255 /342
臺大管理論叢
第
27
卷第
2S
期
255
example, standard deviation of loss ratio for 2000 is calculated using loss ratios from 1996
through 2000. To examine the effect of SOX
12
, we separate the full sample into two sub-
samples: one prior to its implementation (2000-2004) and one following (2005-2010).
3.2 Methodology
We use regression analysis to examine the impact of CEO turnover on reinsurance
demand. To examine the demand for reinsurance, we use three measures: total reinsurance,
reinsurance from affiliated insurers and reinsurance from non-affiliated insurers (e.g., Wang
et al., 2008). We also use Hausman test to determine whether fixed-effects models or
random-effects models should be used since our sample is in the form of panel data. The
results of Hausman test suggest that fix effects should be used for all analyses. Specifically,
we use two-way fixed effects (both firm and year effect
13
) model.
We next discuss the endogeneity issue using Durbin-Wu-Hausman (DWH) test (a two-
stage least squares method, 2SLS) to investigate the relation among CEO turnover,
organizational structure, corporate governance, and reinsurance demand. Potential
endogenous variables are regressed against all the exogenous variables and instrumental
variables in the first stage. The instrumental variables follow previous studies including
Eisfeldt and Kuhnen (2013), Lamm-Tennant and Starks (1993) and Linck, Netter, and Yang
(2008). The instrumental variable in the first regression of CEO turnover is growth of net
written premium
14
(Eisfeldt and Kuhnen, 2013), and the instrumental variable of mutual
variable is relative size
15
(Lamm-Tennant and Starks, 1993) and growth ratio of net written
premiums. The instrumental variable of CEO/chairperson of board duality and board size is
firm age
16
(Linck et al., 2008)
17
and growth ratio of net written premiums, and the
instrumental variable of percentage of independent director on the board is ratio of net
12 We follow Ho et al. (2013) using a two-year lag period to allow the time for the implementation of SOX
to become effective in the insurance industry. According to Green (2006), SOX went into effect in 2004
for the insurance industry.
13 We thank a reviewer for this valuable comment. The year effects fail to significantly possess explanatory
powers. They do serve as control variable purpose.
14 Eisfeldt and Kuhnen (2013) discuss that CEO skills such as the ability to sale growth when using CEO
turnover events from 1992 to 2006. We use the growth ratio of net written premium in the insurance
industry as a proxy variable for sale growth.
15 Relative size is measured as the percentage of a firm’s total premiums earned relative to all firms’
premiums earned.
16 Firm age is measured as the number of years since the firm was established.
17 Linck et al. (2008) utilize a sample of 8,000 public companies of non-financial firms.