Investigation of the predictability of steel manufacturer stock price movements using particle swarm optimisation

It is shown that ensemble classifiers composed of neural networks trained using particle swarm optimisation can uncover a substantial degree of predictability in stock price movements. As in a previous work by the authors use is made here of a training metric, the Matthews correlation coefficient, that has been shown to better handle numerically unbalanced data sets. The work provides a solid basis for the future construction of a trading model. © Springer-Verlag 2013.