Agenda, Volume 10, Number 1, 2002, pages xx-xx
Agenda, Volume 10, Number 1, 2002, pages xx-xx
Impact of Genetic Testing on Life Insurance
Richard Heaney and David Pitt
he Human Genome project generates immense interest in the scientific
community though there are also important issues for the business
community, particularly insurance companies. The dramatic advances in
our understanding of the human genetic code, or human genome, affect our
understanding of the determinants of human longevity and this is critical to the
profitability of life insurance contracts.
Insurance plays an important role in our economy. For example, there were
42 life insurance companies managing approximately AUD $188 billion as at 30
June 2002 and these same companies received AUD $38.1 billion in premium
income over for the year ended 30 June 2002. Life insurance contracts are
increasingly being sold in combination with superannuation where the contract
forms part of the superannuation package. For example up to 85 per cent of all
life office assets and 90 per cent of the premiums were classified as
superannuation business in 2001 (APRA, 2002). Regardless of whether an
individual submits an application form directly to a life insurance company as part
of a stand alone policy, or indirectly via superannuation, the insurance company
faces the question of deciding whether they wish to sell insurance to this
individual.
While life insurance policies can take a number of forms, an essential feature
of these contracts is that they promise the payment of a given amount to certain
beneficiaries when the insured dies. The time of death is critical to the pricing and
profitability of these contracts. For insurance contracts to be profitable they must
be priced so that invested premiums generate sufficient reserves to meet the
payment of death benefits when they fall due.
Failure to adequately model the impact of our increased understanding of the
human genome could have a dramatic impact on the profitability of insurance
contracts. The mapping of the human genome and the rapid development of
genetic testing means that people have access to greater knowledge about their
health and longevity yet this information may not be freely available, particularly
to insurers and annuity providers (Hoy and Polborn, 2000). If individuals have
more information about their health than insurance companies, this can complicate
the pricing of life insurance contracts and annuities. This one-sided access to
information is often referred to as information asymmetry and at its worst
information asymmetry can lead to market failure (Akerlof, 1970). Further, where
the insurer is unable to accurately assess the risk of an applicant it is possible that
prices will be set too high. The ultimate result could be that only those applicants
Richard Heaney is a Reader in Finance and David Pitt is a Lecturer in Actuarial Studies
in the School of Finance and Applied Statistics at the Australian National University.
T
1002 Richard Heaney and David Pitt
most likely to require a payout will purchase the product resulting in low
profitability or losses. This is a critical problem in the pricing of life insurance
products. Life insurers generally attempt to protect themselves through pricing for
average risks though this may become more difficult where information about
health and longevity is restricted in an asymmetric manner. An objective of this
paper is to show that failure to adequately model the impact of our increased
understanding of the human genome – on choices made by consumers, and on the
pricing policies of insurance companies – could have a dramatic impact on the
profitability of insurance contracts.
While Doherty and Thistle (1996) and Hoy and Polborn (2000) provide
economic analysis of the impact of information about the human genome, the
work of actuaries such as Macdonald (1997) provides insight into the problems
that actuaries face in the day-to-day pricing of life insurance contracts. We apply
the MacDonald (1997) model to gain further understanding of the impact of
genetic research on Australian life insurance contracts – in particular, the effects
on profitability of adverse selection by consumers who discover themselves to be
high risk, and who subsequently purchase more insurance than would otherwise
have been the case. The following section provides a brief review of the literature.
It is followed by two sections that respectively describe the model used in the
simulation of the insurance problem faced by Australian insurance companies, and
the results of the simulation. Conclusions are drawn in the final section of the
article.
Insurance Theory
Transactions costs are an important part of the market for life insurance (Gravelle
and Rees, 1985). One of these costs is the cost of identifying the true risk of the
individual. Individuals could be classified into broad categories such as good risk,
where there is little risk of a claim, and bad risk, where the probability of a claim
is high. Insurers generally assess the probability of death through analysis of
objective factors, such as age and occupation, and subjective factors, such as
exercise, diet and habits such as smoking. Each applicant is assessed for these
factors and a premium estimated and charged. Where it is impossible to identify
the risk associated with a group of individuals, adverse selection can lead to bad
risks driving out good risks. This effect could arise where the insurer initially sets
the insurance premium too high for the best risks. Given the high price the best
risk group choose not to insure, leaving only those representing poorer risks for
the insurer. The difficulty for the insurer is that the premium does not allow for
exclusion of the best risk group – with the loss of the best risk group the current
premium is set too low to cover the expected costs of the contract. Over time as
the level of claims follows the actual longevity of the insured group, the insurer is
forced to raise the level of premiums. Again, the better risks in the group choose
not to insure at this higher price and so the cycle continues until the good risks are
priced out of the market and the insurer faces ruin. In effect, the insurer faces a
trade off between the costs of obtaining better information about individuals and
Impact of Genetic Testing on Life Insurance 1003
thus pricing insurance contracts properly and the reduced profitability arising from
the impact of bad risks driving out good risks when insurance premiums do not
properly reflect the risk of the insured group. A regulatory response to this type of
problem occurring in the health insurance area has been the use of community
rating in the pricing of health insurance.
There may be signalling effects arising from the decision of insurance
companies to sort into risk categories. A rational response from low risk
applicants is to signal quality in order to support a separating equilibrium and
receive lower premiums; from high risk applicants the response is to mix signals
in order to keep a pooling equilibrium, whereby the cost of their insurance is
spread across a larger group (and thus lower premiums).
The impact of genetic information on the life insurance market is important
though current research suggests that the net welfare effect of genetic testing is not
clear. For example, Hoy and Polborn (2000), extending the model of Doherty and
Thistle (1996), show that the private value of being informed is positive for the
individual while the social value of the information could be either positive or
negative. Their model includes the impact of genetic testing and it is assumed that
there is an incentive for an individual to undertake the test to obtain further
information about their health with the knowledge that the insurer does not have
access to this information. Essentially there are three groups in the model, high
risk, low risk and the uninformed. Hoy and Polborn (2000) focus on the benefits
to those in the uninformed group who choose to test, assuming that the current
price of insurance is only attractive to high-risk individuals. If the uninformed
choose the test and are found to be bad risks then they can insure and so they are
better off. If they are good risks then they can choose not to insure. In this partial
equilibrium model the uninformed are better off with the availability of the test
because they have the option to purchase a contract at the old price if they test
positive. There are also spillover effects associated with those who choose not to
take the test where the actual risk of this group differs from the average risk for
the pool. For example the spillover effect is positive if the initially uninformed
(who are tested and subsequently buy insurance) are lower risk than the existing
customers. Hoy and Polborn (2000) argue that this source of asymmetric
information is not unusual as there is legislation in a number of countries
specifically set up to protect the rights of the individual to genetic testing results
and to deny the insurance companies control over this information. This debate
continues in Australia and is surveyed in Otlowski (2002).
Thus the impact of the introduction of genetic testing is not clear. It is argued
by Hoy and Polborn (2000) that if information is distributed symmetrically
between insurers and customers, the introduction of the test is welfare reducing for
consumers assuming that markets stay open (alt hough price may vary) if we
ignore the benefits that arise from the use of genetic testing including improved
medical treatment of these conditions. With asymmetric information the net
welfare effects of the test could be positive, negative or a mixed case where those
who undertake the test gain and those that choose not to undertake the test lose.
1004 Richard Heaney and David Pitt
This theoretical work on insurance and asymmetric information helps to
understand the implications of genetic testing and highlights the impact of
asymmetric information but it tells us little about the actual pricing of life
insurance where genetic testing is available to individuals but not to insurers. In
the following section we focus on empirical testing. Actuaries have an important
role to play in valuing these instruments through statistical modelling and we use
the Markov model developed by Macdonald (1997) in analysis of the impact of
genetic testing on breakeven insurance premiums. As indicated above, knowledge
of the mortality of a group of individuals is critical to the accurate pricing of
insurance offered to the members of this group and, given information asymmetry,
genetic testing could have a dramatic effect on the pricing of life insurance.
Actuarial modelling
Actuarial research into the financial impact of genetic testing has gained
momentum in recent years. MacDonald (1997,1999) has published a number of
papers advocating the use of multi-state Markov models to explore the impact of
uncertainty on traditional life insurance products where an individual has access to
information such as genetic testing results but the insurer does not have access to
this information. This is particularly important to insurers where the tests provide
highly predictive information about mortality.
A thorough assessment of the impact of genetic testing involves consideration
of four major factors. The first factor reflects the level of insurance that the
applicant might prefer. The level of insurance could vary with the existence of a
genetic predisposition to a particular disease. It could also vary with whether the
applicant has been genetically tested. A final source of variation lies with whether
the applicant has been genetically tested and found not to have a higher probability
than normal of contracting a particular disease (negative results). The second
factor is the prevalence and predictive accuracy of genetic tests. The third factor
is the proportion of those who have genetic tests and who return a positive result
indicating that they have a higher probability than normal of contracting a
particular disease. The final factor is the extent to which people who receive a
positive genetic test for a particular disease have an increased propensity to
purchase insurance.
The most significant financial impact of genetic testing is thought to occur for
life insurance products such as term insurance and associated riders such as dread
disease insurance. Dread disease insurance is a rider that may be added to a
contract containing death insurance. This rider provides the insured with a benefit
if they incur one of a selection of serious medical conditions listed in the policy.
Genetic testing is more important for term insurance than whole of life
insurance because in the case of whole of life insurance the insured is guaranteed
payment of the amount insured on death. The only question for whole of life
insurers is the timing of the payment. Under a term insurance contract the amount
insured is paid if the insured dies before the term of the contract and so the insurer
prices the contract with a view to both the timing of death and the likelihood of
Impact of Genetic Testing on Life Insurance 1005
death before expiry of the contract. Thus where the insured has a genetic
predisposition for higher mortality they are more likely to purchase a term
insurance contract because the premium is lower for term insurance contracts than
for whole of life contracts. This is because lower risk is ordinarily presented to
the insurer for term insurance contracts.
Commonly term insurance is provided for a period of 10 or 20 years. These
types of insurance contracts are often called risk-based products. In the Australian
setting the majority of the term insurance business is renewed each year, or yearly
renewable. These insurance products give the insured the option to renew the
insurance contract each year and the insurer is obligated to renew the coverage at
the request of the insured as long as there is no material change to the risk
presented by the insured. This arrangement provides considerable potential for
adverse selection against the insurer and so it is important to model the impact on
life insurance contract premiums.
A benchmark Markov model with no genetic test
First, let us consider the case of a customer who purchases life insurance when
genetic testing is ignored. This model will be used as the benchmark case for
analysing the effects of adverse selection on the profitability of term insurance
contracts. Consider a typical individual who purchases insurance cover at age 30
for the first time. This life insurance coverage provides payment of the sum
insured, typically $100,000, on the death of the insured individual and the
payment is made to the estate of the insured individual. We will assume that this
individual has purchased cover for 20 years payable by annual premiums. The
insured renews the policy annually by paying a level annual premium for twenty
years or until earlier death. The situation can be illustrated as in Figure 1 that
shows transitions between three ‘states’. State 1 is where all people who have not
purchased life insurance are situated. If an individual purchases life insurance,
they move from State 1 into State 2. This move between states is called a
transition and in Figure 1 this transition is labelled transition t1. In this model
people who have purchased life insurance, and therefore reside in State 2, can of
course die. Death causes them to move to State 3 and, as shown in Figure 1, make
transition t3. Individuals who have not purchased life insurance, and therefore
reside in State 1, can also die and move to State 3. This movement from the
uninsured state (State 1) to the dead state (State 3) is labelled transition t2. If the
consumer makes an annual decision whether to renew the contract, the transition
probabilities reflect the likelihood of the event (transition) occurring within a
particular year for a person of certain age.
In this three-state model where we ignore the impact of genetic testing the
annual premium paid by an insured individual depends only on two key factors.
The first and most significant is the magnitude of transition intensity t3 – the
mortality of insured individuals. In practice these mortality rates (which vary with
age) are determined by considering the experience of the particular insurer and
1006 Richard Heaney and David Pitt
also by consulting published Australian mortality tables. The other important
factor is the investment income that the insurer is assumed to be able to earn on
the premiums paid by the insured to the insurer. The insurer can earn substantial
amounts of investment income from the prudent investment of premium income.
This is because under term insurance the premiums are paid to the insurer long
before any insurance payment is made, if it is ever made, to the estate of the
insured individual. The assumed rate of investment income is again determined
by reference to the recent investment returns earned by the insurer on the funds
that are backing the relevant insurance portfolio. In this analysis we have ignored
the impact on premiums charged to consumers of commissions and other expenses
that the insurer would ordinarily incur.
Figure 1: Markov Model for Term Insurance in the Absence of Genetic
Testing
A Markov model with genetic testing
We now consider the case where the impact of genetic testing is taken into
consideration in the determination of suitable insurance premiums. MacDonald
(date?) models individuals as occupying a series of states and making transitions
from state to state until they die. The important linkages modelled by MacDonald
are identified in Figure 2. As in the previous diagram, each box in Figure 2
represents a state in which either a consumer or potential consumer of life
insurance could be situated. In any interval of time the consumer is able to make a
transition along any of the arrows. Each transition has a simple description.
Transition t1 occurs when an individual who has not had any genetic testing
purchases insurance. Transitions t2 and t3 occur after a genetic test is performed
t1
State 1
Not Insured
State 2
Insured
State 3
Dead
t2
t3
Impact of Genetic Testing on Life Insurance 1007
on the consumer. The consumer makes transition t2 if the test result is negative
indicating that they do not suffer from the conditions identified in genetic testing.
The consumer makes transition t3 if the test result is positive and the genetic tests
identify the consumer as suffering from the tested ailments. Transitions t4 and t6
occur when an individual purchases life insurance after testing. While the
transition t4 is reserved for those individuals who purchase insurance after
returning a negative genetic test, transition t6 is made when purchasing insurance
for those individuals who returned a positive genetic test. The remaining
transitions (t5, t7, t8, t9 and t10) occur on the death of an individual from any of
the five other states in the model.
Figure 2: Markov Model for Assessing the Impact of Genetic Testing on
Insurer Profitability.
State 1:
Not Insured
No Genetic
Test
State 6:
Dead
State 2:
Insured.
No Positive
Test
State 3:
Tested and
Result is
Negative
State 4:
Tested and
Result is
Positive
State 5:
Insured and
Positive
Test
t1 t2 t3
t4
t5
t6
t7 t8 t9 t10
1008 Richard Heaney and David Pitt
It is clear that the model in Figure 2 is an extension of the model in Figure 1.
States 1, 2 and 3 in the model that ignores genetic testing (Figure 1) correspond to
states 1, 2 and 6 respectively in the model where genetic testing is considered
(Figure 2).
An important feature of this model is that the likelihood of some of the
transitions (those other than death, which obviously vary with age) depend only on
the state currently occupied by the insured. The model ignores information about
past movements between particular states of the model. For example the
probability that an individual who returned a positive genetic test two years ago
will buy insurance in the next week is the same as the probability that an
individual who has just returned a positive genetic test will purchase insurance in
the next week. In Figure 2 this is the same as saying that the chance of moving
from State 4 to State 5 is unrelated to the amount of time spent in State 4.
In addition to considering the probabilities of transitions between states,
consideration must also be given to the payments made where applicable while
residing in states and on the transition between states. In Figure 2 the premiums
are paid to the insurer while the insured is in states 2 and 5. This is because when
an individual is in state 2 or state 5 the individual has life insurance. On transition
from state 2 or state 5 to state 6 (that is on the death of an insured whether or not
that individual had returned a positive genetic test) a payment is made to the estate
of the insured person. This payment is the amount of life insurance purchased by
the insured.
The model aims to determine the impact on insurer profitability of varying
the modelled probabilities of transition between particular states. It is of particular
interest to understand the impact of adverse selection, reflected by the increased
likelihood that individual will purchase insurance if the individual knows that they
have a genetic predisposition to higher mortality than average. This is modelled
by increasing the probability of making the transition t6 to a multiple, greater than
one, of the chance of making transition t5. Another issue of concern to insurers is
the level of insurance sought by those who have returned a positive genetic test
result compared with those who have returned a negative test result or who have
not had a genetic test at all. This can be explored in this model by allowing the
insurance amount that the insured chooses to vary. The relevant insurance amount
is the amount paid on transition from state 5 to the dead state, and it can be
modelled to be double or four times the amount paid on transition from state 2 to
the dead state.
Pricing of Life Insurance Contracts
To gain some idea of the impact of changes in the level of information asymmetry
on the profitability of insurance policies we vary the size of the amount insured
and the probability of purchasing insurance for those who have returned a positive
genetic test and therefore exhibit higher mortality. We then focus on the breakeven
premium for the group of individuals insured. This is the premium charged
to all those insured within a particular group regardless of whether a genetic test
Impact of Genetic Testing on Life Insurance 1009
has been conducted and irrespective of the results of the test where it is
undertaken. The model has been run using Australian mortality data based on the
IA95-97 Life Table. This life table is produced by the Institute of Actuaries of
Australia Mortality Committee and is based on the mortality experience of the
major life insurers operating in Australia during the years 1995 to 1997. The
interest rate is assumed to be 5 per cent per annum compounding continuously in
the model and expenses have been ignored in this analysis.
The projected increase in mortality resulting from a positive genetic test
clearly needs to be estimated for inclusion in the model. This increase is
represented by the difference between the probabilities of making the transitions t9
and t10 and the other transitions to the dead state, namely t5, t7 and t8. For the
purposes of this analysis a 50 per cent increase in mortality has been assumed for
those who are insured and who have returned a positive genetic test. An extreme
value of 0.90 and a less extreme value of 0.30 have been used for the conditional
probability that an individual, given that they have returned a positive genetic test,
will buy insurance.
A summary table of the transition intensities is shown below in Table 1.
Table 1: Transition Intensities used in the Term Insurance Analysis in
the presence of Genetic Testing
Transition Description Transition Intensity
Transition t1 0.50
Transition t2 0.20
Transition t3 0.05
Transition t4 0.05
Transition t6 0.30 and 0.90
Transitions t5, t7 ,t8 at age 30 0.00113
Transitions t5, t7 ,t8 at age 45 0.00139
Transitions t5, t7 ,t8 at age 60 0.00568
Transitions t9 and t10 at age 30 0.00170
Transitions t9 and t10 at age 45 0.00209
Transitions t9 and t10 at age 60 0.00852
Once transition probabilities are assigned it is necessary to select the amount
that individuals would choose to be insured for. With these inputs identified, the
statistical model generates the level of premium required for the insurer to break
even. This model provides considerable flexibility in helping insurers to make
informed decisions about the level of premiums that should be charged. The
1010 Richard Heaney and David Pitt
model enables the insurer to vary transition probabilities as well as the sum
insured and then consider the impact of these choices on the insurer’s break even
level of premiums.
The critical factor for an insurer is the impact of genetic testing on the
profitability of insurance contracts where the insured is aware of the impact of
genetic test results though the insurer is not. In Table 2 we report the results of
varying the age, term of the insurance policy and the amount insured given that the
individual has returned a positive genetic test. The table provides the increased
premium rates required for conventional term insurance policies sold to people
aged 30 or 40 and who hold life insurance contracts for terms of either 10 or 20
years. For comparison the table reports the increases in required premiums for
values of t6 equal to both 0.90 and, in brackets, for t6 equal to 0.30. This shows
how sensitive the required premium increases are to the extent of the modelled
adverse selection. From Table 2 it is clear that the most significant premium
increases occur when those insured, who have returned a positive genetic test,
request higher amounts of life insurance coverage. This increase in premiums is
required because large insurance payments will be paid more often as the
probability of death for individuals who returned a positive genetic test is higher
than for other individuals on average.
Table 2: Premium Rate Increases Required in The Presence of
Adverse Selection
Sum Insured of
Positive Test
Age 30
Term 10 yrs
Age 30
Term 20 yrs
Age 40
Term 10 yrs
Age 40
Term 20 yrs
Normal 4.5%
(3.0%)
2.7%
(2.0%)
4.5%
(2.9%)
2.7%
(1.6%)
2 * Normal 11.0%
(6.8%)
7.2%
(5.7%)
11.1%
(6.9%)
7.3%
(5.4%)
4 * Normal 22.3%
(14.0%)
17.6%
(13.1%)
22.5%
(14.3%)
17.2%
(13.6%)
Note: Values in brackets show the required increase in premium rates when t6,
the transition intensity for those who have returned a positive genetic test,
is 0.30. The values not in brackets show the required increase in premium
rates when t6 is 0.90.
As those with a positive genetic test who chose to buy insurance increase the
amount insured, the break-even premium increases markedly regardless of the age
of the insured or the term of the insurance contract. For example if those insured
Impact of Genetic Testing on Life Insurance 1011
with a positive genetic test choose a normal level of insurance then the adjustment
to the overall break even premium for the insurer with the introduction of genetic
testing is fairly small, either 2.7 per cent or 4.5 per cent in Table 2. In contrast
where the amount insured is quadrupled by those insured who have received a
positive genetic test, the overall break even premium increases by as much as 22.5
per cent. Further, increases in the term of the life insurance contract tend to
reduce the impact on the break-even premium. For example given that the insured
is aged 30, we note that the required percentage increase in break-even premium
reduces from 22.3 per cent to 17.6 per cent with an increase in term from 10 years
to 20 years. Thus the impact of genetic testing on the pricing of life insurance can
be substantial where the insurer does not have access to the results of the genetic
tests.
Conclusions on Pricing of Life Insurance Contracts
These examples show that genetic testing is capable of changing the way the
financial services industry operates. In particular, risk-based insurance products
are highly vulnerable to information asymmetry. The implication from the
theoretical literature is that without a solution to the information problem, we can
get a rational response from insurance companies to raise premiums that may
accentuate adverse selection effects as more of the low risk applicants withdraw
from the insurance market.
This paper has brought together some of the work by both economists and
actuaries in an Australian context and includes a simple modelling exercise
designed to highlight the impact of information asymmetry on break-even life
insurance premiums. The main message from the paper is that while genetic
testing presents a further form of information asymmetry for the insurer, it only
generates significant financial implications when
· those individuals who have returned a positive genetic test not only have
an increased likelihood of purchasing insurance but also
· request significantly higher amounts of insurance coverage.
A clear policy recommendation arising from this work is that insurers be
allowed the right to require access to genetic testing information under certain
circumstances – a particular circumstance addressed in this paper is where the
applicant requests a sum insured well in excess of the level ordinarily sought by
potential purchasers of life insurance. In this case it is critical to the profitability
and solvency of insurance companies that they have access to genetic test results.
One fear might be that potential customers, who are unable to get high amounts of
life insurance coverage with one insurer, might purchase standard amounts of life
insurance coverage from a large number of different insurers and hence create
serious problems for the life insurance industry. Contracts of life insurance
require applicants for life insurance to disclose whether they already have life
insurance with any other insurers and also whether they have been rejected for life
1012 Richard Heaney and David Pitt
insurance cover in the past. This means that, apart from cases of fraud, the
insurance industry can stop people entering into multiple life insurance contracts.
The key message of this paper indicates that failure to address this
information asymmetry could prove costly to the life insurance industry and could
also threaten the viability of a comprehensive, well functioning insurance market.
References
APRA (2002), title? http://www.apra.gov.au/Statistics/Life-Insurance-Market-
Statistics.cfm
Akerlof, G. (1970), ‘The Market For 'Lemons': Qualitative Uncertainty And The Market
Mechanism,’ Quarterly Journal of Economics 84:488-500.
Doherty, N. and P. Thistle (1996), ‘Adverse Selection with Endogenous Information in
Insurance Markets’, Journal of Public Economics 63: 83-102.
Gravelle, H. and R. Rees (1985), Microeconomics, Longman, London and New York.
Hoy, M. and M. Polborn (2000), ‘The Value of Genetic Information in the Life Insurance
Market’, Journal of Public Economics 78: 235-252.
Macdonald, A. (1997), ‘Current Actuarial Modelling Practice and Related Issues and
Questions’, North American Actuarial Journal 1:24-35.
Macdonald, A. (1999), ‘Modelling the Impact of Genetics on Insurance’, North American
Actuarial Journal 3:83-105.
Otlowski, M. (2002), ‘Genetic Testing and Insurance: The Case for Regulation’, Agenda
9:335-354.
This article was stimulated by a recent conference on Genetics and Financial
Services organised by the Centre for Actuarial Research at the ANU’s School of
Finance and Applied Statistics, the Institute of Actuaries and the Securities
Institute. The authors would like to extend thanks to the reviewers for their helpful
comments.
Richard Heaney and David Pitt
he Human Genome project generates immense interest in the scientific
community though there are also important issues for the business
community, particularly insurance companies. The dramatic advances in
our understanding of the human genetic code, or human genome, affect our
understanding of the determinants of human longevity and this is critical to the
profitability of life insurance contracts.
Insurance plays an important role in our economy. For example, there were
42 life insurance companies managing approximately AUD $188 billion as at 30
June 2002 and these same companies received AUD $38.1 billion in premium
income over for the year ended 30 June 2002. Life insurance contracts are
increasingly being sold in combination with superannuation where the contract
forms part of the superannuation package. For example up to 85 per cent of all
life office assets and 90 per cent of the premiums were classified as
superannuation business in 2001 (APRA, 2002). Regardless of whether an
individual submits an application form directly to a life insurance company as part
of a stand alone policy, or indirectly via superannuation, the insurance company
faces the question of deciding whether they wish to sell insurance to this
individual.
While life insurance policies can take a number of forms, an essential feature
of these contracts is that they promise the payment of a given amount to certain
beneficiaries when the insured dies. The time of death is critical to the pricing and
profitability of these contracts. For insurance contracts to be profitable they must
be priced so that invested premiums generate sufficient reserves to meet the
payment of death benefits when they fall due.
Failure to adequately model the impact of our increased understanding of the
human genome could have a dramatic impact on the profitability of insurance
contracts. The mapping of the human genome and the rapid development of
genetic testing means that people have access to greater knowledge about their
health and longevity yet this information may not be freely available, particularly
to insurers and annuity providers (Hoy and Polborn, 2000). If individuals have
more information about their health than insurance companies, this can complicate
the pricing of life insurance contracts and annuities. This one-sided access to
information is often referred to as information asymmetry and at its worst
information asymmetry can lead to market failure (Akerlof, 1970). Further, where
the insurer is unable to accurately assess the risk of an applicant it is possible that
prices will be set too high. The ultimate result could be that only those applicants
Richard Heaney is a Reader in Finance and David Pitt is a Lecturer in Actuarial Studies
in the School of Finance and Applied Statistics at the Australian National University.
T
1002 Richard Heaney and David Pitt
most likely to require a payout will purchase the product resulting in low
profitability or losses. This is a critical problem in the pricing of life insurance
products. Life insurers generally attempt to protect themselves through pricing for
average risks though this may become more difficult where information about
health and longevity is restricted in an asymmetric manner. An objective of this
paper is to show that failure to adequately model the impact of our increased
understanding of the human genome – on choices made by consumers, and on the
pricing policies of insurance companies – could have a dramatic impact on the
profitability of insurance contracts.
While Doherty and Thistle (1996) and Hoy and Polborn (2000) provide
economic analysis of the impact of information about the human genome, the
work of actuaries such as Macdonald (1997) provides insight into the problems
that actuaries face in the day-to-day pricing of life insurance contracts. We apply
the MacDonald (1997) model to gain further understanding of the impact of
genetic research on Australian life insurance contracts – in particular, the effects
on profitability of adverse selection by consumers who discover themselves to be
high risk, and who subsequently purchase more insurance than would otherwise
have been the case. The following section provides a brief review of the literature.
It is followed by two sections that respectively describe the model used in the
simulation of the insurance problem faced by Australian insurance companies, and
the results of the simulation. Conclusions are drawn in the final section of the
article.
Insurance Theory
Transactions costs are an important part of the market for life insurance (Gravelle
and Rees, 1985). One of these costs is the cost of identifying the true risk of the
individual. Individuals could be classified into broad categories such as good risk,
where there is little risk of a claim, and bad risk, where the probability of a claim
is high. Insurers generally assess the probability of death through analysis of
objective factors, such as age and occupation, and subjective factors, such as
exercise, diet and habits such as smoking. Each applicant is assessed for these
factors and a premium estimated and charged. Where it is impossible to identify
the risk associated with a group of individuals, adverse selection can lead to bad
risks driving out good risks. This effect could arise where the insurer initially sets
the insurance premium too high for the best risks. Given the high price the best
risk group choose not to insure, leaving only those representing poorer risks for
the insurer. The difficulty for the insurer is that the premium does not allow for
exclusion of the best risk group – with the loss of the best risk group the current
premium is set too low to cover the expected costs of the contract. Over time as
the level of claims follows the actual longevity of the insured group, the insurer is
forced to raise the level of premiums. Again, the better risks in the group choose
not to insure at this higher price and so the cycle continues until the good risks are
priced out of the market and the insurer faces ruin. In effect, the insurer faces a
trade off between the costs of obtaining better information about individuals and
Impact of Genetic Testing on Life Insurance 1003
thus pricing insurance contracts properly and the reduced profitability arising from
the impact of bad risks driving out good risks when insurance premiums do not
properly reflect the risk of the insured group. A regulatory response to this type of
problem occurring in the health insurance area has been the use of community
rating in the pricing of health insurance.
There may be signalling effects arising from the decision of insurance
companies to sort into risk categories. A rational response from low risk
applicants is to signal quality in order to support a separating equilibrium and
receive lower premiums; from high risk applicants the response is to mix signals
in order to keep a pooling equilibrium, whereby the cost of their insurance is
spread across a larger group (and thus lower premiums).
The impact of genetic information on the life insurance market is important
though current research suggests that the net welfare effect of genetic testing is not
clear. For example, Hoy and Polborn (2000), extending the model of Doherty and
Thistle (1996), show that the private value of being informed is positive for the
individual while the social value of the information could be either positive or
negative. Their model includes the impact of genetic testing and it is assumed that
there is an incentive for an individual to undertake the test to obtain further
information about their health with the knowledge that the insurer does not have
access to this information. Essentially there are three groups in the model, high
risk, low risk and the uninformed. Hoy and Polborn (2000) focus on the benefits
to those in the uninformed group who choose to test, assuming that the current
price of insurance is only attractive to high-risk individuals. If the uninformed
choose the test and are found to be bad risks then they can insure and so they are
better off. If they are good risks then they can choose not to insure. In this partial
equilibrium model the uninformed are better off with the availability of the test
because they have the option to purchase a contract at the old price if they test
positive. There are also spillover effects associated with those who choose not to
take the test where the actual risk of this group differs from the average risk for
the pool. For example the spillover effect is positive if the initially uninformed
(who are tested and subsequently buy insurance) are lower risk than the existing
customers. Hoy and Polborn (2000) argue that this source of asymmetric
information is not unusual as there is legislation in a number of countries
specifically set up to protect the rights of the individual to genetic testing results
and to deny the insurance companies control over this information. This debate
continues in Australia and is surveyed in Otlowski (2002).
Thus the impact of the introduction of genetic testing is not clear. It is argued
by Hoy and Polborn (2000) that if information is distributed symmetrically
between insurers and customers, the introduction of the test is welfare reducing for
consumers assuming that markets stay open (alt hough price may vary) if we
ignore the benefits that arise from the use of genetic testing including improved
medical treatment of these conditions. With asymmetric information the net
welfare effects of the test could be positive, negative or a mixed case where those
who undertake the test gain and those that choose not to undertake the test lose.
1004 Richard Heaney and David Pitt
This theoretical work on insurance and asymmetric information helps to
understand the implications of genetic testing and highlights the impact of
asymmetric information but it tells us little about the actual pricing of life
insurance where genetic testing is available to individuals but not to insurers. In
the following section we focus on empirical testing. Actuaries have an important
role to play in valuing these instruments through statistical modelling and we use
the Markov model developed by Macdonald (1997) in analysis of the impact of
genetic testing on breakeven insurance premiums. As indicated above, knowledge
of the mortality of a group of individuals is critical to the accurate pricing of
insurance offered to the members of this group and, given information asymmetry,
genetic testing could have a dramatic effect on the pricing of life insurance.
Actuarial modelling
Actuarial research into the financial impact of genetic testing has gained
momentum in recent years. MacDonald (1997,1999) has published a number of
papers advocating the use of multi-state Markov models to explore the impact of
uncertainty on traditional life insurance products where an individual has access to
information such as genetic testing results but the insurer does not have access to
this information. This is particularly important to insurers where the tests provide
highly predictive information about mortality.
A thorough assessment of the impact of genetic testing involves consideration
of four major factors. The first factor reflects the level of insurance that the
applicant might prefer. The level of insurance could vary with the existence of a
genetic predisposition to a particular disease. It could also vary with whether the
applicant has been genetically tested. A final source of variation lies with whether
the applicant has been genetically tested and found not to have a higher probability
than normal of contracting a particular disease (negative results). The second
factor is the prevalence and predictive accuracy of genetic tests. The third factor
is the proportion of those who have genetic tests and who return a positive result
indicating that they have a higher probability than normal of contracting a
particular disease. The final factor is the extent to which people who receive a
positive genetic test for a particular disease have an increased propensity to
purchase insurance.
The most significant financial impact of genetic testing is thought to occur for
life insurance products such as term insurance and associated riders such as dread
disease insurance. Dread disease insurance is a rider that may be added to a
contract containing death insurance. This rider provides the insured with a benefit
if they incur one of a selection of serious medical conditions listed in the policy.
Genetic testing is more important for term insurance than whole of life
insurance because in the case of whole of life insurance the insured is guaranteed
payment of the amount insured on death. The only question for whole of life
insurers is the timing of the payment. Under a term insurance contract the amount
insured is paid if the insured dies before the term of the contract and so the insurer
prices the contract with a view to both the timing of death and the likelihood of
Impact of Genetic Testing on Life Insurance 1005
death before expiry of the contract. Thus where the insured has a genetic
predisposition for higher mortality they are more likely to purchase a term
insurance contract because the premium is lower for term insurance contracts than
for whole of life contracts. This is because lower risk is ordinarily presented to
the insurer for term insurance contracts.
Commonly term insurance is provided for a period of 10 or 20 years. These
types of insurance contracts are often called risk-based products. In the Australian
setting the majority of the term insurance business is renewed each year, or yearly
renewable. These insurance products give the insured the option to renew the
insurance contract each year and the insurer is obligated to renew the coverage at
the request of the insured as long as there is no material change to the risk
presented by the insured. This arrangement provides considerable potential for
adverse selection against the insurer and so it is important to model the impact on
life insurance contract premiums.
A benchmark Markov model with no genetic test
First, let us consider the case of a customer who purchases life insurance when
genetic testing is ignored. This model will be used as the benchmark case for
analysing the effects of adverse selection on the profitability of term insurance
contracts. Consider a typical individual who purchases insurance cover at age 30
for the first time. This life insurance coverage provides payment of the sum
insured, typically $100,000, on the death of the insured individual and the
payment is made to the estate of the insured individual. We will assume that this
individual has purchased cover for 20 years payable by annual premiums. The
insured renews the policy annually by paying a level annual premium for twenty
years or until earlier death. The situation can be illustrated as in Figure 1 that
shows transitions between three ‘states’. State 1 is where all people who have not
purchased life insurance are situated. If an individual purchases life insurance,
they move from State 1 into State 2. This move between states is called a
transition and in Figure 1 this transition is labelled transition t1. In this model
people who have purchased life insurance, and therefore reside in State 2, can of
course die. Death causes them to move to State 3 and, as shown in Figure 1, make
transition t3. Individuals who have not purchased life insurance, and therefore
reside in State 1, can also die and move to State 3. This movement from the
uninsured state (State 1) to the dead state (State 3) is labelled transition t2. If the
consumer makes an annual decision whether to renew the contract, the transition
probabilities reflect the likelihood of the event (transition) occurring within a
particular year for a person of certain age.
In this three-state model where we ignore the impact of genetic testing the
annual premium paid by an insured individual depends only on two key factors.
The first and most significant is the magnitude of transition intensity t3 – the
mortality of insured individuals. In practice these mortality rates (which vary with
age) are determined by considering the experience of the particular insurer and
1006 Richard Heaney and David Pitt
also by consulting published Australian mortality tables. The other important
factor is the investment income that the insurer is assumed to be able to earn on
the premiums paid by the insured to the insurer. The insurer can earn substantial
amounts of investment income from the prudent investment of premium income.
This is because under term insurance the premiums are paid to the insurer long
before any insurance payment is made, if it is ever made, to the estate of the
insured individual. The assumed rate of investment income is again determined
by reference to the recent investment returns earned by the insurer on the funds
that are backing the relevant insurance portfolio. In this analysis we have ignored
the impact on premiums charged to consumers of commissions and other expenses
that the insurer would ordinarily incur.
Figure 1: Markov Model for Term Insurance in the Absence of Genetic
Testing
A Markov model with genetic testing
We now consider the case where the impact of genetic testing is taken into
consideration in the determination of suitable insurance premiums. MacDonald
(date?) models individuals as occupying a series of states and making transitions
from state to state until they die. The important linkages modelled by MacDonald
are identified in Figure 2. As in the previous diagram, each box in Figure 2
represents a state in which either a consumer or potential consumer of life
insurance could be situated. In any interval of time the consumer is able to make a
transition along any of the arrows. Each transition has a simple description.
Transition t1 occurs when an individual who has not had any genetic testing
purchases insurance. Transitions t2 and t3 occur after a genetic test is performed
t1
State 1
Not Insured
State 2
Insured
State 3
Dead
t2
t3
Impact of Genetic Testing on Life Insurance 1007
on the consumer. The consumer makes transition t2 if the test result is negative
indicating that they do not suffer from the conditions identified in genetic testing.
The consumer makes transition t3 if the test result is positive and the genetic tests
identify the consumer as suffering from the tested ailments. Transitions t4 and t6
occur when an individual purchases life insurance after testing. While the
transition t4 is reserved for those individuals who purchase insurance after
returning a negative genetic test, transition t6 is made when purchasing insurance
for those individuals who returned a positive genetic test. The remaining
transitions (t5, t7, t8, t9 and t10) occur on the death of an individual from any of
the five other states in the model.
Figure 2: Markov Model for Assessing the Impact of Genetic Testing on
Insurer Profitability.
State 1:
Not Insured
No Genetic
Test
State 6:
Dead
State 2:
Insured.
No Positive
Test
State 3:
Tested and
Result is
Negative
State 4:
Tested and
Result is
Positive
State 5:
Insured and
Positive
Test
t1 t2 t3
t4
t5
t6
t7 t8 t9 t10
1008 Richard Heaney and David Pitt
It is clear that the model in Figure 2 is an extension of the model in Figure 1.
States 1, 2 and 3 in the model that ignores genetic testing (Figure 1) correspond to
states 1, 2 and 6 respectively in the model where genetic testing is considered
(Figure 2).
An important feature of this model is that the likelihood of some of the
transitions (those other than death, which obviously vary with age) depend only on
the state currently occupied by the insured. The model ignores information about
past movements between particular states of the model. For example the
probability that an individual who returned a positive genetic test two years ago
will buy insurance in the next week is the same as the probability that an
individual who has just returned a positive genetic test will purchase insurance in
the next week. In Figure 2 this is the same as saying that the chance of moving
from State 4 to State 5 is unrelated to the amount of time spent in State 4.
In addition to considering the probabilities of transitions between states,
consideration must also be given to the payments made where applicable while
residing in states and on the transition between states. In Figure 2 the premiums
are paid to the insurer while the insured is in states 2 and 5. This is because when
an individual is in state 2 or state 5 the individual has life insurance. On transition
from state 2 or state 5 to state 6 (that is on the death of an insured whether or not
that individual had returned a positive genetic test) a payment is made to the estate
of the insured person. This payment is the amount of life insurance purchased by
the insured.
The model aims to determine the impact on insurer profitability of varying
the modelled probabilities of transition between particular states. It is of particular
interest to understand the impact of adverse selection, reflected by the increased
likelihood that individual will purchase insurance if the individual knows that they
have a genetic predisposition to higher mortality than average. This is modelled
by increasing the probability of making the transition t6 to a multiple, greater than
one, of the chance of making transition t5. Another issue of concern to insurers is
the level of insurance sought by those who have returned a positive genetic test
result compared with those who have returned a negative test result or who have
not had a genetic test at all. This can be explored in this model by allowing the
insurance amount that the insured chooses to vary. The relevant insurance amount
is the amount paid on transition from state 5 to the dead state, and it can be
modelled to be double or four times the amount paid on transition from state 2 to
the dead state.
Pricing of Life Insurance Contracts
To gain some idea of the impact of changes in the level of information asymmetry
on the profitability of insurance policies we vary the size of the amount insured
and the probability of purchasing insurance for those who have returned a positive
genetic test and therefore exhibit higher mortality. We then focus on the breakeven
premium for the group of individuals insured. This is the premium charged
to all those insured within a particular group regardless of whether a genetic test
Impact of Genetic Testing on Life Insurance 1009
has been conducted and irrespective of the results of the test where it is
undertaken. The model has been run using Australian mortality data based on the
IA95-97 Life Table. This life table is produced by the Institute of Actuaries of
Australia Mortality Committee and is based on the mortality experience of the
major life insurers operating in Australia during the years 1995 to 1997. The
interest rate is assumed to be 5 per cent per annum compounding continuously in
the model and expenses have been ignored in this analysis.
The projected increase in mortality resulting from a positive genetic test
clearly needs to be estimated for inclusion in the model. This increase is
represented by the difference between the probabilities of making the transitions t9
and t10 and the other transitions to the dead state, namely t5, t7 and t8. For the
purposes of this analysis a 50 per cent increase in mortality has been assumed for
those who are insured and who have returned a positive genetic test. An extreme
value of 0.90 and a less extreme value of 0.30 have been used for the conditional
probability that an individual, given that they have returned a positive genetic test,
will buy insurance.
A summary table of the transition intensities is shown below in Table 1.
Table 1: Transition Intensities used in the Term Insurance Analysis in
the presence of Genetic Testing
Transition Description Transition Intensity
Transition t1 0.50
Transition t2 0.20
Transition t3 0.05
Transition t4 0.05
Transition t6 0.30 and 0.90
Transitions t5, t7 ,t8 at age 30 0.00113
Transitions t5, t7 ,t8 at age 45 0.00139
Transitions t5, t7 ,t8 at age 60 0.00568
Transitions t9 and t10 at age 30 0.00170
Transitions t9 and t10 at age 45 0.00209
Transitions t9 and t10 at age 60 0.00852
Once transition probabilities are assigned it is necessary to select the amount
that individuals would choose to be insured for. With these inputs identified, the
statistical model generates the level of premium required for the insurer to break
even. This model provides considerable flexibility in helping insurers to make
informed decisions about the level of premiums that should be charged. The
1010 Richard Heaney and David Pitt
model enables the insurer to vary transition probabilities as well as the sum
insured and then consider the impact of these choices on the insurer’s break even
level of premiums.
The critical factor for an insurer is the impact of genetic testing on the
profitability of insurance contracts where the insured is aware of the impact of
genetic test results though the insurer is not. In Table 2 we report the results of
varying the age, term of the insurance policy and the amount insured given that the
individual has returned a positive genetic test. The table provides the increased
premium rates required for conventional term insurance policies sold to people
aged 30 or 40 and who hold life insurance contracts for terms of either 10 or 20
years. For comparison the table reports the increases in required premiums for
values of t6 equal to both 0.90 and, in brackets, for t6 equal to 0.30. This shows
how sensitive the required premium increases are to the extent of the modelled
adverse selection. From Table 2 it is clear that the most significant premium
increases occur when those insured, who have returned a positive genetic test,
request higher amounts of life insurance coverage. This increase in premiums is
required because large insurance payments will be paid more often as the
probability of death for individuals who returned a positive genetic test is higher
than for other individuals on average.
Table 2: Premium Rate Increases Required in The Presence of
Adverse Selection
Sum Insured of
Positive Test
Age 30
Term 10 yrs
Age 30
Term 20 yrs
Age 40
Term 10 yrs
Age 40
Term 20 yrs
Normal 4.5%
(3.0%)
2.7%
(2.0%)
4.5%
(2.9%)
2.7%
(1.6%)
2 * Normal 11.0%
(6.8%)
7.2%
(5.7%)
11.1%
(6.9%)
7.3%
(5.4%)
4 * Normal 22.3%
(14.0%)
17.6%
(13.1%)
22.5%
(14.3%)
17.2%
(13.6%)
Note: Values in brackets show the required increase in premium rates when t6,
the transition intensity for those who have returned a positive genetic test,
is 0.30. The values not in brackets show the required increase in premium
rates when t6 is 0.90.
As those with a positive genetic test who chose to buy insurance increase the
amount insured, the break-even premium increases markedly regardless of the age
of the insured or the term of the insurance contract. For example if those insured
Impact of Genetic Testing on Life Insurance 1011
with a positive genetic test choose a normal level of insurance then the adjustment
to the overall break even premium for the insurer with the introduction of genetic
testing is fairly small, either 2.7 per cent or 4.5 per cent in Table 2. In contrast
where the amount insured is quadrupled by those insured who have received a
positive genetic test, the overall break even premium increases by as much as 22.5
per cent. Further, increases in the term of the life insurance contract tend to
reduce the impact on the break-even premium. For example given that the insured
is aged 30, we note that the required percentage increase in break-even premium
reduces from 22.3 per cent to 17.6 per cent with an increase in term from 10 years
to 20 years. Thus the impact of genetic testing on the pricing of life insurance can
be substantial where the insurer does not have access to the results of the genetic
tests.
Conclusions on Pricing of Life Insurance Contracts
These examples show that genetic testing is capable of changing the way the
financial services industry operates. In particular, risk-based insurance products
are highly vulnerable to information asymmetry. The implication from the
theoretical literature is that without a solution to the information problem, we can
get a rational response from insurance companies to raise premiums that may
accentuate adverse selection effects as more of the low risk applicants withdraw
from the insurance market.
This paper has brought together some of the work by both economists and
actuaries in an Australian context and includes a simple modelling exercise
designed to highlight the impact of information asymmetry on break-even life
insurance premiums. The main message from the paper is that while genetic
testing presents a further form of information asymmetry for the insurer, it only
generates significant financial implications when
· those individuals who have returned a positive genetic test not only have
an increased likelihood of purchasing insurance but also
· request significantly higher amounts of insurance coverage.
A clear policy recommendation arising from this work is that insurers be
allowed the right to require access to genetic testing information under certain
circumstances – a particular circumstance addressed in this paper is where the
applicant requests a sum insured well in excess of the level ordinarily sought by
potential purchasers of life insurance. In this case it is critical to the profitability
and solvency of insurance companies that they have access to genetic test results.
One fear might be that potential customers, who are unable to get high amounts of
life insurance coverage with one insurer, might purchase standard amounts of life
insurance coverage from a large number of different insurers and hence create
serious problems for the life insurance industry. Contracts of life insurance
require applicants for life insurance to disclose whether they already have life
insurance with any other insurers and also whether they have been rejected for life
1012 Richard Heaney and David Pitt
insurance cover in the past. This means that, apart from cases of fraud, the
insurance industry can stop people entering into multiple life insurance contracts.
The key message of this paper indicates that failure to address this
information asymmetry could prove costly to the life insurance industry and could
also threaten the viability of a comprehensive, well functioning insurance market.
References
APRA (2002), title? http://www.apra.gov.au/Statistics/Life-Insurance-Market-
Statistics.cfm
Akerlof, G. (1970), ‘The Market For 'Lemons': Qualitative Uncertainty And The Market
Mechanism,’ Quarterly Journal of Economics 84:488-500.
Doherty, N. and P. Thistle (1996), ‘Adverse Selection with Endogenous Information in
Insurance Markets’, Journal of Public Economics 63: 83-102.
Gravelle, H. and R. Rees (1985), Microeconomics, Longman, London and New York.
Hoy, M. and M. Polborn (2000), ‘The Value of Genetic Information in the Life Insurance
Market’, Journal of Public Economics 78: 235-252.
Macdonald, A. (1997), ‘Current Actuarial Modelling Practice and Related Issues and
Questions’, North American Actuarial Journal 1:24-35.
Macdonald, A. (1999), ‘Modelling the Impact of Genetics on Insurance’, North American
Actuarial Journal 3:83-105.
Otlowski, M. (2002), ‘Genetic Testing and Insurance: The Case for Regulation’, Agenda
9:335-354.
This article was stimulated by a recent conference on Genetics and Financial
Services organised by the Centre for Actuarial Research at the ANU’s School of
Finance and Applied Statistics, the Institute of Actuaries and the Securities
Institute. The authors would like to extend thanks to the reviewers for their helpful
comments.
http://cbe.anu.edu.au/research/papers/pdf/2003-01.pdf
