I recently started joined the Monte Carlo Simulation, Excel and ModelRisk LinkedIn Group and found a very interesting discussion thread on the type of probability functions people use for there daily Monte Carlo modeling. Since I use RiskAMP, an Excel based Monte Carlo simulator, I found this thread very interesting indeed.
Here’s an answer from Lan Ge, scientific researcher at WUR:
I used to analyze risks related to animal epidemics and used distributions like: Beta, Binomial, Exponential, Erlang, Gamma, Normal, Lognormal, Pert, Poisson (of course!), Student, Weibull, Uniform (Discreet Uniform)..right now I am analyzing a broader range of risks (production, investment, planning), I notice that I am using more and more normal, lognormal, Pert, triangle, and student.
What I use in my modeling is Normal, Poisson, Pert, Lognormal, and Binomial a lot. What do you use?
In my day job as a project manager who uses a bit of monte carlo simulation I find getting the distribution right to be nothing more than icing on the cake. The more important problem is have you modeled risk appropriately i.e. do you understand the types of chains of events that create risk in the first place? If you can answer this question well, the selection of distribution really does not matter. You can use a basic triangular distribution and still have a workable risk model
There is a nice paper on this here: http://www.intaver.com/Articles/Article_EventChainMethodologyDetails.pdf
I was also going to add that if you are really into curve fitting distributions to data there is some software that will help you do that called stat-fit: http://www.geerms.com/
I have never tried it myself, but you may want to give it a go. An interesting thing i have been tempted to do with this piece of software is to feed it some market profiles for different styles of markets (bullish, bearish, sideways, etc) markets and then use it to give you some potential distributions. You could then build a montecarlo simulation that shows you different scenarios.
Caprica,
You hit the nail on the head, I spend more time figuring out the various risk events than actually doing the distribution. The trick is trying to find all those hidden risk variables before they go “subprime” on you.