Prognozowanie kursu walutowego z miarami globalnego ryzyka i zmiennymi makro
Exchange rate forecasting with measures of global risk and macroeconomic variables

Anna Sznajderska (Institute of Econometrics, Warsaw School of Economics)


We test out-of-sample forecasting performance of machine learning models applied to 
monthly data for G10 currencies against the US. For that purpose we integrate the 
direct forecast approach to deep NN framework in which fundamentals implied by 
economic theory are used in the role of predictors. We compare our forecasts to 
random walk and purchasing power parity panel reversion models, which were found to 
be competitive benchmarks in previous investigations. Our results indicate that the 
literature might overpromise on the usefulness of ML methods in FX forecasting, 
which confirms that the exchange rate disconnect puzzle remains puzzling. We also 
point that adding regularization techniques and interaction variables helps in 
bringing NN to challenging benchmarks.