There are various methods of including a subset of predictive
variables for returns prediction.
You can try stepwise regression, which picks successively the most
predictive variable in a linear regression, and stops when an
additional variable does improve R^2.
You can also try a regression tree, with a similar principle.
Both are available from Matlab Statistics toolbox.
Hope this helps!
We won’t know a priori which piece of information is important, and what weight it should have.