The Application of Neural Networks in Stock-Exchange Investing

Radoslaw Ogrodnik

Master's thesis

Abstract:

The experiment performed showed that predicting stock movements accurately with a neural networks is a very challenging task. Even obtaining low error in the training process, does not indicate that one will receive high quality predictions in the future. Similarly potential gain obtained be the network in the analysed period does not have to repeat in the next one. The reason behind, is that situations which occurred in the past, do not have to repeat. Moreover there are numerous factors that may influence index's future behaviour which simply cannot be even taken into account by an artificial model. However, even if it is difficult to create a network which would provide an output accurate enough to be used as an only indicator, neural networks are still an interesting tool and may perform well as a guidance for an investor.

Abstract:

The experiment performed showed that predicting stock movements accurately with a neural networks is a very challenging task. Even obtaining low error in the training process, does not indicate that one will receive high quality predictions in the future. Similarly potential gain obtained be the network in the analysed period does not have to repeat in the next one. The reason behind, is that situations which occurred in the past, do not have to repeat. Moreover there are numerous factors that may influence index's future behaviour which simply cannot be even taken into account by an artificial model. However, even if it is difficult to create a network which would provide an output accurate enough to be used as an only indicator, neural networks are still an interesting tool and may perform well as a guidance for an investor.

Keywords

artificial neural network, multilayer perceptron, data overfitting, data generalization, neural network creation, market forecasting, stock-exchange forecasting, radial basis function

Keywords

data generalization, data overfitting, neural network creation, radial basis function, multilayer perceptron, stock-exchange forecasting, artificial neural network, market forecasting

Language used: English

Date on which the thesis was submitted / produced: 13. 5. 2015

Thesis defence

• Date of defence: 12. 10. 2015
• Supervisor: Tomáš Buus
• Reader: František Poborský

Vysoká škola ekonomická v Praze

Master programme:
Finance and Accounting for Common Europe