Johnny Wong
Wk 9 Life Insurance
Types of policies:
Endowment Always pays, whether at death before a certain
term, or at survival to term
Pure endowment Only pays if survival to specified term
Term Insurance Pays on death within specified term
Life Annuities Regular payment until death
EPV of premiums
Suppose there are annual premium payments of P
Premiums are only paid if the person is alive (duh)
They are paid upfront (in case they die during the year)
Thus we have:
Probability of paying premium = probability they are still alive
= kpx
We need to discount by vk
EPV
Principle of equivalence
Equate EPV of premiums with EPV of benefit payments
In a perfect world, the company will make no profit or
loss
Generally though, the principle of equivalence is not
adhered to:
Companies want to profit
Companies need leeway for unfavourable events (solvency)
Probability of ruin
Often you will be given information about an insurance
companies policies and asked to compute how much
reserves they have to maintain a low risk of insolvency.
Almost always:
Use binomial distribution to find mean and variance
Use CLT to approximate to normal
Use orange book to find normal probabilities
S = 100,000
Benefit if claim
n=10,000
Number of policies
p=0.00489
Probability of claim
P=500
Premium income
X = number of claims
X ~ Bin(10,000 0.00489)
C = Capital required
X ~ Bin(10,000, 0.00489)
E[X] = 10,000x0.00489 = 48.8
Var[X] = 10,000 x 0.00489 x (1-0.00489) = 48.66
We want:
Pr(100,000X > 10,000x500 + K) = 0.001
By CLT
Looking at pg 162 of orange book, you will see that
Pr(z > 3.0902)=0.001
Thus
Algebra
K=2,045,622
The company needs at least 2,045,622 for them to have
less than 0.001 chance of insolvency
Recurrence relation
At time t, a premium P will be received by the company
At time t, two things can happen in the space of one
year:
Person dies and you pay the Benefit, after that there are no
more premiums of benefits. This happens with probability qx+t
Person survives and you still have policy liability at t+1, this
has a probability of px+t
Investment policy
In insurance, your inflows and outflows are very
mismatched. Customers pay early on, but you have to
pay them back later (when they claim)
You cant just leave the money sitting around because
you can invest and make more money.
However, you cant invest too much money because
then you wont be able to pay the customer back when
they claim.
Thus you must have a balance of when and how much
to invest in.
Severity
Log normal
Weibull
Frequency
Most often modelled by Poisson distribution:
Poisson distribution
Why is Poisson favoured over say Binomial?
One reason is that Poisson doesnt have a limit on how
many events can occur over a given time period.
Can be Thinned
If you change the time period, you can change the
parameter by the same scale
So if something is Poi(2) over one year, it is Poi(1) over
half a year
Timing
Related to frequency
Exponential distribution models the time until the next
Poisson event
If an even has frequency modelled by Poisson(), then
the time between events is exponentially distributed
with parameter
Modelling of severity
Given a claim has occurred, how much is the customer
claiming?
Often modelled by log normal or Weibull
Log Normal
Weibull
Note that in the orange book, the mean and variance
arent given to you, youll have to compute them from
the Moments
That is, if you want to find the mean, you want E[X1] so
you set k=1
If you want variance, Var[X] = E[X2]-(E[X])2
k=2
k=1
Premium rating
Lets consider a simple one year insurance policy.
End of policy
Benefits are
paid
Start of
policy
Accident
happens
Assuming X
are iid
Unearned premiums
Same logic as unearned revenue in ACCT
For insurance companies, revenue=premiums
When premiums are paid, they cover a certain time
period (usually 1 year)
Start of
policy
Earned
premium
Now
End of
policy
Unearned
premium
Earned premiums
If you look at this equation long enough, it makes sense
Outstanding claims
When a customer makes a claim, it is not paid
immediately, there is a delay between the accident and
the claim settlement
Up until now, we have assumed that a claim during the
year will be settled (benefits paid) at the end of the
year.
In general insurance, the delays can be longer.
When a claim needs to be settled, it is called an
outstanding claim.
It is important to accurately forecast claims so the
insurance companies know how much to keep in their
Excess/Deductible
Customer pays the deductible, insurance company pays
the rest.
E.g. if your claim is $300 and the excess is $200, you
pay $200 and the insurance companies pays $100
Prevents people doing stupid stuf just to get money.
If S is the aggregate loss, and D is the deductible:
Finding E[Y]
The amount
insurance company
will pay given loss is
S
Summing
over all S
that would
require
benefit
payment
The
probability
that the loss
Lower limit of
integral is D because
below the
deductible, the
company pays
nothing.
Superannuation
Defined benefits
Defined contributions
Salary growth
Inflation
Retirement, injury, death
Salary scales
As you work, your salary will increase:
Response to inflation
Promotions
Industry awards
Inflation
Inflation rate denoted by f p.a.
Continuously
compounding salary
growth rate
Salary growth
)
DONT memorise these, they arent that hard to derive.
From here, you can work out the expected value and variance of
the growth rate.
(Using orange book to find the mean and variance of log-normally
distributed random variables)
Defined benefits
Salary at 65 is $250,000
If
k=6
And you have been with the fund for 30 years, (n=30)
Benefit at retirement will be
Ethics
Just dont
Exam tips
Break down questions
Use your orange book (remember, unannotated, no
stickers)
Know your orange book
Understand, dont memorise
MC can be done quicker, maybe leave towards the end?
Linear interpolation should be used when calculating
normal probabilities.
i.e. (1.013)=0.7x (1.01)+0.3x (1.02)
Good Luck!