Comment by stevenae

Thanks for the reply! I am outside the forecasting sphere.

RMSLE gives proportional error (so, scale-invariant) without MAPE's systematic under-prediction bias. It does require all-positive values, for the logarithm step.