Dear Behzad Jawan Bakhsh

You have a micro-panel, which is quite good point of departure. You have basically a Differences-in-Differences framework.

You can use the following specification:

Y_it = intercept + treatment + time_fixed_effects + firm_fixed_effects + other_time-varying_controls + disturbance

I would just include time fixed effects through a year dummy (e.g., year2009 = 0 if year == 2007 and year2009 == 1 if year == 2009) and you can deal with the firm fixed effects using the within-group estimator (for instance). You shold cluster the standard errors at the firm level. Treatment is a variable that takes the value 1 for treated firms when they are treated and 0 othewrise.

In order to be able to estimate the causal effect of the crisis, you should have both untreated and treated firms in the second year. Otherwise, if all your firms in 2009 are treated (and all firms in 2007 are untreated) you cannot separte the "financial crisis effect" from the "year effect".

If that is the case, you can just estimate the effect of "year 2009" (which you can call financial crisis) as follows:

Y_it = intercept + year2009 + firm_fixed_effects + other_time-variying_controls + disturbance

Use clustered standard errors at the firm level again.

You can find more information about these models and how to implement them in Stata in the 5th chaper of Mostly harmless Econometrics (Angrist & Pischke, 2008).

https://www.stata.com/bookstore/mostly-harmless-econometrics/

I hope it helps. If it is useful, mark your question as solved and recommend the answer, please.

José-Ignacio

You may use standard firm fixed effects with time dummy for year 2009 or take a first difference of all variables and estimate the first differenced model without time dummy. In the second approach you will be estimating a cross-sectional regression equation where the variables are difference between t and t-1 hence, you need not bother about time invariant unobservable factors affecting a firm.

the pooled regression model

absolutely not!! T=2 is TOOOO short to use panel data models. I agree with Shreya Biswas, you can use a time dummy variable in a classic model, or you can even run two models for t=1 and t=2 and compare the two models.

Random effect model.

The command in stata is:

xtset timevar

xtreg y x, re

Can I just take the firms in the treatment group for years 2007 and 2009 and and employ First Difference Model in order to analyze the impact of financial crisis on R&D intensity?

Dear Behzad,

I will also suggest you use Standard fixed effect bicause your T is not sufficient.

Yes, you can take the forms in the treatment grouio for 2007 and 2009 and apply first difference model to analyze the impact of financial crisis on R & D intensity.

It's a research, you can approach it in various ways.

All the best.