Real/Dollar Exchange Rate Forecast for the Brazilian Economy Using Predictors Macroeconomics and Supervised Machine Learning.
Exchange Rate; Supervised Machine Learning; Macroeconomic Fundamentals; Kitchen-Sink Regression.
In order to utilize recursively supervised machine learning techniques via Ridge, LASSO, and Elastic Net to enhance Kitchen-Sink regression in predicting the exchange rate between the Brazilian Real (BRL) and the United States Dollar (USD), a combination of predictors of macroeconomic fundamentals including uncovered interest rate parity,
purchasing power parity, monetary fundamentals, Taylor rule, and terms of trade was employed to examine their explanatory power relative to the benchmark model (Random Walk). Only out-of-sample predictive power was examined to ascertain the robustness of the approach, between 2013:01 to 2022:12. The findings indicate that the efficient Kitchen-Sink model is unable to predict the exchange rate statistically/economically. On the other hand, only the UIP predictor demonstrated a good long-term performance, surpassing the benchmark, and furthering utility gains for risk-averse investors.