Standard ARDL assumes Linearity whereas NARDL assumes non-linearity so the former permits the effects of the variables to be same. For example a 1% increase in X has the same 1% decrease in X. The latter is asymmetric in the sense that the partial sums of positive and negatives changes that introduced asymmetry in the specification are unbalanced. Positive can have a different effect on the dependent variable from that of the negative.

L: Y = 1 X

N-Linear Y=1 X+ X-

@Ahmadi,

i thought that you answer is already mentioned in my write up. What do X+ and X- indicate?

@Ahmadi,

You have earned my respect. Honest people tend to achieve a lot. I’m sure you will

/Dear Arize, if the former(ARDL) which assumes linearity permits effects of the variables to be same, what form does the latter(NARDL) assume and or permit on relevant variables?

The very construction of NARDL permits to incorporate the possibility of asymmetric effects of positive and negative changes in explanatory variables on the dependent variable, unlike the case of ARDL, wherein the possible impact of explanatory variables remains same. however, in case the impact of segregated components of an explanatory variable is found to be same, then NARDL model boils down to the standard symmetric ARDL model. in addition, NARDL method provides graphs of cumulative dynamic multipliers used to trace out the adjustment patterns following the positive and negative shocks to explanatory variables. More importantly and interestingly the model is simple and comprehensive enough to permit any asymmetry switching from short-run to long-run or vice versa.

Good points Javet!

Thank you sir @Chuck A Arize

Javed Bhat can you explain more or give a resource about the difference between standard symmetric ARDL and NARDL. In the meantime, what is the difference between asymmetric adjustment towards equilibrium and dynamic multiplier? thanks sir.

For a detailed discussion kindly refer to the original paper.

https://www.google.com/url?sa=t&source=web&rct=j&url=https://papers.ssrn.com/sol3/Delivery.cfm%3Fabstractid%3D1807745&ved=2ahUKEwjCgIjd6OXfAhUBU30KHe9sCAUQFjAJegQIBRAB&usg=AOvVaw27XZwLOf6KHRXaM5P5NzpB&cshid=1547212797670

I agree with Mr. Mahmudi Murjani. The NARDL allows to model non-linear impact of independent variables on the dependent variable. Some times due to presences of assymetric impact, the usual ARDL may not able to capture the same and bound test may show absence of co-integration.

Rajesh Kavediya Javed Bhat excellent point!

The main difference is what is given by Javed Bhat.

Agreed with the explanation of Javed Bhat.

Does the above discussion point to the fact that NARDL supercedes ARDL and should be used invariably instead of Ardl?

i agree with javed bhat explanation for NARDL model.

Yes NARDL captures the model extensively than ARDL does

The ARDL cointegration technique is used in determining the long run relationship between series with different order of integration (Pesaran and Shin, 1999, and Pesaran et al. 2001). The reparameterized result gives the short-run dynamics and long run relationship of the considered variables while Specifically, the NARDL model is a single-equation error correction model that can accommodate asymmetry in the long-run equilibrium relationship and/or the short-run dynamic coefficients via the use of partial sum decompositions of the independent variable(s).