The RATS Software Forum

Hi,

I am interested in replicating a paper by Chang and Serletis (2015): "Oil, Uncertainty, and Gasoline Prices".
This is the URL:
https://econ.ucalgary.ca/manageprofile/ ... -Jan15.pdf

specifically, I am interested in figure 6 and table 5 of the paper. To compute nonlinear impulse response functions as in Kilian and Vigfusson (2013) after accounting for the effect of oil price uncertainty.

Thank you

Have you checked with Serletis? He does a lot of his work with RATS.

I emailed him earlier today and waiting for his reply!

The Kilian-Vigfusson methods are shown at https://estima.com/forum/viewtopic.php?f=8&t=2242.

Hi Tom,

The Kilian and Vigfusson (2013) code doesn't account for the uncertainty effect in the garch-in-mean VAR. So, I am not sure how to compute the nonlinear IRFs using a bivariate GARCH-in-mean VAR.

Thank you

The GIRFs for the GARCH model are from some of Serletis' earlier work

https://estima.com/forum/viewtopic.php?f=8&t=1189

It looks like this paper combines the two ideas.

It does, but I wasn't sure whether the nonlinear IRFs account for uncertainty effect or not! In Kilian and Vigfusson they don't.

So, would I be able to run 10,000 simulations on RATS?

Hi,

Regarding Elder-Serletis(2010) VAR-GARCH-M paper, after running the code for the IRF where can I find the results for the plots? Is it in the series: GDPTOPLUS and GDPTOMINUS? I just need to get the results and draw the figures using different software.

Thank you for your time,

za2015 wrote:Hi,

Regarding Elder-Serletis(2010) VAR-GARCH-M paper, after running the code for the IRF where can I find the results for the plots? Is it in the series: GDPTOPLUS and GDPTOMINUS? I just need to get the results and draw the figures using different software.

Thank you for your time,

That's correct. Those are the responses to positive and negative shocks (of size equal to the square root of the sample variance of the oil price change) with the "M" effect included in the model. The same with "nom" at the end of the name are the responses in the model without the "M" effect.

Hi Tom,

I noticed that when running the code for the IRF using stock returns instead of gdp, the series for the stockreturnstoplusnom and stockreturnstominusnom does not show on the screen. i think the name is too long that's why. Is there a way that I can fix this problem without changing the name of the variable?

Here is the code:

*
compute shock=sqrt(meanhoil)
compute %parmspoke(allparms,withMParms)
@CalcIRF ||shock,0.0||
set stockreturnstoplus 1 nsteps+1 = irf(t)(2)
@CalcIRF ||-shock,0.0||
set stockreturnstominus 1 nsteps+1 = irf(t)(2)
*
* And without M effects
*
compute %parmspoke(allparms,noMParms)
@CalcIRF ||shock,0.0||
set stockreturnstoplusnom 1 nsteps+1 = irf(t)(2)
@CalcIRF ||-shock,0.0||
set stockreturnstominusnom 1 nsteps+1 = irf(t)(2)
*

Thank you for your time,

Hi,

I have another question about Elder-Serletis (2010) paper:

Are the bands established with 95% confidence intervals?

Thank you,

za2015 wrote:Hi Tom,

I noticed that when running the code for the IRF using stock returns instead of gdp, the series for the stockreturnstoplusnom and stockreturnstominusnom does not show on the screen. i think the name is too long that's why. Is there a way that I can fix this problem without changing the name of the variable?

Here is the code:

*
compute shock=sqrt(meanhoil)
compute %parmspoke(allparms,withMParms)
@CalcIRF ||shock,0.0||
set stockreturnstoplus 1 nsteps+1 = irf(t)(2)
@CalcIRF ||-shock,0.0||
set stockreturnstominus 1 nsteps+1 = irf(t)(2)
*
* And without M effects
*
compute %parmspoke(allparms,noMParms)
@CalcIRF ||shock,0.0||
set stockreturnstoplusnom 1 nsteps+1 = irf(t)(2)
@CalcIRF ||-shock,0.0||
set stockreturnstominusnom 1 nsteps+1 = irf(t)(2)
*

Thank you for your time,

Only the first 16 characters matter so, although you can use the really long names, they will actually be treated as stockreturnstopl and stockreturnstomi. You can call them whatever you want---there's no reason they have to start with stockreturns; returnstoplus and returnstominus will work just as well.

za2015 wrote:Hi,

I have another question about Elder-Serletis (2010) paper:

Are the bands established with 95% confidence intervals?

Thank you,

From the comments in the program:

* This will do 16- and 84%-ile bands around the median response which is roughly
* the equivalent of +/- 1 s.d. bands, but allowing for asymmetry.

I have another question, Tom:
Is there a way I can add a square on the graph that shows the significance at the 5% level, as in the attached graph?

Thank you,

Hi Tom,

How can I adjust the code to show the 95% confidence intervals in the series?

Thank you for your time,