Risk Analysis by Simulation (Monte Carlo Simulation) : Michael Wood, October, 2006 (Minor Revisions January 2009)
Risk Analysis by Simulation (Monte Carlo Simulation) : Michael Wood, October, 2006 (Minor Revisions January 2009)
Risk Analysis by Simulation (Monte Carlo Simulation) : Michael Wood, October, 2006 (Minor Revisions January 2009)
Some theories and models are deterministic: they give you one definite, certain answer. Others
are probabilistic: they only provide answers in probabilistic terms. These models can either be
built using mathematical probability theory, or using simulation - sometimes called Monte Carlo
simulation because the results depend on chance in the same way as the results of gambling in a
casino in Monte Carlo. Simulation is both easier and more powerful in almost all contexts.
The idea of simulation is to express uncertainties in the model as probabilities, and then
use these probabilities to generate lots of possible futures. These futures can then be used to
estimate the probability that the real future will turn out one way rather than another.
There are three methods of estimating probabilities for the model. Sometimes past data
may be available (eg for the probability of a storm of a given degree of severity). Occasionally,
arguments about equally likely outcomes may do the trick (eg the probability of a coin landing on
heads is 1/2 since heads and tails are equally likely). When neither of these is possible we have to
use subjective estimates. Ideally you should ask several people for their estimates. Obviously the
probabilities should be realistic if you want to get meaningful results.
In practice, computers are essential for running simulations. @RISK is an add-in to Excel
which can be used perform Monte-Carlo simulation.
The notes below are a very brief introduction only. There is much more available in the
on-line help built into the package, and in the manual.
1
normal distribution) on the left hand side, move the hairs in the diagram, and enter numbers in
some of the cells on the right hand side. To enter a function in the cell, click on Apply.
Alternatively, you can use the Paste function icon (fx): this gives you a little more
information about what the parameters mean but you dont get the diagram.
Useful functions include RiskNormal, RiskUniform, RiskTriang (so called because the
distribution is triangular), RiskPert and RiskDiscrete.
The first three should be fairly obvious. RiskPert is a rounded off triangular distribution.
RickDiscrete is a bit more complicated to use. Before using it, you need to put two
columns of numbers somewhere in your workbook. The first - {x} - represents the discrete
(separate) values of the variable, and the second - {p} - represents probabilities of these values.
For example:
x p
0 0.2
1 0.3
2 0.5
You can then fill in the cell references in the function box. Right click on the cell and use @Risk-
Define distribution to see what the function looks like.
When you paste an @Risk function in the spreadsheet the number which appears in the
cell will be the mean or something similar. It will probably make it clearer how the simulation
works if you change the setting so that a randomly generated number from the distribution
appears in the cell. To do this, click on the Simulation settings icon (red graph with a square
under it), then click sampling and click Monte Carlo under Standard recalc. Now, each time you
press F9, the spreadsheet will be recalculated using different randomly generated numbers.
The @risk functions are saved as part of the spreadsheet (.xls) file.
Step 2. Mark the output cells you want the simulation to analyse
Next select the cells you would like analysed (get the cursor on these cells or highlight them), and
then click on the Add output icon (the one with the single red down arrow).
2
entered in the worksheet. To do this, get to the main menu in the Excel window, and click @Risk
Model List outputs and inputs. Now select all the inputs in the explorer window on the left of
the screen (use the Ctrl key), right click on the selected inputs, and click Correlate distributions.
You can now enter the correlation in one of the two appropriate places in the matrix. (The same
value should appear automatically in the other.) Then click Apply, and start the simulation. To
change the correlations you need to go to the @Risk correlations Sheet, which will have been
inserted in the Excel workbook
Other functions
There are many other facilities offered by @Risk. Use the manual or the on-line Help to explore.
Reading
@Risk on-line Help
Goodwin, P., & Wright, G. (1998). Decision Analysis for Management Judgment (2nd
edition). Chichester: Wiley.
Clemen, R. T. (1996). Making hard decisions (2nd ed). Belmont, Calif: Duxbury Press.
Clemen, R. T. & Reilly, T (2001). Making hard decisions with DecisionTools Suite Belmont,
Calif: Duxbury Press. (Includes a student version of @Risk.)
Palisade Corporation (2002). Guide to using @risk. Newfield, NY: Palisade Corporation.
(There should be three reference copies in the library.)