LAGS Instruction

LAGS list of lags

This is one of the subcommands of SYSTEM for defining a vector autoregression (VAR). Use it to list the lags of the endogenous variables in the VAR. Note that you need to use LAGS 1 TO P to get a full P lag VAR.

Wizard

The Time Series—VAR (Setup/Estimate) Wizard provides an easy, dialog-driven interface for defining and estimating var models.

Parameters

list of lags

The set of lags of each of the endogenous variables which will go into each created equation. Usually these will be consecutive lags as shown below, but you can skip lags. For example:

LAGS 1 2 3 6 12

Example

system(model=canmodel)

variables cangnp canm1 cantbill canunemp canusxr usagnp

lags 1 to 4

det constant

end(system)

defines a six-variable VAR with four lags of each variable plus a constant in each equation.

Notes

If you use VARIABLES and LAGS to define your VAR, each equation will include the same list of lags for each variable listed on VARIABLES. You cannot, for instance, use a different set of lags for the dependent variable or leave lags of variable z out of the variable x equation. You can get such flexibility by defining each equation separately using the instruction EQUATION. However, you cannot use a prior with such models.

If you use the ECT instruction to define an error-correction model, use the actual lag length that you want for the original, undifferenced model—RATS will automatically drop one lag when estimating the differenced error correction form. ECT models do not support non-consecutive lists of lags. If you use ECT in your SYSTEM definition, RATS will always define the model using all lags from 1 to L, where L is the longest lag specified on the LAGS instruction. For an error correction model with non-consecutive lags, you will need to create the differenced variables and error correction terms manually and specify the model in error correction form directly, without using ECT.