The Market Takes Months to Read an Earnings Report

Of all the challenges to the efficient market hypothesis, this is the one that should not exist.

A company announces earnings. The number is a huge, public, unambiguous surprise, splashed across every terminal on the planet. The efficient market hypothesis says the price should adjust immediately, within seconds, and then go back to being unpredictable.

What actually happens is that the price jumps on the day, and then keeps drifting in the same direction for weeks and months afterwards. Good news firms keep drifting up. Bad news firms keep drifting down. This is post-earnings-announcement drift, or PEAD, first noticed by Ball and Brown back in 1968, and it has been a scandal ever since.

Victor Bernard and Jacob Thomas set out in 1989 to settle what was causing it.

The problem: is it a mistake, or is it a risk premium?

By the late 1980s the drift was undeniable. The question was what it meant, and there were exactly two live answers, which is why the paper's subtitle is "Delayed Price Response or Risk Premium?"

• The risk premium defence. Perhaps stocks with big positive earnings surprises are somehow riskier, and the drift is just fair payment for bearing that risk. If so, the market is fine, and our model of risk is merely incomplete. This is the answer that saves market efficiency.
• The delayed response accusation. Perhaps investors are simply slow. The information is right there and they fail to fully process it. If so, the market is not efficient in any useful sense.

Everything hinged on which one. Bernard and Thomas ran the tests, looked hard for the risk-based explanation, and could not find it. The drift did not look like compensation for risk. It looked like people being slow.

The key idea via analogy: nobody notices that surprises come in streaks

The most compelling part of their work is not the rejection of the risk story. It is the specific mistake they identified investors making, which they developed further in follow-up work. It is beautifully concrete.

Here is the setup. Corporate earnings surprises are autocorrelated. If a company beats expectations this quarter, it is more likely than not to beat again next quarter. Business conditions are persistent: a company that just had a great quarter is usually in an environment that produces another good quarter. Anyone who studied a few years of earnings data would see this immediately.

Now, what would a rational investor do on hearing a big positive surprise? They would think: "Excellent. And since surprises tend to run in streaks, I should also raise my forecast for next quarter." They would price in both the current good news and its implications for the future, all at once, today.

Investors did not do this. Bernard and Thomas found that the market priced the current surprise as if it were a one-off, a bolt from the blue with no implications for what comes next. Then, three months later, when the company reported another good quarter (which was largely predictable from the first one), the market acted surprised all over again, and the price jumped again.

Their phrase for it is exact: investors "fail to recognize fully the implications of current earnings for future earnings."

The drift is not a mysterious risk premium. It is the market repeatedly being astonished by something it had every opportunity to see coming. The price catches up in instalments, one quarter at a time, as the predictable future arrives.

Why it mattered

• It is arguably the cleanest violation of market efficiency ever documented. This is not a subtle statistical pattern in obscure microcaps. It is the market failing to correctly price the most watched, most analysed public information there is, on the largest companies in the world.
• It survived everything. PEAD has been re-tested for decades, in dozens of countries, with every risk model anyone could invent. It keeps showing up. It is the anomaly that will not die.
• It made earnings surprise a core quant signal. Standardised unexpected earnings, and its cousins built on analyst revisions, are among the most widely used signals in systematic equity investing. This literature is the reason.
• It gave behavioural finance a mechanism, not just a mood. "Investors are irrational" is a weak claim. "Investors ignore the autocorrelation in quarterly earnings surprises" is a specific, falsifiable, arithmetic mistake. That precision is what makes it persuasive.

The honest limitations

• The costs eat much of it. The drift is strongest in smaller, less liquid, less analysed stocks. Once you subtract realistic trading costs and shorting costs, a large chunk of the paper profit disappears, which is probably part of why it has persisted.
• It has weakened. The drift appears smaller in recent decades than in the original samples, consistent with the general pattern that publication and algorithmic trading erode anomalies. Weakened, though, not gone.
• Measuring the surprise is not trivial. "Unexpected earnings" requires a model of what was expected. Analyst consensus, time-series models and other approaches give different answers, and the strength of the drift depends on which you use.
• The risk story was rejected, not disproved. Bernard and Thomas found no risk-based explanation using the models available. A sufficiently creative future risk model might still account for it. Most people find this unlikely, but it is the honest caveat.

The one-line takeaway

Bernard and Thomas showed that the drift after an earnings surprise is not a hidden risk premium but a plain human failure: investors treat each earnings surprise as a bolt from the blue, ignoring that surprises come in streaks, so the price arrives at the truth in instalments over the following months.