The University of Michigan released its final April reading on Friday: 49.8 on the Consumer Sentiment Index, the lowest number in the survey's seventy-four-year history. It is worse than the trough of June 2022, when post-pandemic inflation peaked. It is worse than April 2009, when unemployment was approaching 10 percent. It is worse than any month of the financial crisis, the dot-com bust, the 1990 recession, or the early-1980s Volcker shock.
The unemployment rate today is 4.3 percent. Year-over-year inflation is 3.3 percent. Real wages have grown. Three of the four lowest sentiment readings ever recorded have occurred within the past nine months.
One of these things does not match the others. The conventional explanation—that consumers know something the data does not—has had a long run, but it has now collapsed under the weight of its own implications. If the Michigan index is to be believed, the United States in April 2026 is suffering its worst economic moment since the Eisenhower administration.
Nobody actually believes this, including the people answering the survey. What they are doing instead is something the Federal Reserve needs to acknowledge before it sets interest rates off the result.
What the survey actually measures now
The Michigan Survey of Consumers asks respondents how they feel about the economy. For most of its history, the answers tracked recognizable economic variables—income, employment, prices, the stock market—with reasonable fidelity. They no longer do.
The April 2026 final reading shows Democrats at 31.8 and Republicans at 87.1, a 55-point spread with Republicans now the in-party. The trajectory is what matters. In February 2009, in the depths of the actual financial crisis under Obama, the partisan gap was four points (Democrats 59.6, Republicans 55.1)—Democrats slightly higher as the in-party, but the partisan effect was tiny against the backdrop of a real economic emergency that both sides could see. By the June 2022 inflation peak under Biden, the gap had grown to 33 points (Democrats 66.4, Republicans 33), still with Democrats higher as the in-party but with the partisan amplification now overwhelming the underlying economic signal. By April 2026, with the in-party flipped to Republicans, the gap is 55 points and the absolute level of the index has hit an all-time low. Whatever the survey is measuring now, it is not what it measured in 2009.
This is not a recent or speculative finding. Carola Binder of Brookings documented in early 2023 that during the Biden inflation surge, the entire distribution of Democrats' inflation expectations remained essentially unchanged in 2021 and 2022 while the distribution of Republicans' expectations shifted upward. The partisan gap was not the result of Democrats and Republicans having different information. The same prices were on the same shelves. It was the result of partisans using the survey as an opportunity to broadcast their political affiliation. Binder noted with restraint: if Democrats and Republicans really had had drastically different inflation expectations during 2021 and 2022, they should have made very different investment and consumption decisions. They did not.
Menzie Chinn at Econbrowser has shown that during the inflation surge, realized inflation tracked Democrats' expectations far more closely than Republicans'. Republicans, in other words, were systematically wrong. Now Democrats are systematically wrong, and the magnitudes are larger.
The University of Michigan's own researchers have acknowledged the issue in an April 2025 white paper. Their defense, in essence, is that independents continue to track national medians and that aggregate trends move in parallel across parties. This is true. It is also a concession: the partisan components are no longer reliable indicators of partisan economic conditions, only of partisan affiliation.
There is a corollary that bears saying out loud. The Conference Board's Consumer Confidence Index rose to 92.8 in March, its third consecutive monthly increase, while the Michigan index was setting record lows. Two surveys, the same population, the same month, opposite signals. They are measuring different things now. Michigan, with its detailed questions about prices and personal finances, has become more sensitive to media diet than to economic conditions. The Conference Board, more focused on labor markets and present circumstances, has held up. The divergence is the data.
Why this matters for monetary policy
The Federal Reserve's longer-run goals statement, reaffirmed at Jackson Hole in 2025, places "well-anchored longer-term inflation expectations" at the center of the Committee's price stability mandate. Wednesday's FOMC statement explicitly listed "inflation expectations" among the inputs the Committee will use in setting policy. At his March 2025 press conference, Chair Powell cited the Michigan five-year inflation expectations figure of 3.9 percent directly when explaining the Committee's caution about cuts. The April 2026 readings put year-ahead inflation expectations at 4.7 percent and long-run at 3.5 percent, the highest since late 2025. The Fed held rates unchanged on Wednesday at 3.5 to 3.75 percent. Inflation expectations were part of the reason.
If those expectations are partly a partisan signaling device rather than a forecast, the Fed is making rate decisions off a contaminated input. The contamination is not subtle. A 4.7 percent year-ahead expectation reading that emerges from Democrats projecting 7 percent and Republicans projecting 2.5 percent is not the same statistical object as a 4.7 percent reading that emerges from broad consensus. The first is a tribal flag with a number stamped on it. The second is a forecast. The Fed, treating both alike, ends up tightening or holding for the wrong reasons in the wrong direction at the wrong time.
Powell himself has flirted with the implication. Asked about the Michigan numbers in March 2025, he noted that "policymakers have not seen that in other surveys or in bond markets." This was the right observation and the wrong response. If the headline survey diverges from professional forecasts, market-implied breakevens, and competing consumer surveys, the appropriate institutional response is not to mention the divergence at the press conference and continue weighting Michigan as before. It is to formally downgrade Michigan as a policy input.
The market knows what the survey does not
There is, conveniently, an alternative, and I have written about it twice for Bloomberg this year. In ["Can Kalshi Take On Inflation Too?"](https://www.bloomberg.com/opinion/articles/2026-01-02/can-kalshi-take-on-inflation-too) (January 2), I examined the prediction market Kalshi's claim to forecast CPI more accurately than the Wall Street consensus. In ["Kalshi and Polymarket Are Economic Oracles"](https://www.bloomberg.com/opinion/articles/2026-02-27/kalshi-and-polymarket-are-economic-oracles) (February 27), I reviewed two academic papers on prediction-market forecasting performance, including the Federal Reserve's own *Kalshi and the Rise of Macro Markets*. The April sentiment data has sharpened the case considerably.
The Fed paper is worth dwelling on, because the FOMC has been ignoring its own staff. The researchers compared Kalshi prices against the New York Fed's Survey of Market Expectations, Bloomberg's consensus forecasts, and federal funds futures. Kalshi prices were of comparable accuracy to the surveys overall and beat the Bloomberg consensus on year-over-year CPI specifically. The mode of the Kalshi distribution has matched the realized federal funds rate by the day of each FOMC meeting since 2022—better than surveys, better than futures. Most importantly for monetary policy, Kalshi was particularly strong at quantifying tail risks, which is precisely what the Fed needs for inflation-anchoring decisions and which surveys handle worst.
The deeper finding from the same paper is that Kalshi delivers something surveys cannot: the full distribution of expectations. The economy is driven by hundreds of millions of decisions by individuals, not by the median professional forecaster. When a Kalshi market on December CPI shows a bimodal distribution with peaks at 2.55 and 2.65 percent rather than a single bell curve, that is not noise. As I argued in the January column, that bimodality almost certainly reflects partisan rooting—Trump supporters betting on the lower number, opponents on the higher. The same partisan signaling that contaminates Michigan inflation expectations is visible in Kalshi prices too.
December CPI came in at 2.7 percent year-over-year, above both Kalshi peaks. The market was biased low, with both partisan camps underestimating, though the higher-expecting peak at 2.65 percent was closer to the actual print than either the lower peak at 2.55 percent or the Michigan one-year-ahead reading of 4.2 percent at the time, which overshot reality by a percentage and a half. The result was not a triumph for prediction markets and it was not a vindication of surveys. It was a reminder that even the contaminated Kalshi distribution carried more usable information than the cleaner-looking single number from the most-cited consumer survey. The bimodality itself was the signal: the market was telling anyone willing to read it that partisan disagreement, not measurement error, was driving the spread.
The difference is what happens next. On Kalshi, partisan bettors face arbitrageurs who care about money rather than politics. If the true probability of a 2.59 percent reading is materially higher than the price, professional traders will buy contracts at 2.59 and sell at 2.55 and 2.65 until the distribution converges toward something closer to the truth. The mechanism is imperfect—the Kalshi paper documents a favorite-longshot bias and I noted in the January column that arbitrage was incomplete—but the mechanism exists. There is a force pulling Kalshi prices back toward reality. There is no equivalent force in the Michigan survey. A respondent who tells the surveyor she expects 8 percent inflation pays no penalty when actual inflation comes in at 3 percent. The signaling is free.
This is the structural point. Surveys ask people what they think. Markets ask people to put money on what they think. Both can be contaminated by partisan affiliation, and as my January column documented, both currently are. Only one has a mechanism for cleaning up the contamination over time, and only one produces forecasts that the Fed's own research staff has certified as competitive with the most expensive professional consensus on Wall Street.
A modest proposal
The fix is straightforward and does not require abandoning the surveys. As I noted in the February Bloomberg column, the optimal forecast in tested cases combines prediction-market prices with expert judgment in roughly 40-60 weights. The Fed should formally adopt the same approach for inflation expectations. The Michigan inflation expectations series should be downgraded relative to its current weighting in policy assessments, given the partisan contamination documented by Brookings's Carola Binder, the University of Michigan's own April 2025 acknowledgment, and the divergence from market-implied measures and the Conference Board. Kalshi-implied probabilities should be elevated alongside TIPS breakevens and the New York Fed's Survey of Consumer Expectations. The FOMC should explain the change publicly, both because transparency about policy inputs is itself anchoring, and so that future presidents understand which numbers carry weight and which do not.
This will not happen quickly, because central banks rarely admit that a long-relied-upon input has decayed. But the cost of not doing it is real. The Fed is currently holding rates higher than the underlying data warrants, in part because the Michigan survey is producing readings that reflect Democratic dismay at the Trump administration rather than household inflation forecasts. Two years ago it produced readings that reflected Republican dismay at the Biden administration. The errors run in both directions and they accumulate. The Federal Reserve does not need to weight a partisan mood ring at the same level as bond markets and prediction markets when setting the price of money for an economy of three hundred forty million people.
The Michigan index is not broken. The respondents are doing exactly what political scientists would predict that polarized partisans do when handed a microphone. The error is the Fed's, for continuing to listen to a microphone the speakers no longer use to communicate economic information. The country has invented a better measure. The Fed's own researchers have certified it. The FOMC's job now is to act on what its staff already knows.
