Application of Deep Learning Methods in Time Series for the Forecast of the Macroeconomic Situation in Latin America

• Autores: Alegre Ibáñez, Víctor Augusto, Lozano Aparicio, Jose Martin
• Formato: artículo
• Fecha de Publicación: 2022
• Institución: Universidad de Lima
• Repositorio: Revistas - Universidad de Lima
• Lenguaje: español
• OAI Identifier: oai:revistas.ulima.edu.pe:article/5817
• Enlace del recurso: https://revistas.ulima.edu.pe/index.php/Interfases/article/view/5817
• Nivel de acceso: acceso abierto
• Materia: Deep Learning; GDP forecasting; CEPAL; Neural Network; Aprendizaje Profundo; Pronóstico de PBI; Redes Neuronales

Deep learning methods can be applied to generate predictive models. We worked with the gross domestic product (GDP) of six Latin American countries: Argentina, Brazil, Chile, Colombia, Mexico, and Peru, using annual and quarterly macroeconomic indicators from the World Bank and the Economic Commission for Latin America and the Caribbean (ECLAC), respectively. For the pre-processing of the data, we decomposed the quarterly series into trend, seasonality, and residual and used them as additional characteristics to provide more information to the models. In addition, outliers resulting from the impact of the COVID-19 pandemic on the world economy were replaced. Multilayer perceptron, convolutional neural networks, LSTM, GRU, and SeqToSeq models were built for each country and their series’ frequency, then evaluated by continuous cross-validation and MAE, RMSE, and MAPE metrics. The optimal models vary for each case.