Threshold Regression AR

[spiatkow]

Hello,

I am trying to estimate threshold regression. I am a new Eviews user. I am also not very familiar with statistics. Thus, I would truly appreciate your help. For my study, I use a pooled, cross-sectional time series data set comprised of the seventeenth Western European nations. The data included in the sample encompassed 56 years (1956-2012), which yielded 895 country/year combinations for the analyses. My data have missing values for GDP, and misses few years for suicide rates for some countries.
For my main model, I estimate change-to-change fixed effect model. This model is estimated with Generalized Least Squares using cross-section weights (PCSE), and with cross-section weights. As noted above, I use fixed effects for nation (cross-section fixed effects). The results are the same if I use cross-section Fixed effect or “c” constant under Cross-specific coefficient . In addition, I included AR(1) given the issues of autocorrelation.
Here is my main model:
d(suicide_rate_st?) ar(1) (d(log(gdpcap_bank?)))
In this model, a unit change in IVs (logged GDP per caapita) produce B unit change in the DV (suicide rate) and this effect is the same throughout the range of the IVs. Now, the argument is that there may be certain thresholds that must be exceeded before socio-economic fluctuations (GDP) affect suicide rates, and that a threshold model would have to specify (or estimate) the values of the IVs (the thresholds, in this case GDP) at which the effect change.
I am not sure, however, how to estimate that. When I tried to estimate threshold regression in the same “unstacked page”, I am getting an error “ gdpcap_bank? is not defined” Thus, I assumed that I should use “stacked page”. In “stacked page” I created new object, select estimate, then under “estimation settings: methods” I selected threshold regression. In the box, then under “equation specification, dependent variable followed by list of threshold repressors”
d(suicide_rate_st) ((d(log(gdpcap_bank))))
Then, under “threshold model specification”
((d(log(gdpcap_bank))))
Now, I have few questions. First, given the main model described above, where should I include my intercept – under “dependent variable followed by list of threshold repressors” OR “List of non-threshold repressors”? How do I include fixed effects for nations? In main model I was getting in the output intercept for each nation, for example 4010--C for county with code 4010. I tried to copy that (4010—C) to the box “List of non-threshold repressors”” but I am getting an error “Illegal lag or index specification for coefficient c”
How do I address the issues of AR(1)? I know there is a SETAR model, but my threshold variable is GDP. Should I keep GDP as my threshold variable and include lagged sucicide under the list of threshold repressors d(suicide_rate_st) d(suicide_rate_st(-1)) ((d(log(gdpcap_bank)))). Would that address the autocorrelation?
Finally, my results change depending on number of thresholds, and trimming percentage. No thresholds are selected when I used all the threshold specifications, with exception to “Fixed Number – global” and “Fixed Number – sequential” I know in these two option I choose the number of thresholds, but how can I determine this number and trimming percentage? What number would be appropriate? Is there any test for it? Or is it better to use other options, even if they give an output “no threshold selected” Perhaps my missing values is the reason for this output?
Finally, should the output and probability be significant to actually say that there is a certain threshold? What if the probability is not significant?
I would also appreciate any references related to this topic.
As a final note, if you do not mind, I have a quick question about interactions. If GDP has a negative effect on suicide and divorce has a positive effect on suicide, and then the interaction between GDP and divorce on suicide is negative, does it mean that the effects of GDP on suicide rates is greater as divorce go down (or up?? ). Same sig, negative main effect of GDP, nonsig. Positive Democracy index, and interaction between these two Democracy*GDP is positive and sig. Does it mean that that the effects of GDP on suicide rates is greater as democracy go down (or up?? ).
Thank you very much for your time and help.
I am sorry it is a lot, but it will help me tremendously.
Best,
Sylwia

[EViews Gareth]

EViews does not support threshold models for panels.

[spiatkow]

Gareth,

Thank you for your quick reply.

Sylwia

[spiatkow]

Dear Gareth,

I have another quick question. I included interactions between GDP and Divorce Rate in the model that I described in my previous post
d(suicide_rate_st?) ar(1) (d(log(gdpcap_bank?))) (d(divorce?))) (d(log(gdpcap_bank?)))*(d(divorce?)))

Is there any option in Eviews to plot the conditional effects/marginal effects of this interaction? I know it is in STATA as postestimation procedure. However, I cannot find it in Eviews. Would you please advice where I can find it in Eviews?

Thank you so much!
Sylwia