Predicción del PBI real nacional trimestral: redes neuronales autorregresivas vs metodología Box-Jenkins

• Autor: Angulo Elorreaga, Luis Alfredo
• Formato: tesis de grado
• Fecha de Publicación: 2013
• Institución: Universidad Nacional de Trujillo
• Repositorio: UNITRU-Tesis
• Lenguaje: español
• OAI Identifier: oai:dspace.unitru.edu.pe:20.500.14414/8660
• Enlace del recurso: https://hdl.handle.net/20.500.14414/8660
• Nivel de acceso: acceso abierto
• Materia: Redes Neuronales, Metodología de Box- Jenkins, Autorregresivas,, PBI trimestral

In this paper contains a comparative approach between artificial intelligence_x000D_ techniques called neural networks, specifically Neural Networks autoregressive (ARNN)_x000D_ versus the Box-Jenkins methodology (ARIMA), in modeling macroeconomic_x000D_ series Gross Domestic Product (GDP) in quarterly periods of our country. We_x000D_ worked with a total of 74 data GDP quarterly national real based on prices in 1994,_x000D_ including from the 1st quarter of 1994 to the 2nd quarter of 2012 which were_x000D_ obtained from the official website of the National Institute of Statistics and_x000D_ Information. This will build the best model for quarterly GDP forecast both the Box-_x000D_ Jenkins methodology with neural networks as autoregressive and performed the_x000D_ comparison between these two models. The best model with autoregressive neural_x000D_ networks was the AR-NN model 5, whose structure is given by the second and fourth_x000D_ lags as input variables and two nodes in the hidden layer, and the best model with the_x000D_ Box-Jenkins methodology is the model SARIMA (2,1,2) x (2,1,2) 4 with constant._x000D_ The comparison between these models was performed using the model selection_x000D_ criteria such as: the adjusted determination coefficient ( ̅ ) for the calibration period,_x000D_ the Akaike Information Criterion (AIC) and the Schwarz Information Criterion_x000D_ (SIC), both for the calibration period as predictors. It was found that the model_x000D_ obtained by the Box-Jenkins methodology has better fit both the calibration period as_x000D_ in the prediction, thus gives best quarterly GDP predictions that the model obtained_x000D_ by autoregressive neural networks