I don't know if I can perform a panel data analysis or dynamic panel data or a GMM if I have only 4 years of data and thousands of individuals (insolvent firms).
USE YEARS AS REPEATED MEASUREMENT: There are 4 years, makes the observations be firms 1 - firm1000 and construct model and tests on that basis for year 1. Then repeat the measurement from model constructed from year 1 in years 2, 3 and 4. in this case, the sample size equals the number of firm. Number of firms = column.
USE FIRMS AS REPEATED MEASUREMENT AND YEARS AS NUMBER OF OBSERVATION: This design will fail for sample sample size where n = 4. Once could not get proper distribution of the data in order to make sense of data pattern since the researcher proposes panel data analysis (looking for general pattern). The repeating observation of 1000 times would not cure sample defect. This may lead to Type I error (insisting of being correct solely on the basis of 1000 repetition of a defective sample).
Do panel in stacked data as shown by Paul in earlier comment. Best of luck.
You are talking about 3 different things: panel structure, method of estimation (GMM) dynamic panel as particular type of panel. For GMM estimation in general see Blundell, Wansbeek,Bond, etc. With 4 periods it is obvious that you cannot take intoaccount dynamic panel and neither GMM for dynamic....
You can run panel data analysis (fixed effects or random effects) with 4 years data and 1000 firms or more observation. If you are considering using panel GMM, I suggest you increase the time period to at least 5 in order to get better results. For example, in difference-GMM, the time period reduces by one period after taking first difference. Moreover, you may not adequately pass the second order serial correlation test with just 4 years data.
Yes, you can perform the panel data analysis, either static (POLS, FE @ RE) or dynamic (Diff GMM @ System GMM) panel data analysis techniques.
The main criterion for choosing between the two alternatives is by looking at the coefficient of the lagged dependent variable. The significance of the lagged dependent variable would indicate the need to go for dynamic model, as it (dynamic model) is more appropriate and useful when the dependent variable depends on its own past realizations (Brañas-Garza et al., 2011). If the Coefficient for the lagged dependent variable is not significant (p-value > 0.05), static model is more appropriate, otherwise dynamic model is to be preferred (p-value
Bogdan-Vasile Ileanu please do you have any reference about the minimal number of waves (or years) so that the data could be used for a dynamic panel data ?
- Badges
- Science topic