The net impact of corporate seasonality on the accuracy of earnings forecasts published by financial analysts

BEBR

FACULTY WORKING
PAPER NO. 89-1623

The Net Impact of Corporate Seasonality
on the Accuracy of Earnings Forecasts
Published bv Financial Analysts

Suzanne M. Luttman
Peter A. Silhan

Ths Library of the
M 2 5 i990

Unhrj

College of Commerce and Business Administration
Bureau of Economic and Business Research
University of Illinois Urbana-Champaign

BEBR

FACULTY WORKING PAPER NO. 89-1623

College of Commerce and Business Administration

University of Illinois at Urbana- Champaign

December 1989

The Net Impact of Corporate Seasonality on the Accuracy of
Earnings Forecasts Published by Financial Analysts

Suzanne M. Luttman
Assistant Professor of Accounting
University of Colorado at Boulder

Peter A. Silhan

Associate Professor of Accountancy

University of Illinois at Urbana- Champaign

Helpful comments on earlier versions of this paper were received from Larry
Brown, Nick Dopuch, Jim Gentry, Jim McKeown, Paul Newbold, Katherine Schipper,
and Dave Senteney.

Digitized by the Internet Archive

in 2011 with funding from

University of Illinois Urbana-Champaign

http://www.archive.org/details/netimpactofcorpo1623lutt

THE NET IMPACT OF CORPORATE SEASONALITY ON THE ACCURACY OF
EARNINGS FORECASTS PUBLISHED BY FINANCIAL ANALYSTS

Abstract

This study investigates the net impact of corporate seasonality (CS) on
the accuracy of forecasts of quarterly earnings per share (EPS) published by
financial analysts. Using a multivariate model to control for other factors,
the results generally do not indicate an association between the degree of CS
and EPS forecast accuracy.

The Net Impact of Corporate Seasonality on the Accuracy of
Earnings Forecasts Published by Financial Analysts

1. Introduction

The potential impact of corporate seasonality (CS) on the reliability
and usefulness of interim financial reports has been a source of concern.1
Maingot (1983, p. 139), for example, warns that "seasonal fluctuations t can
cause amounts for one period to be misleading unless care is taken to explain
seasonal effects." Corporations also have expressed concern over the
"distortion caused by seasonal factors" (Leftwich, Watts and Zimmerman, 1981,
p. 55). Given these concerns, APB Opinion No. 28 (AICPA, 1973) and SEC
Accounting Series Release No. 177 (SEC, 1975) encourage highly seasonal
companies to disclose the effects of seasonality on their accounting results.
To accomplish this, various approaches, including seasonal adjustment, have
been discussed as a means of providing such information (Frank, 1969; Foster,
1977, p. 2; FASB, 1978, par. 315). To date, however, such voluntary
disclosures are rare and seasonal adjustment has not been adopted at the
corporate level.2

The purpose of this study is to provide evidence on the potential need
for measuring and disclosing the effects of seasonality on quarterly
accounting results at the corporate level. Unlike the U.S. government, which
seasonally adjusts much of its published time-series data (e.g., gross
national product, corporate earnings, housing starts, consumer prices, and
unemployment rate) , corporations have not supplemented their quarterly

disclosures with seasonally- adjusted data. Since an important objective of
interim reporting is to provide information for predicting corporate earnings
(FASB, 1978), this study investigates the net impact of CS on the accuracy of
forecasts of quarterly earnings per share (EPS) published by financial
analysts. CS effects are measured here in terms of EPS forecast errors and CS
is defined as the degree (in percentage terms) of variability in quarterly
accounting outputs which is attributable to the seasonal component in those
time series.

Prior research has shown that (1) quarterly earnings typically have both
a seasonal component and an adjacent component (e.g., Watts, 1975; Foster,
1977, Griffin, 1977; Brown and Rozeff, 1979; Bathke and Lorek, 1984), (2)
investors take CS into account when using quarterly data to predict quarterly
earnings (e.g., Foster, 1977; Bathke and Lorek, 1984), (3) EPS forecast
accuracy varies by quarter (e.g., Collins, Hopwood and McKeown, 1984), and (4)
investors react to unexpected quarterly earnings (e.g., Foster, 1977; Bathke
and Lorek, 1984; Mendenhall and Nichols, 1988). However, no previous study
has assessed the net impact of CS on financial analyst forecast (FAF)
accuracy. The current study helps fill this void.3 It is based on the
premise that the degree of seasonality (a characteristic of time series data)
might favorably or unfavorably affect FAF performance (an outcome based in
part on such data) . The net impact of CS on FAF performance is an empirical
issue because it cannot be determined a priori. This aspect of the study is
discussed in the next section.

2. Motivation

The seasonal component of an economic times series represents "the
intrayear pattern of variation which is repeated constantly or in an evolving
fashion from year to year" (Shiskin, Young and Musgrave, 1967, p. 1).
Although seasonality represents pattern, which tends to improve forecast
accuracy, "some believe that for a seasonal business, the variations in
quarterly revenue and the difficulty in relating cost to revenue for parts of
a year seriously impair the usefulness of any measure of interim period income
from operations" (FASB, 1978, par. 311).

In this study, usefulness is assessed in terms of EPS forecast accuracy.
Shillinglaw (1961, p. 223) cautioned that seasonality "may affect interim
statements in such a way as to reduce their usefulness in income forecasting,
and it is to these influences that we must devote our attention. " To test
this assertion, it is hypothesized that if CS affects EPS forecast accuracy it
could do so in two directions, resulting in either lower (Alternative 1) or
higher (Alternative 2) forecast errors across firms. In addition, it might
have no real or observed effect on forecast errors across firms (Alternative
3).

Under Alternative 1, CS could reduce forecast errors if the CS component
were predicted with a high degree of accuracy. Under this alternative,
forecasters would be expected to adjust fully for CS effects on accounting
results and to exploit this information to improve EPS forecasts. In seasonal
firms, the CS component would be expected to lead to lower forecast errors,
since the seasonal component represents pattern.

Under Alternative 2, CS could increase forecast errors if the CS
component were predicted with a low degree of accuracy. Under this
alternative, forecasters would find it difficult to identify and interpret the
CS component. In seasonal firms, the CS component would lead to higher
forecast errors, since it would contain estimation error.

Under Alternative 3, CS could have no effect on forecast errors across
firms if both of the above effects existed and cancelled each other out among
the firms sampled or CS errors were offset by other errors. This alternative
is the null hypothesis.

Alternative 2 and Alternative 3 are plausible because the process of
predicting corporate earnings remains subjective and accounting practices vary
considerably across firms. The ability of individuals to adjust for CS is at
issue because (1) predictions of a stochastic time-series, such as corporate
earnings, involve judgment and (2) research has shown that individuals have
difficulty estimating stochastic processes (Eggleton, 1976, 1982; Doktor and
Chandler, 1988). This aspect of FAF performance is borne out by several
recent studies which indicate that (1) FAFs are not necessarily unbiased
(e.g., Ricks and Hughes, 1985; Biddle and Ricks, 1988; Lys and
Sivaramakrishnan, 1988; Mendenhall and Nichols, 1988; O'Brien, 1988) and (2)
FAFs do not completely dominate time -series models in terms of providing
information used by investors (e.g., O'Brien, 1988). Together, these studies
imply that a positive association between CS and FAF errors (Alternative 2)
cannot be ruled out a priori because systematic forecast errors have been
shown to exist in other contexts.

Alternative 2 also cannot be ruled out because the Financial Accounting
Standards Board (FASB) has noted that the "ability of users of interim reports

to ascertain turning points, or change in trend, is obscured if an enterprise
is subject to significant seasonal influence" (FASB, 1978, par. 314). This
statement implies that CS could be a problem in certain years when conditions
are changing and extrapolation errors are more prevalent. It also implies
that forecasters sometimes may have difficulty estimating the trend component
of a seasonal time series. This would result in higher forecast errors for
seasonal companies.

In effect, the current study serves to provide evidence that was called
for by Foster (1977), who examined six extrapolative models to determine their
relative predictive ability in terms of (1) predicting step-ahead quarterly
sales, expenses, and earnings, and (2) estimating market associations with
each forecast model. He found that seasonal time-series models were more
closely associated with aggregate market reactions than nonseasonal time-
series models and concluded that "the capital market appears to be adjusting
for seasonality by employing a forecast model that incorporates seasonal
patterns in quarterly earnings" (Foster, 1977, p. 18). He noted, however,
that additional research would be required to examine further the various
claims made by some individuals that seasonal earnings could be "misleading"
and "confusing" to investors (Foster, 1977, p. 16). The current study is
designed to serve this purpose by investigating the net impact of CS on EPS
forecast errors.

3. Research Design

The current study uses two measures of seasonality: the degree of
seasonality in sales (DSS) and the degree of seasonality in earnings (DSE) .

Each measure represents the percentage of quarter-to-quarter variation
explained by the seasonal component of each time series. The Census X-ll
Model, which is discussed in Section 3.3.2, was used to generate these

measures .4

A multivariate model was used to control for other factors. This model
is an adaptation of the general model formulated by Albrecht et al . (1977)
adapted in part by Baldwin (1984) and Brown, Richardson and Schwager (1987) ,
for example, and endorsed by Brown, Foster and Noreen (1985, p. 125) and
Foster (1986, p. 287). Albrecht et al. (1977) viewed FAF accuracy within a
multivariate framework and suggested the following variables as potential
determinants of FAF accuracy: (1) earnings variability, (2) corporate age, (3)
corporate size, (4) detail of information, (5) corporation industry, (6) lead
time to terminal date, (7) calendar year of forecast, and (8) forecaster. In
addition to CS , this study considers factors (1), (2), (3), (4), (6) and (7)
as potential determinants of EPS forecast accuracy.5 Time series data from
the quarterly industrial COMPUSTAT file (COMPUSTAT) are used to measure CS ;
FAF data from the Value Line Investment Survey (Value Line) are used to
measure EPS forecast accuracy.

3.1 Hypothesis

The following null hypothesis (Alternative 3) was tested using multiple
regression analysis.

H„ There is no association between the degree of corporate

seasonality and the quarterly income forecast errors of financial
analysts .

3.2 General Model

The following general model was used:

FE - f(CS, PVAR, SIZE, TIME)
where

FE = EPS forecast error,

CS - Degree of corporate seasonality (DSS, DSE) ,
PVAR - Past year-to-year variability of earnings,
SIZE = Firm size, and

TIME = Number of days between forecast date (FDATE) and
subsequent earnings announcement date (ADATE) .

Since FAF accuracy varies across years (O'Brien, 1988) and CS effects on
FAF accuracy could vary across years and quarters, regression models were
estimated for every quarter represented in a seven-year FAF sample (1980-86).
Since the same firms were used across years, joint generalized least squares
(JGLS) was used to jointly estimate sets of regressions as systems of
seemingly unrelated regressions (SUR) . Zellner (1962) notes that gains in
estimation efficiency can be achieved by using SUR, which takes into account
the fact that cross -equation error terms may not be zero. In all, there were
eight SUR estimations needed for this study (four quarterly systems, Q1-Q4,
for each CS measure, DSS and DSE).

3.3 Variables

3.3.1 EPS Forecast Error. FE, the dependent variable, is the absolute
value of the difference between forecasted EPS and actual EPS scaled by the

8
absolute value of forecasted EPS. Expressed in percentage terms, this metric
can be denoted as follows:

FE- (|(FEPS - AEPS) / FEPS | ) x 100

where FEPS - forecasted EPS, AEPS - actual EPS, and | | - absolute value
operator. Values in excess of 300 percent were truncated at 300 percent.6
Value Line was used to provide FEPS and AEPS data for the first, second, third
and fourth quarters (Ql, Q2 , Q3 , Q4) of the current year. In effect, these
four FAFs represented step -ahead- one (t+1) through step -ahead- four (t+4)
forecasts. All data were adjusted for stock splits and stock dividends.

3.3.2 Corporate Seasonality. CS , the variable of interest, was measured
both in terms of corporate sales (DSS) and corporate earnings (DSE) . In this
study, DSS and DSE are expressed in percentage terms. Each represents the
relative contribution of the seasonal component (S) to the variance of the
original series for span one, the interval between adjacent observations. The
Census X-ll Model (Shiskin, Young and Musgrave, 1967) was used to compute
these two measures for each year (1980-86) using a time window of the
preceding ten years of COMPUSTAT data.

The X-ll Model, which is used to seasonally adjust a wide variety of
economic time series, decomposes time-series data into three components:
seasonal, trend-cycle, and irregular. The seasonal component reflects the
intrayear variation which is repeated from year to year; the trend-cycle
component reflects the long- terra trend and the business cycle; the irregular
component reflects the residual variation in the data. In this study, the
additive formulation of the X-ll Model was used on both sales and earnings to

ensure comparability between DSS and DSE and to accomodate negative earnings
numbers which precluded using the multiplicative alternative. This additive
decomposition can be represented as follows:

X = S + C + I
where X is the variable of interest (sales or earnings), S is the seasonal
component, C is the trend- cycle component, and I is the irregular component.

In effect, then, DSS and DSE are summary measures of the relative
contribution of S to the variability of each series (sales or earnings) . Such
measures are generated as standard outputs of the X-ll Model.7 Since the
expected sign of the association between CS and FE cannot be determined a
priori (see Section 2), a two-tailed test is used for this variable.

3.3.3 Past Year-to-year Earnings Variability. PVAR represents past
year-to-year earnings variability, which has been shown to affect FAF
performance (Barefield and Comiskey, 1975; P incus , 1983). It is measured
using the Value Line Earnings Predictability Index (VLPI) . To mitigate the
effects of quarter-to-quarter seasonality and provide investors with a measure
of past EPS variability, Value Line computes this index from the year-to-year
standard deviation of the percentage change in the quarterly earnings series
over the past five to ten years. This seasonality-free index is scaled from 5
Co 100, such that 100 represents a highly predictable (low PVAR) company.
Since the expected sign of VLPI is negative . a one- tailed test is used for
this variable.

3.3.4 Firm Size . SIZE is measured as the natural log of the previous
year's annual sales (LnSALES) . This variable is included as a proxy for the

10
amount of information made available by companies and the amount of effort
expended by financial analysts in predicting corporate earnings. SALES is
used as a proxy for size in the Fortune 500 and has been used as a proxy for
size in various studies (e.g., Schiff, 1978). Brown, Richardson and Schwager
(1987), Mendenhall and Nichols (1988), Bathke, Lorek and Willinger (1989) and
others have found that EPS forecasts generally are more accurate for large
firms than small firms. Since the expected sign of SIZE is negative . a one-
tailed test is used for this variable.

3.3.5 Time Lag. TIME, the number of days between the FDATE and ADATE,
is included to control for FAF performance differences due to the potential
acquisition of new information by financial analysts subsequent to the
publication of an EPS forecast. Crichfeld, Dyckman and Lakonishok (1978),
Bamber (1987), O'Brien (1988) and others have observed that FAFs generally
become more accurate as the earnings announcement date approaches. Since the
expected sign of TIME is positive . a one-tailed test is used for this
variable .

3.4 Data Sample

Every firm (1) was listed in both Value Line and. COMPUSTAT, (2) was a
December fiscal-year company throughout the sample period, (3) remained in its
designated four-digit COMPUSTAT industry classification code throughout the
sample period, (4) had complete COMPUSTAT quarterly sales and earnings before
extraordinary items from 1970-1 to 1985-IV, (5) had a complete set of Wall
Street Journal EPS announcement dates, and (6) had no FE denominators (FEPS)
between -.05 and .05. Criterion (6) was used to mitigate outliers due to

11

small denominators (see Bamber, 1987). Satisfying these criteria were 174,
173, 183 and 193 firms with complete data for the first, second, third and
fourth quarters (Ql, Q2 , Q3 , Q4) , respectively.

In all, there were 197 different firms representing 46 two-digit, 83
three-digit, and 88 four-digit Standard Industrial Code (SIC) categories.
Since each quarter required seven years of complete data for the JGLS
analysis, there were 1,218, 1,211, 1,281, and 1,351 FAFs for quarters Ql
through Q4, respectively. Four data items were required for each FAF (FEPS,
AEPS, FDATE, ADATE) . In addition, for each firm in the 197-firm sample, there
were three variables (DSS, DSE, SIZE) computed once for each year (Ql through
Q4) from COMPUSTAT data through the end of the preceding year and one variable
(VLPI) recorded once for each year from Value Line on the date that the four
quarterly FAFs were made. There were 1,379 individual measurements recorded
for each of these four variables (197 firms x 7 years).

3.5 Regression Models

Two regression models were used to test for CS effects:

LFE, - (30 + /?,DSS, + /32VLPI, + &SIZE, + 0/TIME, + €, (Model 1)
LFE, - j30 + 0.DSE, + &VLPI, + 03SIZE, + 04TIME, + 6, (Model 2)

where LFE, - LnFE for firm j , DSS, - degree of seasonality in sales for firm
j , DSE, = degree of seasonality in earnings for firm j , VLPI, - Value Line
earnings predictability index for firm j , SIZE, = LnSALES for firm j , and
TIME, = ADATE - FDATE for firm j. Each variable on the right-hand side of

12
these two cross -sectional models was measurable before the dependent variable
became known on the earnings announcement date (ADATE) .

Since FE tends to be skewed, LFE, the natural log transformation of FE,
was used to satisfy the distributional assumptions of the two regression
models. Also, since DSS and DSE tend to be highly correlated, these two
measures were not both included in a single model to avoid multicollinearity
problems.8 Regression diagnostics indicated that (1) the models as specified
did not violate the distributional assumptions of the regression analysis and
(2) the independent variables that were included in each model did not exhibit
multicollinearity problems.

4. Empirical Evidence

Empirical evidence on the association between CS (DSE and DSS) and FAF
performance (LFE) is presented in this section. This evidence does not
support the proposition that high CS tends to impair FAF performance
(Alternative 2) nor does it support the proposition that high CS generally
tends to improve FAF performance (Alternative 1) .

4.1 Descriptive Statistics

Table 1 provides a seasonality profile of the sample. Based on the full
197 -firm sample (i.e., every firm that was used in at least one quarter), it
indicates that on average DSS is slightly higher than DSE (58.0 versus 55.9).
It also identifies a number of industries with high CS (SIC 20, food and
kindred products; SIC 27, printing and publishing; SIC 49, electric, gas and
sanitary services) and low CS (SIC 10, metal mining; SIC 26, paper and allied

13
products; SIC 48 communications). None of the two-digit SIC categories
exceeded 10 percent of the full sample and only six two-digit SIC categories
exceeded 5 percent of the full sample. Thus the composition of this sample is
relatively diverse both in terms of seasonality and industry representation.

Table 2 provides some additional statistics on the four quarterly
samples (Q1-Q4) used in the regression analysis. As expected, LFE and FE
increased with TIME from 2.883 and 38.520 (Ql) to 3.248 and 51.032 (Q4) ,
respectively. As indicated by the standard deviations and means of LFE versus
FE and SIZE versus SALES, the log transformations of FE and SALES reduced the
coefficients of variation associated with those variables.

4.2 Regression Results

DSS and DSE were used in cross-sectional regressions with LFE as the
dependent variable and VLPI , SIZE, and TIME as other explanatory variables.
These regressions were estimated jointly by quarter using JGLS . The results
of these regressions indicate that CS generally did not affect FAF performance
(Alternative 3) .

Table 3 presents the regression results for Model 1 which includes DSS
as an independent variable. Out of 28 cross-sectional tests, the regression
coefficient for DSS was significant once in Ql (1984), twice in Q3 (1982,
1986), and three times in Q4 (1982, 1984, 1986). The sign was positive once
(in Ql) and negative five times (in Q3 and Q4) . Using a one-tailed test (a <
.10), VLPI was significant 28 times, SIZE was significant seven times and TIME
was significant 13 times. The system R-squares ranged from .146 (Q3) to .206
(Ql).

14
Table 4 presents the regression results for Model 2 which includes DSE "
as an independent variable. These results are similar overall to the Model 1
results. Out of 28 cross-sectional tests, the regression coefficient for DSE
was significant once in Ql (1984), twice in Q3 (1982, 1983), and once in Q4
(1985). The sign was positive once (in Ql) and negative three times (in Q3
and Q4) . Using a one-tailed test (a < .10), VLPI was significant 28 times,
SIZE was significant seven times, and TIME was significant 13 times. The
system R-squares ranged from .146 (Q3) to .208 (Ql) .

Overall, then, except for the first quarter of 1984, which indicated a
statistically significant positive sign for both DSS and DSE, the results of
the multiple regression analysis do not support the proposition that CS
adversely affects FAF performance (Alternative 2) . In addition, except for
five DSS quarters (Q3:1982, Q3:1986, Q4:1982, Q4:1984, Q4:1986) and three DSE
quarters (Q3:1982, Q3:1986, Q4:1985), the results do not support the
proposition that CS improves FAF performance (Alternative 1) . Consequently,
the results generally indicate that the null hypothesis (Alternative 3) could
not be rejected.

5. Conclusions

This study investigates the proposition that CS could either favorably
or unfavorably affect FAF performance (measured in terms of forecast errors,
FEs). Two potential CS-FE linkages were examined: (1) the association
between DSS and EPS forecast accuracy and (2) the association between DSE and
EPS forecast accuracy. In both cases, a multivariate approach was used to
control for other potential determinants of EPS forecast accuracy.

15
The results indicate that except in one quarter (out of 28 quarters) , on
average FAF performance was not adversely affected by CS . Therefore, it
appears that additional disclosures are not needed to remedy a positive
association between CS and FE (Alternative 2). However, since CS represents
pattern (which tends to improve forecast accuracy) and there generally was no
negative association between CS and FE (Alternative 2) , the results may also
suggest that financial analysts are not exploiting this pattern very well.
Perhaps, then, additional emphasis should be placed on the CS component and
future research should be designed to address this issue further. Also, since
the results (Alternative 3) could be due to a variety of factors, additional
research is needed to determine why the net impact of CS on EPS forecast
errors is essentially zero, and not negative as expected under Alternative 1.
By using a multivariate approach, the results also serve to enhance our
general understanding of the joint determinants of EPS forecast accuracy.
Knowledge of these determinants is useful for research requiring measures of
the market's expectation of earnings (Brown, Richardson and Schwager, 1987, p.
50) . Noteworthy is the impact on FAF performance of past year-to-year
earnings variability, PVAR, as indicated by the Value Line earnings
predictability index, VLPI . This index, which is based on past year-to-year
earnings variability, was statistically significant in every cross-sectional
regression. It appears, then, that even for forecasting on a quarter-to-
quarter basis, the major source of inaccuracy in FAFs is year-to-year earnings
volatility, rather than the extent of any seasonal patterns (which seem to be
reasonably anticipated) . Research designs which need to measure earnings
surprise or control for ex ante predictive ability for some other reason
therefore should consider controlling for PVAR, which affected FAF performance

16
more often than SIZE or TIME. These results also suggest that future studies
attempting to measure or control for SIZE effects should consider PVAR as a
potential omitted variable.

17
FOOTNOTES

1. See Blough (1953), Capon (1955), Shillinglaw (1961), Green (1964),
Taylor (1965), Rappaport (1966), Frank (1969), Backer (1970), Bollom and
Weygandt (1972), Coates (1972), Edwards, Dominiak and Hedges (1972),
Reilly, Morgenson, and West (1972), Bollom (1973), Kiger (1974),
Nickerson, Pointer and Strawser (1975), Foster (1977), Carlson (1978),
Schiff (1978), Leftwich, Watts and Zimmerman (1981), Fried and Livnat
(1981), Maingot (1983), Bathke and Lorek (1984), and Burrowes (1986).

2. An examination of the most seasonal firms in the current sample (upper
third) revealed no such disclosures. Seasonality was mentioned briefly
by only 12.1 percent of these firms (eight cases) and not mentioned at
all by 87.9 percent.

3. By comparing seasonal and nonseasonal models of quarterly data, Foster
(1977) provided evidence which indicated that investors adjust for
seasonality, but he did not examine the impact 'of seasonality on the
accuracy of EPS forecasts published by financial analysts. Collins,
Hopwood, and McKeown (1984) provided some initial evidence which
indicated that seasonal firms might be associated with lower FAF errors
than nonseasonal firms. However, because the focus of their study was
not on seasonality per se, there were no controls for other determinants
of FAF accuracy and no statistical tests were performed on the observed
differences .

4. Previous studies concerned with measuring CS typically have relied on
categorical measures derived from various sources (e.g., Frank (1969),
Kiger (1974), Schiff (1978), Collins, Hopwood and McKeown (1984), Lorek
and Bathke (1984), Bathke, Lorek and Willinger (1989)).

5. Factor (3) serves as a proxy for factors (2) and (4). Factor (5) was
not included because CS and other variables are aligned with industry
membership. Factor (7) is implicit in the design which treats each year
separately. Factor (8) was not incorporated in design since Value Line
data were used exclusively. In addition, number of lines of business
and exchange listing were examined in a pilot study but dropped from
further consideration when results indicated no effects.

6. The percentage of truncations was less than 2.26 percent (114 out of
5,061 observations). Influence diagnostics indicated that the results
were not driven by these or any other observations .

7. Recently, Dugan, Gentry and Shriver (1985) suggested that such measures
might provide insights within an auditing context.

8. Pearson product moment correlation coefficients were significant for
DSS-DSE (averaged .626) and DSE-VLPI (averaged .382). All other
pairings were insignificant.

REFERENCES

Albrecht, W.S., 0. Johnson, L. Lookabill, and D.H. Watson. "A Comparison of
the Accuracy of Corporate and Security Analysts' Forecasts of Earnings:
A Comment." Accounting Review 52 (July 1977) : 736-740.

American Institute of Certified Public Accountants. "Interim Financial
Reporting." APB Opinion No. 28 (AICPA, 1973).

Backer, M. Financial Reporting for Security Investment and Credit Decisions
(National Association of Accountants, New York, 1970).

Baldwin, B.A., "Segment Earnings Disclosure and the Ability of Security

Analysts to Forecast Earnings Per Share." Accounting Review 59 (July
1984): 376-389.

Bamber.L.S. "Unexpected Earnings, Firm Size, and Trading Volume Around

Quarterly Earnings Announcements." Accounting Review 62 (July 1987):
510-532.

Bathke, A.W.,Jr. and K.S. Lorek. "The Relationship Between Time-Series Models
and the Security Market's Expectation of Quarterly Earnings,"
Accounting Review 59 (April 1984): 163-176.

Bathke, A.W.,Jr., K.S. Lorek and G.L. Willinger. "Firm-Size and the Predictive
Ability of Quarterly Earnings Data." Accounting Review 64 (January
1989): 49-68.

Barefield, R. and E. Comiskey. "The Accuracy of Analysts Forecasts of Earnings
per Share." Journal of Business Research 13 (July 1975): 241-252.

Biddle, G.C. and W.E. Ricks. "Analyst Forecast Errors and Stock Price Behavior
Near the Earnings Announcement Dates of LIFO Adopters." Journal of
Accounting Research 26 (Autumn 1988): 169-194.

Blough, C.G. "Some of the Dangers Inherent in Quarterly Financial Statements."
Journal of Accountancy 95 (February 1953): 221-222.

Bollom, W.J. "Towards A Theory of Interim Reporting for A Seasonal Business: A
Behavioral Approach." Accounting Review 48 (January 1973): 12-22.

Bollom, W.J. and J.J. Weygandt . "An Examination of Some Interim Reporting

Theories for a Seasonal Business." Accounting Review 47 (^January 1972):
75-84.

Brown, L.D., CD. Richardson and S.J. Schwager. "An Information J

Interpretation of Financial Analyst Superiority in Forecasting
Earnings." Journal of Accounting Research 25 (Spring 1987) 49-67.

Brown, L.D. and M.S. Rozeff. "Univariate Time -Series Models of Quarterly

Earnings per Share: A Proposed Premier Model." Journal of Accounting
Research 17 (Spring 1979): 179-89.

Brown, P., G. Foster and E. Noreen. Security Analyst Multi-Year Earnings
Forecasts and the Capital Market (American Accounting Association,
1985).

Burrowes, A.W. , "A Survey of Sophisticated New Zealand Users of Financial
Statements and Their Interim Reporting Requirements," Ph.D.
Dissertation, University of Nebraska-Lincoln, 1986.

Capon, F.S., "The Need for Interim Statements," Canadian Chartered Accountant
(October 1955): 277-82.

Carlson, T.E. "Needed: A New Interpretation for Interim Reporting." Management
Accounting (January 1978): 19-22, 38.

Coates , R. "The Predictive Content of Interim Reports- -A Time Series

Analysis." Journal of Accounting Research 10 (Supplement 1972):
132-144.

Collins, W.A. , W.S. Hopwood and J.C. McKeown. "The Predictability of Interim
Earnings Over Alternative Quarters." Journal of Accounting Research 22
(Autumn 1984) : 467-79.

Crichfield, T. , T. Dyckman and J. Lakonishok. "An Evaluation of Security
Analysts' Forecasts." Accounting Review (July 1978): 651-668.

Doktor, R.H. and S.M. Chandler. "Limits of Predictability in Forecasting in

the Behavioral Sciences." International Journal of Forecasting 4 (1988):
5-14.

Dugan, M.T., J. A. Gentry, and K.A. Shriver. "The X-ll Model: A New Analytical
Review Technique for the Auditor." Auditing: A Journal of Practice &
Theory 4 (Spring 1985): 11-22.

Edwards, J.W., G.F. Dominiak, and T.V. Hedges. Interim Financial Reporting
(National Association of Accountants, New York, 1972).

Eggleton, I.R.C. "Patterns, Prototypes and Predictions: An Exploratory

Study." Journal of Accounting Research 14 (Supplement 1976): 68-131.

Eggleton, I.R.C, "Intuitive Time-Series Extrapolation." Journal of
Accounting Research 20 (Spring 1982): 68-102.

Financial Accounting Standards Board. Interim Financial Accounting and
Reporting: Discussion Memoradum. (FASB, May 25, 1978).

Foster, G. "Quarterly Accounting Data: Time-Series Properties and Predictive
Ability Results." Accounting Review 52 (1977): 1-21.

Foster, G. Financial Statement Analysis (Prentice-Hall: Englewood Cliffs,
N.J., 1986).

Frank, W. "A Comparison and Evaluation of Three Interim Income Concepts."
Social Science Quarterly (March 1969): 864-875.

Fried, D. and J. Livnat. "Interim Statements: An Analytical Examination of
Alternative Accounting Techniques." Accounting Review 56 (July 1981):
493-509.

Green, D. "Towards a Theory of Interim Reports." Journal of Accounting
Research 2 (Spring 1964): 35-49.

Griffin, P. A. "The Time Series Behavior of Quarterly Earnings: Preliminary
Evidence." Journal of Accounting Research 15 (Spring 1977): 71-83.

Kiger, J.E. "Volatility in Quarterly Accounting Data." Accounting Review 49
(January 1974) : 1-7.

Leftwich, R.W. , R.L. Watts and J.L. Zimmerman. "Voluntary Corporate

Disclosure: The Case of Interim Reporting." Journal of Accounting
Research 19 (Supplement 1981): 50-77.

Lorek, K.S. and A.W. Bathke , Jr. "A Time -Series Analysis of Nonseasonal

Quarterly Earnings Data." Journal of Accounting Research 22 (Spring
1984): 369-379.

Lys , T. and K. Sivaramakrishnan. "Earnings Expectations and Capital

Restructuring: The Case of Equity-for-Debt Swaps." Journal of Accounting
Research 26 (Autumn 1988): 273-299.

Maingot, M. "Published Interim Reports in the United Kingdom." International
Journal of Accounting 18 (Spring 1983): 133-149.

Mendenhall, R.R. and W.D. Nichols. "Bad News and Differential Market Reactions
to Announcements of Earlier-Quarters Versus Fourth-Quarter Earnings."
Journal of Accounting Research 26 (Supplement 1988): 63-86.

Nickerson, C.A., L.G. Pointer, and R.H. Strawser. "Attitudes of Financial

Executives Toward Interim Financial Reporting." CPA Journal 45 (March
1975): 21-24.

O'Brien, P.C. "Analysts' Forecasts As Earnings Expectations." Journal of
Accounting and Economics 10 (1988): 53-83.

Pincus, M. "Information Characteristics of Earnings Announcements and Stock
Market Behavior." Journal of Accounting Research 21 (Spring 1983):
155-83.

Rappaport, A. "Towards a Theory of Interim Reports: Modification and

Extension." Journal of Accounting Research 4 (Spring 1966): 121-126.

Reilly, F.K. , D.L. Morgenson and M. West. "The Predictive Ability of
Alternative Parts of Interim Financial Statements." Journal of
Accounting Research 12 (Supplement 1972): 105-124.

Ricks, W.E. and J.S. Hughes. "Market Reactions to a Non-Discretionary

Accounting Change: The Case of Long-Term Investments." Accounting
Review 60 (January 1985): 33-52.

Securities and Exchange Commission. "Notice of the Adoption of Amendments to
Form 10 -Q and Regulation S-X Regarding Interim Reporting." Accounting
Series Release No. 177 (SEC, 1975).

Schiff, M. Accounting Reporting Problems: Interim Financial Statements.
(Financial Executives Research Foundation, New York, 1978) .

Shillinglaw, G. "Concepts Underlying Interim Financial Statements."
Accounting Review 36 (April 1961): 222-231.

Shiskin, J., A. Young and J. Musgrave. "The X-ll Variant of the Census Method
II Seasonal Adjustment Program." Technical Paper No. 15, United States
Department of Commerce, February 1967.

Taylor, R.G. "A Look at Published Interim Reports." Accounting Review 40
(1965): 89-96.

Watts, R. "The Time Series Behavior of Quarterly Earnings." Working Paper,
University of Newcastle, April 1975.

Zellner, A. "An Efficient Method of Estimating Seemingly Unrelated Regressions
and Tests for Aggregation Bias." Journal of the American Satistical
Association 57 (June 1962): 348-368.

Table 1
Seasonality Profile of Sample

No. of Average

SIC Industry

20 Food and kindred products

27 Printing and publishing
49 Electric, gas, and santitary services

36 Electrical equipment and supplies

37 Transportation equipment
35 Machinery, except electrical

28 Chemicals and allied products
32 Stone, clay and glass products
10 Metal mining
26 Paper and allied products
48 Communications

All other (35 2-digit codes)
Total Sample

Firms

DSS%

DSE%

8

78.6

79.2

14

76.0

77.3

18

73.5

71.5

10

67.7

48.3

8

62.1

54.2

10

60.4

45.3

19

59.1

58.3

8

56.0

55.3

5

45.5

25.4

10

38.4

40.0

5

27.4

46.3

82

52.6

52.8

197

58.0

55.9

Table 2

Means and Standard Deviations of Variables (1980-86)

01

Mean (SD)

_Q2_

Mean (SD)

m.

Mean (SD)

j24

Mean (SD)

LFE

DSS%
DSE%
VLPI
SIZE
TIME

FE%
SALESa
FDATE"
ADATEb

2.883 (1.299) 2.914 (1.265) 3.042 (1.245) 3.248 (1.242)

58.608 (26.261)
57.551 (26.629)
58.460 (26.959)
6.947 (1.323)

58.221 (26.479)
56.812 (27.114)
57.460 (27.038)
6.959 (1.280)

57.841 (26.145)
56.040 (27.742)
56.291 (27.232)
6.945 (1.306)

57.662 (26.261)

56.168 (27.076)

55.869 (27.799)

6.905 (1.303)

71.945 (27.544) 162.621 (27.531) 253.960 (26.972) 360.698 (30.040)

38.520

2.556

41.484

113.429

(58.569)
(4.889)

(26.852)
(8.044)

38.288

2.345

41.854

204.475

(56.492)
(3.750)

(26.620)
(7.833)

42.172

2.505

42.188

296.148

(59
(4

(26
(8

026)
786)
289)
149)

51

2

41

402

032
394
560
258

(66.319)

(4.618)

(26.509)

(14.626)

Firms

n

174

n

173

n

183

n

193

a In $ billions
b Julian date

Note: LFE = Ln Forecast Error, DSS = Degree of Seasonality in Sales,
DSE = Degree of Seasonality in Earnings, VLPI = Value Line Earnings
Predictability Index, SIZE = LnSALES , TIME = ADATE - FDATE ,
FE = EPS Forecast Error, SALES = Annual Sales, FDATE = Forecast Date,
ADATE = EPS Announcement Date .

Table 3
Regression Coefficients and Significance Levels: Model 1

LFE, = (30 + jS.DSSj + /32VLPI, + /33SIZE, + 0/TIME, + €i

First Quarter (n - 174 per year; system R2 = .206)

Year Intercept

DSS

VLPI

SIZE

TIME

1980

3.2159

(.000)

.0021

(.513)

-.0208

(.000)

.0395

(.737)

.0069 (.011)

1981

4.1500

(.000)

.0049

(.107)

-.0200

(.000)

-.1857

(.001)

.0114 (.000)

1982

4.4022

(.000)

.0029

(.415)

-.0230

(.000)

-.0926

(.101)

.0056 (.048)

1983

4.0064

(.000)

-.0013

(.711)

-.0216

(.000)

-.0234

(.374)

.0064 (.031)

1984

4.1707

(.000)

.0066

(.022)b

-.0261

(.000)

-.0026

(.481)

-.0027 (.839)

1985

4.0328

(.000)

.0049

(.148)

-.0155

(.000)

- . 1446

(.015)

.0077 (.009)

1986

4.4015

(.000)

-.0012

(.724)

-.0227

(.000)

-.0751

(.130)

.0048 (.062)

Second Quarter (n = 173 per year; system R2 = .184)

Year

Intercept

DSS

VLPI

SIZE

TIME

1980

3,

.6021 (.000)

.0004 (

.900)

-.0180 (

.000)

.0339 (,

.717)

.0012 (

.337)

1981

3,

.2582 (.000)

.0047 (

.154)

-.0193 (

.000)

-.0324 (

.318)

.0035 (

.138)

1982

4,

.3076 (.000)

-.0033 (

.302)

-.0241 (

.000)

-.0309 (

.319)

.0042 (

.086)

1983

4

.9415 (.000)

-.0007 (

.819)

-.0257 (

.000)

-.0478 (

,220)

-.0004 (

.552)

1984

4

.1632 (.000)

.0008 (

.806)

-.0201 (

.000)

-.0281 (

.333)

-.0010 (

.632)

1985

4,

.2559 (.000)

.0014 (

.679)

-.0149 (

.000)

-.0421 (

.277)

-.0013 (

.652)

1986

3

.6835 (.000)

-.0031 (

.387)

-.0263 (

.000)

-.0566 (

.224)

.0067 (

.023)

Third Quarter (n = 183 per year; system R2 =- .146)

Year Intercept DSS VLPI SIZE

TIME

1980

2

,4684

(

.003)

.0044

(

.163)

-.0184

(

.000)

.0626

(

.846)

.0029

(,

.157)

1981

3

.4308

(

.000)

-.0016

(

.610)

-.0210

(

.000)

.0146

(

.593)

.0034

(■

.125)

1982

3,

.8710

(

.000)

- .0084

(

.006)a

-.0185

(

.000)

.0137

(

.587)

.0040

(

.091)

1983

3,

.6073

(

.000)

.0023

(

.477)

-.0192

(

.000)

-.0623

(

.173)

.0031

(

.166)

1984

5

.9504

(

.000)

-.0033

(

.314)

-.0202

(

.000)

-.0485

(

.224)

-.0058

(

.969)

1985

5

.1044

(

.000)

-.0006

(

.858)

-.0143

(

.000)

-.1025

(

.055)

-.0012

(

.652)

1986

4

.5234

(

.000)

-.0061

(

.042)b

-.0155

(

.000)

-.1536

(

.006)

.0032

(

.124)

Fourth Quarter (n = 193 per year; system R2 = .167)

Year Intercept DSS VLPI SIZE

TIME

1980

3,

,7117

(

.001)

.0031

(

.341)

-.0152

(

.000)

.0065

(

.541)

-.0004

(

.560)

1981

1,

.6066

(

.101)

.0021

(

.429)

-.0220

(

.000)

.0633

(

.875)

.0068

(

.002)

1982

3,

.4301

(

.001)

-.0052

(

.073)c

-.0214

(

.000)

.0136

(

.591)

.0044

(

.041)

1983

3,

.6626

(

.001)

-.0020

(

.512)

- .0200

(

.000)

-.0711

(

.124)

.0034

(

.098)

1984

5

.2396

(

.000)

-.0056

(

.066)c

-.0204

(

.000)

-.1090

(

.039)

.0002

(

.476)

1985

4

.3799

(

.000)

-.0029

(

.351)

-.0118

(

.000)

- . 1044

(

.050)

.0013

(

.308)

1986

4

.2139

(

.000)

- .0049

(

.084)c

-.0191

(

.000)

-.1047

(

.036)

.0032

(

.087)

1 DSS significant at .01 level (two- tailed test)

b DSS significant at .05 level (two-tailed test)

DSS significant at .10 level (two-tailed test)

Table 4
Regression Coefficients and Significance Levels: Model 2

LFE, - (30 + 0,DSE, + jS2VLPI, + /33SIZE, + 0/TIME, + €l

First Quarter (n =■ 174 per year; system R2

208)

Year Intercept

1980 3.3249 (.000)

1981 4.4347 (.000)

1982 4.4783 (.000)

1983 3.9363 (.000)

1984 4.2633 (.000)

1985 4.2935 (.000)

1986 4.3713 (.000)

Second Quarter (n ■■

Year Intercept

1980 3.7298 (.000)

1981 3.4531 (.000)

1982 4.1508 (.000)

1983 4.8940 (.000)

1984 4.1346 (.000)

1985 4.4817 (.000)

1986 3.6289 (.000)

DSE

0002 (.957)
0005 (.870)
0038 (.308)
0002 (.959)
0095 (.001)a
0028 (.406)
0015 (.671)

VLPI

.0208 (.000)

-.0197 (.000)

-.0241 (.000)

-.0216 (.000)

-.0293 (.000)

-.0162 (.000)

-.0222 (.000)

SIZE

.0416 (.746)

-.1827 (.002)

-.1013 (.918)

-.0231 (.375)

-.0113 (.418)

-.1507 (.012)

-.0722 (.139)

173 per year; system R2 = .184)
DSE VLPI

-.0035 (.274)

.0030 (.402)

-.0023 (.488)

.0000 (.988)

.0021 (.524)

-.0015 (.672)

-.0035 (.360)

-.0166 (.000)

-.0204 (.000)

-.0234 (.000)

-.0258 (.000)

-.0210 (.000)

-.0140 (.000)

-.0252 (.000)

Third Quarter (n = 183 per year; system R2 = .146)
Year Intercept DSE VLPI

1980 2.6467 (.001)

1981 3.4031 (.000)

1982 3.5925 (.000)

1983 3.7163 (.000)

1984 5.8611 (.000)

1985 5.3525 (.000)

1986 4.4992 (.000)

0032 (.341)

0020 (.541)

0081 (.013)b

0011 (.752)

0029 (.394)

0040 (.213)

0083 (.006)*

-.0192 (.000)

-.0204 (.000)

-.0165 (.000)

-.0196 (.000)

-.0197 (.000)

-.0130 (.000)

-.0134 (.000)

SIZE

0346 (.722)
0341 (.310)
0268 (.342)
0478 (.220)
0297 (.323)
0457 (.260)
0532 (.237)

SIZE

0659 (.858)

0129 (.418)

0296 (.682)

0665 (.160)

0437 (.247)

1040 (.051)

1411 (.009)

TIME

0067 (.013)

0111 (.000)

0056 (.049)

0064 (.031)

0026 (.837)

0071 (.014)

0047 (.067)

TIME

0012 (.326)

0033 (.151)

0044 (.075)

0003 (.543)

0009 (.617)

0018 (.707)

0065 (.027)

TIME

0026 (.189)

0035 (.123)

0041 (.086)

0031 (.162)

0058 (.969)

0017 (.713)

0028 (.150)

Fourth Quarter (n = 193 per year; system R2 - .164)
Year Intercept DSE VLPI

SIZE

TIME

1980

3

8337

(

.001)

.0020

(

566)

.0158

(

.000)

.0070

(

.544)

-.0005

(

.572)

1981

1

7313

(

.075)

.0001

(

973)

-.0221

(

.000)

.0649

(

.880)

.0068

(

.002)

1982

3

2553

(

.002)

-.0047

(

122)

-.0201

(

.000)

.0243

(

.658)

.0044

(

.041)

1983

3

.5433

(

.001)

-.0006

(

.837)

- .0204

(

.000)

-.0711

(

.125)

.0034

(

.099)

1984

4

8960

(

.000)

-.0014

(

.653)

-.0206

(

.000)

-.1038

(

.048)

.0003

(

.447)

1985

4

4369

(

.000)

-.0057

(

,065)c

-.0099

(

.000)

-.1008

(

.055)

.0012

(

.322)

1986

3

9784

(

.000)

-.0031

(

291)

-.0187

(

.000)

-.0982

(

.046)

.0033

(

.080)

a DSE significant at .01 level (two-tailed test)
b DSE significant at .05 level (two-tailed test)
c DSE significant at .10 level (two- tailed test)

i

t