Dear Allstat users,
here a question about the correct interpretation of variances in Dynamic
Linear Models.
A DLM is basically formed by two equations:
1.
the observational equation which describes the relationship between the
observation "y" and an unknown parameter "a"; this equation can
generally be written as y(t) = F(t)a(t) + sigma(t)
where "F" is a regression vector and "sigma" is the observational variance
2.
the state equation which rules the updating of the unkown parameter
itself; it can be written as a(t) = M(t)a(t-1) + omega(t)
where "omega" is called the system variance
I usually apply DLM to environmental time series (rainfall and
temperature); is there a way to understand qualitatively the meaning of
these two variances and a way to evaluate a roughly sensible range of
their values? Can I do something simply looking at the original time
series and the way I expect the model behaves?
I hope this question is clear enough
thank you for your help
Stefano
|