Patrick J. Coe and James M. Nason, "Long-run Monetary Neutrality and Long-Horizon Regressions", Journal of Applied Econometrics, Vol. 19, No. 3, 2004, pp. 355-373. This file contains details of an appendix to the paper, data files, and program files. Email enquiries to pcoe@ucalgary.ca APPENDIX: lrmnapp.pdf The appendix describes our data, presents background results for the paper, outlines the construction of the standard error necessary to calculate the VAR-long-horizon regression test of LRMN, presents our Monte Carlo bootstrap methods, and reports Monte Carlo results not included in the paper. DATA FILES: au.mat, cn.mat, uk.mat, us.mat The data come in 4 separate ASCII files. Each file contains a T*3 matrix where T is the sample size, which differs across countries. In each case the first column contains the monetary base (b), the second column contains the monetary stock (m) and the third contains real ouptut (y). All variables are in natural logarithms. The samples are for Australia, 1900/01 to 1993/94 (T=94), Canada 1872 to 1994 (T=123), the United Kingdom 1871-1993 (T=123) and for the United States 1869-1997 (T=129). The appendix describes the data sources. These files, which are in DOS format, are zipped in the file coenason-data.zip. Unix users should use "unzip -a". PROGRAM FILES: These are Gauss (version 4.0) programs that were run on a Gateway PC. They typically require the GAUSS packages MAXLIK & OPTMUM. In general the user needs to set the economy, econ = 1 (AU), 2 (CN), 3 (UK) or 4 (US), the monetary aggregate, magg = 1 (b) or 2 (m)and whether to pre-whiten the HAC estimator (pw=1). These files, which are in DOS format, are zipped in the file coenason-progs.zip. lrmn1.prg This file does the following: 1) estimates beta_OLS,k and beta_VAR,k for k=1,2,...,30 using sample data. 2) performs bootstrap experiment under the null. 3) performs bootstrap experiment under the sample alternative. Output is written to lrmn1??#.out where ?? = au, cn, uk or us and # = b or m. The output in these files produces tables 1 to 6 of the paper and tables A10 and A11 of the appendix. This program also produces a file ecv95??#.mat which contain a 30*2 matrix of empirical 95% critical values. The first column is for beta_OLS and the second for beta_VAR. The rows begin with k=1 and end with k=30. These empirical critical values (for beta_OLS) are used to produce figures 1-4 of the paper. The final output from this program is a file lrmn1??#.mat. This file contains a 10000*8 matrix. The first column contains the estimates of beta_OLS,30 from each repitition of the bootstrap under the null and the second column contains the estimates of beta_VAR,30 from the same bootstrap experiment. The third and fourth columns contain the estimates of the numerator and denominator of beta_VAR,30 at each repetition respectively. Columns five through eight contain the same four objects, but under the alternative. Columns 3 and 7 of these files are used to produce figure 5 of the paper (in the case of the money stock) and figure 1a of the appendix (in the case of the monetary base). This program calls upon the following subroutines: fslrd.g, hodlrd.g, svarmo.g, svarivj.g fslrd.g & hodlrd.g also call upon bartwin.g, bartpw.g, grdbetak.g, lrmn2.prg As lrmn1.prg but HAC estimator is pre-whitened. Output files are lrmn2??#.out. The output in these files produces tables A3 to A8 of the appendix. This program requires the same sub-routines as lrmn1.prg lrmn3.prg This program calculates size adjusted power against a proportional non-neutrality. That is the LR response of output to money is set at 1. This is done for both the OLS and VAR estimators. Output files are lrmn3??#.out and are used to produce table A12 of the appendix. This program requires the same sub-routines as lrmn1.prg (except svarivj.g). It also requires the files containing empirical critical values, evc95??#.mat. lrmn4.prg This program calculates the size adjusted power of the re-scaled t-ratio proposed by Valkanov (2003). 1) Calculates t/T statistics using sample data. 2) Generates empirical distribution of t/T under null. 3) Calculates size adjusted power of t/T statistics. Output files are lrmn4??#.out and are used to produce table A9 of the appendix. This program requires the same sub-routines as lrmn1.prg. lrmn5.prg This program performs bootstrap experiments that correct for bias associated with LRPS. 1) Estimate SVAR(1) under LRPS & calculate bias. 2) Performs bootstrap experiment under null, having subtracted bias. 3) performs bootstrap experiment under sample alternative, having subtracted bias. Output files are lrmn5??#.out and are used to produces tables 7-10 of the paper. This program requires the same sub-routines as lrmn1.prg, except it uses fslrdps.g (hodlrdps.g) instead of fslrd.g (hodlrd.g). lrmn6.prg The programs calculates the size and size-adjusted power of the bias adjusted VAR estimator. 1) Bias calculated as average of the cov term under the null across 10000 bootstrap samples. 2) Performs bootstrap experiment under null, having subtracted bias. 3) performs bootstrap experiment under sample alternative, having subtracted bias. Output files are lrmn6??#.out and are used to produces tables 11-12 of the paper. This program requires the same sub-routines as lrmn1.prg, except it uses hodlrdba.g instead of hodlrd.g and doesn't use fslrd.g. fig1to4.prg Produces figures 1 to 4 of the paper. Requires data files, ecv95??#.mat files & fslrd.g fig5.prg Produces figure 5 of the paper. Requires lrmn1??m.mat, ndens.g and weightn.g. fig5.prg Produces figure A1 of the appendix. Requires lrmn1??b.mat, ndens.g and weightn.g. Subroutines: fslrd.g: calculates long-run derivatives using OLS methodology of Fisher & Seater. hodlrd.g: calculates long-run derivatives using VAR estimator of Hodrick. bartwin.g: calculates optimal weighting matrix using the Newey & West correction. bartpw.g: as bartwin.g but with pre-whitening. grdbetak.g: computes gradient of VAR estimator at horizon k fslrdps.g: as fslrd.g but with "price stability" bias adjustment. hodlrdps.g: as hodlrd.g but with "price stability" bias adjustment. hodlrdba.g: as hodlrd.g but with bias adjustment in covariance matrix. svarmo.g: returns the likelihood function of the over-identified SVAR(1). svarivj.g: estimates just identified SVAR(1) by IV. ndens.g: normal density estimator. weightn.g: computes weights for normal kernal. ***** Addendum (2006-12-06) We are grateful to Professor Gary Shelley of East Tennessee State University for pointing out two typos in our original computer code. These typos pertained to the construction of the residuals and to the updating of the lagged values of output and money in our bootstrap experiments. While checking our code, we also found one more typo in lrmn5.prg whereby the long-run response of money to output was set at zero rather than one. These errors have all been corrected. It is worth noting that our original conclusions remain. That is, the long-horizon regression test of long-run monetary neutrality suffers serious size distortion and has power that approximates size.