Hey,
I am conducting an event study with a short event window (lets say 3 or 5 days). I have a lot of variables, but most of them are fixed over that short period of time (GDP per capita, GDP growth etc) because they are not reported more frequently. This leaves me with abnormal returns that differ during those five days, but the independent variables are mostly the same in that period. Regular regressions will therefore result in a coefficient of zero. My panel dataset contains over 20 countries with abnormal returns of 3 to 5 days, and about 20 independent variables, which are mostly fixed. I have seen in other papers that it is possible to run this type of regression, but I cannot seem to figure out which type and how I should structure my data differently. So far tutorials, manuals and methodology sections of papers have given me no result. Could someone point me in the right direction?
Thanks in advance
I do not completely understand what you are trying to do with your data. However, in our recent WBER paper, we do an event student with panel data. The process is described in the paper: https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhaa022/5975048?guestAccessKey=8c4574df-be1a-439a-b18d-2f349d996f01
If within variation is very low, perhaps is worthy to try random effects panel techniques... Both survival analysis and fixed-effects panel data might not be the most appropiate tools in this framework.
Jesse A Matheson Thank you for your response. I have read your paper and what I want to do is pretty much the same. Except my dependent variable is the abnormal return (which covers five days). This is where my problem lies. Let me use population density as an example since I that is in your paper and I want to use that variable as well.
Population density is measured every 6 months or year, so you have data from 1997 - 2009 (about 13 - 26 data points I guess?). My event study only covers five days, so that means I only have one data point for population density. Hence, a normal regression would result in a coefficient of zero.
So I have this problem with GDP growth, GDP per capita etc. as well. Does this clear my problem up?
Jesse A Matheson As an example I've added some fictional data. But my dataset kind of looks like this.
Erik van D Hi Erik. Looking at this, there are a couple of things that you can do. I assume that CAR in the data set is your outcome, the variable that you are interested in for your event study will be a dummy for each period before and after the event. So if the event is in day 3, you will have dummies for -2, -1, 1, 2 (excluding 0).
You have two choices with the rest of the variables. The first is to include them in your regression and run a random-effects specification (as suggested by José-Ignacio Antón). The second choice is to forget these variables and just include country-specific dummy variables. This is what I would recommend, I see no advantage to using variables that have no within-variation.
An event study only looks at how an outcome for a specific country changes over time, it does not make use of differences between countries. Therefore, variables that do not change over the time period you are interested in are not much use; country dummies explain everything they do.
Jesse A Matheson Thank you for your response, Jesse. I am ready to run the random-effects, but I am interested in your other option. You mention forgetting these variables and including country-specific dummy variables only, would that be a probit model? In this case, when would a variable (lets say GDP per capita) get a 1 or 0?
Thanks for your help!
- Badges
- Science topic
- Similar topics
- Biological Science
- Histology
- Fixatives