Interpreting results form dummy variable lag/lead

[Zorodzai Murinye]

I am analysing the effects of earthquakes on economic growth, for a panel dataset including 180 countries over the 1901-2010 period. For now I have created dummy variables as event indicator: value 1 for every year each country received an earthquake. To assess impact I have done a 10year lag/lead as my event window (that is 10 years before earthquake and 10 years afterwards). While I see a significant decline in economic growth when the strike happens, I also see an upward trend before and after the strike. How can I remove this upward time trend (especially before the strike). Please note when I did my regression, I did control for year and country FE.

Also, I would like to control for size of earthquake. How can I include these interactions on my dummy variables?

[Clyde Schechter]

Your graph is unclear because it doesn't actually say what is on the two axes. But I'll guess that the horizontal axis is years (with earthquake year being zero) and the vertical axis is economic level (not growth): that seems to make the most sense. You can model this easily enough by having two variables:

time_trend, ranges from -10 to + 10 and represents the number of years before or after the earthquake, the earthquake year itself having value zero. (i.e. time_trend = actual_year - year_of_earthquake.)

pre_post indicator: 0 before earthquake, 1 the year of the earthquake itself and all years thereafter.

Then you would regress economic level (or whatever it is that the vertical axis of the graph you show represents) on c.time_trend##i.pre_post. The coefficient of pre_post will represent the immediate shock effect (no pun intended) of the earthquake on economic level. Then run -margins pre_post, dydx(time)- to get the rates of growth of economic level over time before and after the earthquake. From the appearance of your graph, it looks as if including time fixed effects is not necessary here, and even gets in the way of seeing what is going on because they pick up much of the time trend, which then appears as the sequence of coefficients of the time fixed effects. (If there are particular years where there is a major economic shock, say 2008, you might include indicators for specifically those years, but not for every year.)