might knowingly
select uncertain tax positions to maximize its expected return (tax
reduction) and minimize variance through diversification in the
39
fashion discussed in the non-tax context. The author believes such is
feasible outside of the FIN 48 rubric when one examines the cash-tax
impact of uncertain positions, as opposed to their initial income
statement treatment under FIN 48. Exposition on the topic would be
highly dependent upon the technical aspects of the tax statutes,
regulation, and guidance, as applied to various particular fact patterns.

37
SAM SAVAGE, DECISION MAKING WITH INSIGHT 40 (Thomson South-Western
College Publishing, 2d ed. 2003).
38
Savage & Van Allen, supra note 33, at 34.
39
See Harry Markowitz, Portfolio Selection, 7 J. FIN. 77 (1952).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 649

For purposes of illustrating the portfolio effects of uncertain

income tax positions were one not burdened with FIN 48, assume a
firm has ten positions of the type described in the prior example
(perhaps it takes the same position in ten different states). In each
case it has a 70% chance of sustaining $30x and a 30% probability of
sustaining $50x. Simulating the potential outcomes of all ten cases
combined for 1,000 trials reflects a potential average benefit of $359x,
with a 25% likelihood of achieving $340x or less, and a 25% likelihood
of securing $380x more of benefit. Adding these independent
uncertainties together for the ten positions, with the same individual
distributions, allows us to see the impact of diversification. The
likelihood of only sustaining $300x in total is the same as every case
^10
settling at $30x, or .7 = 2.8%. Similarly, the likelihood of sustaining
^10
$500x in total is the same as every case settling at $50x, or .3 =
.00059%. But, there are numerous potential combinations in-between
these extremes which is why the histogram of the distribution peaks.
Yet, since the firm does not meet the MLTN threshold for recognition
it may not record any benefit for any of the income tax positions
under FIN 48.

In fact the firm must establish a reserve for $1,000x, a position

that reflects a complete loss on each case. This result, while perhaps
an extreme example, demonstrates the significant impact FIN 48’s
embedded flaw can have on a company’s financial statements. In the
author’s opinion, there is no question a regime which can lead to such
situations provides little incremental utility to investors.

D. Simpson’s Paradox in Risk Assessment

Simpson’s Paradox is yet another example of the flaw of averages.

We see it, or in most cases fail to identify it, when our uncertain
numbers depend upon hidden data traits. Specifically, Simpson’s
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

650 Virginia Tax Review [Vol. 30:631

Paradox “occurs when the direction of an association between two

40
variables is reversed when a third variable is controlled.” This flaw
has the potential to arise in at least two key U.S. income tax
accounting areas related to uncertain tax positions: evaluating state
income tax look back policies and examining outside counsel opinions
and experience. The paper addresses each of these from a practical
perspective. A concrete illustration of Simpson’s Paradox is provided
first to supply the reader with conceptual familiarity.
Perhaps the most infamous example of Simpson’s Paradox
concerns the 1970s University of California, Berkeley (Berkeley) sex
bias case. Berkeley was sued for bias against female applicants to the
graduate programs. The admissions data for 1973 reflected that male
applicants had a greater chance of gaining admittance than females;
and, the conjecture was that this was not due to mere chance. The
figures are:

Applicants Percentage Admitted

Males 8,442 44%

Females 4,321 35%

The aggregation of the admissions data failed to reveal that

generally when examined on a departmental basis Berkeley actually
appeared to have a bias towards admitting females. In four of the six
departments evaluated females actually had a higher rate of admission
than males. The Department details are as follows:

Male Male % Female Female %

Department
Applicants Admitted Applicants Admitted

A 825 62% 108 82%

B 560 63% 25 68%
C 325 37% 593 34%
D 417 33% 375 35%
E 191 28% 393 24%
F 272 6% 341 7%

40
Christopher H. Morrell, Simpson’s Paradox: An Example from a Longitudinal
Study in South Africa, J. STAT. EDUC., Nov. 1999, available at http://www.amstat.org/
publications/jse/secure/v7n3/datasets.morrell.cfm.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 651

Research into this area revealed that males applied more

frequently to less-competitive departments with high rates of
admission whereas females more frequently applied to more
41
competitive departments with lower rates of admission. The key
learning here is that there may exist a confounding variable that is not
apparent when data sets are aggregated and then evaluated.

1. State Income Tax’s Widely Known Administrative Look-Back

Policy

In tax accounting, Simpson’s Paradox may occur when one relies

on widely known administrative practices of taxing authorities, and
general information from service providers, to establish the tax
reserves. For example, when a company has not filed a return for a
jurisdiction the statute of limitations for assessment and collection
does not expire and the company must evaluate its nexus position:
whether it has sufficient contacts to subject it to tax there. Under the
U.S. accounting rules the company may consider whether the state has
a widely known administrative practice of limiting the number of
42
years of audit to a specified number. Specifically, where past taxing
authority administrative practices and precedents in dealing with the
entity or similar ones “are widely understood, for example, by
preparers, tax practitioners and auditors, those practices and
43
precedents shall be taken into account.” The company may
understand in the context of nexus disputes based on its own
experience, or most likely a representation for an accounting or law
firm, that Jurisdiction A will look back only five years in determining
if a tax return and deficiency are due. The company may conclude
based upon the facts and circumstances that it is MLTN a tax return is
not due for more than five years and that a liability for tax exposures
44
for prior years in not required.
For example, in conversation with one consulting practice
regarding State Z it was suggested to the author that Z’s look-back
was five years on average; and, that as the dollars at issue increased
per year the number of look-back years decreased since Z was
concerned about “getting cash in the door” and “not engaging in

41
P.J. Bickel, E.A. Hammel, & J.W. O’Connell, Sex Bias in Graduate
Admissions: Data from Berkeley, 187 SCI. 398 (1975).
42
ACCOUNTING STANDARDS CODIFICATION § 740-10-25-7(b) (Fin. Accounting
Standards Bd. 2009).
43
Id.
44
Id. §§ 740-10-55-94 to -95.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

652 Virginia Tax Review [Vol. 30:631

protracted disputes.” The firm did not initially provide any detail
behind the five year figure or trending assumption. Upon request the
following table of sanitized data was furnished, for which a linear best
fit was determined by the author. The coefficient of determination,
R² = .18, was computed based upon the square of the correlation
coefficient between the firm’s reported data points and predicted
values.
The low coefficient of determination reflects little of the variation
in the response variable is likely explained by the explanatory one.
Almost 80% of variation may be attributable to unknown items.

Dollars at Issue Look-Back

(in millions) Years
$3.5 6
$4 6
$4.5 7
$5.5 7
$6 7
$7 7
$5 2
$6 2
$7 2
$8 3
$9 3

A follow-up conversation was had with the firm at which time it

SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 653

was determined both tax aggressive cases (i.e., tax shelters) were
grouped together with non-tax shelter cases. The average number of
look-back years provided by the firm for shelter and non-shelter cases
was 6.66 (6–7 years) and 2.4 (2–3 years), respectively. This suggests it
indeed makes a difference whether the company falls into one
category vs. the other. Five years on average is not particularly helpful
or equal to the average of either sub-category. It should be noted what
was provided here was an average number of look-back years per
dollar amount at issue. The same provisos and concerns regarding the
use of average values equally applies with respect to those data points
too.
The data sets for those different populations were segregated and
a linear best fit was prepared for each of them. The R² for the tax
shelter and non-tax shelter cases equaled .61 and .75, respectively. So,
a material portion of variance in the response variable could be
explained by the explanatory variable. The details paint a very
different picture than the original information provided by the firm.
This example reflects how Simpson’s Paradox can and does exist in a
very key area of tax accounting. Where one is provided with aggregate
information in the tax context, e.g., a single number of look back
years, with a suggestion regarding trending, consideration needs to be
given to teasing out more detail in the data set to examine whether the
information is faithful to underlying trends. The point is to provide
the user with more information that will inform judgment and lift
potential clouds of confusion as were present in this example.

2. Examining Outside Counsel Opinions and Experience

Another area Simpson’s Paradox may occur in income tax

accounting concerns the measurement of the uncertain tax benefit to
be booked. Where the MLTN recognition threshold has been satisfied
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

654 Virginia Tax Review [Vol. 30:631

the amount of the benefit to be booked is the greatest amount for

which it is more than 50% likely of being realized upon settlement
with the taxing authority which has full knowledge of the facts and
45
circumstances. Accordingly, the rules adopt a cumulative probability
approach that requires one to assign probabilities to potential
46 47
outcomes. Past audit experience may be considered in this context.
The experience of outside service providers frequently influences
managements’ view as to at what dollar amount it is willing to settle a
matter.
From a practical perspective companies consider a variety of data
points including their own experiences and the experiences of outside
law firms and accounting firms in resolving similar issues. It is
common on material issues for companies to solicit input from
multiple firms on their prior audit experiences and sustention rates. In
addition, the companies may use this information as a basis for
selecting their service provider in the tax controversy context. Some
service providers are most likely to provide sustention percentages
only, with little to no additional details. Therein lurks the potential for
Simpson’s Paradox to rear its head.
Assume one is gathering data points from two firms, A and B, to
determine the amount of benefit to book on a tax matter and who to
hire. What if Firm A reports they have dealt with the issue twice, on
average sustaining 71.75% for their clients (75% on case 1 and 68.5%
on case 2)? What if Firm B reports they too have dealt with the issue
twice sustaining 70.95% on average (74% on case 1 and 67.9% on case
2)? Firm A looks slightly better, right? Not necessarily. In this context
it might be helpful to have a sense of the relevant magnitude of the
different tax disputes. How much was at issue in each case? Suppose
the detail was as follows (all dollar amounts in 000s):

45
Id. § 740-10-30-7.
46
Id. §§ 740-10-55-102 to -104 (containing examples which illustrate the
cumulative probability measurement approach).
47
Id. §§ 740-10-55-102 to -107 (illustrating concept of cumulative probability in
the measurement examples).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 655

Average
Case 1 Case 2 of the The Weighted
(Sustention) (Sustention) Average Average48
s
Firm $108/ $1,197/ $1,305/
75% 68.5% 71.75% 69%
A $144 $1,746 $1,890
Firm $912/ $285/ $1,197/
74% 67.9% 70.95% 72.4%
B $1,233 $420 $1,653

Having the dollar figures and evaluating the weighted average for
the firm’s performance might be more informative. Understanding
who performed stronger on larger cases might also be helpful.
Looking at the weighted averages rather than the average of the
averages, Firm B performed almost 3.5% better than Firm A; while
the average of the averages suggested A was the stronger firm. How
one utilizes this information and how it informs judgment will vary
from company to company. This case illustration contains a modest
sustention percentage differential between the two firms to illustrate a
point. The difference may of course be more pronounced in other fact
patterns. The potential existence of Simpson’s Paradox in this context
may provide companies with murky information. Asking service
providers for sanitized data of the type displayed above, with taxpayer
identifiers removed but further granularity, is a sound approach to
getting a better base-line for decision making. This approach would
not of course reveal any unique factors such as whether other issues,
potentially sensitive ones from the taxing authority’s perspective, were
horse-traded with the reported issue and inquiry should be made to
understand the full context of the resolution.
Before leaving this topic it is worth noting that FIN 48’s
measurement framework has the tendency to deviate from the
weighted average approach. Suppose one evaluates an income tax
return position for which it MLTN to be sustained on the technical
merits and the probability distribution looks like the following based
upon past experience, weight of precedent in the taxpayer’s favor, the
firms’ willingness to litigate the issue based on outside counsel’s
experience and advice:

48
For a comparable example in the field of athletics, upon which the instant
example figures were modeled, see KENNETH ROSS, A MATHEMATICIAN AT THE
BALLPARK: ODDS AND PROBABILITIES FOR BASEBALL FANS ( 2004).
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

656 Virginia Tax Review [Vol. 30:631

Amount of the Individual Cumulative Probability

“As Filed” Tax Probability of a that the Tax Position
Benefit Sustained Particular Outcome Will be Sustained
$100x 30% 30%
$75x 15% 45%
$65x 15% 60%
$50x 40% 100%

In this scenario the taxpayer would recognize a benefit of $65x for

financial statement purposes since it is the greatest benefit amount
more than 50% likely to be sustained. The benefit that would have
been recorded under a weighted average approach is $71x (.3 * $100x
+ .15 * $75x + .15 * $65x + .4 * $50x), $6x greater than what the FIN 48
construct would have allowed. It is difficult to definitively conclude
which approach yields a more reflective result. The FIN 48 method
has some surface appeal, since the matter may only be resolved one
time and if the discrete distribution is accepted that resolution should
fall on one of those points not an aggregation thereof. Companies
with an eye toward convergence to IFRS should carefully consider
how their current and future distributions are constructed with the
highlighted difference in mind since the IAS 12 approach leaning on
IAS 37R may reflect aspects of a weighted average point estimate
method.

E. Scholtes Revenue Fallacy in Intercompany Relationships

1. Case Illustration

Stephan Scholtes, of the Judge Business School at the University

of Cambridge, has identified yet another form of the flaw of averages,
the Scholtes Revenue Fallacy. This flaw was identified in the context
of banking and projected revenues, where a bank when using average
values for estimation purposes for future balance growth shows a
49
profit but on average loses money on each account. The flaw is
50
illustrated in the following table. This example is driven by lower
margins on higher balances when compared to the margins on lower
balances.

49
SAVAGE, supra note 3, at 142–43.
50
Reproduced verbatim from SAVAGE, supra note 3, at 143, Table 19.1.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 657

Profit of Average Account is $11, but on Average You Lose $5

Profit after
Balance Margin Net Revenue
Overhead
High
$1,000 2% $20 ($5)
Balance
Low
$200 10% $20 ($5)
Balance
Average $600 6% $20 ($5)
Flawed Average Calculation Based on Average Margin and
Balance
$600 6% $36 $11
This example assumes an overhead cost of $25 per account.

In the tax context the same flaw may exist in projecting future
revenues with potentially perilous results. For example, where two
related parties (defined here as wholly owned subsidiaries of a
common parent) engage in the same trade or business and one serves
as the sales agent for the other a very similar fact pattern arises. In
that case the margins are typically tiered and structured as sales
commissions such that lower transaction amounts receive a higher
commission percentage, and higher transaction amounts receive a
lower commission. Since the parties are related, the seller will
generally be allocated some portion of general and administrative
(G&A) costs at a fixed amount per transaction. Using the same values
from the table above, in this context one would project a profit on
average whereas each transaction would actually be projected to lose
money. Where the seller is carrying NOLs from prior years this can be
a concern from an income tax accounting perspective.
U.S. GAAP requires a reduction in the measurement of deferred
51
tax assets, including NOLs, not expected to be realized. Evaluation
of this matter requires consideration of all positive and negative
evidence to determine whether based upon the weight of the evidence
52
a valuation allowance is required. A valuation allowance is recorded
if based on the judgment of management it is MLTN (i.e. a probability
of more than 50%) that all or some portion of the deferred tax assets
will not be realized. The MLTN threshold presents its own flaw of
averages of the type discussed herein concerning the FIN 48

51
ACCOUNTING STANDARDS CODIFICATION § 740-10-30-16 (Fin. Accounting
Standards Bd. 2009).
52
Id. § 740-10-30-17.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

658 Virginia Tax Review [Vol. 30:631

recognition threshold for uncertain income tax positions. The

valuation allowance shall reduce the deferred tax asset to the amount
53
that is MLTN to be realized. As part of this analysis past
performance is evaluated as part of an assessment of likelihood of
future profitability. Similarly, projections of future income are
examined to determine if there will be sufficient income to absorb the
54
historic NOLs.
In tax as in banking it is not uncommon to use average values for
projections of future profitability. Doing so in this context might
present a picture of potential loss utilization that is not entirely
consistent with what one actually should anticipate. It could have the
unintended result here of not recording a valuation allowance when
one is required, or recording one for an amount lesser than what is
required. This could have an impact on the firm’s income statement.
In conducting valuation allowance analysis for tax, projections
regarding future profitability should be made not based upon
averages; but, by developing meaningful distribution strings in the
modeling context and potentially conducting Monte Carlo simulation
which will reveal these flaws and provide more meaningful
information for the business regarding likelihood of different
outcomes. The models developed can then be re-evaluated and
changed as appropriate.

2. Corollary: The Even-Steven

There are likely a number of different variations of the Scholtes

Revenue Fallacy. One of them appears in state and local taxation. In
the United States some states require the filing of separate tax returns
on a legal entity basis. Where an entity conducts business in a separate
filing state, or has sufficient contacts in that state such that it has
taxing nexus with the state, it must file a legal entity tax return with
the state. In these cases getting intercompany pricing right can have a

53
Id. § 740-10-30-5(e); see also PRICEWATERHOUSECOOPERS LLP, supra note
28, § 5.1.
54
From a technical Accounting Standards Codification Topic 740 perspective,
future projections of income standing alone are not a sufficient indicator to reach the
MLTN threshold regarding attribute utilization. Section 5.1.3.1 of
PriceWaterhouseCoopers’s’ guidebook addresses this notion stating in pertinent part:
“A projection of future taxable income is inherently subjective and generally will not
be sufficient to overcome negative evidence that includes cumulative losses in recent
years, particularly if the projected future taxable income is dependent on an
anticipated turnaround to operating profitability that has not yet been demonstrated.”
PRICEWATERHOUSECOOPERS LLP, supra note 28.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 659

significant impact for tax purposes. For example, consider a business

where A, which is only subject to tax in state Y (a combined state
where it files its tax return with B), and B, which is only subject to tax
in state Z (a separate filing state), are related parties. Assume A
contracts with B to sell its goods in state Z on a commission basis and
there are only two products: one that sells for $1,000 and one that sells
for $500. Given a 10% commission on the lower sales ticket of $500
and a lower commission of 5% on the higher sales ticket of $1,000, B
makes $50 on each transaction. But, this does not include A’s G&A
allocation to B. If there is an allocation of $50 per transaction there is
no profit on either the high or low sales ticket transaction. Yet, on
average, using the average commission and transaction amounts, B is
profitable. This scenario is illustrated below, and is referred to as the
Even-Steven.

Profit of Average Transaction is $6.25, but on Average You Break

Even
Profit
Transaction Net G&A per
Commission after
Amount Revenue Transaction
G&A
High
Sales $1,000 5% $50 $50 $0
Ticket
Low
Sales $500 10% $50 $50 $0
Ticket
Average $750 7.5% $50 $50 $0
Flawed Average Calculation Based on Average
Commission and Transaction Amount
$750 7.5% $56.25 $50 $6.25

This fact pattern arguably to some produces a good result for B,

no income tax in state Z. And since B is not generating a loss, it does
not have to address whether it has a deferred tax asset potentially
subject to a valuation allowance. A does not pay incremental income
tax on the intercompany transaction since A only files returns in state
Y (a combined filing state) and the A & B intercompany transactions
wash in their consolidation. B also thinks to itself, “hey on average we
project we’ll be profitable.” The Even-Steven is a potential trap
taxpayers might fall into. Organizations considering this or those
running into perpetual losses due to intercompany transactions,
should be aware that most states conform to the federal law which
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

660 Virginia Tax Review [Vol. 30:631

provides that where related parties engage in intercompany

transactions, the Secretary “may distribute, apportion, or allocate
gross income, deductions, credits, or allowances between or among
such organizations, trades, or businesses, if he determines that
such . . . is necessary in order to prevent the evasion of taxes or clearly
to reflect the income of any of such organizations, trades, or
55
businesses.” Whether the seller should be in a position to make a
profit in this scenario or whether it is adequate for it to merely break
even for a period of years and generate profits in later periods, is a
question of fact for careful consideration.

IV. CONCLUSIONS

Use of Monte Carlo simulation constitutes a critical tool for firms

seeking to prudently manage estimated income tax requirements and
their overall cash flows. This is particularly true where non-linearities
or the flaw of averages are embedded within one’s income tax
posture. This paper explores cash-tax implications of the flaws of
averages regarding uncertain levels of receipts, utilization of tax
attributes — which may impact one’s effective tax rate, and the
double-whammy of over or under paying one’s estimated taxes. These
examples by no means reflect an exclusive list of the cash-tax flaws of
56
averages. There are bound to be others. However, the techniques
illustrated in the examples herein furnish the reader with a potential
construct for meaningfully managing estimated taxes based on one’s
unique risk profile and management approach. These tools may have
the potential to bolster firm value by informing more efficient use of
funds and are superior to simple point estimates based upon average
values.
The more severe tax flaws of average examined in the income tax
provision context bear lesser opportunity for immediate and material
remediation. A change in the analytical framework for income tax
accounting would be necessary to achieve that, such as a wholesale
revision to the FIN 48 recognition and measurement standard and
MLTN valuation allowance standard. However, this paper does
identify several specific flaws of which the reader should be aware;
since, in some instances a portion of the flaw may be ameliorated by

55
I.R.C. § 482.
56
See, e.g., SAVAGE, supra note 3, at 132. In providing a top-11 list of flaws of
averages the author provides: “The twelfth of the Seven Deadly Sins is being lulled
into a sense of complacency, thinking you now know all of the insidious effects of
averaging.” Id.
SALAMA.FORMATTED.9.DOC 4/26/2011 11:48 AM

2011] The Flaw of Averages in U.S. Corporate Income Taxes 661

mere identification and securing better information.