28 / 372
審計人員之產業專精與客戶租稅規避：中國實證研究
28
coefficient estimates from Model (4), this study constructs an inverse Mills ratio (
IMR
) and
include it as an additional explanatory variable in Model (1), (2), and (3).
Table 11 First-stage Model: Probability of Choosing Industry Expert as Auditor
Variables
(1)
Spec = IMS_D
(2)
Spec = IPS_D
Coefficient
Z-Stat
Coefficient
Z-Stat
Constant
-7.3525***
-17.21
-8.7839
-0.13
Tenure
0.0189**
2.16
0.0046
0.56
Lnauditfee
0.0562***
2.65
-0.0053
-0.27
Tax-ratio
0.5069*
1.73
0.5993*
1.72
Sales
0.2243***
5.80
0.1048***
2.93
Lnage
-0.0039
-0.92
-0.0108**
-2.16
Local
0.0297
0.59
0.1114**
2.24
Big10
1.3786***
24.66
0.2020***
4.26
Soe
0.0680
1.33
-0.0370
-0.72
Size
0.2299***
10.74
0.1621***
7.31
ROA
-0.4667
-1.48
-0.1656
-0.47
Lev
0.2815**
2.14
0.0679
0.51
CFO
0.1925
0.69
-0.1934
-0.68
IND
Controlled
Controlled
Year
Controlled
Controlled
Chi-Square
1881.74***
842.08***
Pseudo R
2
0.3810
0.2754
N
7660
7133
Note: This table presents the first-stage Probit regression result of determinants of firm choosing
industry expert as auditor. Regression (1) use industry market share (
IMS_D
) to proxy industry
expert, regression (2) use industry portfolio share (
IPS_D
) to proxy industry expert. The sample
includes 7,660 firm-year observations for the period 2008-2012 in regression (1), and 7,133 firm-
year observations in model (2). Z-statistics are presented in the column next to coefficient. ***, **,
* stand for a statistical significant level of 1%, 5%, and 10% respectively.
Table 12 reports the second-stage OLS regression results of auditor industry expertise
on tax avoidance. Consistent with the industry expertise of auditor being associated with
higher levels of tax avoidance, this study finds a negative and significant coefficient of
Spec
in the regression of
Spec
(measured by
IMS_D
) on
CETR
, and a positive and significant
coefficient of
Spec
in the regression of
Spec
(measured by
IPS_D
) on
BTD
. In addition, a
negative association between interaction variables (
Spec
*
Fee
and
Spec
*
Tenure
) and efficient
tax ratio (
BETR
and
CETR
) is found in most of these regressions, as well as a positive