Recursion

 

When this option is checked SVAR uses this transformation when it calculates the recursive estimates of the non-zero eigenvalues from the trace tests. It also makes use of this transformation for the Hansen-Johansen fluctuation tests for the non-zero eigenvalues if the option below is check marked.

 

Writes the fluctuation tests (see Hansen and Johansen (1999)) for the eigenvalues into the output file when this option is selected. Also, additional graphs are shown when generating graphs for recursive eigenvalues.

 

Allows you to test the null hypothesis that the cointegration vectors are constant against the alternative that they are not. Requires that you have checked the "Parameter Constancy Tests" box in the main program window.

 

Computes a Nyblom statistic for the constancy of β when β is restricted. Theoretically its asymptotic distribution is unknown, but the small sample distribution can instead be approximated by bootstrapping. This option also controls the computation of the Nyblom test for β under β restrictions when you've imposed restrictions on α.

 

By default SVAR uses the scores directly when computing the Nyblom tests, as suggested by Bruggeman, Donati and Warne (2003). By selecting this option the program uses Taylor expansions of the scores as suggested by Hansen and Johansen (1999).

 

By default SVAR computes recursive eigenvalues and β using data up to period t (t=k,...,T). When you select this option, the full sample (period T) estimates of all coefficients on deterministic variables and lagged first differences are used. Only the remaining parameters are computed using data up to period t (t=k,...,T).

 

When checked all critical values are set to 1 and the Chow test is measured relative to unity.

 

If you would like SVAR to produce recursive estimates of α when estimating β recursively, then you should select this option. The recursions will always use the exactly or over-identifying restrictions on β. If you have chosen to estimate α under linear restrictions and decided to keep those restrictions, then the recursive analysis will estimate α subject to those linear restrictions.

 

This option has a similar effect as the previous option, except that the recursive Π estimates are displayed.

 

When the feature has been selected, the function "Recursive Trace Test" is visible on the Graphics menu of the Cointegration Rank dialog. SVAR will display a critical value in the graphs for the trace tests, but this critical value is only applicable for each individual recursion and not for the sequence of recursive tests as a whole.

 

When computing recursive estimates of α, β, the non-zero eigenvalues from the trace test, etc., it may happen that the largest non-unit eigenvalue of the companion matrix (if you have selected a cointegration rank less than the number of endogenous variables, then the companion matrix does have unit eigenvalues) is greater than unity. In that case, the fluctuation and Nyblom tests will typically not perform well. In addition, the recursive β (and α) parameters will not be easily interpreted. By enabling this feature you get a graph of the recursively estimated largest non-unit eigenvalue for the companion matrix when you graph the non-zero eigenvalues used in the trace tests and the recursively estimated β parameters. Note that explosive eigenvalues for the companion matrix are most likely to occur when you have opted to use the full sample short-run estimates.

 

When SVAR computes fluctuations tests (see Ploberger, Krämer and Kontrus (1989)) and Chow tests under linear restrictions on α, the computation may take a lot of time. If this option is checked, then  SVAR temporarily sets the maximum number of iterations to 100 and the convergence criterion to 1.0e-004 unless the tests are computed for graphical display.

 

Be default this parameter is set to 2/3. Any number between 0 and 0.95 is valid. Note that 0 means: "use as few observations as possible".