In the presence of instabilities in economics, such as the global financial crisis in 2008 and the recent COVID-19 pandemic, we face challenges in forecasting macroeconomic and financial variables. Recently forecasters have turned their attention to Machine Learning (ML), which has become an important estimation and forecasting tool due to availability of “big data” in economics, and an increasing number of macroeconomic studies show successful forecasting performance applying ML methods. ML can be divided into three groups: supervised, unsupervised, and reinforcement learning. In particular, nonlinear supervised ML methods, such as random forest (RF) has gained attention in macroeconomic forecasting. Studies provide evidence that the properties of RF (nonlinearity and variable selection) play a key role in good performance of forecasts in non-normal times. Although the theoretical properties for most ML methods are not well known for time-series, there are numerous empirical evidence of successful macroeconomic forecasting performed by ML combined with large data.