Can't you essentially fabricate this statistic by controlling for the amount of time that counts as "before", assuming the inverted curves happen more often than recessions?
To address the GP’s point, the relationship could be fabricated, but the data don’t indicate that that’s happening.