Although this paper shows the effectiveness of the
RLS-TS in forecasting exchange rate, there are only three
kernel functions have been investigated. One of future work is
to explore more useful kernel functions for further improving
the performance of the RLS-TS models.
The Regularization Least-Squares method is originated from
statistical learning theory and used for classification and
function estimation issues. The hypothesis is that the training
set is drawn from the random independent identically
distributed observations. But for the times series forecasting,
the observations are definitely dependent. So, the future work
should give the new answer, in dependent conditions, of the
four questions proposed by Vapnik [7] in statistical theory.