Todd E. Clark and Michael W. McCracken, "Tests of Predictive Ability
for Vector Autoregressions Used for Conditional Forecasting," Journal
of Applied Econometrics, Vol. 32, No. 3, 2017, pp. 533-553.

This file describes the data and programs used to produce the
application results (from the 7-variable VAR) in the published paper.

The file cm-files.zip contains the following files:

    21439  CMdata.csv
    77677  CMdata.xlsx
    31232  updatedFREDdata_set_for_CM_paper.xls
    20767  VAR7.prg
      785  pctiles2s.src
     1249  regressfcn.src
     4162  VARcondfc.src
     2901  VARestloop.src
				     
The ASCII files (extensions .csv, .prg, and .src) are in DOS format.
The .xls and .xlsx files are binary.

Contact information:

todd.clark [AT] clev.frb.org,
Federal Reserve Bank of Cleveland, (216)579-2015

***********
DATA FILES:  
***********

The data needed to replicate our application results are provided in
CMdata.csv (comma-separated value text file) and CMdata.xlsx (Excel
file).  These files contain 252 quarterly observations on 9 variables,
for the 1947:Q1-2009:Q4 period, with rows corresponding to dates and
columns to variables.  The first row lists the variables. The first
column in each file provides the dates of the observation. Quarters 1,
2, 3, and 4 are labeled as months 1, 4, 7, and 10, respectively.

Data sources: For the analysis in the paper, we obtained all data from
the FAME database of the Federal Reserve Board, except for the
population series, which we obtained from the (publicly available)
Federal Reserve Board's FRB/US model database.  For those interested
in updating the data, we also provide an Excel file
updatedFREDdata_set_for_CM_paper.xls with the corresponding mnemonics
for obtaining updated data series (excluding the population series)
from the FRED database of the Federal Reserve Bank of St. Louis.

A few data details are worth noting.  (1) For some components of GDP,
including private fixed investment, the BEA only reports chain
dollar-based series back to 1999.  So for private fixed investment, we
used the chain-weighted quantity index.  In growth rates, the use of
the index instead of a series in chain-weighted dollars is immaterial.
(2) In the VAR we used real compensation defined as nominal
compensation deflated by the GDP price index.  (3) Regarding the data
available in FRED versus FAME, the consumption series we obtained from
FAME and used in our analysis included three decimal places.  However,
the consumption series available in FRED is reported to only one
decimal place.

**************
PROGRAM FILES:  
**************

cm-files.zip contains programs in RATS (all text files) needed to
produce the application results, reported in Tables 5 and 6 of the
paper.  The program files and their roles are as follows.

VAR7.prg: program that reads in and transforms the data and calls the
procedure files listed below to estimate the forecasting models,
produce the forecasts, compute the test statistics, and bootstrap the
test statistics to obtain p-values.  Note that results conditioned on
each different variable are obtained by changing the switch "condpos"
that appears at the top of the program.  Setting "condpos" to 1
produces results conditioned on the VAR's first variable.  Setting
"condpos" to 2 produces results conditioned on the VAR's second
variable.  And so on.

VARestloop.src: procedure file to loop over time (the forecast sample)
to produce time series of OLS estimates of the VAR

VARcondfc.src: procedure file to produce unconditional and conditional
forecasts from the VAR, using coefficients, error var-cov matrix, and
conditions fed in as inputs

regressfcn.src: function file for estimating test regressions

pctiles2s.src: function file to computing bootstrap percentiles
(two-sided inference)