Real-Time Nowcasting of Economic Growth. Case study of Romanian GDP

This paper investigates the usefulness of high-frequency economic indicators for forecasting Romanian GDP growth within a mixed-frequency framework. Using quarterly GDP data together with monthly indicators related to economic activity and sentiment, the study compares three forecasting methods: a traditional bridge equation, a MIDAS regression model, and a machine learning approach based on LASSO regularization. The results show that the MIDAS model is better at using information available within the quarter, leading to slightly more accurate forecasts than the bridge model. At the same time, the LASSO approach helps reduce problems such as overfitting and multicollinearity, creating a simpler and more stable model that filters out short-term noise. Overall, the findings suggest that including high-frequency indicators can improve the macroeconomic forecasting process.