Fixed effects model with control group

[Noella Bergmans]

Dear all,

I am examining the impact of merger activities on the costs of firms. The purpose is the check whether firms realised cost savings due to mergers over the years 2000 - 2014. I have a control group, which consists of firms which were never involved in acquisitions (either as target or acquirer) and a target group, which consists of firms which participated in merger activities. I want to compare the cost trends of both groups during the specified period. I am planning to use a fixed effects model with firm and time fixed effects. I will use a dummy 'target' which indicates the year in which a firm merged and a dummy 'target_post" for all years after the merger. However, I am a bit confused on how to include the control group in the analysis. I was thinking about using the xtreg command with the fe option while clustering on id number (unique identifier for each firm), but I am not sure if State then includes the control group in the main analysis?

I would appreciate it very much if someone could help me out with this!
Kind regards,
Noella

[Carlo Lazzaro]

Noella:
1) -fe- estimator, in addition to wipe out all time-invariant predictors, is useful to investigate change of the regressand within the same panel as times goes by. Are you sure that this is waht you have in mind? Moreover, with -xtreg,fe- the only fixed effect you can estimate is the one included in -xtset-:

xtset firm year

. With -i.year- you simply add a categorical variabel aimed at investigating the contribution of years to variation in the regressand
2) I'm not clear why you want to start with non-default standard errors. Did you suspect heteroskedasticity and/or autocorrelation in your data (and, by the way, how many panels do you have in your dataset)?
3) Stata will omit the control group if collinear with the fixed effect.

[Noella Bergmans]

Dear Carlo,

Thank you for your quick response. I clustered on id since I detected heteroskedasticity after performing a modified Wald test (with xttest3). I have a total of 1043 panels in my dataset. Besides the merger variables and a set of control variables, I also included 4 lagged merger indicators which refer to two years before, three years before and four and more years before athemerger event respectively (one year before was excluded for the dummy trap). I included these to test the exogeneity assumption, i.e. covariates uncorrelated with idiosyncratic error term. However, one of the lagged merger indicators turned out to be significant. How does this affect my results? My results also change depending on which lagged merger indicator is omitted.

Thank you for your response,
Kind regards
Sebastiaan

[Carlo Lazzaro]

Noella/Sebastiaan: please note the preference on this forum for real given and family name (as peFAQ) and, I would add, for real posters. Thanks.
While your main question is virtually impossible to reply without taking a look at what you typed and what Stata gave you back, you can easily managed the reference category in your categorical variable via -fvvarlist- notation.