In October 1964, Nikita Khrushchev was deposed, and the West’s experts missed it. Asked why he had not predicted the coup, Harvard’s Adam Ulam gave the all-time answer: “If it came as a surprise to Khrushchev, why wouldn’t it have come as a surprise to me?”
That story opens a book review in the December 31, 2025 Journal. It also opens this piece, because we just graded a year of forecasts against what actually happened. Ulam’s defense held up better than most of the predictions.
Rule one: shorter is smarter
Watch what happened to one forecast — year-end 2026 inflation — as the deadline got closer. The nearest actual reading: July CPI printed at 3.4% in the August 13 paper.
| Survey | Year-end 2026 CPI call | Distance from the July actual (3.4%) |
|---|---|---|
| January 2026 | 2.6% | nearly a full point below |
| April 2026 (Apr. 13 paper) | 3.2% | 0.2 point below |
| July 2026 (Jul. 13 paper) | 3.4% | matches |
Same panel, same question, same year. The only thing that changed was the distance to the target.
One honest caveat, since this piece is about cherry-picking. All three of these are calls for December 2026, and December has not happened. The July survey’s year-end number happens to match the July actual, which is exactly the kind of coincidence a scorekeeper should not cash. None of these forecasts comes due until the December reading lands.
The jobs consensus taught the lesson twice. In the February 12 paper, January payrolls came in at 130,000 and, in the Journal’s own words, blew far past consensus expectations. In the August 8 paper, July payrolls went the other way: a 23,000 loss against an expected 83,000 gain, a miss of 106,000. Same experts, wrong in both directions — and only the July miss came with a published expectation to measure it against.
Rule two: the headline number is the least useful part
Oil produced the year’s cleanest grade. In the March 20 paper, Saudi oil officials’ base case had prices topping $180 a barrel if disruptions persisted. Wood Mackenzie put $200 inside the realm of possibility. In the April 2 paper, Société Générale’s Michael Haigh floated $200 too, and admitted, admirably, that he was guessing in unprecedented conditions.
The grade came in the July 13 paper. Oil was $67.02 on the eve of the war. It peaked at $112.95 in April. By July 6 it was back to $68.55.
Note the important part: the stated condition actually held. Shipping through Hormuz was close to a halt again in the July 11 paper, and mediators were still patching a temporary fix in the August 6 paper. The disruption persisted, and oil still finished nowhere near $180. The scenario was right. The number was noise.
Now the fair part, because the rule works in both directions. The April 13 survey called year-end West Texas crude at $79.66, about 18% below that Friday’s price. The August 28 paper had it at $83.53 — within $4, four months before the deadline. The same profession that missed the drama nailed the destination.
The best forecast of the year was barely a forecast at all. The March 20 survey asked how high crude would have to climb, and for how long, to push recession odds past 50%. Answers ranged from $90 to $200, averaging $138, sustained for an average of 14 weeks. A trigger, a range, and a chart of dots. The trigger never hit, and no recession came. That is a forecast doing honest work: it tells you what to watch instead of what to feel.
Rule three: grade the reasoning, and demand a deadline
Goldman Sachs entered 2026 friendly on stocks partly because Fed rate cuts were coming, per the December 31, 2025 paper. The S&P 500 rose 12.9% through the August 28 paper. Cuts delivered: zero. Right answer, wrong reason. Half credit.
Goldman’s own Jan Hatzius said in the April 2 paper that if Hormuz reopened by mid-April, the U.S. would get modestly slower growth but no recession. Hormuz did not cooperate. No recession came anyway. Right answer, failed condition — luck sitting next to judgment, and worth admitting.
Full credit goes to the unglamorous. Kansas City Fed President Jeff Schmid and a contingent of hawks projected no 2026 cuts, per the March 10 paper, and the target range never moved through August. The dull call, made early and held, beat every exciting one.
The deadline rule comes from bitcoin. The December 31, 2025 paper graded a year of price targets: $200,000 calls from Bernstein, Standard Chartered and Bitwise, against a coin around $88,000 and down about 6% on the year. BlackRock’s Larry Fink had floated $700,000 as possible — with no time span attached. That call can never be wrong, because it never comes due. A forecast without a date is not a forecast. It is a mood.
That December 31 review ends near where we will. The reviewer suggests forecasters might serve people better by helping them handle uncertainty than by pretending to abolish it. That happens to be a portfolio’s job description. Diversification is Ulam’s honesty, expressed in percentages.