Tree structures for predicting stock price behaviour


  • Robert A. Pearson



It is shown that regression trees can be used to give useful predictions of the average price movements of individual stocks when the market is regular. While the detailed error estimates may be up to three times greater for a two month prediction than for a one week average they are still less than those obtained assuming a constant price. More qualitative measures, such as the agreement in direction of movement, and local turning points are relatively independent of the period. When it is known, a posteriori, that the market has had a minor correction the model fails. This is consistent with the chaotic, fractal behaviour. With the minor correction that occurred on the ASX during April 2000 the model actually performed better in the qualitative measures than a momentum assumption.





Proceedings Computational Techniques and Applications Conference