If information advantage has a ceiling, where does it come from in the first place? A price is a machine for aggregating information: information scattered among countless people is compressed through trading into a single number, and public information enters prices within minutes. But the machine does not run for free: people gather information only when gathering it is profitable. Markets can therefore only come close to efficiency. Information advantage never disappears entirely; it only shrinks until it just compensates the cost of gathering information.
In "The Use of Knowledge in Society" (1945), Hayek gave an example. Suppose a tin mine somewhere shuts down, or a new use for tin is found, so that tin becomes scarcer. People around the world who use tin do not need to know what happened or why; they only need to see the price of tin rise, and they will economize on tin and switch to substitutes. No single person holds all the information, yet the price aggregates it and coordinates everyone's actions.
Financial markets push this mechanism to its limit. A company's share price aggregates the scattered pieces of information held separately by suppliers, customers, employees, analysts and countless investors. The questions are how quickly and how completely information enters the price, and who drives it in.
In his 1970 review, Fama divided market efficiency into three strengths according to which kind of information prices reflect:
Each form contains the one before it, and each can be tested with data.
Three strengths of market efficiency
The information in each inner ring is contained in the outer one. Tap any ring to see how it is tested and what the tests found.
The tests show that the weak and semi-strong forms broadly hold, while the strong form does not. Jaffe found in 1974 that company insiders earn excess returns when trading their own company's stock; Keown and Pinkerton found in 1981 that the share prices of takeover targets had already risen markedly before the announcement, showing that news had leaked and been traded on beforehand. It is precisely because strong-form efficiency does not hold that countries have legislated against trading on material non-public information: the advantage of inside information is real, and that advantage is exactly what the law aims to block.
The standard method for testing semi-strong efficiency is the event study: align and aggregate the excess returns of many companies around the same type of event (such as an earnings announcement, a stock split or a merger), and observe how prices change before and after the event. In 1968 Ball and Brown used this method to study annual earnings announcements and found that most of the information in good or bad earnings was already reflected in prices before the announcement. In 1969 Fama, Fisher, Jensen and Roll studied stock splits and found that the price adjustment was completed before the split, with no profitable pattern afterwards.
Event studies: lining up thousands of announcements
The horizontal axis is trading days from the announcement; the vertical axis is cumulative excess return. Switch between three typical cases.
Some results are less clean. In 1989 Bernard and Thomas confirmed that the share prices of companies whose earnings beat expectations keep drifting slowly in the same direction for dozens of trading days after the announcement, a phenomenon known as "post-earnings-announcement drift". It is one of the most robust anomalies, and it also shows that prices do not always absorb information in a single step.
How quickly do prices absorb public information? Using transaction-by-transaction data on earnings and dividend announcements, Patell and Wolfson found in 1984 that the main price reaction occurs within a few minutes to a little over ten minutes of the announcement. Studying televised financial commentary on individual stocks, Busse and Green found in 2002 that prices begin to move within seconds of the broadcast; positive comments are absorbed into prices within about a minute, while negative comments take somewhat longer.
For ordinary investors, this means that by the time they see a piece of public news and trade on it, the price has most likely already reflected it.
If prices already reflected all information, no one would be willing to pay to gather information, because whatever they gathered would already be contained in the price and there would be no profit in it. But if no one gathered information, how could prices reflect it? In 1980 Grossman and Stiglitz pointed out that a perfectly efficient market is logically impossible. In equilibrium, prices reflect information only partially, and the bit of inefficiency that remains is just enough to compensate informed traders for the cost of gathering information.
The paradox turns "efficiency" from a state into a process: markets are close to efficient because someone is always pushing them towards efficiency in pursuit of profit.
Informed traders drive information into prices, but they do not want the information to enter prices all at once, because as soon as the price moves their advantage is gone. In 1985 Kyle built a model in which the informed trader splits the order and trades many times, mixing in with a large volume of "noise trading" by people who hold no information. The market maker sees only the total order flow and cannot tell which orders come from the informed trader, so it can only adjust the price step by step according to the imbalance in the order flow. As a result, information enters the price bit by bit and is fully reflected by the end of trading. The more noise trading there is, the better the informed trader can hide and the more it earns, but the price still reaches the true value in the end.
In 1986 Black summed this up: noise makes financial markets possible, and it also makes them imperfect. Without noise trading, the informed cannot profit, so no one gathers information; with noise trading, prices deviate from value.
The way information enters prices determines what financial information products can and cannot offer users.
By the time public news is pushed to users, the price has most likely reflected it; the speed race belongs to professional institutions, not individual investors.
Explaining how a piece of information relates to the user’s holdings, goals and agreement is worth more than reporting it first.
AI must not obtain, infer or spread material non-public information; data sources and scope of use must be traceable.
For frequent trading that chases news, show the costs and the evidence, so that users’ trades rest on their own goals.