GDP nowcasting with machine learning and unstructured data

Nowcasting models based on machine learning (ML) algorithms deliver a noteworthy advantage for decision-making in the public and private sectors due to their flexibility and ability to handle large amounts of data. This article introduces real-time forecasting models for the monthly Peruvian GDP growth rate. These models merge structured macroeconomic indicators with high-frequency unstructured sentiment variables. The analysis spans January 2007 to May 2023, encompassing a set of 91 leading economic indicators. Six ML algorithms were evaluated to identify the most effective predictors for each model. The findings underscore the remarkable capability of ML models to yield more precise and foresighted predictions compared to conventional time series models. Notably, the gradient boosting machine, LASSO, and elastic net models emerged as standout performers, achieving a reduction in prediction errors of 20% to 25% compared to autoregression and various specifications of dynamic factor model. These results could be influenced by the analysis period, which includes crisis events featuring high uncertainty, where ML models with unstructured data improve significance.