235B / Professor Òscar Jordà
An Atheoretical Monetary System
This exercise is fundamentally empirical and is designed to compare the information that is typically extracted from a vector time series of macro-economic data across different techniques.
Data
The data for this exercise is collected in an EViews database and corresponds to the data used in Christiano, Eichenbaum, and Evans (1996), Evans and Marshall (1998), Hoover and Jordà (2001) and Hamilton and Jordà (2002), to name a few. In particular, it consists of 6 variables: EM denoted the logarithm of non-agricultural payroll employment; P denotes the logarithm of the personal consumption expenditures deflator (1996 = 100); PCOM denotes the annual growth rate of the index of sensitive commodity prices issued by the Conference Board; FF denotes the Federal funds rate; NBRX denotes the ratio of non-borrowed reserves plus extended credit to lagged total reserves; and M2 denotes the annual growth rate of the monetary aggregate M2.
Data Transformations
The first involves transforming the data available in the database (which comes in raw form) to match the definitions that I have just given you.
Evans and Marshall VAR
To get started, let’s try to match the impulse response functions in the Evans and Marshall (1998) VAR. This will allow you to check whether you transformed the data correctly. Make sure to select the same sample period, same lag length, and the same Cholesky ordering (essentially that in which I reported the variables). Note that they perform a number of experiments. I am just looking for the basic VAR. Once you have checked everything is OK, expand the sample back to its original size (i.e., including all available data).
Structural Identification with Cholesly Orderings
The first order of business is to get a sense about the structural identification assumptions. You may begin by checking the variance-covariance matrix of the residuals of the reduced-form VAR. It would be advisable to normalize these residuals to have standard deviation 1 and then look at the structure of the matrix: are the residuals strongly correlated or are they independent?
Next, considering that FF, NBRX, M2 can be reasonable interpreted as the monetary block of the economy, experiment with ordering these three variables ahead of the real-economy block EM, P, PCOM. Are there any substantial differences between this set of impulse response functions and the ones you obtained with the Evans and Marshall (1998) order? In your view, why are there/are there not differences?
Other measures of covariation
Now that you have learned a little bit about the type of dynamic behavior implied by a typical VAR, it is time to evaluate what kind of information you learn from other sorts of measures.
Either in EViews or (more efficiently in GAUSS), compute the autocorrelations and partial autocorrelations for this system (note each will be a 6 by 6 table of graphs). If you write the code in GAUSS, you will be in a position to use it at your leisure in your research at a later date. Compare these graphs with the impulse response functions you calculated above and comment on any differences and similarities. What kind of statements are you willing to make given this approach and how does it differ from the usual VAR type of statement?
Comments
Keep these results available. We will be doing further experimentation on this data set.
1