Econ 427, Spring 2010

Problem Set 5 -- Due Thursday, April 8, 2010

Ch 11. Exercises, Problems and Complements:

11. (page 254) [Hints: The data for housing starts, completions is in the Diebold data sets.]

Extra application problem:

Under Supplements, you will find a data file, capital.dat, with data for quarterly seasonally-adjusted capital expenditures and appropriations for U.S. manufacturing firms between 1953 and 1974. Load this data into Eviews for use in this problem. [Hint: You will be forecasting beyond the sample period, so define your new workbank to run through at least 1977:4.]

a. Because the raw data are non-stationary in levels, we will work with then in differenced logarithms (remember these are then approximately growth rates). Use series commands and the dlog() function to create the appropriate transformed series.

b. Create a cross correlogram for the two differenced log series [Hint: create a "group" in EViews consisting of the two differenced log series. Open the group and from the View menu choose "cross correlation"] What do the graphs tell us?

c. Conduct Granger causality tests for the two differenced log series [Hint: open your group and under View choose "Granger causality"] What do the tests indicate? Are they sensitive to the number of lags included?

d. Set up a VAR model for the two differenced log series. [Hint: Under Quick, choose "Estimate VAR"] Select an appropriate lag order for your VAR by estimating a range of VARS with 1 to 8 lags and tabulating AIC and SIC values. [You set the lag order in the "lag intervals for endogenous" box. Entering "1 4" would indicate a 4th order VAR. You want the summary AIC and SIC values at the very bottom of the regression output. Set your estimation sample for 1955:2 1974:4 to allow for lags.] Comment on the economic meaning of the coefficient estimates for your selected VAR.

e. Generate impulse responses [Hint: click "Impulse" button] and interpret.

f. Produce a forecast for the period 1975:1 - 1977:4 for the two differenced log series. [Hint: you will need to set up a model from the two equations of the VAR. Select "Make Model" from the "Proc" menu in the VAR window.]