| |
 |
|
|
Science Forum Index » Statistics - Math Forum » Estimating parameters for MA(q) model...
Page 1 of 1
|
| Author |
Message |
| Robert Buchanan... |
 Posted: Wed Jul 09, 2008 6:14 am |
|
|
|
Guest
|
Hello,
I'm trying to teach myself something about time series analysis and
forecasting. I'm stuck on figuring out how to estimate the parameters
of a MA(q) model (exact likelihood function). To save time and use
mathematical notation I have written up some notes on what I have done
and posted them at
http://banach.millersville.edu/~bob/TimeSeries/MovingAverage.pdf
I'm trying to implement an algorithm for estimating the model
parameters using Mathematica. However the run-times of my algorithm
are very long (tens of minutes), so I wonder if I am approaching the
problem in an efficient way or even a correct way.
It seems there are several phases to an iteration of the procedure for
estimating parameters. First, assuming an estimate for sigma and
theta, estimate the preliminary values of a. Then calculate the new
values of a (this is where my algorithm is spending most of its
execution time). Then calculate sigma from the preliminary and new
values of a. Finally estimate theta. Repeat as necessary.
If someone has some very basic notes or suggestions for implementing
the algorithm for estimating the parameters of a MA(q) model or if you
can spot my misunderstanding of the process from my notes, please let
me know.
Thanks,
Bob Buchanan
j.robert.buchanan at (no spam) comcast.net |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Sat Sep 06, 2008 2:31 pm
|
|