Hi everyone; I have the following problem. I'm trying to fit a time series (linear regression with several dependent variables). And I'm using a Kalman filter; however... the fit is extremely good. What I mean is that the residuals are all extremely close to zero.While it would be great to get such a fit I think I'm doing something wrong. Is this OK? if you multiply the current varying coefficients times the current variables are you supposed to always get exactly the value in the dependent variable?If this is the case; I guess the best forecast for the time series would be to use the latest coefficients for future data?Plz help.

I would be surprised if you get a very good fit, especially with financial data. Could you give more details on your model ?

residuals of what?If it is resids from the latent factor it might mean that the model is overidentified and therefore spurious.

yes.. u get a good fit generally with KF. there are so many degrees of freedom potentially for each of the data point. further u are optimising in -sample. but out of sample, the forecasts will not be that good.

Add noise to your filter, you are probably setting 'Q' to zero. How are you determining your gain?