Marcin Blazejowski, Pawel Kufel, and Jacek Kwiatkowski, "Model
Simplification and Variable Selection: A Replication of the UK
Inflation Model by Hendry (2001)", Journal of Applied Econometrics,
Vo. 35, No. 5, 2020, pp. 645-652.


1. General information

Original series are freely available in the Journal of Applied
Econometrics Data Archive at
http://qed.econ.queensu.ca/jae/2001-v16.3/hendry. We combined three
ASCII files provided by David F. Hendry into one file with both
original series and dummy variables used in our replication analysis.

All files are in ASCII standard (DOS format) and stored in the archive
bkk-files.zip. Data, gretl BACE package and replication code are
organized in the following directories:

dataset:

  UK_Inflation_1875-1991.csv - plain ASCII file in CSV format with
  variables organized in columns

  UK_Inflation_1875-1991.dat - OxMetrics ASCII file 
  
  UK_Inflation_1875-1991.gdt - gretl native data format (ASCII XML file)
  
lib:
  
  BACE.gfn    - BACE package for gretl (ASCII XML file)
  
  library.inp - ASCII file with basic function definition for
  replication in a narrow sense (included by
  full_gretl_replication.inp)

results:
  
  autometrics.out -- Autometrics replication in a narrow sense

  gretl.out -- gretl replication in a narrow sense together with BACE
               results for 'GUMFinal' and 'GUMId'
  
  gretl_BACE_GUM52.out -- BACE results for 'GUM52'
  
  gretl_BACE_GUM52-MPI.out -- BACE results for 'GUM52' obtained in parallel
               computation via MPI on 20 CPUs (not reported in paper)

 
2. Data definitions (in LaTeX like notation)

Y     - real GDP, £ million, 1985 prices
P     - implicit deflator of GDP (1985=1)
M     - nominal broad money, £ million
R_s   - three-month treasury bill rate, fraction p.a.
R_l   - long term bond interest rate, fraction p.a.
R_n   - opportunity-cost of money measure
N     - nominal National Debt, £ million
U     - unemployment
Wpop  - working population
U_r   - unemployment rate, fraction
L     - employment
K     - Gross capital stock
W     - wages
H     - normal hours (from 1920)
P_e   - world prices, (1985=1)
E     - annual-average effective exchange rate
P_nni - deflator of net national income, (1985=1)
P_cpi - consumer price index, (1985=1)
P_o$  - commodity price index, $
m^d   - money excess demand
y^d   - GDP excess demand
s_t   - short--long spread
n^d   - excess demand for debt
e_r   - real exchange rate
pi^*  - profit markup
U^d   - excess demand for labour
p_o   - commodity prices in Sterling
Ixxxx - dummy variable with 1 for any year xxxx
Iwar  - dummy variable for wars and their after-effects with 1 in years:
        1900-1901, 1914-1922, 1939-1946
Ioil  - dummy variable for breakdown of Bretton Woods and two oil crises with
        1 in years: 1970-1971, 1973-1975, 1979-1980
I1    - dummy variable for years < 1945
I2    - dummy variable for years > 1945
Ib    - dummy variable for very big outliers with 1 in years: 1915, 1917,
        1919 and -1 in years: 1921-1922
Il    - dummy variable for large outliers with 1 in years: 1916, 1918, 1920,
        1975 and -1 in years: 1943, 1945, 1973
Im    - dummy variable for medium outliers with 1 in years: 1880, 1900,
        1939-1940, 1970-1971, 1980 and -1 in years: 1881, 1942, 1944


3. Parameter files

autometrics.fl -- OxMetrics batch file with definition of variables
  transformations and autometrics analysis at 1% significance

full_gretl_replication.inp -- gretl script with definition of
  variables transformations along with code for replication in a narrow
  sense and BACE

bace_options.inp -- gretl script with parameters for BACE analysis:
     y_lags = 1 # we have d_p(-1)
     with_const = 2 # intercept can be dropped
     model_prior = 1 # binomial distribution
     prior_avg_model_size = {formula} # this formula makes all possible
       specifications equally probable (uniform model prior)
     alpha = 0.6 # significance level for the initial model
     top_ranked_model = 10 # we print detailed info for 10 best models
     joint = 0 # jointness analysis disabled
     forecasts = 0
     Nrep = 1000000 # total number of MC3 iterations
     PercentOfburnin = 25 # percentage of burn-in draws
     MPI_threads = 1 # we use just 1 CPU
     verbosity = 2 # we are verbose
     progress = 0 # we don't print progress (it's not costless)
     memory = 3 # we assume machine has >= 10 GB RAM; if machine has less
       than 10 GB RAM, memory should be set to 2 or 1


4. Execution

In order to replicate Autometrics results just type in system's
console 'oxmetrics autometrics.fl' in UNIX-like systems or
'oxmetrics.exe autometrics.fl' in MS Windows. Results will be stored
in file 'autometrics.out'.

In order to replicate the whole gretl analysis, i.e. replication in a
narrow sense and BACE^* in three variants, just type in system's
console 'gretlcli -e -b full_gretl_replication.inp >
full_gretl_results.out' in UNIX-like systems or 'gretlcli.exe -e -b
full_gretl_replication.inp > full_gretl_results.out'. Results will be
stored in file 'full_gretl_results.out'.

^* Note, that default value for MPI_threads is 1, which means that
BACE analysis will be executed on 1 CPU. Running BACE package in
parallel through Message Passing Interface (MPI) reguires installation
of related libraries in your operating system (see 'Gretl + MPI'
manual available at
http://ricardo.ecn.wfu.edu/pub/gretl/manual/PDF/gretl-mpi-a4.pdf).


Marcin Błażejowski, Paweł Kufel, Jacek Kwiatkowski
email: marcin.blazejowski [AT] wsb.torun.pl