5. Empirical Result
5.1 Empirical Result for Main Hypothesis
Table 3 through Table 5 shows the empirical results of the overall effect of customer concentration on cost stickiness. Estimates of three different variables of costs, which are SGA, COGS, and OC, are shown in Table 3 through Table 5 with three specifications, respectively. SG&A costs are most commonly used in the literature on cost stickiness since the seminal paper, Anderson et al. (2003). I report the results in Table 3.
First, Column (1) of Table 3 reports the baseline model derived from Anderson et al.
(2013) with the modification that my variable of interest CC is incorporated. Specifically, the coefficient on ∆𝑙𝑛(𝑆𝑎𝑙𝑒) is 0.57971 (t-statistic =112.28), positively significant at the 0.1% level, which is consistent with prior literature that SG&A costs are positively associated with the sales revenue. The estimated coefficient, 0.57971 indicates that SG&A costs increase 0.57 % per 1% increase in sales revenue. The coefficient on
ln(Sale)
*Dec is -0.16276 (t-statistic =-15.20), significantly negative at the 0.1% level,
which is consistent with the results in Anderson et al. (2003) that provides strong support for cost stickiness. The combined value of these two coefficients on ln(Sale) andln(Sale)
*Dec is 0.43434, indicating that SG&A costs decreases only 0.43% per 1%
decrease in sales revenue. The fact that both the coefficient on ln(Sale) and the combined value of the two coefficients are both significantly less than one, indicating that SG&A costs are not proportional to changes in sales revenue, even though SG&A cost driver should be closely related to sales. When customer concentration (CC) is incorporated in the model, the coefficient on my variable of interest, CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.17843 (t-static =5.07), significantly positive at the 0.1 % level. The result indicates
34
that for companies with more concentrated customers, SG&A costs are less sticky when sales decrease. Column (2) of Table 3, I report the results with control for economic explanations that may affect the degree of cost stickiness based on the prior literature.
Likewise, the coefficient on ∆𝑙𝑛(𝑆𝑎𝑙𝑒) is 0.59434 (t-statistic =110.25), positively significant at the 0.1% level, indicating that SG&A costs are positively associated with the sales revenue. The estimated coefficient, 0.59434 indicates that SG&A costs increase 0.59 % per 1% increase in sales revenue. The coefficient on ln(Sale)
*Dec is -0.15420
(t-statistic =-14.09), significantly negative at the 0.1% level, which provides strong support for cost stickiness. The combined value of these two coefficients on ln(Sale) and ln(Sale)*Dec is 0.44014, indicating that SG&A costs decreases only 0.44% per
1% decrease in sales revenue. When customer concentration (CC) is incorporated in the model, the coefficient on my variable of interest, CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.40746 (t-static=10.71), significantly positive at the 0.1 % level. The result provides strong support for the notion that for companies with more concentrated customers, SG&A costs are less sticky when sales decrease. Regarding the control variables, the coefficient on
GDPGrowth is 0.48971(t-static=8.19) significantly positive at the 0.1% level, which is
consistent with prior literature that macroeconomic environment is promising; firms may have sales growth and increase their investment in the production process. The coefficient on ASINT * ln(Sale) is 0.0010 (t-static=18.70), significantly positive at 0.1%, consistent with prior literature. The result provide strong support for the notion that when asset intensity is higher, the degree of cost stickiness increases since the adjustment costs tend to be higher for firm that rely more on asset owned, such as warehouse, plant, and equipment. The coefficient on EMPINT* ln(Sale) is -1.61007 (t-static=-11.60) negatively significant at 0.1% level. This result is not consistent with the notion suggested by the prior literature that when employee intensity is higher, the degree of cost stickiness35
should increases since the adjustment costs tend to be higher for firm that rely more on people employed whom are not easy to removed.
Column (3) of Table 3 reports the results with full control variables with indicator variables for industry fixed effects in order to it control for the potentially unobserved industry specific factors that are related to cost behaviors. The results are very alike.
Specifically, the coefficient on ∆𝑙𝑛(𝑆𝑎𝑙𝑒) is 0.67089 (t-statistic =38.53), positively significant at the 0.1% level, which indicates that SG&A costs are positively associated with the sales revenue. SG&A costs increase 0.67 % per 1% increase in sales revenue.
The coefficient on ln(Sale)
*Dec is -0.14354 (t-statistic =-13.14), significantly
negative at the 0.1% level, which is consistent with the presence of cost stickiness. The combine value of these two coefficients on ln(Sale) and ln(Sale)*Dec is
0.52735, indicating that SG&A costs decreases only 0.52% per 1% decrease in sales revenue. The fact that both the coefficient on ln(Sale) and the combined value of the two coefficients are both significantly less than one, indicating that SG&A costs are not proportional to changes in sales revenue, When customer concentration (CC) is incorporated in the model, the coefficient on my variable of interest,CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.38812 (t-static =10.26), significantly positive at the 0.1 %
level. The result indicates that for companies with more concentrated customers, SG&A costs are less sticky when sales decrease. The coefficients on the control variables are similar as we reported in Column (2) of Table 3.Overall, my results reported in Table 3 cross three specifications show that although consistent with prior literature that SG&A cost behavior reflect a “sticky” pattern, that SG&A as a variable component of total costs decrease less with a sales decrease than they increase with an equivalent sales increase, when companies with more concentrated customers, SG&A costs are less sticky when sales decrease. The results support the
36
operations management view that suppliers with concentrated customer bases achieve better efficiencies by mutual collaboration and information sharing.
Based on argument by Subramanian and Weidenmier (2003) that due to lack of authoritative guidance, the component of SG&A in one company could be assigned to cost of goods sold in another company, I also investigate the stickiness of cost of goods sold (GOGS) and total operating cost (OC) and report the results in Table 4 and Table 5 respectively.
Column (1) of Table 4 reports the baseline results of Anderson et al. (2003) model with the extension of my variable of interest (CC). The coefficient on ∆𝑙𝑛(𝑆𝑎𝑙𝑒) is 1.01948 (t-statistic =219.16), positively significant at the 0.1% level, which is as expected that cost of goods sold (COGS) are highly correlated to the sales revenue. The estimated coefficient, 1.01948 indicates that COGS increase almost 1 % per 1% increase in sales revenue. The coefficient on ln(Sale)
*Dec is -0.04348 (t-statistic =-4.52), significantly
negative at the 1% level, which is consistent with the notion that COGS is sticky. The combined value of these two coefficients on ln(Sale) and ln(Sale)*Dec is 0.976,
indicating that COGS decreases only 0.976 % per 1% decrease in sales revenue. When customer concentration (CC) is included in the model, the coefficient on my variable of interest, CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.29744 (t-static =9.43), significantly positive at the 0.1% level. The result suggests that for companies with more concentrated customers, COGS is less sticky when sales decrease. Column (2) of Table 4, I report the results with control variables for factors that may affect the degree of cost stickiness. Likewise, the coefficient on ∆𝑙𝑛(𝑆𝑎𝑙𝑒) is 1.00977 (t-statistic =206.52), positively significant at the 0.1% level, indicating that COGS are positively associated with the sales revenue. COGS increases nearly 1 % per 1% increase in sales revenue. The coefficient on ln(Sale)
*Dec is -0.06562 (t-statistic =-6.63), significantly negative at the 0.1% level, which
37
provides strong support for cost stickiness of COGS. The combine value of these two coefficients on ln(Sale) and ln(Sale)
*Dec is 0.94415, indicating that COGS
decreases 0.94 % per 1% decrease in sales revenue. The coefficient on my variable of interest, CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.29744 (t-static =9.43), significantly positive at the 0.1% level. The result strongly supports that for companies with more concentrated customers, COGS are less sticky when sales decrease. Regarding the control variables, the coefficients on GDPGrowth , ASINT *ln(Sale) and EMPINT*ln(Sale)are all similar with those reported in Table .
Column (3) of Table 4 reports the results with full control variables and indicator variables for industry fixed effects. The result reflects a very similar pattern: Specifically, the coefficient on ln(Sale)
*Dec is -0.06408 (t-statistic =-6.44), significantly negative
at the 1% level, which is consistent with the presence of cost stickiness. The coefficient on my variable of interest, CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.35024 (t-static =10.21), significantly positive at the 0.1 % level. The result indicates that COGS cost stickiness decrease with the level of customer concentration.I also investigate whether customer concentration has any effect on the cost stickiness behavior of operating costs. Column (3) of table 5 reports the result with full control variable. The coefficient on ∆𝑙𝑛(𝑆𝑎𝑙𝑒) is 0.85941 (t-statistic =75.77), positively significant at the 0.1% level, which is as expected that operating cost (OC) are highly correlated to the sales revenue. The estimated coefficient, 0.85941 indicates that OC increase 0.85 % per 1% increase in sales revenue. The coefficient on ln(Sale)
*Dec is
-0.03978 (t-statistic =-5.59), significantly negative at the 1% level, which is consistent with the notion that OC is sticky. The combine value of these two coefficients onln(Sale)
and ln(Sale)
*Dec is 0.820, indicating that OC decreases only 0.820 % per
1% decrease in sales revenue. When customer concentration (CC) is included in the38
model, the coefficient on my variable of interest, CC*∆𝑙𝑛(𝑆𝑎𝑙𝑒)*Dec, is 0.29645 (t-static
=12.07). The result shows that for companies with more concentrated customers, OC is less sticky when sales decrease. The control variables, the coefficients on GDPGrowth ,
ASINT *
ln(Sale) and EMPINT*ln(Sale)are all similar with those reported in Table .39