Homework 2
1. The file [Link] contains a number of series including the US index of industrial production
(inprod), unemployment (urate), and producer price index for finished goods (finished). All of the series
run from 1960Q1 to 2008Q1.
a. Exercises with indprod
i. Construct the growth rate of the series as yt=log(indprodt)-log(indprodt-1). Since the first
few autocorrelations suggest an AR(1), estimate an AR(1) model and comment your results.
First of all, I create the growth rate series and I name it y1=log(indprodt)-log(indprodt-1).
Then, I estimate the equation y1 ar(1).
The p value is 0, which means an estimate AR(1) may be good, but we need to check the residuals too. For
this, we use the correlogram.
But we see that p value is almost 0, which is significant, and there is remaining information on the residue
we have not subtracted. AR1 then is not sufficient as it does not behave like white noise.
ii. Show that adding an MA term at lag 8 improves the fit and removes the serial correlation.
b. Exercise with urate
i. Examine the ACF of the series. It should be clear that the ACF decays slowly and that the
PACF has a very negative spike at lag 2.
ii. Temporarily ignore the necessity of differencing the series. Estimate urate as an AR(2)
process including a constant. Comment on your results, and pay particular attention to a1
and a2 coefficients. Any thoughts on the sum of a1 and a2.
iii. Form the difference as yt=uratet-uratet-1. Show that the model
yt=a0+a1yt-1+β4εt-4+β8εt-8+εt exhibits no remaining serial correlation and has a better
fit than yt=a0+a1yt-1+a4yt-4+ β8εt-8+εt or yt=a0+ a1yt-1+a4yt-4+a8yt-8+ εt
Autocorrelation in residuals is less significant in the last two
c. Exercise with finished
i. Construct the growth rate of the series as yt=log(finishedt)-log(finishedt-1).
ii. Show that an ARMA(1,1) model fits the data well but that the residual correlation at lag 3
is problematic.
The residual correlation in lag three has a value smaller than the p-value.
iii. Compare the performance of ARMA((1,3),1) and ARMA(1,(1,3)).
The ARMA(1, (1,3)) adjusts better the data because the residuals show no autocorrelation.
iv. Try some other models that you think is appropriate. Write down the model that you think
fits the data best (you are not limited to the above models.).
I will try ARMA (1,(1,8)) because there is a lag in 8.