Thank you for your reply.   I guess, "EG" means "Enterprise Guide"? Then no, I have run it with base SAS (yes, there is a base SAS users as stupid as me out there).   I will terminate the program. ... View more

Hello,   this is enormous. I really had allready implemented it via a %macro %Do Loop. The amount of time if do save is astonishing. Thank you so much for this - I need this very procedure "over an over" so this really changes the game.   As I do have to implement it for more than one variable, would you mind considering if I have transferred it correctly to a setting more closely to the actual one?   data want; set have; retain x_end y_end z_end; if _n_ ne 1 then x=x_end; if _n_ ne 1 then y=y_end; if _n_ ne 1 then z=z_end; x_end = intercept1 + a*x + b*y + c*z ; y_end = intercept2 + d*x + e*y + f*z ; z_end = intercept3 + g*x + h*y + i*z ; run; So, former y is now "x_end"; Especially, the three ifs are a concern of mine - could I achieve a better solution?   But, nonetheless, your very answer means a huge leap indeed for me. Once again, thank you so much.   ... View more

Hello,   First: Sorry once again for my delayed answer.     @Patrick wrote: You're welcome. I couldn't test the code so hope there are no syntax errors in it. One thing appreas to be missing (as I see it):     ** Merge; data slopes; merge intercept (keep=Parameter Estimate where=(Parameter="a") rename=(Estimate=a)) steigung (keep=Parameter Estimate where=(Parameter="b") rename=(Estimate=b)) ; drop Parameter; dte_id = &i.+1; run; "Intercept" and "steigung" have never been referenced in your code - did you just miss them or can I even merge data sets not having been created (so to say: create data sets in the very data step which entails merging)?   So, I changed it to this: ** Merge; data slopes; merge Parameters (keep=Parameter Estimate where=(Parameter="a") rename=(Estimate=a)) Parameters (keep=Parameter Estimate where=(Parameter="b") rename=(Estimate=b)) ; drop Parameter; dte_id = &i.+1; run; (which is, de facto, merging data sets not even having been created - I do not need the data sets "intercept" and "steigung" at all, as I see it now - six steps less).     Otherwise: Once again, a big thank you. I am honestly still stunned. Especially, this "(keep rename ...)" procedure during data steps helps me a lot. If I consider, how many useless data steps are in my "programs". My oh my...   @Patrick wrote: [...]   Give it a go and should you still have serious performance issues then I'd suggest you open a new discussion and just reference this one.   If further performance improvements would be required then you would need to attach a log so we can see what volumes you're dealing with and where the time gets spent. If posting a log please run your code with options "option fullstimer mprint;" set.   Well, this would make me feel even more like a leech. It would feel like "well, here, look you all, for I have written a program to achieve this and that, go tidy it up, as I do not have time for that." I already have a bad conscious because of all the extensive answers I received thus far - and still have multiple questions.         ... View more

Hello. @Ksharp wrote: So both question are the same ? You want draw row vector not column vector ? Yes, both questions are the same - that is, what I meant with "likewise". I should have expressed my question better - I am sorry for that. @Rick_SAS wrote: Your syntax will be simpler and your meaning clearer if you use the RANPERM function in IML instead of the SAMPLE function. Sampling N items without replacement from a set that contains N elements is a permutation of the elements: proc iml; s = ranperm(1:20); /* equivalent to sample(1:20, 20, "WOR"); */ print s;   So, this eventually brings together PG's comment and my desire to solve it via "proc iml".   Thank you for - showing me the right commands in SAS, as well as, - clarifying, what I am actually aiming for (which I should have known by myself).   Once again, I am really happy to see that great kind of a help.     ... View more

Thank you very much to both of you - PG for simplifying and solving my issue (I only want to reorder the sample randomly) and Xia Keshan for dealing with my "proc iml" issue.   Either, I am too clumsy to get your solution to work, or, I do not have expressed my wish well enough. What I want to do, is "draw from the rows of a NxE-matrix ", such that I get N row-vectors (matrix is "drawn empty") and write this draws to a data set with N rows and E columns.   I just came across this solution, which, I hope, does implement what I strive for: proc iml; call randseed(1); use Sashelp.Cars nobs N; varNames = {"MPG_City" "Length" "Weight"}; read all var varNames into x[rowname=Model]; close Sashelp.Cars; obsIdx = sample(1:nrow(x), N); /* sample size=N, rows chosen from 1:NumRows */ s5 = x[obsidx, ]; /* extract subset of rows */ print s5[rowname=(Model[obsIdx]) colname=varNames]; (in the original, sample size is "5"; "nobs N" is added 5 changed to N, by me).   Again, thank you a lot. ... View more

@RW9 wrote: Its not really clear to me, test data in the form of a datastep and required output generally helps.    I am sorry for this obvious dereliction - I should have prepared better, before positing. Next time, I should not let panic seize hold of me, if something does not work properly or I cannot solve directly.   @RW9 wrote: Conditionals in sql are done by using case statements [...] So I filter the data in the from clause before the groups and aggregates are created.  This is a useful technique to learn. [...]   Nonetheless, that was exactly the kind of support I asked for. Thank you.                   ... View more

So, you leave me really stunned. This is enormous - thank you so much, it is really awesome. All this time you have taken.   I indeed do have performance issues, such that I planned running the regressions during night. First and foremost, I have to look up all the statements used by you for I partially never have come across them.   Again, thank you very much indeed. The help received here thus for is truly great.     ... View more

Dear Patrick,   I considered your remark for ~ three days now and still am too bovine to think of an appropriate answer. Sorry for that. For as I do not want to leave your question unanswered (especially faced with the fact I received such great and quick help), I would like to try to show how I implemented the "dte_id".   No doubt that my "program" and whole approach might be horrible and is poles apart from a solution a versed programmer would be capable of coming up with, but still, it seemingly is working for me.     ******************************************************************************* Run 'expanding' regression, copying each paramter-estimates to next observation; *'*'*'*'*'*'*'*'*'*'*' MA_m&u._i&t. := Moving average of predictor variable at end of month [Note: &u. and &t. denote different Modell specifications, "m" for model, "i" for ICC (my 'predictor')] slnrpk1 := realized return at end of NEXT month (therefore "1") vw_ts := Complete value weighted time series; 'Appending' Regression; %macro insample; %DO i=354 %TO 557; %LET rgno = &i.; data vw_ts_m&u._i&t._&rgno.; set vw_ts_m&u._i&t.; if dte_id LE &i.; run; %ODSoff; proc model data=vw_ts_m&u._i&t._&rgno.; endo slnrpk1; exog MA_m&u._i&t.; instruments _exog_; parms a b; slnrpk1 = a+b*MA_m&u._i&t.; fit slnrpk1 / gmm kernel=(bart,1,0) vardef=n; ods output ParameterEstimates=Parameters; run; quit; %ODSon; * Isolate b0; data intercept; set Parameters; if Parameter="a"; run; data intercept; set intercept; keep Estimate; data intercept; set intercept; rename Estimate=a; run; * Isolate b1; data steigung; set Parameters; if Parameter="b"; run; data steigung; set steigung; keep Estimate; data steigung; set steigung; rename Estimate=b; run; ** Merge; data slopes; merge intercept steigung; run; data slopes; set slopes; dte_id = &i.+1; run; *** Attach paramters to "next" data point, in order to be capable of getting forecast and error and such; data Vw_ts_m&u._i&t.; merge Vw_ts_m&u._i&t. slopes; by dte_id; run; *** house cleaning; proc datasets library=work nolist; delete vw_ts_m&u._i&t._&rgno.; run; quit; %END; %mend insample; %insample;    So for that "DO" Loop, I needed my "dte_id", because, I thought, I were to need an interger to implement the loop. ... View more

Once again, I received exactly the help I needed (thus, yes, Patrick, above gives me what I am after). Thank you very much, both of you! @LinusH wrote: How will you use this new id? Why can't you use date directly?   If the data is sorted by date, use a data step, BY date. Then increase your id +1 whenever first.date is true. I want to use this new ID for a "expanding" regression I do have to run. My sample is to be divided into two parts - an "in-sample" part, where I get parameter-estimates and an "out-of-sample" part, where I do apply the parameter-estimates based on the prior information in such a kind, that I use them to obtain an estimate for the next data point (e.g. from January 1970 until December 1998, I run the regression each month, of ret_realzd on estor_ret, with "returns_realized" := average realized returns and "estor_ret" := average estimator for returns; then, in January 1999, I multiply my estimated slope with the then-prevailing estor_ret, add the intercept and look how well I approximate the realized return, by considering the “error”); Then, for the next set of regressions, the period of estimation prolongs by one month, which is to say, that I run the regression from January 1970 up to January 1999; then, I investigate, with the same procedure, how well I can approximate the realized return as of February 1999.   I am considering to implement this procedure with a DO-Loop; From t=70 (e.g., end of first insample-period) to T=130 (total sample-end); with every increment, the data-set used for regression "grows" by one step. So, for me, as someone without any SAS-experience, it is very convenient to have a date_id; I consider to use this in correspondence with my planned DO-loop.   ... View more