Prediction markets have been sold to the public as something almost magical: a living, breathing barometer of collective wisdom, a place where money talks and the crowd is always right. In recent years, platforms like Kalshi and Polymarket have become the new darlings of political forecasting, with boosters calling them “truth machines” that cut through the noise of polls, punditry, and partisan spin. They claim that when people put their own money on the line, they have every incentive to be honest and smart, and the resulting market prices can reveal the true probability of an event better than any survey. And sometimes, they really do deliver. In 2024, Kalshi famously outperformed many traditional pollsters when it came to predicting the outcome of the U.S. presidential race, correctly signaling a decisive victory for Donald Trump in a way that many surveys missed. That success story fed the hype, making prediction markets seem not just clever but almost mystical. But here’s the thing about magic: it’s often just smoke and mirrors. The very same trading activity that produces those brilliant forecasts can also become a stage for bizarre, almost nonsensical behavior that makes you question whether the market is a window into human wisdom or a funhouse mirror reflecting our strangest impulses. The truth is that prediction markets can tell stories that make no sense at all, and the strange saga of the 2025 Ethiopian election markets on Polymarket is a perfect case study in how beautiful theories collide with messy reality.
On paper, the Ethiopian election looked like a perfectly boring political event. In late June, Prime Minister Abiy Ahmed’s Prosperity Party cruised to a landslide victory in the federal elections, winning 438 out of roughly 500 parliamentary seats. The closest rival party mustered just 13 seats. Abiy Ahmed, already a powerful and controversial figure, was set to serve another five-year term. There was no real suspense, no nail-biting finish, and no meaningful chance that the results would be overturned. You would expect prediction markets to reflect that dull certainty: one candidate with extremely high odds, and trading volume focused on the obvious winner. But what actually happened on Polymarket’s offshore platform was something entirely different. The market for who would become Ethiopia’s next leader somehow generated more than $285 million in trading volume, which is an astronomical sum for a single-country political market, let alone one where the outcome was a foregone conclusion. And yet, almost none of that money was bet on the person who actually won. Only about $173,000 was placed on Abiy Ahmed’s victory. Instead, more than $41 million was traded on his chief of staff, Shimelis Abdisa, becoming the next leader. Abdisa was not a serious contender, not a likely successor, and ultimately not even close. The remaining volume was spread across a host of other long-shot candidates, none of whom had any realistic path to power. In a sane world, this is not how a prediction market should behave. But in the strange ecosystem of online trading, where algorithms, bots, and suspicious account networks can distort everything, the market told a story that had no relationship to reality.
The weirdest part, however, wasn’t just that traders were betting on losers. It was how they were doing it. When Chris Park, a former Polymarket trader who now runs MSR Decode, a prediction-market forensic analysis firm, dug into the data, he found something that looked less like genuine speculation and more like a strange, automated dance. Nearly all of the $285 million in trading volume—about $284 million of it—came from pairs of accounts that were repeatedly buying and selling contracts with each other, back and forth, over and over again. This is what experts call wash trading, a form of market manipulation in which the same economic interest is traded between related parties to create the illusion of activity and liquidity. It’s like two people standing in an empty room, applauding each other endlessly, pretending there’s a standing ovation. There was no actual stream of outside money flowing into the market, no wave of genuine conviction from thousands of traders who believed Shimelis Abdisa was the future leader of Ethiopia. Instead, there were accounts passing the same contracts to each other, racking up an eye-popping volume number that made the market look supremely active, even as the real-world political situation remained static and hopeless for the long-shot candidates. Rajiv Sethi, an economics professor at Barnard College who studies prediction markets, looked at the data and was blunt in his assessment. “It’s obviously some form of volume manipulation,” he said. That kind of accusation strikes at the very heart of what makes prediction markets trustworthy. If you can’t distinguish between a genuine signal about future events and a choreographed illusion manufactured by bad actors, then the platform’s much-vaunted ability to reveal the truth becomes dangerously unreliable.
So what did the people running Polymarket do about it? When asked, Shana Bautista, the company’s head of global intelligence and investigations, told WIRED that the team “immediately flagged this activity, investigated, and blocked the accounts from trading.” That sounds reassuring, like the steady hand of a vigilant security team. But it raises an uncomfortable question: if they immediately flagged it, why were these accounts allowed to generate so much volume in the first place? Why did it take more than $284 million in what looks like wash trades before someone pressed pause? Polymarket did not respond to follow-up questions about the timeline, and that silence speaks volumes. In a market where billions of dollars are at stake, and where retail traders are encouraged to trust the platform as a reliable oracle, allowing this kind of manipulation to run unchecked—even temporarily—is a serious breach of confidence. It also suggests that the safeguards on offshore, unregulated platforms may not be nearly as robust as their marketing suggests. The fact that a handful of coordinated accounts could inflate a minor political market to the size of a major international event is not a minor glitch. It’s a structural vulnerability. If prediction markets are to be taken seriously as modern-day oracles, they need to prove they can distinguish between genuine collective intelligence and the output of a few bad actors running a psychological experiment with other people’s attention. Instead, this incident exposed how easy it is to game the system, and how slow the response can be when the house is making money from the very activity it should be policing.
And then there was the timing, which made everything even more inexplicable. The Ethiopian election happened in June, and Abiy Ahmed’s victory was obvious within days. But Polymarket’s rules stated that the Ethiopia market would not be resolved until October, after the prime minister was formally appointed and sworn in by parliament. In principle, that makes sense: a market should stay open until the event is officially concluded, not just when it becomes probable. But in practice, it created a bizarre, multi-month limbo period. For 123 days after the election, with the outcome all but a mathematical certainty, bursts of seemingly irrational trading continued to erupt on the platform. The wash-trading pairs kept buying and selling, racking up volume on candidates who had no hope of winning. Market participants grew increasingly confused and frustrated. On Polymarket’s discussion boards, users posted comments like “THIS NOT RESOLVED?” and “fuckkkk,” expressing the kind of bewildered irritation you might feel watching a traffic light stay red at an empty intersection. Everyone could see the obvious truth: Abiy Ahmed had won, the votes had been counted, and the world had moved on. But the market, that self-proclaimed truth machine, remained stuck in a strange loop, refusing to acknowledge what was happening in the real world. The outcome was no mystery, and the market did nothing to illuminate the future. Instead, it became a monument to technicality and manipulation, a place where the only thing being measured was the patience of the people waiting for its creators to clean up the mess.
What should we take away from all this? First, it’s a reminder that prediction markets, for all their impressive moments, are not magical. They are tools built by humans and vulnerable to human flaws, including greed, deception, and sheer irrationality. A million dollars moving around a market might mean millions of independent minds converging on the truth, or it might just mean a few bad actors swapping the same dollars back and forth to make the platform look bigger and busier than it really is. The story of the Ethiopian election market is not proof that prediction markets are useless; Kalshi’s 2024 success shows they can sometimes outperform traditional methods. But it is proof that they need scrutiny, accountability, and regulation. When a platform allows billions of dollars in fake-looking volume to churn across its servers, and when it stays silent about how many accounts were involved and how much money they made, the trust that powers the entire enterprise begins to crack. For the everyday trader, the lesson is to be skeptical of impressive volume numbers and slick interfaces. For the platforms themselves, the lesson is that transparency isn’t optional; it’s the whole point. If prediction markets want to be taken seriously as barometers of public opinion and tools for forecasting the future, they cannot simply clean up after the damage is done. They must build systems that make manipulation nearly impossible from the start. Otherwise, they risk becoming just another carnival attraction—entertaining to watch, but not something you’d ever rely on to tell you where the world is heading. The truth machine, it turns out, is only as good as the people who press its buttons.
