Predicting Bonds Using the Linear Relevance Vector Machine

As described earlier in this book, a large part of the problem in modelling financial markets is to select the relevant inputs. Even if we restrict ourselves to only 40 input time series, and allowing for a choice of lags of up to 24, this still leads to a selection process involving up to 960 variables. This is difficult to achieve automatically in a single step, so various methods have been developed in an attempt to solve this combinatorial explosion.