Episode 100: Actuaries in Research - Part 2 (Part 1) - Live With The Maverick

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Today, we are celebrating the one hundredth episode of the
Live with the Maverick podcast and we have a very
special guest to commemorate the occasion. David Wilkie is a
highly respected actuary, best known for his pioneering work in
sarcastic modeling, which has had a lasting impact an actuarial science,

(00:25):
particularly in investment and financial forecasting. David's career has been
defined by his contributions to the way actuaries model uncertainty,
providing them with advanced tools to better forecast financial risks
in an increasingly complex world. Throughout his seventy three year career,

(00:46):
David has published over one hundred and sixty research papers
in the field of actuarial science. His scholarly achievements have
been widely recognized by honorary degrees from Harriet Watt University,
Waterloo University and City University of London, gold medals from

(01:08):
both the Institute of Actuaries and the Faculty of Actuaries,
and honorary memberships of the Swiss, Italian, Swedish and South
African Actuarial Societies. In twenty twenty four, David received the
prestigious max Lander Award from the International Association of Consulting

(01:29):
Actuaries for his contributions to the public awareness of the
work of the actual profession and the promotion of the
business of consulting actuaries. Welcome David. Hello, it's great to
have you, and yes, great to have you, and couldn't

(01:49):
think of a better guest to celebrate the one hundred
episode of the series.

Speaker 2 (01:53):
So well, not that age yet.

Speaker 1 (01:58):
Almost They're almost there. So David, you know I tried.
I tried my best to do your career justice in
the introduction, and you know it's no secret that you've
had a distinguished career, making several contributions to the actual
profession and wider fields. Before we discussed those contributions, let's
first take a step back to understand where the journey began.

(02:22):
So how would you describe your early childhood and formative years.

Speaker 2 (02:27):
Well, the best things just for me to tell you
a story about them. My parents were both Scottish doctors,
graduates from Plaskow University. All my ancestors were Scottish. One
grandfather was a headmaster of a school, another one was
the engineer. What the engineer I never knew. The headmaster

(02:51):
had considerable influence on me, but my parents. My father
had got a GP practice in Lancashire. I was born there.
He wanted to change practices after it when I was three,
so my mother and I went to stay with her parents,
this schoolmaster in Glasgow, and there I think probably he

(03:13):
taught me to read and write and a lot of
other things about Glasgow. Then I went to just local
schools in Lancaster where they got her second practice to
begin with. And then because I was quite right at school,
it wanted to send me to boarding schools. First of

(03:34):
all to a preparatory school that was sort of at
ten to thirteen year olds, and then to what in
England is called a public school, which isn't public at all.
That's a private, fee paying school, this one called Rugby School,
quite well known for the gayment has given the world.

(03:54):
Where I was there from ages thirteen to seventeen. I
was quite bright at maths and a good math teacher,
and I got a scholarship to college at Cambridge when
there was just left just under seventeen, and so I

(04:15):
left school at seventeen spent a year as an actuarial
student in Scottish Widows. Now why was I an actuary
student and two reasons. One the maths master at Rugby
had a cousin or relative who was Phil Sparling, who
was an actory, so he knew about them, and on
holiday in Scotland for a number of years where went

(04:40):
the same resort in Rosemark in north of Scotland. My
parents had got to know another actory, Bren Dwell, who
was at that time and of secretary of the our
Standard Life and later on Chief Executive Standard Life and

(05:01):
President of the faculty and a gold medalist of the faculty.
So I knew about actories and started off in Edinburgh
as a student at Scottish For those doing simple actuarial calculations,
there's sorts of things nowadays that you don't need to
students to do it all because it can be done

(05:23):
on a computer more accurately. But I did learn there
you do everything in pairs. If you want to check numbers,
you need to get two people to do it, or
do it in two different ways and check them. So
that's one one bit of learning from the very early days.
I then went to university and because I was committed

(05:44):
mentally to an actuarial career, I dropped and changed at
the university and did a year's mathematics, a year's economics,
which are quite useful, and a year's English. So actually
my degree is in English and that has stayed with
me as a permanent interest, aside from actuatey at work
the rest of my life. Then two years National service,

(06:09):
and because at the university I had to amuse myself
by joining the university a squadron which was part of
the ARA Volunteer Reserve, learning how to fly little air
of planes called chipmunks, single single seater trainers. I went
into the Air Force as a trainee pilot, learned more

(06:31):
about more about flying on protests and then non vampires,
which are an early type of jet fighter. I got
my wings flying vampires. Never used them, of course, as
as as a fighter, I wasn't learning how to use
them as a fighter, just how to fly in them.

(06:53):
At the end of that of getting my wings, I
spent a few months in London in the Air and
history so it was now called the Department of Defense,
in charge of a little team that was putting officers
records on to punch cards. Now during this period I

(07:14):
had also spent a long long vacation in Denmark, in
a fortnight each and four different insurance companies which were
organized for me by Paul Jacobson, the then president and
very distinguished Danish actory. And he took me to see

(07:37):
Stephenson JF. Stephenson, who was very elderly at that time.
He was the first actory to be appointed after university
and was Professor of Actuary of Science at Copenhagen University
in about nineteen oh five, and I felt this was
a great honor to meet someone from that farthertant past.

(08:01):
There's been about nineteen fifty four or so. After national
service in the Air Force, I spent a few months
again doing a little actuarial job overseas, this time in

(08:21):
the Alliance Insurance, the life insurance part of the Alliance
in Stuttgart, which improved my German a bit. I ended
up untiled a Swabian accent from sounds of who I
come from Stuttgart, not Frum somewhere else. But anyway, after that,

(08:42):
I came back to the Scottish Woods and carried on
with my actuarial examinations, working in the same sort of
department with calculating sprender values and premium rates and stuff
like that. Qualified in nineteen fifty nine and faculty and
I had also known about the institute, and I think

(09:05):
i'd had an Institute year book lying around, and I
noticed that as a fellow of the faculty, I could
get a membership of the Institute only by doing the
final year's exams. And at that time there was a
an advanced statistics exam at the institute, so I thought

(09:28):
that was worth learning a bit more about. I'd done
those statistics at university, that had actually done some at
school last term after doing all examined nations and things
the mass master had taught us about probability and normal
distributions and regressions and stuff like that. So I did

(09:50):
the advanced statistics paper, not the institute which you use
of what was a very advanced textbook, I think it
is called Yule and Ken now called Kendlen Stewart, which
is a serious, seriously heavy textbook. Passed that exam past
all the other ones I needed to, so I became
a fellow of the Institute in nineteen sixty. In the meantime,

(10:17):
I had been asked in the Scottish videos, so I
still was to look at a bunch of paper, bunch
of paper tape typewriters called flex writers. And then after
that I was put onto programming in the Franti Pegasus

(10:38):
computer that got, which was a seriously big machine. I
mean it was a huge thing. Was about five wardrobes
of space and had the equivalent of about two five
six bytes of random access memory and thirty two k
bytes of disk drive, so tiny by modern standards, but

(11:03):
it would it would do any calculations that were required.
And it was what people call a touring machine after all,
and touring which will do anything, which is it will
do anything which it is possible to do with a

(11:24):
computer and logically possible to do, even if it takes
enormous amounts of time, enormous amounts of disk drives or
punch paper tape or punch cards or some external medium.
Just to finish my career before coming back to my computers,

(11:45):
I went to them, went to Switzerland for three for
a couple of years, nothing to do with computers there,
but got interestingly so yet another continental insurance company and
a different approach to things, one different approach. Very much
of the Swiss ree was they were wanting pride reassurance

(12:08):
around the world. They went to another country. They said,
they found out what that country wanted and they provided it.
The British insurance companies, trying to do business abroad, tended
to say we have an exit British product, would you
like to buy it, not finding out what product would
suit that country. You see the different approach of Switzerland

(12:30):
and Britain. Anyway, my wife and I decided it was
better to come back to Edinburgh, so I got a
job at Standard Life, very new people too, also in
the programming department, and I was carried on there for
a bit, then moved into economics research manager of Standard Life,

(12:53):
and then to just research actory and then thought I
realized it would be better to join a firm and
consulting actress, so I joined Are Watson's in the south
of England about nineteen eighty five. Worked there for a while,
most of the during research by now, but they decided

(13:16):
it was I had reached my retirement age at the
age of sixty two, which I thought was at least
three years too young, is not twenty three years. But
I then got a part time job consulting as a
consultant at Harriet Watt University and jointly it was a
young actress Patrick Lee helped was the new firm that

(13:39):
he had been setting up called Income Investment Quantitative Analysis,
and that was the Income was quite successful for a
while but has declined a bit by losing contracts and
doesn't do very much at the moment. And I carried
on at Harriet wat going up about one oh for

(14:05):
four days every month, not lecturing regularly, but doing a
supervising PhD students. Actually when there was a standard life
for about the last two years there, I had been
doing some regular lecturing as organized by the professoral department.
Was a standard life to second me as a for

(14:32):
one day a week to work at Harriet Watt. So
that's roughly my employed career. I'm now unemployed but still
doing some research. You would call it retired, but I
did not quite joke that I'm retired or not. I'm

(14:52):
still doing things. Even giving this interview is something that
take a bit of salt. So that's that's my life history,
very briefly.

Speaker 1 (15:07):
Greally appreciate the overview and a few things. Is one,
and I think I mentioned you before. Actually had the
pleasure of visiting Scotland for the first time this summer
and it was a wonderful experience being in Edinburgh for
a few days. I'm going to the Highlands, so good
and then.

Speaker 2 (15:23):
Good it is it is different from England. This is
one I'm very similar and I used to think in
the past that really Britain split into three. There was
a section south of the Trent which was very solidly England,
there was a section in the north of England and
south of Scotland. It was actually rather similar but had

(15:47):
small differences. And there was the Highlands, which were I
mean the East coast of Scotland Council along with the south.
But the Northwest coast is a different country and historically
has been almost a different country because that was where
the Irish Scotts came and was still speaking Irish or
the modern version of it, Scott's Gaelic still do in

(16:09):
the Northwest, and it's almost a different country from southern Scotland,
so you need to visit both parts. My sister now
lives in the island of Lewis, which is about as
far northwest in Scotland as you can get.

Speaker 1 (16:25):
Yeah, yeah, I'm looking forward to heading back. One thing
I noticed in your response, which is very which I
which I've heard several times, and it was the case
for me, is how I got into how you got
into the profession. It sounds like whether it was I
think you said it was your parents who knew someone
who was familiar with the profession. So it seems like

(16:45):
many times you know, we know what we have a
relative or a friend who knows about actual ealth science.
Would you say beyond that, would you say there were
any other reasons, any other motivating factors for you wanting
to become an actuary.

Speaker 2 (16:58):
Well, the maths master said if you're good at mass
you can earn more money. As naturally I've seen it
perfectly obvious.

Speaker 3 (17:03):
Well doing into whether it was whether that is true
or not, I do not know, but actories, many actoris
earn more than many academics.

Speaker 2 (17:14):
So I was never attempted into a university career. I'm
not researching is I now find that researching is by
forty the more than teaching, and I have The research
I've done has been because, as I'll explain, has been

(17:35):
because it was useful to the I thought useful to professional,
to the committees I was working with, and perhaps I
can I didn't continue talking about my experience in that
was that like.

Speaker 1 (17:49):
Next, Yeah, sure, so from it, I'm glad that you
mentioned the research, so that the theme of todays discussion
is actually actually he's in the research part two. So
we're going to talk about research a bit. No, As
you know, you've, as I mentioned in the intro, you've
published over one hundred and sixty academic papers, and I
know you've perhaps collaborated on others.

Speaker 2 (18:08):
Now, when you.

Speaker 1 (18:09):
When you think of your career trajectory, your long term
career trajectory, and take a step back, you know, what
were the keys to building And I know previously you
might have spoken about this and you said it was
not necessarily your intention to be a researcher, But knowing
what you know now about research and what it takes
to be to be successful and to publish papers, and

(18:32):
what were the keys to building a strong foundation which
ultimately enabled you to to be successful in publishing value
added research.

Speaker 2 (18:42):
Well, I had mentioned that I was to some extent
familiar with computer equipment and learned started learning computer programming
or in machine code on p Anti Pegasus computer and
was doing that at standard life. But then I think Fortos,

(19:04):
I'm not quite sure when Fortram was developed, and I
was started programming in nineteen sixty and Fortram may have
been around by then, but I realized that Fortram was
a handier way of doing it than writing in machine code,
which was clumsy. Well it wasn't. It wasn't clumsy, It
was just the laborious and it wasn't very clear what

(19:29):
the code meant. You had to have a lot of
exclamation beside it. So I fairly quickly, although I now
was moving across into a different department, and that was
later on get this right of Saul in the computer
department at Standard Life. But I wanted to use Fortran

(19:52):
and got permissioned to link have a Fortran compiler linked
onto the mainframe for my own use. Well one in
I got involved with the CMI Mortality Committee Continuous Mortality
Investigation Bureaus it's called now then. At first thing I
remember was getting involved with mortality was I think a

(20:16):
paper presented the Faculty of Actories about nineteen sixty by
I could look it up, by Roland Young and Frank
Reddington on the mortality experience of impaired lives people being
rated up. This was the experience of one large company

(20:37):
and everybody in Britain will know which company I meant.
I shan't mention it. It changed its name in effect
that nowadays. But the a particular tests that we're using
was using criteria in actual minus expected a minor C

(20:59):
divided by the square root of A. And I, knowing
my statistics, got up at the meeting and said, excuse me,
shouldn't that be a minor CEO but the square route
of E. And for a newly qualified or possibly even
an actuarial student had still to say that Frank Reddington

(21:21):
got it wrong. Was a bit impertinent. Actually, my comments
don't appear in the subsequent transcription of the meeting, but
she said, yeah, he said to me afterwards, Yes, I
think you're right. It should be a mine CEO would
square route to E, as everybody knows nowadays. But I
then found myself appointed to the CMI committee by in

(21:46):
nineteen sixty four. And now the committee at that time
had only eight members, the two presidents of the faculty
in the institute and three from each side, and I
and Leslie Anderson and Ernest Broadfield. Oh Leslie Anderson was
the president, Frank Reddington was the chairman of the committee.

(22:07):
So I did get to work with him on a
couple of meetings, really not much more than that, and
I was just chugging along on the committee, commenting on
the routine papers that were being produced, and then the
question came up should we not produce new graduated tables?

(22:29):
And what had happened in Britain was at about every
twenty years or so. The committee produced the new graduated
tables in twenty four to twenty nine. The first on
the semi committee was celebrating a one hundred ten is
centenary this year because nineteen twenty four was for the
first year of data, they didn't know when the committee

(22:50):
actually started, but the first year of data was nineteen
twenty four, and then forty nine fifty two, and then
the next one they should be about seventy nine to
eighty two. And Rodney Barnett, as secretary at the time,
had suggested one way of fitting a table was by

(23:11):
minimizing the Kai squared value the summer a minus e
squared over ease, and I thought, well, we can do
He was doing this by repeats approximation desperately tedious to
do it clerically. Immediately sought, no, we can do this

(23:31):
on the computer, and I was probably in almost the
only person on the committee that knew anything about computers
at all. But there are all the other people would
have lacke offices that were starting to use them. But
I knew how to deal with this. I also knew
about maximum likelihood methods, which is very similar to lease squares. Well,

(23:54):
minimum Kai squares is equivalent to least almost equivalent to
least squares, and that almost equivalent to maximum likelihood estimation.
But maximum likelihood is a better way of doing it.
So I set about writing foretram programs to produce these

(24:15):
graduated graduated tables, which eventually got checked and published as
what would we be six nine seventy two tables, and
we had we put in some forecasting for some of them,
doing the forecasting by taking assuming a one year improvement

(24:37):
in mortality equivalent to one year of age for every
twenty calendar years. That was how a nineteen PA ninety
tables were produced. That was doing very simple programming, and

(24:57):
in the course of that I developed different formulae we're
doing it, which got called compass making ones GM formulae,
which is equivalent really to an exponential plus sorry a
polynomial plus the exponential of polynomial. The typical linear model

(25:19):
that might be used would just be new X's the
eat of the power of polynomial, but putting in the
making part of polynomial plus the exponential of polynomials a
better model. It would now be called an additive model
or a general additive model. So I spent some time

(25:44):
doing that and repeated it over different years, and there
was a paper which is widely quoted I see by
forth for McCutcheon and Wilkeith published nineteen seventy four on
mortality graduation, which really introduced, I think introduced maximum likelihood estimation.

(26:07):
People were doing it sometimes just by during graphical graduation.
You put a lot of dots in, you drew a
curve and used read off the values annual values. So
what I also what we were introducing was using MUX
and probably using a plus on distribution for the residuals
rather than for the natural numbers of deaths, rather than

(26:30):
the normal approximation, which is the normal approximation is fine
when there's larger numbers, but when there are smaller numbers,
as there always are on the mortality table. As you
get to the extremely young and extremely old ages for
an insurance company's data. Not the population, of course, but

(26:50):
always was very old ages. The plass On approximation is
better or pass assumption. So that was quite an influential
I think in constructing mortality tables. Now that bit of
research was therefore done in order to satisfy the requirements

(27:13):
of the CMI committee another committee I got appointed to
tricted across by now and Standard Life to Economics Research Manager,
and that comes up. That's relevant when we've come some
point later. The Actress had been involved in investment in

(27:35):
decees for a long time, starting in the nineteen thirties,
and the convention was that the although they were joint,
the London people faculty and institute looked after the sharing disease,
the Scottish people looked after the bond in disease, and
the bond indices had been rather rather simple, too simple,

(27:58):
and we wanted to be some better ones. So I
was on the joint investment committee, and again I knew
how to write programs to demonstrate the industries and how
to fit a yield curve using actually pretty well the
same methods as for the mortality graduation fitting to just

(28:22):
co fitting exercise, but using perhaps a different criteria. I
think just uh lease squares rather than max rather than
the maximum likelihood, because it was it wasn't the yields.
Yields were around the variables. They were just irregular. They
were they should be there, should be smooth, but were

(28:43):
a bit hicul difficulty because of the bonds. And I
wrote I helped to construct the industries which are still
published in the Financial Times every day, the FT now
called it foot see actories, Government securities in disease. And

(29:08):
I wrote the foretren programs and actually supplied them to
the Financial Times too to use them. Well, that was
again using my programming abilities to help the profession to
produce something these new indices, and they have chugged along,

(29:35):
happened the after a few years. I was not involved
with it and then came back into it at the
later stage and Andrew Keirns and I produced a better
yield curve structure because I remember when we produced the
first yield curve it was just over redemption yields. And

(29:56):
I said, I remember looking back at the discussion, I said,
we had sort of doing it in terms of zero
coupon rates. But didn't want to be too far ahead
of the markets because they didn't know what zero coupon
rates were yet. But later on Afrikaans and I produced

(30:16):
a model for zero coupon rates which and power yields,
which have been used by the financial time for the last
twenty years or two. And at that time I thought
it would be useful to a good historic record. There
were probably ft is that they need. Lots of people

(30:38):
in the share market are not interested in yesterday's prices,
that's all past history. All they want to know is
what all the prices would be tomorrow or next week,
not what they were a month ago. And so have
not in the past been good at keeping historical records,
and much better now because as much easier with the computer,

(31:00):
and there's much more interest in the historical records of
share prices than perhaps there wasn't at some points in
the past, so I could put together We wanted to
put together a database of the prices and yields and
the index values on the government securities indices. And Andrew

(31:26):
Kern's got some money from the issue to pay for
students at Harriet what to go and look back at
protocopied versions of the financial times, which weren't all terribly
easy to read, and we constructed a database which has
been published on Harriet what websites until the beginning of
this year and has recently been taken over by the

(31:48):
Bank of England to put on the now what to
call it Economic Statistic Center of Excellence. That's quite useful.
There isn't a comparable historic record available for the shurances,
which are pity, but we could do something about its

(32:10):
the last twenty years or so, but they started in
nineteen sixty two and the early years are quite difficult.
So that was a that also was research to help
the profession, though there was no thing written. This is
simply an after dating gathering data which is a necessary

(32:31):
part of some bits of research. Then in the CMI
committee they had set up an income protection or what
was first called permanent health Insurance now called income Protection
Insurance subcommittee. Income protection provides cash benefits when people are sick.

(32:59):
It doesn't not nothing to do with medical expenses insurance,
but it is a sickness spend it and the chairman
of that that subcommittee a good friend of mine at
the time, Jim Keirns, the father of Andrew Keirns, who

(33:20):
is now Professor Terry Watt. He died suddenly and the
committee didn't know who to appoint as chairman. So left
the committee to appoint itself as chairman. Appointed to choose
its own chairman, Robert Plum. But when I became chairman
of the CMI a little after that, I decided that

(33:40):
it didn't do to have a committee. Was no, well,
we didn't know what was going on from the main committee.
So I appointed the chairman, said the chairman as a
member of all the subcommittees as well, and I went
on to that IP committee and with the help of

(34:03):
hard Waters Harriet Watt University, we worked out the Now
they've worked out a different way of analyzing sickness. Previously
it had been used. They've been using what called the
Manchester Unity Experience, which was constructed by the Manchester Unity

(34:25):
Friendly Society in the nineteenth century. I think and simply
looked at days sick day, sick per person per so
many insured. It didn't look at the sickness rates and
recovery rates. I wasn't thinking in terms of a multiple

(34:48):
state model, going from healthy to sick, then either to
recovery or dead, and from healthy, you might go to
dead as well, and so Harry Howard Waters knew all
about markoff. These markoff were semi markoff models, and it
took a long time between us to graduate the the

(35:13):
recovery rates and mortality rates from the sick and to
get and to find the sickness rates, and to calculate
the model and to explain it to other people. And
one of the problems the three of us were helping
was Phil Bayless was doing some graduation and Howard Waters
check checking it, and I was doing doing other doing graduations.

(35:37):
We disagreed and how what a week was Howard said
of fifty two weeks in the year. Phil Bailey said
were fifty point one eight weeks in the year, which
is correct on average alive from the leap years. I said,
the seven days in the weeks of seven, it's one seven,
three hundred and sixty fifth of a year. Because using

(36:00):
slightly different definitions, we couldn't get agreement with our answers
because when we looked at the rates at weekly intervals,
they weren't the same. So it is quite interesting how
one has to be terribly careful about definitions and how
things are calculated. Anyway that that eventually all of our

(36:25):
work on that eventually got published in CMI Reports number twelve,
and the IP Committee has continued with it. Was that
the new methods since then. In fact, in order to
help it, I produced quite a lot of the programs.
This time I was in INCA. I produced quite a

(36:47):
lot of programs used for the IP Committee. I think
they've modernized it all and moved on something else, but
I've producing foreground program for them on a commercial basis.
But sometime in the early eighties aides turned up and
it so happened that because my wife was doing some research.

(37:13):
She's a sociologist, not a medical doctor. She was doing
research for a PhD with genetically inherited diseases in Glasgow,
Royland Firmary. And although her main interest was an APKD
acute politistic kidney disease, a nasty thing, nasty disease inherited

(37:34):
from one generation to the next, she was also interested
in human philiacs, which was another hereditaryly uh connect you know,
hereditary disease.

Speaker 1 (37:50):
Uh.

Speaker 2 (37:52):
And the chief doctrine him phil Department came in one
day and said, in America, I think a problem them.
I've been learning about this new disease called AIDS or
HIV infection, and we need to see whether we've got
it in with the HUMOPHILIAX here, because the humophiliacs can

(38:14):
have their symptoms greatly alleviated by injections of what's called
factor eight, which is a condensed factor funding in the
blood of other people. But you need a lot of
blood donors, and quite a lot of blood donors were
infected with HIV, and so many humophiliacs received this infected

(38:40):
blood and my wife's pretti a job in this was
to talk to them and explain about safe sex. For example,
she had to it was ending up taking the buttons
of condoms to supply people with and mhm talk about

(39:02):
it with them as a doing sort of social work
job there. But because of that I knew about it.
It was realized it was very It was quite easy
to program it using the IP sickness model, but putting
in the relevant bit for infection, because with sickness you

(39:27):
just you assume that sickness just depends on somebody's age,
and they will get sick or not sick depending on
their age. Stops depends not of other things. But all
that we knew about was age and sex perhaps terms
of years ensured. But with infectious diseases, it also depends

(39:48):
on a number of people infected. You were cross product.
You've got to bring in the number of people in
fact or not infected who could get the disease, the
number repeople infected, and multiply those two numbers together and
not with another coefficient. It's actually very similar to a

(40:09):
marriage model. If you imagine two groups of people who
are eligible for marriage to like on that not married
males not married females of different ages meeting and there's
a small every time they meet there's a chance of
them deciding to get married. So the mathematics behind the

(40:31):
marriage model is the same as mathematics behind an infection model.
I'm not suggesting that marriage is an infection, but that's
the different matter. Anyway, that took up some time as well,
because I got involved with the AIDS working party of
the Institute and I could follow the found it not

(40:58):
too difficult to program making for the assumptions that the
committee wanted. Now, I think we got our estimates of
number of people might die from age grossly too high.
I felt that was better than getting grossly too low.
If you'll get them too high and scare people and

(41:21):
they go in for safe sex, then that's good. If
you say don't worry about it and they don't go
in for safe sex and the numbers are then a
lot higher than you thought they were going to be,
you've done it the wrong way round. So I didn't
mind that the numbers were less than they might have been.
We alerted people to help to alert people to dangers. Actually,

(41:46):
just four years ago when COVID turned up, since I
knew about infection models, it was just very quickly I
could write down and put into Excel a very simple
COVID infection model that lasted with one one line of

(42:09):
EXCELP per day. And that's how I think any found out.
Any Tay found out about that and got me involved
in giving a webinar on it to the AFR R
M section or super for the I A A and
then got me involved in a working part of the

(42:32):
I a A on of the affair section of the
I A A on lessons learned from from the epidemic,
which I can come on to next, and so all
these things of mortality I P and aids and COVID.
I've all just followed one from another because each one

(42:56):
is an extension of the same same sort of thing.
And for COVID, for the Working Party, Shuley Shaheen and
I have been doing something different. What we decided to
do is look at the history of mortality in mortality

(43:17):
rates in countries which had not been affected by the
World Wars but had good data, and we found three Sweden, Spain,
and Switzerland. Spain has a civil war, but it had
lots of deaths in late nineteen thirties, but they were

(43:40):
all in Spain. The trouble with this person Second World
Wars is that so many of the deaths were in
other countries, and huge numbers of people from Britain died
in France and Belgium, and Flanders France and Belgium, huge
numbers of Germans died on the Western Front, on the
Eastern Front as well as on the Western and so

(44:04):
many countries changed their boundaries as a result of the
wars that that made it difficult to use. And additionally,
the Spanish flu epidemic of nineteen eighteen had not been
really recorded in the countries which were fighting, because it

(44:27):
was probably widespread in the trenches, but the deaths in
the trenches on both sides were just put down to
French fever or something. Didn't notice that it was this
new infection. It was in Spain in particular that it
was noticed and identified as a new infection. It didn't
start in Spain. It was just first publicized in Spain,

(44:50):
and sadly the king there that caught it and died.
But what we did here was take the more population
data from the Human mortality database, fit a Lee Charter
model which population people know about that gives two parameters

(45:15):
reach age and one parameter for time. And it's too
simple what we did because it doesn't pay attention to
the age distribution of the infections, which I can come
back to. But what we did was pick up for
or started went back to nineteen oh eight for a

(45:37):
year before the Spanish flu, and finished in nineteen nineteen,
the year before COVID, and we fitted We use the
Kappa the the time series the time coefficient. Analyze that
using time series models, which I think i'll explain later

(46:00):
is a very much the same sort of arithmetic as
for the investment models, where I've got annual annual series
of rates and inflation or something that's just annual series
of mortality, and in order to allow for the big
jump up in nineteen eighteen and back down again in

(46:20):
nineteen nineteen has appeared in these three countries, we had
to use fatter tail distributions. It clearly the common assumption
and the assumption I'd been using in the Wilkie model,
which will come to later, and was for normally distributed residuals. Well,
these clearly were not normally distributed. So we had to

(46:42):
find fatter tail distributions. And the plus and plus distribution
is quite a good one, but even that wasn't proved
to be not quite enough. We found that on our
assumptions for Switzerland and Sweden, if an insurance company had

(47:06):
used this up to nineteen nineteen to forecast nineteen twenty,
it was the actual COVID experience had come within the
one in two hundred probability, But for Spain it was
much bigger. In fact, the both the Spanish flu with
Spain and the COVID in Spain had much heavi were

(47:29):
much heavier, about twice as heavy as in Sweden, and
with Switzerland somewhere in between. So what we're doing on
that front is trying to find suitable fatter tail distributions
our kosis. Cortosis is the fourth moment divided by the

(47:53):
force power of the standard deviations. It's a measure of
the fat tailedness and for a normal distribution it's always three.
For a plus distribution, it's alwas six for a Sculer
plastic can get a bit higher than six, but not
very much higher. But we were getting cortosis of fifteen

(48:14):
or so, so that wasn't satisfactory for the for the
COVID modeling and an alternatives we've been looking out. There
is a mixture model. Assume x one and x twour
two random variables, both for example, normally distributed, but there

(48:36):
might be any distributions the plus scuto plus or something
else hyperbolic and why it's a random variable which is
equal to x one with the probability p and equal
to x two with a probability one minus p or q,
and that can get much fatter tailed. And we're still

(48:57):
working on that as a as a way of MHM
modeling the past mortality and as I mentioned later, this
helps source of modeling past investment experience, which can also
be much faster tailed than normal, and so I've been

(49:18):
talking about that without without bringing the working model in,
because the working model is soon so different and also
requires a raja rather longer story to explain how I
thought of it and how the different bits of it
come in, and a lot of others. But that that
I think has explained my my motivation in at least

(49:42):
first my accidental getting involved with computer programming now called
coding perhaps, but both the program design and the coding,
and how I took to it and used it for
practical purposes in mortality, investment, in dises, income protection, a

(50:10):
aid all for the benefits of the of the profession
and COVID more recently of my own interest and also
for the interests of the profession too, and what we're
doing on the long term modeling of mortality. I hadn't.
I'll go back for a moment just mentioned forecasting. I'd

(50:32):
mentioned that we did do some forecasting, was the annuity
annuity tables in the nineteen eighties, nineteen seventies, and later
on John Cutchan and I did a different method, But
the CMI has now got a much more elaborate forecasting method,

(50:58):
also deterministic, I think it, but And there's the deterministic
way of looking at immortality table is that if simply
you start with one hundred thousand people people and so
many die, that's indeed nort and were in the first

(51:19):
year seventy die, and second year that's d one. But
we know that these are only really only probabilities and
numbers that die around the variables. And if we're wanting
to do some forecasting, we would want to bring in
a plus on number of deaths or normally distributed number
of deaths, which may be important for some small pockets.

(51:43):
I remember when there's at Watson's the actor who was
dealing with the British rail pension fund. So there's it
so happens that British rail is structured at all. The
previous railway companies have kept their own pension funds. They
haven't been the amalgamy. There's pension funds for the old
Great Western Railway pre nineteenth nineteen twenty three I think,

(52:06):
and the old Great Northern and and they have got
about ten people in them. And should we not allow
for some uncertainty and how long they might live the
massid but yes, the safe things, just to value them
as if they're all neurities certainty age one hundred and
five or something like that. But one could use stochastic

(52:29):
methods for valuing very small pension funds or if they
if it was necessary to keep them separate and not
amalgamate them with something else. So that although so much
of it has been disturbinistic and what in the mortality
rates and what I've been doing, you can bring stacastic

(52:50):
modeling into it too.

Speaker 1 (52:54):
That's yeah, that's that's a really good overview of you know,
some of the modeling that you've done. You talked about
mortality forecasting, the investment indices, income protection, and the key
theme there is that these were enabled by your early
adoption and promotion of the power of programming. Now I'm

(53:15):
curious to get your thoughts on the evolution of programming.
We talked about, you know, programming, coding and how that's
done in insurance and risk management today. A couple of
things I'll mention. Of course, open source learning is very common.
And what's interesting and you heard you hear the CEO
of and video he said, so a couple of times
is with this low code no code paradigm that not

(53:37):
just this industry but a lot of industries who use computing,
who rely heavily on computing power, are moving to where
you can actually execute code without writing a single lineup code.
I'm just curious to get your thoughts on that, because
that's a very that's very different, I think today than
it was perhaps when you started.

Speaker 2 (53:54):
Well yes, but come on, come on to that later too.
But even with ai AI, they talk about programs learning
or programs writing routines. Well, these are not. It's not
the somebody has written the coding of a program which

(54:15):
will alter some numbers or some bit of code elsewhere
cerve the machine apparently learns or.

Speaker 4 (54:23):
The the program, the master program that somebody who has
some computer program or computer or coder has written, will take.

Speaker 2 (54:37):
In instructions and constructor knows how to construct coding or
sub routine to do that. But that's because the coder
in the first places has enabled the has set up
a complicated program which can write other programs. Now this happens,

(55:06):
This can happen with a compiler. You write a code
in some code in Fortron, you put it into through
a Fortroun compiler, which constructs a machine code version of it.
Everything everything is in the computer is actually carried out

(55:28):
in machine code. You can have an interpreter program which
where you give it line by line and it doesn't
construct a machine language program, but it takes the code
in line by line and constructs the machine code to
obey it. It's usually much slower than having a precompiled version.

(55:50):
But all of it, whatever the top level program is
doing at the bottom level is just machine code moving around.
Thoughts and one nothing else.

Speaker 1 (56:02):
Yes, No, lastly, one thing I wanted to ask about.
You've been in the profession, or you spent a lot
of years in the profession. As you reflect on your
actuarial career, what do you think was the most significant
development in the profession our computers.

Speaker 2 (56:21):
When I started working the Scottish audios in nineteen fifty one,
the little the actuarial department had about twenty actuarial students
and it split into ten pairs called a boat, and
we went round. We got in a bundle of requests
please calculate surrender values or loan values for these policies

(56:45):
letters hugely from banks or solicitors. We went round gathering
the data from the policy registers or other registers. See
what femiums had been last paid, calculated the surrender value
either use sometimes using log tables or calculators or very
long cylindrical slide rules, and check that we got the

(57:11):
right answers. And the pair of us could do perhaps
thirty a day. That could be done in seconds now
with one person putting in a policy number and accessing
the database. And those jobs have disappeared. Different jobs have
been created in the programming coding departments. So also it

(57:36):
was also possible with the computer to do valuations every year. Previously,
with even with a bunch of cards, it was so
laborious that you might do a proper valuation every three
years or so. You couldn't have declare bonus every year
because that involved the entire staff doing overtime for three
months to do individual calculations. And so I think programming

(58:01):
has made a huge difference to the way that actories think.
And because of what I've shown in as ways of
opening up different ways of doing things, that has made
a great change. Previously you could only do things which
you could do with either an army of clerks or
clever or clever algebra. Now you can do it with

(58:25):
a computer. You know, you certainly the army of clerks.
You don't need to be too clever with the mathematics.
You can do it approximately. That's what I see. That's
a change, But that's perhaps what i've I see and
what I've done. If other people may see other quite
different things.

Speaker 1 (58:45):
Yes, it's hard to argue with that. I think, you know,
very timely observation. So just just just to recap what
we've spoken about so far. We've spoken about you've learned
about David's backgrowing growing up in Scotland, we got into
actual science, how he was an early adopter in utilizing

(59:05):
the power of programming to utilize models to make a
difference not just for insurance companies but in the broader
society as well. And you know, David talked about the
importance of programming and computers more broadly speaking in terms
of being one of the biggest changes within the industry

(59:26):
or within the profession throughout the course of his career.
So this conclusive first part of our interview. In the
second part of our interview, we're going to be going
into the famous Wilki model and exploring the realm of
modeling uncertainty. So going into the realm of sarcastic modeling.
So thanks for joining us and you can continue to

(59:47):
watch in part two.

Speaker 2 (59:50):
Thank you, David, thank you

All Episodes

Episode 100: Actuaries in Research - Part 2 (Part 1)

Episode Transcript

Popular Podcasts

Stuff You Should Know

Cardiac Cowboys

The Joe Rogan Experience

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}Episode 100: Actuaries in Research - Part 2 (Part 1)