Dear Vit Plhal

I suggest the following articles and link, have a look:

1. https://www.jstor.org/stable/2297285

2. https://www.princeton.edu/~otorres/Panel101.pdf

3. https://conference.iza.org/conference_files/SUMS_2013/slides_1_linear_iza.pdf

4. https://stats.stackexchange.com/questions/57240/how-do-i-interpret-the-correlations-of-fixed-effects-in-my-glmer-output

Best Regards

Dr. Fatemeh Khozaei

Analyze again from the beginning.

Using the regression in the Minitab software and the Assistant menu, find Large Residuals and Unusual X values data, and exclude them from the study.

Then take the data back to the panel regression and do the analysis again.

My idea is to have a lot of Large Residuals and Unusual X values data depending on the subject of your study (dividends and share repurchases).

You might check for multicollinearity since it could have adverse effects on your estimated coefficients. If detected, you should deal with multicollinearity sources and then redo your analysis. In this regard, the articles by Kalnins (2018), Lavery et al. (2019), and Mansfield and Helms (1982) can come in handy. For a proper test of serial correlation, you could check out the article by Born and Breitung (2016). Here are the full citations.

Born, B., & Breitung, J. (2016). Testing for serial correlation in fixed-effects panel data models. Econometric Reviews, 35(7), 1290–1316. https://doi.org/10.1080/07474938.2014.976524

Kalnins, A. (2018). Multicollinearity: How common factors cause Type 1 errors in multivariate regression. Strategic Management Journal, 39(8), 2362–2385. https://doi.org/10.1002/smj.2783

Lavery, M. R., Acharya, P., Sivo, S. A., & Xu, L. (2019). Number of predictors and multicollinearity: What are their effects on error and bias in regression? Communications in Statistics - Simulation and Computation, 48(1), 27–38. https://doi.org/10.1080/03610918.2017.1371750

Mansfield, E. R., & Helms, B. P. (1982). Detecting Multicollinearity. The American Statistician, 36(3a), 158–160. https://doi.org/10.1080/00031305.1982.10482818

Good luck,

Correlation might perfectly change when you use fixed-effects. As you know, the results under FE and RE can be totally different.

In terms of correlation, you can compare the correlation between the variables in levels and the variables in changes (or demeaned). If you compute the correlation of the changes in the variable, this should be an expected outcome of the FE model.