Probably density from soerp

[Padarn]

Hi there,

I was just wondering if there is an easy way to use the result of soerp operations as random variables. For example

x=N(0,1)
y=N(1,2)
z=x+y

Then, can I get the pdf of z easily?

EDIT: Actually I see now (from a plot message) that this is not yet possible.

Did you have anything in mind about how you might go about doing this? Something like edgeworth expansions?

[tisimst]

I've been wanting to implement this for a while now, but I haven't been able to get around to it yet.

There are definitely some ways to go about it, primarily with matching distribution moments to some flexible distribution's moments, like the Generalized Lambda Distribution (GLD), the Pearson family of distributions, or the Johnson family of distributions. I have looked into this a couple of times, but I haven't found a really good solution yet that doesn't require a massive record of data tables for matching moments, though it may come down to that. I wish it were easier.

Have you tried my other package MCERP? It's a Monte Carlo approach to error propagation and it is far more capable and can plot both inputs and outputs, estimate probabilities, induce correlations, etc.?

I still plan on addressing this issue because its a fundamental feature that should be available, in my opinion. Thanks for bringing it up!

[Padarn]

Hi tisimst, sorry for the delayed response.

I was actually hoping to use the package as a means of quickly doing some approximate convolutions. I'm guessing you could do this with MCERP, but I think it might be a little slow for my purposes. I'll have a look though, cheers!