Can you trade the oldest anomaly?

When a company reports earnings that surprise the market, its stock doesn't fully reprice that day. It keeps drifting in the direction of the surprise — up after a beat, down after a miss — for weeks afterward. That lag is post-earnings-announcement drift, PEAD.

On paper it looks like free money: the news is public, the direction is known, you just have to ride the drift. That apparent simplicity is why it has been studied to death — and why it is so instructive that it still doesn't pay a retail investor.

The puzzle is why the price doesn't jump to the right level instantly. The leading explanations are about human attention, not arithmetic: investors anchor on stale expectations, digest the implications of a surprise gradually, and only fully update as analysts revise and the next data confirms.

There is a grain of something real here — markets are made of people with limited attention, and a genuine surprise takes time to propagate. That is why PEAD shows up in the data again and again. The question this project cares about is never 'is the effect real?' but 'is what's left, after everyone else has also noticed, large enough to eat?'

PEAD was first documented in 1968 (Ball and Brown) and has been re-confirmed for over half a century. It is arguably the most robust anomaly in all of finance, and for decades it embarrassed the efficient-market hypothesis.

But old and famous is precisely the problem. An edge that thousands of funds have read the same papers about gets arbitraged thin: quants now trade the drift within hours of the release, compressing exactly the lag a slower investor would have harvested. The anomaly persists in attenuated form — but the fat part is gone, captured by whoever is fastest.

We ran PEAD on our survivorship-free Japanese equity panel, out-of-sample and after realistic cost. Before costs, the drift was detectable — consistent with sixty years of literature. After spreads, market impact and the tax on each realised gain, the after-cost edge out-of-sample was not distinguishable from zero.

This is the same shape as our momentum result: a real gross effect that doesn't survive the toll of actually trading it. We published it on the dashboard as a FAIL, alongside everything else that didn't beat a plain NISA index.

First, frequency. Earnings come once a quarter, so a PEAD strategy only gets a handful of fresh signals per stock per year — thin, lumpy, hard to diversify without holding hundreds of names and paying hundreds of spreads.

Second, timing. By the time a retail investor reads the earnings headline, the initial repricing has already happened in seconds. You are buying after the jump, hoping the leftover drift covers your costs — and the leftover is exactly the part professionals have already arbitraged.

Third, liquidity. The drift tends to be largest in smaller, less-covered names — which is also where spreads are widest and your own order moves the price. The places the anomaly looks strongest are the places it is most expensive to trade, a gap that quietly closes the whole thing.

PEAD is a clean case study in a rule that runs through this entire project: a real, decades-confirmed anomaly can still be worthless to you, because what you keep is the effect minus your costs, minus your latency, minus the crowd that got there first.

When you see a strategy built on a famous published edge, assume the published edge is the gross, pre-crowd number — and that your share is whatever survives after speed, spread and tax. For PEAD, our out-of-sample answer was: not enough. The tax-free index holding it failed to beat stays the benchmark.