This guide walks through what a trading day is, why the same session looks different in each city, and how holidays change the plan. It draws on the public record of trading day rules, so every detail here has a checkable base.

What Counts as a Trading Day

The trading day, sometimes called regular trading hours, is the time span that a stock exchange is open. It is separate from electronic or extended trading hours. When a trading day ends, all trading ends. The market is frozen in time until the next trading day begins. We covered a connected angle in What does after-hours trading actually cost?.

The US Baseline: 9:30 to 4:00

The New York Stock Exchange is, as of the year 2026, open from 9:30 AM Eastern Time to 4:00 PM Eastern Time. NASDAQ uses the same 9:30 to 16:00 ET window. Trading days are usually Monday through Friday. The NYSE and NASDAQ average about 251 trading days a year. Readers following this should also see When do stock circuit breakers actually halt trading?.

Where does that number come from? Start with 365.2425 days in an average year. Take 5/7 of it for the share of work days in a week. Then subtract about 10 holidays. The result is roughly 251 trading days.

Some days are shorter. Up to three trading days, July 3, the day after Thanksgiving, and Christmas Eve, are shortened. The exchanges are open from 9:30 AM to 1:00 PM on those days. If July 3 or Christmas Eve fall on a weekend, the shortened day is simply skipped.

Holidays: Closed, Open, or Moved

The exchanges close for New Year's Day, Martin Luther King Jr. Day, Presidents' Day, Good Friday, Memorial Day, Juneteenth, Independence Day, Labor Day, Thanksgiving Day, and Christmas Day. There are also some holidays where trading is permitted, including Columbus Day, Veterans Day, and New Year's Eve.

Weekends shift things a bit. If Juneteenth, Independence Day, Christmas Day, or New Year's Day fall on a weekend, the public holiday is observed on the Friday before or the Monday after. So each holiday still costs one trading day.

Special events can change the schedule too. A state funeral of a head of state can lead to a shortened trading day, or no trading day at all. Juneteenth was declared a national holiday on June 17, 2021. It was first observed as a market holiday on June 20, 2022.

The Same Session, Different Clocks

Each stock exchange has opening hours that are based on specific time zones. People can trade in these exchanges remotely using electronic trading platforms. For those trading in different parts of the world, there are unique trading days based on the hours tied to any given time zone.

Take NASDAQ, which is open 9:30 to 16:00 ET. Anyone outside of the Eastern Time Zone gets a different trading day. In Vancouver, a trading day would run from 6:30 to 13:00. During the part of the year when North America is on standard time, it would be 17:30 to 24:00 in Moscow, and in Shanghai it would be 22:30 to 5:00. Remote traders should find their trading hours based on the stock exchange's hours and time zone.

Conclusion: Anchor to One Clock

Market hours look complex only until you fix one anchor. Pick the exchange and its home time zone, then convert. The US session is 9:30 to 4:00 Eastern, Monday to Friday, with a few holiday breaks. Everything else is that same window seen from another city.

This article is general information, not investment advice. Markets involve risk, including possible loss of principal.

The answer matters for anyone reading crypto news this cycle. A syndicated release is a vendor statement, not reporting. The Markets Insider version carries its own label: the editorial teams were not involved. Everything below should be read with that condition attached.

What does the release actually claim?

Strip the framing and the checkable statements are few. The presale has passed $11.16 million, at a stated token price of $0.0000001898. Staking is described as paying 161% APY. The project describes a product called PepetoSwap that trades Ethereum, BNB Chain and Solana side by side with no fees on orders, plus an AI layer that scans transactions for contract risk. Contracts are described as audited by SolidProof. The release says the person behind the build is a cofounder of the Pepe ecosystem, a token it says passed $7 billion. It also says a Binance listing is being prepared, and attributes that to "word," not to any exchange statement.

Every one of those is a claim by the token's own promoters. None is independently verified in the material at hand. A raise figure is the easiest number to print and the hardest to check from outside, because the source of the count is the party raising the money.

Why does the Hayes forecast sit next to it?

The release pairs its presale news with Arthur Hayes's $1 million Bitcoin prediction, attributed to CoinGape, with the strongest rally expected in late 2027 or early 2028. From a stated Bitcoin price near $83,700, the release itself computes a 12x. It also asserts that Bitcoin is holding that level on 2026's strongest ETF inflows.

Read as analysis, the pairing is the point. A dated price target from a known name gives a promotional timeline a sense of urgency: if the big coin moves, the release argues, presale entries move faster, and the window closes. That is a sales structure, not a measurement. A price target is a single point on a distribution with no stated error bars, no evaluation window, and no track record attached in the text. Treating it as a schedule is the reader's first mistake to avoid.

What does the SHIB comparison do in the argument?

The release leans on a story from CoinTelegraph about a warehouse manager who turned an $8,000 SHIB purchase in 2021 into $1.5 million. It then states that Pepeto stands where SHIB stood before its breakout, and that the best presale entries in past cycles have returned 100x.

This is the part a careful reader should slow down on. The 100x figure has no dataset, no sample definition, and no evaluation period in the text. One survivor story is not a base rate; thousands of tokens that never broke out are not in the anecdote. The release even names the mechanism itself: it says thousands saw the same chance and froze. That framing converts hesitation into a cost, which is exactly what a time-limited offer is built to do. For related coverage, see What does after-hours trading actually cost?.

How should a reader test claims like these?

The discipline is the same one that applies to any vendor claim in financial news. Separate what is stated from what is verified, and check the stated conditions.

• A raise total, a token price, and a staking APY are the issuer's own numbers. A 161% APY is a promised rate, not a realized return, and promised rates in crypto staking are typically paid in the token itself.
• An audit by a named firm, if real, checks code, not economics. It says nothing about demand or whether the token appreciates.
• A listing that exists only as "word" is not a listing. Exchanges announce their own listings.
• Wallet counts and Telegram growth are activity metrics, and activity can be manufactured. Volume, in particular, is a measure with well-known limits.

Readers who want the general method in more depth can start with our earlier piece on reading a presale press release closely, and on what trading volume measures and what it does not. This connects to our earlier piece, A presale press release, read closely: what Apeing's Stage 6 numbers do and don't establish.

What the evidence does and does not establish

What it establishes: a syndicated promotional announcement exists, dated Oct. 2, 2026, stating an $11.16 million presale, a $0.0000001898 price, a 161% staking APY, and a Hayes $1 million-by-2030 call attributed to CoinGape. What it does not establish: that the raise figure is accurate as counted, that the Binance listing is real, that the staking rate is sustainable, or that any past-cycle pattern will repeat.

Crypto markets are volatile and losses are possible, including total loss on presale tokens that never list. The release's own disclaimer says as much. The honest summary of this news is narrower than its headline: a project says it raised money, a trader restated his forecast with a date, and both claims arrived in the same paid envelope. That is a fact about the information supply chain, and it is the fact most worth carrying forward.

This article is general information, not financial advice. Consider your own circumstances or consult a licensed financial professional.

The first thing a careful reader should do is file this where it belongs: it is a promotional announcement about a crypto presale, and every performance number in it is a vendor claim with a condition attached. The same release supplies the useful contrast case. Ethereum, the chain the token runs on, traded at $2,672.08 with a market capitalization near $326.24 billion and $14.69 billion in 24-hour volume — roughly 4.47% of that capitalization changing hands in a day. That is what deep, observable market activity looks like, and it is worth knowing what trading volume measures and what it doesn't before comparing it to a presale ledger.

What the staged pricing is actually doing

The presale runs across 33 stages. Stage 6, named "Fearless FOMO," prices $APEING at $0.0006; the next stage moves to $0.00065, and the stated listing price is $0.01. The release frames this as giving participants "a predefined early-stage price before the next scheduled increase." Mechanically, that is a count-up pricing schedule, and its function is to convert hesitation into urgency. Every stage boundary is a deadline the marketing copy reminds you is approaching.

The schedule also does something subtler: it manufactures a visible gain before any market exists. A buyer at Stage 6 can compute a paper profit against the stated $0.01 listing price without a single token trading. The release does exactly this, highlighting "a potential 1,500% ROI based on its projected listing price." That figure is arithmetic, not a forecast. It holds only if a listing at $0.01 happens and holds, neither of which the release can promise. A stated listing price is an intention, not a market.

What the two headline numbers imply about who is buying

Two figures in the release invite simple division. More than 481 million tokens sold against more than $111,000 raised works out to an average price near $0.00023 per token — well below the current stage price of $0.0006. That gap is consistent with the staged structure itself: earlier stages were cheaper, so the average buyer paid less than today's buyer. It also means the current stage's cohort is paying a premium, in relative terms, over the book as a whole.

The second division is $111,000 across 442-plus holders, or roughly $250 on average. The release's own example table starts at a $250 purchase, so the typical participant in this tally sits at the bottom of the promoted entry ladder. Small average tickets and a holder count in the hundreds describe a very early, very small distribution base — worth stating plainly, because "momentum" copy tends to blur it. For context on how thin early trading conditions can be, see how market liquidity is actually measured. We covered a connected angle in What does after-hours trading actually cost?.

The staking tiers deserve their own read

The project advertises staking with four APY tiers: 10%, 25%, 50%, and 85%, with staking allocated 30% of the total 16.75 billion token supply. An annual percentage yield paid in the token's own units is a promise about token quantity, not about value. An 85% APY in a token whose price is set by a thin, early market can coincide with a falling dollar value. The release presents the tiers as "utility"; the more precise framing is that they are an incentive to hold rather than sell, which serves the presale's price narrative. Where the yield sits and who ultimately eats the cost of it — dilution of the remaining supply — is the question this publication would ask of any token program.

The Ethereum half of the release is the honest half

The ETH data in the same release is ordinary market reporting, and it is the only part with observable, ongoing price discovery. ETH gained 0.22% over 24 hours, trading between $2,655.56 and $2,746.01, with about 122.09 million coins in circulation and no maximum supply cap. It remains 46.02% below its all-time high of $4,953.73, recorded August 25, 2025. A single day's range says little about direction; the durable fact is the scale difference. A $326 billion market with billions in daily turnover is a different category of asset from a $111,000 presale, and the release's own "two narratives" framing quietly concedes it.

What this release cannot establish

• Whether $APEING ever lists at $0.01, or lists anywhere at all; the figure is stated, not demonstrated.
• Whether the 1,500% ROI claim survives contact with a market; it is conditional on the stated listing price and carries no evaluation window or baseline.
• Whether the staking APY tiers are sustainable, funded, or enforceable; the release describes structure, not economics.
• Whether the reported token and dollar totals are accurate; they are the project's own counts in its own press release.

Crypto markets are volatile and losses are possible in any token, and early-stage presales concentrate that risk: no trading history, no observable liquidity, and pricing set by the issuer. The verifiable content of this announcement is structural — stage counts, allocation percentages, wallet mechanics such as purchases being permanently linked to the buying wallet, and a $25 minimum to activate a referral code. The performance content is a projection wearing the grammar of a result. Readers comparing presale pitches against measured markets may find it useful to start with markets coverage that keeps the two categories separate. Readers following this should also see How is market liquidity actually measured?.

This article is general information, not financial advice. Consider your own circumstances or consult a licensed financial professional.

That number is far below the peak. In June 2022, the annual rate hit 9.1%, the highest in this century's records. It has cooled since, but it has not returned to the low, quiet levels of the 2010s, when annual inflation usually ran between roughly 1% and 3%.

This explainer covers what inflation actually is, how it gets measured, and what pushed it up in the first place. For readers who want the measurement details, our separate piece on how the CPI is actually measured goes deeper. Here, the focus is on the idea itself.

What is inflation, exactly?

Inflation is an increase in the average price of goods and services in terms of money. The key word is average. If cucumbers get pricier because shoppers suddenly want more of them, while tomatoes get cheaper, that is a shift in tastes — not inflation. Inflation is about the general trend of prices, and it reflects the value of the currency itself.

As Wikipedia's overview of inflation puts it, when the general price level rises, each unit of currency buys fewer goods and services. The opposite is deflation, a fall in the general price level. Economists also track related terms: disinflation means inflation is still rising but more slowly; stagflation combines inflation with slow growth and high unemployment.

The word has a curious history. It originally referred to an increase in the money supply, not to rising prices. During the American Civil War, when banks printed more banknotes than the metal backing them, "inflation" came to mean currency losing value. The modern price-level meaning took over later.

How is inflation measured?

The standard yardstick is a price index, most often the Consumer Price Index, or CPI. The Bureau of Labor Statistics publishes it monthly. The inflation rate you see in headlines is the percentage change in that index over 12 months.

The arithmetic is simple. Take the CPI for one month, subtract the CPI for the same month a year earlier, divide by the earlier figure, and multiply by 100. As an example, the difference between January 2016 and January 2017 CPI values worked out to an annual rate of 2.5%.

The US Inflation Calculator site's running table of annual inflation rates, built from BLS data, shows how the numbers moved through the recent episode: 7.0% for calendar 2021, 8.0% on average for 2022, 4.1% for 2023, 2.9% for 2024, and 2.7% for 2025. The 2026 monthly readings have ranged from 2.4% in January and February to 4.2% in May, settling at 3.4% for the 12 months ending August.

One wrinkle worth knowing: the monthly change and the annual change are different animals. The monthly rate is seasonally adjusted; the annual rate usually is not. And headline figures get revised and re-benchmarked, which our coverage of why economic data keeps getting revised explains in detail. For related coverage, see Why economic data keeps getting revised after release.

Why did prices surge after 2021?

Economists group the causes of inflation into a few broad buckets. Changes are widely attributed to growth in the money supply, shifts in real demand for goods and services, drops in available supply such as during energy crises, big cuts in central bank interest rates, or changes in what people expect inflation to be. Those expectations can be self-fulfilling: if firms and workers believe prices will rise, they raise prices and wages to match.

The 2021–2023 surge had features of more than one bucket. Demand shifted hard toward goods while supply chains were strained. Energy prices spiked. Money supply and fiscal support expanded sharply. None of these forces alone explains the whole move, and economists still debate the weights.

What the record shows is the shape of it. Annual inflation climbed steadily through 2021, peaked at 9.1% in June 2022, and then declined in steps through 2023 and 2024. The recent 2026 readings — 3.8% in April, 4.2% in May, then 3.5%, 3.4%, 3.4% — suggest a bumpy plateau rather than a straight line back to 2%.

Who feels inflation, and how?

Inflation is not spread evenly. It lands hardest on things people cannot skip. The Bureau of Labor Statistics' own CPI features make this concrete: in an Economics Daily note on school lunch prices, the agency reports that over the past 20 years, white bread rose 87.5%, lunchmeats 58.3%, and lettuce 52.0%. Groceries sit in every household budget, so those increases compound quietly.

Wages are the other side of the ledger. Moderate inflation has real costs: workers are pushed into risky wage negotiations as their purchasing power erodes between contracts, uncertainty about future inflation can discourage saving and investment, and if inflation runs fast enough, consumers may hoard goods. There are upsides too — a little inflation can reduce unemployment when wages are slow to adjust, and it gives central banks room to work. Most economists today favor a low and steady rate, not zero.

Keeping the rate low and stable is usually the job of central banks, which act through interest rates and open market operations. That is why inflation prints move rate expectations — our piece on how Fed rate probabilities are priced and read covers that connection. We covered a connected angle in How are Fed rate probabilities priced and read?.

What this means for readers

Practical takeaway one: always check the window. "Inflation is 3.4%" means the 12 months ending August 2026, not a forecast of next year. Rates are backward-looking by construction.

Practical takeaway two: your personal inflation rate differs from the headline. The index is an average basket for urban consumers. If you spend heavily on categories rising faster than average, your effective rate is higher, and vice versa.

Practical takeaway three: distinguish the level from the rate. A 3.4% inflation rate means prices are still climbing — just more slowly than in 2022. Disinflation is not the same as falling prices. Confusing the two is one of the most common reading errors in economic news.

The limits of what the number can tell you

The headline CPI rate cannot tell you why prices rose, whose basket got more expensive, or whether the trend will hold. It is one averaged index, not seasonally adjusted in its headline annual form, and it says nothing about asset prices, which can inflate separately from goods and services. Economists also track narrower measures — energy inflation, house price inflation, food price spikes — because the average can mask them.

What the evidence here establishes is the definition, the measurement method, and the recent record. What remains genuinely uncertain is the forward path: the sources describe the forces that drive inflation, but the mix of those forces at any moment is a matter of ongoing debate among economists, not settled fact.

The ratio is built from data that arrives late. US exchanges publish short interest figures twice each month, and the published number reflects positions as they stood roughly two weeks earlier, so the metric is always a snapshot of the recent past rather than the present. That lag shapes everything the ratio can and cannot tell you.

How is short interest computed in the first place?

Short interest is the total number of shares that have been sold short but not yet bought back. A short seller borrows shares from a broker, sells them on the open market, and hopes to repurchase them later at a lower price. Until the borrowed shares are returned, they count toward the short interest figure. Think of it as a headcount of pessimists, captured on a specific reporting date.

The raw share count is only the starting point. Brokerage firms report short interest data to FINRA twice per month, around mid-month and at month's end, and NYSE and Nasdaq publish official compilations from those reports. Because a large-cap company with billions of shares outstanding can carry 10 million shares short and still be lightly shorted in relative terms, analysts usually normalize the figure. The most common version divides shares sold short by the float — the shares actually available for public trading, excluding insider holdings and restricted stock. A stock with a 40 million share float and 4 million shares short has short interest of 10% of float.

Both the numerator and the denominator deserve scrutiny. The float itself is an estimate that changes with insider sales, new issuances, and index events — the mechanics of which we covered in What actually happens to a stock when its index adds it? A stale float can quietly inflate or deflate the percentage without any change in actual short positioning. Readers following this should also see What actually happens to a stock when its index adds it?.

What exactly does the short interest ratio measure?

The short interest ratio — also called days-to-cover, or days to cover — divides shares sold short by average daily trading volume, typically computed over a window of about 30 trading days. The result estimates how many days of normal volume it would take for all short sellers to buy back their borrowed shares, assuming they were the only buyers in the market. This connects to our earlier piece, What does trading volume measure, and what doesn't it tell you?.

That last assumption is the metric's central fiction. In reality, short sellers are never the only buyers, volumes change quickly, and covering rarely happens in an orderly single-file queue. As one trading guide puts it, days-to-cover is only a rough guide, because volumes and liquidity can shift faster than the data updates. The ratio is best read as a liquidity-adjusted measure of how crowded the short side is — a cousin of the questions we asked in How is market liquidity actually measured?

What does it mean in practice? A high ratio can arise two ways: heavy shorting, or thin volume. Both raise the same risk for short sellers — that exiting becomes slow and expensive if the price rises. A low ratio means positions could be closed quickly, which signals less exit risk but says little about direction.

Why do backtests and headlines overstate what the ratio signals?

The ratio's most famous association is the short squeeze: a heavily shorted stock rises, short sellers rush to buy shares to limit losses, and their buying pushes the price higher still. The GameStop episode of 2021 is the standard illustration — the share price jumped from $17.25 to $483, and short sellers reportedly lost over $6 billion as they were forced to buy back at much higher prices.

But the squeeze is the tail, not the distribution. High short interest does not mean a stock will rise, and it certainly does not mean it will fall. Prices are set by many factors, and short sellers are frequently right: elevated shorting often reflects genuine fundamental concerns — deteriorating earnings, regulatory trouble — that play out over months. Past short interest levels do not predict future price movements; a heavily shorted stock can keep falling, rally sharply, or drift sideways for extended periods. Reading the ratio as a contrarian buy signal is the same category error as reading any single indicator as destiny.

There is also a structural point worth knowing. Short interest is not purely a bet on direction. Some of it is hedging — pairs traders shorting one stock against a long in another — and some is arbitrage around mergers and conversions. The ratio cannot distinguish conviction shorts from mechanical ones, which is one reason a high number is ambiguous evidence.

What does the evidence say about how short interest behaves?

Research on what drives short interest is active, and one finding is directly relevant to how you should read the number. According to a Federal Reserve discussion paper by von Beschwitz, Honkanen, and Schmidt, growth in passive fund ownership causes an increase in both lendable supply and short interest, while lending fees stay largely unchanged. The authors find that shorting demand rises more than lendable supply in these cases, and that the additional shorting improves price efficiency around negative earnings news.

What this means: some of the short interest you observe is a supply-side artifact of index ownership, not a surge in bearish conviction. A stock added to a passive-heavy index can show rising short interest with no change in sentiment. Comparing a ratio across stocks — or across time for one stock — without accounting for the lending market's structure invites the same window-dependence trap as any rolling statistic.

Limitations: what the ratio cannot establish

This is a methods explainer, so the limits deserve their own section. The short interest ratio cannot do any of the following:

• Time anything. The data is published twice monthly with a settlement-date lag of several days. A ratio computed today may describe positioning from two weeks ago, already changed.
• Separate conviction from mechanics. Hedging shorts, arbitrage shorts, and directional shorts are indistinguishable in the aggregate figure.
• Adjust for volume regime changes. The denominator is trailing average volume. A volume spike or collapse after the reporting date makes the ratio stale — the same caution that applies to any volume-based measure, as discussed in What does trading volume measure, and what doesn't it tell you?
• Predict squeezes. A squeeze requires a catalyst. The ratio measures the fuel, not the fuse, and most heavily shorted stocks never squeeze.
• Support thresholds as rules. Common interpretive bands — below 10% of float described as low, above 20% as high — are conventions from trading commentary, not validated decision boundaries. Treat them as descriptive vocabulary, not signals.

One further caution: vendor platforms present short interest data with their own refresh schedules and float estimates, so the same stock can show different ratios on different services. When a number matters to a decision, check the exchange or FINRA compilation directly and note the settlement date it reflects.

Practical takeaways

Used properly, the short interest ratio answers one narrow question: how hard would it be for the short side to exit? Read it alongside the percentage of float, the trend in short interest over several reporting periods, and the stock's current volume — and always with the reporting lag in mind. It is one input among many, a snapshot of positioning rather than a view of the future. Compared to what, and as of when, are the two questions that keep the metric honest.

Uzu News publishes information and analysis about forecasting methods, not investment advice, and this analysis makes no claim about any live strategy's future performance. Its subject is what a reported backtest can and cannot support as evidence.

Why do backtests overstate live performance?

Two mechanisms do most of the damage. The first is multiple testing: when a researcher tries many configurations on one history and publishes the winner, the winner's margin is partly luck, and the luck does not repeat. Halbert White's Reality Check, published in Econometrica in 2000, formalized the correction for this in econometrics.

The second is selection under scarcity. Financial history offers one realized path per asset. A configuration that fit the particular sequence of the 2010s is fitting noise that will not be regenerated, and the evaluator usually cannot tell the fit from the mechanism.

The 2014 Journal of Computational Finance paper by Bailey, Borwein, López de Prado and Zhu measured the combined effect: the probability of backtest overfitting, meaning the chance that the best in-sample configuration lands in the bottom half out-of-sample, rises toward and past 50 percent as the number of configurations tried on a fixed history grows. Overfit backtests are the expected output of an ordinary research process, not a rare failure of it.

What would a valid backtest report?

The answer is a set of conditions, not a single metric. Each one is checkable in principle, which is what makes the standard usable.

The last row is the one readers most often have to supply themselves. A vendor's stated Sharpe ratio is a vendor claim, carried with its stated conditions; it becomes evidence only when an independent party recomputes it from data the vendor did not select.

How do regulators read the same evidence?

Banking supervisors reached a similar position a decade before the machine-learning literature made it explicit. The Federal Reserve's Supervisory Guidance on Model Risk Management, SR 11-7 of 2011, treats a model's developer as an interested party and requires independent validation: outcome analysis against data the model did not influence, benchmarking against alternatives, and documentation of known weaknesses and their compensating controls.

The parallel is exact enough to be worth stating. The literature's contribution was to quantify how much selection inflates a reported ratio; the guidance's contribution was to make independent checking a structural requirement rather than a courtesy. A backtest that would fail the guidance's independence test should not impress a reader simply because it appears in a paper instead of a bank memo.

What does this leave the reader to do?

Read the conditions before the number. A Sharpe ratio stated with its trial count, sample length, variance across trials, universe, and window is a report; the same number stated alone is a claim. When the conditions are missing, the deflation arithmetic cannot be run, and the honest default is to treat the figure as unverified.

This publication's own judgment, stated once: the persistent problem is not dishonesty but asymmetry of effort. Running a hundred backtests is cheap and quick, and verifying one is slow and expensive, so markets accumulate unverified winners faster than anyone retires them. That asymmetry, more than any single flawed paper, is why claimed backtest performance should be read as a hypothesis awaiting independent testing.

What remains unknown after this analysis is the base rate: no public dataset records how many published backtests have been independently replicated, so the share that survives verification cannot be stated from evidence. The corrections define the inflation. The audit counts do not yet exist at scale.

Uzu News publishes information and analysis about forecasting methods, not investment advice. This explainer defines the vocabulary, shows how the split is supposed to be done, and states what an out-of-sample result cannot establish even when it is done correctly.

What does out-of-sample actually test?

Backtesting is the practice of replaying a strategy or model on historical data and recording its performance. An out-of-sample test scores the model on data that played no part in choosing its features, parameters, or stopping point. The claim being tested is narrow: on data the model has never influenced and never seen, did the measured behavior hold?

The definition has to be stated that strictly because leakage erodes it quietly. If a researcher tuned a hyperparameter on the test set, normalized features using statistics computed over the full history, or stopped training when test performance peaked, the test set is no longer out-of-sample in any meaningful sense. It has become part of the fitting process through the researcher's choices.

A useful copula definition: calibration is the match between a model's stated probabilities and observed frequencies. Calibration measured on training data is nearly uninformative, because flexible models can memorize the frequencies. Calibration measured on held-out data is evidence, bounded by how much held-out data exists.

How is the split supposed to be done?

The textbook procedure is a three-way split: fit on a training set, tune on a validation set, and report once on a test set touched last. Financial data breaks the textbook assumptions, because returns are close to serially dependent and regimes persist. A split placed at a random date leaks information across the boundary through overlapping windows and features computed over trailing periods.

The standard adaptation is walk-forward evaluation: fit on data up to a date, predict the next period, advance the window, and repeat. Marcos López de Prado's 2018 book Advances in Financial Machine Learning describes purged cross-validation, which removes training observations whose label windows overlap the test period. Both approaches reduce leakage; neither eliminates the deeper problem, which is that there is only one history.

That scarcity is the structural weakness. An equity model tested on 2010 through 2020 has seen one long bull market, two sharp drawdowns, and no repetition of either. The evaluator's question shifts from did it pass to what did it pass on.

Why is out-of-sample performance still overstated?

Because the split is usually reported as if it were the only test, when it was the last of many. Halbert White's Reality Check, published in Econometrica in 2000, addressed exactly this: when many candidate models are evaluated on the same dataset, the best one's apparent performance is inflated by selection. The test corrects for the number of trials.

The 2014 Journal of Computational Finance paper by Bailey, Borwein, López de Prado and Zhu gave the problem a working measure: the probability of backtest overfitting, the chance that the configuration chosen as best in-sample lands in the bottom half of configurations out-of-sample. Under their simulations, trying enough configurations on a fixed history makes a false discovery the expected outcome rather than the accident.

Bailey and López de Prado's Deflated Sharpe Ratio, published in the Journal of Portfolio Management the same year, adjusts a reported Sharpe ratio for the number of trials, the variance of trial results, and the sample length. A reported Sharpe ratio of 1.0 earned as the best of 100 trials on three years of data can carry a deflated value near zero. The correction is arithmetic, not opinion.

How should a reader check a paper's claims?

The check is a short list of questions, and a paper that answers none of them has not reported a test in the sense above.

1. How many configurations were tried before the reported one, and were all trials disclosed?
2. What defined the train and test periods, and were features computed without full-history statistics?
3. What was the baseline, and does the reported metric beat it by more than the trial count would explain?
4. What asset universe and date range, and how many distinct regimes does that range contain?
5. Is the performance claim labeled as the authors' own result, or as independently verified?

Question five matters more than it appears. Replication attempts of published market-timing results routinely find weaker effects than the originals, and the site's standing rule applies to reading as much as to writing: claimed performance and verified performance are different things. The same discipline appears in regulation. The Federal Reserve's Supervisory Guidance on Model Risk Management, SR 11-7 of 2011, requires banks to validate models against outcomes the models did not influence, and to document known weaknesses.

What can an out-of-sample test not establish?

It cannot establish that a model will work live. The out-of-sample period is still one draw from a distribution, and the paper's own evaluation window may contain no regime resembling the reader's deployment period. Performance stated on 2015 through 2019 says nothing about a rate shock, whatever the metric.

It also cannot establish absence of overfitting when the research process itself was iterative. If the researcher saw intermediate results and adjusted the design, the number of effective trials exceeds the number disclosed, and every correction formula that assumes a known trial count undercorrects.

Finally, it cannot establish transfer. A model that calibrates well on US equities out-of-sample has been tested on US equities and nothing else. Cross-asset or cross-market claims require cross-asset or cross-market evaluation windows, stated explicitly. What remains unknown after a clean out-of-sample test is still substantial; the honest paper says so in its own limitations section, and the reader should treat one that does not as a warning rather than a promise.

Uzu News publishes information about forecasting, not investment advice, and this report makes no claim about any firm's current capabilities. The methodological angle is the reason the cases belong on this site: both orders turn on the distance between a claimed model and a deployed one.

What did the orders actually allege?

The SEC's order against Delphia alleges the firm stated in public materials that it used machine learning and aggregated client data to train its models, while the client consent and data-use practices described did not support that statement, per the 2024 order. The Global Predictions order alleges the firm called itself the first regulated AI financial advisor and said AI was incorporated into its forecasts, claims the firm could not substantiate, and separately found misstatements about assets under management.

Neither order required the SEC to evaluate whether any model worked. The violation was representational: saying AI drove a process it did not drive. That is the same claimed-versus-deployed distinction this publication applies to vendor benchmarks, applied here by an enforcement agency.

Why does this matter beyond two small firms?

Because the evidentiary standard travels. A model claim that cannot be checked against a deployed system is a marketing claim wherever it appears, in a pitch deck or a paper. The SEC's 2024 actions establish that regulators will price that gap in dollars, at whatever scale a firm's assets put at stake.

What remains unknown is how the commission will handle claims about model performance, as distinct from model use. These orders addressed misdescription; a case over an unverifiable accuracy claim would raise harder measurement questions, and none has yet been brought on those grounds.

What is the multiple-testing problem in signal research?

Multiple testing is the inflation of false positives that occurs when many candidate signals are evaluated against the same data and only the survivors are reported. A single test at the 5% level accepts a one-in-twenty chance of a false discovery. Run the procedure across hundreds of candidates and false discoveries stop being an accident and become an expected yield.

The count in the academic record is the part that can be documented. Harvey, Liu and Zhu assembled 313 articles — 250 published papers and 63 working papers — covering 316 distinct factors, with empirical tests running from 1967 forward. That is the visible denominator. The authors' conclusion from it was blunt: they argue that most claimed research findings in financial economics are likely false.

Their remedy was arithmetic rather than rhetorical. If the literature is a single large family of tests, the significance hurdle for admission to it has to rise with the size of the family. The hurdle they recommend for a newly proposed factor is a t-statistic greater than 3.0.

How is the correction actually measured?

The corrections come from statistics, not from finance, and they differ in what they promise. Family-wise error rate, or FWER, is the probability of making at least one false discovery across the whole set of tests. False discovery rate, or FDR, is the expected proportion of the discoveries you report that are false. Controlling the first is stricter; controlling the second is more permissive by design.

Harvey, Liu and Zhu apply three established procedures to the factor literature: Bonferroni, Holm, and Benjamini, Hochberg and Yekutieli, usually shortened to BHY. They also build a model that accounts for correlation among the test statistics, because the standard procedures assume a level of independence that a library of overlapping accounting ratios does not have.

The practical output of all three is the same object: a higher bar. A t-statistic of 2.1 that would be reported as significant in isolation is, inside a family of several hundred tests, an unremarkable draw. Nothing about the signal changes. What changes is the number of competitors it was silently measured against.

What happened when the anomaly literature was re-tested?

The correction stopped being theoretical when someone rebuilt the anomalies from scratch. Hou, Xue and Zhang compiled 452 anomaly variables and replicated them over a January 1967 to December 2016 sample, published in the Review of Financial Studies in 2020. Their replication used NYSE breakpoints and value-weighted returns rather than the more permissive conventions common in the original papers.

The headline number is a failure rate, not a performance number. In the authors' own words, “65% of the 452 anomalies in our extensive data library, including 96% of the trading frictions category, cannot clear the single test hurdle of the absolute t-value of 1.96.” That is before any multiple-testing adjustment is applied at all.

Applying one makes it worse. The same abstract continues: “Imposing the higher multiple test hurdle of 2.78 at the 5% significance level raises the failure rate to 82%.” Read carefully, that is two separate findings stacked — methodological choices alone kill most of the sample, and the multiplicity correction kills most of what survives. The published replication reports both hurdles side by side for exactly that reason.

Neither paper claims the survivors are tradeable, and neither is a statement about what any signal will do next. They are statements about how much of a published record survives contact with a stricter test.

How do supervisors treat the same problem?

Bank supervisors arrived at the multiplicity problem from the governance side rather than the statistical one. The Federal Reserve's SR 11-7, issued April 4, 2011, defines model risk as “the potential for adverse consequences from decisions based on incorrect or misused model outputs and reports,” and locates it in two places: models with fundamental errors, and models used incorrectly or inappropriately.

Its definition of backtesting is the operative one for signal work. The guidance defines backtesting as “the comparison of actual outcomes with model forecasts during a sample time period not used in model development.” The clause after the comma is the whole test. A backtest run on the data that selected the signal is not out-of-sample, whatever the code returns.

The guidance also requires that validation carry “a degree of independence from model development and use,” and treats outcomes analysis — comparing model outputs to corresponding actual outcomes — as a core validation element rather than an optional one. The supervisory letter is addressed to banking organizations, not to researchers, but the structural point transfers: the party that built the signal is the worst-placed party to certify it.

On the disclosure side, the Securities and Exchange Commission's adviser marketing rule, Rule 206(4)-1, restricts hypothetical performance — the category that covers backtested results in advertisements. Under the agency's compliance guide, an adviser must adopt policies ensuring the hypothetical performance is “relevant to the likely financial situation and investment objectives of the intended audience” and must provide certain information underlying it. The compliance date was November 4, 2022. This is a description of a rule, not legal advice.

Where does the correction fail?

The first failure is the denominator. Every procedure described above needs to know how many tests were run, and the true count is unobservable. Harvey, Liu and Zhu could count 316 published factors; nobody can count the specifications that were tried and abandoned, which is where most of the multiplicity actually lives.

The second is correlation. Bonferroni, Holm and BHY were built for tests that are independent or nearly so. Financial signals constructed from overlapping accounting data are heavily correlated, which makes the naive corrections misstate the true hurdle — the reason Harvey, Liu and Zhu develop a separate framework that models the correlation structure instead of assuming it away.

The third is that a corrected t-statistic is still a statement about a historical sample. It says a pattern is unlikely to be a fluke of that window. It does not say the pattern has an economic cause, that the cause persists, or that the effect survives trading costs. Hou, Xue and Zhang's finding that 96% of the trading-frictions category fails even the uncorrected 1.96 hurdle is a reminder of how much apparent structure lives in the implementation details.

What these methods cannot establish

A multiple-testing correction cannot rescue a signal from a badly specified test. If the sample, the breakpoints, or the weighting scheme were chosen after seeing the result, the corrected statistic inherits the same contamination at a higher threshold. The Hou, Xue and Zhang exercise is instructive precisely because the convention change and the multiplicity adjustment are reported separately.

These procedures also cannot verify claims made outside the published record. A vendor result presented without its evaluation window, sample construction, and the number of variants tested cannot be corrected, because the inputs the correction requires are missing. A claim of that shape is a claim, not a measured result, and remains one until its conditions are stated.

None of the sources here measured live, post-publication returns of the signals in question, and none of them make a statement about future performance. Neither does this article. Uzu News covers forecasting methods; it does not forecast, and nothing above is a recommendation to act on any model output. The papers cited report what happened to a defined set of variables over a stated historical window, on stated conventions — and that, stated exactly, is the entire claim.

The new guidance, issued jointly as Federal Reserve SR letter 26-2 and OCC Bulletin 2026-13, rescinds OCC Bulletin 2011-12 — the 2011 supervisory letter known across the industry by its Federal Reserve designation, SR 11-7 — along with the Comptroller's Handbook booklet on model risk management, a 1997 credit-scoring bulletin, and the 2021 interagency statement on model risk for Bank Secrecy Act and anti-money-laundering compliance. The agencies describe the rewrite as an effort to "clarify model risk management principles" and set out "a risk-based approach" scaled to a bank's size and complexity, according to the OCC bulletin.

What is model risk management?

Model risk management is the discipline of governing, developing, and validating the quantitative tools a bank uses to price, forecast, or decide. The guidance defines a model as "a complex quantitative method, system, or approach that applies statistical, economic, or financial theories to process input data into quantitative estimates," per the OCC bulletin's text. That definition covers credit-scoring engines, stress-test models, fraud-detection systems, and market-risk forecasts alike — any tool converting inputs into a numerical estimate a bank relies on.

The framework rests on three linked activities: model development and implementation, independent validation, and governance that assigns clear ownership. The 2026 guidance keeps this structure from SR 11-7 largely intact. What moves is the applicability standard and the enforcement posture, not the validation vocabulary examiners have used since 2011.

What actually changed from SR 11-7?

Three changes stand out against the 2011 text, based on the OCC bulletin and the Federal Reserve's SR 26-2 letter. First, applicability is now explicitly tiered by size: the guidance states it "is expected to be most relevant to banking organizations with over $30 billion in total assets," though the agencies note it may still bear on smaller banks with concentrated model exposure. Second, generative and agentic AI models are carved out entirely — the bulletin states "generative AI and agentic AI models are novel and rapidly evolving. As such, they are not within the scope of this guidance," with the agencies signaling a separate request for information to come. Third, the guidance now says directly what SR 11-7 left to supervisory interpretation: "This guidance does not set forth enforceable standards or prescriptive requirements; accordingly, non-compliance with this guidance will not result in supervisory criticism against a banking organization."

That third change reframes the document's legal weight without discarding its content. A bank that skips a step described in the bulletin is not automatically out of compliance in the way it would be under a binding rule — but examiners retain the underlying safety-and-soundness authority the guidance was written to support, and the validation practices it describes remain the reference point supervisors will still ask about.

How is a model's performance actually validated?

Validation under the new guidance runs through the same two-part test the industry has used for a decade and a half: conceptual soundness and outcomes analysis, backed by ongoing monitoring.

Conceptual soundness comes first. The OCC bulletin describes it as "assessing and documenting model design (including key modeling choices, assumptions, qualitative judgments, and data selection), construction, and developmental testing." This step asks whether the model was built to answer the question it is now being used to answer — before a single live output is checked.

Outcomes analysis follows, and it is where a model's real-world track record enters the picture. The guidance defines it as work that "compares model outputs to corresponding real-world outcomes to assess model performance relative to model objectives and business use," and it names backtesting specifically as one method: outcomes analysis "can take many forms, including testing conducted during model development, reports or analysis performed as part of ongoing monitoring, or standalone activities such as back-testing or outlier analysis." Backtesting here means comparing a model's historical forecasts against what actually happened over a defined window — the same out-of-sample logic that underlies performance claims anywhere in quantitative finance, not a bank-specific technique.

A third element, ongoing monitoring, tracks a model after deployment rather than at a single validation date, watching for performance drift as market conditions or input data shift away from the environment the model was built and tested in.

Binding the three together is a requirement the guidance calls "effective challenge," carried over conceptually from SR 11-7: work performed "by individuals with the appropriate expertise to conduct a critical and objective challenge, sufficient independence to maintain objectivity, as well as the organizational standing and influence to effect any change." In practice, that means the people validating a model cannot be the people who built it, and their findings need enough institutional weight that a flagged problem actually gets fixed rather than logged and ignored.

Validation does not stand alone; the guidance ties it to governance and inventory requirements built to make the three-part test auditable over time. The bulletin states that "model risk management benefits from clear roles and responsibilities with well-defined accountability," with "sound governance practices" assigning named individuals to each stage of a model's life cycle — and it draws a line around internal audit's role, noting that audit "would generally not duplicate model risk management activities such as model development or validation." Banks are expected to keep "an effective model inventory" containing "sufficient information to understand model risks" across every model in development or in use, plus documentation adequate to "support the tracking of recommendations, responses, and exceptions" if a validator flags a problem.

Vendor and third-party models get their own carve-out inside the same framework rather than a separate one. The guidance calls for "validation of vendor products, either by internal or outside parties," paired with "ongoing monitoring and outcome analysis to assess whether vendor models are accurate, remain fit for purpose, and continue to be reliable" — an acknowledgment that a bank buying a model off the shelf cannot inspect its internal construction the way it could a model built in-house, and so must lean more heavily on outcomes analysis to catch problems conceptual-soundness review cannot reach. Where a bank customizes a vendor model for its own use, the guidance says "sound practice involves appropriately documenting, justifying, and evaluating adjustments made to customize the model," treating each customization as a change that itself needs validation.

Where does this validation approach fail?

The guidance itself is not enforceable, which limits what it can be credited with achieving. Because the agencies state outright that non-compliance "will not result in supervisory criticism," the bulletin cannot be read as a floor that guarantees any particular bank's validation program meets a fixed bar — it describes sound practice without mandating it, and two banks can diverge widely in rigor while both remaining technically consistent with the guidance.

The framework also does not yet reach the models banks are now fastest to adopt. Generative and agentic AI systems are explicitly out of scope because, in the bulletin's own words, they "are novel and rapidly evolving," with a request for information promised but not yet issued as of this guidance's publication. Until that RFI produces its own standard, a bank deploying a large language model in an underwriting or fraud workflow is validating that system against no interagency benchmark specific to how such models fail — only against the general model-risk principles written for more conventional statistical and econometric tools.

Outcomes analysis and backtesting, the guidance's own preferred evidence of performance, carry a structural limit the bulletin does not resolve: they can only be run against a period that has already happened. A validation report built on outcomes analysis says how a model performed on its own historical or recent window — it does not, and cannot, establish how the same model will perform once market conditions move outside that window. The guidance's emphasis on ongoing monitoring is a partial answer to that gap, not a solution to it; monitoring detects drift after it has already begun to degrade performance, not before.

Finally, the $30 billion applicability threshold means the guidance's own text treats validation rigor as calibrated to bank size rather than to a model's actual materiality. A smaller institution running a high-stakes model — a concentrated lending book scored by a single algorithm, for instance — sits outside the guidance's primary intended audience even though the underlying risk the framework describes does not scale down with asset size.

The quantity doing the work in that sentence is the one most published backtests omit: N, the number of configurations a researcher evaluated before reporting the best one. A backtest is a simulation of a decision rule's performance on historical data. It stops being a test and becomes a search the moment more than one rule is evaluated, and a search across enough candidates will find an attractive equity curve in noise alone.

What exactly is backtest overfitting?

Backtest overfitting is the selection of a configuration whose measured historical performance reflects the number of variants tried rather than any repeatable effect. It is not a coding error and not lookahead bias. The rule is fitted correctly to history; the failure is that history was searched many times and only the winner was reported.

Out-of-sample means evaluated on data that was used neither to fit the model nor to select it. The second half of that definition is the part that fails quietly. A holdout period reused across dozens of candidate specifications has been used for selection, whatever the code comments say, and its results carry the same selection bias as the fitting sample.

Bailey and his co-authors put the reporting failure in blunt terms, writing that "not reporting the number of trials (N) involved in identifying a successful backtest is a similar kind of fraud" (p. 466). Their conclusion is equally direct: because most published backtests do not report N, many of them may be overfitted, and the paper warns that "positive backtested performance will often be followed by negative investment results" (p. 468).

How much history does a backtest actually need?

Enough history that the best result among N trials cannot plausibly be noise. The paper's Theorem 2 formalizes this as a Minimum Backtest Length: the years of data needed to avoid selecting a strategy with an in-sample Sharpe ratio of E[max N] among N independent strategies whose expected out-of-sample Sharpe ratio is zero.

The bound is MinBTL < 2 ln[N] / E[max N]². It scales with the logarithm of the trial count and inversely with the square of the Sharpe ratio being claimed, which is why it bites hardest on short samples and modest performance claims. The paper's own illustration, in its Figure 2 on p. 462, sets the expected maximum Sharpe ratio at 1 and asks how many independent trials a given sample length can absorb.

Both figures come from the same paper and the same assumptions: independent trials, Sharpe ratio estimates treated as Gaussian, and a true out-of-sample Sharpe ratio of zero for every candidate. Correlated configurations, which is what a real parameter sweep produces, are not the case the bound is derived for.

Why can more in-sample optimization make out-of-sample results worse?

Because under some conditions the two are negatively linked rather than merely uncorrelated. Where the paper introduces compensation effects, imposed by global constraints or by serial dependence in returns, it reports that "a strongly negative linear relation between performance IS and OOS arises, indicating that the more we optimize IS, the worse the OOS performance" (p. 465).

Its Proposition 3 states the ordering directly: with those global constraints imposed, a configuration with the higher in-sample Sharpe ratio has the lower out-of-sample Sharpe ratio. That inverts the intuition a backtest is usually asked to support. It does not mean optimization always destroys performance; it means the sign of the relationship is a property of the setting, not a constant, and cannot be assumed from the in-sample number alone.

What does the multiple-testing literature add?

It supplies the same correction at the scale of an entire research field. Campbell Harvey, Yan Liu and Heqing Zhu, writing in the Review of Financial Studies in 2016, catalogued 316 factors from 313 articles studying cross-sectional return patterns, then applied multiple-testing adjustments to the reported t-statistics rather than accepting each finding on its own.

Their headline threshold is a raised bar for new work: "a new factor needs to clear a much higher hurdle, with a t-statistic greater than 3.0" (p. 5), against the conventional 2.0. Applied backward to the existing literature, their tabulation finds that of 296 published significant factors, 158 would be considered false discoveries under Bonferroni, 142 under Holm, 132 under Benjamini-Hochberg-Yekutieli at 1% and 80 at 5% (p. 37).

The spread between 80 and 158 is not a defect in the analysis; it is the answer changing with the error rate being controlled. Bonferroni and Holm control the family-wise error rate, the probability of any false positive at all. The Benjamini-Hochberg-Yekutieli procedure controls the false discovery rate, the expected share of false positives among rejections. Neither result names an individual factor as false.

How do bank supervisors treat backtesting?

As monitoring, not discovery. The Federal Reserve and the Office of the Comptroller of the Currency issued joint guidance on model risk management, SR 11-7, on April 4, 2011, applying to supervised banking organizations in proportion to each one's size, nature and complexity. Backtesting appears there as a subordinate practice inside a larger validation structure.

The guidance defines outcomes analysis as comparing model outputs to corresponding actual outcomes, and describes back-testing as one form of it, involving the comparison of actual outcomes with model forecasts. Validation itself is defined as the processes intended to verify that models perform as expected, in line with their design objectives and business uses.

The letter sets out three core validation elements: evaluation of conceptual soundness, ongoing monitoring, and outcomes analysis. The ordering is the point for anyone reading a performance claim. A favorable backtest satisfies at most the third element, and only for the period tested; it says nothing about whether the model's design is defensible or whether it is still behaving as designed today.

What has to accompany backtested numbers in an advertisement?

In the United States, backtested results shown by a registered investment adviser fall under the Securities and Exchange Commission's marketing rule as hypothetical performance, which the rule prohibits in advertisements unless specific conditions are met. The rule's compliance date was November 4, 2022, eighteen months after its May 4, 2021 effective date.

Under the SEC's small-entity compliance guide, an adviser must adopt and implement policies and procedures reasonably designed to ensure that the performance is relevant to the likely financial situation and investment objectives of the intended audience, and must provide certain information underlying the hypothetical performance. The guide also notes that hypothetical performance does not include performance generated by interactive analysis tools, and that it does not qualify for the one-on-one communication exclusion unless provided in response to an unsolicited investor request or to a private fund investor.

That is a description of what the rule requires, not an assessment of any firm's compliance with it. What it establishes for a reader is narrower and useful: a backtested track record is a regulated category of claim, and the disclosures attached to it are part of the evidence, not decoration around it.

What none of this establishes

The Minimum Backtest Length bound is derived for independent trials. Real parameter sweeps are heavily correlated, so the effective number of independent trials is smaller than the raw count of runs, and the bound functions as a discipline on reporting rather than a precise threshold for any particular study.

Neither paper measures live trading results. The Bailey framework is a statistical argument about selection under simulation; the Harvey, Liu and Zhu tabulation is an argument about published t-statistics. Transaction costs, slippage, financing, capacity and market impact are outside both, and a strategy that survives every correction described here can still fail on those grounds.

Nor can any of it identify which specific published result is spurious. The false-discovery counts are properties of a set of 296 findings, and they move by a factor of roughly two depending on which error rate is controlled. No named factor, firm, or product is implicated by those numbers, and this article does not evaluate any.

Finally, a reported trial count cannot be verified from outside. A researcher who reports N is reporting only their own trials; specifications tried and abandoned by others on the same public dataset are invisible to every correction discussed here, which means the adjusted thresholds are floors rather than guarantees. None of the above is investment advice, and no figure in it describes what any market or model will do next.

The diagnostic comes from a 2015 paper by David H. Bailey, Jonathan Borwein, Marcos López de Prado and Qiji Jim Zhu, and it targets a specific failure mode in quantitative finance: a backtest can be optimized, directly or by trial and error, until it fits its own historical sample rather than any persistent market pattern. A high Sharpe ratio alone does not distinguish the two cases. PBO was built to.

What does overfitting mean in a backtested strategy?

Overfitting is what happens when a strategy configuration is tuned until it matches the noise in a specific historical sample rather than a pattern likely to recur. The authors define it precisely: overfitting occurs when the configuration with the best in-sample (IS) performance systematically underperforms the other tested configurations out-of-sample (OOS).

The distinction matters because in-sample performance is not itself evidence of a working strategy. Any sufficiently flexible search process, with enough parameters and enough variations tried, can find a configuration that performed well on a fixed historical window purely by chance. Out-of-sample underperformance is the signature that reveals it.

How is probability of backtest overfitting actually calculated?

The paper's proposed method, combinatorially symmetric cross-validation (CSCV), requires only the time series of backtested returns for each configuration under consideration, not knowledge of the trading rule itself. It splits the historical performance data into S equal submatrices, forms every possible training set from half of those submatrices, and tests each one against its complementary out-of-sample half.

For each split, CSCV ranks how the in-sample winner performed out-of-sample and converts that rank into a logit score. Repeating the process across every combination produces a distribution of those scores; PBO is the share of the distribution below zero, meaning the share of splits in which the in-sample winner ranked below the out-of-sample median. With S set to 16 partitions, the method generates 12,780 combinations, which the authors report is sufficient for acceptable statistical error.

The comparison illustrates the paper's central point: two strategies with comparable, even favorable, in-sample Sharpe ratios can carry very different odds of having been overfit, and the Sharpe ratio alone does not separate them.

Does testing more strategy configurations raise the risk?

Yes, and the authors treat this as the core reason PBO is needed at all. They write that the probability of finding false positives increases with the number of tests conducted on the same data, and they single out a common shortcut, the simple hold-out method, for a specific flaw: it does not take into account the number of trials attempted before a particular configuration was selected. CSCV is built to incorporate that trial count directly, through the relative ranking of every tested configuration against every other, rather than validating one chosen configuration in isolation.

The paper also separates two related but distinct questions. PBO answers whether the in-sample winner is likely to underperform the group's out-of-sample median. A second statistic the authors report alongside it, the probability that the selected strategy's out-of-sample return is negative, can remain elevated even when PBO looks moderate. In one set of examples with in-sample Sharpe ratios between 1 and 3, all of them positive in-sample, the authors found that roughly 78% of the corresponding out-of-sample Sharpe ratios came back negative, illustrating that a strategy can clear one overfitting check and still lose money out-of-sample by a different measure.

Why do regulators also require disclosure around backtested performance?

Backtest overfitting is not only an academic concern. Federal bank regulators and securities regulators treat backtested and hypothetical model results as a distinct risk category requiring independent scrutiny and disclosure. Supervisory guidance jointly issued by the Federal Reserve and the Office of the Comptroller of the Currency on April 4, 2011, known as SR 11-7, requires that model outputs be validated through outcomes analysis, defined as comparing actual outcomes against model forecasts using a sample period not used in development, and that this validation work be subject to critical review by a party independent of the model's development.

On the securities side, a Securities and Exchange Commission risk alert dated September 19, 2022 on examinations of the investment adviser marketing rule addresses hypothetical performance separately from actual track records, requiring advisers who present it to adopt policies ensuring the performance shown is relevant to the likely financial situation and objectives of the audience receiving it, among other conditions. Both regimes start from the same premise CSCV formalizes: a backtest's headline number is not self-validating.

Where does this diagnostic break down?

PBO measures whether a selection process across a defined universe of configurations reproduced overfitting; it does not certify that a strategy will perform well going forward, and the authors are explicit about several limits on what the number can establish.

• A flawed backtest produces a flawed PBO. The authors state that if the underlying backtest used incorrect assumptions, such as wrong transaction costs, the method assesses overfitting relative to flawed information and inherits the flaw.
• The symmetric-split design assumes the return series does not have strong autocorrelation across the sample; the authors note it may not suit every strategy type.
• Structural breaks are only visible within the dataset actually used. CSCV cannot anticipate a regime change that has not yet occurred in the historical sample.
• Results depend on knowing every configuration actually tried. If some tested and discarded configurations are left out of the input, a pattern the authors call the file drawer problem, the reported PBO understates true overfitting.
• Using PBO itself to search for the best-scoring configuration reintroduces the same problem it is meant to detect; the authors warn that any counter-overfitting technique used to select a strategy will result in overfitting.
• A high PBO among several configurations that all perform similarly well can also reflect a group of genuinely skillful strategies clustered together, not proof that all of them failed.

Regulatory frameworks carry a parallel limitation. SR 11-7 acknowledges that institutions may face constrained data and limited ability to use key model validation tools, in which case the guidance calls for closer attention to a model's stated limitations rather than treating validation as complete. No diagnostic in this space, statistical or supervisory, substitutes for live, out-of-sample performance measured after the fact.

The authors' own recommendation for practitioners follows from these limits rather than around them: disclose every configuration tested, not only the one published, and treat PBO as a reporting requirement rather than a search target. They suggest that journal editors and investors could reasonably ask researchers to estimate and report the probability when a backtest is submitted, on the same logic that clinical trials disclose every arm tested, not only the one that worked.

Frequently asked questions

Does a low PBO mean a strategy will work in live trading?

No. A low PBO indicates the selection process among the tested configurations did not show the signature of overfitting; it says nothing about future market conditions, capacity constraints, or execution costs the backtest did not model.

What is the file drawer problem in this context?

It is the risk that configurations tested and discarded before the final version never enter the PBO calculation, which understates how much searching actually produced the reported result.

Does SR 11-7 apply only to banks?

SR 11-7 was issued to institutions supervised by the Federal Reserve and the OCC, but its outcomes-analysis and independent-review framework is widely cited across finance as a reference standard for model validation generally.

This explainer covers how backtest overfitting occurs, how researchers try to measure it, and what disclosure rules now require of firms that advertise hypothetical results. It does not evaluate, endorse, or predict the performance of any specific model.

This matters beyond quantitative trading desks. The same search-and-select dynamic applies to any AI system marketed on the strength of its historical accuracy, including forecasting tools built on machine learning models that scan thousands of feature combinations automatically. The more configurations a system tries, the more likely it is to stumble onto one that fits the past well by coincidence, which is exactly why understanding the mechanism is a matter of basic literacy for anyone evaluating a forecasting claim, not a niche concern for statisticians.

What a backtest actually measures

A backtest applies a trading or forecasting rule to historical price or economic data to see how it would have performed. On its own, that is a reasonable diagnostic step. The trouble starts when a rule is not tested once but searched for, meaning a researcher or an automated optimizer tries many variations of parameters, indicators, or entry and exit conditions until one combination produces an attractive result on the historical sample.

A methodology paper by researchers David Bailey and Marcos Lopez de Prado illustrates the underlying problem with a simple example: a coin is flipped ten times, producing five heads followed by five tails. A researcher could construct a "seasonal" rule to exploit that pattern. Applied to the next ten flips, the rule breaks even, because the original pattern was random noise, not a repeatable signal. The researchers describe the general risk this way: a systematic search through a large space of strategies will eventually turn up one that appears to profit purely from how the random historical data happened to fall, and that chance pattern is unlikely to repeat going forward.

Why more testing makes the problem worse, not better

Intuitively, testing more variations of a strategy should improve the odds of finding a good one. Statistically, it also improves the odds of finding a strategy that looks good purely by chance. This is a version of the multiple-comparisons problem familiar from other fields of science: run enough trials against noisy data, and some subset will clear an impressive-looking bar even if nothing in the data is real.

Bailey and Lopez de Prado formalized a way to correct for this, called the deflated Sharpe ratio. It adjusts a strategy's headline risk-adjusted return downward based on how many independent variations were tested to find it, along with the length of the sample and the shape of its return distribution. In one worked example in their paper, a strategist reports an annualized Sharpe ratio of 2.5 over five years after 88 independent backtest trials. Once adjusted for that search process, the result implies only about 90 percent statistical confidence that the true, underlying Sharpe ratio is even above zero, short of the 95 percent threshold the authors treat as a reasonable bar for allocating capital. The same headline number, discovered with far fewer trials, would have cleared that bar. The number on the chart had not changed. The confidence behind it had.

How this shows up in disclosure rules

Financial regulators have built specific guardrails around this exact risk. The U.S. Securities and Exchange Commission's Investment Adviser Marketing Rule, which took full effect for advisers in November 2022, requires that any hypothetical performance shown to a general audience be accompanied by policies and procedures reasonably designed to ensure the results are relevant to the likely financial situation and objectives of the people seeing them, along with underlying information supporting the figures. Notably, the rule treats hypothetical performance, a category that includes backtested and model-generated results, as inherently higher-risk than actual account performance, precisely because it is produced after the fact and is vulnerable to the kind of search-driven overfitting described above.

The rule has already produced enforcement action. In September 2023, the SEC announced settlements with nine registered investment advisers that had advertised hypothetical performance, including backtested and model results, to the general public on their websites without the required policies in place; two of the firms were also cited for failing to retain copies of the advertisements. The firms agreed to combined penalties of $850,000, ranging from $50,000 to $175,000 each, without admitting or denying the findings. The SEC's enforcement director at the time, Gurbir Grewal, said hypothetical performance advertising "may present an elevated risk for prospective investors whose likely financial situation and investment objectives don't match the advertised investment strategy." The action did not turn on whether any firm's underlying model was accurate. It turned on whether firms disclosed the conditions under which hypothetical results were generated and to whom they were shown. For AI-driven forecasting tools specifically, this distinction carries extra weight: a model can run a parameter search spanning thousands of automated iterations, far more than a human analyst manually testing variations by hand, which widens the same statistical gap between an eye-catching backtest and a defensible one.

What a strong methodology looks like

Researchers and practitioners who take this problem seriously tend to build a few habits into their process. They reserve a portion of historical data that is never touched during model design, and test the finished model against it only once. They report how many variations were tried before settling on a final rule, not just the best-performing one. They treat a strategy's performance on data collected after it was finalized, sometimes called out-of-sample or live performance, as far more informative than the backtest that produced it in the first place. None of these steps make a forecasting method correct. They make its historical track record harder to fake by accident.

None of this means every backtest is meaningless, or that firms disclosing hypothetical results are acting in bad faith. Backtesting remains a standard, useful diagnostic step in building any forecasting method. The point is narrower: the number on the chart, by itself, does not distinguish a well-validated method from a lucky one, and readers evaluating any AI forecasting claim benefit from asking how the result was found before asking what the result was.

Limitations

This explainer describes a known mechanism in quantitative methodology and the disclosure framework built around it; it is not an assessment of any specific model, adviser, or product, and readers should not treat it as one. Backtest overfitting is difficult to detect from the outside: a firm's internal testing process, including how many variations it tried before settling on a published result, is rarely visible to an investor reviewing a marketing page. The deflated Sharpe ratio and similar corrections are estimation tools built on statistical assumptions, including how independent different trials really were, and different analysts applying them to the same data can reach different adjusted figures. Regulatory disclosure requirements address whether hypothetical results were presented appropriately and to the right audience; they do not certify that a given model or strategy will perform any particular way in the future. Historical patterns in financial data can also shift over time for reasons unrelated to overfitting, such as changes in market structure, so even a rigorously tested, out-of-sample-validated model carries no guarantee of future results. Nothing in this article should be read as investment advice, a prediction of any market outcome, or a recommendation for or against any specific method or product.

The July vote completes the first half-year's arc in a single number: a committee that began 2026 with two members pushing to cut ended its seventh meeting with three pushing to raise. The direction of internal dissent has reversed while the policy rate never moved — a reminder that the vote count and the rate are two different series.

What does the 9-3 vote say?

Structurally, that three members judged current policy too loose given their reading of the data — the statement's persistent inflation concern, including the energy-price attribution carried since spring, now strong enough in some members' assessment to prefer a higher range. The published statement records the vote and the direction of each dissent; the members' reasoning arrives in detail with the minutes, on the usual lag. What the vote does not say is anything about the next meeting: dissent patterns through 2026 have already shown the count swinging from 12-4 to 12-0 to 9-3 across three consecutive meetings, and a committee that variable internally is poorly described by any single meeting's split.

How does July complete the half-year sequence?

The full sequence, all published: January 28, hold, two dovish dissents; March 18, hold, one dovish dissent, Middle East language added; April 29, hold, four dissents — three against the easing-bias clause — with inflation tied partly to energy; June 17, hold, 12-0 under the shortened statement format introduced that month; July 29, hold, 9-3 with hawkish dissents. The policy range never changed; the description of the economy and the internal coalition both changed repeatedly. Statement-format changes add a documentation caution for the year: comparing the shortened June-July statements against the longer earlier ones requires reading what was dropped and what was kept, not just the vote counts.

What is the measurable bottom line?

Seven months, five meetings, five holds at 3.50 to 3.75 percent, dissent swinging from dovish to hawkish across them, and a statement text that grew an energy attribution and then shortened its whole form. Those are the auditable facts. Whether the hawkish dissents presage a hike — the question every summary will ask — is a forecast, and forecasts belong to their makers with their conditions attached. The primary documents for every claim in this article are the statements themselves at federalreserve.gov, with minutes and the 2026 meeting calendar alongside them.

What did the SEC's first AI-washing orders actually allege?

Delphia (USA) Inc., a Toronto-based adviser, told clients between 2019 and 2023 that it used artificial intelligence and machine learning to analyze their personal spending and social-media data, feeding the results into stock-picking algorithms. According to the SEC's order against the firm, Delphia collected some client data intermittently but never used that data with artificial intelligence or machine learning, or as an input into its investing algorithms, at any point.

Global Predictions Inc., a San Francisco-based adviser, separately marketed itself as the "first regulated AI financial advisor" and claimed to deliver "expert AI-driven forecasts" — claims the SEC found were not substantiated. Both firms settled without admitting or denying the findings. Delphia paid a $225,000 penalty and accepted a censure; Global Predictions paid $175,000. Both were ordered to cease and desist from further violations of Sections 206(2) and 206(4) of the Investment Advisers Act, the provisions covering fraudulent and misleading statements to clients.

Then-SEC Chair Gary Gensler framed the enforcement rationale in plain terms: "Investment advisers should not mislead the public by saying they are using an AI model when they are not. Such AI washing hurts investors." Enforcement Director Gurbir S. Grewal added the underlying standard the orders apply: "if you claim to use AI in your investment processes, you need to ensure that your representations are not false or misleading."

How is "no model to validate" different from a model that fails validation?

The SEC's orders describe an absence, not an error: no algorithm existed for AI or machine-learning claims to be tested against. That is a different failure than a model that runs, produces forecasts, and is later found miscalibrated — the scenario federal model-risk guidance for banks was written to address.

Model risk management is the discipline of managing "the potential for adverse consequences from decisions based on incorrect or misused model outputs and reports," as defined in joint 2011 guidance from the Federal Reserve and the Office of the Comptroller of the Currency. That guidance assumes a model is already running: it sets out three validation elements — conceptual soundness, ongoing monitoring, and outcomes analysis, the last comparing actual results to model predictions through back-testing over a stated evaluation period. None of the three applies to a claim about a model that was never built. Conceptual soundness requires design documentation to review; outcomes analysis requires outputs to compare against realized results. The Delphia and Global Predictions matters never reach that stage, because the system supposedly generating those outputs, per the SEC's findings, did not exist as described.

What would model risk management guidance have required before a model reached investors?

Interagency guidance requires disciplined development, independent "effective challenge" from staff not involved in building the model, board-level governance, and review at least annually — a structure built for models that exist and produce output, not for marketing copy describing capabilities that were never deployed.

Applied here, the guidance's checklist explains the size of the gap regulators found. Conceptual soundness review would have asked for the model's design and assumptions; there was none to review. Ongoing monitoring would have confirmed the model was implemented and used as intended, including benchmarking against alternative approaches; there was no live implementation to monitor. Outcomes analysis would have compared predicted results to what actually happened over a defined window; there was no track record to benchmark, because the AI system described to clients and regulators in filings, press releases, and marketing material was not operating.

The interagency guidance also calls for "effective challenge": objective review by staff with the incentive, competence, and authority to critically assess a model, independent of the people who built it. In both matters, the SEC's findings suggest that step never had a model to challenge in the first place — the compliance failure sat further upstream, in what the firms told clients and regulators, not in how a model performed once deployed.

What do the SEC's orders not establish?

The orders establish that specific marketing statements were false and settled under securities law; they do not evaluate whether any AI or machine-learning model used elsewhere in the industry actually forecasts well, and they carry no finding on model accuracy, since accuracy was never at issue in either case.

The SEC's Office of Investor Education separately warned, in a January 25, 2024 investor alert, that AI-generated information "might rely on data that is inaccurate, incomplete, or misleading," and urged investors to verify a firm's registration status independently rather than take AI claims at face value. That alert addresses a different risk than the Delphia and Global Predictions orders: platforms actively using AI output to mislead, versus firms claiming AI use that did not occur. Both sit outside the scope of conventional model validation, because in each case regulators examined a marketing or disclosure statement, not a model's calibration.

For a reader evaluating a forecasting claim from any adviser or vendor, the record separates two questions worth keeping apart: whether a model exists and was validated against a stated baseline, and whether its description to clients matches what was actually built and run. The SEC's 2024 orders resolved only the second question, for two firms, under securities marketing law — not the first.

Stress tests are models used as regulations — the output feeds each bank's stress capital buffer directly — so a methodological change is simultaneously a statistics question and a policy question. The averaging change is the latest and clearest example, and its comparability consequences deserve the same methodological treatment this site applies to any modified instrument.

What changed, and why?

Under the pre-2025 design, each year's scenario produced each bank's requirement afresh — scenario severity varied by design, and resulting buffers moved accordingly. The April 2025 proposal, noted in the Fed's own 2025 stress test documentation, would average results over two consecutive years so that a single year's scenario severity no longer drives a discrete jump in requirements; the October 2025 proposals extended the framework, including publishing final scenarios by mid-February, and the November 2025 Federal Register rule added model and scenario transparency provisions. The stated rationale is volatility reduction in requirements — a policy judgment that smoother buffers serve capital planning better than scenario-driven jumps. The tradeoff is inherent: averaging any instrument smooths its signal, by construction.

What does averaging do to the numbers' meaning?

Exactly what averaging always does — the analysis is the same one this site applies to rolling windows. A two-year averaged requirement responds more slowly to a deterioration visible in the current year's test: a weak result is diluted by the prior year's stronger one, and the buffer adjusts halfway in year one and fully in year two, if the weakness persists. Conversely, sharp improvements phase in equally slowly. Comparability breaks across the boundary: 2026 requirements averaged under the new framework are not directly comparable with pre-averaging years, and analysts quoting a year-over-year change in aggregate capital requirements are comparing instruments, not just outcomes — the vintage problem, in regulatory form. The Fed's publications document the framework; the interpretation discipline belongs to readers.

What did the 2026 results themselves show?

Per the June 24, 2026 release: 32 banks tested, aggregate projected losses of nearly $708 billion under the severely adverse scenario, an aggregate peak-to-trough CET1 decline of 1.6 percentage points — a milder projected decline than the 2025 exercise showed for its scenario — with all banks exceeding minimum capital requirements. The scenario design matters to the read: 2026 scenarios were built on December 31, 2025 balance sheets, and scenario severity varies year to year by construction, which is precisely the variability the averaging framework is designed to absorb at the requirements level. Any cross-year comparison of loss projections inherits both scenario differences and methodology differences; the publication itself states the scenario assumptions.

How should readers track the change?

By vintage, as with any revised series: the Federal Reserve publishes the stress test methodology, scenarios and results annually — the 2026 introduction and results at federalreserve.gov — and the averaging framework's first full cycle completes when 2027 requirements blend two averaged years. Comparisons across the boundary should be labeled as such; comparisons within the new regime can begin once two averaged cycles exist. Instruments that change while measuring are the rule in regulation, not the exception — the reader's job, as ever, is to keep the vintage attached to the number.

The number arrives with its vintage label attached: "advance estimate" means it was compiled from partial source data, and the BEA's own documentation assigns it the widest revision bands of the quarter's three estimates. Two more estimates follow, then annual updates and the comprehensive revisions that restate history when methods or sources change.

What does an advance estimate contain?

The advance figure is built from whatever source data were available at the time of preparation: two months of many monthly indicators and one month of others, plus judgments BEA makes to fill gaps — for Q2 2026, the release's own source notes list which components carried imputations, as they do every quarter. Historically, advance estimates of GDP growth are revised by a few tenths of a percentage point on average between the first print and the latest vintage, with larger corrections clustered around turning points. A 1.5 percent advance reading is therefore best quoted as "1.5, advance" — a number with a published error distribution, not a settled fact.

What is the measured context?

Two quarters of roughly 1.5-to-1.6 percent annualized growth describe an economy expanding well below its long-run average pace — the BEA's published long-run real GDP growth runs near 2 to 3 percent over recent decades — and consistent with the slower payroll growth and low unemployment claims environment documented across the first half of 2026. Reading the level against the DXY-style single-number shorthand that headlines encourage is exactly the error this site exists to discourage: the advance estimate is one vintage of one measure, and its components — consumption, investment, trade, government — each carry their own vintage and their own imputation shares, detailed in the release tables.

How should readers track what happens next?

By the published schedule. The second estimate for Q2 follows in late August with corporate-profits data, the third in September, and annual and comprehensive updates thereafter; each revision's size against the advance is itself a documented statistic in BEA's vintage history. The disciplined read of any single advance print is the one the agency itself prints: provisional, scheduled for correction, and comparable only to estimates of the same vintage. The release and its technical notes are at bea.gov.

Turnover appears on every fund fact sheet, gets summarized in one line, and carries a precise meaning that the one-line uses routinely mangle. It is a pace statistic — how fast the book churns — and its consequences follow from arithmetic on costs and taxes rather than from any information about manager talent. High turnover is neither a sin nor a signal; it is a speed, and speed has a price schedule.

How is the ratio computed?

The standard formula takes the lesser of total purchases and total sales — the lesser so that both buying new money and raising cash do not double-count — divided by average monthly net assets. A fund that sold 80 percent of its book and bought 80 percent reports 80 percent, not 160. The ratio is annual and fiscal-year-based, so funds with December and non-December fiscal years are not directly comparable within a calendar year — a small convention that matters when comparing figures pulled from different data services. Excluded from most computations: in-kind creation-redemption activity in ETFs, which is why ETF turnover figures can read oddly low relative to the portfolio's actual position changes.

What are the documented cost consequences?

Trading costs money — commissions, bid-ask spreads and market impact — and those costs sit outside the expense ratio, borne by the fund. The magnitude is measurable in the transaction-cost literature: for liquid large-cap names, costs per trade are small; for small-cap, emerging-market and less liquid bonds, materially larger. The predictable product: turnover times per-trade cost. Two funds with identical expense ratios can therefore carry meaningfully different true costs if their turnover and liquidity profiles differ — a gap that shows up as tracking of nothing in particular for active funds, but as performance drag that academic cost studies have estimated at single-digit to low-double-digit basis points annually at moderate turnover in liquid markets, and worse in illiquid ones.

What are the documented tax consequences?

In taxable accounts, turnover realizes capital gains: every sale of an appreciated position is a distribution waiting to happen. Funds must distribute realized net gains annually, and high-turnover funds in rising markets have historically distributed larger and more frequent gains — a record visible in each fund's distribution history, which is public. The documented asymmetry: index funds' minimal turnover defers gains inside the portfolio indefinitely (the buy-and-hold tax advantage quantified in the after-tax literature since the 1990s and popularized by the index-fund cost analyses of that era), while high-turnover funds convert paper gains into taxable events on a schedule the shareholder does not choose. Tax-managed variants exist precisely to blunt this, at some tracking cost to their strategies.

What doesn't turnover tell you?

It does not measure skill, in either direction. A 5 percent-turnover fund is not disciplined if its static book is simply stale; a 300 percent-turnover fund is not talented because it trades; the ratio is silent on whether the trades added value, which is a performance question requiring return attribution against a stated benchmark. It does not measure risk directly, though very low turnover in an active fund can indicate drift from the stated strategy — styles wander as positions age. And it does not capture trading quality — whether executions harvested or leaked spread — which requires transaction-level cost analysis, the kind institutional investors commission and funds increasingly report under execution-disclosure regimes.

How should a reader use the number?

As a screen for three checkable follow-ups. If turnover is high, check the fund's liquidity profile — what does it hold, and what do trades in those markets cost. If the account is taxable, check the distribution history — several years of realized-gain distributions are the direct observable consequence. If the fund is active and cheap-looking, check whether its stated net performance already embeds the trading costs — reported net returns do include them, which is why after-cost comparisons of net returns, not expense ratios alone, are the honest basis. Turnover is the beginning of those questions, not the answer to any of them.

The primary numbers sit in standardized fund documents — turnover in the fact sheet and prospectus, distributions in the tax forms funds issue annually — searchable through the Securities and Exchange Commission's EDGAR system, with investor-education material at investor.gov. Read the pace, price the pace, and decline any narrative that confuses motion with judgment.

The spread is the most visible number in markets — quoted on every screen, crossed in every trade — and the least often examined for what it contains. A one-cent spread and a thirty-cent spread are not the same object scaled; they usually have different compositions, which is why they respond differently to the same stress.

What are the three components?

Processing cost is the plumbing bill: exchange fees, connectivity, clearing and the marginal cost of quoting. It is the most stable component, tied to infrastructure rather than information, and it sets the floor under a spread in a perfectly calm, information-free market. Inventory risk is the cost of holding the position a quote implies: a market maker who buys from a seller now holds shares whose price can move before they are resold, and the spread charges for that exposure. Adverse-selection risk is the cost of trading against someone who knows more: if some counterparties trade on information, the market maker systematically buys before falls and sells before rises, and the spread widens to recover those expected losses.

The three components are distinguishable empirically. Inventory effects decay as the position is worked off, often within the day; adverse-selection costs concentrate in trades that are followed by price movement; processing costs are flat. Empirical decompositions in the microstructure literature estimate adverse selection as the dominant share of spreads in information-intensive names, with processing costs dominant in the tightest quotes.

Why do spreads widen so fast on news?

Because the components move at different speeds. Processing cost does not change when a headline crosses. Inventory risk rises with volatility — holding stock for a minute during a tariff announcement is not the same risk as holding it in a quiet session. Adverse-selection risk explodes: the probability that the next order comes from someone who has read the news and knows where the price is heading jumps discontinuously. Spreads quote the sum, so they widen within seconds, long before any human reassessment, and they recover as information becomes commonly held — the pattern documented trade-by-trade in event studies of announcement windows.

What did decimalization change?

Before 2001, U.S. stocks quoted in sixteenths — a minimum tick of 6.25 cents. Decimalization, completed in April 2001, cut the minimum to a penny and competition took quoted spreads on large names down to a cent or less, one of the cleanest and most studied market-structure changes on record. Average quoted spreads for liquid stocks fell by roughly a third or more in the academic estimates of the period. What decimalization did not do is eliminate the other components: inventory and adverse-selection costs reappear inside the quoted penny as depth thinned at each level — a trade documented in post-decimalization studies, where narrower spreads came with smaller displayed size. The total cost of trading did not fall one-for-one with the spread; it redistributed between price and quantity.

How do you measure the spread you actually pay?

Two measures matter and they differ. The quoted spread is the displayed difference between best bid and best ask at an instant. The effective spread is twice the signed difference between your execution price and the prevailing midpoint — it captures price improvement, where orders execute inside the quote, and is the standard measure in execution-quality reporting. Effective spreads are systematically smaller than quoted spreads in liquid names because of internalization and price improvement; the gap itself is a documented market-quality statistic, reported in SEC-mandated execution-quality disclosures by brokers and venues.

Where do spread claims mislead?

Three recurring slips. First, quoting the spread without the size: a one-cent spread on a hundred shares and a one-cent spread on ten thousand shares are different economics once depth enters. Second, comparing spreads across decimal and pre-decimal regimes without noting the tick-size change — the 2001 break is a structural discontinuity that ruins naive long-run spread charts. Third, treating the spread as a tax paid to middlemen in full: the processing component is a fee, but the adverse-selection component is a transfer from slower traders to faster ones, which is a different policy conversation — one the SEC's equity market structure filings at sec.gov document in detail.

Read the spread as a price for three services — standing ready, holding risk, and accepting informational disadvantage — priced continuously and visible to the penny. What a spread contains matters more than what it measures, because the contents predict how it will behave on the day the reader finally crosses it.

A 0.7-point drop in the 12-month rate in a single month is a large move by CPI standards, and large moves have mechanical explanations before they have narrative ones. The discipline is always the same: decompose first, interpret second.

What moved the 12-month rate?

Two standard mechanisms operate in a month like June 2026. Base effects: the 12-month calculation compares this June to last June, so an unusually large increase in the index twelve months ago falls out of the window — the 2025 comparison months carry the energy-surge period, and unfavorable comparisons rolling off can move the year-over-year rate without any new monthly change. Monthly mechanics: a 0.1 percent seasonally adjusted monthly print is near-flat, and a month that follows elevated readings with near-flat prices mechanically lowers the annualized trajectory. Energy's role cuts both ways in 2026: the year's FOMC statements described inflation elevated partly on global energy prices, and any reversal in the June energy components would feed directly into the headline arithmetic. The report's component tables — energy, food at 3.0 percent year over year, shelter with its documented lag structure — attribute the movement precisely; this summary points to them rather than paraphrasing them.

What does one month not establish?

Every question the genre immediately asks: whether a trend has turned, whether policy will respond, whether inflation is returning to target. A single monthly CPI print is one observation of a noisy series whose shelter component lags market rents by many months, whose energy swings are governed abroad as much as at home, and whose seasonal adjustment is re-estimated annually. The measured facts are the 3.5 percent 12-month rate, the 0.1 percent monthly change, and the components; trend claims require the sequence of prints, read with vintages attached.

How should readers track the sequence?

By the release calendar and the primary tables: the Bureau publishes the schedule in advance, the full component detail with each release, and its TED summaries — including the June piece reading "consumer prices up 3.5 percent over the year ended June 2026" — at bls.gov. The neighboring readings — 4.2 percent for May, 3.4 percent for July — frame June inside the year's path; the August release would show the July easing continuing. The number of the month is 3.5; its meaning belongs to the sequence.