Structuring bilateral data on excel

[LKie]

Well the panel data definition seems clear to me but regarding the structure of my data that is exactly why I started this discussion as I can't find information on the bilateral data structuring.

EMU dummy should capture the effect of introduction of euro on the FDI flows to host country.
EU dummy should capture the effects of joining European Union on the FDI flows to host country.
ERM dummy should capture the effects of joining The European Exchange Rate Mechanism on the FDI flows to host country.

[EViews Gareth]

Well the panel data definition seems clear to me but regarding the structure of my data that is exactly why I started this discussion as I can't find information on the bilateral data structuring.

EMU dummy should capture the effect of introduction of euro on the FDI flows to host country.
EU dummy should capture the effects of joining European Union on the FDI flows to host country.
ERM dummy should capture the effects of joining The European Exchange Rate Mechanism on the FDI flows to host country.

The panel data definition may seem clear to you, but it isn't clear to anyone else. Once you have explained the panel definition, we can explain how to structure the data.

From your examples, it seems that your cross-sections are EMU, EU and ERM?

[EViews Gareth]

To be a bit more clear, perhaps...

In your original Excel file, in Example 1 you have 7 columns:

Code: Select all

Host Source Distance Hose Source Year FDI_Value

I think we can all assume that Year is the time-element of your panel. Which of those columns represents the cross-sections?

[LKie]

The panel data definition may seem clear to you, but it isn't clear to anyone else. Once you have explained the panel definition, we can explain how to structure the data.

Well to me the panel data definition is simply data set from the few different countries (or firms, persons etc.) which varies across time (but for the same few countries). You probably meant something else but I can't really grasp it..

I think we can all assume that Year is the time-element of your panel. Which of those columns represents the cross-sections?

Yes, 'Year' is the time-element. Cross-sections are 'Host_Country' and 'Source_Country' as actual data on distance, FDI, GDP etc. depend on them.

[EViews Gareth]

The issue is that panel estimation requires only one set of cross-sections. If you have both Host and Source as cross-sections, you have a 3-dimensional panel. You need to reduce it down to 2 dimensions.

[LKie]

I have attached excel file with the way I have arranged FDI data as of now. Is it 3-dimensional panel? I wouldn't be able to run a regression with data arranged like that?

[EViews Gareth]

That looks like a text-book case of a 3-D panel.

For panel estimation to work, each observation (row) has to be uniquely identified by 2, and only 2, columns.

In your case none of the rows are uniquely identified by two columns, they're uniquely identified by 3 columns (B,C,D).

From a data structuring point of view you could combine columns B and C into one column which would satisfy the 2-D panel requirement. But whether that is ok to do econometrically/economically is up to you (and is the point I've been trying to make all along).

[LKie]

Ok now I see what you mean.

Economically combining B and C would make sense I guess. My variable of interest is dummy EMUt which is 1 if host country has euro at point of time t and 0 if it doesn't. and the FDI values are for the host country only so the dummy should capture euro effect on the host country's FDI value in the case if B and C are combined.

However, the issue is how I would adjoin variables for GDP which are different for host and source country at any given point of time? For example, combined cross-section BGR_AUT would have two GDP values: 1 for Bulgaria and another for Austria.

[EViews Gareth]

Yep, that's an issue

[startz]

Perhaps you might create a variable "host_gdp" which always has the host country's gdp no matter who is the source country.

[LKie]

EViews Gareth,

Why exactly 3-D panel can't be estimated? It seems that this paper https://ideas.repec.org/p/ces/ceswps/_2123.html did it using panel least squares of Eviews 5.0. (Section 4.3).

Startz,

Thanks for suggestion, I will look into it.

[EViews Gareth]

They didn't use random effects, and I'm guessing they manually created dummies for fixed effects.