Can anyone suggest a lecture/paper/textbook that covers an event study (e.g., exogenous policy change) using count time-series (or panel) data? Or alternatively, just a general guideline as to what approaches I can take.
$\begingroup$
$\endgroup$
6
-
$\begingroup$ Could you please give us some more information about the real problem you want to solve? $\endgroup$– kjetil b halvorsen ♦Commented Jan 24, 2015 at 21:50
-
$\begingroup$ Thank you for your interest @kjetil_b_halvorsen. To simplify: y_(i,t,) is an outcome variable (counts truncated at 0) and at some time T there is a known exogenous shock that impacts y. As a first step I just want to run a simple NB regression with dummies for levels and slopes before delving more deeply into the problem. $\endgroup$– econstatCommented Jan 26, 2015 at 18:37
-
$\begingroup$ Maybe you could have a look at the R strucchange package (and its documentation, which would have relevant references). $\endgroup$– kjetil b halvorsen ♦Commented Jan 26, 2015 at 18:56
-
$\begingroup$ I have actually. Data set is too large for R to handle. Tried everything I could to get around that problem also to no avail. Stata also has xtnbreg. I had to stop it after let it run for 2 days. The data is daily over 4 years with thousands of panels. That's why I was hoping for other possibilities. $\endgroup$– econstatCommented Jan 26, 2015 at 19:14
-
$\begingroup$ Secondly, I also have the problem of daily patterns with weekends the counts being really low and including dummies doesn't seem to alleviate that problem much. $\endgroup$– econstatCommented Jan 26, 2015 at 19:15
|
Show 1 more comment
1 Answer
$\begingroup$
$\endgroup$
Check out the following paper which may be helpful to understand about event studies with panel data. https://www.aeaweb.org/articles?id=10.1257/aer.104.10.3038 In general what you are trying to do in an event study design is to understand whether the "event" of interest brings about a sharp change in your outcomes of interest. A very simple way to model this is: $$Y_{it} = \alpha_{i} + \beta_{t} + \sum\limits_{\tau=-k}^k \beta_{\tau} D_{\tau,it} + \varepsilon_{it}$$ Here the i is unit and t is say the time (like year). $\tau$ refers to the years before or after the event takes place. $\beta_{\tau}$ gives the effect of the event $\tau$ years before/after it takes place depending on negative or positive values of $\tau$. You can add unit specific time trends to get rid of confounding trends that will impactyour outcome as well. You would essentially like to ensure that there is nothing else changing at the time that event takes place. Hope this helps! Please ask more questions if not clear.
