Question regarding regression type and output

[Theodor Hansson]

Hi everyone,

Trying to figure out if this OLS regression makes sense. I’m aware that the R-squared output is pretty low, but in general, I would like to know if I can make a regression and set the independent variable as Change in Institutional ownership structure Quarter 1 (For Special Purpose Acquisition Company). The dependent variables are abnormal returns for the same period, the size of the transaction value, SIC (industry) and Firm Age.

The output indicates that there are significant variables such as ARQO1, Firm Age and totaltransaction value.

I do this for three quarters and for quarters zero to three. For SPACs and for IPOs. I test for insider ownership as well, but I wonder if someone could see any potential by doing these OLS regressions. Or should I try some other regressions?
Kind Regards,
Theodor Hansson

[Carlo Lazzaro]

Theodor:
are you dealing with a cross-sectional or a panel dataset?
If the latter were the case, I"d point you out to -xtreg,fe-.

[Theodor Hansson]

Carlo: Thank you for your reply. I believe that the set is panel dataset, since I have more than ONE unit, (not many) and at different points in time, in other words I do have more than one unit for different point of times. All listed de-SPAC (325 firms) in U.S market from 2020-2023 (stock prices and return and so on) and traditional post-IPO (approx 650 firms).

So if I understand your right, I would instead try commands in Stata for xtreg with fixed effects? Do I need to structure the dataset in Excel in some way to get a output from the xtreg, fe? Every recommendation or hint will help me a lot.

Kind regards,

Theodor

[Clyde Schechter]

I think the distinction between panel data and cross-sectional data that Carlo was getting at is not just a matter of multiple units. The issue is about whether you have units that provide data at multiple time periods, as opposed to each time period having a different set of units. For example, in some periodic national surveys, a panel of people is recruited and those people then provide data in each (or nearly each) of separate waves at different time periods. That's panel data. By contrast, in other national surveys a different group of people is surveyed every year (and perhaps a few people end up in more than one wave of the survey, but just infrequently and by happenstance.) That is cross-sectional data.

Do I need to structure the dataset in Excel in some way to get a output from the xtreg, fe?

First, no, you do not need to change the structure of your data. Given that you were able to perform the -regress- and incorporate all units at all times for which complete information was available, the data set is definitely structured properly. What you do need to do is identify what the unit here is. I'm guessing it's the firm and that your data set has some variable which identifies firms. Let's call that variable firm_id. Then you need to run the command -xtset firm_id- before you can use -xtreg, fe- (or any other -xt- command) to tell Stata that the panels are defined by firms.

That said, even if your data did require substantial restructuring in order to do the analysis you need, you should never do things like that in Excel. I assume that you are engaged in this project for some serious purpose and there is at least some chance that you or others will rely in some material way on your findings. (If you are just playing around for fun, or to get acquainted with Stata commands, then you can disregard the rest of this.) That kind of work requires a complete and accurate audit trail of every step in the way. This is not only so that others can double-check your work, but so that you will be able to reproduce it in the future if the need arises, or if the need to perform essentially the same analysis on a new set of data comes your way. Excel leaves no audit trail. In addition, doing data transformations in Excel is quite error prone because it is far too easy to make mousing or typographical errors when selecting ranges of cells to apply operations to. No, serious data management and analysis should only be done with software that leaves a complete record of every step from beginning to end. Stata does that with your do-files or log-files. All the other commercial statistical packages have some equivalent to those.

Excel is fine for visual display of tabular data. Many people like the graphs Excel produces (although, personally, I don't.) It's also a reasonable lingua franca for exchange of data sets among people who do not use the same statistical software, since nearly everybody has Excel and all statistical software I know of can both read and write Excel files. But it is poor practice to use it for data management or analysis in a professional setting. I would never rely on any analysis that was carried out, even in part, in Excel, and you shouldn't either.

[Theodor Hansson]

Clyde: Thank you for your reply. Do you think it is possible to create an interaction variable for Deal Size (Total transaction Value) * (multiple) with IPO Abnormal return than an interaction variable for Deal size * SPAC Abnormal return?

[Clyde Schechter]

Well, from a Stata coding perspective, you can make any interaction you want. The question is whether that interaction makes sense in terms of finance/economics. That's out of my field and I can't advise you on that.

[Carlo Lazzaro]

Theodor:
as I learnt corporate finance in the pre-EBITDA era, the only advice I can give you is to check the quantitative literature in this field and see what interactions were performed in previous researches that covered your same topic.
As Clyde wisely highlights, Stata allows us to code (almost) everything, but it's up to us to code something that makes sense in our research community.