Error correction model.

[Yangyang]

Hi

I have tried ECM model, GDP as the dependent variable, and Unemployment rate as the independent model.
However, the residuals is not stationary, actually, it has a clear upward trend.
One suggestion is that include a linear trend in the equilibrium relation to control the error.
How can i add the linear trend in RATS.

Thank you so much!

Yang

[TomDoan]

Hi

I have tried ECM model, GDP as the dependent variable, and Unemployment rate as the independent model.
However, the residuals is not stationary, actually, it has a clear upward trend.
One suggestion is that include a linear trend in the equilibrium relation to control the error.
How can i add the linear trend in RATS.

Thank you so much!

Yang

First of all, are you sure that (log??) GDP and the UR are cointegrated? I'm not sure I've ever seen anyone come to that conclusion. If they aren't then there is no "equilibrium relation" to fix.

[Yangyang]

Hi

Thank you for your explanation!

I did the co-integration test between GDP and Unemployment rate in log.
The error term from the regression shows no stationary, actually there is
a upward trend in the error term.

I took the first difference, and then did the co-integration regression. But my
supervisor suggested me to add a linear trend to control the stationarity of the
error.
The problem is that i can not make sure if the error is stationary or not
after i add the linear trend.
Thank you!

[TomDoan]

Hi

Thank you for your explanation!

I did the co-integration test between GDP and Unemployment rate in log.
The error term from the regression shows no stationary, actually there is
a upward trend in the error term.

I took the first difference, and then did the co-integration regression. But my
supervisor suggested me to add a linear trend to control the stationarity of the
error.
The problem is that i can not make sure if the error is stationary or not
after i add the linear trend.
Thank you!

That certainly sounds like they're not cointegrated. Do you actually have an economic model for what's going on? It sounds like you're treating these models as generic X and Y rather than GDP and unemployment.

[Yangyang]

I use the Error correction model to capture the relationship between GDP and unemployment rate.
they are obviously not co-integrated in log, one way is to run cointegration regression in differences, which i did.
However, my supervisor told me it might be wrong if i run the cointegration regression in difference, he suggested
me to add one linear trend in the model to control the error.
Is that accessible?
Thank you!

[TomDoan]

You might want to read through the thread at

http://www.estima.com/forum/viewtopic.php?f=6&t=1524

Several other people had the same general problem as you --- trying to let the time series behavior of the data lead the analysis rather than the economics.

[Yangyang]

Hello,

Thank you for your kind advice!

Yang

[Yangyang]

Hello

I have a question about the ECM.
I run the co-integration regression between GDP and unemployment rate, the equilibrium error from that is not stationary. Can i still run the error correction model? Experiences told us non-stationary data may lead to spurious regression.

Thank you for your time!

Best Regards!
Yang

[TomDoan]

You can, but it won't really make any sense. If you run a regression with a stationary variable as the dependent variable and a non-stationary variable as an explanatory variable, the coefficient on the non-stationary variable has to converge to zero asymptotically.

Why are you talking about the "equilibrium error", when there is no equilibrium to start?

[Yangyang]

Hello

I did the Engle and Granger test between GDP and unemployment rate, however, it turns out that they are not co-integrated each other. I adopted your idea to run co-integration regression in first difference, but my supervisor asked me to still estimate ECM. He said estimating ECM on non-stationary data will increase the risk of spurious regression, but not necessary. And also when the equilibrium error from the equation is negative, that makes sense. What I believe is every test should based on the formal statistic test.

Thank you for your time!

Best Regards!
Yang