TF regressor <options> ;
The TF statement specifies a regressor that has a transfer-function relationship with the dependent series. A transfer function is useful for capturing the contributions from lagged values of the regressor. Box and Jenkins (1976) popularized ARIMA models that have transfer-function inputs. In the UCM procedure, you can specify a transfer function of the following type (assuming the regression variable is x):
This transfer function is specified by using the ratio of polynomials in the backshift operator B. The numerator polynomial orders () are positive integers, possibly with gaps (for example, 1, 3). The numerator term
signifies the delay of order d. The denominator polynomial can have two factors: a nonseasonal factor,
, and a seasonal factor whose season length is s,
. The orders of the terms in the denominator factors cannot have gaps; that is, if 5 is the maximum order of the nonseasonal factor, then all terms of orders 1 through 5 are present. By design, the denominator factors are restricted to be stable polynomials (their roots are strictly larger than 1 in absolute value). As an example, consider the following transfer function specification:
You can specify this transfer function as follows:
tf x num=(1 2) den=2 sden=1 s=4 delay=3;
Since the numerator polynomial orders do not have any gaps, the following simpler specification is also available:
tf x num=2 den=2 sden=1 s=4 delay=3;
Because the denominator factors do not permit gaps in their orders, only the maximum orders need to be provided in their specification.
A state space representation of a transfer-function relationship is described in the section State Space Form of a Transfer Function Relationship. You can specify multiple TF statements, each one with a separate regressor. A regressor that is specified in any transfer function specification must not appear in any other regression specifications, such as in the right-hand side of the MODEL statement or in the RANDOMREG and SPLINEREG statements.
Note: The mathematical form of the transfer function considered by PROC UCM is similar to the one considered in the ARIMA procedure (Chapter 7, ARIMA Procedure). However, there are some differences:
The sign convention of the coefficients of the nonzero-order terms in the numerator polynomial in the UCM procedure is opposite to that of the ARIMA procedure.
The ARIMA procedure permits multiple polynomial factors in both the numerator and the denominator. The UCM procedure permits only one numerator factor and at most two denominator factors.
The ARIMA procedure permits full control over the terms present in each of the polynomial factors. The UCM procedure does not permit such fine control over the terms in the polynomials.
In the UCM procedure, you cannot fix the coefficients of the numerator polynomial. They are always estimated from the data.
In the UCM procedure, if both nonseasonal and seasonal factors are present in the denominator, you must specify starting values for their coefficients either for both factors or for neither.
You can specify the following options in the TF statement: