Amateur questions: Cross-sectional analysis

[Marina Jenny]

Hello everyone!

I have surely really amateur questions, but I am struggling with a simple cross-sectional analysis. I am fairly new to STATA and have been reading a lot but guess I only get more confused the more I read/watch on Youtube.

I tell you my case:

In my sample I have 150 pension funds with quarterly observations of returns, funding ratio and invested capital from 2015 - 2020. So time-series-wise only 24 observations, however I have a good amount cross-section-wise (150).

Now what I do have is I have one datasheet where I basically have the different quarters in the header, so basically each quarter (Q12015, Q22015 etc) is a new variable (table view, see below). And I have the same data but where I just have the variable "Period" and under this variable all the quarters repeatedly per fund. Firstly, I am unsure which one of these two sets is better for me. I want to make like plots of the performance (returns/funding ratio quarterly) per fund over time, so that I can observe differences. In order to do so, I am unsure which dataset to use, as both ways I get no nice plots - maybe because of the 150 funds or maybe because I miss to somehow sort or group them..

"Tabel-view":

Other view:

Additionally, I was wondering how I can easily do a cross-sectional regression to detect (reasons) differences in my cross section. I thought of dividing my sample into fractiles,

Sorry for this probably rather beginner's questions.. but any help is highly appreciated. I have been trying to figure this out for hours.

Thanks a lot in adavance already!

Marina

[Ken Chui]

The current question in #1 is not easy to adress because it does not tell us what goals the analysis is trying to achieve. I think you may be examining the data somewhat prematurely. Perhaps try to close the spreadsheet software, take a break if you must, and come back with either an empty file in a word processor or a piece of paper and a pen, and start drafting your analysis questions. For example:

1. Do the mean funding ratios in 2015Q1 and 2020Q4 differ?
2. What do the return fund risk look like if I plot that as a time trend? On average and maybe for each company? (Or if there are too many companies... maybe just top 30 in terms of volume?)

You can draft as many as you wish. (I'd suggest 3-5, as each question can potentially spin off to bigger endeavor.) And they should collectively achieve the main goal that motivated you to collect these data to begin with. A couple more pointers to think about:

1. Instead of "nice plot," describe what the plot looks like. It's dangerous to play with the data freely and try to stumble upon a plot that "speaks to us," this method is prone to our own biases.
2. Perhaps we are not using the term "cross-sectional" the same way. But this data set is a classic example of "what data are not considered cross-sectional" because the companies were followed up along time. I sensed that there is a plan to boil the time trend down to a single difference so that each company is a unique row, but that plan of difference computation needs to be clearly defined. And also this approach would lead to a great loss of information, if not end up misleading.

I hope this helps.

[Marina Jenny]

Hi Ken Chui, thank you very much for your input! It does help