DRAFT STANDARDS FOR THE DEVELOPMENT OF SPREADSHEETS

Anthony Asher
10 August 2000
1 SPREADSHEETS CAN BE ASSETS
Spreadsheets have always been used in actuarial work. The increasing power of PC packages
is allowing them to become more sophisticated. As the spreadsheet can frequently be re-used,
it is important that actuarial departments develop standards in order that new users can
understand them readily.
I have, however, been unable to find anything that sets out the standards required. This
document is a draft of such standards for use in our University courses, and potentially, in
work situations. It assumes a modern package such as Excel or Lotus that allows for multiple
spreadsheets in a workbook.
My experience, reinforced by the questioning of actuarial students over a number of years is
that, while spreadsheets are universally used in actuarial departments, few are actually seen as
company assets and continually developed for re-use. This appears to me to involve a waste of
time and effort. Spreadsheets are adequate for many of the tasks of actuarial modelling, but I
cannot see why they should be redeveloped every time the job passes from one member of
staff to another.
Spreadsheet standards are not really different from the requirement of all work to be
adequately documented. They require discipline, but pay dividends. They turn expenses into
investments.
2 FILING
Before even beginning with standards for workbooks, there is no point in developing
something to be used again if it cannot be found. It seems to me that an office requires a
standardised filing system that should be used for the directory trees on each and every
computer. It should differentiate between such functions as valuation, analyses of experience,
correspondence by other department, etc. Each new member of staff needs to be trained in its
practice.
3 PURPOSE OF THE WORKBOOK
It is possible to explain the purpose of the workbook under "workbook properties" and this
should be used if it is standard practice to do so. It is, however, also helpful to repeat and
elaborate such an explanation in the first set of workbook sheets. Such an explanation should
include:
The purpose of the workbook.
Any unusual maintenance / updating required before use.
Reference to results that future users may find helpful.
Reference to any other documentation.
An index to the other sheets of the workbook. These other sheets should ideally have
easily remembered mnemonic names (so as many as possible appear on the screen at
once.)
Who is currently responsible for its maintenance, and dates of major updates. It may be
wise to limit access to the file or restrict the ability to update it.
List of dates of previous major changes, who was responsible and the location of the
past versions and backups.

3 INPUTS AND PARAMETER

The next sheet can contain the inputs and parameters required. As the robustness of many
models to changes in input values is one of their main points of interest, it is critical to be able
to change the inputs easily. It should be possible to "parameterise" every one of the variables
of interest.
The parameters should all be named, so that formulae can be read more easily. (Names should
only be used in input sheets, as they can become difficult to trace otherwise.) A list of
commonly used abbreviations is attached.
E.g. Interest rates are invariably described as "i" in actuarial formula, so if one interest rate
is used in the workbook, it can be named "i" in this second sheet. If more interest rates are
used the can be called "i1", "i2" etc or given more descriptive names.
One of the aims of having an input sheet is to ensure that each inputted item is given only one
value, and all references to the item are based on the original. This means it is critical that
nothing is manually copied from one part of the spreadsheet to another. Using different
colours can be helpful - if everyone in the office is familiar with the colour scheme. (Some
10% of males including the author - are green or red colour blind, so people should not be
expected to distinguish between greens and reds.)
This also means that any text explaining the inputs should not make reference to numbers as
part of the text.
E.g. Do not say: "I have used a rate of interest of 9% as being 2% more than inflation, and
then put 9% in a neighbouring cell. Rather set it out so that the numbers are in the cells that
have been named, and are used for reference purposes:
"Rate of inflation:
7%
Real rate of interest:
2%"
Tables (vectors or matrices) should be in this first input spreadsheet (or in separate input
sheets) and be named and accessed by lookup functions. Frequently used tables, such as
mortality tables, inflation and past investment returns, can be stored in a common workbook
on a network server and accessed from there. They should be displayed - if possible - in both
tabular and graphical form in the "input" spreadsheet so they can be checked for
reasonableness.
All inputted items that come from other sources should be referenced. A reference requires a
name, date, author and location.

4 WORKING SHEETS
The actual working spreadsheets can then follow. A few guidelines can be suggested:
Headings should be kept as brief as possible as they will otherwise clutter the page and
any graphs that are created from the data. If a longer commentary is necessary, then a
comment can be attached to the cell.
Formulae (especially logical ones such as nested if statements) should not be long. It
is as easy to create new columns (or rows) as complicated formulae, and very much
easier to follow when you are trying to interpret a spreadsheet someone else has
created. Columns and rows showing intermediate values can be reduced (but not
hidden as this becomes confusing) in order to allow as much as possible to be visible
at one glance.
Where alternative notation is available to a formula, use the closest to standard
notation. (e.g. 2^3 is preferable to POWER(2,3))

When copying formulae down and across, the copy should go right to the end of the
column or row. If it is necessary to change formulae mid-row or mid-column, then a
new spreadsheet should be started. This is to avoid inadvertently copying changes
over the wrong formulae. If this is particularly awkward, then a warning should be
included in the first cell of the row or column.
One wants to see as much of the spreadsheet at one time as possible, and there are
more rows than columns on a screen. It frequently happens that useful data is off the
page to the right - while the different rows that can be seen include formulae that have
been copied down. In these circumstances, it makes sense to transpose the data so that
formulae are copied along the rows (and one can, therefore, understand what is out of
sight) and each different row gives different information. All working spreadsheets
should be similarly transposed.
The headers and row titles should be "frozen" so that they remain visible when the
numbers are scrolled.
The spreadsheets should be linked so that references are automatically updated.
Calculation should also be set on automatic when the workbook is saved.

Repetition of cells should be avoided refer to the place where the results were originally
generated.

Spreadsheet default settings should be maintained in order to make it easier for new
users to follow. E.g. Cell references should use the standard format A1, $C$10 etc
rather than [1:-5] or whatever other options are available..

5 OUTPUT
The output should preferably be on a separate sheet - even if it is merely a single number.
If it desirable to see the effect of different input parameters on the output, then windows can
be created to view both simultaneously. It is helpful to create a record of the effect of the
different input values in a separate output scenarios spreadsheet with a table setting out the
different input parameters and output values. This not only saves future users from repeating
work, but may give them insights they otherwise may not have gained.
Output tables should also be graphed when appropriate. Graphs should be labelled clearly and
have a clear background as they are easier to see and much quicker to print.
Any printed output should include the name of the spreadsheet, the date and the "author".
6 TESTING
The testing of the spreadsheet is critical.
-

Setting all rates to zero can often check the essential integrity of the workbook. (It may
be necessary to set interest rates just above zero to avoid dividing by zero.)
In many cases, obvious errors can easily be seen if each set of both input and output
data is graphed.
Other tests at extreme values of the parameters can also prove useful.
Where possible, results should be checked in more than one way. Such checks can be
a built in part of the workbook and be reported in the output sheet.
Spreadsheets will not automatically provide the required level of accuracy. Especial
care should be taken with the results of multiple iterations, the tails of statistical
distributions and differences between large numbers.

7 PRODUCTION USAGE
If the workbook is to be used for "production" - especially by non-technical people, then it
may need to be made more aesthetically pleasing and require more explanation.
Drop down boxes can be used to facilitate accurate inputs.
The cells that require inputs should be highlighted, and all others write-protected. Print areas
should be pre-set and also protected if possible. Output and input should be printed together,
and if more than one page is required should include a run number or time on each page.

8 TYPICAL ACTUARIAL ABBREVIATIONS

A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z

Assurances / Assets at time 0

Actual ... (when combined with other letters)
Bonus rate on bonus/ Assets at time 1
Cx / Claims in accounts / Capital
Dx
Expenses in accounts/ Expected ...
Fund in accounts
Investment income in accounts
Present value of capital payments
Lx / Liabilities
Mx
Nx
easily muddled with zero
Premium(s) in accounts / Price
Rx
Sx
Period of investigation
Solvency margin in assets
tVx - reserve
tWx - paid up sum assured
Zillmer allowance

CPI Consumer Price Index

(E)DS (Expected) death strain
dy Dividend Yield
ey Earnings Yield
NP Net Premium but not NPV as net present value
PF Policy Fee
ALSI All share index of the JSE
ASX Ausralian
FI Financial and Industrial Index

a annuities
b bonus rate on SA
c renewal expense as % of P
d dx - discount rate
e renewal expense allowance
f inflation rate
g dividend growth% / coupon%
h % married
i nominal interest rate
j real interest rate
k Commission loading as %
l lx
m frequency in year
n policy term
o easily muddled with zero
p px
q qx
r r
s sx
t time / tax rate
u profit margin in P
v 1/(1+i)
w withdrawal rates
x age
y age of second life
z age of third life