*Example 10.5
READ (DATA9-2) Year r M D
* Replicate Figure 10.7
GRAPH M YEAR /LINEONLY
* Generate the variables in Step 1 of the example
GENR Y=LOG(M)
GENR Y1=LAG(Y)
GENR DY=Y-Y1
GENR DY1=LAG(DY)
* Step 2-3
* Generate a time trend t.
GENR T=TIME(0)
* The first observation on Y1 is not known, since it is a lagged variable.
* DY depends on Y1, so the first observation on DY is not known either.
* Since DY1 is a lagged DY, the first TWO observations are not known.
* Suppress the first two observations.
SAMPLE 3 36
* Estimate the unrestricted model.
OLS DY T Y1 DY1
* The output reports the T RATIOs for all coefficients, including Y1
* Step 4
* Save the ESS in the variable ESSU, for use in calculating the Fc statistic.
* Use the temporary variable $sse.
GEN1 ESSU=$SSE
* Also save n-k in the variable df. Use the temporary variable $df.
gen1 df=$df
* Estimate the restricted model.
OLS DY DY1
* Calculate the value of Fc, using the formula at the bottom of p. 458.
* The ESS of the restricted model is in the temporary variable $sse.
GEN1 F=($SSE-ESSU)/2/(ESSU/DF)
PRINT F
*
* Alternatively, the Dickey-Fuller tests can be generated automatically
* using the COINT command.
* COINT automatically adjusts the sample for the appropriate lags.
SAMPLE 1 36
* We can specify the number of lagged dependent variables DY by the NLAG option.
COINT Y /NLAG=1
* The appropriate specification is CONSTANT TREND (since there is a
* time trend in the model).
* The first test statistic reported here is the t-test.
* The third test statistic is the F-test on the null hypotheis that the coefficient
* for both t and Y1 is zero.
* The critical values reported in COINT are asymptotic critical values, while those
* in the text are small-sample critical values for n=25 and n=50 respectvely. As a 
* result, these critical values are not the same.
STOP