Panel data set - Financial Crisis and Time Fixed Effects

[Vlugge Jaapie]

Dear all,

First of all, thank you for asking and answering on this forum. It already helped me a lot.

At this very moment I am working on earnings management around IPOs before and during the financial crisis. I have data (2003-2010) of for IPO companies two quarters before, and four quarters after the IPO. I want to do research whether earnings management during the financial crisis is higher or lower or equal.

In my multple regression I add a dummy variable for crisis (years 2007, 2008, 2009 and 2010). But now my prof. says that he misses time fixed effects in my regression. But I always thought that they were not needed because I look at the crisis.

Should I include (besides my crisis dummy) also time fixed effects in my model?

Thanks in advance.

Vlugge Jaapie

[Clyde Schechter]

Whether your model requires time fixed effects is not a statistical question. It is a substantive question in finance. Time fixed effects eliminate extraneous outcome variation that is attributable to year-by-year random shocks in the data generating process. If those shocks are appreciable, including them improves the efficiency of your analysis. If those shocks are also associated with your predictors of interest, then omitting them could also lead to omitted variable bias. So in either of these circumstances, they would be needed. Assuming your professor knows something about finance, I would defer to his advice on this.

Do be aware, though, that when you include time fixed effects in a model that has a crisis variable, in addition to the usual one reference year being omitted due to colinearity with the constant term, a second reference year will also be omitted due to colinearity with the crisis variable. This is normal, and is in no way a problem.

[Carlo Lazzaro]

Vlugge:
welcome to the list.
As an aside to Clyde's helpful remarks:
- as per FAQ (that everybody is requested to read before her/his post), please post what you typed and what Stata gave you back. This approach outperforms tons of words aiming at describing (often unsuccessfuly) what's the matter with poster's analysis;
- if a theoretical justifications calls for -fe- specification, please note that it will get rid of time-invariant predictors, which might be of your interest (eg: industries, which are not assumed to change across time).