(EViews10):ARDL Models (General-to-Specific) #ardl #ecm #boundstest #cointegration #lags

from crunch econometrics thank you for joining me in the continuation of our series on a RDL models today we are considering general to specific approach in eviews general to specific approach we are simply here to look at how to modify ANOVA parameterize model to a parsimonious model what do we imply by an ova parameterized model this is a model having too many lags of a variable or a model with too many regressors or a model with too many regressors and their respective lags so what will be the consequences of such a model upon estimation you are likely to have many insignificant coefficients due to multicollinearity of the regressors so how can it be corrected rather than removing those variables arbitrarily it is important to subject them to scientific testing you can use it at war test or the likelihood ratio test so what is the world test all about the word test for coefficient significance know like Putin says is that these coefficients are equal to zero against the alternative data coefficients and not equal to zero and if a coefficient is equal to zero is simply means it is nor anyone's been in the model so what would be your rejection criteria you reject the null hypothesis if the probe value or the f statistic is lower or equal to 0.05 the second test is the likelihood ratio test for redundant variables the null hypothesis in this case is that the variables are redundant against the alternative that the null hypothesis is not true so what is the decision criteria you reject the null hypothesis if the value on the adjusted r-square from the path ammonius model is lower than that from the ova parameterised model so you can see that in the war test the F statistic will inform your decision the probe value of the F statistic will inform your decision unlike the likelihood ratio test where the value of the adjusted R squared will inform your so having given you all this preambles let us now go over to eviews to work out an example let me pause here by saying before you watch this video make sure you have watched the previous videos on ard ill because i'm still using the same variables and the link to this variable i have included it in the description for this video here are the three variables the log of manufacturing outputs the log of import and the real exchange rate the data span is from 1981 to 2014 last 34 years what's in this example I will only be using the log of imports as the dependent variable while the log of manufacturing value-added and the real exchange rates will be the regressors so let us quickly go on to quick click on estimate equation and here I list out the variables the methods I change it to a RDL in this dialog box I live the dependent variable and the regressors maximum lakhs at for the idea and I come to this place transpacific ation I select model 3 mother 3 is unrestricted constant and new trend you can see here the method is still a rowdy L my sample size is as shown I click OK so on the screen I have the results from the ova parameterised model and the way eviews does it each of these coefficients are serially numbered the first coefficient which is the lag of the log of import X number one the second lag the coefficient of the second lag of import text number two and the Sirian on brain goes on two constants with text number 15 because I have 15 coefficients on this one as you can see you can also see their respective standard errors the T statistics and the probability values I mentioned in one of my previous videos that the T statistics will make no sense or will have no relevance without the applicable probability value in other words it is the p-value that gives relevance or statistical significance to t statistic and from what we can see here as of the 15 coefficients aids are not statistically significant I'm not counting the constants I'm willing limiting it to the regressors so I'm having eight coefficients of the regressors not been significant and after 15 questions in total that is 53% 53% is way too high and that is not acceptable so I have to do something to this model so I begin by subjecting it to the wall test but if I go to the wall test and it will show you that indeed eviews Itachi's serial number in to these coefficients I go to view and click on representations and here you can see C 6 here this is a coefficient for the first leg of MVA you can see c10 here is the coefficient for the level of real exchange rates and you can see C 15 is the coefficient of the intercept let me conduct the war test coefficient Diagnostics a select war test those are the selected coefficients make sure you pick the correct one so I click OK on the screen is the results of the war test and we consider probability value at 0.8 1 this is clearly above 0.05 and in this case we cannot reject the null hypothesis that these coefficients are indeed equal to zero so I have to resubmit that model without despite variables and let us see how the coefficients would turn out to be so I copy out the entire equation I click on representation then I highlight all these I copy it I click on estimates I pasted under methods I change it to this quest remember I said that arrow DL is been estimated by the OLS technique so this is the entire equation so here I begin to remove those coefficients that are indeed equals so one of such is the second lag of the logo of imports I remove that that's a C to be careful to know what you are removing if you are not sure you can easily confirm from what you are violet a to appear again I'm going to remove C six six seven and see it so I'm removing c6 c7 and c8 these are the first second and third lags of logo monitoring outputs so here they are I'm removing c11 c11 is the first leg of real exchange rate having removed those variables I'm clicking okay but what you are seeing on the screen now is the results of the pass ammonius moodle it is now a parsimonious model because I have removed those variables whose coefficients are equal to zero consider this looks better all of them are statistically relevant and significance except the level of real exchange rate which is still a 0.422 so I think I'm okay with this model and I will just leave it the way it is all these coefficients are significant except the level of real exchange rates this is okay so now let us test using the likelihood ratio test I click on quick estimate equation and these are the equations again I'm starting all over I change this real deal again I don't change anything except here we are modified to model 3 and I click OK you can see the outputs of the ova parameterized moodle here we want to conduct a likelihood ratio test to test for the redundancy of variables in this model as you can see just like I explained earlier on each of them are statistically not significant so we need to test their relevance in the module for the likelihood ratio test and what decision is based on the outcome of the adjusted r-square in this case the a squad for the over parameterize model is 0.66 to 5 so half talk on docked in the lateral ratio test I have to compare the outcome of the past - model with this object medicine over parameterize model so let's go to view coefficient Diagnostics I select redundant variables test the busan here make sure you copy them correctly so that you don't begin to test a wrong variable I click OK the result from the redundant variables test as you can see on your screen our F value look at the profile of the F statistic is exactly what we obtained under the wall test let's scroll down a bit you can also see the outcome of the past - Moodle is exactly what we obtained before under the wall test where all the variables are statistically significant except the level of the real exchange rate but notice under the likelihood ratio test it is the adjusted r-square that informs our decision and what is the null hypothesis the null hypothesis is that the variables are redundant against the alternative that they are not redundant and you can only reject the null hypothesis if the value of the adjusted r-squared from the past - Moodle is lower than the over parameterize model so if the value of their Djoser a square is not lower then you cannot reject the null in this case the value of the adjusted r-square is higher than what we have in the over parameterize model in this case we cannot reject the null hypothesis that those variables are indeed redundant so at the end of the day we are going to settle for this passive - model this is much better than what we had before so if you have a situation whereby you did a regression and most of your variables and not status kind of significance begin to subject them to either the war test or the likelihood ratio test whereby you test for the significance of the coefficients or the relevance of the variables don't just start moving the variables arbitrarily from the model because there are no significance you have to test them scientifically I hope this tutorial has been helpful thank you for stopping by subscribe for more videos from crunch econometrics leave us your comments and suggestions on how to improve the quality of our tutorials

Channel: CrunchEconometrix

Views: 19,026

Rating: 4.8435755 out of 5

Keywords: how to correct for parsimony eviews, how to correct for parsimony in eviews, how to correct for parsimony stata, how to correct over-parameterised model eviews, how to correct over-parameterised model

Id: zh3RZvRR7Fk

Length: 11min 25sec (685 seconds)

Published: Mon Mar 26 2018