Page 119 - 34-1

NTU Management Review Vol. 34 No. 1 Apr. 2024

After the screening process, we obtain an unbalanced final sample of 7,455
observations. The sample has 354 unique publicly-traded stock insurance companies, 225
unique privately-held stock insurance companies and 383 unique mutual insurers. The
sample period is from 1999 to 2010. The sample period stops in 2010 because 5 future

years of data are needed to estimate the loss reserve error.
Table 2 indicates the proportion of the sample using in-house actuaries to certify
loss reserves. The proportion of insurers in the sample using in-house actuaries to certify

33
loss reserves is 28% for our sample and remains stable throughout our sample period.
This ratio is 52% for publicly-traded stocks, compared to 17% for mutuals and 9% for
privately-held stocks.
Besides, it is possible that insurers systematically choose an in-house actuary. For
example, managers who are more inclined to manage earnings reporting through loss

reserve estimation (i.e., weak insurers) might tend to select an in-house actuary because
(s)he might be easier to influence. Thus, the results might be influenced by selection bias.
Therefore, we conduct Durbin-Wu-Hausman tests for endogeneity surrounding the choice

of using an in-house actuary. 34
Following Lawrence, Minutti-Meza, and Zhang (2011), we use a PSM to control
for differences in insurer characteristics for insurers using an in-house actuary for reserve
certification and for those using an external actuary (Lennox, Francis, and Wang, 2012).
PSM is a statistical tool attempting to estimate the effect of a treatment by controlling for

the variables that predict obtaining the treatment. That is, PSM matches pairs of firm-years
that have similar firm characteristics but differ in the choice of using an in-house actuary.

mutuals, publicly-traded stocks or privately-held stocks and that they be domiciled in the U.S. result
in a sample size of 11,918. Deletion of observations with extreme errors in their loss reserves then
reduces the sample to 11,741. Dropping observations for insurers that cede all premiums and/or write
greater than 25% of their premium in workers compensation, accident & health, surety, credit, and/or
reinsurance further reduces the sample size to 8001. We continuously reduce the sample size to 7,567
observations with the requirement that the sample firms have data for all of the control variables.
Finally, dropping firms that are present for only one year in the FGLS estimation leads to a final
sample of 7,455 observations.
33 The 28% is based on the number of insurers that use in-house actuaries. This is not the same as the
total dollar amount of reserves associated with in-house actuaries in the U.S., which earlier was stated
as 60% on average.
34 The results of these tests are reported on later in the paper.

111