s-news
[Top] [All Lists]

bootstrapping time series

To: s-news@wubios.wustl.edu
Subject: bootstrapping time series
From: Priya Kulkarni <priyak@pstat.ucsb.edu>
Date: Mon, 17 Feb 2003 00:14:48 -0800
User-agent: Internet Messaging Program (IMP) 4.0-cvs

Hi,

I have simulated 100 time series from ARMA(1,1). 

For each, I have fitted two AR models - one with AIC AR-order and the other 
with 2*AIC AR-order. Later I am generating 99 bootstrap samples from each 
model. 

Here X is a 100*64 data matrix (where 64 is the size of each data series and 
the resample).

When I fit the first model as follows, everything works perfectly.

for (i in 1:N){
.......

x.1<-ar(X[i,],order.max=25)
x.1model<-list(order=c(x.1$order,0,0),ar=x.1$ar)
x.1res<-x.1$resid[!is.na(x.1$resid)]
x.1res<-x.1res-mean(x.1res

x.1<-tsboot(x.1res,x.fun,R=99,l=4,sim="fixed",ran.gen=x.black,ran.args=list(ts=X
[i,],model=x.1model),n.sim=length(X[i,]))

.......
}


Next, for the second model:

for (i in 1:N){
........

x.1<-ar(X[i,],order.max=25)
p[i,]<-min(25,2*x.1$order)
x.2<-ar(X[i,],aic=F,order.max=p[i,])
x.2model<-list(order=c(x.2$order,0,0),ar=x.2$ar)
x.2res<-x.2$resid[!is.na(x.2$resid)]
x.2res<-x.2res-mean(x.2res)
                        
x.2<-tsboot(x.2res,x.fun,R=99,l=4,sim="fixed",ran.gen=x.black,ran.args=list(ts=X
[i,],model=x.2model),n.sim=length(X[i,]))
.......
}

the for-loop terminates the program before going through all N values giving 
the following error : 

Error in .C("S_arimasim",: Could not invert cov: 18
Dumped
Error was while calling subroutine "S_arimasim"


Sometimes it goes over 90+ values (but <100) but most of the times it stops 
quite early (ranging from N=5 to N=40) and I am totally at loss as to why this 
happens. It looks like the problem is in 'arima.sim' and not in 'tsboot'. 

Has anyone seen this problem before? 

All suggestions and ideas are welcome!

Thanks,
Priya


<Prev in Thread] Current Thread [Next in Thread>