CODEWITHSUNDEEP
rtime-series
ThallyHo
ThallyHo

Reputation: 2833

calculate auto regressive AR1 model coefficient over a rolling window

I would like to fit an autoregressive AR1 model to a time series data over a rolling window to extract the coefficient for each time step but over a fixed window, say 100.

I can calculate the coefficient for each tilmestep using something like this

dat <- data.frame(Last = runif(10000), time =seq(1, 10000, 1))

roll.ar<-NULL

##loop to calculate the rolling change  
for(i in 2:length(dat[,1])){
    dat.t<-subset(dat, time<=unique(sort(dat$time))[i])     
    roll.ar[[i]]<-ar.ols(dat.t$Last, aic = FALSE, order.max = 1, dmean = FALSE, intercept = FALSE)$ar[1]
}

But this doesn't use a fixed window - i.e. each successive calculation uses more data than the last. Any guidance gratefully received.

Upvotes: 0

Views: 458

Answers (1)

G. Grothendieck
G. Grothendieck

Reputation: 269644

Below we define a series of 20 values and compute the ar1 coefficient using a window of 10.

library(zoo)
set.seed(123)
dat <- data.frame(Last = runif(20)) # test input

AR <- function(x) {
  ar.ols(x, aic = FALSE, order.max = 1, dmean = FALSE, intercept = FALSE)$ar[1]
}
rollapplyr(dat$Last, 10, AR, fill = NA)
##      [1]         NA         NA         NA         NA         NA         NA
##      [7]         NA         NA         NA -0.2932866 -0.3146008 -0.3142310
##     [13] -0.2622454 -0.4233658 -0.1284332 -0.5366662 -0.6037607 -0.3294886
##     [19] -0.1573014 -0.1964080

Upvotes: 1

Related Questions