The RATS Software Forum

The GSTART, which is what is used by the MAXIMIZE, needs to be at least 1973:3 to allow space for two lags. This is discussed on page 293 of the version 8 User's Guide.

Hi Tom,

Running the code for daily data and accounting for the day of the week effect give me this result:

Code: Select all

MAXIMIZE - Estimation by BFGS
Convergence in    49 Iterations. Final criterion was  0.0000034 <=  0.0000100
Daily(5) Data From 1986:01:06 To 2009:12:31
Usable Observations                      6258
Skipped/Missing (from 6259)                 1
Function Value                    -22258.4669

   Variable                        Coeff      Std Error      T-Stat      Signif
************************************************************************************
1.  B                             0.005857982  0.004410815      1.32810  0.18414668
2.  BVEC(1)(1)                   -0.017187612  0.014597341     -1.17745  0.23901671
3.  BVEC(1)(2)                    0.060123708  0.021574885      2.78675  0.00532403
4.  BVEC(1)(3)                    0.027542465  0.022599354      1.21873  0.22294747
5.  BVEC(1)(4)                    0.000000000  0.000000000      0.00000  0.00000000
6.  BVEC(2)(1)                    0.005762853  0.004441507      1.29750  0.19445938
7.  BVEC(2)(2)                    0.046190846  0.014068071      3.28338  0.00102570
8.  BVEC(2)(3)                    0.008901474  0.024577363      0.36218  0.71721617
9.  BVEC(2)(4)                    0.019904373  0.010230970      1.94550  0.05171458
10. GARCHP(1)(1)                  0.439479548  0.164816862      2.66647  0.00766520
11. GARCHP(1)(2)                 -0.075708348  0.223201709     -0.33919  0.73446471
12. GARCHP(1)(3)                  0.648279094  0.251471458      2.57794  0.00993904
13. GARCHP(1)(4)                 -0.441756737  0.235726974     -1.87402  0.06092786
14. GARCHP(1)(5)                 -0.005387029  0.168472383     -0.03198  0.97449140
15. GARCHP(1)(6)                  0.104161327  0.009531225     10.92843  0.00000000
16. GARCHP(1)(7)                  0.890580041  0.009501739     93.72811  0.00000000
17. GARCHP(2)(1)                  0.115927112  0.024203096      4.78976  0.00000167
18. GARCHP(2)(2)                 -0.030752563  0.032670752     -0.94129  0.34655766
19. GARCHP(2)(3)                  0.028886206  0.035109360      0.82275  0.41065049
20. GARCHP(2)(4)                 -0.011782383  0.034019800     -0.34634  0.72908798
21. GARCHP(2)(5)                  0.011430565  0.025440282      0.44931  0.65320828
22. GARCHP(2)(6)                  0.094695751  0.007201808     13.14889  0.00000000
23. GARCHP(2)(7)                  0.893204379  0.008040614    111.08659  0.00000000

SIC for VAR   49376.60170
SIC for GARCH-M   44700.50774

Does it mean that I have daily effect on Mondays and Wednesday?

Also,

If I want to check if I still have arch effect can I still run this code:

Code: Select all

dec vect[series] ustd(%nvar)
clear(zeros) ustd
dec series[vect] garchu
compute %%garchinit()
gset garchu gstart gend = bb*yvec(t)-%%garchmu(t)
do time=gstart,gend
   compute %pt(ustd,time,%solve(%decomp(hh(time)),garchu(time)))
end do time
@mvarchtest
# ustd


Many thanks

Does it mean that I have daily effect on Mondays and Wednesday?

Yes. That would be the interpretation.

Ooops I am sorry.

So , If I understood you correctly, I need to set zero the daily variances that are negative (i.e. GARCHP(1)(2), GARCHP(1)(4), GARCHP(1)(5), GARCHP(2)(2), GARCHP(2)(4)) correct? And the ones with positive signs, if they are significant it means we have daily effect. Am i right?

This is the code ans the sries is attached.

Again, I truly appreciate all your help.

Dear Tom,


I need to run the IRFs after estimating the GARCH model taking into account the day of the week effect.

How can I adjust this line in the code below?

compute hirf(i+1)=garchp(1)(3)*hirf(i)+garchp(1)(2)*%if(i==1,eps0(1)^2,hirf(i)

Many Thanks

The deterministics don't affect the IRF's, so you just need to make sure you get the right slots for the two "GARCH" lag coefficients.

Hi Tom,

I am running this code for the IRFs



And getting this error

If you do Show Last Error on the Edit Menu, it will take you to the instruction which causes the problem, which is:

Code: Select all

[u]  compute hirf(i+1)=garchp(1)(7)*hirf(i)+garchp(1)(6)*%if(i==1,eps0(1)^2,hirf(i))[/u]

You have forum post underscore tags on the instruction.

Thanks a lot Tom, but I don't see what is wrong with :

compute hirf(i+1)=garchp(1)(7)*hirf(i)+garchp(1)(6)*%if(i==1,eps0(1)^2,hirf(i))

garchp(1)(7) and garchp(1)(6) stand for the GARCH lags!

There's nothing wrong with it. Your program had at the start of the instruction and at the end. Those will give syntax errors.

Thanks a lot Tom.

Hi Tom,

Using quarterly or monthly data, we usually include a full year of lags given that the primary effect of oil prices occurs at one year (based on the arguments by Hamilton and Herrera (2004)). Now, for daily data, since I have a large sample size (over 6000), the AIC/BIC criterion will do for selecting the sufficient number of lags in the mean equation. Can you please help me testing it in RATS? I have read Chapter 4 of Leamer's Specification Searches (Wiley, 1978) but still not sure how related to this.


Many thanks.

Just use @VARLagSelect (http://www.estima.com/forum/viewtopic.php?f=7&t=1261) on the basic VAR with the "M" variables.

Do you mean I need to add @VARLagSelect to my original code as such:

Just apply the standard @VARLagSelect to the model without any of the "M" terms. You're just trying to pick a defensible number of lags. If you tried to pick the number of lags based upon the full model, it would probably take a couple of days to estimate all the required models, and you would probably end up with the same answer.