[Emna Trabelsi]

Dear Eric,

Thank you for having clarified the issue of the presence of a unit root in the case of a binary variable. I will be grateful to you if you suggest some references (i.e., article) discussing the introduction of dummies in a linear regression as break points.

Best,

Emna

[Joro Kolev]

Hi Emna,
1) muticollinearity is not a problem--it does not give rise to biased results or incorrect standard errors--and when it is a problem (you are not able to estimate separately an effect you are interested in) the only solution is collecting more data.
2) omitted variables check out the Ramsey RESET test: " -estat ovtest- performs two versions of the Ramsey (1969) regression specification-error test (RESET) for omitted variables. This test amounts to
fitting y=xb+zt+u and then testing t=0. If the rhs option is not specified, powers of the fitted values are used for z. If rhs is specified,
powers of the individual elements of x are used.

"

[Eric de Souza]

Answering #16: I don't have a reference at hand, but you just need to write down the equation to see it. If d is the step dummy variable, and the equation is:E( y | x) = a + c.d + b.x , then d will move the expected value of y given x up (down) when d =1 (0)

[Emna Trabelsi]

Dear Joro,

I would like to ask about conducting an ardl model instead of OLS regression with Newey-West or Prais standard errors. I know that ardl model does not require a stationarity pretest. How to control for serial correlation and heteroscesticity in that case?
Thanks in advance for your reply.

Best,
Emna

[Joro Kolev]

Hi Emna,

In my paper that I love to cite so much (Kolev, G. I. (2011). The" spurious regression problem" in the classical regression model framework. Economics Bulletin, 31(1), 925-937.)
I do observe that including lags as in eq(10) which is some form of autoregressive distributed lag model (simulation results showing correctly sized tests, last column of Table I) resolved the spurious regression problem. I am attributing this equation to Durbin, J. (1970). Testing for serial correlation in least squares regressions when some of the regressors
are lagged dependent variables. Econometrica, 38, pp. 410–421.

All in all you are probably right that using the user written -ardl- will resolve the spurious regression problem, and you do not need to pre-test for unit roots.

The way how autoregressive distributed lag models work is that you include enough lags so that you do not have autocorrelation in the error anymore. So the problem reduces to simply choosing the correct lag order, and for what I can see -ardl- has automatic lag selection. So I think you do not need to worry about autocorrelation.

For what I can see, -ardl- does not allow the option Robust. I do not see any reason for this, it is just a linear model so robust standard errors should be appropriate if desired.

But I am not an expert on -ardl-, and I believe that one of the authors of this package is a contributor to Statalist, Sebastian Kripfganz , so he might be able to say something more informed on the topic.

Dear Joro,

I would like to ask about conducting an ardl model instead of OLS regression with Newey-West or Prais standard errors. I know that ardl model does not require a stationarity pretest. How to control for serial correlation and heteroscesticity in that case?
Thanks in advance for your reply.

Best,
Emna

[Sebastian Kripfganz]

I would like to ask about conducting an ardl model instead of OLS regression with Newey-West or Prais standard errors.

An ardl model is estimated by OLS. Newey-West standard errors can be obtained following the procedure on slide 41 of my 2018 London Stata Conference presentation:

• Kripfganz, S., and D. C. Schneider (2018). ardl: Estimating autoregressive distributed lag and equilibrium correction models. Proceedings of the 2018 London Stata Conference.

[Emna Trabelsi]

An ardl model is estimated by OLS. Newey-West standard errors can be obtained following the procedure on slide 41 of my 2018 London Stata Conference presentation:

• Kripfganz, S., and D. C. Schneider (2018). ardl: Estimating autoregressive distributed lag and equilibrium correction models. Proceedings of the 2018 London Stata Conference.

Dear Sebastian,

Many thanks for your kind response. Your presentation is very insightful and gives answers to many questions.

Best,
Emna