[R] Newbie help: Data in an arma fit

[Gad Abraham]

caffeine wrote:
> I'd like to fit an ARMA(1,1) model to some data (Federal Reserve Bank
> interest rates) that looks like:
> 
> 
> ...
> 30JUN2006, 5.05
> 03JUL2006, 5.25
> 04JUL2006, N                  <---- here!
> 05JUL2006, 5.25
> ...
> 
> 
> One problem is that holidays have that "N" for their data.  As a test, I
> tried fitting ARMA(1,1) with and without the holidays deleted.  In other
> words, I fit the above data as well as this data:
> 
> 
> ...
> 30JUN2006, 5.05
> 03JUL2006, 5.25
> 05JUL2006, 5.25
> ...
> 
> 
> and the ARMA coefficients came out different.   My question is: Should I
> delete all the holidays from my data file?   What exactly does R do with the
> "N" values in the fit for the ARMA coefficients?
> 
> As a related question, the weekends don't have entries (since the FRB is
> closed on all weekends).  Does the fact that my data is not regularly spaced
> pose a problem for ARMA fitting?

A few comments:

* Is the time series stationary? You can't fit ARIMA to nonstationary data.

* One thing you could try is linear regression of interest rate on time 
and indicator variables for day of week and special days like holidays. 
Then fit an ARIMA to the regression residuals.

* Any specific reason why ARMA(1,1)? Have you looked at the acf and pacf 
of the time series?

Cheers,
Gad

-- 
Gad Abraham
Department of Mathematics and Statistics
The University of Melbourne
Parkville 3010, Victoria, Australia
email: g.abraham at ms.unimelb.edu.au
web: http://www.ms.unimelb.edu.au/~gabraham