For my thesis, I am using panel data with stock returns and other firm data. I first used an event study to calculate abnormal returns (with event window of 7 days so 7 observations for 500 firms) which I want to use as a dependent variable in the analysis after.
Specifically, I want to look at the impact of an event on firm valuation, but differentiating between sectors (see which sectors did better than others). My first approach was to make a firm fixed effects model with clustered standard errors as follows:
abnormalreturns_it = constant_it + SectorDummy1_i + SectorDummy2_i + error_it
Problem: my sector dummies get dropped due to collinearity as a result of the fixed effects (since they do not vary over time). This means I cannot get a coefficient per sector. I also thought about random effects. However, Hausman test says that fixed effect is the only possible option between the two.
I did try a suggestion saying that one could include an event dummy variable (0 before the event and 1 after) and interact it with the sector dummies. So now, while my industry dummies still get dropped, the interaction term stays. However, I am not sure if the coefficients and t-stats from this are still valid? This is the current model:
xtreg AR MV TTF ENSCORE i.TR2##i.WAR, fe vce(cluster c_id)
where AR is abnormal returns and i.TR2##i.WAR is an interaction between industry classification and the event dummy
Other options i heard of:
- a cross-sectional regression is possible, except that I would have to drop variables that change over time.
- mixed effects model but i have yet to look into that.
I appreciate any help. Thank you in advance.
