CODEWITHSUNDEEP
rtime-seriesautoregressive-models
T_stats_3
T_stats_3

Reputation: 155

Generating Non-Stationary Time Series in R

I wish to generate an AR(3) model with parameters (3,0,5) for each of the phi's. This is non-stationary and arima.sim gives the error 

Error in arima.sim(model = list(ar = c(3, 0, 5)), n = 50) :'ar' part of 
model is not stationary

Is there a way this can be done in R?

Upvotes: 1

Views: 2754

Answers (1)

Mark S
Mark S

Reputation: 613

What process variance (SD) did you want to use? If unspecified, arima.sim() will use a default of 1 so I'll assume that's also the case you want here.

# empty vector for process
xx <- vector("numeric",50)
# innovations (process errors)
ww <- rnorm(50)
# set first 3 times to innovations
xx[1:3] <- ww[1:3]
# simulate AR(3)
for(t in 4:50) { xx[t] <- 3*xx[t-1] + 5*xx[t-3] + ww[t] }

If you don't want Gaussian errors, replace ww with some other distribution.

Upvotes: 3

Related Questions