Help for Thesis Regression

[Hidde Hartman]

Hello everyone,

I am writing my Master Thesis and doing research on the influence of Independency Indicators of acquiring firm on their short-term stock return before and after an m&a event.
These indicators are noted by categorical values (A, B, C, D) and I have taken them as dummies into my regression. To also research my moderator of total assets I have made interaction terms between these dummies and total assets to look into the effect of this moderating variable. Below is the regression that I ran. Could someone please tell me whether I did the regression correctly for the above indicators and moderators?

Would mean a lot if I could get some help on this.

	(-10,10)	(-5,5)	(-1,1)
VARIABLES	CAR	CAR	CAR
Ownership Concentration	-0.000603	0.000953	0.000292
	(0.00148)	(0.000800)	(0.000500)
Director Independency Indicator A	-	-	-
Director Independency Indicator Amin	0.363***	0.152***	0.0170
	(0.0511)	(0.0546)	(0.0571)
Director Independency Indicator B	-	-	-
Director Independency Indicator Bmin	0.378***	0.272***	0.0186
	(0.0408)	(0.0294)	(0.0191)
Director Independency Indicator D	2.792	1.730	0.354
	(2.237)	(1.343)	(0.283)
Acquirer Total Assets	-0.0619***	-0.0593***	-0.0586***
	(0.0185)	(0.0124)	(0.00582)
Predeal Acquirer Market Cap	-0.00498*	-0.00499**	-0.00670***
	(0.00288)	(0.00238)	(0.00190)
Method of Payment in Bonds	-0.0335	0.0836***	0.0383*
	(0.0279)	(0.0230)	(0.0224)
Method of Payment in Business Assets	-0.0758***	-	-
	(0.00924)
Method of Payment in Cash	0.00324	0.105***	0.0491***
	(0.00637)	(0.00588)	(0.00419)
Method of Payment in Cash Reserves	-0.0131	0.103***	0.0459***
	(0.0135)	(0.00996)	(0.00643)
Method of Payment in Cash Assumed	0.00416	0.109***	0.0570***
	(0.0180)	(0.0164)	(0.0105)
Method of Payment in Converted Debt	-1.107	-0.578	-0.104
	(0.707)	(0.391)	(0.0752)
Method of Payment in Deferred Payment	-0.00435	0.0940***	0.0366***
	(0.0203)	(0.0157)	(0.0106)
Method of Payment in Earn Out	-0.00162	0.0987***	0.0393***
	(0.00800)	(0.00708)	(0.00491)
Method of Payment in Liabilities	-0.00293	0.0999***	0.0485***
	(0.00768)	(0.00696)	(0.00521)
Method of Payment in Other	-0.0527	0.0439	0.0201
	(0.0832)	(0.0499)	(0.0321)
Method of Payment in Services	-	0.0758**	0.00936
		(0.0300)	(0.0275)
Method of Payment in Shares	0.0100	0.109***	0.0430***
	(0.0104)	(0.00875)	(0.00622)
Ownership Concentration x Total Assets	9.38e-05	-7.81e-05	-1.05e-05
	(0.000180)	(9.76e-05)	(6.00e-05)
Indicator A x Total Assets	0.0233***	0.0589***	0.0419***
	(0.00627)	(0.00672)	(0.00352)
Indicator Aplus x Total Assets	0.0486***	0.0635***	0.0586***
	(0.00586)	(0.00765)	(0.00365)
Indicator Amin x Total Assets	-0.0841***	0.0102	0.0567***
	(0.0169)	(0.0143)	(0.0141)
Indicator B x Total Assets	-	0.00540	-
		(0.00965)
Indicator Bplus x Total Assets	0.0497***	0.0649***	0.0585***
	(0.00601)	(0.00766)	(0.00371)
Indicator Bmin x Total Assets	-0.0506***	-	0.0593***
	(0.0132)		(0.00652)
Indicator D x Total Assets	-0.195	-0.0857	0.0279
	(0.203)	(0.123)	(0.0264)
Constant	0.152	-0.106	0.00853
	(0.149)	(0.0767)	(0.0467)
Observations	1,730	1,730	1,730
R-squared	0.122	0.092	0.054

[Daniel Schaefer]

No one here can really tell you if you did a regression "correct" and generally questions like this are discouraged (see the FAQ above). At a minimum, you'll need to provide the line of code used to estimate your model wrapped in code tags (see the # symbol in the editor), and you should just copy the output for your model from Stata (again, wrapped in code tags) rather than use a table like this.

That said, it is interesting that you omit values for indicators A and B. Are these two indicators collinear? It looks like you don't include indicator C at all. Are these actually a single categorical variable that you've dummy encoded (and made indicator C the reference category)? I wonder why the missing values throughout?

[Carlo Lazzaro]

Hidde:
as an aside to Daniel's good point, is your dataset cross-sectional or else?
Or did you perform the very same regression on three different cross-sectional samples of the same size?

[Hidde Hartman]

.

[Hidde Hartman]

No one here can really tell you if you did a regression "correct" and generally questions like this are discouraged (see the FAQ above). At a minimum, you'll need to provide the line of code used to estimate your model wrapped in code tags (see the # symbol in the editor), and you should just copy the output for your model from Stata (again, wrapped in code tags) rather than use a table like this.

That said, it is interesting that you omit values for indicators A and B. Are these two indicators collinear? It looks like you don't include indicator C at all. Are these actually a single categorical variable that you've dummy encoded (and made indicator C the reference category)? I wonder why the missing values throughout?

Thank you Daniel for the clarification on the guidelines of this forum. I am new here so this helps.

The indicators A and B where indeed omitted due to colinearity and the indicators are, like you said, dummy encoded single categorical variables.

[Hidde Hartman]

Hidde:
as an aside to Daniel's good point, is your dataset cross-sectional or else?
Or did you perform the very same regression on three different cross-sectional samples of the same size?

Thanks for responding Carlo. Yes, my dataset is cross-sectional and the same regression was done on the three event windows.

[Carlo Lazzaro]

Hidde:
but if you have three event windows, I assume that your sample, which is composed of different firms (N) was measured three times (T) during the study period.
If N and T do exist in your dataset you're dealing with a panel.

[Daniel Schaefer]

I don't want to distract to much from Carlo's line of inquiry (whether or not you have a time component is important), but I'm not sure you quite understand what I was asking in #2. "Dummy encoding" has a technical meaning: it is how Stata will code a single categorical variable with more than 2 categories behind the scenes - it is not the same as having a set of 0/1 indicators. I thought you might be trying to do something like that by hand. It looks like you have a series of indicator variables instead. Just call indicators "indicators" or "dummy variables". You don't need to say that they are categorical - it just confuses things.

If indicator A and B are perfectly colinear, they provide exactly the same information. You should include one or the other but not both in your model. You still haven't said why you omit indicator C from the results.

Please do provide the line of code and the regression output directly from Stata in a future post. The extra information would probably help clarify a few things.

[Hidde Hartman]

I don't want to distract to much from Carlo's line of inquiry (whether or not you have a time component is important), but I'm not sure you quite understand what I was asking in #2. "Dummy encoding" has a technical meaning: it is how Stata will code a single categorical variable with more than 2 categories behind the scenes - it is not the same as having a set of 0/1 indicators. I thought you might be trying to do something like that by hand. It looks like you have a series of indicator variables instead. Just call indicators "indicators" or "dummy variables". You don't need to say that they are categorical - it just confuses things.

If indicator A and B are perfectly colinear, they provide exactly the same information. You should include one or the other but not both in your model. You still haven't said why you omit indicator C from the results.

Please do provide the line of code and the regression output directly from Stata in a future post. The extra information would probably help clarify a few things.

Thank you Daniel. Both indicator A and B are omitted by STATA due to collinearity, even when I only use one of them. Indicator C is not used because it was not in this dataset. The most important question that I have is whether it is a valid method to use the dummy variables in the interaction terms (IndicatorAxTotalAssets) of the regression? And how to interpret them?


This is the line of code that I used:

regress CAR OwnershipsconcentrationxTotalAssets IndicatorAxTotalAssets IndicatorAplusxTotalAssets IndicatorAminxTotalAssets IndicatorBxTotalAssets IndicatorBplusxTotalAssets IndicatorBminxTotalAssets IndicatorDxTotalAssets Acquiredstake A Amin B Bmin D ln_Acquirortotalassets ln_Predealacquirormarketcap Bonds BusinessAssets Cash CashReserves CashAssumed ConvertedDebt DeferredPayment EarnOut Liabilities Other Services Shares, vce(robust)

[Hidde Hartman]

Hidde:
but if you have three event windows, I assume that your sample, which is composed of different firms (N) was measured three times (T) during the study period.
If N and T do exist in your dataset you're dealing with a panel.

Thank you Carlo but I have aggregated the CARs over the entire event window without breaking them down into specific time periods within that window, and each firm only has one CAR value in the dataset.

[Hoa Ngo]

Please kindly ignore this post and read my second post below. Thank you so much

[Hoa Ngo]

Hi, I am new to this forum so I am sorry if I post in the wrong topic but as I read guidelines in FAQ, I should post in threads with the same issues? So, I decided to post here, if not, I could post in a new topic. My question is also regarding to my thesis. I am doing my master thesis, too. But I do not know whether my understanding is correct or not because I am wondering about the correlation. For my topic, I am examining the impacts of international recognized quality certificates on exporting firms towards energy management measures adoption. and I use probit model ). To choose exporting firms, I use binary numbers (1 = firm has export activities, 0 = otherwise), for owning international recognized quality certificates ( 1 = firm has this certificate, 0 = otherwise), and for energy management measure adoption ( 1 = firm adopts, 0 = otherwise). the description is as below:

#summarize energymanagement employees EX interantionalcertificates
Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
energymana~t | 997 .2577733 .4376281 0 1
employees | 1,028 164.8959 647.6998 3 11742
EX | 1,028 .2412451 .4280469 0 1
interantio~s | 999 .1861862 .3894518 0 1



This is correlation:

#corr employees EX interantionalcertificates

I assume that there is difference in impact on energy management measures adoption between exporting firms and non exporting firms when they own international recognized quality certificates, So I create interaction terms between international recognized quality certificate variable and export variable as: international quality certificate variable * export variable and then I run probit model as below:

# probit energymanagement employees othermanufacturing garments otherservices metalproducts construction EX interantionalcertificate
> s excertificates

As you can see, "export", "owning international recognized certificate" and their interaction term has meaning. But what I wonder is that "owning international certificates" also could promote enterprises to export ( the image is as below) but the correlation is weak as I think, so I can ignore its relationship among these independent variables ( I am so sorry, I am new with probit and read many different resources from not only this forum but also other academic papers so I am not sure I am correct).

#probit EX interantionalcertificates

and the correlation between dependent variable and independent variables are weak also, as below:

# corr energymanagement employees EX interantionalcertificates

Normally, as I understand, I just need to check relationship between independent variables to avoid the multicollinearity, but in this case, I do not know whether my idea that creating interaction terms is bad or not because "owning international certificates" variable has positively impacts to export as well,

This forum is so useful for me and I learned a lot from this forum,

Thank you

[Carlo Lazzaro]

Hoa:
I think you have to create a brand new post with an informative title, as Hidde's post focuses on a very different issue (you share the thesis topic only). Thanks.

[Carlo Lazzaro]

Hidde:
thanks for clarifying.
I do not know whether what follows is ritual or not in your research field, but I woud chose one out the three timespan (say, -10+10) as a baseline analysis and the remaining ones as two scenario analyses.