What Pundits, Pollsters, and Portfolio Managers Have in Common
It’s August in America, heavy, slow, and humming. The fields are full but not yet ready for the harvest. The heavy air blurs the horizon, and in the country the steady drone of insects is as ubiquitous as the constant jabbering of pundits on cable television.
Political correspondents call August the silly season. Congress is in recess, and millions of families are on vacation. As real news grows scarce, the political prediction season moves into high gear. Pundits sharpen their midterm narratives while rhetoric ramps up for the main event: the federal elections. Every tension between the parties and the branches of government, real or imagined, inflates to fill the space where actual news used to be.
This article isn’t about who wins in November. It’s about narrower, more useful questions: Do predictions—about politics, about sporting events, about the Oscars, about the movement of stock prices—actually have any value?
Can anyone reliably predict what will happen? If they can get that right, can they predict what the outcome will actually mean? Investors might want to pay particular attention here, because they swim in an ocean of predictions.
We’ll examine the prediction industry, look for useful parallels, and offer some advice. What we’ll try not to do is make any predictions; we’ll stick with the data instead.
Predictions Aren’t Free
Prediction is not a hobby; it’s an industry—a large, well-capitalized, rapidly growing one, with its own firms, its own analysts, its own quarterly earnings calls. In the United States alone, the business of measuring and forecasting public opinion generates somewhere between $3 and $4 billion a year, and that figure is only the polling slice of a much larger market-research industry worth tens of billions more. Layer in the newest entrant—prediction markets like Kalshi and Polymarket, where people now wager real money on everything from elections to Fed decisions—and the total climbed to an estimated $36 billion in trading volume in the first quarter of 2026 alone.
That is not a rounding error. That is a genuine, thriving economy built entirely around selling certainty in an uncertain world. Like any industry, its size is not evidence of its accuracy; it’s evidence of its demand. People pay for certainty, at scale, whether or not certainty is the thing actually being delivered. Spoiler alert: it isn’t.
There’s a strange thing true certainty does to a market: it destroys the market’s reason to compensate anyone for risk. Certainty, priced honestly, is worth almost nothing. A risk premium is compensation for not knowing—the extra return demanded for holding an outcome that could go either way. Remove the uncertainty entirely, and that particular premium goes with it.
A Treasury bond still pays a yield, but that yield compensates for time and opportunity cost, not risk. Its default risk premium is effectively zero, because the outcome isn’t in doubt. Extend that logic to a genuinely certain forecast: if an outcome were truly, perfectly knowable, whatever premium exists in its price would reflect only the cost of waiting, never the cost of being wrong.
Despite all of this, consumers continue to pay handsomely for certainty. What underlies this dichotomy is not complicated, once you understand it. What’s being sold was never actual certainty. It was the feeling of it, a product priced at a premium precisely because the underlying good doesn’t exist. There’s a deeper irony buried in here too.
To sell certainty, a forecaster must first be certain about their own certainty. That means confident not only in the outcome, but in the reliability of the method that produced the forecast. That’s a claim layered on a claim, each one unverifiable until the very event it’s predicting has already happened. Which raises the only question that actually matters, and the one the rest of this piece is built to answer: When forecasters sound certain, how often does that confidence turn out to have been validated?
Before examining whether that confidence is earned, it’s worth asking what it would even mean for it to be earned—and then checking the record.
Uncertainty Is a Feature, Not a Bug
Let’s start with the most rigorous study ever conducted on expert political judgment. Psychologist Philip Tetlock spent nearly two decades tracking predictions made by 284 professional forecasters—political scientists, economists, journalists, and other people whose actual job was providing certainty. Eventually, he accumulated 82,361 individual forecasts, checked one by one against what actually occurred. The average expert performed roughly in line with random chance. Many did worse than a simple statistical model that made no attempt at expertise at all. The finding is famous enough now to have earned its own meme: the average pundit forecasts about as well as a dart-throwing chimpanzee.
Markets tell a parallel story, and this one comes with names attached. CXO Advisory Group graded 6,582 individual market-direction calls made by 68 named financial experts over roughly fourteen years, checking each one against what actually happened. The average accuracy over thousands of iterations was 47 percent, worse than a coin flip. Only 5 of the 68 experts topped 60 percent. Jim Cramer landed around 47 percent, while Goldman Sachs strategist Abby Joseph Cohen came in at 35 percent. These are not anonymous, hedged estimates. These are named professionals, graded on the record, against outcomes nobody disputes.
Then there’s the story people think they remember about elections, which isn’t quite the story that happened. In 2016, several prominent models gave Hillary Clinton overwhelming odds of victory. The Princeton Election Consortium put her above 90 percent, and multiple aggregators had her near 98. But the model that actually earned its reputation that year, FiveThirtyEight’s, gave Trump roughly a 3-in-10 chance, high enough that his eventual win fell well within the range the model itself described as plausible. The failure in 2016 wasn’t really a polling failure. It was a confidence failure: the loudest, most certain voices were the ones furthest from being right, while the most hedged forecast was graded, after the fact, as the least wrong.
2024 rhymed with 2016. Nate Silver, the man vindicated in 2016 for his caution, put the race at 50.015 percent to 49.985 percent and called it, without any real exaggeration, “closer than a coin flip.” He also flagged a high probability of a swing-state clean sweep as the single most likely outcome. Both calls held up: Trump swept all seven battleground states, and Silver’s state-by-state polling was accurate in 48 of 50 states nationwide.
What actually surprised people wasn’t a model failure—it was that the broader public conversation, hungry for a confident storyline, had drifted well past what the most careful data ever supported, then treated the result as a shock precisely because it had stopped listening to its own uncertainty.
Run these three data sets together, and a pattern emerges that has nothing to do with politics or finance specifically. It’s never the hedge that fails. It’s the false confidence.
An Unsolvable Problem
None of this is an argument for ignoring the news, or pretending elections don’t matter. It’s an argument for treating every confident forecast, whether political, financial, or algorithmic, as information about the forecaster, not information about the future.
Pundits, portfolio managers, and prediction markets alike are all trying to solve the same unsolvable problem: calling the outcome of a system that changes the moment it’s observed.
The family office approach isn’t to out-predict the predictors. It’s to build a strategy resilient enough that being wrong about the next big event simply doesn’t matter, one where the plan was never counting on anyone’s certainty in the first place, including its own.
Before his rematch with Rocky Balboa in Rocky III, a reporter asked Clubber Lang to predict the fight. He didn’t hesitate, but delivered a deathless line: “My prediction? Pain.” You don’t have to be a Rocky fan, or a Sly Stallone fan, or even a boxing fan, to appreciate why that landed like a knockout punch. All you have to have done is lived for a while.
Clubber wasn’t wrong; the fight proved it. He also wasn’t really predicting the outcome. He was predicting the nature of the contest itself. That’s the one kind of forecast this piece has found any evidence for: not who wins, not by how much, but that uncertainty has a cost, and it comes due whether or not you saw it coming. The only real defense is a plan built to absorb it.
The business of prediction will keep doing what it does, with sharper tools, more data, and, increasingly, AI models trained to sound certain, but no more actual certainty than it ever had. Your plan doesn’t need it to predict future events. It needs to be ready for the uncertainty certain to come.
A steadier way to move through prediction season, and a long investing career, starts with a real conversation backed by real information, not another forecast. A complimentary Taxes First, Then Math™ analysis could be the perfect place to start managing uncertainty.
