Publication: PRONÓSTICO DEL PRECIO DE ACCIONES DE LA BANCA CHILENA MEDIANTE REDES NEURONALES ARTIFICIALES
Loading...
Date
2017
Authors
SIMI SALINAS, STEFANO ANDRÉS
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
La redes neuronales artificiales (RNA) han sido ampliamente estudiadas en las últimasdécadas para preprocesamiento de imágenes, clustering, problemas de clasificación y en elámbito de la predicción. Este último, ha sido abordado en muchas ocasiones por el mercadoaccionario, para que inversionistas y corredores de bolsa puedan obtener beneficios en eltrading de acciones.En esta memoria, se modeló el precio de cierre diario de cuatro acciones del sector bancariochileno mediante redes neuronales artificiales, con un benchmark de modelos linealesclásicos de la literatura, como lo son los modelos autorregresivos (AR). El prónostico serealizó mediante un análisis de ventana móvil, con un paso hacia adelante. Además, seimplementó un modelo híbrido que permitió combinar los resultados del modelo AR con laRNA. Las redes neuronales utilizadas fueron multicapa prealimentadas con propagaciónhacia atrás. Los pronósticos obtenidos fueron medidos con métricas de desempeño conocidas,como MSE, RMSE, MAE y MAPE para cada modelo en el estudio.Los resultados arrojaron que, en tres de cuatro acciones del estudio, los modelos ARfueron superiores en su capacidad predictiva a las RNA y al modelo híbrido.
Artificial neural networks (ANNs) have been extensively studied in the last decades fortasks such as image preprocessing, clustering, classification problems and in the predictionfield. The latter has been approached on many occasions by the stock market, so thatinvestors and stockbrokers would profit in the stock trading.In this report, the daily closing price of four shares of the Chilean banking sector wasmodeled using artificial neural networks, with a benchmark of classical linear modelsused in the literature, such as autoregressive models (AR). The forecast was made by arolling-window analysis, with one-step ahead. In addition, an hybrid model that allowed tocombine the results of the AR model with the RNA was implemented. Multilayer FeedforwardNeural Networks with Backpropagation were used.The forecasts obtained were measured for each model in the study, using knownperformance metrics, such as MSE, RMSE, MAE and MAPE. The results showed that, inthree out of four of the studied actions, the predictive capacity of AR models were superiorto the predictive capacity of RNA and to the hybrid model’s one.
Artificial neural networks (ANNs) have been extensively studied in the last decades fortasks such as image preprocessing, clustering, classification problems and in the predictionfield. The latter has been approached on many occasions by the stock market, so thatinvestors and stockbrokers would profit in the stock trading.In this report, the daily closing price of four shares of the Chilean banking sector wasmodeled using artificial neural networks, with a benchmark of classical linear modelsused in the literature, such as autoregressive models (AR). The forecast was made by arolling-window analysis, with one-step ahead. In addition, an hybrid model that allowed tocombine the results of the AR model with the RNA was implemented. Multilayer FeedforwardNeural Networks with Backpropagation were used.The forecasts obtained were measured for each model in the study, using knownperformance metrics, such as MSE, RMSE, MAE and MAPE. The results showed that, inthree out of four of the studied actions, the predictive capacity of AR models were superiorto the predictive capacity of RNA and to the hybrid model’s one.
Description
Catalogado desde la version PDF de la tesis.
Keywords
ANALISIS VENTANA MOVIL , MODELO AR , PRONOSTICO UN PASO HACIA ADELANTE , REDES NEURONALES ARTIFICIALES , SECTOR BANCARIO CHILENO