Given estimates of , , and , forecasts of are computed from the conditional expectation of .
In forecasting, the parameters F, G, and are replaced with the estimates or by values specified in the RESTRICT statement. One-step-ahead forecasting is performed for the observation , where . Here is the number of observations and b is the value of the BACK= option. For the observation , where , m-step-ahead forecasting is performed for . The forecasts are generated recursively with the initial condition .
The m-step-ahead forecast of is , where denotes the conditional expectation of given the information available at time t. The m-step-ahead forecast of is , where the matrix .
Let . Note that the last elements of consist of the elements of for .
The state vector can be represented as
|
Since for , the m-step-ahead forecast is
|
Therefore, the m-step-ahead forecast of is
|
The m-step-ahead forecast error is
|
The variance of the m-step-ahead forecast error is
|
Letting , the variance of the m-step-ahead forecast error of , , can be computed recursively as follows:
|
The variance of the m-step-ahead forecast error of is the left upper submatrix of ; that is,
|
Unless the NOCENTER option is specified, the sample mean vector is added to the forecast. When differencing is specified, the forecasts x plus the sample mean vector are integrated back to produce forecasts for the original series.
Let be the original series specified by the VAR statement, with some 0 values appended that correspond to the unobserved past observations. Let B be the backshift operator, and let be the matrix polynomial in the backshift operator that corresponds to the differencing specified by the VAR statement. The off-diagonal elements of are 0. Note that , where is the identity matrix. Then .
This gives the relationship
|
where and .
The m-step-ahead forecast of is
|
The m-step-ahead forecast error of is
|
Letting , the variance of the m-step-ahead forecast error of , , is
|
|
|
|
|
|