Panel Logit

[esperanza]

Regards to every member of the forum,

I would be very grateful if someone advises me on how to run a panel logit with EViews5,
I am quite familiar with simple logit, but now I need to run a regression on a panel of six Asian countries in order to create an Early Warning System (EWS) for currency crises prediction. I have monthly data for about 5 years, all of the countries in the sample were actually in crises since 1997, July. As for explanatory variable X, I will take 12-month standart deviation of exchange rate and/or annual change in foreign exchange reserves. As for Y, I am slightly confused, probably merely Y=1 for all months during crisis years (1997-1999), Y=0 in all other cases will be assumed.

The difficulty is that I am confused on how to prepare the data, import and run a regression properly.
The purpose of the regression is similar to the one used by Lestano J. in the attached research paper.

Thank you for any useful help.

[esperanza]

please, have a look at the data attached, I am still confused in how to prepare/import it to EViews5,
this urgent, I have deadline next week...

thank you in advance.

[esperanza]

have anyone ideas on how to run logit on the panel of 7 countries for 1995:1m-2009:12m? in the file attached above?

[startz]

What happens when you read your data into EViews and use the logit command?

[esperanza]

Startz, thank you for the reply! When I just try to open my series as an 'Equation', I recieve 'Alpha series in specification [var name] @EXPAND can convert to catagorical dummies' message. This happens in the very beginning so that I cannot even choose the type of regression to be run 'logit'.

Here is the file:

[startz]

Try just entering at the command line

Code: Select all

logit y x1 x2 @expand(categoryVariable)

using your real variables of course.

[esperanza]

should it be as follows:

logit yindo x1indo @expand(x1indo)

it does not help, neither does: logit yindo x1indo @expand(categoryVariable), as you have suggested

[startz]

You have two separate problems. The small one is that your variables are stored as alphas rather than as the usual numeric series. To get started, try

Code: Select all

logit yindo c @val(x1indo)

The larger problem is that you probably want all your countries listed in one long variable with a separate variable indicating the country. One manual way to do this is to create a new undated workfile with 7 times 180 observations. Create a new variable x1 and copy and paste each country in order. Same thing for any other variables you have. Then create a new variable countryName that has a unique identifier for each country. All this will be a pain, but probably won't take more than an hour.

[esperanza]

Thank you, startz, for the constructive advice. As I have understood from your post, it is only possible to estimate logit for each separate country for time series data, but no panel logit is available? Another thing I wish to understand is what 'alphas' are. This is the first time I encounter this concept.

[startz]

Thank you, startz, for the constructive advice. As I have understood from your post, it is only possible to estimate logit for each separate country for time series data, but no panel logit is available? Another thing I wish to understand is what 'alphas' are. This is the first time I encounter this concept.

No, you can do a panel logit. It's just not built into EViews as an automatic procedure so you're going to have to set it up manually.

[esperanza]

Is it a long procedure to set it up manually? what should be done for it, could you please specify.

[startz]

The larger problem is that you probably want all your countries listed in one long variable with a separate variable indicating the country. One manual way to do this is to create a new undated workfile with 7 times 180 observations. Create a new variable x1 and copy and paste each country in order. Same thing for any other variables you have. Then create a new variable countryName that has a unique identifier for each country. All this will be a pain, but probably won't take more than an hour.

[esperanza]

startz, thank you. As I understood, I should place all countries in a row for each variable, so that I get single 'y' series and separate long time series for each x. now, I think everything is clear, thank you once again for your support, startz!