Propuesta de título para tesis de investigación: indicador de sentimiento para la predicción de la variación en el precio de las acciones IPSA utilizando técnicas de text mining
Cargando...
Archivos
Fecha
2018
Profesor/a Guía
Facultad/escuela
Idioma
es
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Andrés Bello
Nombre de Curso
Licencia CC
Licencia CC
Resumen
La hipótesis de mercados eficientes contempla vacíos al definir el comportamiento del mercado bursátil basado en factores racionales, sin considerar factores irracionales, como es el campo de las finanzas del comportamiento. El sentimiento de los inversores es emitido mediante opiniones, donde en esta investigación se utilizaron métodos computacionales para el procesamiento de tuits, midiéndose el sentimiento de las personas por medio de un léxico, sin embargo, pese a que se ajustó respecto a su eficiencia en el corpus textual, el idioma español generó perturbaciones en el sesgo de sus mediciones, debido a la escases de avances tecnológicos con respecto al idioma. En esta investigación se desarrolló un indicador de sentimiento para la predicción de la variación en el precio de las acciones del IPSA, basado en tres modelos de clasificación de aprendizaje automático supervisado, utilizando minería de textos por medio de datos proveniente de Twitter. Se elaboró un gráfico QQ plot, el cual determinó que tanto la variación en el precio de las acciones del IPSA, como el sentimiento de las personas, provienen de una distribución similar, Es decir, la muestra de la variación en el precio de las acciones vs la del sentimiento de las personas, como prueba empírica, tienen una concordancia en la frecuencia de las características de ambos conjuntos, demostrando que provienen de una distribución similar.
Lo anterior se respaldó por medio de la implementación de los algoritmos de aprendizaje automático supervisado Suport Vector Machine (SVM), Redes Neuronales Artificiales (RNA) y Random Forest (RF), para clasificación, teniendo como dos variables independientes, además del sentimiento, la autoregresión respecto a la variación en el precio de las acciones del IPSA. El desarrollo de los modelos se basó en dos casos, donde el caso 1 considera el sentimiento y el caso 2 no considera lo considera. SVM no otorgó resultados concluyentes, sin embargo, RNA y RF confirieron resultados promisorios involucrando el sentimiento de las personas, aumentando su precisión y mejorando el modelo. Pese a no obtener resultados determinantes, no solo por el idioma, sino también por una cultura que no se concentra mayormente en los movimientos del mercado bursátil, se observó que la influencia del sentimiento de las personas no se considera inocua ante la variación del precio de las acciones del IPSA, respaldando el campo de las finanzas del comportamiento, aprobando la hipótesis, es decir, existe una correlación entre el sentimiento de las personas y la variación en el precio de las acciones. Sin embargo, esto indica se pueden obtener resultados determinantes en base a los avances tecnológicos, ampliando el espectro a la mayor cantidad de idiomas y dependiendo de la cultura de cada región.
The efficient markets hypothesis contemplates gaps when defining the behavior of the stock market based on rational factors, without considering irrational factors, such as the field of behavioral finance. The sentiment of the investors is expressed through opinions, where in this research computational methods were used for the processing of tweets, measuring the feeling of the people through a lexicon, however, although it was adjusted with respect to its efficiency in the textual corpus, the Spanish language caused disturbances in the bias of its measurements, due to the scarcity of technological advances with respect to the language. In this research, a sentiment indicator was developed for the prediction of the variation in the price of IPSA shares, based on three models of supervised machine learning classification, using text mining by means of data from Twitter. A QQ plot was drawn up, which determined that both the variation in the price of IPSA shares and the sentiment of the people come from a normal distribution, that is, there is a correlative behavior between both sets of data. This was supported through the implementation of the Suport Vector Machine (SVM), Artificial Neural Networks (RNA) and Random Forest (RF) algorithms, having as two independent variables, in addition to feeling, the autoregression with respect to the variation in the IPSA share price. The development of the models was based on two cases, where case 1 considers the feeling and case 2 does not consider what is considered. SVM did not give conclusive results, however, RNA and RF conferred promising results involving people's feelings, increasing their accuracy and improving the model. In spite of not obtaining determinant results, not only by the language, but also by a culture that is not concentrated mainly in the movements of the stock market, it was observed that the influence of the feeling of the people is not considered innocuous before the variation of the price of the actions of the IPSA, supporting the field of behavioral finances, approving the hypothesis. However, this indicates that decisive results can be obtained based on technological advances, expanding the spectrum to as many languages and depending on the culture of each region.
The efficient markets hypothesis contemplates gaps when defining the behavior of the stock market based on rational factors, without considering irrational factors, such as the field of behavioral finance. The sentiment of the investors is expressed through opinions, where in this research computational methods were used for the processing of tweets, measuring the feeling of the people through a lexicon, however, although it was adjusted with respect to its efficiency in the textual corpus, the Spanish language caused disturbances in the bias of its measurements, due to the scarcity of technological advances with respect to the language. In this research, a sentiment indicator was developed for the prediction of the variation in the price of IPSA shares, based on three models of supervised machine learning classification, using text mining by means of data from Twitter. A QQ plot was drawn up, which determined that both the variation in the price of IPSA shares and the sentiment of the people come from a normal distribution, that is, there is a correlative behavior between both sets of data. This was supported through the implementation of the Suport Vector Machine (SVM), Artificial Neural Networks (RNA) and Random Forest (RF) algorithms, having as two independent variables, in addition to feeling, the autoregression with respect to the variation in the IPSA share price. The development of the models was based on two cases, where case 1 considers the feeling and case 2 does not consider what is considered. SVM did not give conclusive results, however, RNA and RF conferred promising results involving people's feelings, increasing their accuracy and improving the model. In spite of not obtaining determinant results, not only by the language, but also by a culture that is not concentrated mainly in the movements of the stock market, it was observed that the influence of the feeling of the people is not considered innocuous before the variation of the price of the actions of the IPSA, supporting the field of behavioral finances, approving the hypothesis. However, this indicates that decisive results can be obtained based on technological advances, expanding the spectrum to as many languages and depending on the culture of each region.
Notas
Tesis (Ingeniero Civil Industrial)
Palabras clave
Acciones (Bolsa), Minería de Datos, Emociones