260 /302
顧客與供應商關係與成本結構
260
R&D may facilitate production, and reduce overhead costs, which in turn can affect cost
asymmetry. Following Anderson et al. (2003), we include two variables,
ASINT
and
EMPINT
, as proxies to measure the magnitude of adjustment cost.
ASINT
is asset
intensity, calculated for each firm-year observation as total asset divided by sales.
EMPINT
is employee intensity, calculated for each firm-year observation as the number of
employees divided by sales. We also include the interactions between these two variables
with Δ
ln(Sales)
. We expect the coefficients on
ASINT
and
EMPINT
to be negative because
firms with greater asset intensity have more rigid cost structure and firms with greater
employee intensity have higher adjustment cost. We also include controls for industry
fixed effect (
IndFE
) as well as interactions between industry indicators and Δ
ln(Sales)
.
5. Sample Selection and Empirical Result
5.1 Sample Selection
We identify customer-supplier relationships using the COMPUSTAT Segment File,
which includes data on customer name, type, and revenue contributed to the supplier firm.
The COMPUSTAT Segment File is based on FASB’s and SEC’s requirements that public
firms disclose revenue derived from each major customers representing more than 10
percent of their total sales. Our sample begins in 1976, which is the first year when major
customer data is available and ends in 2015. We gather other financial information about
suppliers and customers from COMPUSTAT. Following Banker et al. (2014), we restrict
our sample to manufacturing firms (four-digits SIC codes range between 2000 and 3999).
Following Irvine et al. (2016), we remove customer-supplier observations that customers
are not identified as company type since government customers are considered as low-risk
customers to the supplier firms. We eliminate customer-supplier observations that
suppliers’ sales, SG&A or operating costs are missing. We also exclude customer-supplier
observations for which SG&A costs exceed sales because these observations express
unusually large commitments of SG&A resources. To control for the potential effect of
outliers, we winsorize the data at the top and the bottom 1 percent.
5.2 Descriptive Statistic and Empirical Results
Table 1 presents the descriptive statistics of variables used in H1. The number of
observations is different among variables due to data availability, which leads to different
sample size for different variables used in the regression. In this study, although we