"Averaging" many to just 1 coefficient in OLS.

[thomaskm]

Dear forum participants,

My goal is to measure inflation persistence for a set of countries. The set of countries is divided into two sections: 'Discretionary countries' and 'Rule based countries'
I use the following regression specification:

d(inflation)=c+b1d(inflation(-1)).

My problem is: i want to find two values of b1. One for all the discretionary countries, and one for all the rule based countries.
I know that what i can't do is take the average of all the seperate regression coefficients of b1.

I don't want to aggregate the inflation data for the set of countries, as some countries (obviously) have higher inflation than other countries, so that would induce a higher weight in the measurement for high inflation countries compared to low inflation countries.

Is there any other way to find two seperate values of b1? Help would be greatly appreciated!

Kind Regards,

Thomas

[startz]

Create a dummy variable for the two sets of countries. Then add in the dummy and an interaction term.

[thomaskm]

So the dummy would look something like this? d:=1 if discretionary, 0 if rule-based. ?
How do I add in a dummy in Eviews? I thought of this option before but i couldn't find any tutorial anywhere on using dummies in Eviews. I'm quite new to Eviews and time series in general.
Is there any chance you can show me what the procedure is so I know how to do it? The eviews file is attached in my first message, if you want to fill it out there.
Your help is much appreciated.

Thomas

[startz]

So the dummy would look something like this? d:=1 if discretionary, 0 if rule-based. ?
How do I add in a dummy in Eviews? I thought of this option before but i couldn't find any tutorial anywhere on using dummies in Eviews. I'm quite new to Eviews and time series in general.

Thomas

You have it right.

A dummy variable is no different from any other variable. You may have to set up a panel structure since you have both a cross-country and time series dimension.

[thomaskm]

Thanks for you advice. Looks like i got a lot further now.
I stacked the data as the tutorial describes here: http://forums.eviews.com/viewtopic.php?t=74&f=7

Then, I did a regression including the dummy variable on the entire dataset. I used a dummy: d1=1 if Rule based, d1=0 if Discretionary.

The output is in the attachment (check regression: inflation_persistence, i used first difference to achieve stationarity).

Could you please tell me if i did the regression right, or if i'm missing something.

In case i did it right, would the following interpretation of the regression be right?
b1=0.616 for rule-based countries and 0.616-0.0556 for discretionary countries.

Thanks a lot for your help so far,

Thomas

[startz]

Unfortunately, I don't have access to EViews at the moment.

[thomaskm]

I see, here's a screenshot then!

[startz]

You're interpreting it correctly, but I would recommend including the dummy on its own as well as the interaction.

[thomaskm]

Thank you so much for your quick and helpful responses today, you've been a great help!

Kind regards,

Thomas