Why Base Effects Create False Inflation Signals
A year-over-year inflation rate is a comparison, not a measurement — and the half of it that was known twelve months in advance moves markets as if it were news.
Ask a sophisticated investor what the inflation rate measures and they will say: how fast prices are rising. That is not quite what the number is. The year-over-year rate is a comparison between two price levels twelve months apart, and a comparison can change for two entirely different reasons — because the recent end moved, or because the distant end did. The second reason produces every property of an inflation signal except the one that matters: it contains no information about what prices are doing now. The market's recurring error is not misreading current price behaviour; it is reacting to moves in a ratio whose denominator was fixed a year ago.
The framing we use for every inflation print is news versus arithmetic. Each month, the year-over-year rate updates in exactly two ways: a new month of price changes enters the twelve-month window, and an old month falls out. The entering month is news — it is the only part of the print that describes the present. The departing month is arithmetic — its value has been public for a year, its exit date has been known since it printed, and its effect on the annual rate could have been computed by anyone with a spreadsheet the day it was released. When the departing month was extreme, the annual rate moves mechanically — down if a spike is rolling out, up if a soft patch is — regardless of what the economy did in the most recent thirty days. That mechanical component is the base effect, and calling it a "signal" is a category error the market commits on schedule.
The mechanics: a ruler that moves
Picture the price index as a line and the annual inflation rate as the slope measured between two points a year apart. The market narrates changes in that slope as "inflation accelerating" or "inflation cooling," language that implies the recent end of the line changed direction. But hold the recent end perfectly steady — identical month-over-month price behaviour for months on end — and the annual rate will still fall if the year-ago period contained a spike, and still rise if it contained a lull. The ruler's back end is sliding along the historical data, and the reading changes because the ruler moved, not the line. Base effects are symmetric and they are patient: a large shock guarantees, twelve months later, an equal and opposite mechanical pull on the annual rate, on a date printed in the calendar. Disinflation delivered this way requires no policy, no demand cooling, and no change in anyone's behaviour — which is precisely why it proves nothing about any of them.
The standard correction is to shorten the ruler: three-month or six-month annualised rates weight the recent end and dilute the base. These momentum measures are the right tool for asking what the inflation impulse is now, but they buy relevance with noise — shorter windows amplify volatile categories and lean harder on seasonal adjustment, which is itself least reliable after unusual years. The mature approach is not to crown one measure but to hold the hierarchy: month-over-month is the raw news, short-window annualised rates are the smoothed news, and the year-over-year rate is the news plus a year of arithmetic. Each answers a different question, and most bad inflation takes begin by asking one question with another question's measure.
Why forecastable arithmetic still moves markets: three layers of why
The illustration: one episode, both directions
The post-pandemic inflation episode ran the base-effect machine in both directions inside three years, which makes it the cleanest teaching case on record — stated here as established history, without figures. In the first phase, the price collapse during the initial lockdowns created an abnormally low base; a year later, the annual inflation rate surged partly because current prices were recovering and partly because the comparison point was a crater. "Base effects" became, for a time, the centrepiece of the argument that the surge was transitory — an argument that was directionally right about the arithmetic component and wrong about the demand-and-wage component riding underneath it, which is exactly the decomposition failure this framework exists to prevent.
In the later phase the machine reversed. The energy spike that followed the invasion of Ukraine spent twelve months in the base and then rolled out, dragging headline annual inflation down rapidly during a period when much of the underlying services basket was still repricing briskly. Investors who read that mechanical descent as the inflation process resolving were early on duration for the right-looking reason and the wrong actual one. Both phases teach the same lesson from opposite directions: the annual rate's most dramatic moves are frequently the ones that carry the least information, because dramatic moves in the annual rate are what extreme bases manufacture.
The strongest case against this framework
The serious objection is that the base month is not fiction — it happened. A household whose costs jumped during the base period and then plateaued genuinely experienced the erosion the year-over-year rate records; the annual comparison is exactly the right measure of realised purchasing-power loss, of what wage negotiations should index against, and of what the political system experiences as "inflation." And since central banks state their targets in annual terms, the objection continues, the annual rate — arithmetic and all — is the policy-relevant number by construction. Calling its movements false signals mistakes the map the authorities actually navigate by for a distortion of some purer territory.
We accept nearly all of this and sharpen the framework rather than retreat. The year-over-year rate is the correct measure for cash-flow questions — realised real-income erosion, cost-of-living adjustment, the political temperature of inflation. It is the wrong measure for impulse questions — what the price-setting process is doing now, which is what the policy path beyond the next meeting, and therefore the discount rate on long-duration assets, actually depends on. Central banks themselves behave this way: annual targets are the accountability device, but the deliberation leans on momentum and breadth measures precisely because the annual rate arrives a year late to its own story. The framework's claim was never that the annual rate is meaningless; it is that the question decides the measure, and the market's standing error is answering the discount-rate question with the cash-flow measure.
What this changes about reading the data
The practical discipline is to make the arithmetic explicit before the print, so that only the news can surprise you. The base calendar is public information: the months rolling out of the window over the coming year, and their extremity, are known with certainty. From that alone you can sketch the mechanical path of the annual rate under a neutral assumption for incoming months — where it will fall without anyone earning it, and where it will rise without anything deteriorating. Held against consensus, that sketch tells you how much of the expected narrative is pre-scripted. Sterling's Embedded Intelligence maintains exactly this split on every inflation series it reads: the arithmetic path, computed in advance, and the news, which is the only component allowed to update Sterling's view of the inflation process. The habit sounds trivial; its effect is not. It converts the annual rate from a signal into a stage set, and redirects attention to the momentum and breadth measures where the process actually shows itself.
How to apply this framework
- Pre-compute the base calendar before every inflation cycle. The months rolling out of the annual window, and their extremity, are known with certainty. Sketch the mechanical path of the annual rate in advance so only the incoming month can genuinely surprise you.
- Split every print into news and arithmetic before forming a view. Attribute the change in the annual rate to the entering month versus the departing one. A move dominated by the departing month is the ruler sliding, not the economy turning.
- Answer impulse questions with momentum measures, cash-flow questions with the annual rate. Short-window annualised rates carry the current impulse; the annual rate carries realised erosion. Most inflation misreads are one question answered with the other's measure.
- Track when arithmetic starts feeding behaviour. A base-driven narrative that migrates into measured expectations or wage demands has become real. Fade repricings the calendar will reverse; respect the ones the expectations channel is converting into process.