Last year's inflation just got rewritten

On 10 February the Bureau of Labor Statistics published its routine annual seasonal revisions to the consumer price index, and in doing so rewrote the inflation record for late 2022. The story behind a months-long rally in risk assets had been simple. Monthly inflation had fallen sharply in the fourth quarter, the argument went, so price stability was coming back fast. The revised figures show that much of that improvement was a seasonal quirk. Core CPI for October, November and December was revised up, and underlying price pressure turns out to have barely eased.

How the calendar is removed

Seasonal adjustment is meant to strip out predictable calendar effects, such as holiday shopping or summer travel, so that the trend shows through. After the upheaval of the pandemic, the standard models struggled to tell lasting shifts from temporary noise.

The method works by comparing each month with the same month in earlier years. If prices have tended to jump every January, some of this January's increase is treated as normal and removed; if they have tended to fall in December, a December fall is discounted. Each February the statisticians re-estimate those factors with a fresh year of data and revise the seasonally adjusted figures for the previous five years. The raw, unadjusted prices do not change. What changes is the judgement about which part of each month's move was the season and which was the trend.

The trouble is that the past few years have been anything but normal. Prices collapsed in the spring of 2020, surged in 2021 and then jumped again in the first half of 2022 as energy costs soared after Russia's invasion of Ukraine. A model looking for repeating patterns saw big increases early in the year and concluded that early-year months were naturally hot. It therefore adjusted those months down and, since the factors must balance over a year, pushed the later months up once the new data were included.

Three per cent was really four and a half

Core inflation in late 2022 did not slow to an annualised pace near 3 per cent, as first reported. On the revised data it stayed above 4.5 per cent. The winter disinflation that had charmed markets came from faulty statistical weights, and demand had not really weakened.

The headline figures shifted too. December's all-items index, first reported as a small monthly fall that had delighted markets, now shows a small rise. The difference sounds trivial, but monthly changes are what traders annualise to judge momentum, and a tenth of a point a month is more than a percentage point over a year.

Bond desks caught out

That is embarrassing for the bond desks that had priced in an early end to Fed tightening. Trades built on the old data are now being unwound one by one.

The positions most exposed are those that bet on a quick fall in short-term interest rates: long positions in two-year Treasury notes, bets on a steeper yield curve, and options that pay out if the Fed cuts in the second half of the year. Inflation-linked bonds come out better, since a stickier underlying trend supports the compensation investors demand for future inflation. Equity valuations, which had risen on hopes of cheaper money, are vulnerable for the same reason.

This has happened before

Statistical artefacts have misled markets before. In 2015, economists spent months arguing that repeated weakness in first-quarter GDP reflected faulty seasonal adjustment rather than a real slowdown, and the Commerce Department eventually overhauled its methods. The lesson then, as now, is that a few months of adjusted data can carry a pattern that exists only in the model. Investors who bought the soft months of late 2022 as proof of disinflation made the same mistake in reverse.

Powell was right to wait

For the Fed, the revisions justify its refusal to go along with market bets on rate cuts. Jerome Powell and his colleagues had warned repeatedly that a few soft months did not prove inflation was beaten.

The Fed's own focus had already shifted to where the revisions bite hardest. Mr Powell said late last year that the key to the outlook was core services excluding housing, a category dominated by labour costs, and that category has shown little sign of cooling. Goods prices have fallen as supply chains recovered, and rents on new leases have eased, but neither of those tells the Fed much about wage-driven inflation in the service economy.

The costs of error are also lopsided. If the Fed eases too early and inflation reignites, it must tighten again from a position of lost credibility, which is what happened repeatedly in the 1970s. If it holds too long, the cost is a weaker economy, which it can address by cutting later. Faced with that asymmetry, a revision pointing to stickier inflation settles the internal argument in favour of more tightening.

What the optimists can still claim

There is a fair counterargument. Seasonal revisions do not change year-on-year inflation, which is calculated from unadjusted prices. Consumer prices rose 6.5 per cent in the year to December, down from a peak of 9.1 per cent in June, and core inflation was 5.7 per cent. That decline is real, and nothing published on 10 February undoes it. Energy prices have fallen, used-car prices are dropping and shipping costs have collapsed, all of which will keep pulling the annual rate lower.

But a central bank setting policy cares about the pace at which prices are rising now, and the annual rate is dominated by what happened months ago. On the revised data, the recent pace is too fast for comfort, and the fall in annual inflation has come mainly from energy and goods rather than from the services that respond to wages. The January consumer price report, due next week, will be the first reading on the new seasonal factors and will show whether that pace is slowing.

The new data show inflation is sticking. Markets cheered a statistical illusion, and underlying inflation never really fell in late 2022. The Fed is now likely to keep raising rates through the spring, and traders who priced in cuts will keep paying for it. Expect the Fed's March projections to show a higher peak rate than the 5.1 per cent median published in December, and expect short-dated bonds to keep losing ground until the monthly core figures fall back decisively.