Initializing Kalman Filter by training sample

Dynare help (legacy posts)
1

Dear all,

In paper reading I’ve realized that many empirical economists, instead of using initial value to start Kalman filter by some guess values, deploy training sample to reach that end. For example, the sample range is from 1960Q1 to 2012Q4; the training sample starts from 1950Q1 to 1959Q4. In most common case, beta and sigma by OLS will be used as initial values for kalman filter and ML estimation.

So I read Dynare manual, and only find presample option. I have no idea whether this option is designed to do the same as training sample. If not, then how am I supposed to do?

Thanks in advance.

richardgu26

2

You are correct. The presample periods are used to initialize the Kalman filter. Only time periods after the presample enter the actual likelihood computations.

The manual will be updated soon.

3

Good to know! Thanks for your quick reply!

richardgu26

[quote=“jpfeifer”]You are correct. The presample periods are used to initialize the Kalman filter. Only time periods after the presample enter the actual likelihood computations.

The manual will be updated soon.[/quote]