*This preview shows
page 1. Sign up
to
view the full content.*

**Unformatted text preview: **reened the data for missing observations of either SG&A costs or sales
revenue in the current or preceding year and deleted observations if SG&A
costs exceeded sales revenue. The total number of remaining observations
is 64,663 for 7,629 ﬁrms, an average of about 8.5 observations per ﬁrm.
Panel A of table 1 provides descriptive information about annual revenues
and SG&A costs for the complete 20-year sample. The mean value of SG&A
costs as a percentage of sales revenue is 26.41% (median = 22.62%, standard
deviation = 17.79%). Panel B of table 1 provides information about the frequency of ﬁrm-periods when revenue fell (relative to the previous period)
and ﬁrm-periods when SG&A costs fell. Revenue fell in 27.01% of the annual
ﬁrm-periods in the sample and SG&A costs fell in 24.98% of the ﬁrm-periods.
The mean value of revenue decreases is 17.45% (median value = 10.99%,
standard deviation = 18.64%) and the mean value of decreases in SG&A
costs is 15.67% (median value = 10.07%, standard deviation = 16.40%). 3.2 ESTIMATION RESULTS We estimated the model using ordinary least squares (OLS). We eliminated extreme observations from the estimation by excluding an observation if the value of any variable was in the top or bottom 0.5% of its distribution (Chen and Dixon [1972]), resulting in a reduction of 705 observations
to 63,958 observations. White’s [1980] test indicated that heteroskedasticity
was not a problem for our loglinear model. We applied the Belsley, Kuh, and
Welsch [1980] diagnostic to test for multicollinearity in the pooled estimation. None of the condition indexes exceeded 5, well below the suggested
cutoffs. We evaluated serial correlation in the data on a ﬁrm-by-ﬁrm basis.
The Durbin-Watson [1951] test statistic revealed signiﬁcant (at the 5% level)
positive autocorrelation for less than 3% of the ﬁrms, indicating that it was
not necessary to correct for serial correlation in the data.
4 Noreen and Soderstrom [1997] specify a similar model and conduct a test for asymmetric
cost behavior with respect to activity increases and decreases. Using data for hospital overhead
costs, their results provided weak evidence of asymmetric behavior (negative signs on their interaction term for activity decreases for 12 of 16 accounts, but overall not signiﬁcantly different
from zero). 54 M. C. ANDERSON, R. D. BANKER, AND S. N. JANAKIRAMAN
TABLE 1
Summary Statistics All the reported numbers are in millions of dollars. The distribution of sales revenue and
SG&A costs is for a population of 64,663 ﬁrm-year observations from 7,629 ﬁrms in the 1999
Compustat data set that satisfy the following selection criteria: no missing observations of
sales revenue (annual Compustat item #12) for the current and preceding year, no missing
observations of SG&A costs (item #189) for the current and preceding year, and no ﬁrm-years
in which SG&A costs exceeded sales revenue. In panel B, observations with a...

View
Full
Document