UNIVERSITY OFILLINOIS LIBRARY
AT URBANA-CHAMPAIGNBOOKSTACKS
THE HECKMAN BINDERY, INC.North Manchester, Indiana
JUST FONT SLOT TITLE
H CC 1W
H CC 1W
H CC 1W
H CC 7W
3RaCULTY/ORKINGPAPER
8 19887 NO. 1468-1485
330B385<"CV">no. 1468-1485cop. 2
<IMPRINU. of ILL.LIBRARY
BANA
BINDING COPYPERIODICAL: D CUSTOM STANDARD ECONOMY THESIS
BOOK: D CUSTOM D MUSIC D ECONOMY AUTH 1ST DRUBOR TITLE I.D.
SAMPLE
NO. VOLSTHIS TITLE
LEAD ATTACH
ACCOUNT LIBRARY NEW
ACCOUNT NAME
UNIV OF ILLINOISACCOUNT INTERNAL I.D.
FOIL MATERIAL
2 WHI 488
ISSN.
B019124NOTES
STX4COLLATING
35
ADDITIONAL INSTRUCTIONS
BINDING WHEEL SYS. ID.FREQUENCY
76
1/A
Dept=STX4 Lot=201 Item J NM=[;1CR2ST3CR MARK BY # B4 91]
SEP SHEETS PTS. BD PAPER TAPE STUBS CLOTH EXT
INSERT MAT
PRODUCT TYPE
Hf |QM
I FILLER
SPECIAL PREP. LEAF ATTACH
ACCOUNT LOT NO
ACCOUNT PIECE NO
VOL IHI'i
fLE
COVER SIZEJ J
X
PIECE NOU 17'
47
001247981
BEBRFACULTY WORKINGPAPER NO. 1476
Financial Ratios, Fundamental Firm Characteristics,
and Measurement Models: Issues and Evidence
Thomas J. Frecka
David A. Ziebart
College of Commerce and Business Administration
Bureau of Economic and Business ResearchUniversity of Illinois, Urbana-Champaign
BEBRFACULTY WORKING PAPER NO. 1476
College of Commerce and Business Administration
University of Illinois at Urbana -Champaign
August 1988
Financial Ratios, Fundamental Firm Characteristics, andMeasurement Models: Issues and Evidence
Thomas J. Frecka, Associate ProfessorDepartment of Accountancy
David A. Ziebart, Assistant ProfessorDepartment of Accountancy
Financial support for this project was provided by theDepartment of Accountancy and the Bureau of Economic andBusiness Research of the University of Illinois at Urbana-Champaign. We appreciate comments provided by participants inthe Accountancy Forum at the University of Illinois.
Do not quote. Comments are welcome.
Digitized by the Internet Archive
in 2011 with funding from
University of Illinois Urbana-Champaign
http://www.archive.org/details/financialratiosf1476frec
Financial Ratios, Fundamental Firm Characteristics , andMeasurement Models: Issues and Evidence
Abstract
This study examines the relationships among a firm's
financial ratios and its underlying financial dimensions using a
measurement model approach. Both exploratory and confirmatory
factor analytic techniques are used to link observable financial
ratios to underlying fundamental dimensions. This study
illustrates a causal modeling approach for evaluating the
representativeness of financial ratios as measures of underlying
firm attributes.
KEY WORDS: Financial ratios; Factor analysis; Causal Models;
Measurement Models: Covariance structures.
Financial Ratios, Fundamental Fira Characteristics, andMeasurement Models: Issues and Evidence
I . I ntroduction
The purpose of this study is to examine the relationships among a
firm's financial ratios and that firm's underlying financial dimensions
using a causal measurement model approach. Financial ratios are
empirically linked to underlying fundamental economic dimensions using
both exploratory and confirmatory techniques. Accordingly, this study
will illustrate a causal modeling approach for evaluating the
representativeness of financial ratios as measures of underlying firm
attributes. In addition, the differences between exploratory and
confirmatory methods will be discussed and illustrated.
A causal model portrays the causal links and chains between the
various components of the process researched (Abdel-Khalik and Ajinkya
[1979, pp. 20-23]). It is unique in its effort to develop a structural
network of causal relationships built upon theoretical underpinnings.
In the causal modeling process a model structure is developed and the
estimation solution is constrained by the parameter requirements of the
theoretical structure. For this study, this means that certain
financial ratios are constrained to load on certain underlying
fundamental financial dimensions (such as liquidity, leverage, or
profitability) and not load on others. This portrays a measurement
structure in which the observed financial ratios are measures of the
unobservable fundamental dimensions. These links between the observed
financial ratios and the underlying financial dimension are
hypothesized, estimated, and tested.
The importance or a causal moaeiing approacn in tnis rinanciai
ratio context relates to how we use ratios to understand the economic
attributes of the firm, how such attributes change over time, or how the
attributes compare with those of other firms. An appreciation of the
underlying dimensions that certain financial ratios measure is important
in a variety of decision contexts including the effects of firm
attributes on the risk and return of securities, on the terms of lending
agreements, or on the evaluation of the relative financial health of a
firm. Determination of the ratios which best measure certain financial
dimensions would enhance the usefulness of ratio analysis and provide
insight into the propriety of using alternative ratios to measure the
same underlying firm dimension.
Traditionally, economic attributes such as liquidity, leverage,
profitability, and activity have been used to portray and evaluate the
financial condition and performance of the firm. Financial ratios have
been constructed to measure these attributes and a number of financial
ratios are purported to measure them. In general, these attributes are
not precisely defined, not directly observable, and not directly linked
by deductive theory to the types of decisions contexts mentioned above.
In addition, the degree of measurement error inherent in various
financial ratios as measures of the underlying attributes has not been
previously assessed.
Since it is not possible to directly observe the economic dimension
of the firm, financial ratios are commonly used as the indicators or
measures of the unobservable financial dimensions of interest. Given a
measurement theory perspective, each financial ratio can be viewed as a
linear combination of the underlying economic dimension and an error of
measurement component.
Given the lack of precise definitions for the source variables (the
constructs which represent various financial dimensions of a firm), it
is not surprising that a large number of financial ratios have been
proposed to measure the financial attributes of the firm. However, the
suitability of the various financial ratios as measures of the
underlying economic dimensions of a firm has not been ascertained. The
validity of a financial ratio as a measure of an underlying financial
dimension is dependent upon the ratio's representativeness (the degree
to which the observed ratio is a measure of a particular dimension and
pnot another dimension) and the degree of measurement error.
In describing the process of identifying "useful" ratios as it
developed in the 1930's, Horrigan [1968] states:
"In this approach, a priori analysis and/orempirical evidence were rarely provided to substantiatean author's claim that his particular selection of ratiosrepresented an efficient collection of ratios for analyzingfinancial statements. Rather, the author's group of
selected ratios — and sometimes accompanying absolute andrelative criteria — were promulgated solely on theauthority of his experience in statement analysis." (p. 288)
Later, "usefulness" was evaluated on the basis of a financial ratio's
ability to predict or explain economic phenomena such as bankruptcy or
security returns. However, if one considers that it is the underlying
firm dimension, measured by the financial ratio, which is the attribute
of interest, then the measurement error issue is critical. In
situations in which the financial ratio is significantly imprecise,
(significant measurement error exists) the usefulness of that ratio in
explaining or predicting the event of interest may be largely
diminished. In these situations one encounters the traditional errors-
in-variables problem.
Given the potential for a large number of ratios, the idea that the
underlying constructs are of interest, and the benefits of parsimony,
various data reduction techniques have been employed. Unfortunately,
these reduction techniques have been based on statistical analyses void
of theoretical underpinnings and measurement structure considerations.
The most widely used technique in the data reduction studies is
exploratory factor analysis. Examples include Pinches and Mingo [1973],
Libby [1975], and Gombola and Ketz [1983]. Unfortunately, an
exploratory approach is significantly limited in its ability to
determine the appropriate measurement structure and to evaluate the
propriety of alternative model configurations.
An approach that allows testing of alternative measurement
configurations is confir matory or restricted factor analysis. Its
primary advantage is that it allows one to test specific hypotheses
regarding the measurement structure. In this study, we compare and
contrast the application of exploratory and confirmatory approaches.
The use of an exploratory approach in earlier studies can not be
criticized. It was not until the late 1960's that both computer
hardware and software were reasonably available for conducting
exploratory analyses (Jackson and Borgatta [1981]). The ready
availability of algorithms to estimate restricted or confirmatory factor
analysis models and the inclusion of the confirmatory approach in
standard statistics texts is a rather recent development (Marridi
[1981]). For example, the maximum likelihood estimation procedure used
in this study was introduced by Joreskog in 1973 and only became widely
available in the late 1970's.^ Recently, confirmatory factor analytic
methods and procedures have been added to widely used user-friendly
statistical packages such as SPSS Fornell [1982] refers to these
confirmatory techniques as "a second generation of multivariate
analysis .
"
In this study, we initially conduct an exploratory analysis.
Alternative exploratory measurement models are presented and evaluated.
We subsequently conduct a confirmatory analysis. In our confirmatory
analysis we hypothesize a particular measurement model configuration,
estimate the model parameters, and then test both the individual
parameters and the overall model configuration. Alternative
measurement model configurations are then estimated, tested, and
evaluated. We then compare and contrast the confirmatory results with
the exploratory results.
From the results we answer the following questions:
(1) What are the underlying dimensions of the data; i.e., what are
the appropriate factors, the ratios that measure each factor, and what
is the number of factors that reasonably explains the observed ratios;
(2) Are the underlying factors correlated, or linked in some way;
(3) Are the underlying factors well specified; i.e. , do the ratios
load significantly on the factors they purport to measure;
(4) Which of the ratios that measure a given factor is the "best"
measure; i.e., it has the smallest measurement error; and,
(5) Does the observed covariation among the ratios lend credence to
the hypothesized measurement structure?
The importance of this research relates to what we can learn from
using a formal measurement model to represent the relationships among
financial ratios and the underlying financial dimensions. Unlike
previous studies which have used only exploratory techniques described
as "brute empiricism," our study statistically tests the fit of the
measurement model structures through over-identification and compares
5
the results to those obtained using the traditional exploratory
approach. In general, the results from our confirmatory analysis
provide greater insight concerning the ratios that are needed to capture
the underlying economic dimensions of the firm.
We find that the underlying financial dimensions are correlated.
This result contrasts with most exploratory studies which have assumed
that the underlying factors are orthogonal. Further, while our analysis
leads to the same number of factors using both exploratory and
confirmatory approaches, the implied measurement model structures are
substantially different. Another notable result is that the underlying
size dimension is significantly correlated with the other dimensions.
This is surprising since ratios are espoused as a method to control for
size differences across firms. Our results indicate that the size
dimension is positively correlated with the profitability, cash flow,
activity, and capital structure dimensions and negatively correlated
with the liquidity dimension.
The remainder of the paper is organized as follows. Part II
provides a description of the measurement theory approach that is
inherent in confirmatory factor analysis and a description of the
modeling process. In part III we describe the data and the empirical
results. Part IV is a summary and includes suggestions for further
research.
II . Measurement Models - Techniques and Issues
A measurement model can be depicted as a set of underlying
constructs or factors in which the observable measures or indicators are
a linear function linking the observable indicators to the factors.
This relationship can be conceptualized as:
x = L X + e
where x is the vector of observable measures, X is the vector of
unobservable constructs, L is a matrix of regressors linking x to X, and
e is a matrix of measurement errors. The variables are assumed to be
mean deviated so that there is no intercept term in the expression.
It is important to note that this illustration of a measurement
model, while depicted in a regression framework, is fundamentally
different from a regression. In a normal regression situation, the x
and X variables are both observable and one attempts to estimate the
regressor L. In a measurement model analysis the x are observable but
the X are not observable. This implies that in addition to estimating L
and e one must also estimate the unobservable factors X. In this
study, the x are financial ratios, the X are the underlying
financial dimensions, and the e are the measurement errors of the ratios
as measures of the dimensions.
Two different approaches can be taken regarding the estimation of
the underlying measurement structure linking the observable variables to
the unobservable constructs. The approach used in most previous studies
of financial ratios is to conduct an exploratory factor analysis or
principal components analysis. This traditional approach places
relatively few, if any, restrictions on the measurement structure
configuration and lets the data delineate the implied measurement
model
.
Unfortunately, the application of a truly exploratory approach may
lead to an implied measurement model structure which violates both the
postulate of factorial causation and the postulate of parsimony (Kim and
Mueller, [1978] pp. 43-45). These two postulates are neccessary to
minimize the seemingly unsurmountable uncertainties inherent in the
linkage between the factor (measurement model) structure and the
observed covariance structure of the data. The postulate of factorial
causation may be violated in a pure exploratory study since the model is
developed with little or no reference to substantive a priori knowledge
about the data or the model structure. Violation of the postulate of
parsimony may result in an exploratory analysis when one chooses a more
complex model over a consistent but more simplistic model. This may
occur because the exploratory analysis has failed to consider the latter
model configuration.
The postulate of factorial causation is based on the assumption that
the observed variables are linear combinations of the underlying factors
and that the covariation between the observed variables is the result of
the observed variables being linked to one or more common factors (Kim
and Mueller, [1978] p. 78). In essence, it is the underlying factors
which cause the observed covariation among the variables. The postulate
of parsimony requires that if two models equally explain the data, then
the simpler model is more appropriate (Kim and Mueller, [1978] p. 79).
The failure to consider the postulate of factorial causation can
result in a serious indeterminacy of one covariance structure but
various factor loadings. In a financial ratios context, this problem
results in the inability to specifically measure the linkage between a
particular financial ratio and the dimension it purports to measure. As
an example, one may be able to find alternative measurement structures
in which the sign and magnitude of the coefficient linking the observed
financial ratio to the underlying financial dimension construct is
dependent upon the rotation method employed in the analysis. Choice of
rotation method may be dependent upon a somewhat arbitrarily chosen
8
criterion. This results in an implied measurement configuration which
is the result of the criterion chosen rather than any theoretical
underpinnings
.
The second indeterminacy, related to the postulate of parsimony,
concerns the situation of one covariance structure but alternative
numbers of factors or measurement configurations. Normally in an
exploratory study, the number of factors in the final solution is based
on the magnitudes of the eigenvalues. The general rule of thumb is to
only include as factors the eigenvectors with eigenvalues greater than
or equal to one. The number of factors may be a somewhat arbitrary
choice and the resulting measurement model configuration is dependent
upon the choices made by the researcher either implicitly or explicitly.
In addition to the problems of factorial causation and parsimony,
the issue of an orthogonal versus an oblique solution can seriously
impact representative validity. With an orthogonal solution, the
approach used in most previous studies of financial ratios, the
underlying factors are assumed to be independent of each other. The
factors are allowed to covary and are not assumed to be independent when
an oblique solution is employed.
The application of factor analytic techniques without an
appreciation for the implied measurement configuration can be
problematic. Foster [1986] points out the potential difficulties of the
traditional exploratory approach:
Factor analysis can, if used in an uncritical manner,become brute empiricism in the extreme. However, whenused with recognition of its limitations (for example, thepotential excessive reliance on factors suggested by data)and of the judgement calls necessary in its application(for example, how many separate factors to identify), it
can be a useful addition to the tools used in financialstatement analysis.
The other approach to the empirical investigation of measurement
structures is a confirmatory approach. With this approach, a particular
causal structure is hypothesized and an a priori measurement model
configuration is imposed on the analysis; the number of factors, the
presence of covariation among the factors, and the determination of
which variables load on which factors are specified. A confirmatory
approach imposes a structure of factorial causation on the analysis.
Accordingly, specific hypotheses regarding the factor structure are
introduced into the analysis since parameters are specifically
constrained. The probability that the imposed structure will be
supported by the observed covariance structure is less, if factorial
causation is not in operation. 5 In a confirmatory analysis, the
parsimony issue usually does not arise since the number of factors is
based on theoretical underpinnings.
Figure 1 illustrates the exploratory and confirmatory approaches to
the empirical investigation of causal measurement structures.
INSERT FIGURE 1
Under an exploratory approach, the solution method, including rotation,
orthogonal versus oblique factors, and the method to determine the
number of factors must be considered. In many instances where "canned"
computer programs are used, these choices are constrained by the
program. In other instances, the researcher either implicitly or
explicitly choses the method without considering the implied
measurement structure.
In the first step of an exploratory study, the solution is computed
and an initial measurement model configuration is implied by the
results. The observed variables can then be regressed on all of the
10
factors found in the first step to determine the sensitivity of the
variables to changes in the factors. This regression step is
particularly useful if the analysis is not an orthogonal solution. For
an orthogonal solution the factor pattern matrix and the factor
structure matrix describe the correlations between the factors and the
variables. However, when an oblique solution is obtained the matrix of
coefficients do not represent the correlations. The measurement
configuration is suggested by the regression coefficients that are
statistically significant. This allows one to determine the
appropriate pattern matrix as well as "identify or name" the underlying
factors. Given that most exploratory models are "just-identified" or
even "under-identified" one cannot statistically test the underlying
measurement model structure. An exploratory approach can only provide
minimal self-validating information regarding the choice among
alternative measurement model configurations.
In contrast, Figure 1 illustrates that the confirmatory approach
hypothesizes a model structure as the basis for the analysis. The
factor analysis is then conducted within the solution constrained by the
hypothesized model configuration. By restricting the analysis such that
the number of factors and their loadings are tested on the sample data,
a confirmatory analysis can provide self-validation. By overidentifying
the model one can statistically evaluate the model configuration and
make comparisons to alternative structures.
The measurement model parameter matrices used throughout the
remainder of this paper are defined as follows:
L is an m by n matrix of the coefficients relating the latent
financial dimension to the observed financial ratio;
1 1
P is an n by n variance-covariance matrix for the latent
financial dimensions;
e is an m by m variance-covariance matrix for the measurement
errors
.
The specification of the parameters in each of these matrices
determines the measurement model configuration imposed on the analysis.
The loadings or pattern matrix, L, depicts the number of factors and
restricts which ratios load on which factor. In an exploratory analysis
all ratios are allowed to load on all factors, whereas in a confirmatory
analysis only selected ratios are allowed to load on particular factors.
For example, by specifying L to be an m by 8 matrix and also specifying
that every element of the matrix is estimated depicts that the
measurement model has 8 underlying financial dimensions and that each
financial ratio loads on each and every one of the underlying financial
dimensions
.
Through specification of the elements in the covariance matrix among
the factors, P, the factors can be restricted to be orthogonal (all of
the off-diagonal elements are restricted to be zero). Alternatively,
specifying P to be full and estimating all of the off-diagonals results
in an oblique solution.
The measurement error matrix, e, can be constrained to be full and
all elements estimated - this denotes that the underlying factors are
not the only systematic source of variation among the variables.
Constraining e such that only the diagonals are non-zero represents that
only the factors on which the variables load are the systematic sources
of variation among the variables. A number of very different
measurement model structures can be constructed through alternative
7specifications for the elements in the various parameter matrices.
12
III. Statistical Procedures and Results
The observable financial measures chosen for this study are a group
of eighteen financial ratios and three size measures. The sample
consists of December 31, 1985 year-end manufacturing firms across a
number of industries. To be included in the sample the firm must be
listed on the Compustat tape. The ratios and size measures chosen are
the following:
cash / current liabilitiescash / salescash / assetsquick ratiocurrent ratiocash flow / salescash flow / assetslong term debt / equitytotal debt / equityinterest coveragecash flow / interestnet income / salesnet income / equitynet income / assetsearnings per shareasset turnoverreceivable turnoverinventory turnoverassetssalesmarket value of equity
This set of financial ratios corresponds closely to the set (or a
subset) of the ratios previously studied by Stevens [1973], Johnson
[1979], or Gombola and Ketz [1983]. In addition, this set of ratios is
also representative of the ratios normally listed in a discussion of
financial statement analysis (see Foster [1986, pgs. 110-111 ] , Lev
[1974], or Van Home [1980]). Both the exploratory and confirmatory
analyses are conducted on this data set.
13
Statistical Procedures
The exploratory analysis uses SHAZAM with an orthogonal varimax
rotation for the solution and is conducted on the correlation matrix.
This choice of an orthogonal varimax solution is consistent with
previous exploratory studies. Although a quartimax rotation would
enhance the factorial interpretations of the observed financial ratios,
the use of a varimax rotation aids in the interpretation of the factors
and is the technique that has been previously applied in most of the
financial ratio literature.
For the confirmatory analysis, the statistical procedure is a Full
Information Maximum Likelihood (FIML) approach. The particular
estimation procedure chosen is LISREL: Analysis of Linear Structural
Relationships by the Method of M axi mum Likelihood by Joreskog and Sorbom
[1978]. This maximum likelihood estimation algorithm requires that the
variables be normally distributed so that the estimated parameter
coefficients as well as the overall model structure can be statistically
tested. This requirement is not neccessary for the exploratory
analysis; however, since we wish to compare the results of the
exploratory and confirmatory analyses the normalized sample is used for
both.
To normalize the data, a logarithmic transformation is applied to
all of the ratios and then the distributions are trimmed, as suggested
by Gnanadesikan [1977, pgs. 121-195] , to provide robust estimates of the
means and variances. 9 The distribution for each ratio is then tested
for normality using the omnibus test (based on skewness and kurtosis)
suggested by Bowman and Shenton [1975]. All of the distributions for
the individual ratios meet the omnibus test for normality criteria at a
level of .05 or better. The preliminary sample consists of 425 firms
14
before trimming. A large number of firms are deleted from the sample
since a firm must be eliminated if an observation on any one of the 21
financial ratios is deemed an outlier. The normalizing of the data
reduces the sample to 193.
Exploratory Analysis
In order to demonstrate the benefits of a theory-derived
confirmatory approach, we present the exploratory analysis first. In
the exploratory analysis no measurement structure is considered "a
priori" and the implied measurement model structure is completely data
derived. Glymour, et. al. [1987] note that exploratory factor analysis
"does not incorporate any of the user's prior knowledge about the causal
process that generates the data."
Using a simple principal components procedure, the eigenvalues and
the cumulative percentages of variation explained by the eigenvalues are
provided in Table 1.
INSERT TABLE 1
Note that the general rule of thumb regarding the appropriate number of
factors to retain in the final solution is to include a factor only if
its eigenvalue is equal to or greater than one. This means that the
factor must explain more than its own proportional variation.
Application of this criteria results in six factors retained for our
data set.
Once the number of factors has been determined, the next step is to
choose the method to be employed in the simplification of the factor
structure. This involves the choice of an oblique solution versus an
orthogonal solution. In addition, the choice of the rotation method to
be applied to the axis must be made. An orthogonal solution is chosen
15
since most previous exploratory studies have used it. A varimax
rotation is employed in order to be consistent with previous studies.
Based on these choices, we estimate the exploratory measurement
model configuration; a six factor model in which the factors are
constrained to be independent. 10 The next step is a regression of the
observed financial ratios on the six estimated orthogonal factors. This
regression allows an assessment of the implied measurement model through
an indication of the degree to which the variations in the observed
ratios are explained by the six factors they are purported to measure.
Note that this model depicts each ratio as measuring all of the
underlying factors since each ratio is regressed on all of the
factors. The results of these regressions are provided in Table 2.
INSERT TABLE 2
In order to have a more meaningful measurement model, a measurement
structure linking individual financial ratios to particular individual
factors (rather than all of the factors) must be imposed. The choice of
the ratio that is to be modeled as an indicator of a particular
underlying dimension can be made in two ways. From a purely exploratory
point of view, one can base this choice on the magnitude of the
correlation between the ratios and the factors regardless of the sign.
An alternative approach, which assumes that the covariation between the
underlying factor and its measure should be positive, is based on the
degree of positive correlation. Some previous studies have used only
the positive factor loadings while others like Johnson [1979] have used
both positive and negative loadings.
A measurement structure configuration based on the degree of
correlation, regardless of sign, between the ratio and the underlying
16
factor is depicted in Figure 2. This model is labeled El.
INSERT FIGURE 2
Note that this measurement structure is very different than one might
expect. It loads ratios that are purported to measure very different
attributes of the firm on the same factor. Two of the six factors do
not show any ratios loading on them. This result is due to using the
magnitude of the correlation between the ratio and the factor employed
to determine the loadings. For instance, the cash flow to sales ratio
is loaded on factor one and has a correlation on -.968, the correlation
with factor six is -.964. Likewise, the receivable turnover ratio has a
correlation of -.554 with factor 1 and .553 with factor two. It is
difficult to compare these results to those of previous studies since
most previous studies only report the single factor loading that the
researcher has chosen; factor loadings of the ratios with other factors
are not provided.
Model E2, based on the magnitude of the positive correlation between
individual ratios and the factors, is depicted in Figure 3. In this
case, no ratio loads on factor five.
INSERT FIGURE 3
For some of the ratios the use of absolute correlations versus
positive correlations makes little difference. The factors chosen for
the ratios to measure are the same under both methods for the following
ratios
:
cash / current liabilities,cash / sales,cash / assets,quick ratio,current ratio,inventory turnover.
17
In order to determine the extent to which the ratios are adequate
measures of the underlying factors we regressed each ratio on the factor
with (1) the largest absolute correlation, and (2) the largest positive
correlation. These results are provided in Table 3.
INSERT TABLE 3
A comparison of the two implied exploratory models, El and E2, shows
significant differences in both the representativeness of the ratios and
the proportion of variation in the ratios explained by the underlying
factor.
Confirmatory Analysis
We initially use an a priori measurement model configuration
suggested by Foster [1986 pgs. 58-70]. This structure has eight factors
which underlie the set of financial ratios used in this study. The
eight factors (underlying economic dimensions of the firm) and the
associated financial ratio measures are as follows (the numbers, 1-21,
associated with each of the ratios will be used rather than the names of
the ratios throughout the remainder of this paper):
Cash position :
1. cash / current liabilities2. cash / sales3. cash / assets
Liquidity :
4. quick ratio5. current ratio
Cash Flow :
6. cash flow / sales7. cash flow / assets
Capital Structure8. long term debt / equity9. total debt / equity
Debt coverage :
10. interest coverage11. cash flow / interest
18
Profitability :
12. net income / sales
13. net income / equity14. net income / assets15. earnings per share
Turnover or activity :
16. asset turnover17. receivable turnover18. inventory turnover
Size :
19. assets20. sales21. market value of equity
In the first confirmatory analysis, the underlying financial
dimensions are assumed to be independent of each other and the financial
ratios are allowed to load only on the dimensions they are intended to
measure. This assumption is consistent with the previous work of
Stevens [1973], Johnson [1979], and Hopwood and Schaefer [1986]). The
parameter matrices which represent this model, Ml, are as follows:
L=
Xl
X2
X3
X4
X5
X6
x7
X8
xl
Ll
X2 L
2x3
L3x4
L4
x5 L5
X6 L
6x? L?x8 L8x9 L9
x 10 L 10x ll L llx12 L 12
x13 L 13x 14 L 14x15 L 15
x 16 ^16x 17 L 17x18 L
18x 19 L 19x 20 L20x21 L21
P(8 by 8) = identity matrix
e(21 by 21) = identity matrix
19
This model is depicted graphically in Figure 4 and it illustrates that
the e's are uncorrelated. This means that the underlying factors are
the only systematic source of variation in the observed ratios.
INSERT FIGURE 4
Model Ml is tested for goodness of fit using a chi-squared test.
This test is applied to determine the model's ability to create a
covariance matrix, S«, that replicates the observed covariance matrix,
So. The chi-squared value for Ml is 4365.5 (189 degrees of freedom).
This high chi-squared value indicates that the measurement structure is
a poor representation of the causal structure underlying the observed
covariation among the financial ratios.
However, as Joreskog and Sorbom [1979] and Bentler and Bonet [1980]
point out, sample size may bias the chi-squared value and lead to
incorrect conclusions. A preferable procedure is to perform incre mental
fit tests based on a comparison of alternative measurement models.
Since the difference in chi-square values for two different model
configurations is also asymptotically distributed as a chi-square
variate, the hypothesis of equivalence between the two model structures
can be tested. We utilize this test procedure below.
An appropriate comparison is to evaluate Ml in relation to a null
model, Mn, defined as a restricted model with no structure implied;
i.e., the elements of L are all set to zero. The chi-square value for
the null model is 8062.9 (210 degrees of freedom). The hypothesis of
model equivalence is rejected at a very high level of significance.
A measure of the proportion of the generalized variance inherent in
the observed data set which is explained by a model, the nor med fir
index , can be computed (Bentler and Bonet [1980]). The normed fit index
20
for Ml is (1 - 4365.5/8069.9), or 46%. The normed fit index can be
interpreted similarly to the coefficient of determination; 46 per cent
of the generalized variation in the observed variance/covariance matrix
is explained by the hypothesized model structure.
As indicated previously, Ml and most of the previous research
(Stevens [1973], Johnson [1979], or Hopwood and Schaefer [1986]) have
assumed that the financial dimensions are uncorrelated. It is very
difficult to justify this assumption given the inter-relationships among
the financing, investing, and operating activities of a firm which are
captured by the underlying financial dimensions. For example, it is
unrealistic to assume that the profitability dimension and the turnover
(activity) dimension are unrelated or that the cash flow dimension is
not correlated with the turnover dimension.
Our second confirmatory model, M2, utilizes the same specifications
for L and e as Ml but allows the underlying financial dimensions to
covary; P is specified to be symetric and full with all of the off-
diagonal elements estimated. This model is depicted in Figure 5.
INSERT FIGURE 5
The chi-square value for M2 is 3661.9241 with 161 degrees of freedom.
Given that more parameters are being estimated, the incremental fit
needs to be assessed since one expects the fit of the model to improve.
M2 has a normed fit index of 55% (recall that the normed fit index for
Ml is 46%). A comparison of M2 to Ml can be made using the same
approach that was used to compare Ml to Mn. This comparison assesses
the propriety of the orthogonality assumption. The null hypothesis of
model equivalence between Ml and M2 is rejected at a significance level
better than .005, based on a chi-squared variate of 704 with 28 degrees
of freedom. This result indicates that the orthogonality assumption is
21
unwaranted. A measurement model allowing covariation among the
underlying financial dimensions is a significant improvement over an
orthogonal model.
However, the adequacy of a model configuration is dependent upon the
individual parameter estimates as well as the overall fit of the model.
The parameter estimates and t-ratios for M2 are provided in Table 4.
INSERT TABLE 4
The parameter estimates indicate some problems with M2. Given that the
analysis is conducted on the correlation matrix, the parameter estimates
for L, P, and e should be less than or equal to 1.00. Note that the
estimates for L5 , L
18 , e5> and e lgare significantly greater than 1.00.
In addition, the estimates for L«g, L 17 , and L«g are insignificant.
This suggests that although the overall model fit seems to indicate a
representative model, some of the parameter estimates are not allowable
and the model structure may be misspecif ied.
Given the difficulties with the M2 parameter estimates, we estimate
an alternative model, M3. This model combines the cash position
dimension of Ml and M2 with the liquidity dimension. This seems
appropriate since cash represents the most liquid asset of a firm. The
debt coverage dimension of Ml and M2 is eliminated in M3. The interest
coverage ratio is a measure of profitability while the cash flow to
interest ratio is a measure of cash flow. These changes decrease
the number factors to six. Note that this is consistent with the number
of factors in the exploratory analysis. M3 is a six factor model in
which the six factors and the associated measures are the following:
22
Profitabilityinterest coverage (x 10 )
net income / sales (x-^)
net income / equity (* 13 )
net income / assets (x^)earnings per share (Xj
5 )
Liquiditycash / current liabilities (Xj)
cash / sales (x2 )
cash / assets (Xg)
quick ratio (x^)
current ratio (x5 )
Cash Flowcash flow / sales (x
g )
cash flow / assets (x-7)
cash flow / interest (X11)
Activityasset turnover (x^g)
receivable turnover (x-jy)
inventory turnover (Xjg)
Sizeassets (xiq)
sales (x2q)
market value of equity (x,i)
Leveragelong term debt / equity (Xg)
total debt / equity (xg)
Each factor is allowed to covary with the other factors. Figure 6
represents this configuration.
INSERT FIGURE 6
The chi-squared value is 4032.406 with 175 degrees of freedom. A
comparison to the null model (Mn) results in the hypothesis of
equivalence being rejected. The incremental fit index indicates that
M3 recreates 50% of the generalized variation in the observed data
matrix. A comparison to M2 indicates that the overall fit of the model
is somewhat poorer; however, the individual parameter estimates are all
appropriate. The factor loading coefficients are all less than or equal
to one and they are all statistically significant except the coefficient
23
linking the asset turnover ratio to the activity dimension. The
parameter estimates for M3 are provided in Table 5.
INSERT TABLE 5
In M3 the factors are allowed to covary. The off-diagonals of the P
matrix indicate that many of the factors are correlated at fairly high
levels. Profitability is positively correlated with liquidity and cash
flow and negatively correlated with leverage. The second factor,
liquidity, is positively correlated with profitability and cash flow and
negatively correlated with leverage. While being positively correlated
with profitability and liquidity, cash flow is also positively
correlated with both size and activity and negatively associated with
leverage. The size dimension is linked positively with cash flow and
activity. The leverage dimension is negatively correlated with
profitability, liquidity, and cash flow.
The covariation between size and the other five factors is of
significant interest. Financial ratios are expected to control for
size differences among firms yet these results indicate that size is
correlated with the other dimensions. To assess the significance of
the covariation between size and the other dimensions, M3 is re-
estimated constraining the size dimension to be orthogonal to the other
five dimensions. The resulting chi-square value for the modified model
is 4078.43 with 180 degrees of freedom. Orthogonalizing the size
dimension results in a model that fits significantly poorer (4078.43
with 180 degrees of freedom compared to 4032.406 with 175 degrees of
freedom.
The matrix that represents the proportion of the variation in the
observed ratios which is not explained by the underlying factor is e.
An analysis of the elements in this matrix indicates that the underlying
24
dimension explains a significant proportion of the variation in the
observed financial measures. It also indicates that the degree of
measurement error varies significantly across the different ratios as
well as across the underlying dimensions. The measurement error
variance is less than 10% in eight instances, between 10% and 50% in six
instances, and greater than 50% for the remaining seven measures.
The magnitudes of the t-statistics for the measurement coefficients
and the magnitudes of the measurement error variances portray the
"representative faithfulness" of the individual ratios as measures of
the underlying financial dimensions. The "best" measure for
prof itabiltiy is the net income to assets ratio. Earnings per share has
the greatest measurement error of any of the ratios used to measure
profitability.
For liquidity there are three ratios that are indicated to be very
adequate measures; cash to current liabilities, cash to sales, and cash
to assets. The cash to assets ratio has the smallest error variance.
Unfortunately, the two more traditional liquidity measures, the quick
ratio and the current ratio, have larger error variances than the cash-
based ratios.
The "best" indicator of cash flow is the cash flow to assets ratio
while the best measure of the size dimension is total assets. The
activity dimension does not seem to be well specified and it is not
measured very well by any of the three ratios. The inventory turnover
ratio has the smallest error variance. The long term debt to equity
ratio has the smallest error variance of the leverage measures.
These results indicate that certain ratios may be better measures
of the underlying economic dimensions than other ratios. Therefore, in
25
choices regarding the ratios to use for firm evaluation, prediction, or
explanation purposes, one should consider the measurement error issue
and choose the ratio with the smaller measurement error. This will
reduce the measurement error problem and enhance the ability to observe
significant coefficients. The underlying financial dimensions and the
financial ratio with the smallest measurement error variance are listed
in Table 6.
INSERT TABLE 6
To further pursue the postulate of parsimony, iterative model
building is undertaken to determine if simpler models can be found that
represent the data as well as M3. In this process, alternative causal
measurement model configurations are estimated and tested. Based on the
results of the simpler models, certain restrictions are relaxed (a more
complex model is developed), the model is again estimated and the chi-
squared value is obtained. A list of the various measurement model
configurations we examine, a short description, and the associated chi-
squared values are provided in Table 7.
INSERT TABLE 7
Mil may be a better representation of the underlying measurement
structure than M3. However, Mil has 5 underlying financial dimensions
and aggregates the activity ratios along with the leverage ratios.
Empirically, these two sets of ratios may be highly correlated but
theoretically they measure very different aspects of the firm. The
normed fit index for Mil is 51%, an insignificant improvement over the
50% for M3.
In the application of causal modeling techniques, the choice of the
most appropriate model must be based not only on the statistical
properties but also theoretical and conceptual underpinnings. Model fit
26
can almost always be improved by relaxing the model structure and
estimating additional parameters. Unfortunately, the model can be
relaxed to the degree that the results are sample dependent and the
parameters have no meaning. Given our data, it appears M3 provides a
reasonable fit from a total model perspective, the coefficient estimates
are appropriate, the model is relatively parsimonious, and the model
seems congruent with theory.
In summary, our confirmatory results indicate that a six factor
oblique model adequately represents the causal measurement structure
underlying the set of financial ratios chosen for this study. The six
latent financial dimensions in the model are: profitability, liquidity,
leverage, activity, cash flow, and size. The model is more parsimonious
than some other model choices; six factors versus eight factors, and the
results indicate that the assumption regarding independence among the
financial dimensions (as presumed in most previous studies) is
unwarranted both theoretically and empirically.
Comparison of Confirmatory versus Exploratory Results
Recall that in the confirmatory analysis we estimate the measurement
error variance for each of the twenty-one financial measures. In order
to compare the measurement error variances of the confirmatory model M3
to the exploratory model E2, the measurement error variances for the
exploratory model are computed as 1 - r . As the error component gets
larger (the correlation between the ratio and the factor get smaller)
the ratio becomes a less valid measure of the underlying factor. A
comparison of the error variances for M3 to those of E2 is provided in
Table 8. This provides some insights into the differences between the
27
data derived exploratory model and the confirmatory model based upon
theoretical underpinnings.
INSERT TABLE 8
In the theory-based measurement model M3, the underlying factor
explains a greater proportion of the variation in the individual ratios
than the exploratory model E2 for all but three of the financial ratios.
These three cases are the quick ratio, the current ratio, and the asset
turnover ratio.
The chi-squared value for M3 is 4032.406 with 175 degrees of
freedom. We can conduct a statistical comparison of the exploratory
models, El and E2, with M3 by computing the chi-squared values for
them. The chi-squared value for El is 6169.659 with 189 degrees of
freedom while the chi-squared value for E2 is 5268.087 with 189 degrees
of freedom. Both El and E2 are significantly poorer fitting models
than the confirmatory model M3. El and E2 have normed fit indices of
.24 and .35, respectively, whereas M3 has a normed fit index of .50.
M3 is more consistent with the observed covariance structure in the
data.
IV. Conclusion
In this study, we illustrate a causal modeling approach for relating
financial ratios to fundamental firm attributes and compare results to
a more traditional exploratory analysis. The use of a confirmatory
approach enables us to answer questions that are not addressed by
previous exploratory studies. Specifically, we are able to:
(1) Define and test a causal measurement model structure for a set
of financial ratios;
28
(2) Determine that the latent economic attributes of the firm are
correlated;
(3) Determine that the economic attributes are proxied quite well by
certain financial ratios; and
(4) Determine the relative degree of measurement error associated
with each individual ratio.
One of the primary benefits of a causal modeling approach is that it
allows us to determine the measurement properties of individual
accounting variables. For the most part, measurement error issues have
been ignored in selecting financial ratios or accounting variables in
various explanatory contexts. However, an understanding of the
measurement properties of accounting data appears beneficial in any
study in which the financial attributes of the firm are used as
exogenous variables.
The use of causal modeling, which is conceptually and theoretically
based, is in its infancy in accounting and there is a need for a great
deal of future research. Some suggestions for future research are
summarized below.
First, within a financial ratio context, there is a need to apply
this methodology to different time periods, different ratios, and
different samples to determine the generalizability of our results. In
this study we look at only twenty-one ratios for a manufacturing firm
sample for one year. Additional industries and additional attributes
need to be examined.
Second, measurement error is an important issue in explanatory
studies where the properties of individual model coefficients are
important. A causal modeling approach can be used to estimate the
measurement error associated with different exogenous variables. An
29
important category of explanatory studies that could potentially benefit
from a causal measurement approach are studies in which attempts are
made to link market variables to fundamental (accounting) variables.
Lev and Ohlsen [1982] call for more research of this type and causal
modeling provides an explicit approach for examining these links.*
Third, causal modeling enables one to determine the degree of
measurement error, but the issue concerning the sources of the
measurement error is not addressed. For example, we determine that
earnings per share measures profitability with a high degree of error.
However, whether the error is due to the rules for computing earnings
per share or some other sources is not addressed. Such issues would
seem to be important from both a policy and a practical point of view.
Finally, causal measurement modeling would seem to have a role in
predictive studies. While explanatory power rather than evaluation of
individual model coefficients is the main objective in such studies, the
inclusion of high measurement error variables may lead to models that
are highly sample sensitive. Research is needed to examine the effects
of measurement error in predictive studies such as bankruptcy, mergers
and acquisitions, and bond rating changes.
30
Endnotes
1. Controversy regarding the role of statistics in causal inference has
existed for some time. See Holland [1986] and Rubin [1986] for a
discussion of the propriety of causal inference in statistical analysis
and the conditions under which a causal statement can be made.
2. Mock [1976] used a measurement model perspective to suggest that
accounting data are measures of underlying unobservable constructs.
Ohlson [1979] alluded to the use of accounting numbers to represent
unobservable economic attributes of a firm in an analytic model
describing security valuation relative to the stochastic behavior of
accounting numbers. Empirical applications of a measurement model
approach are provided by Ziebart [1983], Lambert and Larcker [1985],
Ziebart [1987a], and Ziebart [1987b].
3. In the errors-in-variables problem the explanatory variable is
measured with error. The variation in the dependent variable is
explained by the square of the coefficient times the variance of the
explanatory variable and the prediction error variance. The presence of
measurement error, assumed to be independent and normally distributed,
in the explanatory variable increases the variance for the explanatory
variable and causes the coefficient to be biased downward. As the
measurement error variance increases the coefficient shrinks and becomes
less significant. Accordingly, one should choose to use financial ratio
with the smallest measurement error variance.
4. Computer programs, such as M ILS , which can handle very large data
matrices have only been made available recently.
5. The postulate of factorial causation is usually not an issue since
the measurement structure is based on a more fundamental or theoretical
system tying the observed variables to the underlying unobservable
constructs
.
6. Identification status of the model refers to the degree to which
there is enough information in the observed data relationships to solve
for the unknown parameters.
7. Note the large number of possible model configurations available for
a reasonable size data set. In this study there are 210 covariances and
the possible model configurations is a very large number.
8. For a more complete description of this estimation procedure see
Joreskog and Sorbom [1978] or Ziebart [1987a].
9. See Frecka and Hopwood [1983] for a discussion of the effects of
outliers on the distributions of financial ratios and the propriety of a
trimmed approach.
10. Details regarding the initial factor matrix, the rotated factor
matrix, and the communali ties can be obtained from the authors.
11. Given the sensitivity of the estimation algorithm to the initial
starting values, alternative starting values are introduced. The
parameter estimates are consistent and the inappropriate values are not
the result of the initial starting values.
12. For additional diagnostic assessments, the sigma (model derived
correlation) matrix, the differences between the observed correlation
matrix and the sigma matrix, and the estimated correlation matrix of all
the ratios on all of the underlying financial dimensions can be obtained
from the authors.
13. For an example of linking fundamental accounting variables to
market reactions see Ziebart [1987a] or Ziebart [1987b].
References
Abdel-khal ik, A.R. andAjinkya, B. (1979), Kapirir.il Research in
Accounting: A Methodological Viewpoint, Sarasota, Florida:
American Accounting Association.
Bentler, P.M., and Bonett, D.G. (1980), "Significance Tests andGoodness of Fit in the Analysis of Covariance Structure,"
Psychological Bulletin, 68, 588-606.
Bowman, K.O., and Shenton, L.R. (1975), "Omnibus Test Contours for
Departures From Normality Based on v'bj and b2," Biometrica,
62, 243-250.
Fornell, C. ed. (1982), A Second Generation of Multivariate Analysis- Volume 2: Measurement and Evaluation, New York, N.Y.:
Praeger Publishing
Foster, G. (1986), Financial Statement Analysis, second edition,Englewood Cliffs, N.J.: Prentice Hall
Frecka, T.J., and Hopwood, W.S. (1983), "THe Effect of Outliers onthe Cross-Sectional Distributional Properties of FinancialRatios," The Accounting Review January 1983, 115-128.
Glymour, C, R. Scheines, P. Spirtes, and K. Kelly (1987),Discovering Causal Structure: Artificial Intelligence,Philosophy of Science, and Statistical Modeling, Orlando,Florida: Academic Press
Gnandesikan, R. (1977), Methods for Statistical Data Analysis ofMultivariate Observations, New York: John Wiley.
Gombola, M.J., and Ketz, J.E. (1983), "Financial Ratios Patternsin Retail and Manufacturing Organizations," FinancialManagement, Summer 1983, 45-56.
Holland, P.W. (1986), "Statistics and Causal Inference," Journalof the American Statistical Association, Vol. 81 No. 396,945-960.
Hopwood, W.S., and Schaefer, T.F. (1986), "Incremental InformationContent of Earnings and Non-earnings Based Financial Ratios,"unpublished working paper, Florida State University.
Horrigan, J.O. (1968), "A Short History of Financial RatioAnalysis," The Accounting Review, Vol XLIII No. 2.
Jackson, D.J., and Borgatta, E.F. (1981), Factor Analysis andMeasurement in Sociological Research, Beverly Hills, Ca.:
Sage Publications.
Johnson, W.B. (1979), "The Cross-sectional Stability of FinancialRatio Patterns," Journal of Financial and QuantitativeAnalysis, December 1979, 10-35.
Joreskog, KG., and Sorbom, D. (1978), LISREL IV: Analysis ofLinear Structural Relationships by the Method of MaximumLikelihood, Chicago: International Educational Services.
Joreskog, K.G., and Sorbom, D. (1979), Advances in Factor Analysisand Structural Equation Models, Cambridge, Ma.: Abt Books.
Kim, J., and Mueller, C.W. (1978), Factor Analysis - StatisticalMethods and Practical Issues, Beverly Hills, Ca.: SagePublications
.
Lambert, R.A., and Larker, D.F. (1985), "Golden Parachutes,Executive Decision-Making, and Shareholder Wealth," Journalof Accounting and Economics, 7, 179-203.
Lev, B. (1974), Financial Statement Analysis, Englewood Cliffs,
N.J. : Prentice-Hall
.
Lev, B., and Ohlson, J. (1982), "Market-Based Empirical Researchin Accounting: A Review, Interpretation, and Extension,"Studies on Current Research Methodologies in Accounting: ACritical Evaluation, supplement to Journal of AccountingResearch, 20, 249-322.
Libby, R. (1975), "Accounting Ratios and the Prediction ofFailures: Some Behavioral Evidence," Journal of AccountingResearch, Spring 1975, 150-161.
Marradi, A. (1981), "Factor Analysis as an Aid in the Formationand Refinement of Empirically Useful Concepts," in
Factor Analysis and Measurement in Sociological Research,edited by Jackson, D. and Borgotta, E., Beverly Hills, Ca.:
Sage Publications, 11-50.
Mock, T.E. (1976), Measurement and Accounting Information Criteria- Studies in Accounting Research * 13, Sarasota, Fl.:
American Accounting Associaion.
Ohlson, J. (1979), "Risk, Return, Security-Valuation and the
Stochastic Behavior of Accounting Numbers," Journal ofFinancial and Quantitative Analysis, 16, 317-336.
Pinches, G.E., and Mingo, K.A. (1973), "A Multivariate Analysis of
Industrial Bond Ratings," The Journal of Finance, March 1973,1-18.
Rubin, D.B. (1986), "Which Ifs Have Causal Answers," Journal ofthe American Statistical Association, Vol. 81 No. 396, 961-
962.
Stevens, D.L. (1973), "Financial Characteristics of Merged Firms:A Multivariate Analysis," Journal of Financial andQuantitative Analysis, March 1973, 149-158.
Van Home, J.C. (1980), Financial Management and Policy, EnglewoodCliffs, N.J.: Prentice-Hall.
Ziebart, D.A. (1983), Modeling the Utilization of Accounting Data- An Empirical Study, Unpublished Dissertation, MichiganState University.
Ziebart, D.A. (1987a), "The Effects of Annual Accounting Data onStock Returns and Trading Activity: A Causal Model Study,"Journal of the American Statistical Association, Vol. 82 No.
399, 733-738.
Ziebart, D.A. (1987b), "The Information Content of AnnualAccounting Data: An Empirical Modeling Approach UsingStructural Equation Techniques," Advances in FinancialPlanning and Forecasting, forthcoming.
t-i CO *4—i TJ«
—• f- ~iOJ **
-. _t o
CM ** ri rt
-^ 0) w —
.
o u >» o
CO O CD O i.O
CM co e- o inOS O CN* OS m
O CM 05 O O COOS m o> CO o ^*
c- t* O *P 05 lO
oi o«o a o> ^[0 © ,-1 ,-. ~* c-
(O CD o in OS CM
t* h (\j ifi t* o*r cm co --« t- coco t» co o CO CO
t> rt —( CM —> CMO O CO CO to toC» CM ^ CO « ^
t*io ^ o f m—» 05 05 O CM CM05 co t> co o t-
T CO Irt (D C« «N N O CD - «305 O OS C- CO CO
T - fl
Z" 1 . 1 -3S (- U u u u --------05 o o o o o o
. y y o u y u
~SJ T S3 T SJ
h 4) V 1) D U V
-------CO o o o o o o
- y y y y y o
fl Q T E T
U W «-* *-* *J -• «-»
L. L. U i- t- (-
7 7""""0. a — — — —
O (0 HI (0 fl <Q
X - 1 OJ - - -o *- t- *- *- *- **
03 - - u - - —CO o o o o o o
- y o y y y y
------co S3 to fl to S3aaa a a a
A (0 (0 10 10 A
** Q> QJ CU OJ QJ OJ- - - - - -
,-. u u u u u uCO o o o o o o
• y o o o y y
- - t- - - -(Q (Q 10 (0 m 10
a. a. a. a. a. &
CO (0 W S3 SJ S3
in oj eu qj qj qj cj
W* U _ - - - U
00 o o o o o oo y o y y o
ii —, —
(
—, ^ ^ ,-,
S3 CO (0 (0 (0 (0
U - - - - -(0 S3 ffl S3 S3 (0
a. a a a. a. a
CO tO S3 S3 t0 tO
OJ OJ OJ CU CU OJ OJi - L, - U - -— --.-_.--as o o o o o o
• o y y y o ot| _ ,_, ^H ~H —. ^
CO S3 S3 (0 33 S3
C_ I- t- L. '- -fl 10 (0 (9 O (0
a. a, a a a. a.
S3 S3 S3 (0 S3 S3
CO CU CJ OJ CU OJ OJ
to t- U i- u i- -CO l- U U — - -as o o o o o o
- y y y u y y
(0 tO S3 S3 CO S3
- L. _ - _ -S3 CO (0 (0 CO (0
a a a a. a. a
o <n o m o mt* CM CO C- CO «-»
© O O lO " C)
y y y u y o
CU CJ OJ CJ 4* OJ
B O o O O O Ow y y y y y y
30 — CM CO ^ m CO
\ u u u u c- uo o o o o o
J w w — w w *J
o u y y y y y-^ fl .O S3 S3 S3 S3
-^ « CO 0) — lO
v co rt o t» ^^r co o « to cm
> y y y y y y
™ CU 3j CU 5J OJ OJ
OJ o o o o o oy y y y y y
J3 -^ CM CO ** irt COOJ— ___---
o o o o o og w w *J w -J wu y y y y y yCU S3 S3 S3 S3 9 S3
c- ^» cO C* O »-"
CO CO O 05 tO 0)CM V O •* C- O
y y y y y y
«- %4 <^ <** «- <M>, OJ 3j OJ D OJ OJ
*J o o o o o o»* y y y y y y
— cm co ^> io co
t- t- u u t- uW O O O O Oj3 *^ W w <-< w *->
cu y y y y y y^ S3 SJ S3 S3 S3 S3
CO O ^ CO CM COO ^T ^P CO C- COCO V CO o « --•
y y y y y y
*M ^1 lM <M ^ <WCU OJ OJ QJ OJ OJO o O O O O
oj y y y y y y
OJ -* CM CO ^ Irt CO
CJ> O Tf t- CO Oi(7) CO rr O Wi COCO O O CO O "-t
y y y y y y** *• IM (m «- lM**- '— «*- '—
.
'*-< <*-(
y) OJ ;y cU OJ CJ 0)
OJ o o o o o oc. y y y y y y
Ol CM O CJ> C- OCO CO C- CO O oCO CM CO CO O CO
CU
•h h n CO tj ifi (O
"v. u t- C- (- t- uo o o o o o
3 *J W 4-» *J 4-» 4-*
o y y y y y y^^ t0 S3 (0 (0 (0 (0
c
H
B
n
l
S B
1
II
OJ OJ 0) CJ OJ
u y y y u ut*~" '*-! <bri (^ "4-i (tori
OJ OJ OJ QJ OJ CJ
» o o o o o ooj y y y y y y
^ cm co *r <n co
to co co in o coo> co cm cm m o>*r> cm in co co o
ii n ii ii ii
y y y y y y
> cu cu OJ OJ OJ OJ*-i o o o o o 3—j y y y y y y
—i cm co <j* in to
U U L U L LCU o o o o o oe *_j — *j *- *- _-
o y y u y y yy S3 CO S3 S) S3 sj
co co co o — «eo m •<? co eft oco as o> cm o co
CM 0) <* m CO CMa> cm m in m mco co os co t> ^r
Tf o co co c- to•— co in o co cm05 OS OS CO 00 CO
O f Tf CO f to~« in OS tj> *-> COCM CO CM O O O
CO f CO n -- tt00 T cm in CM oin co « o cm ^
CO CM CO Irt CM TCO CM CM O CO CO05 t> CO 05 CO 05
» S3 S3 S3 S3 SI
CO OJ QJ cj Qj OJ cu»- t- u l. L. U UCO - u u u - u05 O O O O O O
y y y y y y
<0 S3 SJ S3 S3 S3
U E. L. L. L. L.
a a S3 s: sj sj
SJ S3 SJ S3 S3 S3
CM CU QJ OJ OJ OJ OJ
cc t- u u u u uLO u u t- U t- UOJ o o o o o o
- y y y y y y
U L. L. !_ U fafl fl S3 S3 S3 S!
a a a. a a. a
co e S3 sj co cc
CO OJ OJ cu QJ QJ CJ
X ----- -CO c- U U U t- i-.
OS o o o o o o• y y y y y y
H —
i
^-i ^ —
'
—
i
—
<
L. L U L, L, L— T "" ""
— T— — — — — —
S3 «3 CO S3 (0 S3
O OJ QJ QJ OJ QJ OJ^t u u u u u uCO u U U U L. ut> o o o o o o
• y y y y y y
ii —i —
i
—
4
.—
<
—
-
—
»
U U L. (- U US3 S3 (0 S3 (0 S3
a a. a a a a
CO (0 SJ S3 fl (0
CM QJ OJ CU QJ QJ CJ
r- ----- -— -_-_--c- o o o o o o
- y y y y y y
L. - - '- L -*" f "" *7 Ta a a a a a
S3 53 S3 S3 SJ S3
T-t OJ QJ OJ OJ OJ QJ-j ----- _c- ----- -OS o o o o o o
y y y y y y
______S3 S3 S3 S3 tO S3
a a a a a a
« co cm o cc ino » o ^? T Tco co in o cm «
in CO CM 05 CO T?o »- m t> co o^ 1/) CO « N n
co in os co co co05 os ^ m o mco in in o co cm
O) t> t> eo t coo co m CM o *r—
« co -^ o o oCO OS ^ m CM CMin t> to cm •-" inco o o o -i o
CM CM CO O CO Oco t> Tp m in oCO - CM CO O CO
II II II li
. i . _ L . .
w y y y y y y—• *- tb. <to. i— •— (».— — — — — — —.a cu a,, u q a> cu
S3 o o o o o o—
«
y y y y y u
y y y y y y
*- <fc- -to- ^-i «- <»_
OJ CJ CU QJ QJ cuo o o o o oy y y y y y
y y u y u y**- <M 4-i <•_ <*-. <*-.
**- <M <*- <H *- Vri
OJ OJ CU CJ OJ CUo o o o o oy y y y y y
y y y y y y
OJ qj Oj qj OJ QJO o O O O Oy y y y y y
y y y y y y
OJ QJ QJ QJ OJ CJO O O O O Cy y y y y y
y y y y y y
^•b <U(M IWlM** <M CW <** <»— V*OJ QJ QJ QJ QJ CUO O O O O O
co y y y y y ycu
z — cm co *f m coQJ
t- o c o o o o3 w — w w w wy y y y y y u
(9 S3 S3 S3 S3 S3
CD
cm co -» in co
_ - - - - - -—
i o o o o o o!0 w w *j w *- wus y y y y y y
S3 S3 S3 S3 S3 S3
«s ^- cm co v m COwQJ U L. U U U L.
W O O O O O OUS *-• w W w *-• *J
ss y y y y y y(0 S3 S3 S3 S3 S3
—i cm co *j- in to O —• CM CO t m CD w -• cm co ^ in co
^ u u u t- u uo o o o o o
3 *V W *> W w wo y y o y y y^ CO S3 S3 S3 <0 S3
cm o o ^ o> t>co m 01 co cj> ocj) o c- c- m v
N O 0) n O ffl
O rt V ON (O0) O) (O (D V (fi
CO to cm a> CM CO« CO CO CM -« f-CJ) CO CO O tO CM
<Q K| 9 O 10 !Q
h V 4) V V V Cd
— -- - - - - -10 s- u U U l~ ua> o o o o o o
• o cj y o cj o
CO CO CO CO TO fO
- - - - - U10 9— — Jj- — — -i
fl fl Id O (O H]
-t U i 1, 1.' U Ij
o U u - - i- i-
CM !-• U U U U Uo> o o o o o o
. o o CJ CJ u o
CO IQ tO fO CO CO
CM •— -^ — -* —' •-*
U *-" W -M -W w *->
_ - u t* - -(0 CO CO CO CO fl
a a a a a a
CO (0 CO fO CO S3
CO 1 D Qj CU V DL ----- -«-i U L, t- L. U UCJ) O O O O O O
• o o cj cj o o
II —. -^ —. —« —. —
(
(0 33 CO 3 (0 CO
U L. U t- U UCO CO (0 CO CO Sa, a. a. a. a a.
co co a> m co min co co t* to nIrt CO CM CM -^ ^
CO C— t> CJ) CO CJ)
t- ** CO O CM CMIT) CO CM CM —« CM
en f co in « coco v U) o CM COCO IO CM CO CM O
O y (J O O CJ
1) 4; CU D 1) CUo o o o o oo CJ u CJ u u
U U U CJ CJ u
U D y i; 1; 1;
o o o o o oO y cj <j CJ O
- — _—_——- 1.
3J 1, v U -CT O o O O O O0) U CJ CJ O CJ CJ
« cm co tj> m co «-• CM CO f in CO— cm co ^» in co
a3 ----- --^ o o o o o ota *j w *-• w -j w> CJ O CJ O CJ CJ
CO (0 CO to to CO
a
O O CO -D CO COO O CO O CM ^>CJ) 35 CJ) C- CO *h
Q C O fl (CJ 10
U0 - _ L 01 U UOS t- u - u u uIfl - L- - - U UJ) o o o o o o
• U CJ U U CJ u
CO CO (0 CO
- - I- - - L"1 'C ^ T T Oa a a a a. a
CJ) CJ) CO CO CJ) CJ)
C- O CO CJ) co cmco ^ m ^ o o
« CJ CJ o CJ o
t, i« i i, i, a. uw o o o o o o0) o o o o CJ CJ
cm co «t m a
<* co -v o *r t-co o m co cj> cm03 CM *r ^ —t CM
IQ fi 10 |Q (0 d
- i 1 -j "j ; cj ij
C) t- u u u u uCO L. U U U U Ut- o o o o o o
• u u U tj u y
to to co co <o to
_ - - - - _fl fl to & f& (0ao.ao.aa
~« t^ CO O CJ) C-co o in to cj) i-i
t» i-i CM CM O »-•
y) O d u 1) ^ cy
O) o o o o o oU <-> O CJ CJ CJ CJ
—• cm co ^r lo to
v o cm r* 00 «tr co cj) r- cj) oo m co to tt co
to CO to 10 w to
ir.- - C, i 1 - X' l:
tt U U U U U UC - i- L. U '_ Uec o o o o o o
CJ u CJ CJ CJ CJ
to CO CO CO CO CO
u t_ U (- u uco to to to to to
a a a a a a
01 cm t- o o to—
• c- ^ o in coO N ^ VN IT)
. 1
1
1 II
1
n u
s B B cV 91 0} Bl 01
u u CJ o CJ CJ
"—- — _ —
^ Cy 1 li '.' ii
o o o o o oU CJ cj cj cj o
0) cm co rf m co
O L. u u u u uc o o o o o oU <-> W W *J W *J3 CJ CJ CJ CJ U CJ•- (0 to to CO to CO
in ** co co o cofr- w t- in -* toco in to m o f
CO to to to CO CO
s - --------CO o o o o o o
• o O CJ CJ CJ CJ
------co to co ra to to
a a a a a a
h CO V CM CM Tin « in « o cmCM fl> CM V CM CO
H H * 1
1
1
c B B = n s4) u 0> 0) 0)
CJ u CJ u o
t_ o aj u u cj cj
0) o o o o o o> O cj u O O CJ
•^ cm co ^ m to
- O tO (M f OCM CO — CO CM COin if, o f co t
CO rO to CO co to
to CO 3 to to to
CJ) CJ) CO CO V «O) O O •- CM •-•
CO TT O CO CM CO
1! fl il t II
B B B B0) 0) B 0) 0)
U o CJ CJ u CJ
II jj (j jj l Ufc- O o o O 3 O4) U (j CJ O O O
u ^ cm co ^ in to3- -_-_- _
o o o o o o> w w w *- — wU cj cj u cj cj yC <0 CO CO CO CO .-o
r\ r\ r\
10
o
aXDd
Li
r~\ 0|^~\
« in <o
u u uo o 3
4-1 uo u uffl ra ea
<M
o
~H CM C«J t* 1-. ^ r-.
ttltftAAAA AAAA
tt
¥ x x x
AAA AX
tC- «-i —
•
M
^ /~\ ^~\
axwho
f?\
•ooz>.uo
«Bu
s~\
f-\
A A A A A A
/-\
i- —
•
x -A x >:
tA A
10 — r*
CO — ~
V *r CO CMm CO o> CO *To TT CO a> o
co « —*
— *—' in
J kj tJ01 ^ ^ _ ^ rt
o n oft
O O ~o o o
o m co coO O V »o co o «-i
o r- co co ~+o r* -* o c-O tt CO O O
O CO o> *f o ~+o o co -* in coo i co n o w
TT CO > O 0) Hc- -< m co ** o>o o o o o o
Oi/:t*w«nco005CMCOCOCOCJOOONNNNOO
o agft U
CM ^CO -H ~4
<-» *-» CO
~^ Ift
*-'"—'«
C w *-»
,-, _ ^ _ w _ ce v v ai
** *J 4* tm IB O > (ft —
<
0> —
*
\fi CO CO10 « ^ m
OS ^ -h
CO -^ «
—
*
*T .»>«-* O
— ^ 00 03 -« —
'
^ i-i CO —
tA nJ -JOft rt «« rt »-l
O C- TJ- CJo *^ o 0>O CM ff «
** CO CO (T5
CJ CO CO COCO O «-» CNJ
- i i
ooo
CJCM
aen o a
in
n 10 -. -<
—
«
w
oTJ
z>h
s
tttttttto co o *r m
A AAA AAA AAAtf
V CO CO CO
C- —
'
HECKMANBINDERY INC.
JUN95-T.-PW-^ N MANCHESTER.
INDIANA 46962