Re: [Gretl-devel] The mysterious stack() function

In my reply, at http://lists.wfu.edu/pipermail/gretl-devel/2018-September/009010.html to Sven's post on the kinda weird stack() "function" at http://lists.wfu.edu/pipermail/gretl-devel/2018-September/009009.html I finished by saying, And I thereby left the business unresolved! Here's an attempt at resolution. It turns out that the problem for which stack() provides a solution is pretty easily handled by series -> matrix -> reshape matrix -> series The problem is described in section 4.5 of the Gretl User's Guide. That is, we import some panel data that are organized "by variable": T columns hold the time periods and N rows the "units" or individuals. More than one variable may be given in such a file, in which case we assume the k variables are stacked vertically so the data block as a whole has N*k rows and T columns. How do we make a gretl panel dataset out of this? The matrix solution goes as follows, where I assume the "by variable" data are in foo.csv (with no extraneous rows) and there are N=10 individuals. (The value of N is the only "extra" information, not present as such in foo.csv, that is needed.) <hansl> open foo.csv N = 10 # number of individuals set skip_missing off matrix X = {dataset} set skip_missing on T = cols(X) # number of periods NT = N * T # total panel series length k = nelem(X)/NT # "real" number of series X = mshape(X', NT, k) nulldata NT --preserve setobs T 1:1 --stacked-time-series list Panelvars = null loop i=1..k --quiet Panelvars += genseries(sprintf("x%d", i), X[,i]) endloop store panelized.gdt Panelvars </hansl> So here's a suggestion: in section 4.5 of the Guide I delete the discussion of the fix using the mysterious stack() in favour of an example and explanation of the method illustrated above. Should I do that? If so, stack would then be undocumented, but we could keep it for the time being for backward compatibility. Allin