Why We Underreact to Some News and Overreact to the Rest

Every day, people are bombarded with new information: an earnings report, a weather forecast, a medical test result, a poll number. Rational decision theory says that whenever fresh evidence arrives, we should adjust our beliefs by exactly the amount that the evidence warrants—no more, no less. In practice, decades of research in economics, finance, and psychology have shown that human beings rarely hit that mark. Some of the time we shrug off important signals, a behavior economists call underreaction. Other times we pile onto news that barely matters, producing overreaction. What has puzzled researchers for years is why the same species that ignores one piece of evidence will wildly exaggerate the significance of another. A new study published in Econometrica offers a strikingly simple answer: it depends on how we frame the question in our own minds.

The research, conducted by Yucheng Liang of Carnegie Mellon University’s Tepper School of Business, Tony Q. Fan of Lehigh University’s College of Business, and Cameron Peng of the London School of Economics and Political Science, compared two belief-updating tasks that, on paper, look almost identical. In the first, known as an inference problem, a person observes informative signals about some hidden state of the world—say, whether a company is fundamentally healthy—and revises their beliefs about that state. In the second, a forecast-revision problem, the person sees exactly the same signals but must update their beliefs about a future outcome that depends on the underlying state, such as next quarter’s profits. If people processed information consistently, the two tasks should produce the same updating behavior. They do not.

Across the experiments, participants systematically underreacted to signals when they were inferring the underlying state, but overreacted to the very same signals when they were revising forecasts about future outcomes. The authors call this divergence the inference-forecast gap, and its implications ripple far beyond the laboratory. It suggests that the long-standing debate about whether humans are conservative information processors or impressionable extrapolators may be misframed: people can be both, depending on which mental category a problem falls into. The direction of bias is not a fixed personality trait but a consequence of how the belief-updating problem is represented.

“Economists have long used notions of underreaction and overreaction to explain puzzles in macroeconomics and finance, but have never reached consensus on why people underreact in some environments and overreact in others,” Liang notes. That lack of consensus has real consequences. Models of financial markets, macroeconomic expectations, and consumer behavior often bake in an assumption about how people respond to news, and picking the wrong assumption can distort predictions about asset prices, inflation expectations, and demand. The new findings suggest that the missing ingredient is not a deeper parameter of human psychology but a more careful accounting of the task itself.

The mechanism behind the gap, according to the authors, lies in the rules of thumb—simplifying heuristics—that people deploy in the two settings. When the problem is framed as inference about a hidden state, participants tend to behave conservatively, treating each signal as noisy and insufficiently diagnostic, which produces the dampened, underreacting response long documented in the literature on probability revision. When the same information is framed as a forecast revision, participants appear to lean on extrapolative shortcuts, letting the recent signals carry more weight than they statistically deserve, which generates overreaction. The signals are identical; the cognitive machinery applied to them is not. The gap, in other words, is largely driven by different simplifying rules used in the two tasks rather than by any difference in the information itself.

“Our findings suggest that this discrepancy may arise because people consider some problems inference tasks and others forecast-revision tasks, leading them to approach the two in fundamentally different ways,” says Fan, who led the study. “This implies that to understand biased reactions to information in the field, it is essential to first understand how people mentally represent the belief-updating problem they face.” That framing shifts the research agenda. Instead of asking whether people, in general, overreact or underreact to news, researchers and practitioners should first ask which representation a person is likely to adopt when confronted with a particular information environment. A stock analyst reading a earnings surprise, a voter hearing an economic statistic, and a doctor interpreting a test result may each be running a different mental program on the same data.

The study connects to a vigorous experimental literature that has documented context-dependent belief updating for years. Laboratory studies have found that people underreact when updating beliefs about binary states, yet overreact when forming expectations about series that appear to trend, such as stock prices or sports streaks. Field evidence mirrors the split: survey-based measures of professional forecasters often show sluggish revision of long-run views even as short-term predictions overshoot. What the new paper contributes is a controlled comparison that holds the information constant and varies only the elicitation—the type of belief being measured—thereby isolating the task representation as the driver of the bias. “We show that the direction of belief-updating biases depends on the type of belief elicited and the nature of the question,” Peng explains, “and by connecting the inference-forecast gap to the use of different simplifying rules of thumb, we highlight the role of complexity and incorrect mental models in explaining belief-updating biases.”

The emphasis on complexity and incorrect mental models is important because it moves the explanation away from motivated reasoning or emotional noise and toward the architecture of everyday cognition. Real-world belief-updating problems are rarely stated as cleanly as Bayes’ rule requires. People must decide what the signal means, how it relates to the outcome they care about, and how much of their prior view to retain. Each of those steps invites a shortcut, and the shortcuts that feel natural for one framing can be badly miscalibrated for another. An investor who treats a strong quarterly report as merely one noisy data point about firm quality is underreacting in the inference sense; the same investor who extrapolates that report into a rosy five-year forecast is overreacting in the forecast sense. Both behaviors can coexist in one person on the same afternoon, which is precisely what the inference-forecast gap predicts.

For practitioners, the findings carry practical weight. Financial regulators and market designers care about overreaction because it fuels bubbles and crashes, and about underreaction because it delays the incorporation of information into prices. Communication strategists, from public health officials to central bankers, want to know whether their audiences will absorb or exaggerate a given announcement. The study suggests that the answer depends less on the content of the message than on whether recipients mentally code it as evidence about a state of the world or as a cue about what happens next. Designing information releases, survey questions, or forecasting protocols with that distinction in mind could reduce systematic bias, or at least anticipate its direction.

The work also sharpens the theoretical conversation. Classic models of noisy expectation formation often assume a single, stable updating distortion. The inference-forecast gap implies that any such model is incomplete unless it specifies the belief being formed. Future research, the authors suggest, can map which environments push people toward the conservative inference mode and which push them toward extrapolative forecast mode, and can test whether training, experience, or better problem framing closes the gap. For now, the study offers a memorable lesson: the bias in our beliefs is not written solely in the news we receive, but in the question we silently ask ourselves when the news arrives. Same signal, different frame, opposite error—and that single insight may finally explain why humans manage to be too skeptical and too credulous at the same time.

Subject of Research: Belief updating biases in inference versus forecast-revision tasks

Article Title: Economists examine how people infer, revise beliefs in face of new information

Article References: Economists examine how people infer, revise beliefs in face of new information. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: belief updating, behavioral economics, inference-forecast gap, underreaction, overreaction, Econometrica, decision making, heuristics, forecasting, experimental economics, Carnegie Mellon University, financial expectations