0

I'm trying to understand my "decomposition of additive time series" graph. Here's my code:

dbs_discs <- ts(RC$Disconnects, frequency =12, start=c(2013,1))
discs_dbs <- decompose(dbs_discs)
plot(discs_dbs)
discs_dbs

and my results:

$trend
          Jan      Feb      Mar      Apr      May      Jun      Jul      Aug      Sep      Oct      Nov      Dec
2013       NA       NA       NA       NA       NA       NA 301.8891 302.4746 302.6317 303.1842 304.2663 304.2212
2014 304.6779 306.3847 309.0182 310.5303 309.9420 309.1160 307.1276 304.2277 302.4454 301.2108 300.1494 299.7908
2015 299.5936 299.2328 298.4888 297.8479 297.3363 296.2674       NA       NA       NA       NA       NA       NA

As a result, my trend graph shows nothing plotted until mid 2013. Is there a reason why it's showing NA? What does it mean? Why would there be no values?

Thanks!

Jane Lee
  • 1
  • 4

1 Answers1

1

It seems the decompose function uses a 12-month 2-way moving average to determine the trend component of the series. (See ?filter and the code underneath decompose). That is, the trend value in July 2013 will be the moving average for the 6 months before and 6 months after (inclusive).

If you want to perform trend-cycle decomposition but don't want to trim off your end-points, perhaps it's worth looking at the mFilter package, which implements several filters. Note that in basically all trend-cycle decompositions there are end-point issues (ie. mistaking trend and cycle), so buyer beware.

Jim
  • 172
  • 1
  • 8
  • Thanks Jim. This was really helpful. I also found out about the STL function, which decomposes my time series data w/o giving me any NAs. Do you know what the difference is? Thanks – Jane Lee Oct 19 '15 at 02:43