Impact tiers: how Kitalpha computes them

Why it matters

Open the newest released consumer price index row on the calendar and, beside the agency and the reference period, a pill reads high (the row capitalises the word) with a score of 2.05 basis points. The latest GDP row carries a pill of its own. A first-time reader tends to take these as somebody’s judgement of how much a release matters, or as a warning about the next one. Neither is what the field records. The methodology page publishes the arithmetic behind every pill, and this lesson works through it on the CPI row.

The concept

An impact tier is a label, high, medium or low, attached to an indicator on the calendar. On Kitalpha it is the output of a fixed calculation on stored records, and nothing else goes into it: no editor scores the release, no survey ranks it, and no reader’s circumstances are consulted. The calculation has four parts, and every part is published with its numbers.

The first part is the measure. On each of the indicator’s past release days, take the closing 10-year yield from the U.S. Treasury’s par yield curve, the yield the government pays to borrow for ten years, struck once each afternoon, and subtract the previous trading day’s close. Keep the size of the change and drop its sign, so a move up and a move down of the same size count the same. A basis point is one hundredth of a percentage point. This absolute move is a record of what happened on the day, with no view about why.

The second part is the baseline. Yields move a little on every day, release or none, so the release-day average is compared with the same average over quiet days: trading days on which none of the tracked indicators released. The average absolute move on the indicator’s release days minus the average on quiet days is the excess. An excess near zero says release days looked like any other day; a large excess says they did not.

Two further rules apply. A day on which two tracked indicators released is excluded from both, because a closing yield cannot say which release the move belonged to; the days that remain are the clean sample. And the same excess is computed a second time on the 2s10s spread, the 10-year yield minus the 2-year yield, which is the curve’s shape rather than its level, because a rate decision moves short yields far more than long ones. The published score is the larger of the two excesses, and a score-from field records which one it was.

The third part is the window. The score is computed over the trailing five years, the regime a reader is standing in, since the market’s habit of reacting to a release changes over decades. Where five years hold fewer clean days than a published minimum, the method falls back to the full history and the release’s page says so.

The fourth part is the tier itself: the score bucketed against two fixed cutoffs in basis points, published on the methodology page and frozen. Below the minimum clean sample no tier is published at all; the row shows no pill, which is a different statement from low.

Read this way, a tier answers one narrow question: over the stated window, how much more did the Treasury curve move on this indicator’s release days than on a quiet day? It is backward-looking, it describes the bond market rather than any reader, and two indicators’ scores sit on one yardstick and can be compared directly. The bars below place three published scores side by side; look for the lengths against one another and against zero, remembering that each bar is an excess over a quiet day, not a raw move.

Worked example

Take the CPI row released 2026-09-11, whose pill reads high, beside the methodology record and the GDP row. The steps below read the quiet-day baseline from the methodology, take it off CPI’s own average move to reproduce the 10-year excess, set the published score beside that excess, put GDP’s score next to CPI’s, and finish with the size of CPI’s clean sample. Every figure is a record field or a subtraction of two such fields.

What to read off the steps. The third line is the concept: the excess is the release-day average with an ordinary day taken out, which is why a score can be small even though yields move every day. The fourth line agrees with the third to within a hundredth of a basis point when the 10-year excess was the larger: each published input is rounded to two decimals, and the pill was cut when the tiers were last recomputed. The score-from field holds level, which the row prints as “from the 10-year level” or “from the 2s10s curve”; the second means the spread’s excess set the score. The sixth line compares two indicators the only way the method allows: excess moves in basis points on one measure. The last line is the count behind the pill: below the published minimum there would be no pill at all.

Faded example

Now a third record: the producer price index row released 2026-09-10, whose pill reads low. Its published score and CPI’s are given below. Complete the comparison the worked example made for GDP, this time for PPI, then reveal the answer and compare.

Retrieval check

Mark your confidence before each answer. Every option carries an explanation; read the ones you rejected too.

Your summary