The Price of an Opinion: what surveys and markets each know
There are two ways to find out what people expect from the economy. You can ask them, or you can watch what they do with their money. Right now, the two are telling opposite stories. The gap is the trade.
Sentiment says recession. Behavior says expansion.
The University of Michigan’s sentiment index printed 44.8 in May 2026, the lowest reading in the survey’s 74-year history, below the depths of the Great Recession and the COVID crash. Over the same stretch, real consumer spending kept growing, labor markets held, and equities pushed to record highs.
Even the two flagship surveys disagree with each other: the Conference Board’s confidence index sat near 93, around its historical average, in the same months Michigan was printing all-time lows. When two instruments asking similar questions land 48 points apart, the instrument is part of the story.
A Chicago Fed analysis found the once-reliable link between the Michigan index and real spending growth has collapsed to nearly zero since 2020. Part of that is mood, media, and partisanship: sentiment now swings sharply at every change of administration. Part of it is mechanical: economists Ernie Tedeschi and Ryan Cummings estimate the survey’s move to online collection alone made readings roughly 11% more negative. People are simply gloomier typing into a screen than talking to a human.
People spend through their own pessimism
This is the deepest problem with stated sentiment: respondents can tell a surveyor the economy is the worst they’ve ever seen, then book the flight, finance the car, and spend through the holidays. A Boston Fed study concluded years ago that once you control for income, wealth, and other fundamentals, sentiment adds only marginal power for predicting consumption.
Households panic. Breakevens shrug.
Nowhere is the split cleaner than inflation. Household 1-year inflation expectations spiked as high as 6.6% in spring 2025, driven by tariff headlines and grocery-aisle salience. Meanwhile 5-year-forward TIPS breakevens barely moved, holding near 2.2%, inside their pre-pandemic range (Richmond Fed).
Read together, the two lines say something neither says alone: households are anxious about prices now, but capital does not believe the Fed has lost the long game. Hot perceived inflation plus anchored priced inflation is a very different regime from both measures rising together, which would signal genuine de-anchoring and a far more dangerous policy problem.
One caveat cuts the other way: a raw breakeven is not a clean forecast. It bundles expected inflation with risk and liquidity premia. ECB researchers show that once premia are stripped out, market-implied expectations align far more closely with professional surveys. The teal line is disciplined, but it is not pure.
When they disagree, lean toward the money (mostly)
A Federal Reserve Board study of fifteen years of rate expectations found market-implied paths (OIS forwards) modestly beat Blue Chip survey forecasts on average, but the market’s edge widened dramatically in exactly the episodes where the two diverged. In 8 of 10 major divergence episodes, the survey had the largest forecast errors.
The election record tells the same nuanced story. Five presidential cycles of Iowa Electronic Markets data (1988 to 2004) show prices beating contemporaneous polls about 74% of the time, with the widest edge months out, when polls are noisiest. But Erikson and Wlezien found market prices largely follow the polls once polls exist. And a 2025 Vanderbilt analysis of $2.4 billion traded in the 2024 election found Polymarket the least accurate of the four major exchanges, pricing the eventual outcome correctly in just 67% of its markets, with the weakest results in thin, speculative contracts. The same cycle produced a well-documented whale who moved prices with tens of millions of dollars in directional bets.
The most robust finding is the least glamorous: combination wins. In the long-running PollyVote research program, blending polls, prediction markets, models, and expert judgment cut forecast error by half or more versus the typical single method.
Know which thermometer you’re holding
| Survey-Based | Market-Based | |
|---|---|---|
| Mechanism | Direct elicitation: structured questions to a sample | Inference from prices set by capital at risk |
| Sample | Broad, can be representative; reaches ordinary households | Self-selected traders; skews affluent, male, terminally online |
| Update speed | Weekly to monthly snapshots | Continuous, real-time |
| Failure mode | Cheap talk: partisanship, mood, mode effects, no cost to error | Expensive noise: risk premia, illiquidity, whale distortion |
| Tells you | What people feel, and why: perception and distribution | The probability-weighted consensus of informed money |
| Blind spot | Behavior: people spend through pessimism | Explanation: a price is a conclusion without a reason |
Divergence is not noise. It’s information.
When the amber and teal lines split, don’t average them reflexively. Interrogate the gap. Ask which instrument’s known failure mode best explains the disagreement. Sometimes it’s a broken survey. Sometimes a distorted price. And sometimes, as now, the gap is the most honest description available of an economy where how people feel and what money believes have genuinely come apart.
Use for perception, distribution, and the why. They reach people markets never sample and detect the mood that eventually shapes politics, wage demands, and demand for your product, even when it doesn’t show up in this quarter’s flows.
Use for probability-weighted consensus, especially at longer horizons and around discrete events. Adjust for risk premia where possible; discount thin markets and single-whale prints.
Agreement raises conviction. Divergence is a signal in itself. The current read: hot perceived inflation, anchored priced inflation, and a consumer who complains like it’s 2009 but spends like it’s 2019. Position for the flows, monitor the feelings.