[Nick Cox]

That's not a precise criterion so far as I can see. Suppose market growth has always been positive. Do you want the calculation to be based on the last 4, 5, 6, ... years?

[laura wolff]

I would want the calculation to be based on all previous years, and it should show me a 1 for consecutive runs of positive values, if there have been more than 3 positive years. So if growth has always been positive, I'd have a 1 for each year starting in year 4. Is this something I can do using tsegen and rall?

[Nick Cox]

The criterion still looks ambiguous to me. What do you want if there have been 4 years of previous positive growth but not 5?

What you can do is calculate the minimum over the last # years of growth. If that minimum is positive, then all growth in that interval was positive.

Here's a sandbox example;

webuse grunfeld
gen growth = D.invest
tsegen min_4_prev_growth = rowmin(L(0/3).growth)

You can map that last variable to an (0, 1) indicator if desired.

Alternatively, identify spells of growth using tsspell (SSC).

[Zhang Hui]

Dear all,

As shown below, stake is the percentages of firm i investment. pe_id is the id of firms.
How can I calculate the rolling window of stake during an one-year or two-year rolling window?

Please kindly note that there some “gap”s in "stake".
Thank you.

investmentdate	stake	industry_id	pe_id	year
1995-8	0	49	944	1995
1995-1-1	0	67	1543	1995
1995-9	8.4	89	1541	1995
1995-10-1	0	181	1543	1995
1995-10-19	30	187	1040	1995
1995-6	0	198	1444	1995
1995-6	0	198	1339	1995
1996-1-18	0	25	1543	1996
1996-10	0	102	1444	1996
1996	18.9	148	1490	1996
1996-9-1	36.7	153	1543	1996
1996-12	0	187	1040	1996
1997	0	93	1541	1997

[Clyde Schechter]

What does rolling window of stake mean? Do you mean the mean of stake over a rolling window? Or the standard deviation? Or some other statistic? And do you want this to be done separately by industry? Or by pe_id? Or just overall? Your question is completely unclear.

Also, with 24 posts under your belt, it surprises me that you have not already been introduced to the -dataex- command for showing example data. In the future, when showing data examples, please use the -dataex- command to do so. If you are running version 15.1 or a fully updated version 14.2, it is already part of your official Stata installation. If not, run -ssc install dataex- to get it. Either way, run -help dataex- to read the simple instructions for using it. -dataex- will save you time; it is easier and quicker than typing out tables. It includes complete information about aspects of the data that are often critical to answering your question but cannot be seen from tabular displays or screenshots. It also makes it possible for those who want to help you to create a faithful representation of your example to try out their code, which in turn makes it more likely that their answer will actually work in your data.

When asking for help with code, always show example data. When showing example data, always use -dataex-.

[Alessandro DiNola]

Does this work also to compute 10-year centered rolling standard deviations? I have a panel with t=year and id=firm. Suppose I wanna compute the standard deviation of a firm-level variable x over a 10 year rolling window, centered. I cannot use mvsumm (with or without end option), so can I use tsegen? More specifically, suppose I wanna assign to each t, the standard deviation of the window {t-4,t-3,..,t+5}. I tried the following

  tsegen x_sd_10 = rowsd(L(1/4).x F(1/5).x), by(id)

but it doesn't work as expected

[Clyde Schechter]

By specifying L(1/4).x F(1/5.x) you leave out x itself, which is probably why you didn't get what you expected. You can fix that with L(0/4).xF(1/5.x), which will then include x. (L0.x is just x itself.)

Alternatively, you can do this with -rangestat-, by Robert Picard, Roberto Ferrer, and Nick Cox, or with -asrol-, by Attaullah Shah. Both are available from SSC. -rangestat- is a more general program from a math/stats perspective, whereas -asrol- is more tailored specifically to the kind of problems that come up frequently in finance and economics.

Added: The code for this using -rangestat- would be

rangestat (sd) x_sd_10 = x, by(id) interval(time -4 5)

where you replace time with the actual name of your time variable.

[Alessandro DiNola]

Dear Clyde, thank you very much for your suggestion, it fixed the problem! Following up on your comment about rangestat, are these two codes equivalent?

tsegen `x'_sd10 = rowsd(L(0/4).`x' F(1/5).`x', 10)
rangestat (sd) `x'_sd_10_bis = `x', by(id) interval(t -4 5)

where x is the name of the variable and t is the time variable (year). In particular I would like to compute the rolling standard deviation only on companies that have at least 10 years of data (i.e. the window has to contain 10 non-missing values).

[Nick Cox]

Just add (count) to the statistics calculated by rangestat and (if you must) drop what you don't want.

(I don't get these little rules: 10 values means that the SD is worthwhile, but 9 means it is useless?)