Hello,
I am new to this forum and i am currently writing a MSc thesis for finance.
I am researching the effect of CEO overconfidence on the effect of seasoned equity announcements (SEO) and short-term stock returns.
My regression model looks like this.
Dependent variable: Cumulative abnormal returns (CAR) post- SEO announcement
Independent variable: Overconfidence dummy
control variables; size, book-to-market, leverage, return on assets, issue size, underpricing, ceo age, ceo gender (all continuous, except for gender) (size is expressed in logarithm)
N: 349 observations
I have computed the CARs for firms having announced SEOs during 2010-2020, therefore my data is cross sectional.
My thesis supervisor suggested i maybe could add firm/industry and/or year fixed effects.
Therefore, i have the following regression command: reg CAR Overconfidence MktCap BtM Lev Roa IssSize UndPr Age Male i.Industry i.Year, vce(robust)
However, this command returns very small coefficients (max. 0.05) and all of the variables including the constant are statistically insignificant, overall regression significance p>F returns a dot.
I have tried experimenting with the regression, removing year fixed effects and/or industry fixed effects and removing the robust standard errors.
The most significance i get is when i add year fixed effects and robust s.e., 3/8 variables become significant (Overconifdence, IssSize, UndPr), the overall regression significance p>F becomes 0.000.
How could removing year fixed effects be explained or interpreted?
And also, as only 3 variables are significant with very extremely low coefficient values (max. 0.04), how should the main results of my research then be interpreted?
Regarding this, do you have any tips as how to increase significance, or how to check if the regression is performed correctly? Could there maybe be issues with my data?
Looking forward to your answers
Kind regards,
Darya
I am new to this forum and i am currently writing a MSc thesis for finance.
I am researching the effect of CEO overconfidence on the effect of seasoned equity announcements (SEO) and short-term stock returns.
My regression model looks like this.
Dependent variable: Cumulative abnormal returns (CAR) post- SEO announcement
Independent variable: Overconfidence dummy
control variables; size, book-to-market, leverage, return on assets, issue size, underpricing, ceo age, ceo gender (all continuous, except for gender) (size is expressed in logarithm)
N: 349 observations
I have computed the CARs for firms having announced SEOs during 2010-2020, therefore my data is cross sectional.
My thesis supervisor suggested i maybe could add firm/industry and/or year fixed effects.
Therefore, i have the following regression command: reg CAR Overconfidence MktCap BtM Lev Roa IssSize UndPr Age Male i.Industry i.Year, vce(robust)
However, this command returns very small coefficients (max. 0.05) and all of the variables including the constant are statistically insignificant, overall regression significance p>F returns a dot.
I have tried experimenting with the regression, removing year fixed effects and/or industry fixed effects and removing the robust standard errors.
The most significance i get is when i add year fixed effects and robust s.e., 3/8 variables become significant (Overconifdence, IssSize, UndPr), the overall regression significance p>F becomes 0.000.
How could removing year fixed effects be explained or interpreted?
And also, as only 3 variables are significant with very extremely low coefficient values (max. 0.04), how should the main results of my research then be interpreted?
Regarding this, do you have any tips as how to increase significance, or how to check if the regression is performed correctly? Could there maybe be issues with my data?
Looking forward to your answers
Kind regards,
Darya
Comment