EViews.com

yingtan wrote:

ninjagirl wrote:Hello,

I wanted to find a consistent estimate of the Threshold as per Chan 1993. I can't seem to figure out how I can do that using the add-in. Any help? Im consistently getting threshold as 0.


Please help.

just delete 0, which is set as default, from the threshold value at the add-in

Hello,

I am trying to estimate the TAR and MTAR models.
1) However, after reading the docs supplemented with add-in, and previous posts, it is not clear for me, what estimates it provides for above and below treshold.
Logically, I assume that it provides regression of residiuals (saved from long-run regression) to lagged residuals above and below threshold plus to other differenced lagged residiuals. like equation (9) in research? If this is correct, then why do we need exogeneous?

or

Alternatively, does it provide estimates of error correction models? because the add-in provides option to include "exogeneous" variables which influence estimates of ECM model only? like (20, 21).


2) Quote from add-in docs: "By design, residuals of the error corection model are used as the treshold series in the analysis". what does it mean?

Please, can someone clearly explain for beginner user level thess issues.

Many thanks

ascent74 wrote:I am trying to estimate the TAR and MTAR models.
1) However, after reading the docs supplemented with add-in, and previous posts, it is not clear for me, what estimates it provides for above and below treshold.
Logically, I assume that it provides regression of residiuals (saved from long-run regression) to lagged residuals above and below threshold plus to other differenced lagged residiuals. like equation (9) in research? If this is correct, then why do we need exogeneous?

We do not need exogenous variable per se. You may need to control the relationship between endogenous variables (e.g. x1t and x2t). For instance, when modeling the relationship between two emerging market currencies, it is usually wise to use VIX as a regressor (say, z1t). Whether you put exogenous variable(s) into your model, residuals of the long-run equilibrium relationship (equation 2) is of interest here.

ascent74 wrote:Alternatively, does it provide estimates of error correction models? because the add-in provides option to include "exogeneous" variables which influence estimates of ECM model only? like (20, 21).2) Quote from add-in docs: "By design, residuals of the error corection model are used as the treshold series in the analysis". what does it mean?Please, can someone clearly explain for beginner user level thess issues. Many thanks

It means that, if you use exogenous variable(s) in the model, they are evaluated in Eq. 2 (from original study) and that residuals of this equation still refer to disturbance term Mu(t).

Dear trubadur

I'm trying to test for cointegrating TAR and M-TAR between to stock market indices a with constant and VIX as exogenous variables. However, when I try to run the TARCOINT with two endogenous variables, two exogenous (C and the logarithm of the VIX) as well as with the simulations for the appropriate critical values and inferences, there appears to be an error "EQ1 is not defined". I've also tried to group the variables but there's another error coming. I'm sure that the bug is somewhere with me, but I surely need Help.

Thanks in advance !

Ifranky wrote:Dear trubadur

I'm trying to test for cointegrating TAR and M-TAR between to stock market indices a with constant and VIX as exogenous variables. However, when I try to run the TARCOINT with two endogenous variables, two exogenous (C and the logarithm of the VIX) as well as with the simulations for the appropriate critical values and inferences, there appears to be an error "EQ1 is not defined". I've also tried to group the variables but there's another error coming. I'm sure that the bug is somewhere with me, but I surely need Help.

Thanks in advance !

It is difficult to say anything without seeing the workfile. I am guessing it has something to do with the constant term. The model already includes a constant, so you do not have to provide it explicitly (it would cause a singularity problem).

Dear Trubador,
i would like to ask about the asymmetric error correction model that i estimate after i have conducted the MTAR procedure through the tarcoint. Is there a problem if the coefficients of the Ut+ and Ut- ( i use monthly data) are 0,30 and 0,09? is that a sign of explosive series and is not good?

Thank you very much in advance

Dear all,

I am estimating TAR and MTAR (with threshold=0 and consistent) for two prices.

My question is if it is possible to make Eviews displaying the value of the AIC/SIC for each model I estimated.

Thank you in advance,

Bests

trubador wrote:Yes, it is possible. However, as I really do not see any particular use of it, I do not think it would worth the time to modify the code. Information criteria are only used in the selection of appropriate lag length, so you can always get that information at the outset without the need to use the add-in

Dear Trubador,

I'm asking for AIC/SBC because in some works these IC were used for determining the best fitting model among TAR, consistent TAR, M-TAR and consistent M-TAR.

Is the modified code available in the forum or do I have to recode it myself? (I have no idea about how to do it).

Thank you very much

You do not need to recode the whole process, I believe. Once you've estimated the model through add-in, you can write it in the equation form and estimate with OLS. The output will provide you all the necessary information criteria.

Dear Trubador,

Thank you for your answer. It would be great to have not to recode, but in running a single equation how do I specify the above and below threshold variables?

I tried to do the following (for a M-TAR model)

1. Saving the residuals from the Engle Granger first step for cointegration relationship;

2. Differencing the residuals ("genr dres = d(resid01)");

3. Generating the above and below threshold variables: "genr above=dres(-1)>=0"; "genr below=dres(-1)<0" [when Chang's method was applied, instead of 0 I typed the estimated threshold value].

As dependent variable I set the differenced residuals, followed by "above" "below" and "dres(-1 to -4)" (in Eviews' MTAR estimation 4 lagged residuals were included), that is:
dres above below dres(-1 to -4).

The OLS estimates are not the same, they are quite far from the estimates of the MTAR, so I assume I did something wrong.

If you could suggest the right way to estimate threshold models through OLS it would be very helpful.

Thank you very much,

Bests

TARCOINT add-in is now updated to version 1.1 as of June 23, 2016.

Fixed a minor bug causing momentum option not to work properly when using the command line syntax. Supporting files are provided to guide users through some post-estimation analyses.

Dear all,
I am using MTAR model using tarcoint add-ins in order to look for asymmetric unit root in single variable. My variable is profit margins of S&P 500 index.
Is it correct to use the model in this way and compare the results later to ADF results? What i did was to put variable name twice in the endogenous variables box and run model. I would like to ask you whether i did something wrong or correct. Thank you in advance

Dear all.
I am struggling with Threshold Erorr Correction Model in Eview. My model is delta P1t = a*delta P2t + b*delta P1(t-1) + c*delta P2(t-1) + d*[P1(t-1) - e*P2(t-1) - f] + error term. where a,b,c, d, e, f are diffent in each regimes and definded by P1t - P2t --> threshold variable and tc (transaction cost) ---> threshold value. tc is given.
I read some comments on internet and someone said that Tarcoint add-in can handle that. But I figure out tarcoint add in doesnt allow insert threshold variable.
Now I'm stuck. I dont know how to run this model. if anyone can handle it. Please tell me. I highly appreciate it

Best grateful.
trugia

trubador wrote:Yes, tarcoint uses Engle-Granger specification.

Is there a way for residuals of Gregory&Hansen specification to become the treshold series in the analysis? For instance by including appropriate dummy variables into the exogenous box when estiamting Tarcoint?

Dear all,

I would like to ask whether we can use tarcoint add-in for panel data. Thank you.