Glossary

Definitions for the terms used across this site. Everything here is measured with public U.S. government data; see what is an event study? for how the tests work.

Economic releases

CPI (Consumer Price Index) — the U.S. government's main measure of consumer inflation, published monthly by the Bureau of Labor Statistics. The headline year-over-year figure is what most people mean by "the inflation rate." Markets watch it closely as a driver of Fed policy.

Core CPI — CPI excluding food and energy prices, which are volatile. Often seen as a cleaner read on the underlying inflation trend.

PPI (Producer Price Index) — inflation measured at the wholesale level: prices received by producers, one step up the supply chain from CPI. Published monthly by the Bureau of Labor Statistics.

PCE (Personal Consumption Expenditures price index) — an alternative inflation gauge published by the Bureau of Economic Analysis. It is the measure the Federal Reserve's 2% target is officially defined against, though it arrives about two weeks after CPI.

NFP / Jobs report (Employment Situation) — the monthly U.S. labor-market report from the Bureau of Labor Statistics. "Nonfarm payrolls" (NFP) is its headline line: the net change in jobs. Widely considered the single most important scheduled release.

Jobless claims — the weekly count of new applications for unemployment insurance, from the Department of Labor. Timely and high-frequency, but noisy.

Retail sales — the monthly measure of consumer spending at retailers, from the Census Bureau. The main read on the U.S. consumer.

GDP (Gross Domestic Product) — the broadest measure of U.S. economic output, published quarterly by the Bureau of Economic Analysis. The "advance" estimate is the first and most market-relevant release; it describes a quarter that has already ended.

FOMC (Federal Open Market Committee) — the Federal Reserve committee that sets the U.S. policy interest rate. It announces decisions eight times a year, each followed by a press conference.

Markets

Treasury yield — the interest rate on U.S. government debt. Yields move opposite to bond prices. Because Treasuries are considered risk-free, their yields anchor borrowing costs across the economy.

The yield curve / maturities (3M, 2Y, 5Y, 10Y, 30Y) — Treasuries are issued at different maturities, from 3 months to 30 years. Plotting their yields gives the "yield curve." The short end (3-month to 2-year) is driven mainly by current and expected Fed policy; the long end (10- and 30-year) reflects longer-run growth and inflation expectations. The belly is the stretch in between, roughly the 2- to 5-year, which is where most of the reactions on this site show up.

2s10s spread — the 10-year yield minus the 2-year yield, a widely watched gauge of the curve's shape.

Broad U.S. dollar index (DTWEXBGS) — the Federal Reserve's trade-weighted measure of the dollar's value against a basket of foreign currencies. We use this rather than proprietary indices because it is public domain.

WTI and Brent crude — the two main benchmark crude oils: West Texas Intermediate for the U.S. market and Brent for the international market. We use the daily spot prices as published by the U.S. Energy Information Administration.

Statistics and method

Basis point (bp) — one hundredth of a percentage point. A move from 4.00% to 4.08% is 8 basis points. Yields are quoted this way because their moves are small.

Event study — the method behind every result here: gather all instances of an event, measure the market's move around each, and compare to normal. See what is an event study?.

Baseline — what a "normal" move looks like: the distribution of market moves across all trading days, not just event days. Every result is measured against it.

Direction — the average signed move around an event. If it is near zero, the event does not move the market in a predictable direction.

Volatility (size) — the average absolute move around an event, compared to the baseline. A ratio above 1.0 means the market moves more than usual; below 1.0 means it is quieter than usual.

Ratio (×) — how large the event-day move is relative to a normal day. "×2.35" — the jobs report against the 2-year Treasury, the largest robust reading on the site — means about 2.35 times a normal day.

p-value — roughly, the probability of seeing a result this strong if the event actually had no effect. We use p below 0.05 as the threshold for calling an effect detectable; results sitting near the line are reported as not established rather than rounded up.

Significance — a result is "significant" when it is unlikely to be a fluke (low p-value). Note that statistical significance is about reliability, not size or importance.

Multiple-testing correction (Benjamini-Hochberg / FDR) — an adjustment applied when testing many event-and-market combinations at once, because some will look significant by chance. A result is called robust only if it survives this correction. See the full grid.

Surprise (model-based) — on this site, how far a reading came in from the average of its own previous three readings. It is not a market consensus forecast: those are proprietary products we do not license. Because our proxy treats part of a predictable move as if it were news, it understates any real relationship rather than inventing one. Always read it with that caveat.

Larger than (percentile) — where a single reaction ranks against every other release of the same kind since 2000. "Larger than 95%" means only about one release in twenty produced a bigger move; "larger than 14%" means it was one of the quietest. Note that this ranks the raw move against every release since 2000, while the "vs normal" ratio compares it to a normal day in its own year — so a small ratio can sit next to a middling percentile when the year itself was volatile.

Robust vs nominal — a result is nominally significant if its own p-value clears 0.05, and robust only if it also survives the multiple-testing correction across the whole grid. We publish both labels, because the gap between them is where most false findings live.

The verdict on each card says which one it is:

• Significant (green) — clears its own p-value and survives correction across the grid. Across every release we test, only the jobs report earns this on volatility.
• Nominal only (amber) — clears its own p-value but does not survive correction. CPI, PPI and the FOMC decision sit here: the release-day move is real-looking on its own, but once you account for having tested every release against every market, it is the kind of result chance produces. We show it rather than hiding it, and we do not call it a finding.
• None (red) — does not clear its own bar. An honest null, and with a large sample that is a real result rather than a shrug.

These labels are read straight from the test output at build time, not written by hand, so a card cannot drift away from the grid it summarises.

Look-ahead protected — a test that never uses information unavailable at the time of the event. Our windows start from the prior day's close, before the release is public.

Null result — a finding that an event does not have a detectable effect. With a large enough sample (like our 150 jobless-claims releases), a null is an informative result, not just an absence of one.

Related

• What is an event study? — the method in full.
• How we test — and the full results grid — every term above, applied.
• About Macro or Noise — what this site is for.

Historical statistics for informational purposes only, not financial advice.