More accurate than polls, more dangerous than expected: The Federal Reserve's view on prediction markets

Source: The Token Dispatch

Author: Prathik Desai

Original Title: The Signal and the Noise

“Forecasts often tell us more about the forecasters than about the future.”

— Warren Buffett

Money can filter out nonsense. Supporters argue this is precisely why prediction markets are reliable. We’ve seen people accurately forecast the 2024 U.S. presidential election results on Polymarket and Kalshi. However, prediction markets themselves are not new, and their success in political outcome predictions is not the first of its kind.

In October 1988, a group of economists at the University of Iowa used a small, real-money prediction market to satisfy their academic curiosity. They launched a presidential election futures market where participants could buy contracts: if George H. W. Bush won, the contract pays $1; if Michael Dukakis wins, it pays $0. On election eve, Bush’s contract traded at 53 cents, while traditional polls showed a tight race. Ultimately, Bush won with 53.4% of the vote and an 8-point margin.

Since that academic experiment, these real-money futures markets have outperformed traditional polls in every election held more than 100 days in advance. In U.S. presidential elections since 1988, prediction markets have been closer to the final result 74% of the time than polls.

This success stems from a mechanism that forces people to express genuine beliefs backed by real money—something surveys can never do. Those who truly believe Bush will win buy and hold contracts. For casual participants, there’s little incentive to spend $50 supporting a claim they don’t believe in. When thousands of traders act this way, information converges into a price that reflects the collective’s true belief, rather than a small, unrepresentative sample.

That small academic experiment in Iowa, operating on a modest budget, has now evolved into an institutionalized infrastructure.

Last week, a working paper authored by Federal Reserve economists highlighted that the largest regulated prediction market in the U.S., Kalshi, can serve as a valuable real-time benchmark for policymakers. The same week, Lynn Martin, President of the New York Stock Exchange (NYSE), stated that the world’s largest prediction market, Polymarket, influenced the S&P futures on election night by pricing Donald Trump’s victory earlier than any news outlet. Subsequently, Kalshi announced a partnership with a trading platform handling $2.6 trillion in daily institutional volume.

In this deep dive, I will explore whether prediction markets can serve as reliable indicators for policy decisions and what risks they entail.

Prediction Markets as Policy Tools

The paper finds that Kalshi’s forecasts are statistically similar to Bloomberg’s consensus expectations, with nearly identical errors in predicting core CPI and unemployment rates. In fact, the study also found that Kalshi’s predictions for core CPI significantly outperformed Bloomberg’s estimates.

@FederalReserve

Despite similar statistical performance, Kalshi’s unique advantage is its ability to provide more frequent, real-time probability curves for macroeconomic indicators like GDP growth, core CPI, and unemployment. For estimates like inflation, Bloomberg consensus data is only available months before release. This results in lower-frequency traditional estimates that can’t reflect real-time expectation updates, leaving a long gap.

Kalshi not only predicts outcomes but also offers real-time uncertainty ranges and tail risk assessments. In early April 2025, uncertainty around trade policy temporarily boosted inflation expectations. Although this uncertainty ultimately did not materialize, Kalshi priced in the dynamic of this change in real time. Monthly Bloomberg estimates can never capture such subtle shifts.

@FederalReserve

Today, when Federal Reserve officials speak at FOMC meetings, Kalshi’s market odds fluctuate in real time. They price each statement, providing policymakers with a perspective on how traders interpret the communicated expectations.

For example, when Christopher J. Waller made dovish comments before the July 2025 FOMC meeting, the probability of no rate cut dropped to 75%. After a stronger-than-expected jobs report in June, that probability quickly rose back above 90%. The entire expectation, supported by real money, is presented to policymakers in a way that no other tool currently can match.

Who Is Trading in These Markets?

Before trusting prediction markets too much, it’s important to examine who is trading and what the volume represents.

From September 2024 to January 2026, trading volume on Polymarket for FOMC meetings increased elevenfold, from $59 million to $660 million. In total, Polymarket’s FOMC market handled $2.6 billion, surpassing the combined volume of its culture, economics, geopolitics, and science categories.

So, who is making such large trades on FOMC? While it’s hard to identify on anonymous platforms like Polymarket, we can speculate: it’s likely macro hedge fund analysts involved in drafting labor reports, or money market fund managers who profit if rates don’t cut.

Why them? The Iowa market worked because the people who acted consistently and invested significant funds outnumbered those merely gambling without reliable information. Recognizing the risk of over-hedging, I believe that when real stakes and capital are involved, those with credible information tend to gather in the market, leading to more accurate price discovery.

What to Watch Out For

This doesn’t mean prediction markets are perfect policy gauges.

Probabilities in prediction markets also reflect traders’ risk preferences, not just their expectations of outcomes. For example, when Kalshi prices the probability of unfavorable CPI data at 15%, while traditional surveys price it at 10%, two factors could explain this gap:

1. Prediction markets might be pricing in real-time information overlooked by Bloomberg consensus.
2. Traders might be paying a premium in prediction markets to hedge against adverse outcomes.

Policymakers must understand what this gap signifies before treating these signals as policy cues.

While Kalshi’s macroeconomic signals seem reliable, over 85% of its total nominal volume comes from the sports category.

@Dune

Currently, at least 20 federal lawsuits challenge the regulatory arbitrage of prediction markets through nationwide sports betting.

The reliability of Kalshi’s FOMC market partly depends on sports betting, which provides liquidity through active traders, narrow bid-ask spreads, and market-making infrastructure. Although macro markets operate independently, they benefit from this foundation. If sports betting faces regulatory pressure and disappears, the platform would lose its liquidity engine that maintains narrow spreads and continuous prices. Thinner macro markets become easier to manipulate with less capital.

The Fed’s paper suggests using Kalshi as a monitoring tool, not a decision input. But publicly stating this intention is itself problematic.

The author recommends more use of Kalshi to interpret incoming data and monitor real-time Fed communications. However, since referencing prediction markets is public, it could create feedback loops.

For example, a policymaker might see Kalshi’s pricing of a 15% chance of rate cuts, lower than their desired message. In response, they might soften their language in the next speech, which could then trigger volatility in global traditional interest rate markets. The issue is that while Kalshi’s FOMC market has a $660 million volume, the federal funds futures market exceeds hundreds of billions. The former requires only a relatively small position to influence odds. An informed participant with enough capital to realize that moving Kalshi could influence Fed statements (even if not directly controlling them) could leverage small positions to sway a much larger market. This could turn policy communication into an instrument of manipulation.

This highlights the difference between the 1988 Iowa futures market and the prediction markets of 2026. Back then, Iowa economists only aimed to see if a market with real stakes could produce better predictions than surveys. Policy wasn’t as closely scrutinized, so manipulation was less of a concern.

At that time, prices reflected genuine beliefs because they didn’t influence the world. They merely allowed insightful individuals to monetize their views. Once the Fed publicly announces (if it ever does) its intention to use prediction markets as policy inputs, this property disappears. It also introduces a “performative” element to trading.

However, incorporating prediction market odds into the policy toolkit isn’t a mistake. Financial commitments still filter out noise. Informed participants continue to lead price discovery. The resulting signals, in speed and distribution, surpass surveys in real-time reflection of expectations. For FOMC markets, this is especially true: both sides have participants with genuine hedging capacity, and markets that frequently price real-time events better reflect current expectations.

Policymakers should mandate transparency of open-source data as a prerequisite for formal adoption. If data can’t be audited, manipulation may go unnoticed. They must understand that signals and noise originate from the same source. Those with real money and genuine beliefs can tell you their thoughts in real time.

For those powerful enough to game the system, this window of opportunity didn’t exist during decades of Iowa academic experiments. Today, that window is wider than ever.