Longdata
Exercise 2 - univariate analysis
Due Mon., Sept. 22
1. Use pgmarma22.prg to generate simulate data (program is next on web
site).
Generate 200 obs of ar(2), ma(2) and arma(2,2) data. Data should be
stationary - graph to make sure non-explosive and looks stationary, do
not do any testing. The exercise is easier if you use relatively large
values of the four parameters (
a. For each, what should the ACF and PACF look like?
b. Using eviews, for each series, what do the ACF and PACF look like.
c. Estimate the true model (using quick, estimate equation, then
equation is something like "y c y(-1) y(-2) ma(1) ma(2)" for
arma(2,2) and obvious changes for smaller models
d. How do the results compare to the true parms
e. Test the residuals to make sure they appear uncorrelated
f. Now mistakenly take one of your models and do not estimate one term
(eg for AR(2) estimate an AR(1). Look at the correlogram of the
residuals and see if you can pick up the error.
2. Use a series from either prespop.wf1 (US pres popularity and economic data) or brneal.wf1 (british
pop data and economics)
Using the various methods, identify the order of the ARMA process and then estimate the model and then
check residuals to make sure are IID. (Note: series should be stationary, so if you end up picking on that looks
non-stationary, try a first difference)