Prediction of tender price index directional changes

A multivariate discriminant analysis is described, aimed at predicting the direction of changes in the Hong Kong tender price index by utilizing the patterns of change in eight leading economic indicators. Two discriminant functions are derived which distinguish between 'upward', 'constant' and 'downward' index trends with a high degree of success. The predictive power of the discriminant model is tested by means of a simulated ex post holdout sample of eight index values. By comparing the group centroids, seven of the cases are correctly classified. The hit rate of the 'upward' and 'constant' groups is 100%, while the 'downward' group has a hit rate of 75%, suggesting the 'downward' trend to be a more difficult movement to predict. Despite this, the overall predictive results are considerably better than those that would have occurred by chance alone.