Data set and programs to reproduce the forecasts "Fundamentals and exchange rate forecastability with simple machine learning methods"

Data set and programs to reproduce the forecasts in "Fundamentals and exchange rate forecastability with simple machine learning methods" by Amat, Michalski and Stoltz. There are three samples: 1973-2014: based on end-of-month (EOM) data from the International Monetary Fund (IMF); 1999-2017: based on end-of-month (EOM) data from the International Monetary Fund (IMF); real-time data for fundamentals 1973-2014: based on FRED averaged exchange rate data (from the St. Louis Federal Reserve). We present data that we used for all the forecasts. The files with the extension "uirp" are to be used for the UIRP model while those with an extension "mon" for monetary and all fundamentals together given the restrictions on data availability. There is one main file for real time data permitting estimation for the entire 1999-2017 period and others that are trimmed to correspond to the revised data that ends at some month in 2014 for comparison with revised fundamentals. Each data file has the following layout (25 columns): col 1: date (day / month / year) col 2: the exchange rate for that month (average or end-of-month) col 3: the no-change prediction for the next month s_hat_(t+1) = s_t where s_hat_(t+1) is the forecasted exchange rate and s_t is the current exchange rate. col 4-25: predictions formed by each of the fundamentals of the exchange rate for the next month of the form s_hat_(t+1) = s_t + f_t where f_t is the fundamental at time t. col 4-14: predictions with the fundamentals from one country in the currency pair col 15-25: predictions with the fundamentals from the other country [the order is the same as for the other currency] Forecasts formed with the fundamentals, listed only for the first currency: col 4: actual 1-month price level change col 5: 12-month price level change col 6: 6-month price level change col 7: actual 1-month money stock change col 8: 12-month money stock change col 9: 6-month money stock change col 10: money market interest rate col 11: change in the lagged money market interest rates col 12: actual 1-month output (industrial production) level change col 13: 12-month output level change col 14: 6-month output level change "NaN" means that the fundamental to form the forecast was unavailable. To generate our basic tables with results (for example Table 1), we used hence data from columns 5, 8, 10, 11 and 13 (for the first currency in the pair). Commented codes are to reproduce the results in Scilab.