Noise in financial markets

Noise in financial markets describes the imperfect, uneven information that enters prices and forecasts—and the errors that can obscure it. The book’s account turns on a distinction: dispersed judgments can combine into a powerful signal when people retain independent perspectives and a mechanism aggregates them, but information does not become useful merely because it exists. Markets may collect it; organizations may instead overlook it, or mistake confidence for knowledge.

Information, Error, and the Crowd

The book frames collective judgment as a mixture of information and error. When a large enough group of diverse, independent people estimates a quantity or probability, their individual errors can cancel in the average, leaving the information each person contributes. This is not a guarantee that a crowd will be right: if its members have little relevant information, aggregation cannot manufacture it. The claim is narrower and more striking—that a crowd can hold, across many minds, a nearly complete picture no individual possesses alone.

The market reaction to the Challenger disaster illustrates the argument. Shareholders and potential shareholders were, in effect, estimating how much less four companies were worth after the shuttle exploded. The book suggests that even if no trader knew for certain that Thiokol was responsible for the O-ring failure, the market collectively registered that responsibility. Its judgment worked because it combined private interpretations rather than requiring any one participant to possess the whole explanation.

That account identifies four conditions for a wise crowd: diversity of opinion, independence of judgment, decentralization that permits people to use specialized or local knowledge, and aggregation through a mechanism capable of turning separate judgments into a collective decision. Noise is not simply eliminated by gathering more opinions. The structure of the gathering matters: dependent opinions can reproduce one another’s errors, while independent contributions can offset them.

Prices as Forecasts

Sports betting provides a visible example of judgment being revised through a market. At the Mirage, bookmaker Robert Walker adjusted the point spread when early bets suggested that bettors favored one side too heavily. Before the 2001 Super Bowl, two early $3,000 bets on the Baltimore Ravens were enough to move the opening spread from Baltimore by two and a half points to three. The opening number was the bookmaker’s; the line that developed thereafter increasingly reflected bettors’ collective judgment.

A spread is not a crystal ball, and it predicts final scores only weakly. Yet consistently beating the final spread is difficult even for well-informed gamblers. Favorites cover about half the time and underdogs about half; the market shows no simple, persistent error such as systematically underestimating home teams. A documented exception offered a limited opportunity: home underdogs in weeks fifteen through seventeen of the National Football League season had historically been a good bet. More broadly, the Mirage’s closing line was described as the most reliable available forecast of NFL outcomes roughly three-quarters of the time. Across major sports, research generally finds betting markets relatively efficient at incorporating available information; in horse racing, final odds also provide reasonably good estimates of each horse’s chance of winning.

The Iowa Electronic Markets offer a different test. Its election contracts paid according to election outcomes, including contracts tied to a candidate’s share of the popular vote. Across forty-nine elections from 1988 to 2000, election-eve prices missed the result by an average absolute error of 1.37 percentage points in presidential elections, 3.43 points in other U.S. elections, and 2.12 points in foreign elections. The market generally outperformed major national polls, and on the day polls appeared it was more accurate than three-quarters of 596 polls released during the presidential elections in that period.

The market was small—never more than about eight hundred traders—and its participants did not mirror the electorate. Most were men, and a disproportionate, though declining, share lived in Iowa. They were not simply predicting their own votes; their aggregation nevertheless forecast the choices of voters across the country. Its forecasts also moved less wildly than poll results, changing sharply mainly when new information appeared. The contrast shows one way markets can filter fluctuating judgments without claiming perfect foresight.

Diversity and the Limits of Expertise

Markets often contain diversity by virtue of their size and open entry: participants differ in risk tolerance, time horizons, investing styles, and information. Small teams and formal organizations cannot count on that variety arising automatically, and a few biased members can exert disproportionate influence. In such settings, who is included can change the group’s access to information before any discussion begins.

Political scientist Scott Page tested this point with computer-simulated groups of ten or twenty problem-solving agents, each equipped with a different set of skills. Some agents performed better individually than others, but groups mixing stronger and weaker problem-solvers almost always outperformed groups composed only of the strongest. Selecting a group randomly and letting it work could do as well as, or better than, spending time identifying the best individual agents and assembling them alone.

The result does not make intelligence irrelevant: successful groups still included high-performing agents, and none of the simulated agents was ignorant. It shows instead that individual ability and group capacity are not the same. If highly capable people tend to bring similar tools, a team of them may leave perspectives and skills unused. Adding people who are less accomplished by one measure but differently equipped can enlarge what the group collectively knows. In this account, diversity reduces one source of error not by making each judgment flawless but by preventing a group from sharing the same blind spot.

When Information Is There but Unused

The intelligence failures preceding September 11, 2001, expose the other side of aggregation. After the Second World War, intelligence work did not settle into the centralized system “Wild Bill” Donovan had envisioned. Instead, the intelligence community became a fragmented collection of agencies with overlapping responsibilities, including the Central Intelligence Agency, the National Security Agency, the National Imagery and Mapping Agency, the National Reconnaissance Office, the Defense Intelligence Agency, and the military services’ intelligence arms. Although the CIA director was nominally in charge, he exercised little supervision in practice; much of the funding came from the Department of Defense. The Federal Bureau of Investigation, responsible for domestic law enforcement, operated almost entirely outside that community.

Before September 11, the system had also failed to anticipate the 1993 World Trade Center bombing, the 1998 U.S. embassy bombings in Kenya, and the bombing of the USS Cole in Yemen. A congressional inquiry later found that agencies had failed to capitalize on the individual and collective significance of available information, missed opportunities to disrupt the September 11 plot, and overlooked material that might have improved the chances of uncovering and preventing it. Yet retrospective judgment has limits. Roberta Wohlstetter’s study of Pearl Harbor argues that warning signals were submerged in what she called “the buzzing and blooming confusion”; recognizing which signals mattered was difficult even after the attack made their significance apparent. The book therefore treats strategic surprise as a genuine problem, not proof that every missed clue should have been obvious in advance. It leaves open whether a different organization could at least have produced a more concrete estimate of the threat.

The Columbia shuttle investigation makes the distinction between available evidence and usable information more immediate. After foam struck the shuttle’s left wing during ascent, engineers estimated the debris’s size and speed and used the Crater algorithm to assess how deeply it might penetrate the thermal-protection tiles. The Debris Assessment Team had no firm conclusion, but it reported reason for concern. Its request for on-orbit images had not been approved, limiting the evidence available to the analysis.

Managers also claimed that images would not have sufficient resolution to reveal the damaged area, although none had the necessary security clearance to know the resolution and none asked the Department of Defense, which would take the photographs. The decision was thus made with little or no knowledge presented as certainty. When the assessment team later offered five scenarios, it qualified its likely-safe conclusion by emphasizing that the analysis was profoundly limited: the engineers did not know precisely where the foam had hit, and Crater had been designed for debris hundreds of times smaller. Management focused on the reassuring conclusion rather than the uncertainty surrounding it. The episode shows that information can fail to guide a decision not because it is absent, but because a group does not gather, transmit, or weigh it adequately.

Noise, Bubbles, and Market Efficiency

A market can also mistake other people’s activity for evidence about value. In a Caltech experiment, students traded shares for fifteen five-minute periods. Each share paid a dividend of 24 cents per period, making its maximum value calculable: $3.60 before the first period, then $3.36 after the first dividend, with the value declining by 24 cents at each period’s end. Despite that transparent schedule, the trading price jumped to $3.50 and stayed there almost until the experiment ended. Even after a share’s calculated value fell below $1, participants continued paying $3.50 for it.

Participants explained that they knew the price was too high but expected to buy, collect a dividend or two, and resell at the same price to someone else. The gap between a share’s known dividend value and its persistent trading price makes the experiment a demonstration of how belief in resale can overwhelm information about value. The book’s broader distinction is therefore essential: aggregation can make a market a useful forecaster when participants contribute independent information, but prices can also carry noise when traders follow one another’s activity rather than the underlying facts. A market’s collective judgment is powerful, not infallible.