Determinants of the Demand for Reinsurance for the … Review Vol. 34 (Oct 2015), 125-138 Vincent Y....
Transcript of Determinants of the Demand for Reinsurance for the … Review Vol. 34 (Oct 2015), 125-138 Vincent Y....
Management Review
Vol. 34 (Oct 2015), 125-138
Vincent Y. Chang
Determinants of the Demand for Reinsurance
for the U.S. Property-Liability Insurance Industry:
Quantile Regression Analysis
(Received Feb 19, 2013; First Revision Dec 11, 2013; Second Revision Apr 9, 2014;
Third Revision Jun 6, 2014; Accepted Aug 7, 2014)
1. Introduction
Reinsurance is an important risk management device for primary insurers, especially when that risk leads to catastrophies such as the 921 earthquake in Taiwan, Hurricane Katrina in the U.S., and the earthquake in Sichuan province, in China. Insurers always treat re- insurance as a hedge mechanism to overcome non- diversifiable risks. In addition, reinsurance can be used to limit the insurer’s underwriting risks and reduce the probability and expected costs of potential bank- ruptcy. Thus, reinsurance is often regarded as an in- dispensable and effective risk management mechan- ism for primary insurers.
It is critical for property-liability insurers to purchase sufficient reinsurance contracts. The short- term contract characteristic forces managers to pay more attention to a firm’s liquidity. Because reinsure- ance can be treated as a substitute for firm liquidity, as well as for firm capital, property-liability insurers actually rely on reinsurance to reduce short-term fi- nancial pressures and liquidity risks. Furthermore, short-term insurance policies may generate insuffi- cient cash reserves, encouraging managers to demand further reinsurance to mitigate the impact of cata- strophic risk on the near-event horizon.
The current literature provides insightful evi- dence of factors for determining insurers’ demand for reinsurance (Carneiro and Sherris 2005; Chang and Jeng 2015; Cole and McCullough 2006; Garven and Lamm-Tennant 2003; Kader and Adams 2007; Mayers and Smith 1990; Shiu 2011; Wang et al. 2008; Yanase 2015). However, most of analyses adopt the ordinary least squares regression (OLS), or the two-stage least squares regression (2SLS) methods, which model the relationship between the covariates of firm-specific characteristics and the conditional mean of demand for reinsurance, by examining issues related to demand for reinsurance. To the best of my knowledge, rela- tively few papers have studied the determinants of an insurer’s demand for reinsurance using a quantile re- gression (QR) approach or a two-stage quantile re- gression approach (2SQR), even though QR could pro-
Vincent Y. Chang is an Assistant Professor of Depart-
ment of Insurance & Department of Finance (joint ap-
pointment), Chaoyang University of Technology. Email:
[email protected]. The author gratefully acknowl-
edges the two anonymous reviewers for their construc-
tive comments and the financial support of the Ministry
of Science and Technology (National Science Council)
(MOST (NSC) 100-2410-H-324-010). The author would
like to thank Enago (www.enago.tw) for the English lan-
guage review.
vide an alternative dimension to the study of demand analysis and contribute to the analysis of heavy-tailed data
1. Thus, this study aims to fill an important void
in the literature.
It is predicted that insurers with higher levels of reinsurance tend to be safer and experience less financial pressure and liquidity risk than insurers with lower reinsurance levels. Thus, one can predict that the insurer’s behavior, for instance, including risk be- havior and financial and operating strategies, should present a significant difference in the lower and high- er reinsurance quantiles. Thus, the QR approach natu- rally provides useful functions for examining the de- terminants of insurer’s demand for reinsurance across various quantiles, especially for insurers within the lower quantiles.
Prior studies suggest that insurer leverage and
liquidity maintenance may represent endogenous in-
fluences on demand for reinsurance (Chang and Jeng
2015; Shiu 2011)2. Therefore, in this study, the 2SQR
paradigm is implemented to correct for endogeneity
bias (Amemiya 1982; Chen and Portnoy 1996; Kim
and Muller 2004; Powell 1983)3. Instead of focusing
on the average effects of covariates on the demand
for reinsurance, this study applies the 2SQR approach
to explore the potentially differential effects across
the demand distribution. Using pooled, time-series, and
1Koenker and Bassett (1978) introduce the QR approach.
Koenker and Hallock (2001) further propose that there
are several advantages to this approach. For example,
the QR approach seizes the critical features of the data-
base used and offers a complete picture of the covariate
effect when a set of percentiles is modeled (Chang and
Tsai 2014). 2I sincerely thank the anonymous referee who proposed
that this study should provide the endogeneity test a-
mong demand for reinsurance, leverage, and liquidity. In
this study, the Hausman’s specification test indicates that
the endogeneity influences of insurer’s leverage and li-
quidity on demand for reinsurance do exist. This study de-
tails the endogeneity test in the methodology section. 3The QR estimation bias due to endogenous variables is
as a result of ignoring the dependence between the ex-
planatory variable and the unobserved error term. Since
the error term of reduced-form of the endogenous vari-
able is correlated with the structural error term of the de-
pendent variable, it can be regarded as an omitted con-
trol variable in the traditional QR model. The QR esti-
mation has large MSE (mean square errors) as the de-
gree of endogeneity is higher. Thus, this study uses 2SQR
to correct for the endogeneity bias from endogenous var-
iables.
Management Review, October 2015 126
cross-sectional data for U.S. property-liability insur- ers from 2006 to 2010, the 2SQR results show that the influence of some firm-specific characteristics, for instance liquidity and loss development, are not con- stant across quantiles, which contradicts the findings of prior studies (Chang and Jeng 2015; Cole and Mc- Cullough 2006; Hau 2006; Wang et al. 2008). The evidence presented here indicates that 2SQR analysis could provide a richer and more insightful explana- tion for the lower and higher quantiles of the rein- sureance distribution than by just using the 2SLS ap- proach. In addition, the evidence indicates that other firm-specific characteristics (e.g., leverage, profitability, organizational form, and concentrations) exert an im- pact on the consistence of results with the 2SLS ap- proach. However, the magnitude of these influences on demand for reinsurance differs significantly across various reinsurance quantiles. These findings again enhance the main argument proposed in this study.
The contributions of this study are as follows. First, this study is the first article to analyze the deter- minants of the insurer’s demand for reinsurance with- in a particular quantile of the overall demand distri- bution, and it provides more meaningful information than the traditional 2SLS approach, while also com- pensating for the inherently biased estimations of the latter model. Thus, the present study fills an important analytic gap in the literature by applying the 2SQR analysis. Second, the 2SQR approach can analyze the influence of covariates at a specific point in the re- insurance distribution, which provides more insight- ful evidence for the lower and higher quantiles of the demand distribution than the 2SLS approach. Specif- ically, this study finds that liquidity and the loss of development are not constant across various quantiles, which contradicts the findings reported in previous studies (Chang and Jeng 2015; Cole and McCullough 2006; Hau 2006; Wang et al. 2008). In addition, the magnitude of the impacts for other firm-specific char- acteristics on demand for reinsurance varies signify- cantly across the quantile range, even if the signs of empirical results are consistent with the previous lit- erature. The evidence further concurs with the main hypotheses, which are developed and presented below. Third, the insightful information obtained using the 2SQR approach provides an incentive for policyhold- ers, policy makers, and/or regulators to assess the insurer’s demand for reinsurance through these firm- specific characteristics, especially in the case of insur- ers in the lower reinsurance quantiles.
The rest of this paper is structured as follows. Section II describes the primary hypotheses and vari- ables used in this study. Section III briefly discusses the 2SQR approach. Section IV interprets the data, presents model fitness checks, and reports the em- pirical results. Section V concludes the article.
2. Hypotheses Development and Variables Description
2.1 Hypotheses Development
Note that the expectations of all exogenous vari- ables for the traditional 2SLS approach may be dis- tinct from the 2SQR approach. For instance, the hy-
potheses of bankruptcy cost, agency cost, risk bear- ing, and renting capital all propose that a higher- leverage insurer tends to increase their demand for re- insurance (Chang and Jeng 2015; Cole and McCul- lough 2006; Garven and Lamm-Tennant 2003; Shiu 2011; Wang et al. 2008). However, the impact of lev- erage on demand for reinsurance may differ between insurers in lower and higher quantiles. Since the neg- ative marginal effect of reinsurance on underwriting risk is likely to diminish in the higher reinsurance quantiles, the effect of leverage on reinsurance is like- ly to increase as the level of reinsurance increases. For each additional unit of leverage, an insurer with a high level of reinsurance will need “more” reinsure- ance to maintain its underwriting risk at a target level. In contrast, the inverse argument will be presented re- garding the lower reinsurance quantiles. For each ad- ditional unit of leverage, an insurer with a lower level of reinsurance will need “less” reinsurance to satisfy its target level of underwriting risk, because the nega- tive marginal effect of reinsurance on underwriting risk is likely to increase for the lower reinsurance quantiles.
Furthermore, the substitute hypothesis indicates that higher liquidity insurers tend to demand less re- insurance (Chang and Jeng 2015; Hau 2006). How- ever, it is predicted that the impact of liquidity main- tenance on demand for reinsurance in the lower quan- tiles also differs from demand in the higher quantiles. Likewise, since the negative marginal effect of rein- surance on underwriting risk is likely to lessen in the higher reinsurance quantiles, the effect of liquidity maintenance on reinsurance is likely to decrease in the higher quantiles. Thus, for each increment of li- quidity maintenance, an insurer with a high level of reinsurance will need “less” reinsurance to uphold its target level of underwriting risk. On the contrary, based on the argument similar to those above, to sus- tain its target level of underwriting risk, an insurer with a low level of reinsurance will need “more” reinsur- ance for each increment of liquidity maintenance in the lower reinsurance quantiles. The reason is that in- surers in the lower reinsurance quantiles encounter more financial pressures or bankruptcy costs. This forces insurers to gradually increase their demand for reinsurance, even if they possess liquid assets. Accord- ingly, it is predicted that the determinants of demand for reinsurance across the reinsurance distribution are also different for other explanatory variables
4. Conse-
quently, this study sets out to test the following hy- potheses:
Hypothesis 1: The impact of firm-specific charac-
teristics on demand for reinsurance
is different for firms at opposite ends
of the demand distribution.
Hypothesis 2: The predictions of determinants of re-
insurance differ between the 2SLS
approach and 2SQR approach.
4Overall, one can surmise that the influence of other ex-
planatory variables (e.g., organizational form, profitabi-
lity, tax issues, and loss of reserves) on demand for rein-
surance is also distinct at the lower and higher reinsure-
ance quantiles.
Reinsurance: Quantile Regression Analysis 127
The traditional analysis of demand determinants
employs the OLS/2SLS approach (Chang and Jeng
2015; Cole and McCullough 2006; Garven and Lamm-
Tennant 2003; Mayers and Smith 1990; Shiu 2011;
Wang et al. 2008; Yanase 2015). The 2SLS approach
relies on an a priori distributional assumption of the
dependent variable. In addition, the 2SLS approach
always accepts a homogeneous influence of the de-
pendent variable, which increases the estimation bias,
especially when the dependent variable is heterogene-
ous.
In contrast to the 2SLS approach, the QR/2SQR
approach, which presents a more complete picture of
the covariate effect, allows us to examine the differ-
ential effects across the demand for reinsurance dis-
tribution when a set of percentiles is modeled. Thus,
one could conduct an assessment at the tails of the de-
pendent variable by identifying the determinants sep-
arately. In addition, the 2SQR approach is robust and
less sensitive to the presence of outliers or skewed
tails. To sum up, the 2SQR approach is more effi-
ciently provide insightful information than the 2SLS
approach, especially at the tails of the reinsurance dis-
tribution. This study then hypothesizes that
Hypothesis 3: The 2SQR approach complements the
2SLS approach in terms of providing
efficient estimation and insightful in-
formation.
2.2 Description of Variables
Prior studies always adopted the reinsurance ratio as a proxy for the insurer’s demand for reinsure- ance (Chang and Jeng 2015; Cole and McCullough 2006; Garven and Lamm-Tennant 2003; Mayers and Smith 1990; Shiu 2011; Wang et al. 2008; Yanase 2015). In accordance with prior research, the insur- er’s reinsurance ratio (Reins), which is defined as (affiliated reinsurance ceded + nonaffiliated reinsure- ance ceded)/(direct business written plus reinsurance assumed), is also used.
The literature suggests that several firm-specific characteristics, such as bankruptcy, tax issues, profit- ability, concentration level, organizational form, liquid- ity risk, solvency concerns, and business mix, may affect the insurer’s demand for reinsurance (Chang and Jeng 2015; Cole and McCullough 2006; Garven and Lamm-Tennant 2003; Mayers and Smith 1990; Shiu 2011; Wang et al. 2008; Yanase 2015). Accord- ingly, these firm-specific characteristics and hypothe- ses are then introduced as follows.
With regard to bankruptcy characteristics, three variables are adopted: firm size, leverage, and loss development. In accordance with the hypotheses of bankruptcy cost, agency cost, risk bearing, and rent- ing capital, leverage is predicted to be positively re- lated to the demand for reinsurance (Chang and Jeng 2015; Cole and McCullough 2006; Garven and Lamm- Tennant, 2003; Shiu, 2011; Wang et al. 2008). It is pre- dicted that insurers writing more business relative to their surpluses should have a higher insolvency prob- ability; therefore, greater leverage tends to produce a
higher demand for reinsurance. The measurement of the insurer’s leverage is defined as the direct business written to surplus (Leverage)
5.
As suggested in previous studies (Chang and Jeng 2015; Christensen, Hoyt, and Paterson 1999; Cole and McCullough 2006; Gaver and Paterson 1999; Grace 1990; Petroni 1992; Shiu 2011; Wang et al. 2008; Weiss 1985), potential financial constraints also play an im- portant role in demand for reinsurance. Insurers with a positive loss development (i.e., under-loss reserving) will purchase more reinsurance contracts to surmount their potential financial constraints, whereas insurers will purchase less if they have a negative loss devel- opment (i.e., over loss reserving). In addition, Harring- ton and Danzon (1994) indicate that insurers may hide their underreported claim liability and capital in- adequacy by using reinsurance. To sum up, the posi- tive relationship between loss reserve and demand for reinsurance is expected. Following Cole and McCul- lough (2006), two-year loss development (2_years_ loss), which is defined as the development in esti- mated losses and loss expenses incurred 2 years be- fore the current year and prior year (scaled by poli- cyholders’ surplus), is chosen to proxy the insurer’s potential financial constraints.
Hau (2006) proposed that an insurance contract can provide additional liquidity to a nonfinancial firm, which implies that the relationship between corporate insurance demand and liquidity is substitutive. Based on his argument, this study extends his rationale to the insurance industry and argues that reinsurance con- tracts provide extra liquidity for primary insurers. In addition, insurers with more liquid assets have an in- creased ability to mitigate unexpected and/or emer- gent demand for cash, thus reducing their demand for reinsurance (Chang and Jeng 2015). Therefore, it is predicted that the insurer’s liquidity is negatively re- lated to its demand for reinsurance. The liquidity (Liq) measurement is defined as the sum of cash and in- vested assets divided by total assets.
To reduce the probability of insolvency, small insurers tend to purchase more reinsurance contracts than large insurers (Chang and Jeng 2015; Cole and McCullough 2006; Garven and Lamm-Tennant 2003; Hoyt and Khang 2000; Mayers and Smith 1990; Shiu 2011; Wang et al. 2008; Weiss and Chung 2004; Yan- ase 2015). Thus, a negative relationship between firm size and demand for reinsurance is predicted. In this study, the natural logarithm of total assets is used to measure firm size (Size).
5As the anonymous referee points out, many prior stud-
ies also use liabilities to assets as a proxy for leverage.
This study also use liabilities to assets to reexamine the
2SQR for robustness check. Overall, the results are con-
sistent to the main findings of Table 3. This study does
not tabulate those specific results here. For a consistent
comparison, in this paper we follow previous studies
(Chang and Jeng 2015; Cole and McCullough 2006;
Garven and Lamm-Tennant 2003; Shiu 2011; Wang et al.
2008) and use the ratio direct business written to surplus
to measure the insurer’s leverage. I sincerely thank the
anonymous referee for pointing out this important issue.
Management Review, October 2015 128
This study also uses tax-exempt investment in-
come relative to total investment income (Tax_ex) as
a proxy to capture the influence of expected tax liabil-
ity and/or tax-favored assets. Smith and Stulz (1985)
and Mayers and Smith (1990) indicate that insurers
can reduce their earnings volatility through reinsure-
ance purchasing, thereby reducing their expected tax
liability. Garven and Lamm-Tennant (2003) propose
that insurers can benefit from reinsurance purchasing
in terms of mitigating the costs of huge unexpected
losses and enhancing investments in tax-favored as-
sets. Thus, a positive relationship between the tax-
exempt factor and the demand for reinsurance is pre-
dicted. However, the empirical findings of Adams, Hard-
wick, and Zou (2008), Graham and Rogers (2002),
and Shiu (2011) do not support the positive influence
of tax-exempt factors on demand for reinsurance. In
addition, Adams, Hardwick, and Zou (2008) and Shiu
(2011) further propose that insurers with a higher mar-
ginal tax rates will purchase less reinsurance to de-
crease expected tax liabilities. To sum up, the ex-
pected sign between the tax-exempt factor and the de-
mand for reinsurance is mixed.
Insurers with higher profits tend to demand less
reinsurance because they have higher payment capac-
ity and cash flows to confront loss claims and finan-
cial pressures (Chang and Jeng 2015; Cole and Mc-
Cullough 2006; Mayers and Smith 1990; 2004; Pow-
ell and Sommer 2007; Shiu 2011; Wang et al. 2008).
To control for profitability impacts on demand for re-
insurance, the insurer’s profitability, that is, the in-
surer’s return on assets (ROA), is adopted, which is
defined as net income plus tax and interest expenses,
divided by total assets.
The degrees of business concentration (Bus_H)
and geographic concentration (Geo_H) also influence
the insurer’s reinsurance decision. The business Her-
findahl Index is defined as the sum of the squares of
the ratio of the dollar amount of direct business writ-
ten in a particular line of insurance to the dollar a-
mount of direct business across all 26 lines of in-
surance; in addition, the geographic Herfindahl Index
is defined as the sum of the squares of the ratio of the
dollar amount of direct business in state j to the total
amount of direct business across all states (Chang and
Jeng 2015; Cole and McCullough 2006; Kim, Mayers,
and Smith 1996; Mayers and Smith 1990; Shiu 2011;
Wang et al. 2008). The argument for risk diversify-
cation indicates that insurers with a higher concen-
tration in a given line of business, or in a given geo-
graphic area, may have higher incentives to purchase
more reinsurance. In contrast, the real services hy-
pothesis indicates that reinsurers not only provide
protection against unexpected losses but also supply
real services via specialized knowledge and econom-
ies of scale. Thus, to obtain more technical support
from reinsurers, less business- or geography-concen-
trated insurers may demand more reinsurance. In
addition, Shiu (2011) indicates that more concentrated
insurers could specialize in, and may underwrite less
volatile lines of business, or less-risky contracts and,
therefore, demand less reinsurance. In sum, the pre-
dictions between business and geographic concentra-
tions and the demand for reinsurance are undetermined.
Since the organizational form hypothesis indi-
cates that stock insurers can benefit from diversify-
cation (outside shareholders) and/or from a lower cost
of raising external capital, this suggests that stock
insurers tend to purchase less reinsurance than mutual
insurers (Mayers and Smith 1990). On the contrary,
the agency cost hypothesis proposes that insurers’ pur-
chases of more reinsurance could mitigate the under-
investment problem among residual claimholders (Ad-
ams 1996; Shiu 2011). Consistent to the agency cost
hypothesis, Adams (1996) and Shiu (2011) find that
stock insurers purchase more reinsurance than mutual
insurers. Summing up, this study proposes that the pre-
diction effect of the organizational form is mixed. The
stock dummy variable (Stock) is used to control for
the organizational form effect, which is defined to be
1 if the insurer is the stock insurer, and 0 otherwise.
To control for the systematic difference between
a single firm (nonaffiliated insurer) and a firm be-
longing to a group (affiliated insurer), the single dum-
my variable (Single) is used. It equals 1 if the insurer
is nonaffiliated, and 0 if it is affiliated. A negative
relationship between the single dummy variable and
the demand for reinsurance is expected because insur-
ers belonging to a group are able to shift profits with-
in the group and thus, reduce tax payments (Chang
and Jeng 2015; Cole and McCullough 2006; Wang et
al. 2008).
It should be noted that commercial lines (Com_
line) are essentially more risky and contain higher
claim volatility than personal lines6. Based on the risk
taking viewpoint, insurers that write more commer-
cial lines also tend to purchase more reinsurance. In
addition, a higher claims volatility pattern also forces
insurers to rely more on reinsurance contracts (Ad-
ams, Hardwick, and Zou 2008; Graham and Rogers
2002; Graham and Smith 1999). Consequently, it is
expected that insurers that write greater volumes of
commercial lines are more likely to purchase rein-
surance. The commercial lines ratio is defined as: (di-
rect premium written of fire, allied lines, farm owners
multiple peril, commercial multiple peril, mortgage
guaranty, ocean marine, inland marine, financial guar-
anty, medical malpractice-occurrence, medical mal-
practice-claims made, earthquake, group accident and
health, credit accident and health (group and individ-
ual), other accident and health, workers’ compensa-
tion, other liability -occurrence, other liability-claims
made, products liability-occurrence, products liability-
claims made, commercial auto liability, aircraft (all
perils), fidelity, surety, burglary and theft, boiler and
machinery, and credit), divided by direct business
written.
6I sincerely thank the anonymous referee who proposed
that the underwriting portfolio has been considered to have
an effect on demand for reinsurance. Thus, this study
uses a commercial lines ratio to control underwriting mix
effects on the insurer’s demand for reinsurance.
Reinsurance: Quantile Regression Analysis 129
Insolvency concerns also have an important in-
fluence on demand for reinsurance7. Chen, Hamwi, and
Hudson (2001) and Shiu (2011) point out that insur-
ers with a higher probability of insolvency tend to
purchase more reinsurance. The reason is that an in-
solvent insurer with higher levels of reinsurance is
better positioned to confront higher external financial
costs in the capital markets. This study adopts the
natural logarithm of the RBC ratio (LnRBC) to proxy
an insurer’s degree of insolvency. The RBC ratio is
defined as (Total Adjusted Capital *100)/(2*Author-
ized Control Level), where the items are collected
from the five-year historical annual statement from
the National Association of Insurance Commissioners
(NAIC).
The literature also suggests that the variation
effects among the different lines of business may in-
fluence the insurer’s reinsurance decisions, because
some lines of business may have particular effects on
their demand for reinsurance; for instance, liability
related lines, in which higher agency costs give the
insurer greater incentive to purchase more reinsure-
ance. Consequently, this article also uses premiums
written in each line of business (W(i))8 to control for
the impact of variations in lines of business on de-
mand for reinsurance (Chang and Jeng 2015; Cole
and McCullough 2006; Mayers and Smith 1990; Shiu
2011; Shortridge and Avila 2004; Wang et al. 2008).
Finally, year dummies are also included to control
time-series heterogeneous effects as a whole.
3. Methodology
Based on the logic of Cummins and Sommer (1996)
and Baranoff and Sager (2002; 2003), Shiu (2011) con-
cludes that the insurer’s leverage has endogenous
influence in determining the insurer’s demand for
reinsurance. Thus, insurers tend to have a target un-
derwriting risk as a result to determine their demand
for reinsurance and leverage simultaneously. Chang
and Jeng (2015) further propose that liquidity main-
tenance also produces an important endogenous effect
on demand for reinsurance. Likewise, for approach-
ing underwriting risk targets, insurers determine their
demand for reinsurance, liquidity, and leverage en-
dogenously and simultaneously9. According to the ar-
7I sincerely thank the anonymous referee for considering
the solvency concerns that affect the insurer’s reinsure-
ance decision. 8The variable is defined as the ratios of premiums writ-
ten in each line of business to premiums written in all 27
lines of business. To avoid a singular matrix in the re-
gression, Mayers and Smith (1990), Cole and McCul-
lough (2006), Wang et al. (2008), and Chang and Jeng
(2015) suggest the model should exclude the financial
guaranty, international, and reinsurance lines. 9Accordingly, this study further implements a Hausman
endogeneity test to confirm whether the insurer’s lev-
erage and liquidity are endogenous with its demand for
reinsurance. First, the explanatory variables of the lev-
erage equation are identified according to Cummins and
Sommer (1996), Baranoff and Sager (2002; 2003), Shiu
gument of Chang and Jeng (2015) and Shiu (2011),
the 2SQR paradigm is performed to correct the en-
dogeneity bias, by replacing the endogenous regres-
sors with their fitted values of the reduced-form equa-
tions in this study (Amemiya 1982; Chen and Portnoy
1996; Kim and Muller 2004; Powell 1983). The tradi-
tional 2SLS approach only captures the average ef-
fects of covariates on demand for reinsurance. How-
ever, the 2SQR approach not only explores poten-
tially differential effects across the reinsurance dis-
tribution but also provides further insightful informa-
tion on how the insurer’s demand for reinsurance is
determined. Therefore, this study intends to comple-
ment the literature gap in terms of the 2SQR approach.
The 2SQR estimation procedure is introduced
as follows. Consider a liner model describes as:
1 0 2 1 1 2Y Y Z , (1)
1202 ZY , (2)
where is the dependent variable of interest, is
the endogenous regressor, is an exogenous vector
of , is the exogenous vector of , is the un-
observed structural error term of , and is the
disturbance terms of . Let '21,,1 ZZZ be a vec-
tor of all exogenous variables. The endogeneity of
is due to dependence between and , conditional
on .
In the first stage, the reduced-form equations
are set as10
:
VZY 2 (3)
From equation (3), the fitted values could be gener-
ated, i.e., 2ˆ ˆY Z .
(2011), and Chang and Jeng (2015). These variables in-
clude demand for reinsurance, liquidity, firm size, busi-
ness and geographic Herfindahl indices, single firm dum-
my, intra-group Herfindahl Index, New York-licensed
dummy, marketing channel dummy, organizational form
dummy, return on capital, natural logarithm of the RBC
ratio, and year dummies. The explanatory variables that
are used in the liquidity equation are obtained from the
existing literature (Bruinshoofd and Kool 2002; Chang
and Tsai 2014; Chang and Jeng 2015; John 1993; Kim,
Mauer, and Sherman 1998; Opler et al. 1999; Shiu 2006).
These variables contain demand for reinsurance, lev-
erage, business and geographic Herfindahl indices, 2
years’ loss development, firm size, stock dummy, single
firm dummy, return spread, firm growth rate, cash flow
volatility, reserves of liability and property lines, claims
of liability and property lines, and year dummies. The
result of Hausman’s specification test (F value = 423,
rejecting the null hypothesis) indicates that the endog-
eneity influences of the insurer’s leverage and liquidity
do exist. 10Instrument variables for the reduced-form equations
consist of all exogenous, control, and year dummy vari-
ables of demand for reinsurance, liquidity, and leverage
equations.
1Z
1Y 2Z2Y
1Y
2Y
2Y
Z
1Y 2Y
Management Review, October 2015 130
Table 1 Summary Statistics of Numerical Variables Variables Minimum Mean Median Maximum Std Dev
Reins 0.0000 0.3878 0.3449 0.9632 0.2903
Leverage 0.0419 1.7788 1.1542 12.4190 2.0024
2_years_loss -0.0005 -0.0000a -0.0000b 0.0004 0.0001
Size 14.5017 18.3070 18.2111 22.9579 1.8366
Liq 0.2827 0.8284 0.8621 0.9927 0.1350
Tax_ex 0.0000 0.2899 0.2301 0.9566 0.2625
ROA -0.1631 0.0395 0.0408 0.2296 0.0592
Bus_H 0.1403 0.5796 0.5122 1.0000 0.3033
Geo_H 0.0420 0.5684 0.5272 1.0000 0.3857
Stock 0.0000 0.6781 1.0000 1.0000 0.4672
Single 0.0000 0.4093 0.0000 1.0000 0.4918
Com_line 0.0000 0.6310 0.8383 1.0000 0.3990
LnRBC 0.2425 2.1947 2.1403 4.7270 0.7673
Obs. 5,626
The dependent variable is Reins, which is defined as (affiliated reinsurance ceded + nonaffiliated reinsurance ceded)/
(direct business written plus reinsurance assumed). Leverage it is defined as the ratio of direct business written-to-
surplus. The variable of 2_years_loss is two-year loss development and equals the development of estimated losses
and loss expenses incurred two years before the current and prior year, scaled by policyholders’ surpluses. The proxy
of firm size is Size, and is defined as the natural logarithm of total assets. The Liq is defined as the sum of cash plus
invested assets-to-total assets. The variable of Tax_ex is defined as tax-exempt investment income relative to total
investment income. And ROA is defined as net income plus tax and interest expense, divided by total assets. The
Bus_H variable is the line of business Herfindahl Index, which is defined as the sum of the squares of the ratio of the
dollar amount of direct business written in a particular line of insurance to the dollar amount of direct business across
all 27 lines of insurance. Conversely, the Geo_H variable is the geographic Herfindahl Index, which is defined as the
sum of the squares of the ratio of the dollar amount of direct business in state j to the total amount of direct business
across all states. The organizational form dummy variable is Stock, which is to indicate stock or mutuals. It equals 1
if the insurer is a stock, and 0 if it is a mutual. The single dummy variable is Single, which is to indicate an affiliated
or nonaffiliated insure. It equals 1 if the insurer is nonaffiliated, and 0 if it is affiliated. Com_line represents the
proportion of the insurer’s commercial lines of business. Finally, LnRBC is defined as the natural logarithm of RBC
ratio. In this study, total firm-year observations are 5,626.
Note a: The mean value of 2_years_loss equals to −4.6591*10−5 rather than 0.
Note b: The median value of 2_years_loss equals to −3.2263*10−5 rather than 0.
In the second stage, in equation (1) is re-
placed by ; then, the 2SQR estimator of is
the solution to the following minimization problem:
(4)
To examine the distinct influences on deter-
minants of the insurer’s demand for reinsurance be-
tween higher and lower demand for reinsurance quan-
tiles, the equality test of estimated parameters of each
variable across quantiles will be implemented. In ad-
dition, diagnostic tests of model fitness are also con-
ducted to test whether the distribution of the distances
of demand for reinsurance is normal and whether the
quantile regression fits the data appropriately11
.
4. Data and Empirical Results
4.1 Data
An analysis is performed using data from pro-
perty-liability insurance companies in the NAIC for
the years 2006-2010, which originally comprised 3,007
11This study provides the Quantile-Quantile probability
plots and a histogram of the standardized residuals to
examine the model fitness issue.
insurers. To be included in the sample, each company
needed a complete data set for each single year. Thus,
insurers with missing data were deleted, and 2,060
insurers remained. Companies with unreasonable val-
ues (or illogical values) for all variables were also
excluded from the empirical analysis12
. Subsequently,
1,869 insurers remained. Insurers that operated as pro-
fessional reinsurers and whose reinsurance accounts
for more than 75% of their total premium written
were also excluded (Chang and Jeng 2015; Cole and
McCullough 2006; Powell and Sommer 2007; Shiu
2011). Finally, the exclusions left this study with com-
plete data for 1,773 insurers and 5,626 firm-year ob-
servations. In addition, to control for the outlier prob-
lem, the variables used in this study were winsorized
at the 1st and 99th percentiles, except for the dummy
variables, to avoid the influences of extreme values.
Summary statistics for all variables of the pool-
ed time-series and cross-sectional data are shown in
Table 113
. The mean value of reinsurance ratio (Reins)
is 38.78%, with 29.03% standard deviation. The aver-
age liquidity (Liq) is 82.84%, which proposes that in-
surers are conservative, and ranges from a minimum
12For instance, the reinsurance ratio <0 and >1, leverage
<0, the geographic and business Herfindahl Index >1
and/or <0. 13Table 1 does not report the numbers of other control
variables and year dummies.
1 0 2 1 21
1 0 2 1 1 2
1 0 2 1 1 2
1 0 2 1 1 21
K ˆR Y Y Z
ˆY Y Z
ˆMin Y Y Z
ˆ ( ) Y Y Z
2Y
2Y ˆ ( )
Tab
le 2
D
escri
pti
on
of
Each
Rein
su
ran
ce D
ecil
e
Des
crip
tion
of
ind
epen
den
t v
ari
ab
les
for
each
rei
nsu
ra
nce
dec
ile (
10
deci
les)
Qu
an
tile
(by
Rei
ns)
R
ein
s L
ever
ag
e 2
_ye
ars
_lo
ss
Siz
e L
iq
Ta
x_
ex
RO
A
Bu
s_H
G
eo_
H
Sto
ck
Sin
gle
C
om
_li
ne
Ln
RB
C
Dec
ile
1
0.0
04
13
7
1.1
18
06
0
-0.0
00
058
17
.24
56
66
0.8
76
40
3
0.2
52
01
8
0.0
45
07
2
0.8
02
13
5
0.7
12
64
1
0.6
72
59
8
0.6
92
17
1
0.6
12
62
6
2.1
74
68
9
Dec
ile
2
0.0
62
20
8
1.0
81
46
4
-0.0
00
060
18
.87
12
30
0.8
75
57
5
0.3
39
53
4
0.0
47
72
0
0.6
09
42
3
0.6
27
61
2
0.5
59
50
3
0.5
15
09
8
0.5
95
03
2
2.2
37
08
5
Dec
ile
3
0.1
28
99
2
1.0
66
56
0
-0.0
00
068
18
.20
30
86
0.8
78
65
2
0.2
88
59
0
0.0
44
93
7
0.5
92
17
2
0.6
48
27
0
0.4
84
90
2
0.6
07
46
0
0.6
83
69
4
2.2
56
22
6
Dec
ile
4
0.2
04
42
3
1.1
33
03
6
-0.0
00
054
18
.27
68
56
0.8
68
61
8
0.2
71
06
4
0.0
38
91
6
0.5
66
24
3
0.6
30
89
0
0.4
77
79
8
0.5
91
47
4
0.7
24
01
5
2.1
74
00
1
Dec
ile
5
0.2
96
46
3
1.0
57
48
2
-0.0
00
052
18
.35
01
09
0.8
53
70
8
0.2
79
92
5
0.0
40
73
2
0.5
63
62
0
0.5
89
53
0
0.6
02
13
1
0.4
81
35
0
0.7
14
53
7
2.1
18
12
7
Dec
ile
6
0.3
97
60
2
1.2
08
59
3
-0.0
00
054
18
.38
62
95
0.8
44
43
4
0.2
76
41
2
0.0
38
23
9
0.5
60
10
2
0.5
70
56
0
0.7
14
03
2
0.3
73
00
2
0.6
83
20
3
2.1
12
46
3
Dec
ile
7
0.5
0112
8
1.3
53
57
5
-0.0
00
042
18
.56
23
12
0.8
24
77
3
0.2
83
95
5
0.0
34
42
7
0.5
47
62
9
0.5
27
98
2
0.7
81
52
8
0.2
75
311
0.5
85
21
5
2.0
97
26
6
Dec
ile
8
0.6
18
94
8
1.9
67
29
6
-0.0
00
041
18
.62
94
98
0.7
78
79
8
0.3
14
13
1
0.0
33
53
6
0.5
16
05
3
0.4
80
24
8
0.8
20
60
4
0.2
18
47
2
0.5
53
05
7
2.0
81
01
5
Dec
ile
9
0.7
62
80
7
2.8
56
31
3
-0.0
00
024
18
.38
01
66
0.7
61
911
0.2
90
28
1
0.0
38
18
8
0.5
21
43
0
0.4
67
96
1
0.8
24
15
6
0.2
04
26
3
0.5
94
05
2
2.1
53
74
8
Dec
ile
10
0
.90
307
8
4.9
61
83
7
-0.0
00
014
18
.16
22
35
0.7
2110
7
0.3
03
21
9
0.0
33
55
5
0.5
17
64
3
0.4
27
94
8
0.8
44
64
3
0.1
33
92
9
0.5
64
13
2.5
44
63
4
This
tab
le p
rese
nts
the
mea
ns
of
each
rei
nsu
rance
dec
ile
for
all
var
iable
s. T
he
dec
ile
1 i
s th
e gro
up o
f th
e lo
wes
t R
ein
s gro
up,
wher
eas
dec
ile
10 i
s th
e hig
hes
t R
ein
s gro
up.
The
Rei
ns
is
def
ined
as
(aff
ilia
ted r
einsu
rance
ced
ed +
nonaf
fili
ated
rei
nsu
rance
ceded
)/(d
irec
t busi
nes
s w
ritt
en p
lus
rein
sura
nce
ass
um
ed).
Conce
rnin
g a
bout
Lev
erage,
it
is d
efin
ed a
s th
e ra
tio o
f dir
ect
busi
nes
s w
ritt
en-t
o-s
urp
lus.
The
var
iable
of
2_ye
ars
_lo
ss i
s tw
o-y
ear
loss
dev
elopm
ent
and e
qual
s th
e dev
elopm
ent
of
esti
mat
ed l
oss
es a
nd l
oss
expen
ses
incu
rred
tw
o y
ears
bef
ore
the
curr
ent
and p
rior
yea
r, s
cale
d b
y p
oli
cyhold
ers’
surp
luse
s. T
he
pro
xy
of
firm
siz
e is
Siz
e, a
nd i
t is
def
ined
as
nat
ura
l lo
gar
ithm
of
tota
l as
sets
. T
he
Liq
is
def
ined
as
the
sum
of
cash
plu
s in
ves
ted
asse
ts-t
o-t
ota
l as
sets
. T
he
var
iable
of
Tax_ex
is
def
ined
as
tax
-exem
pt
inves
tmen
t in
com
e re
lati
ve
to t
ota
l in
ves
tmen
t in
com
e.
And R
OA
is
def
ined
as
net
inco
me
plu
s ta
x a
nd i
nte
rest
expen
se,
div
ided
by
tota
l as
sets
. T
he
Bu
s_H
var
iable
is
the
line
of
busi
nes
s H
erfi
ndah
l In
dex
, w
hic
h i
s def
ined
as
the
sum
of
the
squar
es o
f th
e ra
tio o
f th
e doll
ar a
mount
of
dir
ect
busi
nes
s w
ritt
en i
n a
par
ticu
lar
line
of
insu
rance
to t
he
doll
ar a
mount
of
dir
ect
busi
nes
s ac
ross
all
27 l
ines
of
insu
rance
. C
onver
sely
, th
e G
eo_H
var
iab
le i
s th
e geo
gra
phic
Her
findah
l In
dex
, w
hic
h i
s def
ined
as
the
sum
of
the
squar
es o
f th
e ra
tio o
f th
e doll
ar a
moun
t of
dir
ect
busi
nes
s in
sta
te j
to t
he
tota
l am
ount
of
dir
ect
busi
nes
s ac
ross
all
sta
tes.
The
org
aniz
atio
nal
form
dum
my
var
iable
is
Sto
ck
, w
hic
h
is t
o i
ndic
ate
stock
or
mutu
als.
It
equal
s 1 i
f th
e in
sure
r is
a s
tock
, an
d 0
if
it i
s a
mutu
al.
The
single
dum
my
var
iable
is
Sin
gle
, w
hic
h i
s to
indic
ate
an a
ffil
iate
d o
r nonaf
fili
ated
insu
re.
It e
qual
s
1 i
f th
e in
sure
r is
nonaf
fili
ated
, an
d 0
if
it i
s af
fili
ated
. C
om
_li
ne
repre
sents
the
pro
port
ion o
f th
e in
sure
r’s
com
mer
cial
lin
es o
f busi
nes
s. F
inal
ly,
Ln
RB
C i
s def
ined
as
the
nat
ura
l lo
gar
ithm
of
RB
C r
atio
.
Reinsurance: Quantile Regression Analysis 131
Management Review, October 2015 132
Figure 1 Distribution of Residuals for Reinsurance with 0.1Quantile
Figure 2 Distribution of Residuals for Reinsurance with 0.25 Quantile
Figure 3 Distribution of Residuals for Reinsurance with 0.5 Quantile
Figure 4 Distribution of Residuals for Reinsurance with 0.75 Quantile
Figure 5 Distribution of Residuals for Reinsurance with 0.9 Quantile
Figure 6 Q-Q Plot of Residuals for Reinsurance with 0.1Quantile
Reinsurance: Quantile Regression Analysis 133
Figure 7 Q-Q Plot of Residuals for Reinsurance with 0.25 Quantile
Figure 8 Q-Q Plot of Residuals for Reinsurance with 0.5 Quantile
Figure 9 Q-Q Plot of Residuals for Reinsurance with 0.75 Quantile
Figure 10 Q-Q Plot of Residuals for Reinsurance with 0.9 Quantile
of 28.27% to a maximum of 99.27% of the total
assets. The mean value of leverage is approximately
1.7788 and ranges from 0.0419 to 12.4190. The mean
(median) of two-year loss development (2_years_loss)
is −4.6591*10−5
(−3.2263*10−5
), which indicates that
insurers tend to create over loss reserving. In addition,
Table 1 shows that 67.81% of observations are for
stock insurers and that 40.93% of these are listed as
single insurers in the database. Overall, the summary
statistics for the explanatory variables are similar to
those reported in previous studies (Chang and Jeng
2015; Cole and McCullough 2006; Fier, McCullough,
and Carson 2013; Garven and Lamm-Tennant 2003;
Mayers and Smith 1990; Wang et al. 2008) and indi-
cates that the sample selection in this study is appro-
priate.
Table 2 reports the mean values of explanatory
variables for 10 quantile deciles of the insurer’s rein-
surance distribution. Overall, before considering the
interactive impacts among explanatory variables, most
explanatory variables tend to have different patterns
in the lower and higher reinsurance quantiles. For in-
stance, as reinsurance quantiles increase, the mean
values of Liq, ROA, Bus_H, Geo_H, and Single de-
crease, whereas the averages of Tax_ex and 2_years
_loss increase. Interestingly, Leverage, Stock, and
LnRBC present a U-shape from the lower reinsurance
deciles to the higher reinsurance deciles, whereas an
inverse U-shape emerges for Size and Com_line.
These important features of higher and lower rein-
surance quantiles imply that analysis using the tradi-
tional 2SLS approach may be biased or insufficient.
Thus, this study proposes the 2SQR approach to re-
solve the inadequacies of the traditional 2SLS ap-
proach.
4.2 Empirical Results
Model fitness checks should be done before the
empirical analysis14
. This study implements two diag-
nostic tests of model fitness, a histogram of the stand-
ardized residuals and the quantile-quantile plot diag-
nostic to scrutinize whether the data is well-fitted or
not. Figure 1 to Figure 10 describe the model fitness
patterns for the standardized residuals and the quan-
tile-quantile plot diagnostic. Figure 1 to Figure 5 sug-
gest that higher quantile regressions ( = 0.5, 0.75,
and 0.9) fit the data well, whereas lower quantile re-
14I sincerely thank the anonymous referee for suggesting
this important issue prior to the empirical analysis.
Management Review, October 2015 134
gressions ( = 0.1 and 0.25) do not. In addition, Fig- ure 6 to Figure 10 also propose that the data for high- er quantile demand for reinsurance ( = 0.5, 0.75, and 0.9) follow the line clearly, which indicates a normal distance distribution. However, lower demand for re- insurance quantiles ( = 0.1 and 0.25) deviate from normal in the tails, which presents short tails at both ends of the data distribution. In sum, model fitness checks propose that higher quantile regressions fit the data better than lower quantile regressions and pres- ent a normal distribution for reinsurance residuals.
To make a performance comparison between the 2SLS and 2SQR approaches, this study presents corresponding results from these two approaches in Table 3. The coefficient of Liq is significantly nega- tive in the 2SLS model, which suggests that the sub- stitution hypothesis is supported (Chang and Jeng 2015; Hau 2006). However, the coefficients in the 2SQR model are significantly different in the higher and lower quantiles. For insurers in the lower reinsurance quantiles ( = 0.1, 0.25, and 0.5), liquidity presents negatively to demand for reinsurance. These results are consistent with the results obtained using the 2SLS model. Insurers with more liquid assets have a greater ability to mitigate unexpected and emergent demand for cash, which then enables them to de- crease their demand for reinsurance, even though the reinsurance protection from reinsurers is lower. Con- versely, a significantly positive coefficient is found for insurers in the higher reinsurance quantile ( = 0.9). This result contradicts the hypothesis prediction that an insurer with a high level of reinsurance will need “less” reinsurance to uphold its target level of under- writing risk in the higher reinsurance quantiles. One plausible explanation for this interesting finding is that the higher liquidity insurer, with a higher given level of reinsurance, tends to have increased ability or capacity to underwrite a new or a risky business in advance and therefore, intends to increase its reinsur- ance purchasing as a whole.
In Table 3, the coefficient of the 2_years_loss is insignificantly negative in the 2SLS model. In ad- dition, the results also present a different impact on demand for reinsurance between the 2SLS and 2SQR models. For insurers in the lower reinsurance group ( = 0.1, 0.25, and 0.5), the empirical results (neg- ative coefficients) indicate that insurers with lower loss reserves tend to purchase less reinsurance, which is inconsistent to the expectation (Chang and Jeng 2015; Cole and McCullough 2006; Wang et al. 2008). This study speculates that insurers with lower levels of reinsurance tend to have higher liquidity, lower leverage, and higher-profit insurers
15 such that they
tend to purchase less reinsurance. Conversely, for in- surers in the higher reinsurance group ( = 0.9), loss reserving is positively related to the demand for rein- surance, which indicates that a decrease in loss re- serving will encourage insurers to purchase more rein- surance.
15Table 2 notes that insurers in the lower reinsurance
quantiles tend to have higher liquidity, lower leverage,
and higher profits. As a result, insurers are able to de-
crease their demand for reinsurance even though they
remain in a position of under-loss reserving.
Some independent variables are consistent to
the predictions for both the 2SLS and 2SQR models:
Reinsurance is negatively related to Tax_ex, ROA,
Bus_H, Geo_H, Single and Size, and is positively re-
lated to Leverage, Stock, Com_line, and LnRBC. In
Table 3, both the 2SLS and 2SQR models propose
that the tax-exempt factor is negatively related to the
insurer’s demand for reinsurance, which is inconsist-
ent with the suggestions of Smith and Stulz (1985),
Mayers and Smith (1990), and Garven and Lamm-
Tennant (2003). However, these results are closer to
the findings of Shiu (2011). To decrease expected tax
liabilities, an insurer with a higher marginal tax rate
will purchase less reinsurance. Furthermore, geograph-
ic and business diversified insurers tend to demand
more reinsurance than nongeographic and nonbusi-
ness diversified insurers. These results suggest that
insurers tend to rely on more real services via special-
ized knowledge and economies of scale from rein-
sureers. Moreover, consistent with Shiu (2011), more
concentrated insurers could specialize in, and may
underwrite, less volatile lines of business or less-risky
insured and therefore demand less reinsurance. In
addition, the results of the stock dummy variable
(Stock) indicate that the agency cost hypothesis is
supported for both the 2SLS and 2SQR models (Ad-
ams 1996; Shiu 2011). To mitigate the underinvest-
ment problem among residual claimholders, stock in-
surers purchase more reinsurance than mutual insurers.
As expected, the results displayed in Table 3
show that both the 2SLS and 2SQR models present a
negative relationship between the single dummy vari-
able (Single) and the demand for reinsurance. Single
insurers tend to demand less reinsurance because they
cannot benefit in terms of shifting profits within the
group and thereby reduce tax payments (Chang and
Jeng 2015; Cole and McCullough 2006; Shiu 2011;
Wang et al. 2008). Moreover, consistent with the argu-
ment of greater ability to confront losses and finan-
cial pressures, higher profitability insurers tend to de-
mand less reinsurance than lower profitability insur-
ers. In addition, consistent with the literature, this
study also finds that smaller insurers tend to confront
a higher probability of insolvency and, therefore, pur-
chase more reinsurance contracts than larger insurers.
In Table 3, the coefficients of Leverage are sig-
nificantly positively related to the demand for rein-
surance for both the 2SLS and 2SQR models. These
results are consistent with the hypotheses of bank-
ruptcy cost, agency cost, risk bearing, and renting
capital (Chang and Jeng 2015; Cole and McCullough
2006; Garven and Lamm-Tennant 2003; Shiu 2011;
Wang et al. 2008). In addition, the empirical results
indicate that the impacts of leverage on insurers’
demand for reinsurance differ between the lower and
higher quantiles. The effect of leverage on reinsure-
ance is likely to increase in higher levels of reinsure-
ance. Consequently, for each increment leverage, an
insurer with a high level of reinsurance will need
“more” reinsurance to maintain its underwriting risk
at a target level. In contrast, for each additional unit
of leverage, an insurer with a low level of reinsurance
will need “less” reinsurance in the lower quantiles.
Tab
le 3
E
mp
iric
al R
esu
lts
E
xp
ecte
d s
ign
2S
LS
2S
QR
/ Q
uan
tile
s (
=0.1
, 0.2
5, 0.5
, 0.7
5, an
d 0
.9)
0.1
0.2
5
0.5
0.7
5
0.9
Tes
t of
Eq
uali
ty
(X2 V
alu
e)
Inte
rcep
t
0.8
390
***
0.5
251
***
0.3
658
0.8
138
*
0.6
094
0.2
710
Leve
rage
+
0.0
789
**
-0.0
086
0.0
562
***
0.0
984
***
0.1
199
***
0.1
337
***
50.1
696
***
2_ye
ars
_lo
ss
+
-9.9
097
-2
4.7
115
**
-93.4
328
***
-80.1
537
**
75.3
432
140.5
779
**
49.6
809
***
Siz
e
- -0
.0128
***
-0.0
017
-0
.0008
-0
.0088
*
-0.0
27
2
***
-0.0
241
***
52.9
076
***
Liq
-
-0.6
086
***
-0.5
302
***
-0.5
163
*
-0.8
582
*
0.1
929
0.7
897
**
27.0
149
***
Tax_ex
+/-
-0
.0448
***
-0.0
006
-0
.0320
**
-0.0
685
***
-0.0
396
*
-0.0
386
**
20.4
048
***
RO
A
- -0
.2026
**
0.0
219
-0
.1339
*
-0.2
842
**
-0.6
461
***
-0.6
034
***
68.3
956
***
Bu
s_H
+
/-
-0.1
050
***
-0.0
920
***
-0.1
137
***
-0.1
247
***
-0.0
324
0.0
126
46.7
855
***
Geo
_H
+
/-
-0.0
652
***
-0.0
086
-0
.0313
**
-0.0
428
**
-0.0
982
***
-0.0
856
***
38.6
923
***
Sto
ck
+/-
0.0
717
***
0.0
410
***
0.0
667
***
0.0
836
***
0.1
019
***
0.0
584
***
26.0
646
***
Sin
gle
-
-0.1
458
***
-0.0
140
***
-0.0
779
***
-0.1
869
***
-0.2
292
***
-0.1
917
***
378.7
580
***
Com
_li
ne
+
0.0
779
***
0.0
752
***
0.1
046
***
0.0
787
**
0.0
911
***
0.0
736
**
3.7
495
Ln
RB
C
+
0.1
001
***
0.0
094
*
0.0
665
***
0.1
367
***
0.1
171
***
0.0
870
***
119.5
736
***
Obs.
5,6
26
This
tab
le p
rese
nts
the
resu
lts
of
both
the
2S
LS
appro
ach (
colu
mn 3
) an
d t
he
2S
QR
appro
ach w
ith q
uan
tile
s =
0.1
, 0.2
5,
0.5
, 0.7
5, an
d 0
.9 (
colu
mn 4
~ 8
). C
olu
mn 2
rep
ort
s th
e pre
dic
tion s
igns
of
firm
-spec
ific
char
acte
rist
ics.
Colu
mn 9
show
s th
e te
st o
f co
effi
cien
ts e
qual
ity
for
each
var
iable
acr
oss
quat
nle
s. T
he
dep
enden
t var
iable
is
Rei
ns,
whic
h i
s def
ined
as
(aff
ilia
ted r
einsu
rance
ceded
+
nonaf
fili
ated
re
insu
rance
ce
ded
)/(d
irec
t busi
nes
s w
ritt
en plu
s re
insu
rance
as
sum
ed).
L
eve
rage
it is
def
ined
as
th
e ra
tio of
dir
ect
busi
nes
s w
ritt
en-t
o-s
urp
lus.
T
he
var
iable
of
2_ye
ars
_lo
ss i
s tw
o-y
ear
loss
dev
elopm
ent
and e
qual
s th
e dev
elopm
ent
of
esti
mat
ed l
oss
es a
nd l
oss
expen
ses
incu
rred
tw
o y
ears
bef
ore
the
curr
ent
and p
rior
yea
r, s
cale
d b
y p
oli
cyhold
ers’
surp
luse
s. T
he
pro
xy
of
firm
siz
e is
Siz
e,
and i
s def
ined
as
the
nat
ura
l lo
gar
ithm
of
tota
l as
sets
. T
he
Liq
is
def
ined
as
the
sum
of
cash
plu
s in
ves
ted a
sset
s-to
-tota
l as
sets
. T
he
var
iable
of
Tax_ex
is d
efin
ed a
s ta
x-e
xem
pt
inves
tmen
t in
com
e re
lati
ve
to t
ota
l in
ves
tmen
t in
com
e. A
nd R
OA
is
def
ined
as
net
inco
me
plu
s ta
x a
nd i
nte
rest
expen
se,
div
ided
by
tota
l as
sets
. T
he
Bu
s_H
var
iable
is
the
line
of
busi
nes
s H
erfi
ndah
l In
dex
, w
hic
h i
s def
ined
as
the
sum
of
the
squar
es o
f th
e ra
tio o
f th
e doll
ar a
mount
of
dir
ect
busi
nes
s w
ritt
en i
n a
par
ticu
lar
line
of
insu
rance
to t
he
doll
ar a
mount
of
dir
ect
busi
nes
s ac
ross
all
27 l
ines
of
insu
rance
. C
onver
sely
, th
e G
eo_H
var
iable
is
the
geo
gra
phic
Her
findah
l In
dex
, w
hic
h i
s def
ined
as
the
sum
of
the
squar
es o
f th
e ra
tio o
f th
e doll
ar
amount
of
dir
ect
busi
nes
s in
sta
te j
to t
he
tota
l am
ount
of
dir
ect
busi
nes
s ac
ross
all
sta
tes.
The
org
aniz
atio
nal
form
dum
my
vari
able
is
Sto
ck
, w
hic
h i
s to
indic
ate
stock
or
mutu
als.
It
equal
s 1 i
f
the
insu
rer
is a
sto
ck,
and 0
if
it i
s a
mutu
al.
The
single
dum
my
var
iable
is
Sin
gle
, w
hic
h i
s to
indic
ate
an a
ffil
iate
d o
r nonaf
fili
ated
insu
re.
It e
qual
s 1 i
f th
e in
sure
r is
nonaf
fili
ated
, an
d 0
if
it i
s
affi
liat
ed.
Com
_li
ne r
epre
sents
the
pro
port
ion o
f th
e in
sure
r’s
com
mer
cial
lin
es o
f busi
nes
s. F
inal
ly,
Ln
RB
C i
s def
ined
as
the
nat
ura
l lo
gar
ithm
of
RB
C r
atio
. In
this
stu
dy,
tota
l fi
rm-y
ear
obse
rvat
ions
are
5,6
26.
***, **, an
d *
rep
rese
nt
stat
isti
cal
signif
ican
ce a
t th
e 1%
, 5%
, an
d 1
0%
lev
els,
res
pec
tivel
y.
Reinsurance: Quantile Regression Analysis 135
Management Review, October 2015 136
Furthermore, consistent with the risk taking
viewpoint and higher claims volatility argument, the
results indicate that insurers that write more comer-
cial lines (Com_line) tend to purchase more reinsure-
ance (Adams, Hardwick, and Zou 2008; Graham and
Rogers 2002; Graham and Smith 1999). Finally, both
the 2SLS and 2SQR models suggest that insurers
with higher probability of insolvency tend to pur-
chase more reinsurance, which is consistent with the
argument of insolvency concerns (Chen, Hamwi, and
Hudson 2001; Shiu 2011).
Note that insurers in the higher reinsurance
quantiles are safer than those in the lower reinsurance
quantiles16
. Thus, it is predicted that the determinants
on demand for reinsurance in the lower and higher
quantiles are distinct. In other words, the influence of
the demand for reinsurance for each variable may
present a different sign, or a dissimilar impact (mag-
nitude). This study further implements the tests of e-
quality for each variable across the five quantiles ( =
0.1, 0.25, 0.5, 0.75, and 0.9). Testing these results
indicates that there is a difference among the esti-
mated parameters for all variables across the five
quantiles, except Com_line, which indicates that Hy-
pothesis 1 is supported. Specifically, Leverage, ROA,
Geo_H, Single, and Size have larger impacts on de-
mand for reinsurance in the higher reinsurance quan-
tiles, whereas Bus_H presents an inverse result in the
lower reinsurance quantiles. In addition, the evidence
of Tax_ex, Stock, and LnRBC suggests that the influ-
ence on demand for reinsurance is larger in the me-
dian quantile. This study also finds that the signs of
Liq and 2_years_loss are significantly different be-
tween the lower and higher reinsurance quantiles.
However, the other remaining variables present a con-
sistent sign regarding both lower and higher reinsure-
ance quantiles. Summing up, the evidence suggests
that Hypothesis 2 is, as a whole, weakly supported.
Taken together with previous empirical results,
the evidence presented by this study proposes that the
2SQR approach reports a definite and distinct impact
on the demand for reinsurance for various reinsurance
quantiles. As proposed, the traditional 2SLS approach
was found to contain some inappropriate assumptions
and therefore to produce a biased estimation. In con-
trast, the 2SQR approach provides efficient estima-
tion and insightful information on the determinants of
demand for reinsurance, which could compensate for
16I sincerely thank the anonymous referee who pointed
out that if insurers try to maintain a target level of under-
writing risk, then the argument that “insurers in the high-
er reinsurance quantiles are safer than those at the lower
reinsurance quantiles” may not hold theoretically. In
general, it should be noted that the protections from rein-
surance contracts are higher for insurers in the higher re-
insurance quantiles. Insurers purchase more reinsurance
could mitigate not only underwriting risk but also liquid-
ity risk (or short-term financial pressure), catastrophic
risk, and bankruptcy risk. Even if insurers tend to main-
tain a target level of underwriting risk, they may still
have incentives to demand more reinsurance as a whole.
the inadequacy of the traditional 2SLS approach. Over-
all, this study concludes that Hypothesis 3 is strongly
supported.
5. Concluding Remarks
Instead of using the traditional 2SLS approach, this
study proposes the 2SQR approach for examining the
determinants of the insurer’s demand for reinsurance.
This study finds that the determinants of demand for
reinsurance are driven by the insurer’s specific quan-
tile within the overall demand for reinsurance distri-
bution. The empirical evidence shows that the param-
eters (signs) for liquidity and loss development vary
within different reinsurance quantiles, which differen-
tiates these findings from previous ones (Chang and
Jeng 2015; Cole and McCullough 2006; Hau 2006;
Shiu 2011; Wang et al. 2008). The evidence indicates
that the traditional 2SLS approach may provide an
insufficient or biased explanation for the determinants
of insurers’ demand for reinsurance. In addition, al-
though consistent evidence is presented for other
firm-specific characteristics, the magnitudes of these
effects on demand for reinsurance differ significantly
across all reinsurance quantiles. Again, these new find-
ings, which also differ from the literature, enrich the
function of the 2SQR analysis in the lower and higher
quantiles of the reinsurance distribution. To sum up,
the 2SQR approach is more efficient and can provide
more insightful information than the traditional 2SLS
approach.
Reinsurance is an important issue for the pri-
mary insurers. Policyholders, policy makers, and/or
regulators should pay more attention in their assess-
ments of an insurer’s financial pressures and bank-
ruptcy problems, especially for insurers operating with
lower levels of reinsurance. In sum, it is of impor-
tance for policyholders, policy makers, and/or regula-
tors to assess the insurer’s demand for reinsurance by
considering the empirical implications of both the
2SLS and 2SQR models.
Reinsurance: Quantile Regression Analysis 137
REFERENCES
Adams, Mike (1996), “The Reinsurance Decision in
Life Insurance Firms: An Empirical Test of the
Risk-Bearing Hypothesis,” Accounting and Fi-
nance, 36(1), 15-30.
----, Philip Hardwick, and Hong Zou (2008), “Rein-
surance and Corporate Taxation in the United
Kingdom Life Insurance Industry,” Journal of
Banking and Finance, 32(1), 101-115.
Amemiya, Takeshi (1982), “Two Stage Least Abso-
lute Deviations Estimators,” Econometrica, 50
(3), 689-712.
Baranoff, Etti G. and Thomas Sager (2002), “The
Relationship among Asset Risk, Product Risk,
and Capital in the Life Insurance Industry,” Jour-
nal of Banking and Finance, 26(6), 1181-1197.
---- and ---- (2003), “The Relationship among Organ-
izational and Distribution Forms and Capital and
Asset Risk Structures in the Life Insurance In-
dustry,” Journal of Risk and Insurance, 70(3),
375-400.
Bruinshoofd, Allard and Clmens Kool (2002), “The
Determinants of Corporate Liquidity in the Neth-
erlands,” SSRN working paper, (February 14),
(accessed December 4, 2010), [available at http://
ssrn.com/abstract=300698 or http://dx.doi.org/10.
2139/ssrn.300698].
Carneiro, Luiz A. and Michael Sherris (2005), “De-
mand for Reinsurance: Evidence from Austral-
ian Insurers,” Working Paper, (June 10), (ac-
cessed December 10, 2010), [available at https://
www.business.unsw.edu.au/About-Site/Schools-
Site/risk-actuarial-site/Documents/L.%20Carneir
o%20and%20M.%20Sherris%20-%20Demand%
20for%20Reinsurance%20Evidence%20from%2
0Australian%20Insurers.pdf].
Chang, Vincent Y. and Jeffrey Z. Tsai (2014), “Quan-
tile Regression Analysis of Corporate Liquidity:
Evidence from U.S. Property-Liability Insurance
Industry,” The Geneva Papers on Risk and In-
surance: Issues and Practice, 39(1), 77-89.
---- and Vivian S. Jeng (2015), “The Relationships a-
mong the Demand for Reinsurance, Liquidity, and
Leverage in the U.S. Property-Liability Insur-
ance Industry,” Taiwan Economic Review, Forth-
coming.
Chen, Lin-An and Stephen Portnoy (1996), “Two-
Stage Regression Quantiles and Two-Stage Trim-
med Least Squares Estimators for Structural E-
quation Models,” Communication in Statistics,
Theory Methods, 25(5), 1005-1032.
Chen, Yueyen, Iskandar S. Hamwi, and Tim Hudson
(2001), “The Effect of Ceded Reinsurance on
Solvency of Primary Insurers,” International Ad-
vances in Economic Research, 7(1), 65-82.
Christensen, Theodore E., Robert E. Hoyt, and Jef-
frey S. Paterson (1999), “Ex Ante Incentive or
Earnings Management and the Informativeness
of Earnings,” Journal of Business Finance and
Accounting, 26(7-8), 807-832.
Cole, Cassandra R. and Kathleen A. McCullough
(2006), “A Reexamination of the Corporate De-
mand for Reinsurance,” Journal of Risk and In-
surance, 73(1), 169-192.
Cummins, J. David and David W. Sommer (1996),
“Capital and Risk in Property-Liability Insurance
Markets,” Journal of Banking and Finance, 20
(6), 1069-1092.
Fier, Stephen G., Kathleen A. McCullough, and James
M. Carson (2013), “Internal Capital Markets and
the Partial Adjustment of Leverage, Journal of
Banking and Finance, 37(3), 1029-1039.
Garven, James R. and Joan Lamm-Tennant (2003), “The
Demand for Reinsurance: Theory and Empirical
Tests,” Insurance and Risk Management, 7(3),
217-238.
Gaver, Jennifer J. and Jeffrey S. Paterson (1999), “Man-
aging Insurance Company Financial Statements
to Meet Regulatory and Tax Reporting Goals,”
Contemporary Accounting Research, 16(2), 207-
241.
Grace, Elizabeth V. (1990), “Property-Liability Insur-
er Reserve Errors: A Theoretical and Empirical
Analysis,” Journal of Risk and Insurance, 57(1),
28-46.
Graham, John R. and Daniel A. Rogers (2002), “Do
FirmsHedge in Response to Tax Incentives?”
Journal of Finance, 57(2), 815-839.
---- and Clifford W. Smith (1999), “Tax Incentives to
Hedge,” Journal of Finance, 54(6), 2241-2262.
Harrington, Scott E. and Patricia M. Danzon (1994),
“Price Cutting in Liability Insurance Markets,”
The Journal of Business, 67(4), 511-538.
Hau, Arthur (2006), “The Liquidity Demand for Cor-
porate Property Insurance,” Journal of Risk and
Insurance, 73(2), 261-278.
Hoyt, Robert E. and Ho Khang (2000), “On the De-
mand for Corporate Property Insurance,” Jour-
nal of Risk and Insurance, 67(1), 91-107.
John, Teresa A. (1993), “Accounting Measures of Cor-
porate Liquidity, Leverage, and Costs of Finan-
cial Distress,” Financial Management, 22(3), 91-
100.
Kader, Hale A. and Michael Adams (2007), “Testing
for Trade-offs in the Reinsurance Decision of
United Kingdom (UK) Life Insurance Firms,”
technical paper, European Business Management
School, University of Wales, Swansea.
Kim, Chang-Soo, David C. Mauer, and Ann E. Sher-
man (1998), “The Determinants of Corporate Li-
quidity: Theory and Evidence,” Journal of Finan-
cial and Quantitative Analysis, 33(3), 335-359.
Management Review, October 2015 138
Kim, Tae-Hwan and Christophe Muller (2004), “Two-
Stage Quantile Regression When the First Stage
Is Based on Quantile Regression,” Econometrics
Journal , 7(1), 218-231.
Kim, Woon-Joong, David Mayers, and Clifford W.
Smith (1996), “On the Choice of Insurance Dis-
tribution Systems,” Journal of Risk and Insur-
ance, 63(2), 207-227.
Koenker, Roger and Gillber Bassett (1978), “Regres-
sion Quantiles,” Econometrica, 46(1), 33-50.
---- and Kevin Hallock (2001), “Quantile Regres-
sion,” Journal of Economic Perspectives, 15(4),
143-156.
Mayers, David and Clifford W. Smith (1990), “On the
Corporate Demand for Insurance: Evidence from
the Reinsurance Market,” Journal of Business,
63(1), 19-40.
---- and ---- (2004), “Incentive for Managing Ac-
counting Information: Property-Liability Insurer
Stock-Charter Conversions,” Journal of Risk and
Insurance, 71(2), 213-251.
Opler, Tim, Lee Pinkowitz, Rene Stulz, and Rohan
Williamson (1999), “The Determinants and Im-
plications of Corporate Cash Holdings,” Journal
of Financial Economics, 52(1), 3-46.
Petroni, Kathy R. (1992), “Optimistic Reporting in
the Property-Casualty Insurance Industry,” Jour-
nal of Accounting and Economics, 15(4), 485-
508.
Powell, James (1983), “The Asymptotic Normality of
Two Stage Least Absolute Deviations Estimators,”
Econometrica, 51(5), 1569-1575.
Powell, Lawrence S. and David W. Sommer (2007),
“Internal Versus External Capital Markets in the
Insurance Industry: The Role of Reinsurance,”
Journal of Financial Services Research, 31(2-3),
173-188.
Shiu, Yung-Ming (2006), “Corporate Liquidity: Evi-
dence from the United Kingdom Life Insurance
Industry,” Applied Economics Letters, 13(15), 993-
998.
---- (2011), “Reinsurance and Capital Structure: Evi-
dence from the United Kingdom Non-Life In-
surance Industry,” Journal of Risk and Insurance,
78(2), 475-494.
Shortridge, Rebecca T. and Stephen M. Avila (2004),
“The Impact of Institutional Ownership on the
Reinsurance Decision,” Risk Management and
Insurance Review, 7(2), 93-106.
Smith, Clifford W. and Rene Stulz (1985), “The De-
terminants of Firms’ Hedging Policies,” Journal
of Financial and Quantitative Analysis, 20(4),
566-584.
Wang, Jeniffer, Vincent Y. Chang, Gene C. Lai, and
Larry Y. Tzeng (2008), “Demutualization and De-
mand for Reinsurance,” The Geneva Papers on
Risk and Insurance: Issues and Practice, 33(3),
566-584.
Weiss, Mary A. (1985), “A Multivariate Analysis of
Loss Reserving Estimates in Property-Liability
Insurers,” Journal of Risk and Insurance, 52(2),
199-221.
---- and Joon-Hai Chung (2004), “U.S. Reinsurance
Price, Financial Quality, and Global Capacity,”
Journal of Risk and Insurance, 71(3), 437-467.
Yanase, Noriyoshi (2015), “The Effect of Japanese
‘Keiretsu’ on the Corporate Demand for Rein-
surance,” Journal of Risk and Insurance, Forth-
coming.