How prediction markets reveal what we believe -- and how we behave.

A few years ago, I read Everybody Lies by former Google data scientist Seth Stephens-Davidowitz. The book introduces the power of big data with a distinctly sociological lens, drawing on information from Google, Facebook, and other digital platforms.

Its premise is simple: if you want to truly understand people, their fears, motivations, and desires, do not ask them. Look at what they search.

When people are alone with a screen and a question, something shifts. There is no social cost to asking about our deepest anxieties. Few people have not conducted a late-night search about something that felt urgent in the moment.

In those private interactions, the façade drops. Social signaling fades. There are no filters, no audiences. What emerges is something closer to the truth.

That idea is what makes the book so compelling. Big data is not just more data; it is often more honest data. Traditional sources such as surveys, interviews, and even conversations are flawed. People exaggerate, misrepresent, and smooth over uncomfortable edges. Sometimes, they deceive even themselves.

But human behavior, chaotic as it may be, leaves a trail. Increasingly, that trail is digital.

Today, that insight feels even more relevant. There is now something more intimate than a search query: conversations with artificial intelligence.

A significant share of interactions with large language models involves personal questions: practical guidance, decisions, and dilemmas. The nature of these exchanges has evolved from simple task execution to more personal reflection. Questions about relationships, health, and individual choices are now routine.

These systems receive confessions people might not share elsewhere. A question framed hypothetically is often anything but.

People are still searching for the same things: guidance, clarity, reassurance. The difference is that the interface has changed. There is now something like a therapist, teacher, and advisor available at any hour — patient, accessible, and non-judgmental.

With that comes an extraordinary volume of behavioral data — deeply personal, highly revealing, and largely unguarded.

Remarkable, yes. Also worth considering carefully.

But the same principle that makes AI interactions revealing (that people are more honest when unobserved) applies in another domain entirely.

Markets.

More specifically, prediction markets. Systems built around a single question: what do people truly believe will happen?

Can You Predict the Future?

A prediction market is a marketplace where participants trade contracts tied to future outcomes. If the event occurs, the contract pays out; if not, it expires worthless. These markets can cover everything from elections to economic data to corporate behavior.

The logic behind them echoes the Efficient Market Hypothesis: prices reflect all available information, not because any one participant is fully informed, but because markets aggregate fragmented knowledge into a single signal.

In theory, that signal approximates truth.

Prediction markets extend this idea. They do not just ask participants what they think, they require them to commit capital. If someone has better information, they are incentivized to act. If they are wrong, they incur losses. If they are right, they are rewarded.

Over time, this mechanism should push prices toward the true probability of an outcome.

In practice, these markets often react faster than traditional sources of information and can outperform polling and expert forecasts.

It sounds almost too clean. A decentralized truth machine, continuously updated by people with real stakes in being right.

Except markets are human systems. And humans, as it turns out, are messy. Really messy.

Markets Are Human Systems

Once money is introduced, behavior changes. Prediction markets do not simply measure beliefs, they create incentives to gather information, act on it, and occasionally distort reality itself.

Some participants have taken that to unexpected places.

The most viral example: someone allegedly snuck a battery-powered hairdryer onto a road near Charles de Gaulle airport and held it directly up to a temperature sensor until the official readings spiked well above actual conditions. They’d already placed a bet on exactly that temperature threshold. They walked away with $34,000. France’s official weather agency filed a criminal complaint. The sensor has since been relocated.

Then there are the mention traders, people who bet on specific words politicians will say at speeches and press conferences. One 20-year-old store manager in Pennsylvania began livestreaming his Kalshi trades, training himself by watching what he described as “uncountable hours” of Trump footage to anticipate the president’s vocabulary. He attended a Trump rally in December and, by his own since-deleted account, shouted words from the crowd trying to get the president to say them. Somalia. Bitcoin. Hottest. He later claimed it was a joke. The video suggests otherwise.

And then, most seriously: a U.S. special forces soldier was arrested for allegedly using classified military intelligence about the capture of Venezuelan president Nicolás Maduro to make $400,000 on Polymarket. He’s now facing federal charges.

Meanwhile, Bloomberg analyzed every wallet active on Polymarket since early 2025 and found that over 100,000 accounts lost at least $1,000, nearly twice the number that made that much. The majority of profits were captured by a tiny slice of automated bots. Everyone else, in aggregate, lost $131 million.

What makes these cases so interesting isn’t just that people cheated. It’s how they cheated. They didn’t hack the platform. They manipulated the underlying reality the market was trying to measure. Which somehow seems more unsettling. Because we seem to be learning what happens in real time when you build a truth machine and then tell people there’s money in breaking it. Orwellian to say the least. That is a different kind of vulnerability.

Prediction markets were designed to extract honest beliefs by making incorrect views costly. But when the stakes rise, some participants stop trying to be right — and instead try to shape the outcome.

This is the inversion of the Everybody Lies insight. Search data reflects honesty because there are no consequences. Prediction markets aim to produce honesty through consequences. But consequences introduce a second force: the incentive to manipulate.

Traditional equity markets, for all their imperfections, remain anchored to economic activity — companies producing goods and services, generating revenue, and compounding value over time. Even there, structural shifts have introduced distortions, as passive investing and algorithmic strategies increasingly influence price movements.

Prediction markets lack that underlying foundation. They are structurally zero-sum, and often negative sum once fees and intermediaries are considered. Gains for one participant come directly from losses elsewhere.

In that environment, the incentive to outperform others — whether through better information, speed, or behavior — becomes central.

This Is Fine…

There are attempts to address these challenges.

Regulators have begun to focus on issues such as insider information and market integrity. Platforms are also introducing rules to restrict participation by individuals who may have the ability to influence outcomes directly.

Policy frameworks continue to evolve, and the regulatory landscape remains unsettled.

At the same time, some of the most interesting developments are coming from the industries being traded on.

Professional sports leagues, for example, have begun partnering with prediction platforms to monitor betting activity and flag irregular patterns. The same markets used to speculate on outcomes are now also tools for identifying potential misconduct.

It is a notable shift: systems designed to aggregate belief are now being used to detect deviations from expected behavior.

In industries characterized by information asymmetry, access to better data has always conferred an advantage. These developments suggest that the advantage is becoming more structured — and more closely monitored.

All’s Well That Ends Well

Stephens-Davidowitz argued that the most honest data emerges when people do not believe they are being observed; when they ask questions privately, without consequence.

Artificial intelligence has expanded that dynamic dramatically. People now externalize thoughts, concerns, and decisions in ways that generate rich behavioral data.

Prediction markets take a different approach. They enforce honesty by attaching consequences.

Both aim to reveal truth through behavior. Both are effective, in their own ways. But one carries an inherent tension: the moment meaningful incentives are introduced, behavior shifts.

A system designed to measure belief begins to influence it.

Search works because it feels private.

Prediction markets work because visibility — and consequence — are built in.

And when that balance shifts, the system changes with it.

Please read important disclosures at www.penderfund.com/disclaimer.

Standard performance data for Pender Funds can be found at www.penderfund.com/solutions/