# Which model to use when one variable is country specific and the other is company specific?

So basically my study is on a single country with a 10 year time period. The sample is of several companies of that country. My independent variables are of two kinds, one is country-specific macroeconomic and the others are company-specific financial. The dependent variable is also company specific. I am trying to find the joint effect of the two independent variable on the dependent.

The problem here is that since the country is same, hence for a single year all the companies within the sample will have same values for the macroeconomic variables. I don't think time series will work since I am not suggesting that current period y depends on past period y. But I am also not sure whether this is a panel data as I am focusing on a single country.

Which model should I use? Should I use interaction term?

• So you have something like $y_{ij}=\beta_0 + \beta_{1i} x_i + \beta_{2j} z_j$ where $i$ indexes the country and $j$ the company ? – user83346 Sep 27 '17 at 9:22
• No. Because the i index for Y variable does not change as the entire sample is from the same country. This means that value of x will also change only if the time period changes. – Adnan Sep 27 '17 at 12:09
• Also since I wish to find the combined effect, I was wonder if an interaction term could be included or not – Adnan Sep 27 '17 at 12:17
• Could you write down the formule that describes your model? – user83346 Sep 27 '17 at 13:45

## 1 Answer

A standard approach for this is a hierarchical model, in which there is a random company effect so that the years are correlated within company. This is presumably what you meant when you referred to panel data. Alternatively, one could of course treat the records as independent, but that is typically an untenable assumption that leads to inappropriately overconfident conclusions. A random company effect improves a lot over this, but assumes all years are equally correlated for the same company. Time series methods or more sophisticated random effects structures (e.g. allowing for a different correlation over time) relax this assumption and are even more likely to correctly reflect the real correlations in the data.

None of these things get you around the fact that year and macroeconomic variables are mixed up. So, if that variable is of interest, it may be tricky to say much about it. 10 years may well be too short to give you enough variation that is not mixed up with specific years. Other countries (ideally very different ones) with different macroeconomic situations one be one way to try to deal with that, but may not help (availability of data, some trends may be the same globally etc.).

• Thanks. This actually makes a lot of sense. I will try and and read up on hierarchical models. You may be correct that 10 years can be too short to find any visible change in the variable. But I am trying to make a country-specific event study. I want to see whether there were changes in the variables before during and after the event and whether this change was the cause for the event to happen. The hence longer time period and different countries cannot be used. – Adnan Sep 27 '17 at 12:14
• Often it is really nice if you have the same change happening in different countries/states of a country at a different time (and in some not at all) and possibly even at different intensities (e.g. introduction of a minimum wage and changes to its level). – Björn Sep 27 '17 at 14:34