If it did, it would cause continuity errors. Say you have an explosion on-screen, you look away then look back again, and the explosion would keep going from where it was when you stopped looking at it, which you don’t want.
To clarify, when using GPU sprite emitters, the emitter stops drawing particles when the bounds are no longer off screen to avoid wasting resources (and for technical reasons too, primarily the limits of GPU emitters). The emitter continues ticking particles, so the emitter will still eat into computation time depending on how complex your particles are to tick (this will become apparent if Niagara has the same occlusion system as Cascade), but just stops rendering the entire emitter if it is off-screen. This should dramatically lower how expensive your effect is on your GPU, but the CPU will still have a little bit of cost associated with the emitter.