| Abstract: | Technical prediction of stock returns is notoriously extremely difficult. In this report we use the concept of trends as predictor variables. A statistical investigation of the trend concept is presented for stocks on the Swedish stock market. Trend variables are further used as input variables in a nearest‐neighbor analysis to find patterns of trend values that result in non‐random future returns. The nearest‐neighbors algorithm is extended by a selection procedure to find regions in the input space, where the future returns are asymmetrically distributed. The algorithm is applied to a number of international stock indexes and has positive results for some and negative for others. It also has other natural applications when the overall predictability is low, and can only be expected to apply in indeterminate regions of the input space. Copyright © 2000 John Wiley & Sons, Ltd. |
