*
* Example of correlation graphs on pages 217-219,
* 223-227.
*
open data statest.asc
data(format=prn,org=columns) 1 200 y1 y2 y3 y4 y5
*
* These use the procedure BJIDENT. The DIFFS option indicates the number of
* differencings you want to examine, so the second and third of these
* automatically do both y2 and its first difference and y3 and its first
* difference.
*
* The option METHOD=YULE uses the same method of calculation as in the examples
* in the text. (The default is called METHOD=BURG).
*
* You'll notice that the two standard error bands around zero are flat in the
* graphs in the text, but expand with increasing lag in the RATS graphs. This is
* because the RATS standard error bands are computed treating the
* autocorrelations at earlier lags as given at their estimated levels. The flat
* standard error are appropriate under the assumption that the series is actually
* white noise. As a result, the bands are much flatter when the estimated
* correlations are small, so the white noise assumption is nearer to being
* correct.
*
@bjident(method=yule,number=18) y1 101 200
@bjident(method=yule,diffs=1,number=18) y2 101 200
@bjident(method=yule,diffs=1,number=18) y3 101 200
*
* Graph of trend and difference stationary series on page 221
*
set y4hat = 10+.5*t
*
graph(footer="Figure 7.3 Trend Stationary and Difference Stationary Series",key=upleft) 3
# y4
# y4hat
# y5
*
* BJIDENT applied to series 4 and 5. Because we suspect non-stationarity we
* include the diffs=1 option to do the levels and first difference.
*
@bjident(method=yule,diffs=1,number=18) y4 101 200
@bjident(method=yule,diffs=1,number=18) y5 101 200
*
* Unit root tests. These are done with the DFUNIT procedure The main options are
* LAGS=# of extra lags (here zero), and TREND for including a trend in addition
* to the intercept. The range on the DFUNIT is the range of data to use. Since we
* use one lag, in order to run the test over 101 to 200, we tell it the range is
* 100 to 200.
*
@dfunit y1 100 200
@dfunit(det=trend) y4 100 200