The Adaptation Tax — the plain-English companion

the paper (DOI) · code & data repository

Companion to the research paper of the same title. This is education, not investment advice. Nothing here tells you what to buy, sell, or predict. It explains what the paper found, how the checking worked, and what the results do and do not mean.

The claim in one sentence

The yardsticks governments use to define a “normal” economy are averages of the recent past, so when the economy sinks, the yardsticks sink with it — and this paper measures what that costs, in dollars, inside five real institutions.

The scale that re-zeroes itself

Imagine a bathroom scale with a strange feature: it slowly re-zeroes itself to your average weight over the past year. Gain a pound a month, every month, and the scale keeps adjusting its idea of “zero” upward behind you. Step on it after a year of steady gain and it reads: about normal. The scale isn’t broken. It’s doing exactly what it was built to do — and that is the problem.

That re-zeroing scale is a trailing average: an average computed from the recent past, updated as new data arrives. An earlier paper in this research line proved a mathematical property of these averages and tested it on 44 combinations of financial instruments and averaging windows, across 11 instruments and up to 155 years of daily data. The finding: when a series and its own average drift apart, it is overwhelmingly the average that moves to close the gap, not the series coming back. The average chases the data. Always.

That sounds like a technicality until you notice where trailing averages live. They are built into the machinery that governs interest rates, bank regulation, Social Security raises, and pension accounting. In each of those systems, a trailing average is used to define what “normal” is — normal output, normal credit, normal prices, normal asset values. And an average that chases the data will chase a downturn too: as the economy sinks, the official definition of “normal” sinks alongside it. The downturn gets absorbed into the yardstick. This paper measures how much, system by system, using only public data anyone can download.

The filter that declared the recession over

Start with the paper’s opening story. In 2010, the statistical tool central banks use to estimate the economy’s “potential output” showed the US economy as roughly back to capacity. The tool is a smoothing recipe called the Hodrick-Prescott filter; it draws a gentle trend line through the ups and downs of GDP. GDP had climbed back above the filter’s trend line. By the filter’s reckoning, the recession was over.

Fifteen million Americans were still unemployed.

The filter was not malfunctioning. As GDP fell and stayed down, the filter did what a trailing average does: it pulled its trend line down toward the new data, quarter by quarter lowering what counted as “potential.” When GDP recovered to that lowered benchmark, the gap read zero. The economy was not back on its old track; the track had been moved.

The paper measures this on inflation too, where the stakes run through interest-rate policy. Apply the same filter to consumer prices the way a policymaker would have to, using only the data available at each moment. Its estimate of the inflation trend ends up biased in opposite directions depending on which way inflation is moving. The measured asymmetry between rising-inflation and falling-inflation periods is 0.60 percentage points — roughly the difference between reading 3.0% and reading 3.6%. A statistical test built for time-ordered data puts the chance of an asymmetry that large arising from luck at about three in a thousand. The filter also does 85.4% of the gap-closing work when inflation deviates from its trend: the trend comes to the data, not the reverse. Now feed that biased trend into the standard guideline for setting interest rates — the Taylor Rule, a formula that recommends a rate from inflation and output readings. The recommendation tilts: too loose while inflation surges, too tight while it cools.

The speed-limit sign that watches the traffic

After the 2008 crisis, bank regulators worldwide adopted a safeguard with a sensible design: when credit is booming, banks must set aside extra capital — a rainy-day cushion, called the countercyclical buffer — to absorb the losses when the boom turns. The trigger gauge is the credit gap: how far the economy’s total borrowing, measured against GDP, has climbed above its own long-run trend.

You can see the flaw coming. The trend is a trailing average of the borrowing data itself. It is a speed-limit sign that watches the traffic: when everyone speeds up together, the posted limit quietly rises, and by the gauge’s reading nobody is speeding at all.

The paper measures this on the boom that mattered most. During the 2003–2007 credit expansion — the run-up to the crisis this safeguard was designed for — the trend absorbed 81.4% of the boom into its own definition of normal. Only the remaining sliver registered on the gauge. Priced against the capital rule as written, the understatement corresponds to an estimated $154.73 billion in rainy-day capital that the gauge never called for on the paper’s primary bank-data series ($163.02 billion on the fallback series from a second regulator). The safeguard’s own arithmetic hid most of the storm it was built to catch.

The raise that arrives a year late

Every year, Social Security checks get a raise meant to keep up with inflation — the cost-of-living adjustment, or COLA. The statute computes it by comparing last year’s prices to the year before. That is a trailing measure with a built-in delay, and the consequence is easy to state: whenever inflation accelerates, this year’s raise was computed from last year’s slower prices. It is like paying this month’s grocery bill with a budget set from last year’s receipts.

In the paper’s sample, the COLA fell short of the inflation actually experienced in 96% of rising-inflation years. The paper is explicit that, for this system, the direction is nearly guaranteed by the formula itself — the measurement’s job here is to confirm the lag survives real-world details like rounding rules and benefit freezes, and to size it. Across the most recent inflation surge, the accumulated shortfall came to roughly $1,423 per retiree.

The paper then tests the obvious repair: run the same statutory formula four times a year instead of once. Measured against history, the quarterly version tracks actual inflation with 38% less error, at approximately zero cumulative cost — the raises arrive sooner when inflation accelerates and adjust down sooner when it cools, roughly cancelling over a full cycle. But before any results were computed, the design had committed a bar: to be declared better, a reform had to cut tracking error by at least 50%. It cut 38%. The design’s own rule therefore records the quarterly COLA as neutral-but-not-better — a measured candidate, not a demonstrated fix. The number that would have made a better headline was not allowed to become one.

The crash the books smooth away

Public pension funds — the retirement systems of teachers, firefighters, and city workers — report the value of their investments using actuarial smoothing: instead of the current market value, the books carry an average of values over several recent years. The stated purpose is stability, and in calm markets it is harmless.

In a crash it becomes a delay mechanism for bad news. It is like insisting, after housing prices fall, that your house is still worth its five-year average — comforting, checkable by nobody’s actual transaction, and exactly wrong at the moment decisions get made. In fiscal 2009, at the bottom of the financial crisis, the median public pension plan’s reported asset value stood 20.6% above what its investments were actually worth at market prices. And the direction is one-sided where it matters: in each of the 8 downturn years the paper examines, reported values exceeded market values — 8 of 8 — with the gap’s statistical test carried by the crisis year itself. Funding decisions, contribution rates, and political fights over pension health all ran on the smoothed numbers.

The same tilt in five economies

A skeptic should ask: maybe this is an American quirk — something about US data, US statutes, US agencies. So the paper re-runs its two core measurements — the GDP asymmetry and the inflation asymmetry — on five economies: the United States, the United Kingdom, the Euro Area, Japan, and South Korea. The settings are deliberately identical everywhere, with nothing tuned to any one country.

The tilt appears with the same sign in 5 of 5 economies for GDP and 5 of 5 for inflation. And the sizes cluster: the effect varies across countries by only about 14.0% of its typical size for GDP, and 46.8% for inflation. The paper reads that pattern the cautious way: as consistent with a property of the averaging arithmetic itself rather than of any one country’s economy. It also flags that one economy’s GDP result is individually too weak to stand alone on its shorter data sample.

What it adds up to

The paper’s last measurement asks the household question: what does all this mispricing cost per person? Because the systems interact and the errors depend on conditions, the answer requires simulation — a Monte Carlo exercise, meaning thousands of simulated economic paths run through the measured biases to see the range of outcomes. This exercise was run only because the design’s pre-committed decision board supported it, and its framing is careful in a way worth understanding.

During simulated moderate stress, the compound mispricing comes to an estimated $116.36 per US household per year, averaged across all households. That average conceals sharp concentration. For a household that takes out a mortgage at the wrong moment — when the biased trend has tilted rates the wrong way — the exposure runs roughly $3,642 per year, or about $523 per month on a typical mortgage. And the paper is explicit about what kind of number this is: an absolute mispricing total, which sums every dollar of error regardless of direction. A rate error that costs the borrower benefits the lender, and vice versa. So this is a measure of how much money is being moved and lost by mismeasurement, not a claim that every household is worse off by that amount.

The fix that fits in a spreadsheet

For each system inside its scope, the paper’s replacement candidate is the same object it used as the diagnostic: the divergence — a fast trailing average minus a slow one. When the economy is steady, fast and slow agree and the divergence sits near zero. When conditions turn, the fast average moves first, the slow one lags, and the gap between them opens — a signal that fires precisely when the trailing yardstick is starting to lie. It is how you notice a fever: not by asking whether today feels like yesterday, but by comparing your forehead now against your normal.

None of the replacements requires a model or a specialist. A two-window divergence, a plain rolling mean, the existing COLA formula run quarterly — each is computable from public data in a spreadsheet, and the paper demonstrates each on the exact series the institution actually uses. The claim is deliberately modest — not that trailing tools should be abolished; the literature the paper cites finds no single indicator dominates, and the demonstrations are framed as complements. The offer is a spreadsheet-simple companion gauge kept alongside the official one, disagreeing loudest exactly when the official gauge is most wrong.

Three bets on the next downturn

Everything above is measured on the past, and the paper says so plainly. History can flatter a thesis. So the paper ends with three dated predictions about the next downturn, registered publicly with explicit triggers, deadlines, and failure conditions — resolvable by anyone from named public data by July 2031.

First: at the next recession peak, if credit has boomed beforehand, the credit gauge’s understatement will again exceed a stated multiple of the official gap. Second: after the next sustained inflation surge, the first Social Security COLA will again fall short of the inflation retirees actually experienced. Third: at the next 20% stock-market decline, pension books will again show reported values above market values by the stated margin. Each prediction activates only if its triggering event occurs — and the registration states in advance that a prediction whose trigger never arrives counts as inapplicable, not passed. A test that never ran is not a test that was survived. If events go the other way, each failure is recorded in the paper’s public corrections log.

What the checking caught in this paper

The general machinery every paper in this series runs through — the hash-pinned inputs, the machine-checked ledger of numbers, the verification program, the adversarial review — is described once in the series’ shared verification note, which follows every companion on LaggingTruth.com. What belongs here is what the process caught and changed in this paper specifically.

The most consequential catch came from the first round of hostile review — a fresh, memory-isolated AI session working from the paper and its committed materials under a fix-or-rebut protocol. It found three load-bearing statistical defects: the paper’s headline significance tests treated time-ordered economic data as if each observation were independent, which overstates confidence. The remediation was recorded as a dated design amendment. The primary inference for every affected result was moved to methods built for time-dependent data (resampling history in blocks rather than shuffling individual points), and every decision rule that depended on an affected number was re-evaluated. All the affected verdicts survived on the corrected statistics — and the review record, including what the defects were, is committed in the repository. A second targeted review round then regenerated the entire claims ledger from scratch, rebuilt the rendered manuscript, re-ran one experiment end-to-end on a freshly downloaded copy of the data, and certified the remediation complete.

The pre-registered rules also bound the paper’s own hopes. The quarterly-COLA reform missed its committed 50% improvement bar and is recorded as neutral-but-not-better, and one economy’s individually insignificant GDP result is reported as such. A limitation statement that had overclaimed was rewritten during review, and the household-cost exercise was renamed and reframed mid-review to state exactly what its number measures — mispricing exposure, with the transfer-versus-loss distinction and its conditions carried into the abstract. The full analysis was additionally reproduced from a clean checkout in a fresh environment.

What this cannot do

• It is not an accusation of deceit. The bias is a mathematical property of trailing averages, not evidence that any institution is lying. The scale re-zeroes because of what it is, not because someone bent it.
• It does not say “abolish trailing tools.” The replacements are demonstrated as spreadsheet-simple companions to the official gauges, in line with published findings that no single indicator dominates.
• The dollar figures are modeled exposure, not audited losses. They come from a simulation under stated conditions, sum errors in both directions, and include transfers between parties as well as outright losses.
• The predictions may never activate. Each waits on a triggering event — a recession, an inflation surge, a market drawdown — and by rule, a trigger that never arrives makes the prediction inapplicable, not confirmed.
• The sample is the past. Every measurement is retrospective; the three registered predictions are the only forward-looking claims, and they are the paper’s exposure to being wrong in public.

The takeaway

Institutions need a definition of “normal” to act on, and they overwhelmingly build that definition from an average of the recent past. This paper measures the consequence in five systems that touch nearly every household. The average absorbs the very events it is supposed to detect. It declared a recession over while fifteen million were unemployed, and it hid a credit boom worth $154.73 billion in uncalled bank capital. It shorted retirees roughly $1,423 across an inflation surge, and it let pension books float 20.6% above reality at the bottom of a crisis. The same tilt, the same direction, in five economies. The repair on offer is not a grand model but a second gauge simple enough for a spreadsheet, one that speaks up exactly when the official yardstick has started to move. And because measurement claims should be testable by events, the paper leaves three dated predictions waiting for the next downturn to grade them.

This companion is licensed CC BY-NC 4.0. The research paper it accompanies is licensed CC BY-NC-ND 4.0, and the analysis and verification code is MIT-licensed. Education, not advice: nothing in this document is financial advice, an investment recommendation, or a forecast.