new;cls;
maxk = 8;  @数据选择滞后值的设置的最大滞后@
n = 100;
crit_t = 1.645;
p = 1;       @ 1 = with trend, 0 = with constant only, -1 = no constant or trend @
select = 0;  @ 1 = fixed lag, 0 = data dependent lag selection @
y = cumsumc(rndn(n,1));  @生成y序列@
@ or, Read the dat file here. load data[n,1] = data.txt;  @

/* Using a Fixed lag */
if select == 1;
   lag = 4;
   {beta,tval, alpha, tstat} = adf(y,p,lag);   @调用ADF估计程序@
   format /rd/m1 12,4;
   "Using a Fixed lag"
   "Fixed lag    = " lag;
   "T(a-1)       = " alpha;
   "ADF stat     = " tstat;
   "Estimated coefficients :
     (x(t-1), dx(t-1), dx(t-2), ... dx(t-k), const, trend) ";
   beta';
   "t-stat";
   tval';
   goto next;
endif;

/* Using data-driven lag selection procedure */
statkeep = zeros(maxk+1,1);
tvalkeep = zeros(maxk,1);

{beta,tval, alpha, tstat} = adf(y,p,0);
statkeep[1] = tval[1];
j=1;
do while j <= maxk;    @构造数据选择滞后值的一个循环@
   {beta,tval, alpha, tstat} = adf(y,p,j);
   statkeep[j+1] = tval[1];   /* statkeep  for k = 0, 1, 2, ..., maxk */
   tvalkeep[j] = tval[j+1];   /* tvalkeep  for k = 1, 2, ..., maxk */
j = j + 1;
endo;

@ selection using t-test，这个可以看作者的文章，通过abs(tvalkeep[j]) > crit_t来选择@
/* select the lag where the coeff is sig. from maxk to 0. if all coeff are insig., kk = 0  */
j = maxk;
do while j >= 0;
        if (abs(tvalkeep[j]) > crit_t) or (j == 0);
            kk = j;
            j = 0;
        endif;
j = j - 1;
endo;

@ kk = selected lag @
{beta, tval, alpha, tstat} = adf(y,p,kk);
   format /rd/m1 12,4;
   "Using Data dependent selection of the lag";
   "Selected lag = " kk;
   "T(a-1)       = " alpha;
   "ADF stat     = " tstat;
   "Estimated coefficients :
     (x(t-1), dx(t-1), dx(t-2), ... dx(t-k), const, trend) ";
   beta';
   "t-stat";
   tval';
next:
end;

/*** ADF
** Format:  {beta, tval, alpha, tstat} = adf(x,p,l);
** Input:   x     -- time series variable
**          p     -- order of the time-polynomial to include in the
**                   ADF regression.  Set p = -1 for no deterministic
**                   part.
**          l     -- number of lagged changes of x to include in the
**                   fitted regression.
** Output:  alpha       --  estimate of the autoregressive parameter;
**          tstat       --  ADF t-statistic
**          beta        --  coeff. of (x(t-1), dx(t-1), dx(t-2), ... dx(t-k),
                                   constant, time trend)
**          tval        --  t-values
*/
proc (4) = adf(x,p,l) ;
local b,k,z,res,so,var_cov,xx, tval;
local timep,t,m,xmat,nobs,dep,ch,f;
     if (p < -1);
        "Error: p cannot be set < -1";
         retp(0,0,zeros(6,1));
     endif ;
     if (cols(x) > 1);
        "Error: ADF cannot handle a data matrix; cols(x) > 1 (=" cols(x) ")";
         retp(0,0,zeros(6,1));
     endif ;

     nobs    = rows(x);
     if (nobs - (2*l) + 1 < 1) ;
        "Error: l is too large; negative degrees of freedom.";
         retp(0,0,zeros(6,1));
     endif ;

     dep     = trimr(x,1,0);
     ch      = diff(x,1);
     k       = 0     ;
     z       = trimr(lagn(x,1),1,0) ;
     If (l > 0) ;
        do until k >= l ;
           k = k + 1 ;
           z = z~lagn(ch,k) ;
        endo ;
     Endif ;
     z       = trimr(z,k,0);
     dep     = trimr(dep,k,0);
     if ( p > -1) ;
        z = z~ptrend(p,rows(z));
     endif ;
     b       = dep/z ;
     res     = dep - z*b ;
     so      = (res'res)/(rows(dep)-cols(z));
     var_cov = so*inv(z'z);
     b[1,1] = b[1,1] - 1;
     tval    = b ./ sqrt(diag(var_cov));
retp(b,tval,b[1,1],(b[1,1])/sqrt(var_cov[1,1])) ;
endp ;

proc lagn(x,n);
    local y;
    y = shiftr(x', n, (miss(0, 0))');
    retp(y');
endp;

proc diff(x,k) ;
     if ( k == 0) ;
        retp(x) ;
     endif ;
     retp(trimr(x,k,0)-trimr(lagn(x,k),k,0)) ;
endp ;

zhe shi wang shang de ,dan shi shi cuo de

运行程序，GAUSS 提示，undefined symobls
ptrend        d:\gauss9.0\adftest.g(112)