Published predictability falls by more than half once it is published
Across 97 predictors published in finance journals, portfolio returns were about 26 per cent lower out of sample, and about 58 per cent lower once the paper was in print. Half of the decay arrives before anyone could have traded on it.
This is the finding that sets the terms for everything else we do. It comes from McLean and Pontiff, who took 97 characteristics that had been published as predicting stock returns, and measured each one again over three windows: the sample the authors used, the period after that sample but before the paper appeared, and the period after publication.
Why the middle bar is the interesting one
The drop after publication has an easy story. People read the paper, they trade the effect, and the effect gets smaller. That is what an efficient market is supposed to do, and it is mildly reassuring.
The middle bar has no such story. Nobody had read the paper yet. The rule simply stopped working as well the moment it left the window it was discovered in. That gap is not the market learning. It is the original number having been too high in the first place.
A published result is an upper bound, not an estimate. Before we build anything on a paper, we assume the effect is roughly a quarter smaller than stated, and we test whether what is left is worth the cost of trading it.
What it does not say
It does not say the literature is worthless. Forty two per cent of something real is still something real, and a handful of the 97 held up almost intact. It also does not say that any particular effect is dead, because the figure is an average across a wide set.
And it says nothing at all about what anyone should buy. It is a statement about how published research behaves as a body, which is a different kind of claim.
How we use it
Every study we run starts by reproducing the published number on the published sample. If we cannot get near it, we stop and write that down. If we can, we run the same rule forward with no changes, and we expect it to be worse. The question is never whether it decays. It is whether what survives the decay is large enough to matter after costs.