M R I E S R M L I: Ortality Isk and TS Ffect On Hortfall and ISK Anagement in IFE Nsurance
M R I E S R M L I: Ortality Isk and TS Ffect On Hortfall and ISK Anagement in IFE Nsurance
M R I E S R M L I: Ortality Isk and TS Ffect On Hortfall and ISK Anagement in IFE Nsurance
1, 57–90
DOI: 10.1111/j.1539-6975.2012.01496.x
ABSTRACT
Mortality risk is a key risk factor for life insurance companies and can have a
crucial impact on its risk situation. In general, mortality risk can be divided
into different subcategories, among them unsystematic risk, adverse selec-
tion, and systematic risk. In addition, basis risk may arise in case of hedging,
for example, longevity risk. The aim of this article is to holistically analyze
the impact of these different types of mortality risk on the risk situation and
the risk management of a life insurer. Toward this end, we extend previous
models of adverse selection, empirically calibrate mortality rates, and study
the interaction among the mortality risk components in the case of an insurer
holding a portfolio of annuities and term life insurance contracts. For risk
management, we examine natural hedging and mortality contingent bonds.
Our results show that particularly adverse selection and basis risk can have
crucial impact not only on the effectiveness of mortality contingent bonds,
but also on the insurer’s risk level, especially when a portfolio consists of
several types of products.
INTRODUCTION
Recently, there has been a growing interest in mortality risk and its management in
the scientific literature as well as in practice, especially due to the demographic devel-
opment in most industrialized countries. The increasing life expectancy poses serious
problems to life insurance companies selling annuities and to pension funds. These
problems are especially severe because of a scarcity of possibilities to hedge against
this risk. Due to the limited capacity of reinsurance, several alternative instruments
for managing demographic risk, for example, by transferring mortality risk to the
capital market or the use of natural hedging, have been discussed in the scientific
literature and by practitioners. However, mortality heterogeneity as well as infor-
mation asymmetries between the insurance company and the insured about these
57
58 THE JOURNAL OF RISK AND INSURANCE
1
See Finkelstein and Poterba (2002) and Cohen and Siegelmann (2010).
2
In general, adverse selection refers to information asymmetry and hidden characteristics. In
this paper, we follow BDV (2002a) and refer to adverse selection as the observation that due to
mortality heterogeneity and asymmetric information, annuitants experience a lower mortality
than the average population and therefore have a higher life expectancy. Other papers (e.g.,
Coughlan et al., 2007) refer to this as basis risk. In the following analysis, we consider two cases
in order to highlight the importance of mortality information in underwriting, one where the
insurer is not fully informed about the mortality of its annuitants, and one case where adverse
selection can be fully addressed.
3
Other potential sources of basis risk in longevity hedges are stated, for example, by Sweeting
(2007) or Coughlan et al. (2007) and include age mismatch or geographic differences.
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 59
Ngai and Sherris (2011) also use a portfolio-specific mortality factor and, following
Stevenson and Wilson (2008), assume a linear and constant effect of age as the only
impact factor. BDV (2002a) choose a different approach and model annuitant mortal-
ity through a Brass-type relational model for the central death rates. Concerning the
effectiveness of MCBs (or other instruments for transferring mortality risk to capital
markets) under basis risk resulting from adverse selection, other certain aspects have
already been discussed in the literature. Sweeting (2007) discusses the influence of
basis risk when using a survivor swap qualitatively in a utility-maximizing frame-
work and concludes that basis risk is comparatively small, and thus will not hinder
the occurrence of hedging transactions. In terms of the effectiveness of q-forwards4
based on the population mortality for hedging insured lives, Coughlan et al. (2007)
use historical data and conclude that the loss in efficiency is small from a long-term
perspective. Ngai and Sherris (2011) quantify the impact of basis risk in longevity
bonds and q-forwards in a static framework and find that basis risk does not signif-
icantly affect the hedging effectiveness. Coughlan et al. (2010) introduce a general
framework for assessing basis risk in longevity hedges and conclude that it can be
reduced considerably by applying their framework for calibrating the hedge. A more
general concept in this context, the so-called population basis risk, describes the risk
of basing the payout of the risk management instrument on a different population5
and is discussed by Li and Hardy (2009) and Coughlan et al. (2007). Thus, to date,
results in the literature suggest that basis risk in longevity hedges overall has a minor
impact on the effectiveness of the hedge.
The second risk management instrument, natural hedging, has also been studied in
the literature. Cox and Lin (2007) as well as Bayraktar and Young (2007) examine the
impact of natural hedging on pricing. Gründl, Post, and Schulze (2006) and Hanewald,
Post, and Gründl (2011) compare the effects of different risk management strategies on
shareholder value, concluding that natural hedging is the preferred risk management
tool, but only under certain circumstances. Wang et al. (2010) apply the concept of
duration to mortality and derive an optimal liability mix, which is characterized
by a portfolio mortality duration of zero, while Wetzel and Zwiesler (2008) show
that the mortality variance, i.e., the variance due to fluctuations in mortality, can be
reduced by more than 99 percent through portfolio composition. Gatzert and Wesker
(Forthcoming) consider the insurer as a whole and show how to immunize a given
risk level by simultaneously considering the investment and insurance portfolio.
Despite a fair amount of research on mortality risk, the impact of all three mortality
risk components (separately and combined) and basis risk resulting from adverse
selection on the risk level of a life insurance company and on the effectiveness of
different risk management strategies with respect to reaching a desired risk level as
well as hedging against unexpected changes in mortality has not been systematically
4
A q-forward is a standardized mortality contingent swap, based on the LifeMetrics index
by JP Morgan. The LifeMetrics index is distinguished by gender and age for the popula-
tion of United States, England, Wales, the Netherlands, and Germany (for more informa-
tion and current index data, see http://www.jpmorgan.com/pages/jpmorgan/investbk/
solutions/lifemetrics).
5
Potential sources of population mismatch include differences in geographic location, age,
social status, etc.
60 THE JOURNAL OF RISK AND INSURANCE
6
This mortality model is taken as an example and can as well be replaced by other stochastic
mortality models that provide a good fit depending on the concrete application (and the
respective country).
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 61
MODEL FRAMEWORK
Modeling and Forecasting Mortality Risk
Modeling Unsystematic Mortality Risk. One of the most frequently used models for
mortality is the Lee–Carter (1992) model, which consists of a demographic and a time
series part. In this framework, the central death rate or force of mortality μx (τ ) is
modeled through
where a x and b x are time constant parameters, indicating the general shape of mor-
tality over age and the sensitivity of the mortality rate at age x to changes in kτ ,
respectively, where kτ is a time-varying index and shows the general development
of mortality over time, and εx,τ is an error term with mean 0 and constant variance.
Lee and Carter (1992) propose to fit an appropriate ARIMA process on the estimated
time series of kτ
kτ = φ + α1 · kτ −1 + α2 · kτ −2 + · · · + α p · kτ − p + δ1 · ετ −1 + δ2 · ετ −2 + · · · + δq · ετ −q +ετ = k̂τ + ετ ,
k̂τ
where k̂τ is the forecasted realization of the time index used in BDV (2002a) for simu-
lating random numbers of death, thus reflecting the unsystematic mortality risk, and
E x,τ is the risk exposure at age x and time τ , defined as E x,τ = (nx−1 (τ − 1) + nx (τ ))/2,
where nx (τ ) is the number of persons (i.e., the population size) still alive at age x and
the end of year τ .7 The advantages of the BDV (2002a) model are that the restrictive
assumption of homoscedastic errors made in the Lee–Carter (1992) model is given up
and that the resulting Poisson distribution is well suited for a counting variable such
as the number of deaths.8
Modeling Adverse Selection and Basis Risk. Mortality heterogeneity refers to the fact
that mortality rates are not identical for all individuals of the same age x but differ de-
pending on, for example, genetic predisposition or behavior. Individuals are usually
able to gain some information about their individual mortality, for example, through
family history or their general health situation, which may influence their insurance
decisions (see Finkelstein and Poterba, 2002). For instance, a person estimating its
own mortality to be below average will be more likely to purchase an annuity than
a person with below average mortality. Insurance companies generally do not have
access to this information, and thus cannot directly distinguish between individuals
with above or below average health. These circumstances give rise to information
asymmetries and thus the problem of adverse selection, as both the level of mortal-
ity rates as well as their development over time differ between annuitants and the
general population. At the same time, adverse selection also implies basis risk when
hedging against longevity risk due to the difference between the mortality rates of
the reference population used as an underlying for the hedge and the annuitants’
mortality rates. Hence, basis risk arises because the underlying and the hedged pop-
ulation are not perfectly dependent and can thus reduce the hedging effectiveness.9
In the following, adverse selection is modeled through an extension of the brass-type
relational model used by, among others, BDV (2002a)
pop pop
ln μann
x,τ = α + β1 · ln μx,τ + β2 · ln μx,τ · τindex + e x,τ , (1)
9
Here, adverse selection and basis risk are modeled identically through a differing mortality
experience for annuitants and the population as a whole. However, the two terms refer to
different aspects. While adverse selection refers to the difference in mortality experiences
arising from mortality heterogeneity, basis risk refers to the consequences of this difference
when hedging longevity risk through capital markets.
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 63
since the actual relationship between annuitant and population mortality (equal to
the mortality of term life insurance policyholders) is misestimated, the different de-
velopment of mortality rates for annuitants and life insurance policyholders cannot
be fully taken into account when calculating premiums and benefits. However, the in-
surance company may be able to gain information about the average mortality within
the annuitant portfolio under adverse selection, for example, by way of experience
rating. Thus, second, we assume that the insurer is able to perfectly estimate and
thus account for adverse selection effects and consequently to take this information
into account when determining benefits and premiums of annuitants.10 This setting
is referred to as “adverse selection perfectly estimated.”
Modeling Systematic Mortality Risk. Systematic mortality risk is the risk that cannot
be diversified through enlarging the insurance portfolio; that is, it is the risk of unex-
pected deviations from the expected mortality rates applying to all individuals, which
can result, for example, from a common factor unexpectedly impacting mortality at
all ages (see, e.g., Wills and Sherris, 2010). This can, in general, be attributed ei-
ther to unexpected environmental or social influences, impacting mortality positively
or negatively,11 or to wrong expectations about future mortality due to estimation
errors.12 Unexpected common factors that influence lives in a similar way induce
dependencies and thus destroy diversification benefits of large pool sizes. In the lit-
erature, systematic mortality risk is modeled and accounted for in different ways.
Hanewald, Piggot, and Sherris (2011) and Wills and Sherris (2010) characterize sys-
tematic (longevity) risk as uncertain changes in mortality applying to all individuals,
which leads to dependencies between lives due to common improvement in mortal-
ity rates across individuals. Wang et al. (2010) describe systematic risk as a constant
shock to the force of mortality, thus accounting for unexpected changes in mortality
rates, similarly to Milevsky and Promislow (2003) and Gründl, Post, and Schulze
(2006). Furthermore, Cox and Lin (2007) point out that while mortality risk may not
be hedgeable in financial markets, it may be reduced or eliminated by insurers by
means of, for example, natural hedging, reinsurance, asset-liability management, or
mortality swaps.
In the following, systematic mortality risk is modeled through different realiza-
tions of the time trend kτ , where now the error term is taken into account, having
syst
kτ = k̂τ + ετ , which we refer to as the “neutral scenario” as the mean life expectancy
does not change. The factor ετ impacts mortality at all ages in year τ 13 and thus causes
dependencies between lives, which cannot be diversified through enlarging the port-
folio. To study systematic mortality risk in more detail, we conduct scenario analyses
by distinguishing between a longevity scenario, in which mortality is unexpected
low, and a scenario with unexpected high mortality (“mortality scenario”) using the
10
Thus, adverse selection in the sense of hidden information is, in fact, eliminated.
11
Additionally, certain other macroeconomic variables might have an influence on mortality
(see, e.g., Hanewald, 2011).
12
An example of a potential source of estimation error is the choice of the appropriate sample
period, since kτ is rather sensitive towards the specified period.
13
Due to the assumed ARIMA process for kτ , subsequent years are also impacted by the
realization of ετ .
64 THE JOURNAL OF RISK AND INSURANCE
TABLE 1
Overview of Force of Mortality Depending on Included Mortality Risk Component
Without adverse
selection μxpop,unsyst Mortality rates of the reference population
• Only unsystematic risk
μxpop,syst Mortality rates of the reference population with
systematic mortality risk (time trend ksyst )
• Unsystematic risk + systematic risk
With adverse selection μaxnn,unsyst Mortality rates of annuitants
• Unsystematic risk + adverse selection
μaxnn,syst Mortality rates of annuitants with systematic
mortality risk (time trend ksyst )
• Unsystematic risk + adverse selection +
systematic risk
TABLE 2
Balance Sheet of the Insurance Company at
Time t = 0 for i = unsyst, syst
Assets Liabilities
i
Slow (0) MAi (0)
i
Shigh (0) MLi (0)
i
Mbond (0) E (0)
syst,longevity syst,mortality
kτ = k̂τ − |ετ | and kτ = k̂τ + |ετ | .
Summary of Modeled Mortality Risk. Based on the mortality model presented here,
the probability that a male policyholder aged x in calendar year τ dies within the
next year, given he has survived until age x, is calculated by q x (τ ) = 1 − exp(−μx (τ ))
n−1
(see BDV, 2002a, p. 376), and n px = i=0 px+i is the probability that an x-year-old
male policyholder survives for the next n years. Based on the previous modeling of
mortality rates, four different cases can be distinguished depending on the inclusion of
unsystematic risk, adverse selection, and systematic mortality risk, laid out in Table 1.
systematic risk. The term life insurance contract is financed through constant annual
premiums PLi . We thereby assume that the insurance company sells fL ·n = nL (0) term
life and (1 − fL )·n = nA (0) annuity contracts, where fL denotes the fraction of term life
insurance and n the constant number of insurance contracts sold.
On the liability side, MiA(0), or, more general, MiA(t), denotes the value of annuities
at time t and MLi (0) represents the value of term life insurance liabilities at time t = 0
for i = unsyst, syst. To increase the comparability in the numerical analysis, the total
volume per contract14 of both contract types in t = 0 is identical. E(0) denotes the initial
equity contributed by shareholders in t = 0, and Ei (t) is residually determined as the
difference between assets and liabilities. In return for their investment, shareholders
receive a constant fraction re of the positive earnings each year as a dividend, given
by divi (t) = re · max(E i (t) − E i (t − 1); 0), i = unsyst, syst. Furthermore, Mbond
i (t) is the
i
value of the MCB at time t as detailed in the next subsection, Slow (0) stands for the
market value of low-risk assets at time t = 0, and Shigh i (0) represents the high-risk
assets. If the insurance company purchases a MCB, a premium ix,T has to be paid in
t = 0, implying that the initial capital at time 0 available for investment in the capital
market Si (0) is given by
The total value of assets Ai (t) in the balance sheet at time t, in turn, increases by the
market value of the MCB such that
The market value of assets Sij (t), j = low, high, is assumed to follow a geometric
Brownian motion with μ j the being drift and σ j denoting the volatility. Let WlowP
P denote two Brownian motions with correlation ρ under the real-world
and Whigh
measure P on the probability space (, F , P), where F is the filtration generated by
the Brownian motion. Hence, Sij (t) can be expressed as (see Björk, 2004)
1 P
Sij (t) = Sij (s) · exp μ j − σ 2j · (t − s) + σ j · Wj,t P
− Wj,s ,
2
j = low, high, i = unsyst, syst.
i
Si (t) = Shigh i (t) + ni (t) · P i − ni (t) · a − d i (t) · DB
(t) + Slow L L A L
14
Volume here refers to the present value of expected benefit payments in t = 0.
66 THE JOURNAL OF RISK AND INSURANCE
As before, niA(t) is the number of annuitants still alive at the end of year t, niL (t)
the number of term life insurance policyholders alive at the end of year t, d Li (t)
represents the number of deaths of life insurance policyholders during year t, and
Xi (t) denotes the coupon payment for the MCB in year t and Slow i (t − s) = α · Si (t − s)
i
and Shigh (t − s) = (1 − α) · S (t − s). The total market value of assets is then given by
i
i
Ai (t) = Si (t) + Mbond (t), i = unsyst, syst.
A−t
T
MiA (t) = niA (t) · A
a ·s px+t · (1 + r )−s , i = unsyst, syst,
s=1
⎛ ⎞
−t−1
TL
MLi (t) = niL (t) · ⎝ L
DB ·s px+t L
· q x+t+s · (1 + r )−(s+1) − PLi ·s px+t
L
· (1 + r )−s ⎠ ,
s=0
i = unsyst, syst,
with niA(t) being the number of annuity contracts and niL (t) being the number of life
insurance contracts in t. TL and TA denote the maximum duration of the respective
contract type and r is the risk-free rate. The superscripts A and L of the mortality
rates used in the valuation of insurance contracts refer to the respective mortality
rates for annuitants or life insurance holders, which depend on the mortality as-
sumptions (unsystematic risk, unsystematic risk + adverse selection; see Table 1), but
do not include the systematic mortality risk. The contract parameters a, DB and PLi ,
i = unsyst, syst, are calculated by using the actuarial equivalence principle; that is,
expected premiums must be equal to expected benefits. We refer to these premiums
fair fair
as SP A and PL .16 In the presence of systematic mortality risk, the insurance com-
pany additionally demands a risk premium (1+δ), which we assume to be equal for
syst syst
both products, resulting in the premiums PL and SP L . Furthermore, the contract
volume of annuities and life insurance contracts, defined as the present value of the
expected benefit payouts, is fixed to V for both contract types. Thus, the death benefit
15
The time subscript τ in the death and survival probabilities has been dropped from the for-
mulas for ease of illustration; however, all death and survival probabilities remain dependent
on age and time.
16
The probability of default is not taken into account in pricing since we assume that the
insurance benefits will continue to be paid out in case of a default; this is in line with the
current situation in many countries, where benefits are guaranteed by a guaranty fund (see,
e.g., Gatzert and Kling, 2007).
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 67
L −1
T
L −1
T
!
· (1 + r )−(t+1) = ·t pxL · (1 + r )−t ,
fair
V= DB ·t pxL · q x+t
L
PL (2)
t=0 t=0
A−1
T
!
a ·t+1 pxA · (1 + r )−(t+1) = SP A .
fair
V= (3)
t=0
T−1
ix,T = E Xunsyst (t) · (1 + r − λ)−(t+1) , i = unsyst, syst,
t=0
17
This approach seems justifiable despite the nonsuccess of the bond, since, for example, Blake
et al. (2006) state that the failure was likely due to weaknesses in design rather than due to
mispricing.
68 THE JOURNAL OF RISK AND INSURANCE
where r is the risk-free interest rate and λ the risk premium for systematic mortality
risk, if i = syst and λ = 0 if i = unsyst. The actual cash flow Xi (t) at time t depends on
the mortality of the reference population, that is, on the forecasted force of mortality
pop,i
μx , and thus also on whether systematic risk is taken into account in the analysis
or not (i = unsyst, syst). As before, in pricing and valuation, systematic risk is not
taken into consideration in the mortality projection as it represents an unexpected
change to mortality. However, systematic mortality risk is addressed by introducing
the risk premium λ. Moreover, basis risk is involved in the hedge as the behavior of
pop,i
the underlying of the bond (with μx ) is not identical to the development of the
hedged position, that is, the portfolio of annuities (with μann,i
x , see Table 1). Let niref (t)
denote the number of persons in the reference group still alive at the end of year t,
which can be recursively calculated as niref (t) = (niref (t − 1) − dref
i (t)), where d i (t) is
ref
the number of persons who died within year t. This is calculated using
nref (0) equal to an arbitrary number18 and C being the initial coupon payment agreed
upon at inception of the contract. The exposure to risk of the reference population
ref ,i r e f ,i pop,i pop,i
E x,t is given byE x,t = −(niref (t − 1) · q x )/ ln( px ) (see BDV, 2002b). Then, the
annual payoff Xi (t) is equal to
niref (t)
Xi (t) = · C, i = unsyst, syst.
nref (0)
Thus, the actual coupon payment taken into account in risk measurement by means
of actual cash flows is Xsyst (t), whereas for valuation, Xunsyst (t) is used. The value of
i
the MCB at time t, Mbond (t), i = unsyst, syst, is an asset for the insurance company and
given by the expected value of future cash flows discounted to time t and given the
information at time t, multiplied by the number of MCBs purchased at time 0 (n B ),
T−1
i
Mbond (t) = n B · E t Xunsyst ( j) · (1 + r − λ)−( j−t+1) ,
j=t
t = 0, . . . , T − 1, λ = 0 if i = unsyst.
Risk Management Using Natural Hedging. We further examine the effect of natural
hedging, which uses the opposed reaction toward changes in mortality rates of term
life insurances and annuities to immunize a life insurer against systematic mortality
18
nref (0) is merely a scaling parameter that does not impact the result, since the coupon payment
is expressed in relative terms.
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 69
risk. In the literature, natural hedging has been used for minimizing (e.g., Wetzel
and Zwiesler, 2008) or immunizing (e.g., Wang et al., 2010) the risk of an insurance
company in response to unexpected (systematic) changes in mortality. In the following
analysis, we take an immunization approach as in Wang et al. (2010), but follow
Gatzert and Wesker (Forthcoming) by considering the insurance company as a whole
instead of only focusing on the liability side. In the case without adverse selection
and taking the probability of default PD as the relevant risk measure, for example,
the optimal portfolio composition f L∗ is defined as
pop pop,syst pop pop,syst
g ( f L ) =
PD f L ; μx ; μx = PD f L ; μx − PD f L ; μx
!
= PDunsyst ( f L ) − PDsyst ( f L ) = 0,
implying that the probability of default is the same with and without systematic risk.
However, since the risk immunizing portfolio composition does not constitute the
risk-minimizing portfolio, the investment strategy, reinsurance, or MCBs can be used
to achieve a desired risk level, which is at the same time immunized against changes
in mortality as illustrated in the numerical section.
Risk Measurement. To analyze the impact of mortality risk on the insurer’s risk sit-
uation, we consider two downside risk measures, namely, the probability of default
(PDi ) and the mean loss (MLi ), which essentially correspond to the lower partial mo-
ments (LPMs) of order 0 and 1, adapted to account for the long duration of contracts
and to take into account potential default during the contract period. The probability
of default is defined as
where Tdi = inf{t : Ai (t) < L i (t)} is the time of default. Thus, the PD only measures
the frequency of default. The second risk measure, the mean loss, is defined as an
LPM of order 1 at the time of default, discounted to t = 0, that is,
L i (Td ) − Ai (Td ) · (1 + r )−Td · 1 Tdi ≤ T , i = unsyst, syst,
i
ML i = E
where 1{Tdi ≤ T} denotes the indicator function, which is equal to 1 if the condition
in the brackets is satisfied. Thus, this risk measure is the discounted unconditional
expected loss in case of default, which takes into account the extent of the default and
thus reflects the amount by which assets are not sufficient to cover liabilities. It can
thus be interpreted as the average amount of money necessary for funding a case of
default during the contract term.
NUMERICAL ANALYSIS
This section presents results of the numerical analysis. We first discuss the relevant
input parameters and the estimation of mortality rates. Second, all three types of
mortality risk—unsystematic risk, adverse selection, and systematic risk—and their
70 THE JOURNAL OF RISK AND INSURANCE
interaction are analyzed with respect to an insurer’s risk situation with special fo-
cus on adverse selection. Third, the impact of mortality risk on the effectiveness of
risk management instruments is illustrated, thereby considering MCBs and natural
hedging.
19
These parameters were subject to robustness checks. Results are similar when considering a
shorter duration of contracts (e.g., T = 30 or T = 25), younger term life insurance policyholders
(e.g., x = 30), or older annuitants (e.g., x = 70, x = 75).
20
This might be considered too low; however, due to the scarcity of data especially at
high ages, a reliable estimation of the parameters above this age is not possible. For ex-
ample, JP Morgan recommends a maximum age of only 89 in the accompanying soft-
ware for its LifeMetrics index (see http://www.jpmorgan.com/pages/jpmorgan/investbk/
solutions/lifemetrics/software).
21
The standard error of Monte-Carlo simulation for the value of the mortality contingent bond
unsyst
at t = 1, Mbond (1), is about 0.0322 (the expected value is approximately 12; the exact standard
error depends on the path considered and values for the standard error lie between 0.0291
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 71
Estimating and Projecting Future Mortality and Adverse Selection in the United King-
dom. Regarding the empirical estimation of mortality rates, the United Kingdom
is chosen as an example of a typical industrialized country due to the availabil-
ity of mortality rates for annuitants deduced from actuarial mortality tables that
are derived from actual insurance data. Hence, the data basis for the estimation of
mortality for both groups of insured (annuitant data and population data used for
term life policyholders) is the number of deaths and exposure to risk for the United
Kingdom from 1950 to 2009 available through the Human Mortality Database (2012)
and the U.K. annuitant mortality from the CMI from 1947 to 2000 as reflected in
the five mortality tables for the years 1947, 1968, 1980, 1992, and 2000.22 The esti-
mated demographic parameters of the BDV (2002a) model are consistent with the
results stated in the original article by Lee and Carter (1992) and the estimated and
forecasted mortality trend kτ is obtained by applying Box–Jenkins time-series anal-
ysis techniques, which indicated an ARIMA (0,1,0) model23 with drift equal to >
φ = −1.5403 (standard error 0.3056); the standard error of ετ is estimated as 2.3474.
Systematic mortality risk is modeled by simulating random realizations of ετ for each
year. As illustrated in Figure 1, this common factor impacts mortality at all ages and
thus leads to dependencies in the number of deaths at each date τ . Figure 1 exhibits
the correlation between the random number of deaths Dx,τ in the year τ = 2020 for
different ages x.24 Without systematic mortality risk, as shown in of Figure 1a, the
number of deaths Dx,τ for different ages x for a given year τ is generally uncorrelated.
Thus, the benefits of risk pooling apply. However, under systematic mortality risk
(Figure 1b), the common factor causes correlations between the number of deaths for
different ages and the correlation coefficient increases with policyholders’ age.
These results can also be confirmed when considering the correlation between the cash
flows for annuities and death benefits for the cohort of, for example, 1947 (annuitants)
and 1977 (life insurance policyholders) over the contract term, that is, for the year
2012–2047, as shown in Figure 2. Without systematic mortality risk, the correlation at
each point in time t is zero, while under systematic mortality risk, negative correlations
between cash flows of annuities and death benefits can be observed, which increase
over the contract term. These correlations between the number of deaths for different
ages caused by systematic mortality risk in general destroy diversification benefits.
This can further be seen in Figure 3, where the coefficient of variation var(X)/E(X)
is displayed, which provides a relative measure of risk by relating the standard
and 0.0354. These values are calculated for an initial coupon C = 1, while for t = 10, it is about
0.0263 (between 0.0239 and 0.0287).
22
The corresponding graphs exhibiting the demographic parameters exp(ax ), bx , and kτ can
be found on the authors’ webpage; for the estimation procedure using a unidimensional
Newton method, we refer to BDV (2002a). Other annuitant data with more data points are
not publicly available; however, the data still imply significant parameter estimates when
calibrating the time series to estimate the effect of adverse selection.
23
The Schwarz as well as the Akaike information criterion indicated a more complex model
for the ARIMA time series. However, subsequent residual analysis using Box–Ljung test as
well as ACF and PACF analysis showed no significant residual autocorrelation.
syst
24
All calculations of the correlations are based on kτ = k̂τ + ετ , that is, the “neutral” scenario.
Results for the longevity and the mortality scenario are qualitatively similar, except that
72 THE JOURNAL OF RISK AND INSURANCE
FIGURE 1
Correlations Between the Random Number of Deaths at Age x and Age y for Life
Insurance Policyholders and Annuitants, Respectively, in the Year τ = 2020
0.6 0.6
c orrela tio
0.4 0.4
c orrela tio
0.2 0.2
n
n
0.0 0.0
ag
ag
35 35
e
li fe
e
40 95 40 95
li fe
45 90 45 90
in
85 85
su
in
50 nt 50 t
an
su
i ta
ra
ui t
80 80
ra
u
nc
55 n 55 n
75 75 an
nc
an
e
70 ag e
e
70 ag e
po
60 60
po
li c
li c
65 65 65 65
ho
y ho
ld
er
ld
er
FIGURE 2
Correlation Between the Cash Flows for Death Benefits and Annuities for the Cohort of
1977 and 1947 at Age x and y Over the Contract Term
0.0
-0.2
correlation
-0.4
unsystematic
-0.6
unsystematic+systematic
FIGURE 3
Measuring Diversification—Coefficient of Variation for Different Ages and Different
Portfolio Sizes Under Unsystematic and Systematic Mortality Risk
10
25
20
8
15
6
10
4
2
5
0
deviation to the expected value. Here, X denotes the random number of deaths for
the cohort of 1947 (annuitants) or 1977 (life insurance policyholders) for different ages
(i.e., different points in time) with and without systematic mortality risk. The results
show that under unsystematic mortality risk, the coefficient of variation decreases
for an increasing portfolio size; that is, the benefits of risk pooling and the law
of large numbers apply. However, under systematic mortality risk, diversification
benefits are limited and the risk reduction achievable through enlarging the portfolio
is considerably reduced.
Regarding adverse selection, the estimation according to Equation (1) indicates that
annuitant mortality rates improve more rapidly than the population mortality rates,
but this greater improvement decreases over time. The estimated intercept α is equal to
−0.0275 (0.0198), the parameter for the relationship between annuitant and population
mortality (β 1 ) is 1.1618 (0.0123), and the interaction term between year τ index and
population mortality is slightly negative with β 2 = −0.0004 (0.0002) (robust standard
errors in parentheses).25 The estimated standard error of residuals e x,t is 0.1292,
correlations are slightly smaller. The year 2020 was used as an example. The results are based
on 100,000 simulation runs and a population of 10,000 policyholders for each age x.
25
The estimates for β 1 and β 2 are significantly different from zero (on the 1 percent and 5 percent
significance level, respectively), while the intercept is not. The inclusion of β 2 additionally
leads to a slight improvement in R2 by 0.1 percentage point. Note that τindex = 1950 − τ , where
1950 is the first year for which mortality data are used and τ is the year under consideration.
The R2 indicates that more than 98 percent of the variance of annuitant mortality can be
74 THE JOURNAL OF RISK AND INSURANCE
which are also taken into account for each year t and age x in forecasting. In case the
insurer is not be able to perfectly account for adverse selection (“adverse selection
misestimated”), the parameters of Equation (1) are misestimated such that β 1 =
1, β 2 = 0, and α = −0.2779. Concerning the interaction of adverse selection and
systematic risk, further analysis shows that the correlation between lives implied
by systematic mortality risk is reduced through adverse selection, which is due to
the difference between the mortality experience of the population and annuitants
induced by mortality heterogeneity and adverse selection.
The Impact of Systematic Mortality Risk on an Insurer’s Risk Situation. Systematic mor-
tality risk is considered based on different scenarios, where Figure 4 displays results
explained through the model, which is to be expected, since impact factors for mortality
rates of annuitants and the population should generally be similar.
26
The exact portfolio composition for which the insurer’s risk is minimized depends on input
parameters and contract characteristics. For example, if the term life insurance is financed
through a single premium, the risk level for a portfolio with only term life insurance is higher
as compared to a portfolio with only annuities.
TABLE 3
Overview of Assumptions on the Insurer’s Pricing and Risk Measurement
Risk Measure-
Notation in figures Pricing and Reserving Fair Annuity ment
pop pop
q xA = q x q xA = q x
Without adverse selection “unsystematic risk” pop 748 pop
q xL = q x q xL = q x
unsyst fair unsyst fair
PL = PL ; SP A = SP A
pop
q xA = q x q xA = q xpop,syst
“unsystematic risk + pop 748a
q xL = q x q xL = q xpop,syst
systematic risk”
syst fair syst fair
PL = (1 + δ) · PL SP A = (1 + δ) · SP A
a nn∗ : α=0, β1 =1, β2 =0,
qx , adv. sel. misestimated
q xA = 688 q xA = q xann
With adverse selection “unsystematic risk + q xann , if adv. sel. perfectly estimated pop
pop 663 q xL = q x
MORTALITY RISK
adverse selection” q xL = q x
unsyst fair unsyst fair
PL = PL ; SP A = SP A
a nn∗ : α=0, β1 =1, β2 =0,
AND ITS
q xL = q x
+ systematic risk”
syst fair syst fair
PL = (1 + δ) · PL ; SP A = (1 + δ) · SP A
a
Systematic mortality risk is considered in the premium through the loading δ.
IN LIFE INSURANCE
75
76 THE JOURNAL OF RISK AND INSURANCE
FIGURE 4
Probability of Default and Mean Loss Under Different Types of Mortality Risk for a
Longevity Scenario
0.3 0.4
0.3
in 100 T
0.2
in %
0.2
0.1
0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
0.3 0.4
0.3
in 100 T
0.2
in %
0.2
0.1
0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
0.3 0.4
0.3
in 100 T
0.2
in %
0.2
0.1
0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
for the longevity scenario.27 Without adverse selection (Figure 4a), the impact of a
systematic change in mortality on a portfolio of life insurance contracts (fL = 1) is
greater than the impact on a portfolio of only annuities (fL = 0), which is due to the
different types of insured risks.28 However, the impact on a portfolio of annuities is
still not negligible but amounts to an increase of 10.3 percent for the probability of
default and to 9.8 percent for the mean loss. This opposed reaction of life insurance
and annuities in response to systematic mortality risk and the negative correlations
between cash flows induced by systematic mortality risk (see Figure 2) create nat-
ural hedging opportunities that can immunize the risk of an insurance company
against changes in mortality (at the intersection points, where the risk level remains
unchanged despite the unexpected common factor impacting mortality).
Further analyses for the neutral scenario and the mortality scenario showed that even
though the mean life expectancy is not impacted in case of the former, life insurance
contracts are considerably more sensitive toward systematic mortality risk.29 This
result is also supported by the mortality scenario, where the probability of default
increases from 0.03 percent (in Figure 4a) to almost 2 percent for a portfolio with only
life insurance. Thus, while under unsystematic mortality risk, a portfolio of only life
insurance contracts implies a lower risk level than annuities in the considered exam-
ples, the sensitivity of life insurance toward systematic mortality risk is considerably
higher than a portfolio of annuities.
The Impact of Adverse Selection on an Insurer’s Risk Situation. Regarding the impact
of adverse selection on an insurer’s risk situation, which is induced by mortality
heterogeneity among individuals and asymmetric information between insurer and
insureds, in addition to Figure 4b, Figure 5 exhibits the relative change in the risk of
an insurance company due to the presence of adverse selection (perfectly estimated
or misestimated) as compared to the case where only unsystematic risk is included.
When comparing the difference between the case where only unsystematic risk is
considered (“unsystematic”) and the case with adverse selection (estimated perfectly
or misestimated, “unsystematic + adverse selection”), Figures 4b and 5 show that the
risk level considerably increases when taking into account adverse selection. This is
true even if adverse selection is perfectly estimated by the insurer as illustrated in
Figure 5, line “adverse selection perfectly estimated.” In this case, the difference to
the situation with only unsystematic risk still constitutes an increase of 7.8 percent in
27
First, the longevity scenario corresponds to a mean increase in the remaining life expectancy
of a 65-year-old man of about 1.9 years in the year 2012 from 18.5 to 20.4 years. Second,
the mortality scenario implies a mean decrease in the remaining life expectancy of about
1.8–16.7 years. Third, the realization of ε τ is not restricted in the “neutral” scenario, and since
E(ετ ) = 0, the mean life expectancy is not impacted.
28
Life insurances constitute a low-probability risk, which are more heavily impacted by a
change in mortality rates than annuities, which constitute a high-probability risk (see, e.g.,
Gründl, Post, and Schulze, 2006).
29
For a portfolio with only annuities, in contrast, the risk level decreases slightly due to the
inclusion of a loading for systematic mortality risk. Thus, in this case, the required premium
is sufficiently high to cover the costs of systematic mortality risk for a portfolio with only
annuities.
78 THE JOURNAL OF RISK AND INSURANCE
FIGURE 5
Maximum Range of Risk Due to Adverse Selection (Figure 4b) When Adverse Selec-
tion Is Misestimated or Estimated Perfectly (Difference Between “Unsystematic” and
“Unsystematic + Adverse Selection” in Figure 4b(i)) and 4b(ii), Respectively)
40 40
in %
in %
20 20
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
fraction of life insurance f L fraction of life insurance f L
adverse selection adverse selection
misestimated perfectly estimated
Note: The relative change is calculated as the relative increase in the risk measure due
unsyst
to adverse selection, for example, for the probability of default, (PDwith a dver se selection −
unsyst unsyst
P Dwithout a dver se selection )/PDwithout a dver se selection .
the default probability for a portfolio of annuities (fL = 0). For mixed portfolios, the
increase in risk due to the inclusion of perfectly estimated adverse selection can be
even up to 21.7 percent in case of the mean loss. The risk level increases by a fairly
large amount if adverse selection is misestimated, which is mainly due to a lower
than predicted mortality in the annuity portfolio, leading to a greater than expected
cash outflow. For example, for a portfolio with only annuities (fL = 0), the default
probability increases by about 30 percent (see Figure 5, left graph, “adverse selection
misestimated”). The increase in the mean loss for the same portfolio even corresponds
to more than 35 percent and can rise to almost 60 percent for mixed portfolios (see
Figure 5, right graph), which emphasizes the importance of properly forecasting
not only the mortality of the population as a whole, but especially the relationship
between annuitant mortality and population mortality, as an underestimation of
annuitant mortality leads to severely increased risk.
Thus, adverse selection, even if it can be perfectly forecasted and taken into account in
pricing, can considerably increase the risk level of an insurance company, especially
when considering mixed portfolios. However, the results also emphasize that the
impact of adverse selection can significantly be decreased through a reliable forecast
of the relationship between annuitant mortality and mortality of the population as a
whole (and thus the mortality of term life insurance policyholders), which stresses the
importance of developing models for better forecasting this relationship. Including
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 79
The Impact of Basis Risk on the Effectiveness of MCBs in Reducing the Risk Level. In
this subsection, we examine the effectiveness of MCBs for reducing the risk level of
an insurance company in the absence of systematic mortality risk. The use of MCBs
regarding the impact of systematic mortality risk is analyzed in the second subsection.
Figure 6 shows results for different assumptions regarding the MCB and mortality risk
for the probability of default and the mean loss. The line “without adverse selection
(no basis risk)” shows that the effectiveness of an MCB under “ideal” circumstances, in
which the probability distribution of mortality for the population underlying the MCB
and for the hedged insurance portfolio is identical, is examined (i.e., without basis
risk). In this setup, any deviations in the mortality of the hedged and the underlying
reference population are only due to unsystematic deviations in realized mortality
30
When considering a portfolio with younger term life insurance policyholders (x = 30), older
annuitants (x = 70), and a shorter duration of contracts (T = 30), the impact of adverse
selection is further increased. For example, for a portfolio consisting of 50 percent term life
insurance and 50 percent annuities, the mean loss increases by about 100 percent if adverse
selection is misestimated, and by about 13 percent if it is perfectly estimated as compared
to 73 percent and 9 percent, respectively. In addition, all results shown are based on a single
portfolio of life insurance contracts consisting of annuities as well as term life insurance,
where all policyholders belong to the same cohort. Taking into account continuing business
activity, that is, the repeated sale of insurance contracts in which the current mortality can be
more fully acknowledged, may decrease the effects explained above.
31
The size of the loading does not substantially impact the results as shown in sensitivity
analyses.
32
This number is somewhat arbitrary, but since we are merely interested in the relative effec-
tiveness of an MCB under different assumptions concerning mortality, the different scenarios
have to be comparable in terms of amount of purchased hedging instruments.
33
For instance, for a portfolio with 50 percent annuities and 50 percent term life insurance, the
unsyst
insurance company purchases bonds with a total volume of Mbond = 0.5 · 10,000 · 1,000 =
5 Mio, which is equal to a total coupon payment of 5,000 · C = 5,000 · 75 = 375,000 (to be
weighted with the percentage of survivors in the underlying population).
80 THE JOURNAL OF RISK AND INSURANCE
FIGURE 6
The Impact of Risk Management Using MCBs Under Different Assumptions Concerning
Adverse Selection Without Systematic Mortality Risk
40 40
rel. change in %
rel. change in %
30 30
20 20
10 10
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Note: The relative change shown by the dotted line is the relative reduction in the risk
measure achievable through the use of MCBs, for example, for the probability of default,
unsyst unsyst unsyst
(P Dwithout MC B − P Dwith MC B )/P Dwith MC B .
between the population as a whole underlying the MCB and mortality within the
insurance portfolio.
The results show that a significant risk reduction can be achieved through the use of
MCBs. For a portfolio consisting of only annuities (fL = 0), for which MCBs with a
unsyst
total volume of Mbond = 10 Mio are purchased, the probability of default is reduced
by −27.8 percent (i.e., −0.05 percentage points) and the mean loss by −43.3 percent
(i.e., −94 T). This observation indicates that MCBs are more effective in reducing the
severity of default than in reducing the frequency of default. In terms of a relative
risk reduction, the effectiveness of MCBs can be further enhanced through portfolio
composition, despite purchasing fewer MCBs. In particular, for the probability of
default, the maximum risk reduction of −32.2 percent is achieved for a portfolio
with 70 percent annuities (fL = 30 percent), for which MCBs with a total volume of
unsyst
Mbond = 0.7 · 10 T · 1,000 = 7 Mio are purchased. With respect to the mean loss, the
use of MCBs leads to the highest risk reduction of −50.2 percent for a portfolio with
70 percent annuities (fL = 30 percent).
Turning to the effectiveness of MCBs in reducing the level of risk of the insurance
company under basis risk, the results vary depending on the insurer’s ability to es-
timate adverse selection. For a portfolio composed only of annuities (fL = 0), that is,
a typical pension fund, for example, without basis risk, MCBs imply a reduction of
27.8 percent in the probability of default, while the probability of default can only
be decreased by 19.4 percent in the case of a misestimated adverse selection used
in pricing, which constitutes a significant loss in efficiency as compared to the case
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 81
without adverse selection. However, if the insurer is able to estimate adverse selection
perfectly and takes this knowledge into account in pricing, the loss in efficiency com-
pared to the case where no basis risk is included can be reduced, but only by around
3 percentage points in the case of an annuities portfolio. Thus, this result emphasizes
the importance of accounting adequately for basis risk effects when determining the
amount of risk management needed to achieve a desired risk level.
Due to the higher level of risk and the loss in efficiency of MCBs under basis risk, more
MCBs need to be acquired to achieve the same amount of risk reduction in the presence
of basis risk, which comes with greater cost for transferring this risk to the capital
market. To gain an impression about the risk management costs associated with basis
risk (if adverse selection effects are perfectly accounted for), one can calculate the
additional volume of MCBs needed to reach the same level of risk as when basis risk
is absent, that is, when the probability distributions of the mortality of annuitants
and term life insurance policyholders are identical. For example, for a portfolio of 50
percent term life insurance (fL = 0.5) and 50 percent annuities, MCBs with a volume
unsyst
of Mbond = 0.5 · 10,000 · 1,597 = 7.99 Mio are needed under basis risk to achieve the
same probability of default that would otherwise be achieved through purchasing
unsyst
bonds with a volume of only Mbond = 0.5 · 10,000 · 1,000 = 5 Mio when no basis
risk is modeled. This corresponds to an increase in the volume of risk management
of almost 60 percent. For the mean loss, the volume has to be increased by even 65.6
unsyst
percent to Mbond = 0.5 · 10,000 · 1,656 = 8.28 Mio. For a portfolio with annuities
only (fL = 0), an increase in the volume of MCBs by 36.3 percent (41.0 percent) for the
probability of default (mean loss) would be necessary.
The Impact of Basis Risk on the Effectiveness of MCBs for Hedging Systematic Risk. In
the literature, MCBs are also described as an important tool in reducing the impact
of systematic mortality risk. Here, the longevity scenario is considered since the
“survivor bond” by Blake and Burrows (2001) was proposed for hedging the longevity
risk inherent in annuities and pensions. The results for this analysis are displayed in
Table 4 for a portfolio consisting of only annuities. In the case without basis risk,
MCBs can be used to reduce the impact of systematic mortality risk on an insurer’s
risk situation. This is particularly evident for the mean loss, which in the presence of
systematic mortality risk can be reduced by almost 40 percent in the present setting
syst
by purchasing MCBs with a volume of Mbond = 10,36 Mio, using the same coupon as
in the case of unsystematic mortality risk. This result confirms the previous finding
that MCBs prove more useful in hedging the severity of default as compared to
the frequency of default. The risk reduction effect is almost as strong in the case
where adverse selection is perfectly estimated and priced by the insurer (reduction
of 38 percent), even though the implied change in mortality differs for the hedged
population and the reference population underlying the MCB.
However, if adverse selection is misestimated and not perfectly taken into account in
pricing, the effectiveness of MCBs is considerably dampened. An additional analysis
of mixed portfolios further shows that under unexpected low mortality, the insurer’s
risk level is further decreased, since unexpected low mortality leads to lower payout
82 THE JOURNAL OF RISK AND INSURANCE
TABLE 4
Probability of Default and Mean Loss for a Portfolio of Only Annuities (fL = 0) Including
Systematic Mortality Risk (Longevity Scenario)
TABLE 5
Fraction of Life Insurance at Which the Impact of Systematic Mortality Risk on the Risk
Situation of an Insurance Company Is Immunized
for term life insurance contracts and, at the same time, a higher payments from the
MCB.34
34
In contrast, in case of an opposed change in mortality rates, for example, because the rate
of mortality improvement is overestimated, the purchase of MCBs would lead to a further
increase in the risk, since payments from the MCBs decrease at the same time that payouts
for term life insurance increase.
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 83
portfolio composition, the risk level remains unchanged for the modeled unexpected
changes in mortality. The immunization is thereby driven by the negative correlation
between cash flows for death benefits and for annuities over the contract term (see
Figure 2), which is especially pronounced for later contract years and contributes to
the immunizing effects utilized in natural hedging.
If no adverse selection is assumed in pricing and risk measurement, the insurance
company can eliminate the impact of unexpected low mortality on the probability of
default by signing about 27.2 percent life insurance contracts in case of the longevity
scenario (30.0 percent when using the mean loss as the relevant risk measure). In case
of the mortality scenario, the optimal fraction of life insurance contracts is reduced
to 21.3 percent (20.3 percent for the mean loss). The presence of adverse selection
leads to a change in the optimal fraction of life insurance contracts, at which the
risk of an insurance company is immunized against unexpected low mortality. This
is particularly evident in case of the longevity scenario. If adverse selection is not
correctly priced, the optimal fraction is reduced from 27.2 percent to 18.7 percent for
the probability of default and from 30.0 percent to 22.0 percent for the mean loss.
Thus, despite the greater implied change in life expectancy of annuitants, fewer life
insurance contracts are needed to eliminate this effect. In the case of a perfectly fore-
casted adverse selection, the fraction of life insurance contracts is further decreased to
15.5 percent and 18.6 percent, respectively. For the mortality scenario, in contrast, the
impact of adverse selection is overall negligible and the optimal fraction of life insur-
ance remains almost constant. Compared to the longevity scenario, fL is considerably
lower for the mortality scenario, which can be explained by the greater sensitivity of
life insurance toward systematic mortality risk in the mortality scenario. Our immu-
nization approach is similar to Gatzert and Wesker (Forthcoming) and Wang et al.
(2010) and can also be compared to the results in Cox and Lin (2007). The results in
Wang et al. differ from the results found in the present setting in that the fraction of
life insurance contracts is generally higher. For instance, for a 10 percent mortality
shift, the optimal product mix proportion, i.e., the optimal proportion of life insurance
liabilities, lies between 30 percent and 35 percent life insurance.35 In Cox and Lin,
past mortality shocks are used in an example to illustrate the effectiveness of natural
hedging. First, a positive shock to mortality is modeled based on the average life
expectancy improvement rate in historical mortality tables. In this case, the deviation
between the present value of benefits and premiums for both products can be elim-
inated completely through portfolio composition by signing about equal amounts
of life insurance and annuity business. Second, as an example for a bad shock, two
different epidemic scenarios are modeled based on the 1918 flu epidemic. While in
one scenario, natural hedging can contribute to a considerable reduction in cash flow
volatility, in the other scenario an optimal ratio of annuity to life insurance, business
of about 80 percent to 90 percent is found. In Gatzert and Wesker, slightly higher
optimal portfolio fractions are found than those in the present analysis. The differ-
ences regarding the optimal portfolio mix in the previous literature and the present
findings generally arise due to various reasons. First, the definition of the product
mix proportion differs, for example, using the value of liabilities (Wang et al., 2010)
35
The product mix proportion ωlife is thereby defined as ωlife = V life /.V, with V as the total
liability and V life as the life insurance liability (see Wang et al., 2010, p. 476).
84 THE JOURNAL OF RISK AND INSURANCE
36
Wang et al. (2010) also acknowledge the importance of accounting for adverse selection
effects. They conduct sensitivity analysis through implementing different mortality shifts for
annuitants and life insurance and do not focus on a separate model.
TABLE 6
Portfolio Composition and Amount of MCB to Simultaneously Achieve a Certain Risk Level and Immunize This Risk Level Against
Systematic Mortality Risk (Longevity Scenario)
syst
MCB volume Mbond (t) (1- 26.1%) (1–28.1%) (1–17.2%) (1–20.4%) (1–16.1%) (1–17.6%)
· 10 T · 1,378 = · 10 T · 2,574 = · 10 T · 3,498 = · 10 T · 4,491 = · 10 T · 2,551 = · 10 T · 3,646 =
10.18 Mio 18.51 Mio 28.96 Mio 35.75 Mio 21.40 Mio 30.04 Mio
AND ITS
and thus in the presence of basis risk, the optimal fraction of life insurance contracts
decreases by 10 percentage points in case of the PDsyst , while the amount of MCBs
needs to be increased to 21.40 Mio to achieve and immunize the desired risk level.
When adverse selection is not correctly forecasted, the volume of MCBs has to be
increased even further to 28.96 Mio, while the fraction of life insurance is slightly
higher than in the case when adverse selection was perfectly forecasted. These results
are in line with those in the previous subsection, in that under adverse selection,
a smaller fraction of life insurance contracts is needed to eliminate the impact of
unexpected low mortality on the insurer’s risk situation and that more MCBs are
needed to achieve a certain risk level.
Sensitivity Analyses
To examine the robustness of the results with respect to input parameters, sensitivity
analyses were conducted. Concerning the loading δ for systematic mortality risk, we
followed Gründl, Post, and Schulze (2006) and changed the loading to 0.5–5 percent
(instead of 1 percent). The loading has a significant impact on the risk situation of
the insurance company under systematic mortality risk. By demanding a loading of
δ = 5 percent instead of δ = 1 percent, the probability of default can be reduced in
the longevity scenario by about 30 percent for a portfolio with only annuities and
even by more than 70 percent for a portfolio with only life insurance contracts. When
reducing the loading to 0.05 percent, the effects are reversed and the insurer’s risk level
increases under systematic mortality risk. However, the size of the effects is smaller
as compared to a loading of δ = 5 %; for example, the mean loss for a portfolio with
only annuities increases by about 5 percent in the longevity scenario.
When reducing the loading λ of the MCB from 35 to 20 bp, the premium under
syst syst
systematic mortality risk decreases from x,T = 1, 036 to x,T = 1, 020, which results
in a lower risk of the insurance company under systematic mortality risk when MCBs
are used for risk management. However, the effect is rather small. In contrast, a higher
fraction of high-risk assets in the investment portfolio considerably increases the risk
situation of the life insurer. For example, decreasing the fraction of low-risk assets
α from 80 percent to 50 percent almost doubles the risk of an insurance company
for a portfolio with only annuities and can more than triple it for mixed portfolios.
The portfolio composition for which the risk of the insurance company is immunized
against the modeled longevity scenario still ranges between 20 percent and 30 percent
life insurance in the case without adverse selection, while under adverse selection,
the fraction of life insurance contracts needed for an immunization decreases for
a riskier asset strategy. Thus, as shown in Gatzert and Wesker (Forthcoming), the
asset allocation should be taken into account when determining the optimal portfolio
composition. Concerning the impact of adverse selection, the relative increase in risk
due to adverse selection is less pronounced for a riskier asset allocation. However, if
adverse selection is misestimated, the increase in risk can still amount to more than
30 percent and is thus not negligible. Furthermore, the effectiveness of MCBs for
lowering the risk level of an insurance company is slightly reduced for a riskier asset
strategy.
Finally, setting the dividend to shareholders to zero (re = 0 percent) considerably
decreases the risk level of an insurance company as reserves can be built up faster. For
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 87
a portfolio with only annuities and without adverse selection, for instance, the mean
loss and the probability of default are reduced by about 20 percent for re = 0 percent.
In addition, the impact of adverse selection decreases as well, but to a minor extent.
SUMMARY
In this article, we examine the impact of three different components of mortal-
ity risk—unsystematic mortality risk, adverse selection, and systematic mortality
risk—as well as the basis risk in longevity hedges resulting from adverse selection
on a life insurer’s risk level using U.K. data. Furthermore, we study the effectiveness
of two risk management strategies, including natural hedging and the purchase of
MCBs, in the case of a two-product life insurance company offering annuities and
term life insurance contracts.
Our results show that under unsystematic mortality risk, the insurer’s risk level
can generally be reduced by means of portfolio composition. Taking into account
adverse selection in addition to unsystematic mortality risk implies a substantial
increase in the risk of an insurance company. However, the impact of adverse selection
can be considerably reduced through a correct forecast of the relationship between
life insurance policyholder mortality and annuitant mortality, that is, under perfect
information about adverse selection. Concerning the impact of systematic mortality
risk, term life insurances are much more strongly affected than are annuities, which
is due to the different types of risks insured.
Turning to the effect of the three mortality risk components on an insurer’s risk
management, our findings demonstrate that MCBs can contribute to a major reduction
in the risk level, even in the presence of basis risk, that is, if the implied change
in mortality is not identical for the underlying and the hedged population due to
mortality heterogeneity and adverse selection effects. However, the extent of the risk
reduction achievable with a certain volume of MCBs decreases substantially due
to basis risk and if adverse selection is not correctly forecasted. Furthermore, an
improvement in the efficiency of MCBs for mixed portfolios of term life insurances
and annuities can be observed. Thus, our results emphasize the importance of these
factors, which should be taken into account when determining the volume of risk
management activities needed to achieve a desired safety level.
Regarding the usefulness of risk management for reducing the impact of system-
atic mortality risk, our findings show that the effectiveness of MCBs is not severely
hampered if adverse selection is correctly accounted for, that is, under perfect in-
formation by the insurance company about annuitant mortality. This is true despite
the presence of basis risk. The impact of unexpected low mortality on the mean loss,
that is, the severity of default, can be reduced by about one-third through the use
of MCBs if adverse selection is assumed absent or forecasted perfectly. Turning to
the effectiveness of natural hedging under systematic mortality risk for eliminat-
ing the impact of an unexpected change in mortality, our observations show that
despite the different implied level of mortality as well as speed of mortality improve-
ment in the insurance portfolio, natural hedging can still be a feasible and important
risk management tool against unexpected changes in mortality. However, in partic-
ular, in the longevity scenario, adverse selection needs to be taken into account in
88 THE JOURNAL OF RISK AND INSURANCE
REFERENCES
Barbarin, J., 2008, Heath–Jarrow–Morton Modeling of Longevity Bonds and the Risk
Minimization of Life Insurance Portfolios, Insurance: Mathematics and Economics,
43(1): 41-55.
Bayraktar, E., and V. Young, 2007, Hedging Life Insurance With Pure Endowments,
Insurance: Mathematics and Economics, 40(3): 435-444.
Biffis, E., M. Denuit, and P. Devolder, 2010, Stochastic Mortality Under Measure
Changes, Scandinavian Actuarial Journal, 2010(4), 284-311.
Björk, T., 2004, Arbitrage Theory in Continuous Finance, 2nd edition (Oxford: Oxford
University Press).
Blake, D., and W. Burrows, 2001, Survivor Bonds: Helping to Hedge Mortality Risk,
Journal of Risk and Insurance, 68(2): 339-348.
Blake, D., A. Cairns, K. Dowd, and R. MacMinn, 2006, Longevity Bonds: Financial
Engineering, Valuation and Hedging, Journal of Risk and Insurance, 73(4): 647-672.
Brouhns, N., M. Denuit, and J. K. Vermunt, 2002a, A Poisson Log-Bilinear Regression
Approach to the Construction of Projected Lifetables, Insurance: Mathematics and
Economics, 31(3): 373-393.
Brouhns, N., M. Denuit, and J. K. Vermunt, 2002b, Measuring the Longevity Risk in
Mortality Projections, Bulletin of the Swiss Association of Actuaries, 2002(2): 105-130.
Cairns, A., D. Blake, P. Dawson, and K. Dowd, 2005, Pricing the Risk on Longevity
Bonds, Life and Pensions, October: 41-44.
Cairns, A., D. Blake, K. Dowd, G. D. Coughlan, D. Epstein, A. Ong, and I. Balevich,
2009, A Quantitative Comparison of Stochastic Mortality Models Using Data From
England and Wales and the United States, North American Actuarial Journal, 13(1):
1-35.
Cohen, A., and P. Siegelmann, 2010, Testing for Adverse Selection in Insurance Mar-
kets, Journal of Risk and Insurance, 77(1): 39-84.
Coughlan, G., D. Epstein, A. Ong, A. Sinha, J. Hevia-Portocarrero, E. Gingrich, M.
Khalaf-Allah, and P. Joseph, 2007, LifeMetrics, A Toolkit for Measuring and Man-
aging Longevity and Mortality Risks, Technical Document, JP Morgan Pension
Advisory Group.
Coughlan, G. D., M. Khalaf-Allah, Y. Ye, S. Kumar, A. Cairns, D. Blake, and K. Dowd,
2010, Longevity Hedging: A Framework for Longevity Basis Risk Analysis and
Hedge Effectiveness, Discussion Paper PI-1013, The Pensions Institute.
MORTALITY RISK AND ITS EFFECT IN LIFE INSURANCE 89
Cox, S. H., and Y. Lin, 2007, Natural Hedging of Life and Annuity Mortality Risk,
North American Actuarial Journal, 11(3): 1-15.
Dawson, P., K. Dowd, A. Cairns, and D. Blake, 2011, Survivor Derivatives: A Consis-
tent Pricing Framework, Journal of Risk and Insurance, 77(3): 579-596.
Finkelstein, A., and J. M. Poterba, 2002, Selection Effects in the United Kingdom
Individual Annuities Market, Economic Journal, 112(476): 28-50.
Gatzert, N., and A. Kling, 2007, Analysis of Participating Life Insurance Contracts: A
Unification Approach, Journal of Risk and Insurance, 74(3): 547-570.
Gatzert, N. and H. Wesker, Forthcoming, The Impact of Natural Hedging on a Life
Insurer’s Risk Situation, Journal of Risk and Insurance.
Gründl, H., T. Post, and R. Schulze, 2006, To Hedge or Not to Hedge: Managing
Demographic Risk in Life Insurance Companies, Journal of Risk and Insurance, 73(1):
19-41.
Hanewald, K., 2011, Explaining Mortality Dynamics: The Role of Macroeconomic
Fluctuations and Cause of Death Trends, North American Actuarial Journal, 15(2):
290-314.
Hanewald, K., J. Piggot, and M. Sherris, 2011, Individual Post-Retirement Longevity
Risk Management Under Systematic Mortality Risk, Australian School of Business
Research Paper No. 2011ACTL10.
Hanewald, K., T. Post, and H. Gründl, 2011, Stochastic Mortality, Macroeconomic
Risks, and Life Insurer Solvency, Geneva Papers on Risk and Insurance—Issues and
Practice, 36(3): 458-475.
Human Mortality Database, 2012, University of California, Berkeley, and Max Planck
Institute for Demographic Research (Germany). World Wide Web: http://www.
mortality.org or www.humanmortality.de (accessed January 16, 2012).
Lee, R. D., and L. R. Carter, 1992, Modeling and Forecasting U.S. Mortality, Journal of
the American Statistical Association, 87(419): 659-671.
Li, J., and M. Hardy, 2009, Measuring Basis Risk Involved in Longevity Hedges,
Presented at Fifth International Longevity Risk and Capital Markets Solutions Con-
ference.
Lin, Y., and S. H. Cox, 2005, Securitization of Mortality Risks in Life Annuities, Journal
of Risk and Insurance, 72(2): 227-252.
Lin, Y., and S. H. Cox, 2008, Securitization of Catastrophe Mortality Risks, Insurance:
Mathematics and Economics, 42(2): 628-637.
Milevsky, M. M., and D. Promislow, 2003, Can Life Insurance Be Used to Hedge
Payout Annuities? Part 1: Modeling Longevity Risk, Working Paper.
Ngai, A., and M. Sherris, 2011, Longevity Risk Management for Life and Variable
Annuities: The Effectiveness of Static Hedging Using Longevity Bonds and Deriva-
tives, Insurance: Mathematics and Economics, 49(1): 100-114.
Plat, R., 2009, Stochastic Portfolio Specific Mortality and the Quantifica-
tion of Mortality Basis Risk, Insurance: Mathematics and Economics, 45(1):
123-132.
90 THE JOURNAL OF RISK AND INSURANCE
Stevenson, M., and A. Wilson, 2008, Mortality of Public Sector Scheme Pensioners
2005–2007 Update, Presented at the Institute of Actuaries of Australia 4th Finan-
cial Services Forum.
Sweeting, P., 2007, Pricing Basis Risk in Survivor Swaps, Discussion Paper PI-0622, The
Pensions Institute.
Wang, J. L., H.C. Huang, S. S. Yang, and J. T. Tsai, 2010, An Optimal Product Mix for
Hedging Longevity Risk in Life Insurance Companies: The Immunization Theory
Approach, Journal of Risk and Insurance, 77(2): 473-497.
Wetzel, C., and H. J. Zwiesler, 2008, Das Vorhersagerisiko der Sterblichkeitsent-
wicklung—Kann es Durch Eine Geeignete Portfoliozusammensetzung Minimiert
Werden? Blätter der DGVFM, 29(1): 73-107.
Wills, S., and M. Sherris, 2010, Securitization, Structuring and Pricing of Longevity
Risk, Insurance: Mathematics and Economics, 46(1): 173-185.
Copyright of Journal of Risk & Insurance is the property of Wiley-Blackwell and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.