topic Re: cross sectional regression by month- NOT Rolling regression in SAS Programming

cross sectional regression by month- NOT Rolling regression

Saba1 — Thu, 05 Oct 2017 04:44:28 GMT

I have a panel data shown below with "id" identifies different stocks and "month" is monthly dates column. The data set is sorted by id and month.

data have; 
  input id month x1 x2 y;
  datalines;
45  09/30/2004  -0.30  1.34  1
45  09/30/2004  -0.30  1.34  1
45  05/31/2005  -0.08  0.78  0
45  12/31/2008    .     .    0
48  09/30/2001  -0.01  1.88  1
48  02/28/2004   0.78  1.50  0
48  05/31/2005    .    0.54  1
49  01/31/2002   0.05   .    1     
51  03/31/2001  -0.06  1.48  0
55  05/31/2005   0.30  1.21  1
55  05/31/2005   0.30  1.21  1
55  11/30/2007   .      .    0
; run;

I need to run cross sectional regression (proc reg) "by month". So the purpose is to observe variations in coefficient (beta) of "x2" over time i.e. over various months. I am using following proc but there is an Error saying "month" is not sorted in ascending order. When I sort the data by month and run the regression, resulting output gives "0" coefficients for missing values.

proc reg data=have;
 model y=x1 x2;
 by month; run;

Kindly guide me in this regard. I also need to know how to create a data set based on time variant coefficients of only x2 along-with their corresponding months (date), and how to ultimately draw their graph.

Thanks.

Re: cross sectional regression by month- NOT Rolling regression

Ksharp — Thu, 05 Oct 2017 06:35:37 GMT

If I was right, you need make a new variable to flag MONTH.

new_month=month(month);

then use it as a BY variable.

proc reg data=have;
 model y=x1 x2; by new_month; run;

But I noticed your Y variable is 0 or 1. Maybe you need check PROC LOGISTIC .

Re: cross sectional regression by month- NOT Rolling regression

Saba1 — Thu, 05 Oct 2017 16:43:34 GMT

@Ksharp thanks for the reply. Can I use the same By month statement in PROC LOGISTIC? Secondly how to create a data set & graph of only x2 betas (over time)?

Re: cross sectional regression by month- NOT Rolling regression

Ksharp — Fri, 06 Oct 2017 13:39:22 GMT

Yes. I think so.

for your second question, use OUTEST= option of PROC LOGISTIC to save the betas and use PROC SGPLOT to display it by month.

Re: cross sectional regression by month- NOT Rolling regression

Saba1 — Sat, 07 Oct 2017 03:32:20 GMT

@Ksharp right, thanks.