ABSTRACT
Machine learning is a popular branch of data science and includes various algorithms that all aim to learn the links and associations between the variables along with hidden patterns in a dataset. The primary aim of machine learning is not only to learn, though. Learning is assumed to be a success only if its results are generalizable over other datasets. In that sense, the aims and ideas behind machine learning are not so different from those behind the statistical or econometric models. Both the statisticians/econometricians and data scientists try to i) explain the data, and ii) predict the future. That said, there is a fundamental difference between the two areas, which is highly important for the exchange rate literature. Machine learning models allow for a trade-off between bias and variance in order to control the overfitting problem, while unbiased coefficient estimation continues to be an overemphasized target in statistics/econometrics. That is why old school statistics might have the upper hand in explaining the past, but machine learning models are likely to have better chance when it comes to predicting the future.
