How would you design an experiment to test whether a new trading signal improves performance while controlling for market regime changes?