om

i.c
CM2-04: Measures of investment risk Page 1

ng
si
om
m
o
as
.m
w
w
w
Measures of investment risk
Syllabus objectives
2.1 Properties of risk measures

2.1.1 Define the following measures of investment risk:

 variance of return
 downside semi-variance of return
 shortfall probabilities
 Value at Risk (VaR) / TailVaR.

2.1.2 Describe how the risk measures listed in 2.1.1 above are related to the
form of an investor’s utility function.
2.1.3 Perform calculations using the risk measures listed in 2.1.1 above to
compare investment opportunities.
2.1.4 Explain how the distribution of returns and the thickness of tails will
influence the assessment of risk.
2.2 Risk and insurance companies

2.2.1 Describe how insurance companies help to reduce or remove risk.

2.2.2 Explain what is meant by the terms ‘moral hazard’ and ‘adverse selection’.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 2 CM2-04: Measures of investment risk

ng
si
om
o m
0 Introduction

as
.m
w
In financial economics, it is often assumed that the key factors influencing investment decisions

w
w
are ‘risk’ and ‘return’. In practice, return is almost always interpreted as the expected investment
return. However, there are many possible interpretations and different ways of measuring
investment risk, of which the variance is just one, each of which corresponds to a different utility
function.

This chapter outlines a small number of such measures, together with their relative merits, and
then moves on to discuss how insurance can be used to reduce the impact of risk.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 3

ng
si
om
o m
1 Measures of risk

as
.m
w
1.1 Introduction

w
w
Most mathematical investment theories of investment risk use variance of return as the
measure of risk.

Examples include (mean-variance) portfolio theory and the capital asset pricing model, both of
which are discussed later in this course.

However, it is not obvious that variance necessarily corresponds to investors’ perception of

risk, and other measures have been proposed as being more appropriate.

Some investors might not be concerned with the mean and variance of returns, but simpler things
such as the maximum possible loss. Alternatively, some investors might be concerned not only
with the mean and variance of returns, but also more generally with other higher moments of
returns, such as the skewness of returns. For example, although two risky assets might yield the
same expectation and variance of future returns, if the returns on Asset A are positively skewed,
whilst those on Asset B are symmetrical about the mean, then Asset A might be preferred to
Asset B by some investors.

1.2 Variance of return

For a continuous distribution, variance of return is defined as:


 (  x )
2
f ( x ) dx

where  is the mean return at the end of the chosen period and f(x) is the probability
density function of the return.

‘Return’ here means the proportionate increase in the market value of the asset, eg x  0.05 if
the asset value has increased by 5% over the period.

The units of variance are ‘%%’, which means ‘per 100 per 100’.

eg  4% 2  16%%  0.16%  0.0016

Question

Investment returns (% pa), X , on a particular asset are modelled using a probability distribution
with density function:


f (x)  0.00075 100  (x  5)2  where 5  x  15

Calculate the mean return and the variance of return.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 4 CM2-04: Measures of investment risk

ng
si
om
o m
Solution

as
.m
The density function is symmetrical about x  5 . Hence the mean return is 5%. Alternatively, this

w
w
could be found by integrating as follows:

w
15

E[ X ]  0.00075 x 100  (x  5)2 dx
5 
15
 0.00075 75 x  10 x 2  x 3dx
5
15
 75 10 1 
 0.00075  x 2  x 3  x 4 
2 3 4  5

 0.000757,031.25  364.5833

5

ie 5% pa.

The variance is given by:

15
 
Var ( X )  0.00075 (5  x)2 100  (x  5)2 dx
5

15
 0.00075 100(x  5)2  (x  5)4 dx
5
15
 100 1 
 0.00075  (x  5)3  (x  5)5 
 3 5  5

 0.0007513,333.33  (13,333.33)

 20

ie 20%% pa.

Alternatively, we can calculate the variance using the formula: Var( X )  E[ X 2 ]  (E[ X ])2 , where
E[ X 2 ] can be found by integration to be 45%%.

For a discrete distribution, variance of return is defined as:

(  x)2 P(X  x)

x

where  is the mean return at the end of the chosen period.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 5

ng
si
om
o m
Question

as
.m
w
Investment returns (% pa), X , on a particular asset are modelled using the probability

w
distribution:

w
X Probability
7 0.04

5.5 0.96

Calculate the mean return and variance of return.

Solution

The mean return is given by:

E[ X ]  7  0.04  5.5  0.96  5

ie 5% pa.

The variance of return is given by:

Var(X )  (5  (7))2  0.04  (5  5.5)2  0.96  6

ie 6%% pa.

Alternatively, we can calculate the variance using the formula: Var( X )  E[ X 2 ]  (E[ X ])2 , where
E[ X 2 ] is 31%%.

Variance has the advantage over most other measures in that it is mathematically tractable,
and the mean-variance framework discussed in a later chapter leads to elegant solutions for
optimal portfolios. Albeit easy to use, the mean-variance theory has been shown to give a
good approximation to several other proposed methodologies.

Mean-variance portfolio theory can be shown to lead to optimum portfolios if investors can
be assumed to have quadratic utility functions or if returns can be assumed to be normally
distributed.

In an earlier chapter we discussed how the aim of investors is to maximise their expected utility.
The mean-variance portfolio theory discussed in a later chapter assumes that investors base their
investment decisions solely on the mean and variance of investment returns. This assumption is
consistent with the maximisation of expected utility provided that the investor’s expected utility
depends only on the mean and variance of investment returns.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 6 CM2-04: Measures of investment risk

ng
si
om
m
It can be shown that this is the case if:

o
as
.m
 the investor has a quadratic utility function, and/or

w
w
 investment returns follow a distribution that is characterised fully by its first two

w
moments, such as the normal distribution.

If, however, neither of these conditions holds, then we cannot assume that investors make
choices solely on the basis of the mean and variance of return. For example, with more complex
utility functions and non-normal return distributions investors may need to consider other
features of the distribution of returns, such as skewness and kurtosis.

Question

Define both the skewness and the fourth central moment (called the kurtosis) of a continuous
probability distribution.

Solution

The skewness of a continuous probability distribution is defined as the third central moment.

It is a measure of the extent to which a distribution is asymmetric about its mean. For example,
the normal distribution is symmetric about its mean and therefore has zero skewness, whereas
the lognormal distribution is positively skewed.

The kurtosis of a continuous probability distribution is defined as the fourth central moment.

It is a measure of how likely extreme values are to appear (ie those in the tails of the distribution).

1.3 Semi-variance of return

The main argument against the use of variance as a measure of risk is that most investors
do not dislike uncertainty of returns as such; rather they dislike the possibility of low
returns.

For example, all rational investors would choose a security that offered a chance of either a 10%
or 12% return in preference to one that offered a certain 10%, despite the greater uncertainty
associated with the former.

One measure that seeks to quantify this view is downside semi-variance (also referred to as
simply semi-variance). For a continuous random variable, this is defined as:


 (  x )
2
f ( x ) dx

Semi-variance is not easy to handle mathematically, and it takes no account of variability

above the mean. Furthermore, if returns on assets are symmetrically distributed,
semi-variance is proportional to variance.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 7

ng
si
om
o m
Question

as
.m
w
Investment returns (% pa), X , on a particular asset are modelled using a probability distribution

w
with density function:

w

f (x)  0.00075 100  (x  5)2  where 5  x  15

Calculate the downside semi-variance of return.

Solution

We saw in an earlier question that the variance of investment returns for this asset is 20%%.
Since the continuous distribution f (x) is symmetrical, the downside semi-variance is half the
variance, ie 10%%.

For a discrete random variable, the downside semi-variance is defined as:

 (  x)2 P(X  x)
x

Question

Investment returns (% pa), X , on a particular asset are modelled using the probability
distribution:

X probability
7 0.04

5.5 0.96

Calculate the downside semi-variance of return.

Solution

We saw in an earlier question that the mean investment return for this asset is 5%. So the
downside semi-variance is given by:

 (5  x)2 P(X  x)  (5  (7))2  0.04  5.76

x 5

ie 5.76%% pa.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 8 CM2-04: Measures of investment risk

ng
si
om
o m
1.4 Shortfall probabilities

as
.m
A shortfall probability measures the probability of returns falling below a certain level. For

w
w
continuous variables, the risk measure is given by:

w
L
Shortfall probability   f ( x ) dx
where L is a chosen benchmark level.

For discrete random variables, the risk measure is given by:

Shortfall probability   P( X  x )
x L

The benchmark level, L, can be expressed as the return on a benchmark fund if this is more
appropriate than an absolute level. In fact, any of the risk measures discussed can be
expressed as measures of the risk relative to a suitable benchmark which may be an index,
a median fund or some level of inflation.

L could alternatively relate to some pre-specified level of surplus or fund solvency.

Question

Investment returns (% pa), X , on a particular asset are modelled using a probability distribution
with density function:


f (x)  0.00075 100  (x  5)2  where 5  x  15

Calculate the shortfall probability where the benchmark return is 0% pa.

Solution

The shortfall probability is given by:

0
P( X  0)  0.00075 100  (x  5)2 dx
5
0
 1 
 0.00075 100 x  (x  5)3 
 3  5

 0.00075  41.6667   166.6667 

 0.15625

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 9

ng
si
om
o m
Question

as
.m
w
Investment returns (% pa), X , on a particular asset are modelled using the probability

w
distribution:

w
X probability

7 0.04

5.5 0.96

Calculate the shortfall probability where the benchmark return is 0% pa.

Solution

The shortfall probability is given by:

P( X  0)  0.04

The main advantages of the shortfall probability are that it is easy to understand and calculate.

The main drawback of the shortfall probability as a measure of investment risk is that it gives no
indication of the magnitude of any shortfall (being independent of the extent of any shortfall).

For example, consider two securities that offer the following combinations of returns and
associated probabilities:
Investment A: 10.1% with probability of 0.9 and 9.9% with probability of 0.1
Investment B: 10.1% with probability of 0.91 and 0% with probability of 0.09

An investor who chooses between them purely on the basis of the shortfall probability based
upon a benchmark return of 10% would choose Investment B, despite the fact that it gives a much
bigger shortfall than Investment A if a shortfall occurs.

1.5 Value at Risk

Value at Risk (VaR) generalises the likelihood of underperforming by providing a statistical
measure of downside risk.

For a continuous random variable, Value at Risk can be determined as:

VaR ( X )  t where P ( X  t )  p

VaR represents the maximum potential loss on a portfolio over a given future time period
with a given degree of confidence, where the latter is normally expressed as 1  p . So, for
example, a 99% one-day VaR is the maximum loss on a portfolio over a one-day period with
99% confidence, ie there is a 1% probability of a greater loss.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 10 CM2-04: Measures of investment risk

ng
si
om
m
Note that Value at Risk is a ‘loss amount’. Therefore:

o
as
.m
 a positive Value at Risk (a negative t ) indicates a loss

w
w
 a negative Value at Risk (a positive t ) indicates a profit

w
 Value at Risk should be expressed as a monetary amount and not as a percentage.

Question

Investment returns (% pa), X , on a particular asset are modelled using a probability distribution
with density function:


f (x)  0.00075 100  (x  5)2  where 5  x  15

Calculate the VaR over one year with a 95% confidence limit for a portfolio consisting of £100m
invested in the asset.

Solution

We start by finding t , where P( X  t)  0.05 :

t
 0.00075 100  (x  5)2 dx  0.05
5
t
 1 
 0.00075 100 x  (x  5)3   0.05
 3  5

Since the equation in the brackets is a cubic in t , we are going to need to solve this equation
numerically, by trial and error.

3
 1 
t  3  0.00075 100 x  (x  5)3   0.028
 3  5

2
 1 
and t  2  0.00075 100 x  (x  5)3   0.06075
 3  5

0.05  0.028
Interpolating between the two gives: t  3   2.3
0.06075  0.028

In fact, the true value is t  2.293 . Since t is a percentage investment return per annum, the
95% Value at Risk over one year on a £100m portfolio is £100m  2.293%  £2.293m . This means
that we are 95% certain that we will not lose more than £2.293m over the next year.

For a discrete random variable, VaR is defined as:

VaR( X )  t where t  maxx : P(X  x)  p

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 11

ng
si
om
o m
Question

as
.m
w
Investment returns (% pa), X , on a particular asset are modelled using the probability

w
distribution:

w
X Probability

7 0.04

5.5 0.96

Calculate the 95% VaR over one year with a 95% confidence limit for a portfolio consisting of
£100m invested in the asset.

Solution

We start by finding t , where t  maxx : P(X  x)  0.05 .

Now P(X  7)  0 and P(X  5.5)  0.04 . Therefore t  5.5 .

Since t is a percentage investment return per annum, the 95% Value at Risk over one year on a
£100m portfolio is £100m  5.5%  £5.5m . This means that we are 95% certain that we will
not make profits of less than £5.5m over the next year.

VaR can be measured either in absolute terms or relative to a benchmark. Again, VaR is
based on assumptions that may not be immediately apparent.

The problem is that in practice VaR is often calculated assuming that investment returns are
normally distributed.

Question

Calculate the 97.5% VaR over one year for a portfolio consisting of £200m invested in shares.
Assume that the return on the portfolio of shares is normally distributed with mean 8% pa and
standard deviation 8% pa.

Solution

We start by finding t , where:

P( X  t)  0.025 , where X  N(8,82 )

Standardising gives:

 t 8   t 8
P Z       0.025
 8   8 

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 12 CM2-04: Measures of investment risk

ng
si
om
m
t 8

o
Now (1.96)  0.025 from page 162 of the Tables, so:  1.96  t  7.68 .

as
8

.m
w
w
Since t is a percentage investment return per annum, the 97.5% Value at Risk over one year on a

w
£200m portfolio is £200m  7.68%  £15.36m . This means that we are 97.5% certain that we will
not lose more than £15.36m over the next year.

Portfolios exposed to credit risk, systematic bias or derivatives may exhibit non-normal
distributions. The usefulness of VaR in these situations depends on modelling skewed or
fat-tailed distributions of returns, either in the form of statistical distributions (such as the
Gumbel, Frechet or Weibull distributions) or via Monte Carlo simulations. However, the
further one gets out into the ‘tails’ of the distributions, the more lacking the data and, hence,
the more arbitrary the choice of the underlying probability becomes.

Hedge funds are a good example of portfolios exposed to credit risk, systematic bias and
derivatives. These are private collective investment vehicles that often adopt complex and
unusual investment positions in order to make high investment returns. For example, they will
often short-sell securities and use derivatives.

If the portfolio in the previous question was a hedge fund then modelling the return using a
normal distribution may no longer be appropriate. A different distribution could be used to
assess the lower tail but choosing this distribution will depend on the data available for how
hedge funds have performed in the past. This data may be lacking or include survivorship bias,
ie hedge funds that do very badly may not be included.

The Gumbel, Frechet and Weibull distributions are three examples of extreme value distributions,
which are used to model extreme events.

The main weakness of VaR is that it does not quantify the size of the ‘tail’. Another useful
measure of investment risk therefore is the Tail Value at Risk.

1.6 Tail Value at Risk (TailVaR) and expected shortfall

Closely related to both shortfall probabilities and VaR are the TailVaR (or TVaR) and
expected shortfall measures of risk.

The risk measure can be expressed as the expected shortfall below a certain level.

For a continuous random variable, the expected shortfall is given by:

L
Expected shortfall = E max(L  X ,0)   (L  x )f ( x ) dx
where L is the chosen benchmark level.

If L is chosen to be a particular percentile point on the distribution, then the risk measure is
known as the TailVaR.

The (1  p) TailVaR is the expected shortfall in the p th lower tail. So, for the 99% confidence
limit, it represents the expected loss in excess of the 1% lower tail value.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 13

ng
si
om
o m
Question

as
.m
w
Investment returns (% pa), X , on a particular asset are modelled using a probability distribution

w
with density function:

w

f (x)  0.00075 100  (x  5)2  where 5  x  15

Calculate the 95% TailVaR over one year for a portfolio consisting of £100m invested in the asset.

Solution

In a previous question, we calculated the 95% VaR for this portfolio to be £2.293m based on an
investment return of –2.293%.

The expected shortfall in returns below –2.293% is given by:

E[max(2.293  X ,0)]  0.00075

2.293
5  
(2.293  x) 100  (x  5)2 dx

 0.00075
2.293
5  171.975  97.93x  7.707x2  x3  dx
2.293
 0.00075  171.975 x  48.965 x 2  2.569 x 3  0.25x 4 
  5
 0.0462

On a portfolio of £100m, the 95% TailVaR is £100m  0.000462  £0.0462m . This means that the
expected loss in excess of £2.293m is £46,200.

For a discrete random variable, the expected shortfall is given by:

Expected shortfall = E max(L  X ,0)   (L  x)P(X  x)

x L

Question

Investment returns (% pa), X , on a particular asset are modelled using the probability
distribution:

X Probability
7 0.04

5.5 0.96

Calculate the 95% TailVaR over one year for a portfolio consisting of £100m invested in the asset.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 14 CM2-04: Measures of investment risk

ng
si
om
o m
Solution

as
.m
w
In a previous question, we calculated the 95% VaR for this portfolio to be –£5.5m based on an

w
investment return of 5.5%.

w
The expected shortfall in returns below 5.5% is given by:

E[max(5.5  X ,0)]   (5.5  x)P( X  x)

x 5.5

 (5.5  (7))  0.04  0.5

On a portfolio of £100m, the 95% TailVaR is £100m  0.005  £0.5m . This means that the
expected reduction in profits below £5.5m is £0.5m.

However, TailVaR can also be expressed as the expected shortfall conditional on there
being a shortfall.

To do this, we would need to take the expected shortfall formula and divide by the shortfall
probability.

Other similar measures of risk have been called:

 expected tail loss

 tail conditional expectation

 conditional VaR

 tail conditional VaR

 worst conditional expectation.

They all measure the risk of underperformance against some set criteria. It should be noted
that the characteristics of the risk measures may vary depending on whether the variable is
discrete or continuous in nature.

Downside risk measures have also been proposed based on an increasing function of
 L  x  , rather than  L  x  itself in the integral above.
In other words, for continuous random variables, we could use a measure of the form:

L
 g(L  x) f (x) dx
Two particular cases of note are when:

1. g(L  r )  (L  r )2 – this is the so-called shortfall variance

2. g(L  r )  (L  r ) – the average or expected shortfall measure defined above.

Note also that if g(x)  x 2 and L =  , then we have the semi-variance measure defined above.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 15

ng
si
om
m
Shortfall measures are useful for monitoring a fund’s exposure to risk because the expected

o
as
underperformance relative to a benchmark is a concept that is apparently easy to

.m
understand. As with semi-variance, however, no attention is paid to the distribution of

w
outperformance of the benchmark, ie returns in excess of L are again completely ignored.

w
w
Question

Consider an investment whose returns follow a continuous uniform distribution over the range
0% to 10% pa.

(i) Write down the probability density function for the investment returns.

(ii) Calculate the mean investment return.

(iii) Calculate the variance and semi-variance measures of investment risk.

(iv) Calculate the shortfall probability and the expected shortfall based on a benchmark level
of 3% pa.

Solution

Useful information about the continuous uniform distribution can be found on page 13 of the
Tables, including the form of its probability density function, and formulae for its mean and
variance.

(i) Probability density function

Working in % units, the investment return follows a U(0,10) distribution. So the probability
1
density function is f (x)  for 0  x  10 and 0 otherwise.
10

If we work with the returns expressed in decimals instead, then f (x)  10 for 0  x  0.10 and 0
otherwise.

(ii) Mean

The mean investment return is:

1
(0  10)  5
2

ie 5% pa.

(iii) Variance and semi-variance

The variance is given by:

1
(10  0)2  8.33
12

ie 8.33%% pa.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 16 CM2-04: Measures of investment risk

ng
si
om
m
Alternatively, we could evaluate the variance using the integral:

o
as
.m
10
 5  x 2 dx

w


w
10

w
0

Since the uniform distribution is symmetric, the semi-variance is equal to half the variance,
ie 4.17%% pa.

Alternatively, we could evaluate the semi-variance using the integral:

5
 5  x 2 dx
 10
0

(iv) Shortfall probability and expected shortfall

The shortfall probability is given by:

3 3
1 x
SP   10 dx   10   0.3
0 0

The expected shortfall is given by:

3
 3  x  dx   1 3
ES   10
 2

 10 3 x  0.5 x   0.45 %
0
0

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 17

ng
si
om
o m
2 Relationship between risk measures and utility functions

as
.m
w
An investor using a particular risk measure will base their decisions on a consideration of

w
the available combinations of risk and expected return. Given a knowledge of how this

w
trade-off is made it is possible, in principle, to construct the investor’s underlying utility
function. Conversely, given a particular utility function, the appropriate risk measure can be
determined.

For example, if an investor has a quadratic utility function, the function to be maximised in
applying the expected utility theorem will involve a linear combination of the first two
moments of the distribution of return.

In other words, if an investor has a quadratic utility function then their attitude towards risk and
return can be expressed purely in terms of the mean and variance of investment opportunities.

Thus variance of return is an appropriate measure of risk in this case.

Question

(i) State the expected utility theorem.

(ii) Draw a typical utility function for a non-satiated, risk-averse investor.

Solution

(i) The expected utility theorem states that:

 a function, U(w), can be constructed representing an investor’s utility of wealth, w
 the investor faced with uncertainty makes decisions on the basis of maximising
the expected value of utility.

(ii)
U(w)

0 w

Non-satiated investors prefer more wealth to less and so the graph slopes upwards,
ie U(w)  0 .

Risk-averse investors have diminishing marginal utility of wealth and so the slope of the
graph decreases with w, ie U(w)  0 .

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 18 CM2-04: Measures of investment risk

ng
si
om
m
If expected return and semi-variance below the expected return are used as the basis of

o
as
investment decisions, it can be shown that this is equivalent to a utility function that is

.m
quadratic below the expected return and linear above.

w
w
w
Thus, this is equivalent to the investor being risk-averse below the expected return and
risk-neutral for investment return levels above the expected return. Hence, no weighting is given
to variability of investment returns above the expected return.

Use of a shortfall risk measure corresponds to a utility function that has a discontinuity at
the minimum required return.

This therefore corresponds to the state-dependent utility functions discussed in a previous

chapter.

Question

What is meant by a state-dependent utility function?

Solution

Sometimes it may be inappropriate to model an investor’s behaviour over all possible levels of
wealth with a single utility function. This problem can be overcome by using state-dependent
utility functions, which model the situation where there is a discontinuous change in the state of
the investor at a certain level of wealth.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 19

ng
si
om
o m
3 Risk and insurance companies

as
.m
w
3.1 Introduction

w
w
Individuals and corporations face risks resulting from unexpected events.

Risk-averse individuals can buy insurance to remove their exposure to risks. Being risk-averse,
they will be willing to pay more for insurance than the expected cost of claims. The insurance
company is willing to offer insurance primarily because it is able to spread its risks.

3.2 What to insure

Two considerations must be taken into account when assessing the effect of a risk: its
likelihood and its severity. In a formal scenario, a risk matrix or graph is used as shown
below.

Figure 4.1

The figure above shows the frequency-severity dynamics of four possible events.

 Event 1 is a low-frequency-low-severity event. Such an event does not warrant any

worry to a corporation or individual.
For example, a solar eclipse occurs with low frequency, but is also accompanied by a low
level of severity!

 Event 2 is a high-frequency-low-severity event. Such an event would occur many

times but at a low cost each time. The overall cost, due to high frequency, may be
damaging. These types of events may need to be assessed on how they can be
controlled.
For example, a smashed mobile phone screen is common but is typically limited in its
extent.

 Event 3 is a high-frequency-high-severity. Such events are to be avoided.

For example, car accidents involving fatalities happen relatively frequently and have a
high severity.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 20 CM2-04: Measures of investment risk

ng
si
om
m
 Event 4 is a low-frequency-high-severity event. Such events tend to be insured.

o
as
.m
For example, earthquake or hurricane damage.

w
w
In general, low severity events (such as events 1 and 2) are not generally insurable as the

w
cost of management per claim is too expensive. However micro-insurance for poorer
clients and new technologies that enable insuring small items over a short term (such as
gadgets during a holiday) have been gaining ground.

3.3 Pooling resources

Insurance reduces the variability of losses due to adverse outcomes by pooling resources.

Consider a simple scenario of a property that has one hundredth chance of suffering
£10,000 in damages (and 99% of no damages). The expected cost is 1% of £10,000 which is
£100 while the VaR(99.5%) is £10,000.

Question

Explain why the 99.5% Value at Risk is £10,000.

Solution

Let X be the impact suffered by the property, so P( X  0)  0.99 and P(X  10,000)  0.01 . Then
from the definition of Value at Risk for a discrete random variable we have:

VaR( X )  t where t  maxx : P(X  x)  p

But P( X  10,000)  0 and P( X  0)  0.01 , therefore:

t  maxx : P( X  x)  0.005  10,000

This means that the Value at Risk is £10,000.

If ten independent properties with similar characteristics are pooled together, the average
cost is still £100 per property. At the extreme, the probability of all of them suffering the
damage is 0.0110 . The VaR(99.5%) in this case is if one property suffers damage, that is
£100 on average (using a Binomial distribution).

Question

Verify that the 99.5% Value at Risk in this case is £10,000.

Solution

Let X be the total impact suffered by the properties, so X  Binomial(10,0.01)  (10,000) . Then
from the definition of Value at Risk for a discrete random variable we have:

VaR( X )  t where t  maxx : P(X  x)  p

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 21

ng
si
om
m
If no properties suffer damage then X  0 . The probability of this is:

o
as
.m
 10 

w
P( X  0)     0.010 (1  0.01)10  0.9910  0.9044

w
0

w
So: P( X  0)  1  0.9044  0.0956

If exactly one property suffers damage then X  10,000 . The probability of this is:

 10  10!
P( X  10,000)     0.011 (1  0.01)9  0.011  0.999  0.0914
 1 9!1!

So: P( X  10,000)  1  0.9044  0.0914  0.0042

Therefore:

t  maxx : P( X  x)  0.005  10,000

This means that the Value at Risk is £10,000, which equates to one property suffering damage.

In pooling resources, an insurer attempts to group insureds (being corporations or

individuals) within homogeneous groups. In the case of an individual with the ability to
influence into which group they fall, adverse selection can occur. If the insurer is also risk
averse, then the insurance premium needs to include a margin to compensate the insurer
for taking on the risk.

3.4 Policyholder behaviour

Adverse selection describes the fact that people who know that they are particularly bad
risks are more inclined to take out insurance than those who know that they are good risks.

It arises because customers typically know more about themselves than the insurance company
knows.

Adverse selection is sometimes called ‘self-selection’ or ‘anti-selection’.

To try and reduce the problems of adverse selection, insurance companies try to find out
lots of information about potential policyholders. Policyholders can then be put in small,
reasonably homogeneous pools and charged appropriate premiums.

Moral hazard describes the fact that a policyholder may, because they have insurance, act
in a way which makes the insured event more likely.

This is because having the insurance provides less incentive to guard against the insured event
happening. For example, while driving to work one day you realise that you forgot to lock the front
door of your house. If you didn’t have any household contents insurance, you might decide to go
back and lock it. If you had adequate insurance, you might decide to carry on to work. This
difference in behaviour caused by the fact that you are insured is an example of ‘moral hazard’.

Moral hazard makes insurance more expensive. It may even push the price of insurance
above the maximum premium that a person is prepared to pay.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 22 CM2-04: Measures of investment risk

ng
si
om
o m
as
.m
w
w
w
The chapter summary starts on the next page so that you can keep
all the chapter summaries together for revision purposes.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-04: Measures of investment risk Page 23

ng
si
om
o m
as
Chapter 4 Summary

.m
w
w
Measures of investment risk

w
Many investment models use variance of return as the measure of investment risk.


For a continuous random variable: V  (  x)2 f (x) dx


For a discrete random variable: V  (  x)2P( X  x)

x

Variance has the advantage over most other measures that it:
 is mathematically tractable
 leads to elegant solutions for optimal portfolios, within the context of mean-variance
portfolio theory.

The main argument against the use of variance as a measure of risk is that most investors do
not dislike uncertainty of returns as such; rather they dislike the downside risk of low
investment returns. Consequently, alternative measures of downside risk sometimes used
include (in the continuous and then discrete cases):

 semi-variance of return:  (  x)
2
f (x) dx  (  x)2P(X  x)
x

L
 shortfall probability:  f (x) dx  P( X  x )
x L

each of which ignores upside risk.

Value at Risk (VaR) represents the maximum potential loss on a portfolio over a given future
time period with a given degree of confidence (1  p) . It is often calculated assuming that
investment returns follow a normal distribution, which may not be an appropriate
assumption.

For a continuous random variable, VaR( X )  t , where P( X  t)  p .

For a discrete random variable, VaR( X )  t , where t  maxx : P(X  x)  p .

The expected shortfall, relative to a benchmark L is given by E[max(L  X ,0)] .

L
For a continuous random variable, expected shortfall =  (L  x) f (x)dx .
For a discrete random variable, expected shortfall =  (L  x)P(X  x) .
x L

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 24 CM2-04: Measures of investment risk

ng
si
om
o m
as
When L is the VaR with a particular confidence level, the expected shortfall is known as

.m
TailVaR. TailVaR measures the expected loss in excess of the VaR.

w
w
w
It is also possible to calculate the expected shortfall and TailVaR conditional on a shortfall
occurring by dividing through by the shortfall probability.

Relationship between risk measures and utility functions

If expected return and variance are used as the basis of investment decisions, it can be
shown that this is equivalent to a quadratic utility function.

If expected return and semi-variance below the expected return are used as the basis of
investment decisions, it can be shown that this is equivalent to a utility function that is
quadratic below the expected return and linear above.

Use of a shortfall risk measure corresponds to a utility function that has a discontinuity at the
minimum required return.

Using insurance to manage risk

Insurers decide which events to offer protection for based on the frequency and severity of
the event.

The pooling of resources can be used to reduce an insurer’s risk.

Adverse selection describes the fact that people who know that they are particularly bad
risks are more inclined to take out insurance than those who know that they are good risks.

Moral hazard is the change in a policyholder’s behaviour once insurance has been taken out,
which makes the risk event more likely to occur.

© IFE: 2019 Examinations The Actuarial Education Company