The ECM estimation is normally based on OLS. Could you elaborate a bit more on OLS method in your context?

ECM estimation is based on OLS, meaning that an OLS function is still relevant , can a pure OLS regression model still give good results that are acceptable in research as a basic econometric technique? Judging from your answer it means OLS is basic and incorporating ECM into it is just for improvement for particular study, can OLS alone give a standard result for particular research even if other estimation methods are not introduced?

Dear Akram, I have explained further . Thank you.

Furthermore, Akram, during one of our internal PG defenses a participant outside department of finance, commented that any econometric study without use of ARDL today is of a low quality. Is this an accepted dictum?

As I understand from your posts, the OLS is a linear regression model with only contemporaneous explanatory variables and without any dynamics (i.e., lags or leads) estimated using OLS in your context. If this is the case, estimation results might suffer from spurious regression issue if variables are non-stationary or not co-integrated. As commonly documented, a spurious or nonsense regression is a regression that gives wrong statistical evidence of a linear relationship between non-stationary variables. If the variables are stationary or co-integrated, there is no spurious regression issue in the OLS (in your context) but it fails to incorporate short- and long-run dynamics. The ARDL is advantageous over the latter approach in the sense that it incorporates dynamics. The ECM, on the other hand, allows one to analyze short- and long-run dynamics which is important in economic systems.

Dear Akram, If I understand your answers well, OLS estimation first trial that does not gie spurious results, is acceptale, but if its results suffer from spurious regression issue due to non-stationay or not-c0-integrated variablesin te model, then we can embark on application of ARDL tests. Not that OLS method should be neglected. I would like us to collaborate. My emailto is:[email protected]. Pls send me mail

Thank you very much, Prof John Frain, for your convincing answers/explanation. If I understand you well, there is nothing wrong with the traditional OLS techniques and that analysis of OLS estimates where ''classical'' assumptions are valid.

GARCH methods/ARDL testscan only come in if OLS estimates give spurious results thereby showing existence of what you called voanalysislatility (variance). Hence we can bypass using the traditional OLS to apply the Engle-Granger estimate of the cointegrating relationship between I(1) variables that is based on a new theory.

• From your analogy, there is no method that is superior to the other, but one can be more appropriate than the other depending on the operating theories and variables! Please I would like you to comment on my response as well as contact me on tel:+234-8037876614: emailto:[email protected]. Thank you.

I have edited my answer to make it a bit clearer and to remove some typing errors. The revise version is below. GARCH methods model volatility. Integration, unit roots and cointegration are part of the recent theory applicable to the analysis of non-stationary variables. Both of these methodologies are applicable when traditional "OLS" is not valid. They provide answers to different questions. If I can clarify some my answer I would prefer to continue the discussion here where we might get an interesting discussion started.

Amended version of previous answer (deleted)

I would tend to use the term more appropriate rather than more sophisticated. The more sophisticated methods isn the question (apart from GARCH) refer to the analysis of non-stationary data. Take, for example, the Engle-Granger estimte of the cointegrating relationship between I(1) variables. This is estimated using OLS techniques but the coefficient estimates do not have the usual "OLS" distributions. Thus is one is to work with these one must have a new theory that is applicable in the case of I(1) variables. This new theory is not appropriate to the analysis of OLS estimates where the "classical" assumptions are valid.

GARCH methodology is used to model persistence in voanalysislatility (variance). This is not possible using traditional OLS.

The judgement that these methods are superior or better is not appropriate. I would simply use the term "horses for courses"

Thank you, John Frain C, for your clearer clarifications/answers and the new answers you have provided. Exchange rate, inflation, interest rates, etc are among variables that often experience volatility or variability. Apart from OLS estimates that exhibit spurious results, how would one know that a study with its dependent and independent variables would need GARCH model and its other techniques like integration, unit root tests, Granger -causality tests and others to carry out its tests and analysis? I would like you to send me most of your past works that cover these theories and their applicability.

Thank you, John, for your educative answers. Variables like exchange rate, inflation, interest rates, etc are prone to variability or volatility. However, apart from these and OLS estimates that result in spurious results, how do we discover models that would use GARCH techniques or OLS Methods?