2

I am doing an event study in R. I want to use a SUR model, as the event windows are overlapping. Here is some code with which you can create the data I use and what I have done so far: 

#data
library("systemfit")
library("plm")
nederland<- read.table("https://pastebin.com/raw.php?i=93qFnEir", sep=";", header=TRUE)
nedpanel<-pdata.frame(nederland, c("id", "t"))
cyprus<- read.table("https://pastebin.com/raw.php?i=93qFnEir", sep=";", header=TRUE)
cyppanel<-pdata.frame(cyprus, c("id", "t"))
#SUR framework
nederland_sur<-systemfit(returns ~ Price + Pre + Event + Post, method = "SUR",data = nedpanel)
cyprus_sur<-systemfit(returns ~ Price + Pre + Event + Post, method = "SUR",data = cyppanel)

Just a quick explanation on the regression equation: I am trying to single out effects on stock returns (dependent variable) at the event date (event dummy=1). Also, there is a dummy on the day before (=pre) and after (=post) the event day. Price is a control variable.

The problem is that the two datasets, cyprus and nederland, overlap. The data set nederland starts on February 1st 2013 and the data set cyprus starts on February 11th 2013. Both go back 87 days for each entity. This implies that the overlap of the two event windows is 77 days.

Therefore, there might be a correlation of the error terms. In order to obtain correct standard errors, I want to estimate these two events simultaneously, using an SUR model.

What I did so far does not seem correct. How can estimate the event in the Netherlands and the event in Cyprus simultaneously instead of estimating one after the other?

Dave13
  • 55
  • 5
  • 1
    This seems like a request for a tutorial or package recommendation, which is off-topic. (Voted to close). Looks like `systemfit` [has a nice vignette, with at least 2 worked examples](https://cran.r-project.org/web/packages/systemfit/vignettes/systemfit.pdf). Have you read that? Seems like a good place to start. – Gregor Thomas Feb 27 '19 at 14:53
  • 1
    If you want to get help from Stack Overflow, make the question more concrete. Share some sample data (maybe just code to simulate sample data). As little data as you can that still illustrates your problem. Show your best attempt, and we can take that concrete example and go from there. – Gregor Thomas Feb 27 '19 at 14:55
  • Thanks for you comment. I will edit this post and try to make it more concrete. I have read the systemfit vignette on this but unfortunately it did not help. – Dave13 Feb 27 '19 at 15:23
  • @Gregor is this still too abstract? – Dave13 Feb 27 '19 at 16:45
  • 1
    Looks **much** better. – Gregor Thomas Feb 27 '19 at 16:51
  • Can you help? Because nobody else can and it is urgent unfortunately – Dave13 Feb 28 '19 at 08:18
  • 1
    Sorry, I've never used `systemfit` or even `plm`. You could post a similar question on Cross Validated (stats.stackexchange) and maybe find people with more expertise in this area. A question there should probably focus more on the approach than the data preparation---but it would be a good place to validate your approach, maybe get some help or examples on data structure, and possibly get ideas for other approaches as well. – Gregor Thomas Feb 28 '19 at 13:55

0 Answers0