Study on Forecasting the Stock Market Trend Based on Stochastic Analysis Method

Zhang, Deju; Zhang, Xiaomin

doi:10.5539/ijbm.v4n6p163

Cited by 36 publications

(29 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…arithmetic operators, conditional operators or problem specific functions) and leaf nodes would be terminalsexternal inputs, constants, and zero argument functions. On the other hand, Zhang et al 32 applied Markov Chain…”

Section: Stock Market Prediction Modelsmentioning

confidence: 99%

An Intelligent Model for Stock Market Prediction

Hamed¹,

Hussein²,

Tolba³

2012

IJCIS

View full text Add to dashboard Cite

This paper presents an intelligent model for stock market signal prediction using Multi-Layer Perceptron (MLP) Artificial Neural Networks (ANN). Blind source separation technique, from signal processing, is integrated with the learning phase of the constructed baseline MLP ANN to overcome the problems of prediction accuracy and lack of generalization. Kullback Leibler Divergence (KLD) is used, as a learning algorithm, because it converges fast and provides generalization in the learning mechanism. Both accuracy and efficiency of the proposed model were confirmed through the Microsoft stock, from wall-street market, and various data sets, from different sectors of the Egyptian stock market. In addition, sensitivity analysis was conducted on the various parameters of the model to ensure the coverage of the generalization issue. Finally, statistical significance was examined using ANOVA test.

show abstract

Section: Stock Market Prediction Modelsmentioning

confidence: 99%

An Intelligent Model for Stock Market Prediction

Hamed¹,

Hussein²,

Tolba³

2012

IJCIS

View full text Add to dashboard Cite

show abstract

“…En este sentido, autores como Fernández y Sosvilla (1998), proveen evidencias en favor de la predicción no-lineal de los tipos de cambio. Otros autores han utilizado métodos estocásticos basados en cadenas de Markov con el fin de afrontar los estados aleatorios en el comportamiento de estos mercados (Zhang y Zhang, 2009). A pesar de no entregar un valor concreto de pronóstico, su ventaja se da en que permite predecir cambios posibles en los estados de los precios en términos de probabilidad de ocurrencia.…”

Section: Introductionunclassified

Aplicación de las Redes Neuronales al Pronóstico de Precios en el Mercado de Valores

2012

View full text Add to dashboard Cite

ResumenEste trabajo propone un modelo basado en redes neuronales artificiales para el pronóstico de los precios de dos de las principales acciones transadas en mercado de valores colombiano. El modelo propuesto se aplica al estudio de las acciones de Ecopetrol y Preferencial Bancolombia, empresas que negocian en las bolsas de valores de Colombia y Nueva York. Se utilizan dos estructuras de redes incluyendo como entradas la serie de precios diarios en la primera y la serie de precios más el índice del dólar estadounidense DXY en la segunda. Se prueban diferentes configuraciones de redes neuronales utilizando una serie de seis meses, donde los datos de los primeros cinco se utilizan para entrenamiento dejando el último mes para verificar la capacidad predictiva de la red. Los resultados muestran un buen comportamiento de las redes neuronales con bajos errores en su desempeño tanto en aprendizaje como en predicción. AbstractAn artificial neural network model to forecast the price of two of the main shares traded in the Colombian stock exchange is proposed in this work. The model is applied to study the shares of Ecopetrol and Preferencial Bancolombia, companies that trade in the stock exchanges of Colombia and New York. Two network structures including the daily price series in the first and the price series plus the dollar index DXY in the latter are used. Different neural networks configurations are trained using a series of six months, where five months are used as training patterns and the next month is left to test the predictive capabilities of the network. The results show good performance of the neural networks with low training and testing errors.

show abstract

“…The transition matrix is calculated from all historical data without dividing them in time windows and following the known process of transformation to the stochastic matrix for each approach (described in subsection 2.1), and the transition probability matrix is generated. According to Markov process, the predicted transition probability matrix * in different periods is formed by +1 * = * , where t=1, 2,.., n (Zhang and Zhang, 2009). Suppose that the null hypothesis 0 : = * , in case of testing all transitions together then a statistical criterion for testing the hypothesis can be based on the fact that the test statistic…”

Section: Statistical Testmentioning

confidence: 99%