EXAMINERS’ REPORT

CS2 - Risk Modelling and Survival Analysis

Core Principles
Paper A

April 2023
 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

Introduction

The Examiners’ Report is written by the Chief Examiner with the aim of helping candidates,
both those who are sitting the examination for the first time and using past papers as a
revision aid and also those who have previously failed the subject.

The Examiners are charged by Council with examining the published syllabus. The
Examiners have access to the Core Reading, which is designed to interpret the syllabus, and
will generally base questions around it but are not required to examine the content of Core
Reading specifically or exclusively.

For numerical questions the Examiners’ preferred approach to the solution is reproduced in
this report; other valid approaches are given appropriate credit. For essay-style questions,
particularly the open-ended questions in the later subjects, the report may contain more points
than the Examiners will expect from a solution that scores full marks.

For some candidates, this may be their first attempt at answering an examination using open
books and online. The Examiners expect all candidates to have a good level of knowledge
and understanding of the topics and therefore candidates should not be overly dependent on
open book materials. In our experience, candidates that spend too long researching answers
in their materials will not be successful either because of time management issues or because
they do not properly answer the questions.

Many candidates rely on past exam papers and examiner reports. Great caution must be
exercised in doing so because each exam question is unique. As with all professional
examinations, it is insufficient to repeat points of principle, formula or other text book
works. The examinations are designed to test “higher order” thinking including candidates’
ability to apply their knowledge to the facts presented in detail, synthesise and analyse their
findings, and present conclusions or advice. Successful candidates concentrate on answering
the questions asked rather than repeating their knowledge without application.

The report is written based on the legislative and regulatory context pertaining to the date that
the examination was set. Candidates should take into account the possibility that
circumstances may have changed if using these reports for revision.

Sarah Hutchinson
Chair of the Board of Examiners
July 2023

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

A. General comments on the aims of this subject and how it is marked

The aim of the Risk Modelling and Survival Analysis Core Principles subject is to
provide a grounding in mathematical and statistical modelling techniques that are of
particular relevance to actuarial work, including stochastic processes and survival models.

Some of the questions in this paper admit alternative solutions from those presented in
this report, or different ways in which the provided answer can be determined. All
mathematically correct and valid alternative solutions received credit as appropriate.

In cases where an error was carried forward to later parts of the answer, candidates were
given full credit for those later parts.

In higher order skills questions, where comments were required, well-reasoned comments
that differed from those provided in the solutions also received credit as appropriate.

Candidates are advised to take careful note of all instructions that are provided with the
exam to maximise their performance in future CS2A examinations.

B. Comments on candidate performance in this diet of the examination.

The number of questions on the paper was reduced from nine to eight in response to
previous candidate feedback related to time pressure when answering the paper using MS
Word. However, candidates should note that the syllabus and core reading for this subject
is extensive, and the examiners are concerned that a considerable number of candidates
present themselves without sufficient preparation and understanding across the whole
syllabus. Although this exam is set twice a year, the examiners feel that many candidates
are likely to need more than 4-6 months preparation time given the length of the
syllabus.

Although this examination is “open-book,” the examiners would recommend that

candidates prepare for this paper as if they were sitting a traditional “closed-book” exam.
There are two reasons for this. Firstly, access to resources in a time limited exam is no
substitute for understanding and revision. Secondly, this examination tests understanding
of statistics and risk models asking candidates to apply these techniques to certain
scenarios or data sets given in the questions. Successful answers to these questions
require candidates to enter the examination with the necessary understanding of the
statistical methods and then apply a careful problem-solving approach to the data or
scenario given. It is highly unlikely that consulting resources in the examination time
available will help here.

The examination team note that the higher order questions were found by candidates to be
challenging in this sitting. Again, the key here is application of knowledge (for example
about censoring in the Kaplan-Meier estimate or the parameters in the Lee-Carter model
in two of the questions on this paper) to the scenario presented or the results obtained in
earlier parts of the question. It is hoped that the solutions presented below will help
candidates in future sessions appreciate the type of answers that can reach the highest
marks.

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

C. Pass Mark

The Pass Mark for this exam was 51.

1226 candidates presented themselves and 376 passed.

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

Solutions for Subject CS2A - April 2023

Q1
(i)
The chain is irreducible because every state can be reached from any other state [1]

(ii)
We are interested in:

If , [1]

If , [1]
For

[1]

[1]

[1]
[1]
(Candidates do not need to use the i, j notation - other formats are acceptable including
an explanation of where there are zero entries without listing them all)

(iii)
Starting with all the white balls in A, getting will require that a white ball is
drawn from A and a black ball from B for each [1]
Thus, the probability

[2]
[Total 10]

Parts (ii) and (iii) of this question were not well answered by many candidates. In
fact, as a proportion of the marks available, the average mark was lowest for the
whole paper. This is a surprise as the question does not require specialist
technical knowledge, but rather basic knowledge of transition probabilities and a
careful approach to problem-solving. This question is an excellent example of the
comment at the beginning of this report about preparation for open-book
examinations.

Q2
(i)
The likelihood function for the ith year [½]
L (mu_i;di,vi) = exp(-mu_i *vi)* mu_i ^di [1]

Where:
mu_i is the transition rate from unemployed to employed in the ith year [½]

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

di is the number of transitions from state “unemployed” to state “employed”

in ith year [½]
vi is the total observed waiting time in State “unemployed” in ith year [½]

(ii)
Which results in maximum likelihood estimate of:
mu_i^hat = di / vi
Therefore,
mu_1^hat = 5000 /30000 = 0.16667 [1]
mu_2 ^hat = 7000/22000 = 0.31818 [1]

(iii)
The maximum likelihood estimator mu_2^hat has a variance equal to:
Mu2/E[V] [2]
Where:
Mu2 is the true transition rate in the second year
E[V] is the expected waiting time of being unemployed
Mu2 ~= mu_2^hat = 0.31818
E[V] ~=v2 = 22000 [1]
Variance = 0.000014 [1]

(iv)
Estimating
P (not getting employed in the next 2 years) = 2p0 = exp( - int( o to 2) mu x+s ds) [1]
= Exp(-mu1)*exp(-mu2) [1]
= 0.8465* 0.7275 [½]
= 0.61579 [½]
[Total 12]

This question was well answered, and the average mark was the highest across
the whole paper.

In parts (i) and (ii) credit was given to candidates who only started to differentiate
between years 1 and 2 in the second part, and to candidates who expressed all
their answers numerically without full notation. The most common mistake was to
calculate a blended transition rate across the two years for which partial credit
was given if subsequent calculations proceeded correctly with that rate.

Q3
(i)
From the 3 plots in Figure 1
a
The increasing values of a imply that Mortality rates increase with age [½]
Visually the plot appears linear [½]
which implies that the increase is exponential [1]

k
Given that the b-values are positive [½]

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

the decreasing values of k imply that mortality has been improving over the period [½]
either: Visually the plot appears somewhat linear [½]
which implies that improvement rates have been constant [½]
or: the rate of decrease appears less pronounced at the earliest and latest years
suggesting different rates of improvement over the period [1]

b
The increasing values of b imply that mortality improvements have been greatest
for younger ages [½]
The values appear constant in ages 65 - 70 and decrease thereafter [½]
(½ mark for any other reasonable observation about the nature of the graphs)
[Marks available 6, maximum 4]

(ii)
Medical improvements have been greatest for younger ages
Education has had the greatest effect on younger population e.g. related to
smoking advice, general health
Or any other reasonable comment [1]
(Award ½ mark if reasonable comment about older ages instead)

(iii)
exp(-3.35 + 0.0484 * (-13)) = 0.0187 accept any 0.0185 - 0.0189 [1]
(Award ½ mark if parameters read correctly or calculation performed correctly but not
both)

(iv)
projected k, in 1 years: -0.4763 1.64 * (0.0842 + 0.346)1/2 [1]
k = -0.4763 0.9745 [1]
k = (-1.4508, 0.4982) [1]
Therefore, the lower limit of the required confidence interval is
m = 0.0187 exp(0.0484 * -1.4508) = 0.0174 [1]
The upper limit is m = 0.0187 exp(0.0484 * 0.4982) = 0.0192 [1]

(v)
ax is a measure of the average rate at each age over the investigation period. Values
would therefore be relatively high for the 20-30 year group. A plot of ax may show
an “illness bump" [2]

For kt there would be a rapid decline in values around 1980, due to the rapid fall off
of deaths in the 20-30 age range following the cure being introduced to the
population. It may lead to a general underestimation of projected mortality rates if a
simple linear model is adopted for projecting k. The effect will depend on the
relative weightings of deaths in that age group [2]

bx will show large numbers in the 20-30 year band; these characteristics will be
incorrectly projected into the future (assuming the illness has been eradicated),
with improvements at these ages being greatly exaggerated [2]

In summary projected mortality rates are likely to be too low, with the greatest effect
on the 20-30 year age band [1]

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

Other reasonable comments on ages 30+ [1]

[Marks available 8, maximum 5]
[Total 16]

This question was generally poorly answered, particularly parts (i), (iv) and (v).

In part (i) a large range of sensible points about a, b and k were given credit. Many
candidates simply recited definitions of these parameters rather than applying those
definitions to the evidence of the graphs in the question. The best answers combined
a description of the plots with an understanding of the model parameters.

In part (iv) partial credit was given to a wide range of approaches to calculating a
confidence interval. In particular a number of candidates derived a value for
m70;2019 and then built a confidence interval for that rather than building the
interval around k.

In part (v), candidates were not given credit for discussion of the advantages and
disadvantages of cohort models given one is assumed in the question.

Q4
(i)
Allows path to recovery to vary with time since snake bite [½]
Constant transition intensities would seem inappropriate here [½]

(ii)
Transition rate to full recovery falls with duration [½]
Given -2.5 parameter probability of full recovery quickly becomes negligible [½]
Transition rate to recovery with long term effects increases with duration [½]
It seems reasonable that as the duration of sickness increases, the probability of
recovery without long-term health effects decreases and the probability of recovery
with long-term health effects increases [½]
As t increases this transition rates trends to 0.05 [½]
There is no upper limit to the time taken to recover in this model [½]
There is no death state [½]
The transition rate can go negative at some durations which is unrealistic [½]
Other reasonable observations [½]
[Marks available 5½, maximum 3]

(iii)
Pr (person bitten eventually fully recovered)
= integral(0,) Pr(remains sick from 0 to t)
* (transition rate to fully recovered at t) dt [2]
Pr(remains sick from 0 to t) = exp(-integral(0,t)(exp(-2.5u) + 0.05 - exp(-2.5u))du) =
exp(-0.05t) [1]
so integral becomes:
= integral(0,∞) exp(-0.05t) exp(-2.5t) dt = integral(0,∞) exp-(2.55t) dt [1]
= [-exp-(2.55t) / 2.55]:(0,∞) [1]
= 1 / 2.55 = 0.392 [1]

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

(Full credit should be awarded to candidates who give the correct numeric answer and
show some working but not necessarily all of the steps above)
[Total 10]

This is a straightforward Markov jump process question that was reasonably well
answered.

In part (ii) a wide range of suitable comments attracted credit. This (somewhat
akin to question 3 above) is an example of the need to show understanding of
mathematical concepts by applying them to the scenario given in the question.
Being successful in this is one of the key differences between candidates who
passed and those who did not. Once again, taking a closed-book rather than open-
book approach would pay dividends here as the necessary step of applying
knowledge to the scenario is unlikely to be found in resources consulted during an
examination.

Q5
(i)
We have (1 + a * b) * (a + b) / (1 + b^2 + 2 * a * b) = 0.65 (1) [1]
and (1 + a * b) * (a + b) / (1 + b^2 + 2 * a * b) * a = 0.325 (2) [1]
Dividing (2) by (1) gives a = 0.5 [½]

Substituting in (1) gives:

(1 + 0.5 * b) * (0.5 + b) = 0.65 * (1 + b^2 + 2 * 0.5 * b) [½]
i.e. 0.5 * b^2 + 1.25 * b + 0.5 = 0.65 * (b^2 + b + 1) [½]
i.e. 0.15 * b^2 - 0.6 * b + 0.15 = 0. [½]
The roots of this quadratic are 0.268 and 3.732 [1]
We require the root less than 1 in magnitude [½]
which is 0.268 [½]

(ii)
For autocorrelation at lag 1 the sample size to be used in the formula for Fisher’s
transformation is n - 1 [1]
The test statistic ½ * log ((1 + r_1) / (1 - r_1)) is approximately Normally distributed
with mean ½ * log ((1 + rho_1) / (1 - rho_1)) and variance 1 / (n - 4), where r_1 is
the sample autocorrelation at lag 1 and rho_1 is the theoretical autocorrelation at
lag 1 [1]
For a = 0.5 and b = 0, rho_1 = 0.5 [½]
th
The 95 percentile of the standard Normal distribution is 1.645 [½]
We therefore require the least positive integer n such that
½ * log ((1 + 0.65) / (1 - 0.65)) - ½ * log ((1 + 0.5) / (1 - 0.5)) > 1.645 / sqrt(n - 4), [1]
i.e. such that
n > 4 + (1.645 / (½ * log ((1 + 0.65) / (1 - 0.65)) - ½ * log ((1 + 0.5) / (1 - 0.5))))^2. [1]
The least positive integer n satisfying this inequality is 57 [1]
(Full credit was given to candidates who use n-3 instead of n-4. In this case the final
numeric answer will be 56)
[Total 12]

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

This question was not well answered and for the third consecutive session
candidates have not answered Time Series questions as well as expected. A lot of
candidates spent valuable exam time deriving the autocorrelation formulae rather
than applying them, which the examiners suspect was due to the derivations being
consulted during the open-book exam rather than the application of the formulae
having been revised beforehand.

In part (i) the first marks available are for restating the autocorrelation formulae
in terms of the ARMA(1,1) model and then proceeding from there.

Q6
(i)
Since the exponential distribution with parameter \lambda and with expectation
\mu=1/\lambda has tail probability
Exp(-x/ \mu ) then [1]
Exp(-50/\mu)=2/300 so
-50/\mu =log(2/300)=-5.010635 [1]
So
\mu=-50/5.010635=9.978775 [1]

(ii)
Since the threshold exceedance distribution for the exponential distribution is the
same as the original distribution then [1]
[or since the exponential distribution is memoryless, then …]
the random variable U=X-50|X>50 has the same expectation as above, i.e. 9.978775 [1]

(iii)
GPD is preferred if extreme weather events are becoming more likely [1]
and therefore the exceedance distributions are expected to have fatter tails than
those of the exponential [1]
modelling of the tails is seen as more important in a scenario such as this [1]
other sensible comments contrasting the GPD and the exponential [1]
[Total marks 4, maximum 2]

(iv)
If beta =1 the Pareto distribution will have expectation the same as the expected
exceedance amount
\gamma /(gamma -1)= 9.978775 [1]
or
\gamma = (gamma -1)*9.978775 [1]
\gamma= 9.978775/(9.978775-1)= 1.111374 [1]
[Total 10]

The first three parts of this question on loss distributions is relatively straightforward,
so again it is disappointing that these were not well answered.

Part (iv) is more demanding.

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

In part (i) as with other calculation questions in this paper, full credit was awarded
for solutions with the correct final numerical answer (9.9878) and some but not each
step of working given.

Full marks were also awarded in (ii) for reference to the memoryless property of the
exponential distribution.

Q7
(i)
We have discrete data [½]
The hazard depends on duration / time [½]
There is [right] censoring [½]
There is non-informative censoring [½]
The data is suited to a non-parametric approach [½]
Other sensible comment on data suited to K-M approach [½]
[Marks available, 3, maximum 2]

(ii)
At duration t weeks, let dt be the number who fail the task that week
ct be the number censored that week
(see below for application of censoring to this problem)
nt be the ”risk set” - the number of volunteers still on the program
then ht is the hazard of failing the task in week t where ht = dt / nt
[1]
and the Kaplan Meier survival function is S(t) where

S(t) = [½]

The Kaplan Meier estimate assumes that censoring occurs after failure therefore
volunteers who do not arrive for week j need to be included in cj-1 not cj [1]

t nt dt ct ht 1-ht S(t)
1 20 1 0 0.05 0.95 0.95
2 19 2 1 0.105263158 0.894736842 0.85
3 16 2 0 0.125 0.875 0.74375
4 14 0 0 0 1 0.74375
5 14 1 4 0.071428571 0.928571429 0.690625
6 9 1 0 0.111111111 0.888888889 0.613888889
7 8 2 0 0.25 0.75 0.460416667
8 6 1 0 0.166666667 0.833333333 0.383680556
9 5 2 0 0.4 0.6 0.230208333
10 3 1 0 0.333333333 0.666666667 0.153472222

[½] [½] [1] [½] [½] [1]

We seek the largest t at which S(t) >= 0.3 [½]

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

The number of weeks required is 8 [1]

(iii)
hazard unlikely to be zero at week 4 [½]
large amount of censoring between weeks 5 & 6 - would want to investigate why [1]
relatively small sample size [½]
would knowledge of the required number of weeks change behaviour? [½]
the right censoring may well be informative [½]
other types of censoring may be present [½]
different training programs may not be uniformly difficult [½]
other sensible comments [½]
(To obtain full marks on this part some discussion of the censoring here is required)
[Marks 4 available, maximum 3]
[Total 13]

This Kaplan Meier Estimate question was reasonably well answered.

The key to a full correct solution in part (ii) and then to a strong answer to part
(iii) is to understand the role of censoring in this scenario. The ordering of failure
and censoring is one of the K-M assumptions and well prepared candidates
recognised this in their answers.

A wide variety of layouts for the calculations and answers in (ii) were given full
credit. Candidates are reminded of the importance of defining terms when
completing survival model calculations.

Part (iii) was less well answered and again asks candidates to apply knowledge of
the model to the scenario in the question.

Q8
(i)
Using the backshift operator one can show that the corresponding polynomials are
1-a B^2 [1]
and
1+bB [1]
The roots need to be in absolute value less than 1
abs(a)<1 and abs(b)<1 [2]

(ii)
ARMA(2,1) [1]

(iii)
The Yule-Walker equations are
gamma_0=a gamma_2+sigma^2 [1]
and
gamma_k=a gamma_{k-2} for k >= 1 [1]
So

CS2A A2023 © Institute and Faculty of Actuaries

 CS2A- Risk Modelling and Survival Analysis - Core Principles - April 2023 - Examiners’ report

gamma_1=a gamma_1 [1]

gamma_2=a gamma_0 [1]
These imply that
gamma_1=0, gamma_2=a gamma_0 and in general [1]
gamma_k =0 for k odd [1]
gamma_k = a^{k/2} gamma_0 for k even [1]
therefore
rho_k=0 for k odd [½]
rho_k=a^(k/2) for k even [½]
(There are no marks available for deriving the Yule Walker equations from first
principles)

(iv)
Sample acf of the data could have indicated insignificant spikes for odd lags as
for b=0 case those values are zero [2]
AIC/BIC could have also been used to confirm the statistical preference between the
two models [1]
In the parameter estimation process for model (1), some low t-values could have been
produced, particularly for the parameter b, indicating over-parametrisation. [1]
other sensible comments contrasting the fit of the two models [1]
[Marks available 5, maximum 4]
[Total 17]

This question was much better answered than the earlier Time Series question.

Parts (i) and (ii) were well answered.

In part (iii) some candidates used valuable exam time to show a derivation of the
Yule Walker equations which was not required (and gained no marks). Again, in an
open-book exam, application not knowledge based ais required for marks.

In part (iv) a wider range of sensible comments attracted marks including

alternative tests that could have been performed.

[Paper Total 100]

END OF EXAMINERS’ REPORT

CS2A A2023 © Institute and Faculty of Actuaries

Beijing
14F China World Office 1 · 1 Jianwai Avenue · Beijing · China 100004
Tel: +86 (10) 6535 0248

Edinburgh
Level 2 · Exchange Crescent · 7 Conference Square · Edinburgh · EH3 8RA
Tel: +44 (0) 131 240 1300

Hong Kong
1803 Tower One · Lippo Centre · 89 Queensway · Hong Kong
Tel: +852 2147 9418

London (registered office)

7th Floor · Holborn Gate · 326-330 High Holborn · London · WC1V 7PP
Tel: +44 (0) 20 7632 2100

Oxford
1st Floor · Belsyre Court · 57 Woodstock Road · Oxford · OX2 6HJ
Tel: +44 (0) 1865 268 200

Singapore
5 Shenton Way · UIC Building · #10-01 · Singapore 068808
Tel: +65 8778 1784

www.actuaries.org.uk
© 2021 Institute and Faculty of Actuaries