Not a money guy, but isn't measuring accuracy of a price model via least squares misguided?
A model that predicts the market with 99% accuracy (1% LS error), but reacts to price changes in a delayed manner is not very valuable, while the same model that is early is basically a money printer.