I gather that's the idea, but mean squared error and mean absolute error are fairly correlated, so I'm not sure if that would be an advantage or disadvantage.

I'm running a MSE hill climbing thing at the moment, I might give it a go and see if it helps.