- -~ . snAs-‘m- amsw<«\\1“:“1

AN EMPIRICAL EXAMMATION.
0F SNCOME MANIPULATION

Thesis for the Degree of Ph. D.
MICHIGAN STATE UNWERSWY
EUGENE A. IMHOFF, JR.
1973

   
     

LI BRA R Y
Gichigan State
University

REP-‘-

  

qmatmrtiry that the
. thesismtit ’ '
AN EMPIRICAL EXAMINATION OF INCOME MANIPULATION

presented by

Eugene A. Imhoff, Jr.

has been accepted towards fulfillment
of the requirements for

Ph.D _ Business — Accounting
degree 1n _____—

MW 233'!

Major professor

 

:— ‘A
Imomc av ‘

   
   

R
IPIIII'III'. IICIIHI

ABSTRACT

AN EMPIRICAL EXAMINATION OF INCOME MANIPULATION

BY

Eugene A. Imhoff, Jr.

The purpose of this research was to test the ability
of the preposed classification technique to discriminate
between potential income manipulators and non-manipulators.
The proposed classification technique was based on the
strength of the association between reported net income
and reported sales trends.

The first step of the research was to examine the
association between sales and net income data for each of
the 41 firms included in the study. The 41 firms included
were selected primarily on the basis of the quantity of
income statement data they provided. The association
between the sales and net income data for the 41 firms,
as measured by the coefficient of determination, indicated
that major differences in the sales—net income relation-

ships existed among the firms examined.

Eugene A. Imhoff, Jr.

The classification of firms was based on the assump-
tion that firms should naturally have a relatively strong
relationship between reported sales data and reported net
income data, and therefore the net income trend should be
associated with the trend in sales. Hence, firms which
had relatively strong (R2 greater than .75) sales-net
income relationships were classified as non-manipulators
(NM firms), and those with relatively weak (R2 less than
.50) relationships were classified as potential manipu—
lators (PM firms).

Two tests were made to determine the nature and
cause of the differences in the sales-net income relation-
ships. The first test was conducted to determine if
either group had a significantly greater number of income
statement variables which improved the basic sales-net
income relationship. The test results indicated that
the number of variables which improved the basic sales—
net income relationships was not significantly different
for the PM and NM firms. All but one of the 33 classified
firms had at least one variable which was capable of im-
proving the relationship between sales and net income at
the .05 level of significance.

The second test examined the 277 variables of 33

 

.u‘

‘0

III
[II

([3

5|

.u

CI)
1“

Eugene A. Imhoff, Jr.
classified firms in terms of four general behavior patterns.
Using the same definitions of "weak" and “strong" that were
indicated above, the variables were examined for the
presence of one of the following behavior patterns:

(1) strong correlations with both sales and income,

(2) strong correlation with sales and weak correlation with
income, (3) weak correlation with sales and strong corre-
lation with income, and (4) weak correlations with both
sales and income.

Variables which reflected the first two behavior
patterns were not considered to be income manipulating
variables due to the consistency of their behavior with
respect to sales. Variables which reflected the third
and fourth behavior patterns were considered to be capable
of manipulating income due to the inconsistency of their
behavior with respect to sales.

Chi—square tests of the frequency distribution of
each behavior pattern among the PM and NM groups indi-
cated that differences existed in each behavior pattern
at the .05 level of significance. The PM group of firms
had a significantly greater number of variables with the
second, third and fourth behavioral characteristics.

These results tended to support the PM—NM classifications.

Eugene A. Imhoff, Jr.

The final stage of the research was to illustrate how
variables were capable of manipulating reported net income.
Four tax expense variables which were weakly correlated
with both sales and net income were examined. The behavior
of these variables was considered to be indicative of the
type of behavior which would be capable of manipulating
reported net income.

The first conclusion of the research was that major
differences did exist in the strength of the sales-net
income relationships of the firms examined in the study.
These differences were found to exist even among firms in
the same industry classification.

The second conclusion of the research was that PM
firms had a significantly greater number of variables
that were capable of manipulating income than NM firms.

Finally, based on the two conclusions above, the
classification technique proposed and tested in the
research was considered to be capable of discriminating

between potential income manipulators and non-manipulators.

AN EMPIRICAL EXAMINATION OF INCOME MANIPULATION

BY

Eugene A. Imhoff, Jr.

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Accounting and Financial Administration

1973

Copyright by

EUGENE A. IMHOFF, JR.
1973

annéy‘
“J“ .LJ

this 1

v
\§.
ji~.e‘

‘
has}
NJ‘I.

Sails

ACKNOWLEDGMENT

There are many who deserve recognition for their
contributions and assistance which aided me in completing
this thesis. To Dr. R. F. Salmonson, Chairman of my com—
mittee, I owe a great deal more than a simple thank you.
His direct contributions to improve the organization and
readability of the thesis were numerous and showed great
patience. The direction and guidance which he continu-
ously provided me with, always without delay, displayed
both interest and consideration far greater than that
called for by a Chairman, for which I am sincerely
grateful.

To Dr. Richard Lewis, who served on my committee and
provided me with the technical knowledge and assistance
that aided me frequently and significantly in the research,
and will continue to benefit me in the future, I extend my
heartfelt appreciation and thanks.

To Dr. Melvin O'Connor, who served on my committee
and whose comments and guidance particularly in the early
stages helped me to focus my efforts and develOp an appro—
priate mental attitude concerning the dissertatidn, goes

iii

 

r 5::

:AA“ fl!

u4v~ ‘

vq:‘r‘

.ubv‘!‘

the 5

&A Q..

.v‘ 0|

my sincere thanks.

I would also like to eXpress my thanks to my fellow
doctoral students whose comments both in class and in
informal conversations contributed very meaningfully to
the successful completion of the thesis. And in addition
to my committee and fellow students, I would like to
specifically thank Professors Jones, Olson, and Collins
for their constructive criticisms and comments, and
Dr. Ronald Marshall for the hours spent in discussions
which proved fruitful to me. I am also, of course,
indebted to Jo McKenzie whose efforts in physically com-
pleting the thesis provided this end result.

Finally, but far from last in importance, I wish to
thank my wife, Barbara, for her unending support, confi-

dence, patience, and affection throughout.

iv

TABLE OF CONTENTS

LIST OF TABLES. . . . . . . . . . . . . . . .

LIST OF FIGURES . . . . . . . . . . . . . . .

Chapter

I

II

III

INTRODUCTION. . . . . . . . . . . . .

The Issues. . . . . . . . . . . . .
Awareness of the Problem. . . . . .
Import of the Income Statement. . .
Purpose . . . . . . . . . . . . . .
Organization of the Study . . . . .
Chapter I Footnotes . . . . . . . .

AN ELABORATION OF INCOME MANIPULATION

Arriving at Some Definitions. . . .
An Apparent Inconsistency . . . . .
Smoothing Research Problems . . . .
Cycle Prdblems . . . . . . . . .
Disclosure Prdblems. . . . . . .
Economic Conditions. . . . . . .
The Big Bath . . . . . . . . . .
A Proposed Manipulation Guideline .
Summary . . . . . . . . . . . . . .
Chapter II Footnotes. . . . . . . .

A REVIEW OF EMPIRICAL RESEARCH. . . .

Smoothing Research. . . . . . . . .
Gordon, Horwitz, and Myers . . .
Dopuch and Drake . . . . . . . .
Gagnon . . . . . . . . . . . . .
Archibald. . . . . . . . . . . .
Cepeland and Licastro. . . . . .
C0peland . . . . . . . . . . . .
Dasher and Malcolm . . . . . . .

V

Page
viii

ix

ll
15
l7
17
20
28
3O
3O
37
39

41

41
41
44
45
46
47
49
53

Chapter Page

White. . ... . . . . . . . . . . . . . 56
Barefield and Comiskey . . . . . . . . 58
Manipulation Research . . . . . . . . . . 59
Simpson. . . . . . . . . . . . . . . . 59
Copeland and Moore . . . . . . . . . . 62
Summary . . . . . . . . . . . . . . . . . 64
Chapter III Footnotes . . . . . . . . . . 65

IV. RESEARCH QUESTIONS, DESCRIPTION OF THE

RESEARCH DESIGN AND APPLICATION. . . . . . 68
Research Questions. . . . . . . . . . . . 68
Guideline Model Selection . . . . . . . . 71
Time Period Examined. . . . . . . . . . . 72
Data Collection Sources . . . . . . . . . 72
Sample Selection. . . . . . . . . . . . . 73
Variables . . . . . . . . . . . . . . . . 77
Classification System . . . . . . . . . . 79
Analysis Technique. . . . . . . . . . . 83
Application of the Research Design. . . . 88
Summary . . . . . . . . . . . . . . . . . 89
Chapter IV Footnotes. . . . . . . . . . . 90

V. RESEARCH RESULTS AND ANALYSIS . . . . . . . 91
Stepwise Regression Analysis Findings . . 91
Behavior of Income Statement Variables. . 101

Strong-Strong Case . . . . . . . . . . 102
Weak-Weak Case . . . . . . . . . . . . 103
Strong-Weak Case . . . . . . . . . . . 105
Weak-Strong Case . . . . . . . . . . . 106
Behavioral Testing Results. . . . . . . . 107
Strong-Strong Test . . . . . . . . . . 108
Weak-Weak Test . . . . . . . . . . . . 109
Strong-Weak Test . . . . . . . . . . . 111
Weak-Strong Test . . . . . . . . . . . 114
Summary of Behavioral Tests and Some
Further Testing. . . . . . . . . . . . . 116
Summary . . . . . . . . . . . . . . . . . 127

‘11. SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS . 129
Summary of Results. . . . . . . . . . . . 129
Conclusions . . . . . . . . . . . . . . . 135

Recommendations . . . . . . . . . . . . . 137

vi

Chapter Page
Appendix I . . . . . . . . . . . . . . . . . . . . 142
Appendix II . . . . . . . . . . . . . . . . . . . . 145

Bibliography. . . . . . . . . . . . . . . . . . . . 147

vii

Table

5-3

5-4

5-5

5-6

LIST OF TABLES

Page
Two Variable Classification as Smoothers
or Nonsmoothers . . . . . . . . . . . . . . 51
Comparative Smoothness of Income Streams. . 55
Reports with Bath Characteristic. . . . . . 63
Sample Firms. . . . . . . . . . . . . . . . 78
Frequency of Number of Variables
Reported. . . . . . . . . . . . . . . . . . 79
Variables Used in Study . . . . . . . . . . 80
Variables Found to Provide Improvement. . . 95
Data for Sample Firms . . . . . . . . . . . 97
Characteristic Distribution . . . . . . . . 108
Characteristic Distribution . . . . . . . . 109
Characteristic Distribution . . . . . . . . 110
Characteristic Distribution . . . . . . . . 110
Characteristic Distribution . . . . . . . . 112
Characteristic Distribution . . . . . . . . 113
Characteristic Distribution . . . . . . . . 114
Characteristic Distribution . . . . . . . . 115
Characteristic Distribution . . . . . . . . 118
Characteristic Distribution . . . . . . . . 120

viii

LIST OF FIGURES

Figure Page
1 Distribution of R2 Values for the Sales—
Net Income Regressions. . . . . . . . . . . 92
2 Distribution of R2 Values for the Sales-
Net Income Regressions with the Sign
Of R. O O O O I O O O O O O 0 O O I O O O 0 94
3 Distribution of Variables of PM Firms
Negatively Correlated with Net Income . . . 122
4 Distribution of Variables of PM Firms

Negatively Correlated with Sales. . . . . . 123

5 Example of Potential Manipulation
Through Tax Expense Variable. . . . . . . . 124

6 Examples of Tax Expense as a Percent
of Reported Net Income. . . . . . . . . . . 126

ix

CHAPTER I

INTRODUCTION

The first chapter of this dissertation presents the
issues upon which the research was based, the resulting
purpose of the research, and the approach and organiza-

tion of the chapters to follow.

The Issues

 

The degree of confidence in the reliability of the
published financial statement information was the central
issue upon which the research was based. More specifi—
cally, the issue involved the fact that there is consider-
able controversy over the reliability of the process of
income determination.

Charges have been levied by persons in various dis-
ciplines concerning the creditability of the income state-
ment.1 Although the charges have taken many different
forms, a great deal of the criticism has been aimed at
the flexibility of income measurement guidelines. It has
been demonstrated that the flexibility of these measure-
ment guidelines (or accounting methods) will permit many

1

2
different measures of income to be reported for the exact
same set of events.2 As a result, it is possible for a
firm to exert some influence on its reported net income
by its selection, or change in selection, of accounting
methods to be employed in the measurement process. To be
sure, different accounting methods do not exist for the
purpose of allowing firms to exert influence on their
income measurement. The different accounting methods
reflect the fact that all items of a particular classifi-
cation of items are not sufficiently uniform among firms
to be accounted for in the same manner.

For example, all items classified as inventory are
not so similar that a single accounting method, which
served as a guideline for their measurement, has been
found to be generally acceptable. Varying situations have
necessitated the development of different methods of
accounting for inventory which have been pragmatically,
if not logically, derived. To require all inventory to
be accounted for in the same manner would be far too
restrictive and unrealistic to permit general agreement
to be attainable. On the other hand, an expansion of the
classification of items to permit the definition of all
the various types of inventory and the most apprOpriate

method of accounting measurement for each classification

3

could provide a solution to the flexibility prdblem.
However, unless the “appropriate method" were somehow
determined on a case by case basis, the result would
simply be a more detailed and enumerative "use—this—
method-for-this-general-situation" type of guideline
system, a mere expansion of our present situation and
the problems relating thereto.

While criticism of the flexibility permitted in
accounting methods selection and the resulting lack of
rigidity in income measurement continues, two facts
should be stated in defense of such criticism. First,
improvements in reporting requirements have been made in
recent years which now require disclosure of the effects
of changes in accounting methods on reported income.3
This is in addition to the fact that frequent and sig-
nificant changes in accounting methods had previously
been guarded against to a considerable extent by auditing
standards and procedures.

Second, while much is made of the permissiveness of
accounting methods, little mention is made of other
vehicles available to the management whose intention is
to exert influence on its reported income. The manage-
ment of a firm may choose to influence reported income

by the way it handles the recording of non-transaction

4
based events, such as the determination of additions to
"allowance", "reserve", and to contra accounts which
affect revenues and expenses. Also, real events such
as the amount to be expended on advertising or research
and development may be specified or adjusted with the
intention of influencing income. There are numerous ways
in which a firm's management might act to influence re-
ported income without changing accounting methods, if it
so desired.

These two facts do not refute criticism of the
income measurement process. They do imply that changes
in accounting methods are not frequently occurring
events which are flagrantly misused by management, and
that other vehicles for influencing reported income do
exist. By no means do they imply that a potentially
serious prdblem does not exist.

It is clear that a firm's management, through various
actions, is capable of exerting influence on the measure—
ment of various revenue and expense accounts and is
thereby able to manipulate its reported income. And if
situations do exist, where management actions are taken
with the intention of manipulating reported income, then
the income statement becomes nothing more than a measure

of the aspirations of the management that contrived it.

5

The mere fact that such a situation may exist, that the
reported income figures might be influenced by a firm's
management, presents the accounting profession with a
critical weakness that may reduce the credibility and
usefulness of the entire accounting function. To sub—
stantiate the exiStence of the problem and its importance,
the degree of awareness of the problem of the reliability
of the income statement and the significance of the in-

come statement itself must be examined.

Awareness of the Prdblem

It is appropriate to say that unless there is an
awareness of a problem situation, then no problem exists.
The literature of accounting, finance, and business in
general certainly reflects an awareness to the potential,
if not real, prdblem of the reliability of the income
measurement process and the resulting income statement.

The presence of such terms as "Income Smoothing",
"The Big Bath", "Quality of Earnings", and "Income
Manipulation" can be frequently found in such literature.
The general implication of such terms is that management
deliberately adjusts reported income. The most general
term for such activity is income manipulation, which is

the central issue of this research.

6
The frequency with which the general concept of
income manipulation has appeared in the literature is sub-
stantial enough to confirm that an income statement cred-
itability prdblem does exist. A review of the literature
which has been directed at the concept of income manipu—

lation is examined in Chapter III.

The Import of the Income Statement

 

There is a generally accepted theory that the income
statement is currently the most widely used, if not the
most important, segment of accounting data. This theory
is evident by the evaluation of the income statement and
its role in relation to other accounting information pro—
vided by both the 1966 American Accounting Association
committee to prepare A Statement of Basic Accounting
Theory and the American Institute of Certified Public

Accountants Accounting Principles Board Statement No. 4.

 

Substantiation of the apprOpriateness of this theory is
very clearly pointed out by Ball and Brown who, through
their empirical research involving stock market reactions
to various bits of information, arrived at the following
conclusion:

"Of all the information about an indi—

vidual firm which becomes available

during the year, one-half or more is

captured in that year's income number,u5

7

Such conclusions support the theoretical premise that
users of financial data rely considerably on the informa-
tion content of earnings.

There exists additional information which substanti-
ates the importance of the income statement. For example,
in the finance literature most models depicting investor
behavior are based on the present value of the discounted
future stream of dividends. The association between
dividends, earnings per share, and reported income is
substantial and was clearly established by McDonald as a

foundation for his dissertation.6

glasses

The importance of the income statement and the aware-
ness of a problem concerning its reliability have been
established. The purpose of the research reported here
was to test the ability of a prOposed classification
technique to effectively discriminate between firms which
had potentially manipulated reported net income and those
firms which had not manipulated reported net income. The
classification technique was based on a fundamental income
statement relationship which is explained in detail in
Chapter II .

The accounting profession is faced with numerous

8
prOblems today, and while the creditability of the income
statement is only one of the problems, it should be
viewed as a serious one.

The information in an income statement is a major
part of the total output of accounting information. Many
new and different types of information, some of it related
to income reports (such as earnings forecasts), are being
sought by those who use financial data, and the accounting
profession is being called on to provide much of it. If
the profession cannot substantiate the quality of its cur-
rent output in the face of criticism concerning that out—
put, it is doubtful that it can or will play a meaningful
role in providing new and more desirable types of informa-
tion. If the profession cannot clearly and meaningfully
report on what has taken place in the past period, it is
extremely doubtful that it can for example provide meaning-

ful forecasts of what the future will hold.

Organization of the Study
Chapter II of this dissertation reports on the
analysis of income manipulation concepts and definitions,
and considers the role of income smoothing in the manipu—
.1ation framework. The general prdblems which have been

Gancountered in previous manipulation research are

9
considered. A method of classifying firms as potential
manipulators and non-manipulators is then described,
followed by an analysis of how the prdblems encountered
in previous research are dealt with by the classification
method.

As previously mentioned, Chapter III contains a
review of the previous research in the area of income
manipulation. The review is limited to the empirical
research efforts which have been reported in the litera-
ture.

Chapter IV includes a statement of the research ques—
tions examined in the dissertation. The research method-
ology and the data collection techniques which were em-
ployed in the research are then described, followed by a
brief description of how the research methods related to
the research questions.

Chapter V of the dissertation reports the results of
the research testing, an analysis of the results and how
they provide the answers to the two research questions
posed in Chapter IV.

And finally, Chapter VI provides a summary of the
research findings and the conclusions drawn from the

research.

10

Chapter I--Footnotes

For examples, see R. M. COpeland and M. L. Moore,
"The Financial Bath: Is It Common?", MSU Business
Topics, Vol. 20, No. 4 (Autumn, 1972), pp. 63-69:
John H. Myers, "Depreciation Manipulation for Fun
and Profits", The Financial Analysts Journal (Nov.-
Dec., 1967), p. 119: "What Are Earnings? The
Growing Credibility Gap", Forbes, May 15, 1967,

p. 42; Leonard Spacek, What is Profit? (Cambridge,
England: The Institute of Chartered Accountants in
England and Wales, 1970), p. 26.

 

 

 

For examples, see M. C. O'Connor and J. C. Hamre,
"Alternative Methods of Accounting for Long-Term
Nonsubsidiary Intercorporate Investments in Common
Stock", The Accounting Review, Vol. XLVII. No. 2
(April, 1972), pp. 308-319, and also R. J. Chambers,
"A Matter of Principle", The Accounting Review,

Vol. XLI, No. 3 (July, 1966), pp. 443-457.

 

 

Accounting Principles Board, "Opinion No. 20:
Accounting Changes", and "Opinion No. 22: Accounting
for Policy Changes", APB Accounting Principles,
(American Institute of Certified Public Accountants,
New York, New York).

 

See note 1 above.

Ray Ball and Philip Brown, "An Empirical Evaluation
of Accounting Income Numbers", The Journal of
Accounting Research, Vol. VI, No. 2 (Autumn, 1968),
p. 176.

 

 

Charles Le Roy McDonald, An Empirical Examination of
Published Predictions of Future Earnings. Unpublished
Ph.D dissertation, Michigan State University, 1972.

 

CHAPTER II

AN ELABORATION OF INCOME MANIPULATION
AND A PROPOSED GUIDELINE

Arriving at Some Definitions

The literature to date coming under the general
heading of income manipulation can be divided into two

general areas.

1. Literature dealing with the tOpic of
income smoothing.

2. Literature dealing with other types of
income manipulation such as taking a
"bath."

The majority of references to the concept of income
manipulation which appear in the literature are not based
upon empirical research. Of the empirical research that
has been conducted, a great deal of it has been in the
income smoothing area. This point is suggested by COpe—
land who states, "One manipulating goal widely attributed
to management is the desire to smooth reported income."1
Further evidence of the prominence of research dealing

with income smoothing as Opposed to manipulation in

general is provided in the review of empirical literature

11

12
in Chapter III.

An income smoother is usually defined as a firm which
manipulates its reported income to obtain a smooth income
trend. Stated another way, an income smoother is a firm
whose "managers perceive their performance measure to be
a decreasing function of earnings variability".2 These
definitions of income smoothing appear to be clear but
are realistically somewhat difficult to apply in attempt-
ing to empirically determine if a particular firm is an
income smoother. And while variations from these defini-
tions are slight, it is significant that they are not the
only definitions. Although the various definitions of
income smoothing are not identical, and few definitions
explicitly indicate that smoothing income is the result
of an act of management, they all strongly imply that the
smoothing phenomenon is the result of management actions
to manipulate.

Given that income smoothing may be defined as the
result of management actions to significantly decrease
the variability of earnings through the manipulation of

3 it is clear that income

the income statement variables,
manipulation is a necessary but not sufficient condition

for income smoothing to be present. In other words,

income manipulation (defined to mean the result of

13
management actions of any type which are undertaken to
significantly influence reported income) is necessary if
income smoothing is to take place, but income manipulation
itself does not necessarily result in a decreasing vari—
ability in reported income (or income smoothing). Income
smoothing not only necessitates income manipulation, but
it also requires that the manipulation be in the appro—
priate direction. Income smoothing may be viewed as a
special case of income manipulation. Hence, an income
smoother is an income manipulator whose objective is to
manipulate income so that the income pattern through time
presents a smooth trend line.

There is not complete agreement in the smoothing
literature as to what the pattern of the trend line should
be. While some feel that the pattern should be a constant
income trend, others feel the pattern should present a
strictly increasing income trend to reflect a constant
4

or an exponential growth rate.

The first concept is expressed by the formula:

E(I)t+1 = It (1)

where: E(I) expected value of reported

t+1 . . .
income in period t+l
I = actual income reported in
period t.

14
The variability of income using this model is
reflected by the sum of the absolute changes in income
from period to period, since the expected income number in
the coming period is that income which was last reported.
The second concept of income smoothing is similar to
the first except that the actual income in period t is

Operated on by a constant and/or an exponent such that:

B
= +
E(I)t+l aIt e (2)
where: e = residual error term,
a = constant,
B = exponent,
= a ab v
E(I)t+l s o e,
It = as above.

In this model, the expected income number for the coming
period is some multiple of the income which was last re-
ported.

The estimated rate of increase in income could be
based on a time series regression estimate computed from
previous reported income data, or some other basis. The
variability is reflected in the sum of the residual error
terms. The general concept of a constant growth model
has been hypothesized by some to be the "ideal norm of
5

behavior".

There is no strong support for selecting one concept

15
of smoothing over the other. If the theory that firms
have a life-cycle pattern where they experience periods
of growth, stability, and decline is accepted, then the
most apprOpriate smoothing concept in any given instance
would depend on a case by case analysis to determine
what stage of its life cycle a firm was in. The life-
cycle theory rules out the possibility of a single
acceptable concept of a pattern or trend which would
reflect income smoothing. Both concepts have their draw-
backs which are examined in more detail later in this

chapter.

An Apparent Inconsistency

 

Given that the definitions of income smoothing and
income manipulation are acceptable, it has been shown
that income smoothing is a special and more restricted
case of income manipulation. It can also be stated that
the empirical research on income smoothing makes up nearly
all of the research available in the area of income manip-
ulation. This appears to be illogical since it would
seem that the more restricted case of income manipulation
would be more difficult to detect than income manipulation
in general. While there is some agreement that income

smoothing is a special case of manipulation, there is no

l6
reason explicitly or implicitly stated in the literature
to explain why most research has been directed at this
special case of manipulation.6

The situation may be due to the fact that there
exists a generally agreed upon and clearly defined
standard by which income smoothing may be measured, which
is the variability of reported income. The models pre-
sented earlier describe the techniques which have been
used to measure such variability. In defining manipula-
tion as management actions which are undertaken to signifi-
cantly influence reported income, the effect of manipula-
tion on the reported income numbers is unclear. Hence,
income smoothing, by virtue of its having a predictable
effect on reported income, has a degree of specificity
which provides for its potential measurement while income
manipulation does not.

Before income manipulation in general can be ade-
quately researched, a guideline for detecting manipulation
must be developed, since a commonly accepted guideline
does not currently exist. A proposed guideline for
detecting income manipulation is develOped later in this

chapter.

l7

Smoothing Research Prdblems
The prdblems which have a bearing on income smoothing
research and on general manipulation of income as well
will be discussed under the following four categories:
1. Cycle Prdblems
2. Disclosure Problems
3. Economic Conditions

4. Bath Phenomenon

Cycle Prdblems

It has already been established that the life—cycle
theory of the firm presents some problem, to the degree
that the theory is realistic. If firms do have a life
cycle portrayed by a period of growth followed by a period
of stabilization and later decline, then it would have a
definite bearing on the type of smoothing model used to
measure the variability of reported income.

The constant model (1) would fail to detect smoothing
in growth firms, and the exponential or growth model
(2) would fail to detect smoothing in the stable firms,
and both models would fail to detect smoothing for firms
in a state of decline (unless the a coefficient and/or the
B exponent in model 2 were allowed to be less than unity).

In order to select the appropriate model in each case,

1:"

18
some scheme for classifying all firms into one of the
three life cycle stages would be necessary. No recognition
of the life cycle theory has been made in the research
dealing with income smoothing and, to the extent that the
theory is appropriate, the results of such research have
been distorted.

It would be possible to apply both models (1 and 2)
to each firm and select that model which provided the
"best fit" to the reported income streams. Such a tech-
nique, aside from being biased toward Obtaining a maximum
number of inconsistently defined smooth income streams,
has some Obvious prOblems.

If a firm were in the process of changing from one
stage in the life cycle pattern to another during the
period which the reported income data were collected, it
could be misclassified as a "non-smoother" while in fact
it was a smoothing firm. Also, the assumption that such
a technique truly classifies firms into their apprOpriate
stage of the cycle would necessarily have to be made and
supported in order to substantiate the inconsistency in
the models used to measure variability. In addition to
these prOblems, the appropriate classification of the
random sample, or the sample of non-smoothing firms must

‘be made to permit a meaningful "benchmark" for comparison

19

with both model and model 2 smoothers. In summary, to the
extent that the life cycle theory is appropriate it pre-
sents income smoothing research with a considerable prob-
lem to overcome.

Another cycle problem affecting smoothing research is
that of cyclical industries. Some industries are commonly
felt to be cyclical in nature, such as the building mate-
rials industry.7 The inclusion Of firms in cyclical in-
dustries in the sample of firms used for income smoothing
research represents a problem very similar to that of the
life cycle prOblem discussed above. An additional question
regarding firms in a cyclical industry is: Should the
cycle be reflected in the reported income stream, and if
so, to what degree?

While no reference has been made to the effect of the
life-cycle theory, some indirect recognition has been
given to the cyclical trends of reported income in
general. Two studies discovered that a shorter time
period over which the data are collected to fit the
various smoothing models will result in a better fit of
the income data to the trend line than in longer time

periods.8

The implications are that shorter time periods
for test purposes will reduce cyclical effects in reported

income and will decrease the standard error of estimate of

20
income trend lines. As a result, what little considera-
tion has been given to the cycle effects in general has
only served to restrict the usefulness of the "smoother,

non-smoother" classification systems that do exist.

Disclosure PrOblems

 

Perhaps the most pressing problem facing manipulation
research is the lack of disclosure Of sufficient income
statement data. Nearly all income smoothing research has
relied on the examination Of various income statement
variables to either support a "smoother, non-smoother"
classification system derived independently Of the income
statement variables, or use them as a basis for classifying
firms as smoothers or non-smoothers.

The general purpose behind the examination of various
income statement variables is to determine if they result
in income smoothing or income manipulation. Unless income
statement variables are disclosed, it is impossible to
determine if they are being used to smooth or manipulate
income. Therefore, a firm that actively manipulates
income could conceal the means by which it conducted its
income manipulation simply by not disclosing the income
statement variables used to manipulate. This may be

easily accomplished by consolidating manipulated variables

21

and other variables into some aggregated income statement
classification such as "cost of goods sold" or ”selling,
general and administrative expenses". Such massive con—
solidation of income statement accounts will tend to
Obscure any efforts to manipulate, making the detection
Of manipulated variables virtually impossible.

While there are guidelines for disclosure in both
published financial statements and financial statements
presented to regulatory agencies, they generally do not

provide sufficient disclosure for many research needs.

Public Requirements

The American Institute of Certified Public Accountants
(AICPA) has stipulated its general disclosure requirements
in its reporting standards which are a part of the ten
Generally Accepted Auditing Standards to which all audited
financial statements must conform.

The purpose of the standards as established in the
Statements on Auditing Procedure (SAP) NO. 33 follow:

Auditing standards differ from auditing
procedures in that "procedures" relate

to acts to be performed, whereas "stan-
dards" deal with measures of the quality
Of the performance of those acts and the
Objectives to be attained by the use Of
the procedures undertaken. Auditing
standards as thus distinct from auditing
procedure concern themselves not only
with the auditor's professional qualities

22
but also with the judgment exercised by
him in the performance of his examina-
tion and in his report.9

The four Reporting Standards deal with the report by
the auditor on the audited financial statements, the Third
Reporting Standard being:

Informative disclosures in the financial
statements are to be regarded as reason-
ably adequate unless otherwise stated in
the report.

As a result, unless the auditor's report indicates
otherwise, it may be assumed that the disclosure presented
in the published financial statements is "reasonably ade-
quate", without clearly distinguishing adequacy from in-
adequacy. Any sample Of published annual reports will
help to point out how loosely this standard can be inter-
preted. The general income statement format that is
typically presented consists Of "cost of goods sold",
"selling, general and administrative expenses", "tax
expense", and perhaps a few other accounts. The amount

of detailed account information to be found in pub-

lished annual reports is generally minimal.

Regulatory Requirements
While the annual reports provide very little income
account information to shareholders, there are other less

accessable sources which do provide income statement data

23
in some detail. The Securities Act Of 1933 and the Secu—
rities Exchange Act Of 1934 were the initial measures
taken by the government to regulate publicly traded com-
panies.‘ The Securities Exchange Commission (SEC), which
was established as a result Of the 1934 Act, has developed
and made numerous revisions to Regulation S-X which repre—
sents the most rigorous financial reporting requirements
imposed on today's publicly traded corporations. But
while the financial statements and supporting schedules
which are available as a result of regulation S-X provide
much more information than is found in annual reports, the
information is not readily available to the public. The
documents filed by corporations as a result of regulation
S-X are generally on file only at the Offices of the SEC
in Washington D.C.

Some services do Obtain and reproduce some financial
information from these documents (principally the annual
form 10—K), making it more readily available to the pub-
lic. Moody's has, for many years, derived certain finan-
cial facts from SEC reports to include in its annual
reporting service information, and stipulates those cases
where the financial information was Obtained from SEC
documents. More recently, some SEC derived financial

information has become available on the Compustat Tapes.

24

These sources have been invaluable to researchers seeking
detailed financial data not found in annual reports.

But the need for additional information continues.
The public at large, as well as the financial community,
continues to demand more and more information. The fact
that there is a greater demand for more readily available
information is reflected in the recent success of Leasco
Information Products, Inc. The following statements from
an article entitled, ”Disclosure-Decision Making with an
Edge" point out the demand for more financial data and
Leasco's role in filling that demand.

The ability of the financial community
to make sound decisions about its money
has always been largely a matter of how
much information it could lay its hands
on. That has been the principle under-
lying a series of federal acts dating
back to 1933 that require publicly
owned corporations to disclose increas-
ing amounts Of information about their
business.

For many years, the enormous volume of
documents filed by public corporations
(now approximately 10,000) have (sic) glutted
the shelves of the SEC and of libraries

of securities exchanges and other insti-
tutions that maintained c0pies. In early
1968 Leasco signed its first contract
with the SEC and began putting the various
corporate reports--now approaching 90,000
a year- on microfiche, 4- by 6-inch sheets
of film that can hold up to 60 pages each.

This method simplified the physical
handling and storage of these documents,

25

and made acquisition of all basic, un-
edited financial data a one-stOp Oper—
ation.11

While Leasco has thus far done little more than to
increase the availability of existing information, the
following point is well made:

Undeniably, the public interest is best
served when access to information is
increased. The best capital market to
fuel the American economic machine is
an informed investment community. But
until the advent of Leasco's "Disclo-
sure" services, the Objective of a
universal, widely used information
resource was always pushed a little
further out because of the increasing
number and complexity of documents
filed. There was simply no way to
keep apace with the information content
of this vital resource.12

In addition to the "Disclosure" system, Leasco has
develOped two new disclosure services to begin publication

in 1973, the Disclosure Index Journal and the Disclosure

 

Resume Annual.

The Disclosure Index Journal to be pub-
lished monthly, will provide access to
corporate reports released by the SEC
for microfilming by subject and by com-
pany. In addition to the 150 to 200
pages Of these indexes, the monthly
service will begin with a "Current
Awareness" section that will highlight
develOpments of immediate importance

to users. Twice a year, subscribers
will receive a cumulative index covering
all new documents filed during the period.
And once a year, an annual cumulative
index will appear classifying the

 

26

approximately 90,000 documents filed
during the previous twelve months.

The Disclosure Index Journal eliminates
the need to search the whole list of
companies to find those that have re-
ported a specific activity during the
period. A quick look under the rele-
vant heading in the index will provide
the names of all those that have re—
ported such activities. Brokers,
security analysts, bankers, institu—
tional and individual investors,
lawyers, accountants, scholars, and
all the other users Of corporate in—
formation will now be able to save
countless personnel hours formerly
devoted to slogging through company
files.

The demand for more information is being heard and
response to that demand is being made. Yet the availa-
bility Of financial information remains as somewhat of a
prOblem to past and current financial research efforts.
While the SEC has improved its reporting requirements
over the years, it is still seeking suggestions as to
what additional financial information could and should
be required. The practice Of consolidating basic account
data into an aggregated account persists even in SEC

reporting.

Comparability
Another disclosure prOblem that has had a bearing on

income smoothing research is the lack of uniformity in

reportel
ted to
smoothe
of a si
sale 0:

sarily

cial j
concex
plate
latio

firms

Consj

the (
1957
earn
acco
"ear
in p
incc
Prob

the

27
reported financial information. Some studies have attemp-
ted to accept or reject a hypothesized "smoother, non-
smoother" classification scheme by examining the effects
of a single income statement variable such as "gain on
sale Of marketable securities".14 Such research is neces-
sarily limited to those firms reporting such a variable.
The general inadequacy Of uniformly disclosed finan-

cial information is reflected by the fact that no research
concerning income manipulation has been based on a com—
pletely random sample Of firms. Firms included in manipu-
lation research are typically screened, and only those

firms which have the necessary financial data are included.

Consistency

Still another disclosure prOblem is inconsistency in
the composition of aggregated accounts. For example, from
1957 - 1962 Chrysler Corporation included its "equity in
earnings Of unconsolidated subsidiaries" in an aggregated
account titled "other expenses". Since 1962 these
"earnings" have been included in an account titled "equity
in net earnings of unconsolidated subsidiaries". These
inconsistencies make for difficult analysis and research
prOblems, and tend to restrict the usefulness of some Of

the detailed financial data that are available.

28
The lack of the quantity and availability of finan-
cial statement data, and the inconsistency and lack Of
comparability in existing financial data have all combined

to make disclosure a significant research prOblem.

Economic Conditions

Research aimed at detecting income smoothing has not
been fruitful. A low degree Of variability in reported
income streams is felt to be the goal and characteristic
of an income smoothing firm. Yet even if a firm did desire
to manipulate its reported income to reflect a smooth in-
come trend, there are certain events which might tempo-
rarily overrule such an Objective. There are various
economic situations that occur which might make it too
difficult or less desirable to maintain the smooth income
trend.

Situations such as the "credit crunch" experienced
during 1969-70 and 1965-66 affect the entire economic
community and make income smoothing more difficult and
less necessary. The management of a firm that reports a
decline in reported income in a year when most other firms
are also reporting declines does not have to be overly
defensive in explaining the declining figures providing

they are reasonable under the circumstances. Economic

lll

29
situations which create general decline (as well as others
which may stimulate economic activity) will tend to in-
crease the variability in reported income streams despite
possible management Objectives to the contrary, and thereby
weaken the criteria for the determination Of income
smoothing.

Other economic situations may influence overall
earnings variability Of a smaller segment of the economic
community. A change in tariff restrictions on imported
steel may have a similar impact on the steel industry. A
decline or increase in new resource discoveries may create
comparable economic situations for Oil, gas, copper,
gypsum, and various other industries which are heavily
dependent on natural resources. Here again, the vari-
ability Of reported income should increase as a result of
such situations and tend to distort smooth income trends.

Economic conditions such as were described above pre-
sent potential problems for income smoothing research, and
may help to explain why two smoothing studies found the
variability of income streams increased as the number Of

a a l 5 ’
years examined increased.

30

The BigyBath

 

An event which will inevitably tend to increase the
variability Of reported income streams is the phenomenon
Often referred to as "bathing". The "big bath" theory
generally implies a situation where a firm has a substan-
tial loss or decline in profits to report, and will add
to the loss or decline by charging Off previously deferred
expenses to the current period, and perhaps establishing
excessive reserves for possible future losses. As one
author states:

Management will Often seize on a period

in which losses or comparatively adverse
results are reported to create such re-

serves (reserves for future costs and

losses). A new management team will
Often desire to make a "clean sweep".

16
While the "bath" phenomenon implies manipulation of
past and/or present and/or future reported income figures,

it does not present a decreasing effect on the variability

of reported income.

A Proposed Manipulation Guideline

 

It would be useful if a means of determining which
firms manipulated or smoothed reported income could be
developed. Yet manipulation or smoothing of reported
income is virtually impossible to prove without an admis-

Sion of such Objectives on the part of a firm's management.

31
The most rigorous research methodology, no matter how
logically the hypothesis is developed, or how statisti-
cally significant the results, cannot prove income
smoothing or income manipulation.

The Objective Of the guideline develOped here is not
to prove income manipulation. But it is expected to yield
some additional insight into the income manipulation con-
troversy and serve as a useful means Of separating poten-
tial income manipulators from non—manipulators.

The classification system prOposed is based on a
fundamental income statement relationship. The relation-
ship is that which exists between sales and net income.
The source Of income is revenue, since without revenue
there can be no income. And since revenue is typically
recognized at the time of sale, the vast bulk of reported
revenue consists Of sales.

It seems logical to extend this concept further by
stating that there should be some relationship between
sales and net income, and therefore changes in the
reported sales pattern should be associated with the
Pattern of reported net income. The relationship is
fundamental. It deals with two of the most frequently
rEported pieces of financial data. TO the extent that a

3r81ationship does naturally and logically exist between

32
sales and net income, a standard by which manipulation of
reported income may be judged is provided.

The basic premise of this research is that there
should be a strong functional relationship between sales
and net income (defined throughout the remainder of this
study to be net income after taxes but before extra-
ordinary items and the taxes applicable thereto) over a
relevant range of sales activity. That is, over some
relevant range of sales that is pertinent to an individual
firm, a particular sales level should be a general indi-
cation of what the resulting level of reported net income
will be. Also, changes in a firm's level Of net income
should be related to the changes in the level Of sales.

The 3 priori premise is that there are firms which
have a strong functional relationship between sales and
net income. Such firms are believed to be non—manipu-
lators, since the changes that occur in reported income
streams are highly associated with changes in reported
sales. There is an additional belief (which is later
eXplained in more detail and which was tested as part of
the research) that such non-manipulating firms will possess
income statement variables that are also highly associated
with the changes in reported sales. Generally when the

changes in income statement variables are all highly

33
associated with changes in the pattern Of sales, the vari-
ables are construed to be behaving rationally and the non-
manipulator classification is apprOpriate.

On the other hand, if a firm is manipulating its
reported income, the functional relationship between
reported sales and reported income will be weak. The firm
that possesses a weak relationship between sales and in-
come should also be characterized by the presence Of in-
come statement variables which are weakly associated with
sales. A more detailed discussion of the association be-
tween sales, net income and other income statement vari-
ables is presented in Chapter V.

The basic guideline for determining firms which are
potential income manipulators and firms which are non-
manipulators is the association of the functional relation-
ship between reported sales and net income data, as mea-
sured by regression analysis. A brief review of the terms
and computational procedures of regression analysis as it
applies to this research is provided in Appendix I.

The proposed guideline for classifying firms as
potential manipulators and non-manipulators will overcome
most of the problems previously discussed.

The guideline is quite simple to use. The classifi—

cation system is realistic in that it does not imply, as

 

()

34
other researcher's classifications have, that the system
can positively detect firms that are manipulators or
smoothers. While the use of a "potential manipulator"
classification may appear to be too weak to be meaningful
at first, the real meaning Of such a classification cannot
apprOpriately be determined until the results are examined.
The analysis of the research findings in Chapter V pro-
vided the basis for evaluating the strength of the infer-
ences to be made from the classification system.

The fact that the prOposed guideline system is based
on the behavior of reported income streams relative to the
behavior of sales overcomes the cycle prOblems previously
cited. If a firm's reported income streams are not smooth
due to cycle effects but are either inconsistent with
sales that should reflect cycle effects, or relatively
smooth compared to sales (which again should reflect the
cycle effects), the system will classify the firm as a
manipulator. TO the extent that cycle effects are
reflected in the reported sales streams, the cycle prob-
lem is eliminated.

The disclosure prOblem will remain a problem to the
proposed guideline system, but to a lesser degree. The
classification Of a firm as a potential manipulator (PM)

or a non-manipulator (NM) requires only the sales and

35
net income data. However, to determine the validity Of
this classification system, it is necessary to support
the classifications by examining the behavior Of other
income statement variables.

The quantity Of income statement variables that is
disclosed does represent a constraint on the ability to
support the prOposed guideline. The income statement
variables that are selected will also have to be con-
sistently reported for a given firm over the period
tested. However, the study did not examine the manipula-
tive effect of any income statement variables in particular
and therefore the income statement variables selected for
examination need not be comparable among those firms in-
cluded in the study. In eliminating the comparability
requirement, it is assumed that manipulation Of reported
income can be achieved through any number Of variables,
and that different variables may be used to manipulate
the reported income Of different firms, or of different
periods for a single firm.

TO give apprOpriate recognition to the disclosure
problem, it was considered as the basis for sample
selection as explained in Chapter IV.' If the classifica-
tion technique which was tested is considered by the

reader to be meaningful, its usefulness will be facilitated

36
by the fact that the data necessary for its implementation
are readily available.

The potential problems presented by the impact Of
various economic events and the bathing phenomenon are
circumvented by the prOposed guideline system. TO the
extent that economic conditions having an impact on a par-
ticular firm are reflected in the sales stream of that
firm, the effect Of the conditions is taken into consider—
ation.

If the economy were in a period of recession, the
reported sales Of a firm should change relative to the
impact Of the recession on the firm, and reported income
should change according to the change in sales. A reces-
sion, or other types Of economic conditions and events,
should not distort the relationship between sales and net
income of a given firm. If a firm's net income reacted
more severely to a recession than sales, or if management
decided to take a "bath" in a period of declining sales
performance, the over-reactions reflected by the change
in reported income would be found to be inconsistent with
the change in reported sales. The result would be to
decrease the strength of the sales-net income relationship.

The proposed guideline system might also help to ex-

plain why some smoothing research results were

37
inconclusive. Typical income smoothing research has
classified firms with smooth trends in reported income as
"income smoothers", implying that their income was some—
how manipulated to decrease the variability of the income
stream. Yet it is possible that a smooth income trend
was the logical result of a smooth sales trend, and that
no manipulation of income statement variables was made to
achieve the smooth income trend.

The proposed system would eliminate such apparent
misclassifications since a smooth income trend that is
associated with a smooth sales trend would reflect a
strong sales-net income relationship which characterizes
NM firms. Income smoothing firms could be more precisely
defined as firms with a poor sales-net income relationship
(PM firms) gpd a relatively smooth income stream. It
would be quite interesting to see how this additional
criterion alone would effect the results of previous

smoothing research.

Summary

In this chapter, the meaning of the terms "income
manipulation", and "income smoothing" was discussed and
defined in the context which they are used throughout the

remainder Of this dissertation. Income smoothing is but

38

a special case of income manipulation with which most
manipulation research has concerned itself. Some of the
problems which have had a bearing on the outcome of pre—
vious smoothing research were examined. Finally a prO-
posed guideline designed tO classify firms as potential
manipulators (PM's) or non—manipulators (NM's) which
could overcome most Of the research prOblems previously
cited was presented. A closer look at previous manipula-

tion research endeavors is provided in the next chapter.

10.

ll.

12.

13.

39

Chapter II--Footnotes

R. M. COpeland, "Income Smoothing", Empirical Research
in Accounting: Selected Studies, 1968, p. 101.

 

 

Eugene E. Comiskey and Russel M. Barefield, "The
Smoothing Hypothesis: An Alternative Test", The
Accounting Review, Vol. XLVII, No. 2 (April, 1972),
p. 291.

Perhaps the most explicit support of such a defini-
tion is provided by R. M. Copeland, Op. cit., p. 101.

For example, Dasher and Malcolm, and Gordon, Horwitz
and Myers used the exponential growth model while
Gagnon, Archibald, COpeland and Licastro. and COpeland
all used the constant model (1) (See footnotes l, 3,
10, 15, and 17 in Chapter III for complete citations).

Paul E. Dasher and RObert E. Malcolm, "A Note on
Income Smoothing in the Chemical Industry", Th2
Journal of Accounting Research, Vol. VIII, No. 2,
(Autumn, 1970), p. 255.

R. M. COpeland, Op. cit., pp. 101-102.
Gary E. White, "Discretionary Accounting Decisions and

Income Normalization", The Journal Of Accounting
Research, Vol. VIII, NO. 2 (Autumn, 1970), p. 261.

 

P. E. Dasher and R. E. Malcolm, op. cit., p. 257,
and R. M. Copeland, Op. cit., pp. 114-115.

Committee on Auditing Procedure, Auditing Standards
and Procedure (AICPA, New YOrk, New York, 1963),
p. 15.

 

 

Ibid., p. 16.

"Disclosure-Decision Making With An Edge", The Leasco
Magazine (An internal publication Of the Leasco
Corp.), Vol. 4, NO. 4 (April, 1973), p. 15.

 

Ibid., p. 17.

Ibid., p. 15.

14.

15.

16.

40

Nicholas Dopuch and David F. Drake, "The Effect Of
Alternative Accounting Rules for Nonsubsidiary
Investments", Empirical Research in Accounting:
Selected Studies, 1966, pp. 192-219.

P. E. Dasher and R. E. Malcolm, Op. gi£., p. 257,
and R. M. COpeland, Op. cit., pp. 114-115.

Leonard Bernstein, “Reserves for Future Costs and
Losses: Threat to the Integrity Of the Income
Statement", The Financial Analysts Journal (Jan-
Feb, 1967), p. 147.

 

c.)
p)
'1

M'

‘4

1.

.Se

CHAPTER III

A REVIEW OF EMPIRICAL RESEARCH

Of the literature involving income manipulation, nine
of the eleven research studies have been involved with
income smoothing. While income smoothing is not the
principal concern Of this research, its relationship to
income manipulation warrants consideration. A summary Of
the income smoothing research is provided, followed by a
review Of the two studies concerning manipulation of in-

come. The studies are reviewed in chronological order.

Smoothing Research
Gordon, Horwitz, and Myers1
The earliest research efforts concerning income

smoothing were conducted by M. J. Gordon, B. N. Horwitz,
and P. T. Myers (GH&M). The GH&M study used 21 firms in
the chemical industry as the sample to be tested. The
firms did not represent a random sample. The variable
used to test the smoothing hypothesis was the investment

credit, as accounted for by the 21 selected firms during

41

42
1962 and 1963.

During the 1962—63 period, two different accounting
measurement rules were commonly used: (1) taking the tax
savings from the investment tax credit into income during
the year Of asset acquisition only: (2) distributing the
tax savings over the life Of the acquired asset. Income
smoothing

...was tested by considering whether an
accounting measurement rule (the invest-
ment credit) was selected which tended

to: (1) adjust the firm's percentage
change in earnings per share to the
average percent change in the industry,

or (2) smooth the firm's earnings per
share toward a normal value, or (3) smooth
the firm's rate of return on stockholders
equity.2

In a review Of the GH&M study, COpeland states:

Normal income and smoothed earnings were
defined with considerable rigor .....

The results of their analysis were
inconclusive. Introducing double ex-
ponential smoothing to measure the
first two criteria (stated above) pro-
duced more error than it eliminated,
thus leaving Open to question the
validity Of their evidence.

Conclusions based on the third criterion stated above
were questioned in comments made on the GH&M study by
Zeff, because:

One, a rate of return on net assets is
rarely found in corporation annual re-

ports, suggesting that managers appar-
ently do not intend to convey a notion

43

of the success of operations in terms
Of that criterion. Two, financial
analysts utilize a relationship between
income and market value per share, not
book value per share.

In addition to criticism concerning the three
criteria used to test smoothing, Copeland states:
......the long-run smoothing potential
Of the investment credit is question-

able because the device, once adopted,
commits a firm to report given amounts

at determinable future dates, and this

reporting may have antismoothing effects.

Furthermore, the testing design used by

GHM only examined manipulative adjust-

ments for one year at a time so that

conclusions about smoothing for indi-

vidual firms could not be drawn, i.e.,

some maximizers, minimizers, or un-

systematic manipulators may have been

classified as smoothers.5

In addition to the above study, GH&M also examined

the effects Of five measurement rules on the time series
trend Of reported net income Of United States Steel Corpo-
ration for a twelve year period. The measurement rules
examined included: (1) the method of depreciation used,
(2) the use Of guideline lives for depreciable assets,
(3) the estimates of past service pension costs, (4) the
estimated current pension cost, and (5) the investment
credit. While the managementHs use of these measurement

rules roughly corresponded to the smoothing criteria

established above, the lack of consistency in its

44
selection of alternative measurements resulted in incon-

clusive evidence.

Dopuch and Drake7

 

In a 1966 study, Nicholas DOpuch and David F. Drake
(D&D) examined the effects of dividend income and gains on
the sale Of marketable securities on the variability of
the reported income trend. D&D began with 1000 firms
that reported market values of securities to the SEC.

They then determined those firms whose cash plus market-
able securities represented more than 25 percent of the
total assets for any year between 1954 and 1964, according
to data listed on the Compustat Tapes. This reduced the
sample from 1000 to 630. Of the 630 firms, only 200
reported cash plus nonsubsidiary investments that repre-
sented greater than 20 per cent of total assets for the
entire 11 year period. Through examining information in
Moody's and SEC lO-K reports, the sample was reduced to
12 firms which had substantial investments in marketable
securities and which reported the market values of those
securities for the years 1954-1964.

The examination of these twelve firms revealed that
smoothing was not generally facilitated by the sale of

marketable securities.

45

Except for a few cases when the reported

gains and losses were significant, the

smoothing effects were due to dividend

income rather than to the sale Of secu-

rities.8

While the results Of the D&D study indicated that,

for the firms tested, dividend income represents a more
prominent smoothing tool than the sale of marketable secu-
rities, their conclusions concerning the use of either
method to smooth income were not overwhelming.

There was some evidence of smoothing,

but we did not Observe any gross dis-
tortions of reported income.9

sasneel 0

As part of a study initially reported in 1967 by
Jean-Marie Gagnon, the effects Of the purchase versus
pooling measurement choice On income smoothing was to be
examined. While the principal purpose Of the research
proposed by Gagnon was to determine if there was some
method Of predicting how a merger would be treated (pur-
chase Or pooling of interest), one aspect of the study
did intend to "find Out whether there is any empirical
basis for assuming that managers seek to smooth reported
income".11

The sample used by Gagnon consisted of the examina-

tion of 500 New York Stock Exchange mergers occurring

46

during the period 1955-1958. Smooth income was defined to
be a consistently reported income figure from year to year
(model 1). The results Of Gagnon's research, published
four years later, indicated the selection of the purchase
vs. pooling of interest measurement choice did not corre-
spond to smoothing.12 As suggested in a review Of Gagnon's
prOposed research by Copeland13 and confirmed by Gagnon in
his cOncluding remarks from the completed study, "It is
doubtful that income smoothing can explain accounting deci-

sions having a long term effect on reported income."14

Archibald15

 

T. Ross Archibald, in a study reported in 1967,
examined the effects of a change from accelerated depre-
ciation methods to straight-line depreciation on the
reported trend in income. The sample used consisted of 55
firms which (subsequent to the 1954 change in the Internal
Revenue Code permitting accelerated depreciation) had
adOpted accelerated depreciation methods for both finan-
cial statement and tax purposes, and later reverted to
the straight-line method for financial reporting while
retaining the accelerated method for tax purposes. Since
a change back to the straight line depreciation method

would increase reported income in the year of change, an

47
income smoother would wait until a period Of declining
profits before making the change. As summarized by
COpeland:

Archibald found that 22 Of the 55
switch back firms had lower profits

in the year Of change, but Offered

no conclusion on the smoothing hy-
pothesis. However, any conclusions

on smoothing would have been question-
able On two methodological grounds.
First, Observation was made only once
on one manipulative variable, so that
a pattern Of behavior could not be
determined. Some Of the 22 declining
profit firms may have been maximizers
or randomly acting nonmanipulators.
Moreover, change in depreciation
method calls for an auditor's quali-
fication on the consistent application
of accounting principles. If firms
were really trying to artificially
sweeten profits, it is unlikely that
they would do so in a form that exposes
their actions.16

Archibald, like Gagnon, defined a smooth income
stream to be a constant reported income number from year
to year. His study did not, at the time it was reported,
actually measure the effect Of the depreciation change on
the variability Of reported income. The Archibald study

remains incomplete as Of this writing.

Copeland and Licastrol7

In a 1968 study, Ronald M. COpeland and R. L.

Licastro (C&L) reported the effects of dividends received

48
from unconsolidated subsidiaries on the reported net in-
come figures. C&L felt that such dividend income could
be an effective smoothing technique for firms which
carried unconsolidated subsidiaries on the cost basis
while being able to effectively control the subsidiaries.
In such cases, managers could wait until the end of a
reporting period to determine the level of income before
dividends, then request dividend payment from the con-
trolled unconsolidated subsidiaries in the amount neces-
sary to achieve the desired income level. The inability
of a subsidiary to pay the dividends could be circumvented
by the use Of intercompany loans.

The C&L sample included 20 New York Stock Exchange
firms which had seemingly controlled unconsolidated sub-
sidiaries that were reported on the cost basis. The 11
year period examined was 1954 to 1965. As reported by
COpeland:

For each firm, the sign of the year-to-

year change in earnings was compared to

the sign Of the change in dividends

remitted to the parent. A chi-square
contingency test was applied to the

169 sets of data. The evidence sta-

tistically supported the (null) hy-

pothesis that the dividend-income tech-

nique was not used to smooth income.18

Aside from not yielding significant results, the C&L

study suffers the same methodological problem as the GH&M

49
and the Archibald studies. Examining the effects Of a
variable on the year-tO-year change in income does not
give an indication Of the trend in reported income
streams, and since income smoothing implies a smooth
£5229 in reported income such a test can not possibly
indicate income smoothing.

A weakness that should be noted with all studies up
to this point is that they attempted to detect income
smoothing by examining a very limited number of income
statement variables. In doing so, they made two strong
.2 priori assumptions: (1) the income statement variables(s)
examined is (are) the most likely means of achieving in-
come smoothing, and (2) each variable will be consistently
used to either smooth or not smooth income. While it may
be possible to select variables p_priori which most likely
are able to facilitate a desire to smooth income, there is
no apparent reason to assume that any one particular vari-

able will consistently be used for income smoothing.

Copeland19

Later in 1968, Copeland reported the results Of
another income smoothing study. COpeland's Objectives
were to specify the attributes Of accounting variables

which could be used to smooth, to evaluate earlier

50
smoothing research, and to test several smoothing
hypotheses.
In his analysis Of accounting variables Copeland
determined that smoothing devices should be restricted
to accounting practices or measurement rules.
A smoothing device ought to involve
only accounting interpretation of an
event, not the event itself. Accoun-
ting manipulation is a matter Of
form, not Of substance.20
The sample used in the study was comprised of 19
firms whose published financial statements found in their
annual reports reflected the presence Of at least two
potential smoothing variables, one of which was dividend
income in each case. A smoothing firm was defined as "one
which uses the variable to smooth in a majority of periods
examined", and a variable was said to have smoothed in-
come ”if year-tO-year changes in the variable are such as
to decrease year-to—year variances in income".21
Using these definitions, a test was made to determine
if the classification Of firms (as smoothers and non-
smoothers) based on the dividend-income variable (Variable ,
l) was significantly different than the classification of
firms based on all other variables examined (Variable 2).

If the two classification schemes lead to approximately

the same results, the inference would be that a

51
classification based on one variable would not be signifi-
cantly different than a classification based on a number

Of variables.

The results of the chi-square test, reproduced below,
indicate that the two classifications do not yield com-
parable results at an a = .05 level Of significance

(x2 = 4.27>critical X2 value Of 3.84) .22

TABLE 3-1

Two Variable Classification of Nineteen
Firms as Smoothers or Nonsmoothers During
Sixty-eight Consecutive Four-Year Periods

 

Non-
Smoothers Smoothers Totals

Variable 1 (Dividend—Income) 31 37 68

Variable 2 (All other vari-
ables) 43 25 68

 

Totals 74 62 136

The other major test conducted in the study was to
determine the effect of the length of the time series ex-
amined. Two, four, six and eight year time series were
tested. Using the definitions cited above, the number of
firms classified as non-smoothers by both Variable 1 and
Variable 2 increased from 27% to 43% as the length of the
series increased from 2 to 6 years, and declined to 41%

in the eight year series. The strongest inference to be

52
made from the time series results is that, using Copeland's
definitions and classification scheme, the length Of time
studied will have a bearing on how a firm is classified.
While Copeland does not completely define the accounts
included in Variable 2, he does state that it includes:
.....extraordinary charges and credits,
write-Off of fixed assets or intangibles,
cessation or unusual changes in pension
charges, changes in accounting methods
or procedures, fluctuations in contingen-
cies or self-insurance reserves, changes
in deferred charge or credit accounts,
introduction Of the 53rd week year, and
others.23
The inclusion of many Of these variables, as well as
the dividend-income variable, appears to contradict the
attributes Of smoothing variables defined by Copeland at
the outset of the study. In a footnote Copeland admits
that: "While few of these variables meet all the criteria
previously established, all possess some smoothing poten-
tial".24 This contradiction points out the weaknesses of
the far too restrictive criteria established for smoothing
variables.
While COpeland does not precisely define his testing
procedures, it appears that, since only 19 firms were in-
cluded in the study and the test results cited above

included 68 four year time series, a given firm could be

classified as both a smoother and a non-smoother over the

53
test period. The number of cases where a given firm
might change classifications cannot be determined since
the number of years covered by the study varied between
four and twenty years. The inclusion of a single firm
as a number Of Observations is a questionable procedure,
especially since the sample was so small and was not
randomly selected.

While the results Of the study are interesting, they
are tentative at best. The problem Of supporting the
classification system employed was not considered. The
analysis of a variable's effect on year-to—year variances

in reported income was estimated by COpeland in cases

 

where the dollar amounts of the change in a variable
could not be determined, but the direction Of the change
and its effect on reported income was felt to be
"Obvious".25 While COpeland did an excellent job of
critically reviewing the smoothing research conducted

to that time, he Offered very little in the way of

smoothing research improvements.

Dasher and Malcolm26

In a study reported in 1970, Paul E. Dasher and
RObert E. Malcolm (D&M) differentiated between real and

artificial smoothing as follows:

54

Real smoothing refers to an actual trans-
action that is undertaken or not under-
taken On the basis Of its smoothing effect
on income, whereas artificial smoothing
refers to accounting procedures which are
implemented to shift costs and/or revenues
from one period to another.27

While D&M define real and artificial smoothing, they
do not apply the definitions to their study since the two
both are capable Of smoothing and sometimes hard to dis-
tinguish from one another. The D&M sample consisted of
52 firms in the Chemical and Chemical Preparations Indus-
tries which enabled them to compare their results with the
GH&M study. The sample was non-random.

The four variables examined for their smoothing
effect were: (1) pension expense, (2) dividend revenues,
(3) research and develOpment expense, and (4) extraordinary
items. The model used to measure variability was an ex-
ponential growth model similar to model 2 presented in
Chapter II. D&M, from the least squares estimate of their
growth model, calculated the standard deviation Of the
estimate (SDE) of reported income, and reported income
less the effects of the four smoothing variables (net of

tax), for both an eleven year (1956-66) and a six year

(1961-66) period. The results are reproduced below:

55
TABLE 3-2
Comparative Smoothness of Income Streams

 

6-year ll-year
period period

Average ratio Of reported-income SDE to
geometric mean Of reported income ..... .135 .212

Standard deviation of above ............. .146 .148
Average ratio Of income-before-smoothing
SDE to geometric mean Of reported

income.. .............................. .148 .224

Standard deviation Of above ............. .145 .148

 

The most Obvious and perhaps the most meaningful result is
that the income streams are less smooth for the ll-year
period than for the 6-year period for both reported income
and "income-before-smoothing". This result is consistent
with COpeland's findings.

D&M test the presence of smoothing using the two
SDE's to form a smoothing ratio (SR). As stated in the
study:

This (SR) consists of the SDE of income
before smoothing variables to the SDE of
reported income. An SR greater than 1.0
would indicate greater variability in the
"income before smoothing" and a tendency
toward deliberate smoothing of reported
income. An SR less than 1.0 would tend
to indicate a "nonsmoother".29

Their results revealed that 35 Of the 52 firms had

SR's greater than 1.0 for the ll-year data, and 40 firms

56
had SR's greater than 1.0 for the 6-year data. While the
results appear to be very impressive, they do not provide
any support for the classification system employed.

The use of ratios is also questionable since they do
not provide any measure of the degree of smoothing as it
is defined 3 priori, and they are completely unrelated to
the goodness of fit of the exponential smoothing model
employed. Also, since the geometric mean of reported in-
come is not used to normalize the two SDE's used in the
ratio, it seems logical to assume that the SDE of "income-
before-smoothing" will be greater than the SDE of reported
income (which results in a ratio greater than 1.0) since
the value Of the income-before-smoothing numbers will be
greater than the reported income numbers. Carried to the
extreme, the SDE Of sales over the SDE of reported income

will surely be greater than 1.0!

White30

In another 1970 study, Gary E. White attempted to
determine "whether the motive of income smoothing or in-
come normalization can explain managements' selection Of
accounting alternatives".31 White's analysis considered
two industries, the chemical indus'ry (for comparison with

the GH&M and D&M studies), and the building materials

57

industry. White used both linear and lOgarithmic least
squares regression models to fit earnings per share (EPS)
trends, and selected the model which yielded the best fit
to the EPS data (the model which provided the highest R2)
for each firm over the 10 year period from 1957 to 1966.
The 10 firms in each industry which had the highest R2's
for either model and a positive EPS trend over the entire
test period were classified as income smoothers. The non-
smoother groups were made up Of 10 firms randomly selected
from each industry. The effects of various management
accounting policy decisions on income statement variables
were then examined to determine if the variables were used
to achieve the positive EPS trends of the smoothing com-
panies. In the analysis of his research results, White
states that:

On the whole there is no evidence that

companies in the smooth samples sig-

nificantly achieved their positive

least-squares trends by their choice

of accounting alternatives. This

suggests that smooth trends were

achieved by chance and/or by con-

trolling variables other than the

accounting policy decisions included

in the study.32

It was mentioned earlier that it would be interesting

to see how the proposed guideline would effect previous

smoothing research. One of the firms included in the

58

White study was also included in the current study. White
classified the firm as a smoother due to the goodness Of
fit of the EPS trend to the regression model. This study
classified the firm as a non-manipulator however, since
the smooth upward trend in reported income was strongly
associated with the smooth upward trend in sales. Depend—
ing on the acceptability Of the proposed guideline, this
case suggests that a potential prOblem may exist in the
frequently used method of classifying firms as smoothers
or non-smoothers, that is, based on their reported income

trends alone.

Barefield and Comiskey33

 

A study reported in 1972 by Russel M. Barefield and
Eugene E. Comiskey (B&C) attempted to determine if firms
could achieve smooth income trends through their choice
between the cost and equity method of accounting for sub-
sidiaries. The study included a sample Of 30 firms for
the period 1959 to 1968. The firms were selected based
on the availability of necessary information in published
annual reports and included ten firms which used the cost
basis, 14 firms which used the equity basis, and 6 firms
which used both methods during the 10 year test period.

B&C employed a linear growth model to measure the

59

variability in reported earnings and the effects of the
cost-equity choice on earnings variability. The conclu-
sions Of the B&C study were not inconsistent with previous
research efforts in that they cited disclosure of informa-
tion and sample size as limitations Of the research effort.
B&C reported that:

Analysis of the data seem (sic) to indi-

cate that conditions vary sufficiently

to insure that neither accounting method

applied across subsidiaries will con-

sistently result in smoother earnings.34

This result is consistent with the notion suggested

by COpeland that it is difficult to smooth income streams
over many periods through the selection of a single
accounting method during a single period which commits the
firm to an accounting procedure for future periods. B&C

did find "modest support for the hypothesis that firms

select that method which produces smoother earnings".35

Manipulation Research

Simpson

Richard H. Simpson reported the results of his
research findings in a 1969 article entitled "An Empirical
Study of Possible Income Manipulation".36 The study is

the only empirical research effort known which deals

specifically with income manipulation.

60

In the study, Simpson examined the effects Of four
“accounting practices" on reported income. The four prac-
tices of concern were: (1) the investment tax credit,

(2) unusual gains and losses, (3) investment in the common
stock Of other firms, and (4) Operating losses. These
practices are said to have manipulative qualities since
they all permit varying acceptable accounting treatments
for situations that are not significantly different.
Simpson determined 3 priori the single accounting treat-
ment for each Of the four "practices" which he felt was
"acceptable". For example, he determined that the invest-
ment tax credit should reduce the cost of the qualifying
assets with the benefit thus spread over the assets'
useful life. The flow-through method which takes the
entire credit in the year Of a qualified asset acquisition
was deemed "unacceptable".

A random sample of 85 New York Stock Exchange firms
was selected and then reduced to 49 firms which met
Simpson's sample selection criteria. The selection
criteria were generally aimed at eliminating new firms,
foreign firms, non—industrial firms, and firms which were
subsidiaries Of other firms.

Simpson adjusted the reported incomes for 1964 of

the 49 sample firms to determine what the "corrected

61
income" would have been had the "acceptable" accounting
practice been employed. The firms were then ranked
according to the total absolute percentage Of change re-
sulting from the "unacceptable" practices (pp£p_increasing
and decreasing effects combined) on "corrected incomes".
The firm with the median rank was eliminated and the
remaining 48 firms were divided into two groups Of 24,
one with the lowest ranked firms (whose "unacceptable"
practices had little or no effect on corrected income)
and the other with the highest ranked firms.

Simpson ran a number of tests to try to determine if
the two groups had characteristics which were significantly
different. The tests were aimed at determining whether or
not there were differences in: (l) the overstatement or
understatement Of corrected net income, (2) uncorrected
price-earnings ratios, (3) debt to equity ratios, and
(4) firm sizes as measured by sales level.

Only the first item listed was found to be signifi-
cantly different between the two groups. This result is
certainly understandable since the overstatement or under-
statement of corrected net income was instrumental in
ranking firms and separating them into two groups.

The strength of Simpson's study rests primarily on

his 3 priori determination Of which accounting practices

62
currently considered as being "generally acceptable" are
"acceptable" from his point of view. TO the extent that
others agree with Simpson's point of view on which prac-
tices are acceptable or not acceptable, the study does
provide a method Of determining whether a firm is an in-

come manipulator or not.

Copeland and Moore37

In a recent article, Ronald M. Copeland and Michael

L. Moore (C&M) reported on their research Of a specific
type Of manipulation, the corporate bath. The purpose of
the C&M study was stated as follows:

...to determine (1) how frequently com—

panies utilize discretionary accounting

decisions to reduce income, that is, take

a bath: (2) what are the economic condi-

tions with respect to income or stock

price movements that exist when discre-

tionary accounting decisions are made:

and (3) whether the size of discretion-

ary accounting decisions is related to

the magnitude of the adverse income or

price movement.
The discretionary accounting decisions referred to by C&M
include write—Offs, write-downs, and provisions for future
charges which cannot be objectively determined at the time
the provision is made.

The C&M sample was develOped by randomly selecting

1,000 firms from the 1,800 companies listed on the Compustat

63

Tapes and requesting their annual reports for the years
1966 through 1970. The 907 companies which responded con-
tributed a total of 3,761 annual reports, with the number
varying between 545 and 871 for individual years from 1966
to 1970.

A company was characterized as a "bath company" if it
reported extraordinary items which were based on discre—

tionary accounting decisions and which decreased the in-

 

come (Or loss) before extraordinary items by ten percent

or more. A summary Of the frequency with which the "bath"

characteristic was detected is reproduced below.39

TABLE 3-3

Number Of Reports to Have the
Bath Characteristics

 

Total Bath Col 3
umn
‘ngp Reports Reports Percentage(EBTEfiH—§)
1970 765 58 7.58
1969 871 53 6.08
1968 822 34 4.14
1967 758 33 4.35
1966 545 17 3.12
Total 3761 195

Further testing revealed that there were statisti-
cally significant relationships between those firms which
made discretionary accounting decisions and those which
had income declines or management changes. NO relation-

ship was found between firms with the "bath"

64

characteristic and stock price declines.

Summary

While a good deal Of time and effort has been exerted
on research involving the manipulation Of income, very few
strong inferences or conclusions can be made as a result
Of these endeavors. Some common problems have been: the
lack Of detailed income statement information, the incon-
sistency and lack Of comparability in information that is
available, the inability to test a truly random sample
(due in part to the previously cited information gathering
problems), and the general inability to meaningfully
characterize or otherwise confirm proposed smoother-non-
smoother or manipulator-nonmanipulator classification

systems.

11.

12.

l3.

14.

65

Chapter III--Footnotes

M. J. Gordon, B. N. Horwitz and P. T. Myers,
"Accounting Measurements and Normal Growth of the
Firm" in R. K. Jaedicke, Y. Ijiri, and O. Nielson
(eds.), Research in Accounting Measurement, American
Accounting Association (1966), pp. 221-31.

Ibid., p. 224.

Ronald M. COpeland, "Income Smoothing", Empirical
Research in Accounting: Selected Studies, 1968,
p. 106.

 

S. A. Zeff, "Discussion Comments", in R. K. Jaedicke,
Y. Ijiri, and O. Nielson (eds.), Op. cit., p. 250.

R. M. COpeland, Op. cit., p. 106.

M. J. Gordon, B. N. Horwitz and P. T. Myers, Op. cit.,
pp. 221-31.

Nicholas DOpuch and David F. Drake, "The Effect Of
Alternative Accounting Rules for Nonsubsidiary Invest-
ments", Empirical Research in Accounting: Selected
Studies, 1966, pp. 192-219.

 

Ibid., p. 206.
Ibid., p. 207

Jean-Marie Gagnon, "Purchase Versus Pooling of
Interests: The Search for a Predictor", Empirical
Research in Accounting: Selected Studies, 1967,
pp. 187-204.

 

Ibid., p. 188.

Jean-Marie Gagnon, "The Purchase-Pooling Choice: Some
Empirical Evidence", The Journal of Accounting Research,
Vol. IX, NO. 1 (Spring, 1971), pp. 52-72.

 

R. M. Copeland, Op. cit., p. 108.

J. M. Gagnon, Op. cit., p. 63.

VJ

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

66

T. Ross Archibald, "The Return to Straight-Line
Depreciation: An Analysis of a Change in Accounting
Method", Empirical Research in Accounting: Selected
Studies, 1967, pp. 164-80.

 

R. M. COpeland, Op. cit., p. 107.

R. M. Copeland and R. L. Licastro, "A Note on Income
Smoothing", The Accounting Review, Vol. XLIII, NO. 3
(July, 1968), pp. 540-45.

 

R. M. Copeland, pp;_gip., p. 109.

Ibid., pp. 101-116.

1219-: pp. 104-5.

2229'! p. 111.

Epig., p. 113.

.gpig., p. 111.

Ibid., p. 111 (footnote 28).

‘gpgg., p. 111.

Paul E. Dasher and Robert E. Malcolm, "A Note on
Income Smoothing in the Chemical Industry", Th3

Journal of Accounting Research, Vol. VIII, No. 2
(Autumn, 1970), pp. 253-259.

 

lpig., p. 253.

gpig., p. 257.

g2;g., p. 257.

Gary E. White, "Discretionary Accounting Decisions
and Income Normalization", The Journal of Accounting

Research, Vol. VIII, No. 2 (Autumn, 1970),-pp. 260-
273.

 

 

Ibid.. p. 260.

Ibid., p. 270.

33.

34.

35.

36.

37.

38.

67

Eugene B. Comiskey and Russel M. Barefield, "The
Smoothing Hypothesis: An Alternative Test", Th2
Ascounting Review, Vol. XLVII, No. 2 (April, 1972),
pp. 291-298.

 

Ibid., p. 298.
Ibid., p. 298.
Richard H. Simpson, "An Empirical Study Of Possible

Income Manipulation", The Accounting Review, Vol.
XLIV, NO. 4 (October, 1969), pp. 806-817.

 

R. M. Copeland and Michael L. Moore, "The Financial
Bath: Is It Common?", MSU Business Topics, Vol. 20,
NO. 4 (Autumn, 1972), pp. 63-69.

 

Ibid., p. 68.

CHAPTER IV

RESEARCH QUESTIONS, DESCRIPTION OF THE
RESEARCH DESIGN AND APPLICATION

The purpose of the research underlying this disser-
tation was to determine if the proposed classification
technique could distinguish between potential manipulators
Of reported net income and non-manipulators. This chapter
indicates the research questions that were considered in
attempting to support or refute the proposed classifica—
tion technique. A description of the research design is
then presented, followed by a general indication of how

the research design was applied to the research questions.

The Research Questions

 

The purpose Of the research testing was to determine
if the analysis of income statement variables of the two
groups of firms would provide evidence to either support
or refute their prOposed classification as potential
manipulator (PM) and non-manipulator (NM) firms. In the
regression sense, the differences in the sales-net income
relationships of the two groups were not necessarily

68

69
associated with known differences in the nature Of their
respective income statement variables. The research ques-
tions examined and reported in this thesis were as follows:

1. DO income statement variables Of the PM firms

have a significantly different effect on their
respective sales-net income relationships than
the income statement variables of the NM firms?

2. Do the income statement variables of the PM

firms and the NM firms have significantly
different behavioral characteristics?

The first question was designed to indicate when the
basic regression relationship between sales and net income
was capable Of being significantly improved through the
addition of income statement variables, and if the fre-
quency with which improvement occurred was significantly
different for PM firms than for NM firms. Significant
improvement may be interpreted as an instance where a
variable was capable Of reducing a meaningful portion of
the residual error left by the basic sales-net income
regression.

The second question was designed to indicate when
major differences occurred in the types of behavior
patterns reflected by the income statement variables of

the two groups of firms. A more detailed explanation of

7O
behavior patterns and the means of classifying a variable's
behavior is provided later in this chapter.

If the answers to the two questions posed above had
been negative, it would have implied that the differences
in the sales-net income relationships were either attribu-
table tO factors other than the income statement variables
(i.e. the effective or ineffective utilization of resources),
or that the variables which did affect the sales-net income
relationship were not examined in the study. In this case
the conclusion would have been that the proposed classifi-
cation system was not supportable by the research, implying
the classification system could not discriminate between
potential manipulators and non—manipulators.

If the answers to either of the two questions had
been positive, the classification system would not have
automatically been validated. If the variables of the two
groups did have significantly different effects on the
sales-net income relationships, and/or the variables of
the two groups did have significantly different behavioral
characteristics, it would have still been necessary to

relate the differences to manipulative qualities.

 

(1'

f7)

71

Guideline Model Selection

 

While the prOposed guideline was said to be based on
the sales-net income relationship, the exact form of the
relationship was not specified. In an attempt to deter-
mine an appropriate relationship, a sample Of 15 firms was
examined. Sales and income data for the firms were col—
lected for a 12 to 15 year period. The data were analyzed
by a statistical regression program which calculated the
coefficient Of determination (R2) for a linear and three
curvilinear relationships. The models of the relationships

tested are provided below:

..= .+ . ,+ ,,
(1) Y1] B03 B13 Xi] e13
..= .+ . ..
(2) Log Y13 B03 B13 Xij + e1]
..= .+ . ..+ ..
(3) Y1] B03 B13 Log X1] e1]
(4) Log Yij = Boj + Blj Log Xij + eij
where: Yij = the expected income Of firm j for year i
(the dependent variable),
xij = the reported sales Of firm j for year i
(the independent variable),
Boj = the computed regression constant of firm j,
Blj = the computed sales coefficient Of firm j,
eij = the residual error term of firm j for
year i,
and where; i = l, ..... ,12,...,15 years,

j = 1, ............ ,15 firms.

72

The analysis indicated that the linear relationship

eliminated the greatest amount Of squared error in the

majority Of the firms examined (9 firms), having
in the highest R2 most frequently. In the other
the linear model eliminated nearly as much error
any Of the other three models. As a result, the

model was selected as the regression form of the

The Time Period Examined

 

resulted

six firms

as did

linear

research.

The study was designed to cover a sufficient number

Of years to permit the influence Of various economic con-

ditions and cyclical trends to be included in the test

period. If the influences Of such matters were reflected

in the sales trend, as was suggested in Chapter II, there

would be a substantial number of firms whose sales-net

income relationships remained strongly associated. The

data collected for the study were Obtained from the years

1955 to 1971.

The Data Collection Sources

 

The ideal data collection source would have been the

SEC lO-K reports which provide more detailed account infor-

mation than any other publicly available source known to

the researcher. While it was mentioned earlier that

Leasco provides these reports on microfiche, it was

73

discovered that they are available only from 1968 to the
present. Previous lO-K reports were not available locally.

Other available sources were Moody's and the Compu-
stat Tapes, as mentioned earlier. After examining a num—
ber of lO-K reports and comparing them to the data which
were available on the Compustat Tapes and in Moody's, it
was determined that the amount Of additional data which
could have been obtained from examining the lO-K's
directly as Opposed to the other two sources was marginal
at best. As a result, the Compustat Tapes and Moody's

were the sources of data utilized in this research.

The Sample Selection

 

The firms included in the study were not randomly
selected. The researcher could have randomly selected a
large number of firms and then eliminated those which did
not meet the sample selection criteria, as many smoothing
studies have done. However, such a procedure would have
been of little value since, in the end, such a sample
would have been non-random.

Since this research tested the proposed classifica-
tion Of firms through the analysis Of variOus income
statement variables, the quantity Of variables disclosed

was considered to be the primary selection criterion.

74

Selection Of firms based on the quantity of income state-
ment data provided presented a potential problem which
had to be considered.

It was suggested by COpeland and others that if
information concerning a particular variable is provided
to the public in annual reports, it is probably not being
used to smooth or manipulate reported income. In other
words, to effectively manipulate or smooth income would,
according to some, require that the variable used to smooth
or manipulate not be reported separately. If a disclosed
variable were being used to smooth or manipulate the vari-
able's manipulative effect would be readily ascertainable.
There is probably some truth to this concept and its merit
was considered in designing the study. The effect of
extraordinary items, for example, that were reported in
the financial statements which accompany the annual reports
were not considered for their manipulative qualities.
Also, the data for the variables selected were collected
from sources other than the published annual reports of
firms. However, some Of the variables examined in the
study were also provided in the annual reports of the
sample firms to permit a test Of the manipulative qualities
Of variables which are generally available tO the public.

The impact Of one commonly disclosed variable is referred

75
to in the analysis in Chapter V.

Included in the Compustat Manual is a chart which
lists all information items contained on the Tapes, and
the frequency with which each item was recorded for each
year covered. The two items which were least frequently
reported in 1955 (Research and DevelOpment, and Selling
and Advertising Expenses) were selected from this chart
and a program was written to Obtain the names Of the com-
panies which reported either Of these items consistently
from 1955 to 1971. It was hOped this selection would prO-
vide those companies which reported the greatest amount Of
information.

A printout Of all the income statement information
(for the period 1955 to 1971) of each of the 63 companies
meeting the initial selection criterion was then Obtained
from the Compustat Tapes. These firms were reviewed and
retained in the study if the following criteria were met:

1. The firm had no drastic changes in its com-

postion during at least 13 consecutive years
of the test period (ie., no significant mergers
or acquisitions, etc).

2. The firm's data contained at least five income

statement variables consistently and consecu-

tively reported each year for at least 13

76
years of the 17 year test period.

3. The data available in Moody's combined with

the data from the Tapes resulted in at least
7 income statement variables consistently and
consecutively reported each year for at least
13 years Of the 17 years examined.

This review reduced the number Of firms tO 26.

TO supplement this group of firms, a printout Of all
income statement information for the same period was ob-
tained for all firms in a number of Compustat industry
groups. These data were Obtained from the Tapes in an
attempt to include in the sample a number Of firms in the
same industry to allow some comparisons by industry. The
industry classifications of the first 26 firms selected
were considered in deciding which industries to examine,
as was the number of firms in each particular industry
classification (generally, only industries with less than
ten firms were examined). The criteria cited above were
then applied to the industry groups.

The resulting non-random sample consisted Of 41
firms. The sample contained an information quantity bias,
as well as a bias toward certain industries. These biases
resulted strictly from a desire to examine firms with sub-

stantial account detail and to permit some comparisons

77
within industries. The sample firms and their Compustat
Industry Classifications are listed on page 78, in

Table 4-1 .

The Variables

 

Since the research did not concern itself with any
single variable's manipulative capacity, it was not neces-
sary to have variables uniformly present among the firms
included in the study.

The study also avoided discrimination among "real"
and "artificial" variables. All variables are capable of
being manipulated to the extent that they may be influenced
by discretionary management actions. For example, discre-
tionary actions would include management decisions to
change accounting measurement methods (or accounting
principles), to establish or alter the level of expendi-
tures accounted for in various accounts, or to write—Off
or capitalize certain expenditures. While it would have
been very interesting to discriminate among the various
types of discretionary management decisions, it was fre-
quently impossible to determine what type of action influ-
enced a given variable in a given period.

As previously indicated, the principal criterion

which was considered in determining whether or not a firm

-.lt\ In! FC..<.-.

78

 

.oo mcocamama

uuoammmue ufi<

Hmuauao

mumamoum<

.Ou< a muumm Ou=<
.-

mOHOHnm> scuoz

I
.aaaam .msm a muamao
.msuca a
.cmol%umcfinumz
mamasmumz .wnsm
you“: Hmmum

acnmz|ammum

.9 w .H.<

.<.3.H

mmcwaufi< cmufipms< smm
mOCHHuH< usmwuo amassOuoz
mmcfiapfic Hmcoaumz
mmcfiaua< muamo
mmcfiaufi< cmufiums<
cfixumm uwsam
pmmnxooq

mommm zmmamm

mu0uox Hmumcwo
“mammucu

mwsom zmsufim

umumfiwmm nmmo .Humz
newswommmupp<

macaw: unmamum
.Hwnm wcficMOUImamao
mch .mcoo mcoummmm
meum .m.D

ammum oasnaamm

:

uohmzlammum

.ummson .ummucHIHfio
..

mufiuwsmou

.omom a .meImwsua

.-
Hmuwnumlmwsua

.-
uOanImHmOaamno
usfiN w pmmq
.umfizlmamumz

Hmmum HmGOHumz
sfianwsma paw swoop
Hmmum pachH

meum oosu<

.s.z .Hao uumuamum
mauuso madam:
wuumHHHo

.mnmq umuxmm
nmsoam wcfiuwnom
.uaH Manama
mpfinumu cows:

aHHo

Oucwmsoz

mmasoumm

momuo .m .3

uaoosn .H .m
Hmuasmno son
paamaho uncapped
manuasmnu pmfiaa<
mamumcwz mop .um
.mmm a uawsm cmuauwa<

 

msmz knumsch

mamz mamanu

msmz zuumspca

msmz mamafioo

 

Hue mqm<H

79
was included in the study was that it provided at least 7
variables. A summary of the 41 firms and the number of

variables they reported is provided below.

TABLE 4-2

FREQUENCY OF NUMBER OF VARIABLES REPORTED

 

Total
Number of Variables 11 10 9 8 7 -

Number of Firms 1 2 20 10 8 41

 

 

Since the data for the variables were collected from
two different sources, it was necessary to confirm some
information with both sources to insure that the data per-
tained tO the same reported sales and net income figures
in both cases. In cases where it was ambiguous as to when
a variable might have been included as part of another
variable (ie. "dividend income" from Compustat included in
"other income" in Moody's), the variable in question was
eliminated.

A summary of the 347 variables for which information
was gathered, and the frequency with which each variable

was included is presented in Table 4-3 on page 80.

The Classification System

 

Since the proposed classification system is based On

the assumption that firms with a strong sales-net income

80

TABLE 4-3

VARIABLES USED IN STUDY

 

General Account Title

Frequency of

 

Inclusion

Depreciation Expense 41
Interest EXpense 33
Tax Expense 41
Pension Expense 38
Incentives Compensation Expense 12
Selling and Advertising EXpense 11
Research and DevelOpment Expense 20
Rental EXpense 21
Non-Recurring Expenses 7
Unconsolidated Subsidiaries-Remitted Earnings 6
Maintenance and Repairs EXpense 33
Other Income 30
Other EXpense 7
Dividend Income 7
Dividend and Interest Income 4
Rent and Royalty Expense 2
Royalty, Joint Venture, and Other Income 2
Selling, General and Administrative EXpense 6
Interest Income 3
General and Administrative EXpense 2
Amortization of Special Tools 2
Gain on Sale of PrOperty 6
Others (with only 1 inclusion each) _12_

Total 347

 

relations
tionships
define "V
bans fo
tion (R2
of the a
It
cant di:
examini:
a: fig
test (t
in the
to char
from t}
level C
ference
Where '
the dij
the fac
a23!le (-
be: of
If
“My 8.

R, the n

81

relationship are non-manipulators and those with weak rela—
tionships are potential manipulators, it was necessary to
define "weak" and "strong" relationships. The logical
basis for the definitions was the coefficient of determina-
tion (R2), which provides a relative measure Of the strength
of the association between sales and net income.

It is possible to determine if statistically signifi-

cant differences exist between different R2

statistics by
examining the differences in the standard error Of R (where
R = /R5 = the coefficient of correlation). The statistical
test (the Student t test) to determine if the differences
in the standard error of R were real differences, or due
to chance, was applied to the regression data generated
from the 15 firms in the pre-study sample, using a .05
level of significance. It was discovered that major dif-
ferences in the R2 statistics (ie. R2 = .90 vs. R2 = .20
where .00£R251.00) were necessary in order to infer that
the differences were real. This result was largely due to
the fact that the inference from the test, which would
apply to all time periods, was based on such a small num-
ber of observations (12 to 15 years).

If the classification had been based on statisti-

cally significant differences in the standard error of

R, the majority of the firms would have been eliminated

82
from the study due to their lack Of sufficiently high or

2 statistics. If the classification was not based

low R
on statistically significant differences, then no

statistical inference could be made about periods other

 

than the period tested.

Since the number of years examined would have had to
have been greatly increased to permit the majority of
firms examined to yield statistically significant differ-
ences in R2 for classification purposes, and since the
amount of income statement data available in years prior
to 1955 decreased rapidly, the statistical classification
method was rejected. This was a regretable consequence
since the resulting PM-NM classification of firms was
necessarily based on the Opinion Of the researcher and

therefore is subject to question. If statistical infer-

 

ences were limited to the period examined only, then 33y
difference in the R2 values was, statistically a £231
difference.

Having been forced to an.§ priori definition of weak
and strong relationships to permit the PM—NM classifica-

2

tion of firms it was decided that firms with R values

above .75 would be considered to have sufficiently strong

sales—net income relationships to be classified as NM

2

firms. An R Of .75 meant that 75% of the squared error

83
Of estimating net income was eliminated by basing the
estimate of net income on the sales level associated
with it (using the least squares regression estimate).
In absolute terms, by interpreting the coefficient of
association, this meant that 50% of the relative error

(unsquared) was eliminated by using the regression esti-

 

mate. Hence, firms whose sales-net income relationship

was strong enough to permit the elimination of over 50%

of the relative error in estimating net income were classi-
fied as NM firms.

In order to permit a clear distinction between PM

2

firms and NM firms, a void R range was used to eliminate

borderline firms. Firms with R2 values Of less than .50

2

were classified as PM firms, and firms with R values

between .50 and .75 were unclassified and eliminated from

the study.1

The Analysis Technique

 

The technique used to analyze the data was a stepwise
linear regression routine. The basic equation analyzed

can be expressed as follows:

estimated net income,

where: Y

X1 = reported sales,

84

X2 - X = income statement variables,

“0 = constant,

Bl = sales coefficient,

B2 - Bn = coefficients Of income statement variables,

e = error term.

The regression program was used to calculate the strength

of the association between sales and net income, and to

measure the significance of the income statement variables

on the sales-net income relationship. To determine the

impact Of the variables on the sales-net income relation-

ship, three analytical methods were tested:

1.

a deletion routine which initially included
sales and all income statement variables
examined in the original regression equation

as independent variables, and deleted the in-
come statement variables which did not sig-
nificantly contribute (at an a = .05 level Of
significance) toward explaining the dependent
variable, net income:

an addition routine which began with net income
as the dependent variable and sales as the only
independent variable, then added any of the
income statement variables which significantly

contributed (at a = .05)toward explaining the

85
dependent variable:

3. an addition routine which began with net in-

come as the dependent variable and sales as

the only independent variable, and which

then examined each income statement variable

individually to determine at whatlevel of

significance it would have been added to the

regression equation.
All three methods considered sales as a fixed independent
variable which was not allowed to be added or deleted from
the regression equation. Also, all three analysis methods
provided an intercorrelation matrix which permitted the
examination of the association of any one variable with
all other variables.

Due to the intercorrelation effects of the independent
variables, the deletion routine permitted variables which
were not significant, in and of themselves, to remain in
the final regression equation. Also, it was possible that
the deletion routine eliminated variables which, in and of
themselves, were significant. For example, if the dele-
tion routine was considering eight variables and the last
variable examined was significant by itself, but its asso-
ciation with the dependent variable had already been

"explained" by one or a group of other variables due to

86
the intercorrelation effects, the last variable examined
would have been deleted from the equation.

The first addition routine (2) was also complicated
due to the intercorrelation effects. The program added
variables one at a time according to their level of sig—
nificance in the regression equation. However, if the
first variable added was highly correlated with other
variables, the other variables would not have been added
even though they may have been significant had they been
considered individually.

The second addition routine (3) determined whether
or not each individual variable was significant, without
regard to the intercorrelation effects of other variables.
Since the income statement variables were reported and
collected as individual variables, it was considered most
apprOpriate to analyze them individually. Hence, the last
routine was selected as the most apprOpriate analysis
technique.

It should be pointed out that the intercorrelation
problems encountered with the regression routines
described above were partially due to the fact that one
of the assumptions of regression analysis was violated.
When a number of independent variables are being used to

explain the behavior Of a dependent variable, it is

87
assumed that the independent variables are independent of
one another. Since all of the independent variables in
this study are used to help determine the dependent vari—
able in the format Of the income statement, they are not
clearly independent of one another.

On the other hand, the fact that they are all used to
arrive at the reported net income figure does not neces-
sarily mean that they are dependent on one another in the
regression sense. For example, there is no reason to
believe that the amount expensed for research and develop-
ment is dependent on the amount expensed for depreciation
or for maintenance and repairs.

A certain amount of interdependence was implied by
the fundamental relationship assumed in this thesis. In
general terms, the income statement variables, to the
extent that they were variable in nature, were expected
to change relative to the changes in sales. To the extent
that the variables were fixed in nature, they would be
represented in the regression cdnstant and would not affect
the relationship. These general conditions were sufficient
conditions for sales and net income to be strongly asso-
ciated, but they were not necessary conditions. A firm
could have had an artificially strong relationship between

sales and net income. For example, two variables would

88

appear to be fixed, when considered together, if one
increased by $10 and the other decreased by $10 for every
$1,000 increase in sales. The first variable would have
been positively correlated with sales, the second nega-
tively correlated with sales, and if considered together
they would have been completely uncorrelated with sales.

The above discussion tends to support the regression
analysis technique which was selected. Examining the im-
pact Of each individual variable on the sales-net income
relationship minimized the interdependence problems, since
only two independent variables were considered at each

step Of the regression routine.

Application Of the Research Design
Data were collected for sales, net income and at
least seven income statement variables for each firm for
13 to 16 consecutive years. The data for each firm were
entered into the linear stepwise regression routine. The
regression's output provided the intercorrelation matrix,
the strength (R2) of the basic sales-net income relation-
ship, and the level Of significance of each income state-
ment variable examined. If a variable was significant at
the .05 level it was added to the regression and the re-

vised coefficient Of determination was calculated. If a

89
variable was not significant at the .05 level, the a level
at which the variable would have become significant was
calculated. Hence, the analysis technique yielded a rela—
tive measure Of the degree Of improvement provided by each
variable.

In attempting to answer the first research question,
the stepwise regression analysis was utilized. The fre-
quency with which PM and NM firms possessed variables that
significantly improved the sales-net income relationship
was examined for statistically significant differences.

In examining the second research question, concerning
the behavioral characteristics of the income statement
variables, the intercorrelation matrix was utilized. If
the PM-NM classification Of firms was meaningful, the be-
havior of the variables examined would indicate differences

between the two groups of firms.

Saw

The research questions which, when answered, would
help determine the appropriateness of the proposed classi-
fication system have been presented. The research method-
ology which was employed in answering the questions was

then described and related to the research questions. The

research results and the analysis of those results are

presented in the following chapter.

90
Chapter IV--Footnotes
1. A sensitivity analysis was performed in order to

determine the validity of this elimination and may
be found in Appendix II.

CHAPTER V
RESEARCH RESULTS AND ANALYSIS

The Stepwise Regression Analysis Findings
The stepwise regression routine provided the calcula—
tion of the coefficient Of determination (R2) Of the basic
sales-net income relationship for each of the 41 firms.
The distribution Of R2 values is summarized in Figure l,
on the following page.

The use Of the R2

statistic for analysis purposes was
preferred to the use of R which is not meaningfully defin-
able. The R statistic does, however, indicate whether the

2 statistic

correlation is positive or negative which the R
does not. Since a negative correlation Of any magnitude
(.00 to -1.00) was considered weaker than any positive

correlation, it was necessary to discriminate between R2
values based on positive coefficients Of correlation (R
values) and those which were based on negative coefficients
of correlation. While the mathematical process of squaring
the coefficients of correlation eliminates the possibility

of negative R2 values, the sign of the R statistic was

91

92

 

 

 

 

 

 

 

 

 

 

 

 

 

mosam> NM

om. om. 05. so. on. 04. cm. on. OH. oo.
H H .H

N N .N

.m

I

m m m .n
.o

L

Q

a

 

 

 

mHADmmm onmmmmomm mzoozH HMZImmA<m
UHm<m mmH mom Ammaq<> va ZOHH<szmmHmQ mo mHzmHUHmmmoo mo onHDmHMHmHQ

H MMDUHM

may:
we Hmnaaz

93
applied to R2 throughout the remainder Of the analysis.

Since the net income Of some Of the firms examined was
actually negatively correlated with sales, Figure l is
not completely accurate and is restated to reflect the
negative correlations in Figure 2 on page 94.

Based on the predetermined definitions of PM and NM
firms, the results indicate that 19 Of the 41 firms were
in the PM classification, and 14 Of the 41 firms were in
the NM classification. Of the 8 firms which were not
classified, two were in the .7l-.80 range with R2 values
Of less than .75.

It was expected that some Of the firms would have
sales-net income relationships which were weakly associ-
ated. However, it was totally unexpected that so many Of
the firms in the PM group (8 of the 19) would have sales
and income trends which were negatively correlated. While
the negative correlations are fairly weak, they do indi—
cate instances where increasing (decreasing) sales trends
may be accompanied by decreasing (increasing) net income
trends.

The assumption that PM firms would possess variables
that provided significant improvement to the basic sales-
net income relationship more frequently than the NM firms

proved to be incorrect. Of the 347 variables examined in

94

mmsHm> NM

00. om. 0e. om. om. OH. 00. oa.l o~.n om.l 0¢.I
P

on.1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mHADmmm onmmmmomm mzoozH HNZImmq<m
UHm<m mom .ZOHH<AMMMOU mo HzmHonmmoo ho onm :HHz
.AmmDA<> va ZOHH<szmmHmQ mo mHZMHUHhhmOo mo ZOHHDmHMHmHQ

N mmDUHh

1m

we

we

 

rm

mauwm
mo umnasz

95
all 41 firms, 164 were found to provide significant im-
provement. Every firm except one (Baxter Labs) had at
least one variable examined which was capable Of signifi-
cantly reducing the residual error left by the basic
sales-net income regression. As a result, it was Obviously
not possible to discriminate between PM and NM firms based
on the presence Of a single significant variable.

The next test performed attempted to determine if
there were differences in the number Of significant vari-
ables found in the two groups. The 19 firms classified as
PM firms had a total of 162 variables examined, Of which
82 were significant. The 14 firms classified as NM firms
had a total Of 115 variables examined, of which 50 were
significant. A two by two contingency table of these

results is presented below.

TABLE 5-1

NUMBER OF VARIABLES FOUND TO PROVIDE SIGNIFICANT IMPROVE-
MENT TO THE BASIC SALES-NET INCOME RELATIONSHIP

 

 

 

Number of Variables at PM Firms NM Firms Totals
an a = .05 level which were (R2<.50) (R2>.75)

Significant 82 50 132
Non Significant 80 65 145

 

Totals . 162 115 77

96

A chi-square test was used to examine the null hy-
pothesis: The frequency with which variables are added is
not significantly different for PM firms than for NM firms.
The critical chi-square value was 3.84 for all two by two
contingency tables at an a = .05 level. If the calculated
value Of chi-square exceeded 3.84 in any case, the null
hypothesis was rejected.

The calculated value of chi-square for the results
tabulated above was 1.372, hence it was not possible to
reject the null hypothesis. Table 5-2 provides a list of
the 41 firms, their Compustat industry classifications,
the R2 generated from each firm's sales and net income
data alone, the rank of the R2 from the highest (41) to
the lowest (1), the number of variables examined in each
case, and the number of variables which significantly im-
proved the basic R2 at an a = .05 level.

In answering the first research question, the results
indicated that it was not possible to discriminate between
PM and NM firms based on the existence, or the frequency
of existence, of variables which significantly improve the
basic sales-net income relationship. While the results
were not expected, it was found, upon analysis of the
mechanics of the regression routine, that the selected

method of analysis (stepwise regression) did not provide

97

 

o n mm wmum. .mspsH a
.cmulmuwsunumz mosses uumsmum
N m NN mono. mHmHumumz .mpHm .prm wsHsuooumemso
m n H waq¢.A|V nocfiz meum mch .mcoo msoummmx
q a a maoo.AuV : Hmmum .m.=
m a a some. : Hmwum UHHnsamm
a m mm quo. : Hmmum HmcoHumz
a w m coco.H-O = aHHzmsmH new mmaow
m a HH memo. : Hmmum pamHsH
N o «N Hmoo. sonmzuHmwum Hmwum oosu<
H m cm SsHm. .ummaoo .meucHuHHo .s.z .HHo sumaamum
N m N mmNH.A|v : manuso mcmfimm
m H mm mmmw. mUHumSmoO muumHHHo
o m an mHaa. .amom a .emznmmsun .man Hmuxmm
m m mm Humm. : nmsoam wsfiumeom
H a He «Nam. HmUngmummsue .osH umuHHm
a m NH mNmH. : mpfinumo GOHGD
N m «a whoa. : ciao
a m cm omnm. : Ousmmcoz
o w mm mmom. : mmasoumm
m m mm memo. : mumuu .m .3
e a ma Nana. : uaoasa .H .m
a m an wmwm. : HmUHsmno eon
a m mm mum“. : pasmc%o smuauma<
m m mH mama. HemszmHmuHamno mHmOHsmno pmHHH<
e m mm Hmnn. ocHN a pawn mamumcaz com .um
a a an ceom.. .omfizlmamuuz .mmm a uamsm smufiumsg
usmuawwmmww pmcfismxw NM mo m esousH mamz huumsch osmz.wamsou
mums scans moanmwum> scam wanmHmm
mmenHum> umnssz
mo Hoessz

 

Nlm mqm<a

98

 

 

«0H mam mamuoa
.mll .mnu os osma. .oo maoaamHmH .H a .H.<
s a o mmoo.x-v : .<.3.e
a HH m mamo.Anv : mmsHHuH< chHuma< sum
0 o om «HNN. : mmaHHuH< ucmHuo ummssuuoz
n m NH NONN. : mmsHHuH< HmsoHumz
n m am Noam. : mmcHHuH< muHmn
N a a NHoo.H-V unoamamse HH< mocHHuH< chHuma<
e n mm «one. Hmoauao :qumm umsam
s m m sooo.xue mumamoum< ummBIUOH
m m ma ommm. .004 a muumm ous< mmhmm ammamx
0 OH Hm NeHo. : muouoz Hmumcmu
o a SH SSON. moHUHnm> none: umHmaunO
a a on mom“. : mason zmsufim
m OH OH wamo. : umuwwmmm ammo choaumz
o a wH MOHm. .aHsam .msm a mUHHHo newswommmueua
usmufiwasmwm pwcfismxm m mo mm mEOOcH msmz mmuwach mamz hmwmsoo
mums BOHcs mmapmfium> xcmm upZImmHmm
mmapmwum> umnssz
mo uwnssz

 

 

poacfiucoollmlm mqm<H

99
the expected results because it did not evaluate the vari—
ablesin.the manner anticipated. It was believed that the
regression analysis would indicate a variable was signifi-
cant only when the variable was poorly correlated with the
independent variable (sales) and strongly correlated with
the dependent variable (net income). If a variable were
highly correlated with sales, it was assumed that it would
not be added as a significant variable since the capacity
of the variable to explain the trend Of net income would
have already been accounted for by the fixed sales vari-
able. The assumption was incorrect.

Since the sales variables of the PM firms were poorly
associated with net income (by definition Of PM), the re-
gression routine, in seeking variables which were associ-
ated with the trend in net income, did select variables
which were weakly associated with sales as was anticipated.

On the other hand, when a NM firm had a variable
which was strongly correlated with net income, it was
always strongly correlated with sales as well. This
event was not totally unexpected since it was assumed
p 25125; that variables which were strongly correlated
with net income and weakly correlated with sales would be
variables which had the capacity to manipulate income.

If the classification system were apprOpriate, NM firms

100

should have had a noticeable absence of such variables.
The fact that they did have a complete absence Of such
variables tended to support the classification system.
The regression characteristic which was unexpected, and
which resulted in so many significant variables in the NM
group, was the degree of significance found in variables
which were highly correlated with both sales and net in-
come. It was assumed that such variables would not sig-
nificantly improve the sales—net income relationship due
to their strong association with sales, as previously
explained. However, this assumption proved to be incorrect.

To illustrate the prOblem a case will be explained.
Elmer Perkin CO. had a very strong relationship between
sales and net income for the test period as indicated by
an R2 Of .9562. Yet the regression routine found four
variables which provided statistically significant improve-
ment! The squared correlations of the four variables with
the sales variable were .94, .96, .98, and .98.

It was originally assumed that such variables would
not provide significant improvement since their ability
to explain the behavior of net income would have already
been accounted for by the sales variable. The fault with
the original assumption was that it did not consider the

overriding influence of a variable's association with the

101
dependent variable, net income. All four variables in the
above case had squared correlations with net income that
exceeded the squared correlation of sales to net income
(.96, .98, .98, and .99).

While the regression analysis did fail to discrimi-
nate between PM and NM firms in the manner anticipated,
the analysis did provide some useful information for the
behavioral characteristics study which is reported below.
It should be noted that the failure Of the regression
analysis to discriminate between PM and NM firms based on
either the number Of cases or the frequency with which
significant variables were indicated was the result of
imprOper assumptions concerning the mechanics of the
analysis technique. The behavioral analysis of the income
statement variables which follows helped to prOperly

qualify the results Obtained from the regression analysis.

The Behavior Of Income Statement Variables
The results of the regression analysis indicated that
there were major differences in the strength of the sales-
net income relationship among the firms examined, even
within a particular industry classification. In attempting
to determine what role the various income statement vari-

ables played in the sales-net income relationship

102.

differences, an analysis of their behavior with respect to
both sales and net income was considered to be apprOpriate.

The relationships which a variable may have with
sales and net income can be generalized into four basic
types of relationships. A variable can be (1) strongly
correlated with both sales and net income, (2) weakly cor-
related with both sales and net income, (3) strongly cor-
related with sales and weakly correlated with net income,
or (4) weakly correlated with sales and strongly corre—
lated with net income. Each behavior pattern is now dis-
cussed with respect tO its manipulative potential, and

its presence in both the PM and NM groups.

The Strong-Strong Case

When a variable was strongly correlated with both
sales and net income (an SS variable), it was felt that
the variable was behaving predictably since the changes
in the variable were consistent with the changes in sales
and net income. There was some reason to believe that SS
variables would not be present with equal frequency in
both PM and NM firms. Due to the definition of NM firms,
it was expected that such variables would exist more fre-
quently in this group. Since sales and net income were

poorly correlated in the PM group, it was less likely that

103
a PM firm would possess a variable which was strongly
correlated with both sales and net income. A variable
which was strongly correlated with both sales and net
income in either PM or NM firms was not considered a

variable which would be capable of manipulating income.

The Weak-Weak Case

 

Variables which were weakly correlated with both
sales and net income (WW variables) were considered to be
capable of reducing the strength of the sales-net income
relationship. In this case, it was considered to be
meaningful to distinguish between variables which were
negatively correlated with sales or net income or both
and variables which were positively correlated with both
sales and net income.

Variables which had weak positive correlations with
sales and net income were behaving in a manner which was
unrelated to the trends in sales and net income. While
they could have been variables which were being used to
manipulate income, they also could have been relatively
fixed revenue or expense variables which were insensitive
to changes in sales and net income. (Note that all vari-
ables which were weakly correlated with sales and net in-

come were not necessarily behaving in a relatively fixed

104
manner.)

Variables which were so weakly associated as to be
negatively correlated with sales or net income or both
were considered to be behaving in an unusual manner. An
expense variable which was negatively correlated with net
income (or sales) would tend to decrease as net income
(sales) increased, and increase as net income (sales)
decreased. Such behavior seems suspect, and becomes more
so as the strength Of the negative correlation increases.
These negatively correlated variables were considered to
be Obvious candidates for distorting the assumed relation-
ship between sales and net income, and were also con-
sidered to be capable Of manipulating income. (It would
seem that the variables which had a strong negative cor-
relation with net income would be the variables which
should be examined for income smoothing since they serve
to reduce changes in income by acting against the direction
of the change.)

Unlike the SS variables, there was no reason to
assume the WW variables would not appear among the vari-
ables of both PM and NM firms with equal frequency. If
the principal reason for variables being weakly associated
with sales and net income was simply that they were rela-

tively fixed in nature, both groups should possess WW

105

variables with approximately the same frequency.

The Strong-Weak Case

 

Variables which were strongly correlated with sales
and weakly correlated with net income (SW variables) were
also capable of reducing the strength of the sales-net
income relationship. However, this behavior pattern was
not considered to represent a capacity to manipulate in-
come, since any variable which was behaving consistently
with the sales trend was considered to be behaving
rationally. Instances where a variable behaved consist—
ently with respect to sales but inconsistently with
respect to net income indicated that the sales-net income
relationship had been weakened through some other variable,
or group of variables, to the point where net income was
no longer associated with sales. In other words, while SW
variables were not considered tO be potential manipulators
of net income, they were considered to be indicative of
instances where income might have been manipulated. If
net income had not somehow become disassociated with sales
the SW variables would have prObably been strongly associ-
ated with both sales and net income.

If the classification system was not capable of

discriminating between potential manipulators and

106
non-manipulators, there would be no reason to assume that
SW variables would not be present among the variables of
both PM and NM firms. However, if the PM firms contained
SW variables with greater frequency than the NM firms, it
would tend to confirm the ability of the PM—NM classifi-
cation technique tO discriminate between potential manipu-

lators and non-manipulators.

The Weak-Strong Case

Variables which were weakly correlated with sales and
strongly correlated with income (WS variables) had the
capacity tO reduce the strength Of the association between
sales and net income. These variables also had the poten-
tial to manipulate net income, since they helped to explain
the income trend, and at the same time behaved inconsist—
ently with respect to sales.

Here again, there was no reason to assume that WS
variables would not be among the variables of both PM and
NM firms. However, if the classification technique was
capable of meaningfully classifying firms, the frequency
with which WS variables were present among the NM firms

should have been relatively low.

107
Behavioral TestingyResults

In the analysis presented above, the four general
behavior patterns of income statement variables were dis—
cussed. Tests were conducted to determine whether the
behavior patterns of the variables examined in the PM
firms were significantly different than those Of the NM
firms. A chi-square test was used in each instance, at
the .05 level Of significance.

The first generalized hypothesis was: The frequency
with which PM firms contain a single variable with the
specified behavior pattern is no different than that of
the NM firms. Instances where at least one variable Of a
particular behavior was absent in every firm or present in
every firm in either the PM or NM groups would be indicated
testing this hypothesis.

The second generalized hypothesis was: The number Of
variables which reflect the specified behavior pattern is
no different for PM firms than for NM firms. Testing this
hypothesis provided the total frequency of occurrence for
each behavior pattern in both PM and NM firms, and indi-
cated significant differences.

Since the behavior patterns of variables were dis-
cussed in the context of weak and strong correlations, it

was again necessary to define weak and strong. In order

108
to be consistent with the previous definitions, weak cor-
relations were defined as instances where a variable's
squared correlation (R2 with the sign of R) with either
sales or net income was less than .50. Strong correla-
tions were defined as instances where the squared corre-
lation Of a variable with either sales or net income was

greater than .75.

The Strong-Strong_Test

The frequency with which the two groups of firms
possessed at least one variable which was strongly asso-
ciated with both sales and net income is presented in the

following contingency table.

TABLE 5-3

 

Characteristic PM Firms NM Firms Totals

 

at least one variable was strongly cor-

 

 

related with both sales and net income 0 14 14

no variables were strongly correlated

with both sales and net income 19 0 19
Totals 19 14 ‘3;

 

The calculated value of chi-square from the above table
was 32.99 which was greater than the critical value of
3.84, hence the null hypothesis was rejected. While this

test does not lend strong support to the classification

109
system, it does indicate a complete absence of SS variables
in the PM group of firms.
The number Of variables which were strongly associated

with both sales and net income is summarized in Table 5-4.

TABLE 5-4

 

Characteristic PM Firms NM Firms Totals

 

variables strongly correlated with both
sales and net income 0 62 62

 

variables not strongly correlated with
both sales and net income 162 53 215

 

Totals 162 115 277

 

The calculated value of chi-square was 112.52, hence the
null hypothesis was again rejected. Over one-half of all
the variables (62 Of the 115) of the NM firms were strongly

associated with both sales and net income.

The Weak-Weak Test

 

The frequency with which the two groups of firms had
at least one variable which was weakly correlated with
both sales and net income is presented in the following

table.

110

TABLE 5-5

 

Characteristic PM Firms NM Firms Totals

 

at least one variable was weakly cor—
related with both sales and net income 19 8 27

 

no variables were weakly correlated
with both sales and net income 0 6

0‘

 

Totals 19 14 3

 

The calculated value of chi-square from the results tabu-
lated above was 7.79. As a result, the hypothesis that
both groups had at least one variable which was weakly
correlated with both sales and income with equal frequency
was rejected. All of the PM firms had at least one vari-
able which was weakly correlated with both sales and income,
while six of the NM firms had no such variable.

The number of variables which had the weak-weak

characteristic is indicated in the following table.

TABLE 5-6

 

Characteristic PM Firms NM Firms Totals

 

variable weakly correlated with sales
and net income 57 14 71

 

variable not weakly correlated with
both sales and net income 105 101 206

 

Totals 162 115 277

111

The calculated value of chi-square from the data
tabulated above was 18.69 which was considerably more sig-
nificant than the first test above. While 8 of the 14 NM
firms had at least one variable which was weakly corre-
lated with both sales and income, only 14 of the 115 vari—
ables (12%) of NM firms possessed the weak-weak behavior.
The 57 variables of the PM firms which possessed the weak-
weak behavior represented 35% of the total 162 variables
examined.

While the above distribution reflects highly signifi-
cant differences in the frequency with which the weak-
weak behavior characteristic appeared (d<.OOl), it was
concluded in the initial discussion of the weak-weak case
that there was no reason to expect an unequal distribution
of such variables. It was also suggested that such vari-
ables have the potential to manipulate income. As a
result, if any inferences were to be made from the above
tests, they would tend to support the PM - NM classifica-
tion system. A closer examination of weakly correlated

variables will be made later in this chapter.

The Strong-Weak Test

 

The frequency with which the two groups possessed at

least one variable which was strongly associated with

112

sales and weakly associated with net income (SW variables)

is presented in the following contingency table.

TABLE 5-7

 

Characteristic PM Firms NM Firms Totals

 

at least one variable strongly assoc—
iated with sales and weakly associated 18 2 20
with income

 

no variables strongly associated with
sales and weakly associated with income 1 12 13

 

Totals l9 l4 3

 

The calculated value of chi-square was 21.80 which
indicates a highly significant difference (q<.OOl) in the
distribution among the two groups. Only one of the PM
firms did not have at least one SW variable. Upon closer
examination of this firm it was found that none of the
variables examined had a squared correlation of greater
than .67 with either sales or income.

The number of SW variables which were present in the

two groups is summarized in the following table.

 

 

113

TABLE 5-8

Characteristic PM Firms NM Firms Totals

 

variables strongly correlated with
sales and weakly correlated with 65 3 68
net income

 

variables not strongly correlated
with sales and weakly correlated 97 112 209
with net income

 

Totals 162 115

N
\J
\J

The calculated value of chi-square was 51.10 which
was again significant at an a of less than .001. While 2
of the 14 NM firms had at least one of the SW variables,
only 3 of the 115 variables of NM firms were so charac-
terized. As was previously stated, the presence of SW
variables, while not considered to be manipulated variables
themselves, should be an indication of instances where in—
come has been potentially manipulated. Had the frequency
distribution of such variables not been significantly
different for the two groups, the PM-NM classification
system would have been discredited. The results tabulated
above once again tend to confirm the ability of the classi-

fication to discriminate between PM and NM firms.

114

The weak—Strong Test

 

The variables which were weakly correlated with sales
and strongly correlated with income (WS variables) were
the variables which the regression routine was expected
to add as providing significant improvement to the basic
income statement relationship. The frequency with which
a firm possessed at least one WS variable is tabulated

below.

TABLE 5-9

 

Characteristic PM Firms NM Firms Totals

 

at least one variable that was weakly
correlated with sales and strongly ll 0 ll
correlated with net income

 

no variables weakly correlated with
sales and strongly correlated with 8 14 22
net income

 

Totals 19 14 33

 

The calculated chi-square value of 11.22 indicated
the difference between the PM and NM groups was signifi-
cant at an a level of less than .001. None of the NM
firms had a single WS variable.

The frequency with which WS variables occurred is

reported in the following table.

 

115

TABLE 5—10

 

Characteristic PM Firms NM Firms Totals

 

variables weakly correlated with
sales and strongly correlated with 12 0 12
net income

 

variables not weakly correlated
with sales and strongly correlated 150 115 265
with net income

 

Totals 162 115 27

 

The calculated chi—square value of 7.95, while being the
lowest calculated value, indicated a difference between
the two groups which was significant at an a level of
less than .005.

The WS variables were expected to be variables most
likely to represent manipulation. In examining the 12 WS
variables of the 19 firms it was discovered that each
variable was added to the stepwise regression routine as
was originally anticipated. However, of these 12 WS
variables, 11 represented the "tax expense" variable and
the other represented a "bonuses" account. There was no
reason to believe that either type of account would not
normally be strongly associated with reported net income.
As a result, none of the 12 WS variables was construed to
have manipulated income. The fact that they were weakly

correlated with sales while being highly correlated with

 

116
income simply indicated that if manipulation occurred, it

todk place before the two expense items were calculated.

Summary of the Behavioral Test Results and
Some Further Testing

 

Each of the eight tests reported indicated that the
differences in the behavior patterns of the income state-
ment variables examined for PM and NM firms were statisti-
cally significant. These results tend to confirm the
assumption that the differences in the strength of the
basic sales-net income relationships, which was the basis
for classifying firms, were at least partially attributable
to the differences in the behavior of the income statement
variables. The answer to the second research question is
clearly affirmative. Yet, it remains unclear as to how
these behavioral differences result in potential manipula-
tion for PM firms.

Recall from the previous behavioral analysis discus-
sion that SS variables and SW variables were ruled out as
being representative of income manipulating behavior due
to their strong associations with the trend in sales. The
WS variables and the WW variables were considered to be
candidates for manipulating income due to the inconsist-
ency of their behavior with respect to the sales trend.

Since the 12 WS variables were found to be

 

 

117
non-manipulative in nature, the existence of potential
income manipulation can not be exemplified by the analysis
of these variables. This fact should not be construed to
mean that WS variables do not have the potential ability
to manipulate income, or that WS variables which may have
served to manipulate income did not exist in the sample
of firms examined. The disclosure of such variables
could have easily been avoided by combining their amounts
with other variables and reporting them under some aggre-
gated account classification.

Aggregating accounts which have dissimilar behavior
patterns will distort the behavior patterns of the indi—
vidual accounts. The analysis of aggregated accounts in
such cases is of limited benefit. Exactly how accounts
should be aggregated and/or reported has, of itself, been
a controversial issue. The 1966 American Accounting Asso—
ciation Committee to Prepare a Statement of Basic Account-
ing Theory suggested an aggregation format based on whether
the amounts in income statement accounts were either fixed
or variable in nature. Such an aggregation technique
would have been extremely helpful for this research effort,
since it would have provided a benchmark for comparing
actual behavior patterns to stipulated behavior patterns.

The only disclosed variables examined in the study

 

118

which might have exemplified manipulative capabilities
were the WW variables. Some of the WW variables could
have represented accounts which were relatively fixed in
nature, which would account for their lack of sensitivity
to changes in sales and net income. However, when the WW
variables were negatively correlated with sales or income P1
or both, their changes with respect to trended sales or

net income or both were in the Opposite direction. Such

 

behavior was considered to be indicative of a capacity to 1‘;
manipulate net income.
To determine if the PM and NM firms had significant
differences with respect to the presence of negatively
correlated variables, a more restrictive version of the
weak-weak test was examined. The results of the first

test are presented below.

TABLE 5—11

 

Characteristic PM Firms NM Firms Totals

 

at least one variable was negatively
correlated with either sales or 19 4 23
net income

 

no variables were negatively correlated
with either sales or net income 0 10 10

 

Totals 19 14 2;

 

119

Since it was assumed that no inherent differences in
the presence of negatively correlated variables should re—
sult from the PM—NM classifications, the null hypothesis
was: The presence of at least one negatively correlated
variable is no different for PM firms than for NM firms.
From the data tabulated above the calculated value of chi—
square was 19.49 which was significantly greater than the
critical value of chi-square (at a - .992 the critical
value of chi—square is 7.879), hence the null hypothesis
was rejected.

As in all of the behavioral pattern tests previously
reported, it was of interest to note when any differences
were found based on the presence of a single variable of
a specified characteristic. In this case, it was dis-
covered that none of the PM firms in the sample had a com-
plete absence of negatively correlated variables.

An additional test was made to determine the number
of negatively correlated variables each group was Observed
to have had, and if the frequency distribution of such
variables was significantly different between groups.

The results of this test are tabulated below.

 

120

TABLE 5-12

 

Characteristic PM Firms NM Firms Totals

 

number of variables negatively corre-
lated with either sales or net income 91 4 95

 

number of variables positively corre-

 

lated with both sales and net income 71 111 182
Totals 162 115 277

 

The calculated value of chi—square (82.87) indicated that
there were highly significant differences.

The 4 NM firms which had negatively correlated vari-
ables each had only one such variable. Of these 4 vari-
ables, 3 were negatively correlated with both sales and net
income and one was negatively correlated with sales only.
The average of the 3 squared correlations with net income
was (-).096, and the average of the 4 squared correlations
with sales was (-).075.

Of the 91 variables which were negatively correlated
with either sales or net income for PM firms, 67 were
negatively correlated with net income, 33 were negatively
correlated with sales, and 9 were negatively correlated
with both. The average squared correlation of the 67
variables which were negatively correlated with net income
was (-).1795, and the average squared correlation of the

33 variables negatively correlated with sales was (-).l632.

 

121
Figures 3 and 4 indicate the frequency distribution of the
negatively correlated variables of the PM firms for net
income and sales respectively.

Since negatively correlated variables have the
capacity to manipulate income, it was possible to examine
these variables and find an example which represented
potential income manipulation. As mentioned in the analy—
sis of the 12 WS variables, it was not considered unusual
that the 11 tax expense variables were strongly correlated
with net income. This relationship was, in fact, the ex—
pected behavior of all tax expense variables. However 4
of the 19 PM firms had tax expense variables which were
weakly correlated with net income, and one of the 4 was
negatively correlated with net income (R2 = -.33).

Figure 5 indicates the behavior of the tax expense vari-
able which was negatively correlated with net income.

It was not the intention of this researcher to imply
that the tax expense variable illustrated in Figure 5, or
any other variable, represented an actual case where income
was manipulated. However, it is evident that the tax ex-
pense variable illustrated, in and of itself, has the
capacity to manipulate income. The tax expense as a per-
cent of reported net income (column 4) ranged from 58% to

103%, with year-to-year variations in the percent as large

 

 

 

122

.MIADI
m”. . 1.\Iu. V It...

5er-

 

mmsam> mm

05.: 00.: om.n oq.l om.l 0N.I OH.I oo.l

 

 

.OH

 

ma .mH

maham
.om mo umnanz

.mN

won

 

 

 

an mm

m so onm meHz um
.mzoozH emz meHz ame<qmmmoo wqm>HH<omz
meme mons mzmHm 28 so mmamaHm<> so onesmHmemHn wozmaommm

m mmauHm

1123

00.1

 

mosam> mm

0m.| 00.: 0m.l 0a.! om.l 0N.I 0a.: 00.:

 

 

 

 

 

roa

maufim
mo umnasz

rmH

 

 

0H

 

0m

x mo zuHm mHHS NM .mm4<m mHHz QmH<qmmmoo w4m>HH<Omz
mmm3 monz mzmHm Em mo mmqm<Hm<> mo onHDmHmHmHQ VUZMDONMm

q MMDUHW

124

 

 

FIGURE 5
(l) (2) (3) (4) (5) (6)
Z change in 2 change in
Reported Reported Column reported reported
Net Tax 3 as a net income tax expense
Income Expense Z of from previ- from previ-
Year (millions) (million) Column 2 ous_year ous year
1957 $396.61 $280.33 71%
1958 341.25 197.77 58 - 13.96% - 29.45%
1959 418.70 308.87 74 + 22.70 + 56.18
1960 381.40 250.32 66 — 8.91 - 18.96
1961 418.16 264.93 63 + 9.64 + 5.84
1962 451.60 333.46 74 + 8.00 + 25.87
1963 472.26 349.55 74 + 4.57 + 4.83
1964 471.43 359.35 76 - .18 + 2.80
1965 407.23 357.66 88 — 13.62 - .47
1966 389.12 334.25 86 — 4.45 - 6.55
1967 313.86 253.42 81 - 19.34 - 24.18
1968 371.87 382.16 103 + 18.48 + 50.80
1969 356.20 343.30 96 — 4.21 - 10.17
1970 328.70 250.40 76 - 7.72 - 27.06
1971 356.50 265.00 74 + 8.46 + 5.83

 

La.

 

1' I; .jth-VL'

—4 11" “‘u‘Lf'. A: a . .I
y

125
as 22% (1967 to 1968).

Although the tax expense variable was selected be-
cause its behavior was different than that which was
normally expected of such a variable (the average squared
correlation with net income of all 41 tax expense vari-
ables examined was +.735), it also provided an illustra- P}
tion of how an income statement variable which would .
normally be disclosed in the published financial reports

could have had a manipulative effect on net income. j

 

An abbreviated analysis, similar to that which was
illustrated in Figure 5, was conducted on the three
remaining WW tax expense variables of the PM firms to
provide further evidence of the apprOpriateness of the
interpretation of WW variables in general. The results
of the analysis are illustrated in Figure 6.

The analysis of negatively correlated variables pro-
vided further evidence of the ability of the classifica-
tion system to discriminate between PM and NM firms. The
behavior of a negatively correlated variable and three
other WW variables provided illustrations of how these
variables may have a manipulative effect on reported net

income.

126

 

 

FIGURE 6
TAX EXPENSE AS A PERCENT OF REPORTED
NET INCOME
Year* Case 1 Case 2 Case 3
l 80.64% 37.00% 81.47%
2 51.91 32.47 81.45
3 72.08 31.07 78.12
4 70.96 36.85 62.34
5 63.92 50.11 52.35
6 74.13 55.97 66.51
7 78.68 55.43 64.43
8 85.04 64.35 70.77
9 89.74 77.34 80.32
10 83.39 84.60 86.59
11 87.16 104.58 85.46
12 94.77 101.77 82.65
13 88.02 99.60 95.28
14 89.28 82.41 93.64
15 103.58 80.23 95.27

 

* year 1 is most recent year for which data was
collected in each case.

NOTE: Case 2 and 3 above, and the case illustrated
in Figure 5 are all from the Chemicals-Major
Industry classification.

—I

 

127
Summary

This chapter reported the results of the research and
an analysis of the results. The stepwise regression rou—
tine was used to test the first research question. The
results of the regression testing indicated that the vari-
ables examined in both PM and NM firms were capable of
improving the basic income statement relationship.

An analysis of the behavior of income statement vari-
ables with respect to sales and net income trends utilized
the intercorrelation matrix to test the second research
question. The behavior analysis indicated which behavioral
patterns were capable of manipulating income and which
were not considered to have income manipulating capabili-
ties. The tests concerning the frequency with which PM
and NM firms contained variables of each general behavior
pattern indicated that the frequency distributions for
each behavior pattern were significantly different for
PM and NM firms. These results indicated that the answer
to the second research question, which asked if PM and NM
firms had variables which had significantly different be-
havioral characteristics, was affirmative.

Further analysis considered the nature of the differ-
ences in the behavior of variables, and indicated that

the PM firms possessed a significantly greater number

 

128

of variables which were considered to be capable of

manipulating income.

q

\l-I". I,

 

 

CHAPTER VI
SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS

The first section of Chapter VI contains a summary of
the results presented and discussed in the previous chap-
ters. A brief review of the conditions which led to the
research tepic is presented, followed by the two research
questions considered and a summary of the results which
were Obtained in answering the two questions. The conclu-
sions drawn from the research results reported are then

presented, followed by some recommendations.

Summary of Results

In the literature of accounting and finance it has
been suggested that some firms manipulate their reported
net incomes. The notion that income is sometimes manipu-
lated represents a substantial criticism of the usefulness
to external users of the income statement which is gen-
erally considered to be the most widely used segment of
published accounting information. A good many critics

have sighted the flexibility and permissiveness of

129

1‘ 1.55.53-1- .’

 

II)! .1 ‘-‘-l~ . .
i.

\. .. . n
' .' n’ '-
’1!

130
generally accepted accounting principles as the major
vehicle for facilitating a firm's ability to manipulate
its reported net income. Others have expressed the
belief that manipulative tools are not necessarily con-
fined to accounting principles. In either case, the bulk

of the criticism has been aimed at the accounting pro—

"4.4.1:”. ‘
e.
-.

fession. ”7

There have been a number of researchers who have

A

attempted to develOp techniques whereby instances of

 

manipulation might be determined. If general agreement
could be reached on exactly what recognizable conditions
constitute income manipulation, the implied desire of
some firms to manipulate their reported net income might
be substantially deterred. However, none of the research
endeavors thus far has been capable of providing a gen-
erally agreed upon definition of manipulation or a tech-
nique for determining when income has been manipulated.
The purpose of this research was to test the ability
of the classification technique herein proposed to dif-
ferentiate between potential manipulators of reported net
income and non-manipulators. The classification of firms
was based on their linear relationships between reported
sales and reported net income over a 13 to 16 year period.

Firms whose reported net income data were strongly

131
associated with their respective net sales data were con-
sidered to be non-manipulator firms. The NM classifica-
tion was based on the assumption that there should be a
strong functional relationship between net income data and
sales data for those firms which have not manipulated re-
ported net income.

Conversly, firms whose reported net income data were
weakly associated with their respective net sales data
were considered to be potential income manipulators. The
assumed relationship between sales and net income in these
instances had somehow become distorted, and firms which
had manipulated net income would, therefore, be among
those firms with weakly associated sales-net income data.

To determine if the prOposed classification technique
was apprOpriate it was necessary to examine the strength
of the sales-net income relationships of a number of firms
for differences. Once differences in the sales-net income
relationships had been established, the behavior of each
firm's individual income statement variables was examined
to determine if it was responsible for the differences.

In determining the effect of the various income statement
variables, two research questions were considered.

The first research question asked if the income

statement variables of the PM and NM firms had

132
significantly different effects on their respective sales-
net income relationships. In answering this question, a
stepwise regression analysis was used to evaluate the im-
pact of each individual variable on the residuals from the
sales—net income relationship of the firm to which it
pertained.

An evaluation of the relative improvement to the
basic sales-net income relationships of the 33 classified
firms provided by their respective income statement vari—
ables was made using the stepwise regression analysis.

All but one of the 33 classified firms had at least one
variable which was capable of significantly improving the
strength of their basic sales-net income relationships.
Of the 162 variables belonging to the 19 firms in the PM
group, 82 were found to provide significant improvement
(at the .05 level of significance). Of the 115 variables
belonging to the 15 firms in the NM group, 50 were found
to provide significant improvement.

Based on these regression results, it was not possible
to discriminate between PM and NM firms on the basis of
either the number of instances or the number of variables
which indicated that improvement in the basic relationship
had taken place. Both groups of firms had variables which

improved their relationships, and the frequency with which

133
such variables were present did not indicate that a sig—
nificant difference existed between the two groups.

The answer to the first research question was that
the income statement variables of the PM and NM firms did
not have significantly different effects on their respec-
tive sales-net income relationships.

The second research question asked if the income
statement variables of the PM and NM firms had signif-
icantly different behavioral characteristics. In answer-
ing this question the data from an intercorrelation matrix
were used. The behavior pattern of each variable was
characterized by its correlation with sales as well as
its correlation with net income.

Income statement variables were considered, defined
and classified into four general behavior patterns and
evaluated with respect to their income manipulating capa-
bilities. Variables whose behavior patterns represented
either strong correlations with both sales and net income,
or strong correlations with sales and weak correlations
with net income were considered to be incapable of manipu-
lating income. Variables whose behavior patterns repre-
sented either weak correlations with both sales and net
income, or weak correlations with sales and strong corre-

lations with net income were considered to be capable of

134
manipulating income.

The frequencies with which PM and NM firms possessed
variables with the four general behavior patterns were
tested to determine if there were significant differences.
The tests revealed that statistically significant differ-
ences (at an a of less than .005) occurred in each of the
four behavior patterns. PM firms were found to have a
complete absence of income statement variables which were
strongly correlated with both sales and net income. Also,
PM firms were characterized as having a significantly
greater number of variables which were: (1) weakly corre—
lated with both sales and net income; (2) weakly corre-
lated with sales and strongly correlated with net income;
(3) strongly correlated with sales and weakly correlated
with net income.

The answer to the second research question was that
the variables of the PM and NM firms did have significantly
different behavioral characteristics. The differences in
the behavior of the variables of the two groups tended to
lend support to the notion that firms classified as PM's
were potential income manipulators and those classified
as NM's were non-manipulators.

An additional test was conducted to determine if there

were differences in the frequency with which negatively

135

correlated variables occurred among the two groups of firms.
Negatively correlated variables were considered to be
strong candidates for manipulating income. The results
indicated that PM firms had a significantly greater number
of variables which were negatively correlated with sales or
net income or both than NM firms, providing additional sup-
port for the PM-NM classifications.

Finally, some illustrations of variables which were
said to be capable of manipulating income were presented
to reveal how they might have been used to influence the

determination of reported net income.

Conclusions

 

The implications of the research results are largely
a function of how they are interpreted by the reader. Two
direct conclusions may, however, be drawn.

First, it was revealed that major differences do
exist in the strength of firms sales-net income relation—
ships, even among firms in the same industry classifica-
tion. The differences were so great in some instances
that they could have been considered real (statistically
significant) differences for all periods rather than for
the years tested alone. This was again true for some

firms in the same industry classification.

 

. i.) . .

11-1

 

136

Second, it was concluded that the income statement
variables examined in the research did play an important
part in generating the existing differences in the sales—
net income relationships of the sample firms for the period
tested. Furthermore, firms classified as potential income
manipulators were found to have a significantly greater
number of variables which behaved unpredictably and incon-
sistently with regard to their respective sales and net
income data. Such behavior was considered to be typical
of the behavior which would characterize manipulated income
statement variables.

Based on the aforementioned conclusions, the prOposed
classification technique was generally considered to be
capable of discriminating between potential income manipu-
lators and non-manipulators. The classification technique,
being dichotomous in nature, is not as refined as it might
be, and the definitions of "weak" and "strong" are not
beyond reproach. However, the research results seem to
strongly infer that the basic concept underlying the
classification technique is not unfounded, and might well
provide a sound foundation for future income manipulation

research.

 

137

Recommendations

 

Based on the results reported in this dissertation,
it is strongly recommended that the information generated
by the classification technique used in this research be
considered as a part of the information set for investment
decision making. Firms with weakly associated sales-net
income data should generally be considered less desirable
investments than firms with strongly associated data, all
else being equal. Whether or not the inconsistent behavior
of income statement variables, which characterizes PM
firms, is by design should not be of major import to the
investor. It is sufficient to realize that the income
statement variables and resulting net income generally
behave in a less predictable manner for PM firms than for
NM firms.

Knowledge of a firm's sales—net income relationship
based on past data may be extremely useful when used in
conjunction with other information. If a firm is expected
to have an increase in sales activity in the coming period
it is often assumed that net income will also increase.
The research results clearly indicate that for PM firms
this would not have always been an appropriate assumption.
Yet, based on past experience, there is a high probability

that the assumption would be apprOpriate for NM firms. If

 

138
the past relationship between a firm's sales and income
data can be expected to continue in the future, then,
given an expected sales value, it would be possible to
make a better estimate of expected net income for NM firms
than for PM firms.

Another recommendation, based primarily on the Opinion
of the author, is that the Securities Exchange Commission
(SEC) or the American Institute of Certified Public
Accountants or both should take some positive action to
improve the existing income statement disclosure require-
ments and reporting format. The solutions to many of
the questions and prOblems which burden the accounting
profession today will require extensive empirical research
by its members. This research endeavor, as well as many
others, has been constrained by both the quantity and
quality of the existing income statement data. To relieve
these constraints it will be necessary to develOp a more
meaningful disclosure format.

One possible format might provide for the aggregation
.Of accounts on the basis of their association with sales.
For example, variables which had a correlation with sales
over the past 5 or 10 years (including the current year)
of between .75 and 1.00 could be combined and reported

together, those having a correlation with sales of between

 

139
.50 to .74 could be combined and reported together, and
so on. It would also be possible to further sub—divide

these aggregated accounts according to their particular

functions, such as cost of goods, selling, administrative,

and so on. The resulting aggregations would amplify the
behavior of groups of accounts with respect to sales and
all other groups of accounts, and would prohibit aggrega—
tions of accounts whose behavior patterns tend to offset
one another.

While this suggestion may be too drastically differ-
ent to be considered feasible, there are other possibili-
ties which might provide meaningful improvements to the
present reporting format. The income statement format
recommended by the American Accounting Association's Com—
mittee to Prepare A Statement of Basic Accounting Theory
(ASOBAT), by differentiating between variable and fixed
income statement components, would provide new and useful
information, particularly for research concerned with the
behavior of income statement variables. As a case in
point, if the accounting profession expects to have a

meaningful role in meeting the demand for forecasted

earnings statements, it will be necessary to gain a better

understanding of the behavior of the elements which are

used to determine earnings. The ASOBAT format would also

 

 

140
be of considerable benefit to research which considers
the analysis of past results and trends in income state-
ment data, as this research has.

While the format of the income statement is thought
to be in need of improvement, it is also believed that it
should generally disclose more information. At a minimum,
the maximum amount of actual income statement data (exclu—
sive of exhibits, etc.) which is reported by a firm to any
source (ie. the SEC, the CAB, etc.), should be the same
amount of information reported to all other sources. This
requirement would make the information that is presently
available equally accessable to all users and potential
users of those data, with no additional cost or effort
required on the users part.

To improve the uniformity of income statement disclo-
sure, it could be required that all firms in any given
industry report essentially the same information. For
example, if research and develOpment is considered to be
a significant expense in the chemical industry or the
aerospace industry, all firms in those industries should
be required to disclose the amount expended. In other
industries where research and develOpment expenditures
are not material in amount, disclosure would not be

necessary. Some significance limits could be stipulated,

 

141
possibly based on a variable's percent of net sales, to
help determine which accounts should be disclosed and
which could be aggregated.
The possible improvements which could be made to
existing income statement format and disclosure require-
ments are too numerous to be considered here. It is

sufficient to say that improved requirements in these

areas would facilitate more meaningful accounting research,

and that the research is a necessary element for effec-
tively solving many of the controversial issues in

accounting.

 

 

APPENDIX I

Regression Association

 

The fundamentals of the regression analysis used in
this research are explained in this appendix. The be-
havior of reported net income and its association with
various reported elements of the income statement were of
principal interest. Reported net income (Y) was, there-
fore, the dependent variable used throughout. To explain
the behavior of reported net income, its association with
various income statement variables (the independent Xj
variables) was examined. The reported sales variable was
considered to be the principal independent variable (X1),
and was a fixed element in the regression analysis in that
it was always considered as an independent variable in the
regression equation. The regression analysis evaluated
the ability of the individual independent variables of
each firm to explain the trend in the dependent variable
(reported net income) for the period examined. The
mechanics of the regression association and the definitions

of the statistics generated are provided below.

142

143

Let X-- = the j independent variables,
Y = the mean value of the dependent variable,

YC = the estimated values of the dependent
variable computed by the regression
equation based on the actual values of
the independent variables (in simple
linear regression Yc would be the least
squares line),

Yi = the actual values of the dependent vari-
able,
i = l,....,n = the number of observations

(years) for both Xi and Yi'

 

 

 

 

Then
2 (Y1 " Y)2 2 . -
. = 0 = the variance of the Y. from Y
1=1 n Y 1
= the total variance,
2 (Y1 - YC-)2
i=1 = 02 = the variance of Y- from the
n Y . 1
C Yc- estimates,
1
= the unexplained variance,
n (Yc- - 3.02
2 1 _ 2 _ . _
i=1 — o 2 — the variance of Y from Y,
n Y-X Ci
given the Xi's
= the explained variance.
Since 0 = 02 + 02 2
Y Ye Y-X
2 2
we can say 0
.1 = OYC + OY-XZ
02 "-2— 2
Y °Y OY
: 1 = 2 2
- °YC + °Y-x2
2 O2
CY Y
2 2

III
l-"
ll
X
+
w

144

where K2 . .
the unexplained variances

2
0

Yb _

02 — the total variances
Y

 

x the amount (ratio) of squared
error unable to be reduced.

2
Y-X = the explained error
the total error

 

= the amount (ratio) of squared
error reduced by using Yc. to
estimate Yi given the 1
values Of Xij.

As most commonly defined:

R2 = the "coefficient of determination", which
provides a relative measure of the strength
of the association between the Xj's and Y,
K2 = the "coefficient of non-determination",

R = the "coefficient of correlation" (undefined)

K = the "coefficient of alienation" which indicates
the relative percent of unsquared error unable
to be reduced by the regression equation,

1-K=A = the "coefficient of association“ the relative
percent of unsquared error able to be reduced
by the regression equation.

SOURCE: Materials prepared and provided by Dr. Richard
Lewis, Professor of Marketing, Michigan State
University.

APPENDIX II

Sensitivity;Analysis

 

To test the effect of the eight firms eliminated from
the study on the research results, a form of a sensitivity
test was performed on the definitions of PM and NM firms.
Without changing the definitions of weak and strong be-
havior patterns, the eight firms were added first to the
PM group and then to the NM group to determine their impact
on the results of the chi-square tests reported in the
research. The following table summarizes the results of

the sensitivity tests.

145

146

 

 

 

 

 

Case X2 Values
(Sales-Net Income) Presence Test Frequency Test
W - W Case .
(l) as reported 7.79 18.69
(2) if PM < .75 6.41 16.19
(3) if NM > .50 6.81 14.77
S - W Case
(1) as reported 21.80 51.10
(2) if PM < .75 27.34 42.60
(3) if NM > .50 . 11.32 44.33
W — S Case
(1) as reported 11.22 7.95
(2) if PM < .75 10.14 7.84
(3) if NM > .50 7.12 6.99
S - S Case
(1) as reported 32.99 112.52
(2) if PM < .75 32.90 136.59
(3) if NM > .50 22.58 70.83

 

Generally speaking, the results of the chi-square
tests are less significant when the eight firms are added
to either the PM or the NM group. This would imply mixed
behavior patterns on the part of the variables of the eight
firms, indicating no strong tendency to act like either the
PM or the NM groups as originally specified. The chi-
square values calculated from the sensitivity tests were,

however, all significant at the .05 level.

SELECTED B IBLIOGRAPHY

Accounting Principles Board, APB Accounting Principles,
New York, New York: American Institute of Certified
Public Accountants.

Archibald, T. Ross, "The Return to Straight-Line Depre-
ciation: An Analysis of a Change in Accounting
Method , Empirical Research in Accounting: Selected
Studies, 1967, 164-80.

Ball, Ray and Brown, Philip, "An Empirical Evaluation of
Accounting Income Numbers", The Journal of Accounting
Research, Vol. VI, No. 2 (Autumn, 1968), 169-78.

 

Chambers, R. J., "Matter of Principle", The Accounting
Review, Vol. XLI, No. 3 (July, 1966), 443-57.

 

Comiskey, Eugene E. and Barefield, Russel M., "The
Smoothing Hypothesis: An Alternative Test", The
AccountingyReview, Vol. XLVII, No. 2 (April, 1972),
291-98.

Committee on Auditing Procedure, Auditing Standards and
Procedure, New York, New York: American Institute
of Certified Public Accountants, 1963.

Committee to Prepare a Statement of Basic Accounting
Theory, A Statement of Basic Accounting Theory,
Evanston, Illinois: American Accounting Associ-
ation, 1966.

COpeland, R. M., "Income Smoothing", Empirical Research
in Accounting: Selected Studies, 1968, 101-116.

. and Licastro, R. L., "A Note on Income
Smoothing", The Accounting Review, Vol. XLIII.
No. 3 (July, 1968), 540-5.

. and Moore, Michael L., "The Financial Bath:
Is It Common?", MSU Business Topics, Vol. 20, No. 4

 

147

148

(Autumn, 1972), 63—7.

Dasher, Paul E. and Malcolm, Rdbert E.,

"A Note on Income
Smoothing in the Chemical Industry", The Journal of
Accounting Research, Vol. VIII, No 2 (Autumn, 1970),
253-9.

DOpuch, Nicholas and Drake, David F.,

"The Effect of
Alternative Accounting Rules for Nonsubsidiary

Investments", Empirical Research in Accounting:
Selected Studies, 1966, 192-219.

Gagnon, Jean-Marie,

"Purchase Versus Pooling of Interests:
The Search for a Predictor", Empirical Research in
Accounting: Selected Studies, 1967, 187-204.

"The Purchase-Pooling Choice: Some Empirical
Evidence", The Journal of Accounting Research, Vol.
IX, No. 1 (Spring, 1971), 52-72.

Jaedicke, Robert K.,

Ijiri, Yuji, and Nielson, Oswald,
Research in Accountigg Measurement, Menasha,
Wisconsin: American Accounting Association, 1966.
McDonald, Charles LeRoy, An Empirical Examination of
Published Predictions of Future Earnings.

Unpub-
lished Ph.D. dissertation, Michigan State University,
1972.

Mendenhall, William, The Design and Analysis of Experi-
ments, Belmont, California: Wadsworth Publishing
Company, Inc., 1968.

O'Connor, M. C. and Hamre, J.

E., "Alternative Methods of
Accounting for Long-Term Nonsubsidiary Intercorpo-

rate Investments in Common Stock", The Accounting
Review, Vol. XLVII, No. 2 (April,

1972), 308-19.
Salmonson,

R. F., Basic Financial AccountingfiTheory,
Belmont, California: Wadsworth Publishing Company,
Inc., 1969.

Simpson, Richard H.,

"An Empirical Study of Possible
Income Manipulation", The Accounting Review, Vol.
XLIV, NO. 4 (October, 1969), 806-17.

149

Spacek, Leonard, What is Profit? Cambridge, England:
The Institute of Chartered Accountants in England
and Wales, 1970.

 

White, Gary E., "Discretionary Accounting Decisions and
Income Normalization", The Journal of Accounting
Research (Autumn, 1970), 260-73.

 

       

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

JIHIHHIIHHIIIHIHullllIlllfllfllMJlflwlLlWlll

3 1293 03