Skalin Terasvirta JAE 1999 (STAR models/Nonlinear Causality)

[TomDoan]

These are replication files for (part of) Skalin and Terasvirta(1999), "Another Look at Swedish Business Cycles, 1861-1988", Journal of Applied Econometrics, vol. 14, no 4, pp 359-78. They fit LSTAR or ESTAR models to long annual data for quite a few Swedish macro data series and do a "non-linear" causality test, testing the null that a second series doesn't enter into the STAR model (either linearly or through its own STAR effect). This only does a few of the series, but demonstrates the techniques required.

stjae1999.rpf

Estimation of STAR models

(5.25 KiB) Downloaded 1423 times

staje1999causal.rpf

Causality tests

(2.82 KiB) Downloaded 1345 times

st-data.dat

Data file

(31.98 KiB) Downloaded 1308 times

descript.pdf

Data dictionary

(12.09 KiB) Downloaded 1318 times

[adrangi]

Hi Tom. I used this procedure successfully. However, I'm perplexed by the inconsistency of the chi-squared values (too small) and almost zero P-values for all lag lengths in the final causality tests. The P-values don't match the Chi-sq values produced by the procedure. What am I missing? Thanks for any help. Bahram

[TomDoan]

Hi Tom. I used this procedure successfully. However, I'm perplexed by the inconsistency of the chi-squared values (too small) and almost zero P-values for all lag lengths in the final causality tests. The P-values don't match the Chi-sq values produced by the procedure. What am I missing? Thanks for any help. Bahram

Those are F's not chi-squares.

[adrangi]

Thanks much. I see. So, restrictions are being tested. I'll review the program again. Thanks. Bahram

[TomDoan]

Thanks much. I see. So, restrictions are being tested. I'll review the program again. Thanks. Bahram

Sure. It's doing a standard causality test, but on linearized products of lags of the tested variable and the threshold variable.

[adrangi]

Hi Tom. I've used this procedure for nonlinear causality multiple times with success. However, tonight I'm getting error messages on the computation of ndf. I'm attaching that portion of the program and the error message. please have a look and maybe you can offer a solution. Thanks much. Enjoy the XMAS holidays. Best, BA

[TomDoan]

It looks like you have a serious problem with your XTEST variable---offhand, it looks like it's constant. When you have a problem, the first thing to do is always to check the data (AFTER any transformations), and the second is to take the NOPRINT option off the regressions.

[adrangi]

Hi. Thanks Tom. will check the data and take the noprint out. Best, BA