Erik Hjalmarsson and Tamas Kiss, "Long-Run Predictability Tests Are Even 
Worse Than You Thought", Journal of Applied Econometrics, Vol. 37, No. 
7, 2022, pp. 1334-1355.

***** Data ***** 

Our empirical analysis uses monthly international equity returns, in
excess of a local risk free rate. The file "hk-files.zip" contains the
file "data.csv" which includes the original data. The column headers
include variable names as follows:

-- 'Country' is the total monthly equity return
-- 'Country'_rf is the risk free rate in the given country
-- 'SP500_ret' is the total return on the S&P 500 index
-- 'Rfree' is the annual riskfree rate in the US

'Country' takes the following values in the dataset:

Austria
Austrlia (Australia)
Belgium
Canada
Denmark
Finland
France
HongKong
Ireland
Italy
Nethrlnd (the Netherlands)
NewZland (New Zeland)
Norway
Singapor
Spain
Sweden
UK (the United Kingdom)

There are 1140 rows (observations) in the data file, but the number of
effective observations varies by country. 

Data were sourced from Kenneth French's Data Library (for equity
returns) and Macrobond (for the risk free rates). For further details,
please consult Section 6.1 in the paper.

***** Reproducing the empirical results *****

The empirical results presented in the paper (Table 5) can be
reproduced in three steps using the files in "hk-files.zip". 

1. Most of the regression results can be obtained via running
"empirical_analysis.do" in Stata (version 16 or later, the package
"outreg" should be available). Results will be displayed in the
Results window in Stata (please disregard the first two rows in the
Stata output). 

2. The 5% critical values can be obtained by using the "tstat_cv.m"
file in the folder "Matlab" via the following syntax:

  cv=tstat_cv(q,T,delta),

where 'cv' is the desired 5% critical value, 'q' is the horizon, 'T'
is the sample size and 'delta' is the correlation between u and v.

3. Bootstrapped critical values can be obtained by running
"bootstrap_calc_to_paper.m" in the folder "Matlab". Results will be
shown in the command window.  

Any deviations of the results from Table 5 are due to the random
nature of the bootstrap procedure (step 3).