It depends on how the shock influenced the treated and control unit.

If you can argue that the shock influenced all units in the same way, then the inclusion of two way fixed effects (particularly time fixed effects in this case) can help you to rule out the effects of such 'common' shock.

However, if there are reasons to think that the structural shock differently affected treated and control units, then this could result in potential biases in the estimation and identification of the average treatment effect.

There is a novel literature dealing with the correct identification of such effects. I would suggest you reading the following paper:

"VISUALIZATION, IDENTIFICATION, AND ESTIMATION IN THE LINEAR PANEL EVENT-STUDY DESIGN" from Simon Freyaldenhoven, Christian Hansen, Jorge Pérez Pérez and Jesse M. Shapiro. You can find the paper here: http://www.nber.org/papers/w29170

I particularly suggest you to read section 3.1. of the paper as long as accompanying the paper there is the Stata command "xtevent" which is particularly useful to deal with such econometric analyses.

I entirely concur with Francesco Luigi Milone - but I have a 2 questions to aid in providing informed suggestions:

1). Where are these 4 years (in which a structural break occurred prior to policy intervention) located in the entire sample period - in the early years or middle or towards the end - of the sample period?

2). Was the the policy intervention implemented at the middle or towards the end of your sample period? - probably a breakdown in percentiles of the sample sub-period lengths can do.

Sincere apologies for asking such questions - I suppose answers to them can provide valuable information in assessing the nature and magnitude to which the shock influence(d) the treatment and control groups in question.

Thank you so much Dr. Elvis Munyaradzi Ganyaupfu , I attached the fig. here given below....please suggest what need to do regarding structural break.

Hi Md Azizur Rahman

Structural breaks in series have key implications for suitable model specification and estimation. From the trend graph you shared, it looks like/suggests that there was a single structural break in the series.

The traditional approach you can implement in such an instance can be to account for such a structural break in testing for unit roots in the series by incorporating a dummy variable to modify the standard Dickey-Fuller (DF) test to allow for changes in both level and trend. For instance, you can perform such a test using the Kwiatkowski, Phillips, Schmidt & Shin (1992) and Zivot-Andrews unit root test (1992) which allows merely one structural break in the data generation process.

However, please note that use of dummy variable assumes the presence of 1 or 2 structural breaks in the level or trend of the given variable – hence the main setback linked with a dummy variable is that it detects only sharp breaks, while it does not allow for smooth changes and it assumes the number of breaks but without knowledge of the functional form of the break. Combine with the inability to allow for a smooth structural change, imposing the wrong number of structural breaks (in your case, which seems to be a single (1) break) can result in serious model misspecification since several structural changes/breaks occur slowly rather than sharply.

Given this insight that dummy variables are usually unable to properly model structural breaks, you can elect to avoid this drawback by using unit root testing procedures which use Fourier series approximation(s) as proposed by Becker eta al. (2004), Enders & Lee (2004, 2012), Becker, Enders & Lee (2006), Christopoulos & Leon-Ledesma (2010) and Rodrigues & Taylor (2012) – eg. Fourier-ADF unit root test procedure – please search papers or Journal articles with econometric literature for detail on these tests and consideration thereof. Generally, the Fourier series approximation used allows for structural breaks of unknown number and unknown functional form.

Upon completing unit root tests with evidence of stationarity in your series at 5 percent significance level, then there should be nothing to worry – and you can proceed with econometric estimation using the suitable technique and I assume your results should be unbiased.

All the best!

Thank you so much, Dr. I am using panel data. Is unit root test mandatory in panel data?

Panel unit root tests are necessary when panels (N) are longer than time periods (T) – that is when N > T. In such scenarios, panel unit root tests on the series can be conducted using theHarris-Tzavalis (HT) panel unit root test method at level and/or first difference to identify asymptotic distributions of panels (Hlouskova & Wagner, 2006) – eg in line with the sequential limit theorem, where the number of time periods (T) is observed or regarded as fixed, while the number of all panels in the dataset matrix is assumed to approach infinity (Harris & Tzavalis, 1999). Therefore, if N < T, then yes there is no need or rationale to conduct panel unit root tests.

Obviously, you want to find out, whether and how (economic?) policy affects firms's performance and if this policy influence changes. if there is a structural break. In the first instance, you should not concentrate on estimation methods, but using using common sense or choosing theories (you think are appropriate) to find an appropriate specification of a hypothesis to test. You xhould not make the mistake of many econometricianx just to take a data set and apply an econometric (statistical) method without using your brain. You can use the analysis on the time periods before and after the structural break.

Based on the little I have read I can say no,

Because difference-in-difference model does not all the times takes time dimension. Therefore, structural breaks do not harm its results.

Thank you so much Dr. Zainab. Can you please share the citation?

تستعرض هذه المدونة أبحاثا إنمائية وتناقش –مع القراء والباحثين- فكرة التنمية المركَّبة بوصفها فرضية جديدة ومختلفة.. تحتوي المدونة على منشورات علمية وفكرية؛ حيث تتوزع على (10) كتب، وأكثر من (50) بحثا محكما.