# ECON 5420 Lecture Notes - Lecture 6: Federal Reserve Economic Data, Prediction Interval

10 views1 pages

Published on 5 Nov 2018

Department

Economics

Course

ECON 5420

Professor

HOMEWORK 6 Econ 5420, Fall 2017

This homework requires the use of the data set forecast on Carmen. The data includes

the following variables from the Federal Reserve Economic Data base:

construction Seasonally adjusted construction Spending

manufacturing s2i Ratio of sales to inventory in Manufacturing

unemp Unemployment Rate

tbill3mo 3 Month Treasury Rate

month Month

year Year

t Created to set time

1. We want to estimate Unemployment as a function of each of the other variables

(not month, year, t). Begin by looking for obvious time trends and ﬁrst diﬀerence

any variable with a time trend. Which variable(s) has a time trend? (There

should be no seasonal trends. The data is seasonally adjusted. )

2. Conduct Dickey-Fuller Tests on your variables. Which variables have unit roots?

Report a DF stat for each. Take ﬁrst diﬀerences on those with unit roots.

3. Run a regression with your detrended, stationary variables only. Report the

results.

4. Use lagged values1of the variables used in number 3 to create a forecast for

unemployment in December of 2016. Find a conﬁdence interval for that prediction.

(To ﬁnd the values that you will use in the regression look in at the data editor.

August 2016 is the last month. ) Report the regression results.

5. Now create a forecast using the level values despite the fact that they contain

trends and unit roots. Add lagged values of each variable until the last lag is

no longer statistically signiﬁcant, including lags of unemployment. Repeat the

exercise of ﬁnding a prediction and conﬁdence intervals. Write out the forecasting

equation, report the regression results, and show the prediction interval. The true

value is 4.9%. How close was your estimate?(Note: The answer that you get will

depend on which variables you add in which order. All reasonable answers will be

accepted. You can even add lagged errors and time trends if they are signiﬁcant.)

1The values must be lagged because we are predicting outside our sample. We do not know what

any variables are in December 2016.