Why Core Inflation Became the Fed's Focus
Core inflation is not a claim that food and energy don't matter — it is a forecast of where headline will settle, wearing a measurement's clothes, and the reaction function is written on it.
The standing joke about core inflation — that it measures the cost of living for people who neither eat nor drive — misidentifies what the number is for. The joke assumes core is trying to measure the cost of living and doing it badly. It is not trying to measure the cost of living at all. Core inflation is a forecasting device: an attempt to estimate, from the noisy monthly data, where headline inflation itself will settle once its most mean-reverting components have finished mean-reverting. The sophisticated error is subtler than the joke — it is knowing that central banks watch core, without knowing why, and therefore failing to notice the specific conditions under which the rationale breaks down and the reaction function quietly changes what it is anchored to.
The framing to hold: core is a forecast wearing a measurement's clothes. It is published to index precision, decimal places and all, which makes it look like a description of the present. Its function is entirely about the future — it exists because a central bank cannot act on the present. Once you read core as an estimator rather than a gauge, its exclusions stop looking like value judgements about which prices matter and start looking like what they are: a crude but serviceable filter for separating the persistent component of inflation, which policy must respond to, from the transitory component, which policy cannot reach and should not chase.
The logic: three layers of why
One refinement matters for anyone applying this. "Core" as commonly quoted is the crudest available persistence filter — it removes two named categories and keeps everything else, including components with their own severe measurement lags. Statistical agencies and central bank research shops maintain sharper filters: trimmed-mean measures that discard the largest moves in either direction each month regardless of category, median measures, and services aggregates excluding shelter. These generally do the persistence-extraction job better. Core survives as the public anchor for institutional reasons — it is simple, explainable, and stable in definition, and an accountability regime needs a number that does not require a methodology appendix. The investor's takeaway is to read core as the communicated persistence estimate and the sharper filters as the working one, and to pay attention when they disagree — the committee usually sides with the working estimate before the public one catches up.
The illustration: born as an alibi, matured into an estimator
The origin of core inflation is itself the best illustration of both its value and its danger, and it is established history. The concept emerged in the United States in the 1970s, when successive food and energy shocks were tearing through the price level. Its first institutional use had an unmistakable flavour of excuse: policymakers of that era leaned on the argument that the inflation was coming from supply shocks outside monetary control, and the ex-food-and-energy measure conveniently quarantined the worst of the numbers. Used that way — as a reason to look away from headline rather than a tool for forecasting it — core helped rationalise a policy stance that stayed too easy while the shocks it excluded were leaking steadily into wages, expectations, and the sticky components core was supposed to represent. The persistent inflation of that decade was eventually visible in core too, but by then the process had years of momentum.
The rehabilitation came with the harder lesson: the exclusion is only valid while the excluded components are genuinely mean-reverting and genuinely failing to pass through. Modern practice treats core accordingly — not as a quarantine but as an estimator whose validity is continuously checked against second-round evidence. The same test separated the sheep from the goats in the post-pandemic episode: the initial goods-and-energy surge was arguably outside policy's reach, but the moment the shock began showing up in services pricing and wage growth, the alibi reading of core expired and the estimator reading demanded action. The lesson generalises: core used as a reason not to act has a bad historical record; core used as a persistence forecast, with its validity conditions monitored, has a good one. An investor should always ask which of the two uses is currently in fashion at the central bank — because the transition from one to the other is a repricing event.
The strongest case against core
The serious critique comes in two layers. The first is the salience argument: households form their inflation expectations disproportionately from the prices they see most often — fuel and groceries above all — and expectations feed the wage round. If headline drives expectations and expectations drive future core, then core is the output of a process whose input it deliberately excludes, and a central bank watching core is watching the downstream gauge while the upstream one floods. The second layer is structural: the exclusion logic assumes food and energy shocks are temporary deviations around a stable trend. In a world of persistent energy repricing — supply constraints, geopolitical fragmentation, the capital cycle of an energy transition — energy inflation can carry trend information, and excluding it systematically biases the persistence estimate in whichever direction the structural trend runs.
Both points are correct as mechanisms, and neither overturns the framework — they define its boundary conditions. The salience channel is real but conditional: it operates when headline shocks are large and sustained enough to enter bargaining behaviour, which is observable in expectations surveys and wage data rather than a matter of faith. The structural channel is real but identifiable: a mean-reverting shock and a persistent repricing look different over time, and the estimator can be — and in practice is — adjusted when the excluded components stop behaving as assumed. The honest summary is that core rests on an empirical premise, not a truth: that food and energy shocks die out before they breed. The premise holds in most years, fails in identifiable regimes, and the failure is diagnosable in real time through pass-through and expectations data. The investor's job is not to pick a side in the core-versus-headline debate; it is to monitor the premise, because the price of every long-duration asset embeds an assumption about whether the central bank's persistence estimator is currently pointed at the right target.
What this changes for a portfolio
The direct consequence is about which number is tradable. Because the reaction function is written on core, the core print is the one that moves the expected policy path, and the expected policy path is the discount-rate input for every long-duration asset in the book. Headline matters to the portfolio through an entirely different door — real household income, consumer volumes, the revenue mix of consumer-facing businesses. When the two diverge, the divergence itself is the signal: firm headline over soft core is a consumer squeeze without a policy consequence, which tends to be bad for consumer cyclicals and roughly neutral for duration; soft headline over firm core is consumer relief without policy relief, the combination that most reliably wrong-foots investors who trade the headline number expecting the policy consequence. Sterling's Embedded Intelligence enforces this routing on every release: core and its sharper cousins into the policy-path model, headline into the cash-flow model, and the divergence between them into the sector-rotation view. Sterling also tracks the estimator's validity conditions — pass-through from excluded components, expectations behaviour, the gap between crude core and trimmed measures — because the largest repricings in this territory come not from the prints themselves but from the moments the central bank changes which number it believes.
How to apply this framework
- Price policy off core; price the consumer off headline. The reaction function is written on the persistence estimator, so core surprises are discount-rate events. Headline-only surprises are real-income events — different exposures, different trades, same release.
- Watch the gap between crude core and the sharper filters. When trimmed-mean, median, or ex-shelter services measures diverge from the quoted core rate, the committee's working estimate differs from the public one — and policy tends to follow the working estimate first.
- Monitor the estimator's premise, not just its printout. Core's validity rests on excluded shocks dying out before they breed. Pass-through into services pricing, wage growth, and expectations surveys is the evidence that the premise is failing — and that the reaction function is about to migrate.
- Treat alibi-usage of core as a regime warning. When policymakers cite core to justify inaction while headline shocks are visibly entering wages and expectations, history's precedent is unkind. The 1970s version of that argument preceded the era's worst inflation — and its eventual, expensive correction.