ignoring the fact that regression is usually inappropriate on non-stationary time series, if you explore your regression a little more by adding the obvious interaction term (i.e scarcity * time) then scarcity becomes significant again. https://twitter.com/phraudsta/status/1185179285096525824?s=20