Calculation of standard deviation in stochastic simulation

[DannyBachman]

When you do a stochastic simulation, you can get confidence limits and a standard deviation series. How are these calculated?

[EViews Gareth]

Confidence interval is based on standard deviation. Standard deviation is simply calculated in the normal way from the simulation results.

[DannyBachman]

Sorry, you will have to pretend I am especially dense here (not hard to do, really). What is the "normal" way? Is that the sample standard deviation for a given period just the standard formula calculated for the observations for all the simulations in that period? I just want to be sure I understand.

[EViews Gareth]

At each observation of the solve, calculate the standard deviation of the simulated solution values. So you're not calculating a standard deviation across time periods, rather across stochastic solves.

[DannyBachman]

Thanks, Gareth. I think we are saying the same thing (you are confirming what I thought). We are working on getting the confidence limits directly from the simulations (e.g.), picking out the 97.5th highest and lowest simulation) rather than using the calculated standard deviations.

[EViews Gareth]

To be more concrete:

Code: Select all

create(page=page1) d5 1990 1991 rndseed 1 series y=nrnd series x=nrnd equation eq1.ls y c x eq1.makemodel(m) m.stochastic(p=page2) m.solve(s=s) show y_0s pageselect page2 series sd = @stdevpsby(y_0, @obsid) show sd