[R] R: forecasting a binary time series using the VLMC package

[Gauthier Pierard]

I would like to ask some clarifications on the method:

predict.vlmc

My problem is to forecast a binary time series one period ahead. I have a
time series bin2 of length 2000. When using

m2<-vlmc(bin2)
fc2<-predict(m2)


   1. fc2[i] is a prediction for i, not for i+1, is that correct? I am
   aware that the documentation stipulates "Compute predictions on a fitted
   VLMC object for each (but the first) element of another discrete time
   series.", but am still asking to make it 100% clear.
   2.

   I guess that the predictions fc2 are based on the full range [1:2000] of
   bin2, because I fitted a VLMC to the full timeseries on the first line
   above. Therefore, I am actually forecasting each period by already "knowing
   the future", is that correct?
   3.

   In order to forecast while "not knowing the future", can I do the
   following:

   for(i in 1000:1999) {
   retFull2 <- window(retFull, start=1, end=i)
   bin2<- window(bin, start=1, end=i)
   dummy<-ts(c(bin2,0))  #Adding a dummy zero at the end of each window
                     #so that a prediction will be made for i+1 as well
                     #without using i+1 while fitting the model
   m2<-vlmc(bin2) # bin2 granges from 1 to i
   fct2<-predict(m2, dummy)[i+1,1]  #forecasting on an "
artificially-added" i+1 index.}

   I am adding a "dummy" zero at the end of each windowed ts, and
   predicting for i+1 as well. Is it relevant at all? Any suggestions? Any
   practical suggestions on how to best forecast these binary time series?

Many thanks in advance, cheers!!!!!!!!

	[[alternative HTML version deleted]]