Page 1 of 1

Multivariate Local Level Model

Posted: Fri Jan 15, 2016 5:04 pm
by Rantmurph
Hi

I want to estimate a multivariate model with a common "local level". I started out with a bivariate model with separate levels etc.:

(1) y1 = s1 + x1*b11 + x2*b12 + v1, where the y1 local level is a random walk s1 = s1(-1) + w1
(2) y2 = s2 + x1*b21 + x2*b22 + v2, where the y2 local level is a random walk s2 = s2(-1) + w2

(The next steps are imposing a common level s1 = s2, and allowing for missing y2 data at the end of the sample.)

All of the random errors are uncorrelated. I can estimate (1) and (2) individually with DLM, but I am not able to estimate (1) and (2) jointly. I need to do this when I impose the common local level.

My starting values from the separate estimates of (1) and (2) do not appear to work. I have obviously made some silly mistake, but I cannot spot it.

My Excel data file and Rats program file are attached. I would be grateful for any assistance.

Thanks

Anthony

Re: Multivariate Local Level Model

Posted: Fri Jan 15, 2016 7:12 pm
by TomDoan
You're missing the Y option on the last DLM. BTW, I would recommend scaling the Y's up to 100 to get better scales on the variances.