Reid Ji

Right — in-sample is the honest caveat. The finding is the drop itself: a raw signal looks great until you control for need size. So it's a hypothesis to test out-of-sample next — and I'll publish whichever way it goes.