Yet Another Blog in Statistical Computing

I can calculate the motion of heavenly bodies but not the madness of people. -Isaac Newton

SAS Macro for Engle-Granger Co-integration Test

In the coursework of time series analysis, we’ve been taught that a time series regression of Y on X could be valid only when both X and Y are stationary due to the so-call “spurious regression problem”. However, one exception holds that if X and Y, albeit non-stationary, share a common trend such that their trends can be cancelled each other out, then X and Y are co-integrated and the regression of Y on X is valid. As a result, it is important to test the co-integration between X and Y.

Following the definition of co-integration, it is straightforward to formulate a procedure of the co-integration test. First of all, construct a linear combination between Y and X such that e = Y – (a + b * X). Secondly, test if e is stationary with ADF test. If e is stationary, then X and Y are co-integrated. This two-stage procedure is also called Engle-Granger co-integration test.

Below is a SAS macro implementing Engle-Granger co-integration test to show the long-term relationship between GDP and other macro-economic variables, e.g. Personal Consumption and Personal Disposable Income.

SAS Macro

%macro eg_coint(data = , y = , xs = );
*********************************************************************;
* THIS SAS MACRO IMPLEMENTATION ENGLE-GRANGER COINTEGRATION TEST IN *;
* A BATCH MODE TO PROCESS MANY TIME SERIES                          *;
*********************************************************************;
* INPUT PARAMETERS:                                                 *;
*   DATA: A INPUT SAS DATASET                                       *;
*   Y   : A DEPENDENT VARIABLE IN THE COINTEGRATION REGRESSION      *;
*   X   : A LIST OF INDEPENDENT VARIABLE IN THE COINTEGRATION       *;
*         REGRESSION                                                *;
*********************************************************************;
* AUTHOR: WENSUI.LIU@53.COM                                         *;
*********************************************************************;

options nocenter nonumber nodate mprint mlogic symbolgen
        orientation = landscape ls = 150 formchar = "|----|+|---+=|-/\<>*";

%local sig loop;

%let sig = 0.1;

%let loop = 1;

%do %while (%scan(&xs, &loop) ne %str());

  %let x = %scan(&xs, &loop);

  ods listing close;
  ods output FitStatistics = _fit;
  proc reg data = &data;
    model &y = &x;
    output out = _1 residual = r;
  run;
  quit;

  proc sql noprint;
    select cvalue2 into :r2 from _fit where upcase(label2) = "R-SQUARE";
  quit;

  proc arima data = _1;
    ods output stationaritytests = _adf1 (where = (upcase(type) = "ZERO MEAN" and lags = 1) drop = rho probrho fvalue probf);
    identify var = r stationarity = (adf = 1);
  run;
  quit;
  ods listing;

  %if &loop = 1 %then %do;
    data _adf;
      format vars $32. lterm_r2 best12. flg_coint $3.;
      set _adf1 (drop = type lags);
      vars = upcase("&x");
      lterm_r2 = &r2;
      if probtau < &sig then flg_coint = "YES";
      else flg_coint = "NO";
    run;
  %end;
  %else %do;
    data _adf;
      set _adf _adf1 (in = new drop = type lags);
      if new then do;
        vars = upcase("&x");
        lterm_r2 = &r2;
        if probtau < &sig then flg_coint = "YES";
          else flg_coint = "NO";
      end;
    run;
  %end;
    
  %let loop = %eval(&loop + 1);
%end;

proc sort data = _last_;
  by descending flg_coint probtau;
run;

proc report data = _last_ box spacing = 1 split = "/" nowd;
  COLUMN("ENGLE-GRANGER COINTEGRATION TEST BETWEEN %UPCASE(&Y) AND EACH VARIABLE BELOW/ "
         vars lterm_r2 flg_coint tau probtau);
  define vars      / "VARIABLES"                      width = 35;
  define lterm_r2  / "LONG-RUN/R-SQUARED"             width = 15 format =  9.4 center;
  define flg_coint / "COINTEGRATION/FLAG"             width = 15 center;
  define tau       / "TAU STATISTIC/FOR ADF TEST"     width = 20 format = 15.4;
  define probtau   / "P-VALUE FOR/ADF TEST"           width = 15 format =  9.4 center;
run;

%mend eg_coint;

%eg_coint(data = sashelp.citiqtr, y = gdp, xs = gyd gc);

SAS Output

----------------------------------------------------------------------------------------------------------
|                  ENGLE-GRANGER COINTEGRATION TEST BETWEEN GDP AND EACH VARIABLE BELOW                  |
|                                                                                                        |
|                                       LONG-RUN      COINTEGRATION         TAU STATISTIC   P-VALUE FOR  |
|VARIABLES                              R-SQUARED         FLAG               FOR ADF TEST    ADF TEST    |
|--------------------------------------------------------------------------------------------------------|
|GC                                 |      0.9985   |      YES      |             -2.8651|      0.0051   |
|-----------------------------------+---------------+---------------+--------------------+---------------|
|GYD                                |      0.9976   |      YES      |             -1.7793|      0.0715   |
----------------------------------------------------------------------------------------------------------

From the output, it is interesting to see that GDP in U.S. is driven more by Personal Consumption than by Personal Disposable Income.

Advertisements

Written by statcompute

June 28, 2015 at 11:46 am

%d bloggers like this: