Longdata
Assignemnt on integration and cointegration
Due Wed., Oct 8
This is a pretty easy exercise, only involving data.
0. Run spurious.do (on web) and satisfy yourself that the generated x
and y appear related by regression, but are not cointegrated (the
dfuller test on the residuals). Make sure to set the seed so we do
not all get the same results.
1. Using either eviews or stata, and either british or american data,
find one series that each might be integrated (I suggest you use
some economic variable that is a level, not a change). Test for it
being I(1), that is, it has a unit root but its first difference
does not. Do this using the various tests in stata and/or
eviews (all variants of correct tests). Do the various tests differ
very much?
2. Find another economic variable that also might be I(1) and test as
in q. 1. You need not use the entire battery of tests.
3. Test for whther the two series are cointegrated.
4. IF they are, fit a one way error correction model and try the Engle
Granger two step method.