The CPI Print That Moved Everything by a Tenth

The Bureau of Labor Statistics publishes the monthly change in the consumer price index to one decimal place, and almost everything that goes wrong in the reading of an inflation release begins with that convention, which almost nobody trading the first minute has thought about.

Consider two months. In the first, the underlying core monthly change is 0.254%. In the second it is 0.249%. The difference between them is half a basis point, which is to say nothing at all: measurement noise, a sampling quirk, one category behaving oddly in one metropolitan area during a single survey period.

Published, the first is 0.3% and the second is 0.2%. A tenth apart. One confirms disinflation and the other calls it into question in every headline written before nine o’clock, and the yield curve will move on a distinction that is not there in the underlying data.

What each version implies if it repeats

The arithmetic makes the gap between appearance and substance very clear indeed.

Take the rounded figures and compound them over twelve months. 1.003^12 - 1 = 3.66% and 1.002^12 - 1 = 2.43%. Against a target of 2%, those are two entirely different worlds, and a market that has to guess at the path of the policy rate is quite right to care which one it is looking at.

Now compound the underlying numbers instead. 1.00254^12 - 1 = 3.09% and 1.00249^12 - 1 = 3.03%. Six hundredths of a percentage point apart. The same two months, honestly measured, are close to identical, and the reported versions differ by more than a full percentage point of implied annual pace, on the strength of a rounding rule.

None of this is a complaint about the statistics, which are produced carefully to a published methodology by people who are entirely candid about their own margins of error. It is an argument for reading the tables. They are published at the same instant as the headline. They are free. They take perhaps ten minutes.

The reaction is still rational, which is the awkward part

Having made that argument, I should complete it honestly. The size of the moves makes sense once you take the compounding seriously.

Monthly inflation numbers are small. A tenth is a substantial fraction of a typical month. The market is pricing a path, it updates that path on each new observation, and a single observation that shifts the estimate of the trend is worth a great deal more than the observation itself. The policy rate then feeds into every discount rate in every other market, which is why an inflation release moves equities as hard as it moves bonds, and that chain, from the discount rate through to sector behaviour, is worked through in how interest rates affect stocks.

So the rounding does not make the reaction foolish. It makes the reaction fragile. That is a different criticism, and a more useful one. A view formed on a rounded tenth is a view that can be reversed by the next month’s rounding, and it very often is.

The order I read the tables in

Shelter comes first, because it is the largest single piece of the index and it is measured from rents that turn over slowly, so it describes the rental market of several months ago. A print that runs hot because of shelter is reporting on last year’s leases, which is a materially different piece of news from a print that runs hot in something with no lag in it.

Then the categories with a history of swinging on their own. Used vehicles, airfares, certain insurance lines. These can move a monthly figure single handedly and they frequently reverse the following month, so a surprise concentrated in one volatile line is weaker evidence than the same surprise spread thinly across services.

Then core against headline, and both against the three month annualized pace, which smooths some of the monthly noise without discarding the recent trend the way the twelve month rate does.

On the release I have in mind, the surprise sat almost entirely in one services category with a long record of volatility, and it partly reversed the following month. By then the initial move had faded. It had been written about as though it had been obvious. The people who had been most confident at 8:31 in the morning were working with the least information available to anybody that day.

The construction of the basket, the difference between headline and core, and how the release compares with the personal consumption expenditures measure the Federal Reserve actually targets are all in inflation and CPI explained. For the wider schedule, including which releases reliably move markets and which are better read afterwards than at the instant they land, the economic calendar sorts them by impact.