Dear Researchers
I am big confused regarding the reset tests of my Model. I am doing the work on exports effect on output growth in standard production function. Specification of the model is lnrealGDP c Lnlabor + lncapital + lnexports. i am identifying the exports effect on GDP sector wise. I have estimated the separate models for aggregate, primary, manufactured and services sector exports by using ARDL approach but in all cases the t or F statistic values of Ramsey reset tests are very high means rejected the hypothesis that model is correctly specified. I have included the different lags of each variables but the results of Ramsey reset tests are remaining same.
My question what is wrong with my model and yours’s opinion how I will solve this problem
Regards
Irshad Hussain
Ok thanks alot Dear Ali
Can i add the term of trade as additional variables. Can it helps
Ok thanks Respected Mousumi Bhattacharya
Madam i have read your lecturer notes but it tells to include the relevant variables. How you think what are the relevant variables are possible in this framework. I also tried to include the real imports but the sign of import goods is negative. However, most of the empirical literature reported positive sign. Second, after including the real imports, the results of Ramsey reset test is same. what you think can the i tried in other functional form as i have taken all the variables in Natural log
thanks and best regards
Irshad
If you see your model as reduced form then you might include more demand-side variables, e.g. total GDP as an indicator of domestic demand
I think Vince Daly implies Log(total GDP volume) Or Log(real GDP)! Sometimes, outliers in the data cause nonlinearity which is what RESET is suggesting. Consequently, it makes sense to save your residuals and used the SEE of estimated equation to divide the residuals. Then, list the obtained data and identified the largest absolute resid (standardized residuals). Then add a dummy variable into your model. The dummy variable should be one and zero otherwise. Please understand that this will not destroy your model. It simply helps the reader of your work to appreciate the DGP.
Dear Arize
I have taken the gdp in real gdp, not volume of GDP
thanks for your suggestions. But how we divide the residuals you means i save the residual of the estimated model and then all the residual divide by SEE of the model. where found the large value of residual after dividing. From that year i derived dummy like 1 for the year of the larger residual value and so on. for other i put 0. either i am correct or not
Second, i used the ARDL approach to estimate the model but i have observed if i take the dependent variable like log(realGDP) in the model instead of d(log(realGDP) then the value of Ramsey reset test is very high, rejecting the Ho null hypothesis of specification error. How you think what is the problem with this function form?
thanks in advance
regards
irshad
@Sayed
Yes, it is correct and when I referred to GDP volume, I was referring to real (i.e., constant) GDP in for example 2010 prices in indexed form. I was not referring to NGDP (IFS, 99b). You need to try out what people tell you, instead of asking what we think caused it. This is because No one knows the DGP of your data but at least your RESET test says there is some type of nonlinearity or specification error. Use the log(real GDP) not d(log(realGDP).
Dear Arize thanks a lot
But i am going to establish the co integrating relationship among the concerned variables through ARDL approach. As i have observed from the literature that all the authors have used dependent variables in first difference in Unrestricted ECM-ardl model if i use without d i think it can not follow the same structure suggested by pesaran et al. 2001 so guide me in this regard how i handle this issue
I think you misunderstood me: ARDL is specified using the first difference but to test for cointegration using FPSS you should use the log level data. To see, what I mean please can you visit my RG cite and see my "Alternative approach to ARDL --it is consistent with Pesaran et al. (2001). Look in my RG data set. The final ARDL is of the form
dlog(RealGDP C ECm(-1) dlog(realGDP(-1) dlog( of other variables) plus an error term.
Dear Arize
First thanks alot. Yes i know that for establishing using the level data. I estimated the Wald test for co integration. Second where is your RG test please send me address of the site as i explore more detail regarding this issue
See my name above here, click their and go to my contributions then find data set
- Badges
- Science topic
- Similar topics
- Psychology