Dear all,
Does anyone use the S-Plus add-on, "splus fin-metrics". If so I have a
question.
I am still learning how to use fin-metrics and as so am following examples
in books. One example in the book 'Time Series Analysis by State Space
Methods' [by] Durbin and Koopman concerns the well known introduction of
compulsory seat-belt wearing in the front of cars in Jan 1983 in the
Britain. In their example Durbin and Koopman initially fit a structural
model with components: level and trigonometric seasonal (no slope) to the
whole series (1969 - 1984 inclusive) of logged number of drivers who were
killed or seriously injured in road accidents in Britain. They obtain the
following parameter estimates:
estimated variances of disturbances
level: 0.00093585
seasonal trig: 5.0109e-007
irregular: 0.0034160
log-likelihood: 435.295
prediction error variance: 0.00586998
My problem is that splus gives me different values; the estimated
disturbance variances I get are out by only a minute amount which doesn't
concern me but the log-likelihood and prediction error variance are out by a
considerable amount. I get:
168.8588 and 0.9999... respectively.
s-plus tells me that Kalman filtering failed, could this be the problem? and
if so why are my estimated variances of disturbances pretty much correct -
these I used to get the prediction error variance.
hope I haven't gone on for too long, thanks in advance for help
James
|