Inputting and outputting into and from do loop on basis of name vector?

[LauriN]

Hi,

I'm not quit sure how does this work so the header might be misleading or then it's indeed what I'm seeking for.

However, I have this regression procedure which I need to repeat on my sample of 30 firms:

The preregress module formulates dependent and independent variables which are then used in regress module.

I use "read where()" command to input firms from pooled data. Also I have to form every time a vector "resid_firmname = resid". How could I just make a vector out of the firm names and such a loop which would do this procedure firm by firm and also how to formulate such a output vector for every firm?

proc iml;

edit kirjasto.Dax30_hourly09;

/*394 missing values of volume, 264 for ALTG*/

/*BAYG changed to BAYGn on 21.09.2009*/

/*create variables for excess volume regression*/

start preregres (vol, high, low, close) global (x, y);

lnvol=log(vol);

lagvol=lag(lnvol, (1:50));

vola=high/low;

lnvola=log(vola);

lagvola=lag(lnvola, (1:10));

close10=lag(close, -10);

lnclose10=log(close10);

timeorig=T(1:nrow(vol));

tempx=lagvol||lagvola||lnclose10;

xtemp2=tempx[51:(nrow(vol)-10),];

lintrend=T(1:nrow(xtemp2));

constant=j((nrow(xtemp2)),1);

x=constant||xtemp2||lintrend;

y=lnvol[51:(nrow(vol)-10)];

finish;

/*list of firms:

adsg, altg, alvg, basf, bayg, beig, bmwg, cbkg, cong, DAIGn, DB1Gn, dbkg, dpwg, dteg, eong,

fmeg, freg, heig, hnkg, ifxg, lhag, ling, lxsg, mrcg, muvg, rweg, sapg, sdfg, sieg, tkag, vowg

*/

print resid_adsg resid_altg resid_alvg resid_basf resid_bayg resid_beig resid_bmwg resid_cbkg resid_cong resid_daig resid_db1g;

read all var "volume" into volume where(RIC="DB1Gn.DE");

read all var "high" into high where(RIC="DB1Gn.DE");

read all var "close" into close where(RIC="DB1Gn.DE");

read all var "low" into low where(RIC="DB1Gn.DE");

run preregres (volume, high, low, close);

start Regress;                    /* begins module        */

  xpxi = inv(x`*x);               /* inverse of X'X      */

  beta = xpxi * (x`*y);           /* parameter estimate  */

  yhat = x*beta;                  /* predicted values    */

  resid = y-yhat;                 /* residuals           */

  sse = ssq(resid);               /* SSE                 */

  n = nrow(x);                    /* sample size         */

  dfe = nrow(x)-ncol(x);          /* error DF            */

  mse = sse/dfe;                  /* MSE                 */

  cssy = ssq(y-sum(y)/n);         /* corrected total SS  */

  rsquare = (cssy-sse)/cssy;      /* RSQUARE             */

/* print ,"Regression Results", sse dfe mse rsquare; */

  stdb = sqrt(vecdiag(xpxi)*mse); /* std of estimates    */

  t = beta/stdb;                  /* parameter t tests   */

  prob = 1-probf(t#t,1,dfe);      /* p-values            */

  /*print ,"Parameter Estimates",, beta stdb t prob;

  print ,y yhat resid;*/

finish Regress;                   /* ends module          */

run Regress;

create kirjasto.residuals var {resid};

append;

resid_db1g=resid;

close kirjasto.residuals;

[Rick_SAS · Accepted solution]

To "loop over firms", follow the ideas in this article: BY-group processing in SAS/IML - The DO Loop .

Outside the loop, use the SETIN and SETOUT statement to define the data sets that you are reading from and to. (For an example, see Sequential access: Reading one observation at a time in SAS/IML software - The DO Loop )

Use the CREATE statement to define the variables for the output data set.

Then loop over the firms. Inside the loop:

1) READ ALL... WHERE(RIC=firm);

2) Compute the regression model for that firm.

3) Write the statistics to the output data set

[LauriN]

Oh yeas it was pretty straight forward after all. The outputting remained somewhat unclear for me but I realized another solution to the problem:

I initialized matrix with width = number of firms, length = length of output vector (residuals in this case) and replaced the columns in this vector with real residuals in the do-loop:

proc iml;

use kirjasto.dax30h09v4;

read all var{RIC} into c;    /*reading class variables, firm id's*/

firm=unique(c);

res_matrix=j(2226,ncol(firm),1);

do i = 1 to ncol(firm);

read all where(RIC=(firm));

run preregres (volume, high, low, close);

run Regress;

res_matrix[,i]=resid;

end;

create kirjasto.residuals from res_matrix [colname=firm];

append from res_matrix;