“Should Economic Theories Guide the Machine Learning Model in Forecasting Exchange Rate?

This archive contains all code, data, and documentation necessary to fully replicate the empirical results presented in the paper, including all tables and figures featured in the main text and the appendix.
The structure of the replication folder is organized as follows:
1. Code - Contains all Python scripts required to perform the empirical analyses, and generate every table and figure reported in the main text. - Each code cell within the notebooks is sequentially labeled A–Z, followed by AA, BA, CA, and so on up to FA. This labeling scheme enables unambiguous cross-referencing between individual code outputs and the corresponding tables and figures.
2. Excel File - Raw/: Contains the original monthly data for the ten sample currency pairs, spanning January 1999 to July 2022. The file includes bilateral exchange rates and monthly observations for all explanatory variables, as downloaded from publicly available sources. - Processed/: Contains processed data files generated by scripts in the Code/ folder. These files contain, among other items, the MSPE-adjusted statistics required for the Clark-West (CW) tests and the subsequent OLS regressions. They are also used to produce results such as the CDSE plots and the cumulative-return series for the forecast-driven trading strategy described in the main text.
3. Directions for Tables and Figures - Each table and figure is accompanied by detailed instructions explaining how the corresponding empirical result reported in the main text is generated—either through Python code or Excel-based computations. The instructions clearly specify the procedure, methodology, and step-by-step processes required for full replication.