Thesis PROPUESTAS PARA MEJORAR EL PRONÓSTICO DE LA INFLACIÓN EN PAÍSES LATINOAMERICANOS MEDIANTE EL USO DE MODELOS HÍBRIDOS QUE MEZCLAN "FUZZY INFERENCE SYSTEM", MODELO AUTORREGRESIVO DE MEDIA MÓVIL Y DESCOMPOSICIÓN DE WAVELET
Date
2017
Authors
PEIRANO FUENTEALBA, RODRIGO GIOVANNI
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
El pronóstico de la inflación es un tema importante y relevante para las economíasmundiales, ya que el análisis de esta variable macroeconómica puede derivar en decisionesimportantes por parte de los banco centrales y gobiernos de cada país. Pero aún cuandoes un tema relevante existe carencia de investigaciones, de nuevas metodologías respectoal pronóstico de esta variable, aún más si nos extendemos a investigaciones en paises endesarrollo.El presente estudio propone nuevos modelos híbridos, en primer lugar un FuzzyInference System - AutoRegressive Moving Average model (FIS-ARMA) y también unFuzzy Inference System-AutoRegressive Moving Average model, que utiliza variables deentrada las series entregadas por la técnica de descomposición de WAVELET. (FIS-ARMAWAVELET).Los resultados indican que existe un mejor desempeño en término de MeanSquare Error (MSE) de la combinación de estos modelos lineales y no lineales, sobre eluso de un modelo individual ARMA o sobre una red neuronal artificial con diferentesconfiguraciones.
Inflation forecasting has been and will continue to be a relevant issue for the world’seconomies, since dierent decisions and policies by central banks and governments of eachcountry arise from and are related to this macroeconomic variable. However, even given itsrelevance, there is limited research with new methodologies to forecast it, and even lessstudies in emerging economies. In fact, the most commonly used model in forecasting isthe Autoregressive Moving Average (ARMA) model.The present study proposes two new hybrid models, based on Fuzzy Inference Systems(FIS), to forecast inflation in four Latin American countries.The first model is a Fuzzy Inference System - Autoregressive Moving Average model(FIS-ARMA), while the second is a FIS-ARMA but with time series preprocessing usingWavelet Decomposition (FIS-ARMA-WAVELET). The results indicate a higher accuracyin inflation forecasts, in terms of Mean Square Error (MSE), of the proposed models over the ARMA model and Artificial Neural Networks.
Inflation forecasting has been and will continue to be a relevant issue for the world’seconomies, since dierent decisions and policies by central banks and governments of eachcountry arise from and are related to this macroeconomic variable. However, even given itsrelevance, there is limited research with new methodologies to forecast it, and even lessstudies in emerging economies. In fact, the most commonly used model in forecasting isthe Autoregressive Moving Average (ARMA) model.The present study proposes two new hybrid models, based on Fuzzy Inference Systems(FIS), to forecast inflation in four Latin American countries.The first model is a Fuzzy Inference System - Autoregressive Moving Average model(FIS-ARMA), while the second is a FIS-ARMA but with time series preprocessing usingWavelet Decomposition (FIS-ARMA-WAVELET). The results indicate a higher accuracyin inflation forecasts, in terms of Mean Square Error (MSE), of the proposed models over the ARMA model and Artificial Neural Networks.
Description
Catalogado desde la version PDF de la tesis.
Keywords
DESCOMPOSICION DE WAVELET , INFLACION , PRONOSTICO DE SERIES DE TIEMPO , RED NEURONAL ARTIFICIAL , SISTEMA DE INFERENCIA DIFUSO