My fear with this type of 'system' is that it is easily open to curve fitting.
If, for example, you ran numerous backtests on HLT, then DIS, to determine the best parameters for your percentages for each instrument, then this is seriously curve fit to the data and useless going forward. And...