Economics 470: Economic Fluctuations and Forecasting
Homework #6: Answers
This is due on Tuesday, March 12th.
1.Read Thurman and Fisher’s “Chicken, Eggs, and Causality, or Which Came First?” and be prepared to discuss this on March 12th. This paper has been posted on my website and is based upon a technique known as Granger Causality that is described in chapter 11 of your text (outlined especially in 11.7).
2.Online I have placed a data set entitled, “CPI and PPI.” This data consists of observations of the American Consumer Price Index and Producer Price Index. A description of the PPI may be found on: while a description of the CPI may be found at: HINT: Check for unit roots in each series and adjust accordingly prior to doing part a and b.
a. Economists have argued that the PPI, by virtue of its being a measure of the cost of production, might be a good predictor (or cause) of CPI inflation. Perform two Granger Causality tests. Does the PPI “cause” the CPI? Does the CPI “cause” the PPI? What does it mean to “cause.” What macroeconomic implications do you learn from these two Granger Causality tests?
I take the logs of PPI and CPI first.
Both PPI and CPI have unit roots so I take first differences and then estimate a number of VARs to discover what the correct number of lags are. For the answer key, I use a VAR of 12 lags (though this neither minimizes the AIC nor the SIC).
In this case dlncpi Granger causes dlnppi and dlnppi Granger causes dlncpi.
Granger causality implies that past values of one variable impact the current variable even after holding constant past values of the current variable.
b. Compute impulse response functions for PPI and CPI (note, I’m having you figure out how many lags to include in each). What is the impact of a change in the PPI on CPI? What is the impact of a change in the CPI on PPI?
I find:
A positive shock to dlncpi causes positive, future increases in dlncpi that die out slowly over time. A similar thing happens with dlnppi. A positive shock to dlnppi (lower left corner) appears to cause a small, positive shock to dlncpi, though this dies out quickly. The top right diagram is hard to explain; it appears that a positive shock to dlncpi raises dlnppi, somewhat erratically and then causes it to decline and then die out.
3.If the data you are using in your final project involves multiple variables, provide Granger Causality tests of those variables. If your data does not involve multiple variables, provide a correlogramof your preferred forecasting model for the data you are working on for your final project.