1

I have a time series (ts) and used the MARSS package to create a state space model

fit = MARSS(ts)

giving the parameter estimates, the state estimates (fit$states) and their standard errors (fit$states.se)

But these estimates are just for the historic data series.

There is a great tutorial on how to produce these matrix models.

http://cran.r-project.org/web/packages/MARSS/vignettes/Quick_Start.pdf

But how can I use the historic model output matrices to make new matrix estimates and forecast 1,2,3 periods into the future?

Alex
  • 978
  • 1
  • 9
  • 22
adam.888
  • 7,686
  • 17
  • 70
  • 105
  • There is also a great blog:http://tr8dr.wordpress.com/2011/08/03/smoothed-utf/ and it would be very interesting to extend these ideas. – adam.888 Jul 10 '12 at 19:26
  • 1
    Ans:MARSSsimulate(MLEobj, tSteps = 100, nsim = 1, silent = TRUE, miss.loc = NULL) – adam.888 Apr 03 '13 at 17:59
  • I'm not familiar with `library(MARSS)`. But I just finished a forecasting project and made use of `library(forecast)`, specifically `forecast::forecast()`. I'm not sure it handles `MARSS` objects, but worth a shot – alexwhitworth Aug 25 '15 at 02:52

1 Answers1

1

A bit too late, but all you need to do to forecast is to input missing values at the end of the series and then MARSS will automatically fill those missings with the desired forecasts. Forecasting in state space models is equivalent to handling missing values...

slak
  • 11
  • 1