Retrieving AR & ARMA regression coefficients in BRMS model

Interfaces brms
1

Hi,

I am trying to fit an linear model with two random effects and an AR/ARMA effect, with normal family.
I then want to use the regression coefficiënts of the AR/ARMA effects to further sample from the posterior predictive distribution in a forecasting sense. If there is a solution for this, then that would solve the problem as well.

Using the AR() or ARMA() function does not return the wanted coefficiënts, how can I still obtain these?

The model I use is:
brm(log(vol) ~ (1 + arma(time = date_rank, p = 2, q = 1) | onoff : Market))

An example of the data:

Country date_rank date Section vol

1 Netherlands 2 2018-01-08 00:00:00 A 15735.
2 Netherlands 2 2018-01-08 00:00:00 B 755.
3 France 3 2018-01-15 00:00:00 A 93516.
4 France 3 2018-01-15 00:00:00 B 1635.
5 Netherlands 4 2018-01-22 00:00:00 A 45527.
6 Netherlands 4 2018-01-22 00:00:00 B 81913.
7 Netherlands 5 2018-01-29 00:00:00 A 43554.
8 France 5 2018-01-29 00:00:00 B 12503
9 Germany 6 2018-02-05 00:00:00 A 26751.
10Germany 6 2018-02-05 00:00:00 B 11144.

Thanks.

1 Like
2

Hi,
most of your prediction, needs should be satisfied by the posterior_predict/posterior_linpred/posterior_epred functions. You should also be able to extract the coefficients themselves either via the coef funciton or accessing the underlying RStan fit via fit$fit (and then using as.array or extract). But usually using the prediction functions is preferred as it is harder to make a coding error :-)

Best of luck with your model!