Reading Numbers Like an Analyst · Lesson 6 of 6
Percentiles and base rates: where today's value sits in its own history
One concept: percentiles
Why it matters
A headline calls the gap between two Treasury yields “the widest in years” or “unheard of”. Kitalpha’s 2s10s spread record holds that gap: the 10-year yield minus the 2-year yield, today 0.33 pp (33 bp) as of 2026-09-15, with a daily history that starts on 2002-01-02. Is today’s level rare, or ordinary? The figure on its own cannot say. This lesson takes one idea, the percentile, and uses it to place a value inside its own history, so that “unprecedented” becomes a claim you can check against the record rather than a feeling.
The concept
A percentile is a position in a sorted list. Take every daily level a series has recorded, sort them from lowest to highest, and ask what share of them sits below today’s level. That share, written as a number from 0 to 100, is today’s percentile. A percentile of 50 says half the recorded days sat below today’s level; a percentile of 90 says nine days in ten sat below it and one in ten above. When some days sat exactly at today’s level, each such tie is counted as half a day, so a level that matches many past days is pushed to neither end of the scale.
Two things about that definition matter most. First, a percentile is a rank, not a size. It counts days, and it says nothing about how many basis points separate today’s level from any other. Two spreads can have the same percentile while one sits far from its median in basis points and the other close to it, because what decides the percentile is how many days fell on each side, not how far away they were. Second, a percentile is a percentage of days, not a percentage of a value. The all-history figure in this lesson is never “today’s spread as a fraction of its record high”; it is “the share of recorded days that sat below today’s spread”. Keeping those two apart is the single most common slip when reading one.
A base rate is the plain-language cousin of the same idea: how often something has occurred in the record. Asking for the base rate of a level like today’s is asking what share of the history sat above or below it, which is exactly what the percentile answers. A level near the 50th percentile, the median, is as common as a level gets; a level at the 99th has been matched or exceeded on about one recorded day in a hundred; only a percentile at the very top or bottom of the scale means no recorded day sat on the far side of it. “Unprecedented” is therefore a claim about the percentile, and it is a demanding one.
The window is part of the claim. The same level can be ranked against every recorded day or against only the most recent five years, and the two percentiles need not agree, because the levels that were common recently may differ from the levels that were common over decades. Neither window is the right one; each answers a different question, and an honest reading names which set of days it counted. Kitalpha’s spread pages carry the full daily history as a chart and a table; the percentiles in this lesson are worked out from that history at the moment the page is built, and the day count and start date are printed beside them so the window is never hidden.
The histogram below sorts the 2s10s spread’s last 1,500 trading days of levels into bins a quarter of a percentage point wide, covering 2020-09-15 to 2026-09-15. Look for the median line: half of those days sit in the bars to its left. A level’s percentile is how far along the bars it sits, counted from the left edge, and a level with tall bars on both sides of it has plenty of company.
| Bin from | to | Count |
|---|---|---|
| -1.250 | -1.000 | 5 |
| -1.000 | -0.750 | 79 |
| -0.750 | -0.500 | 119 |
| -0.500 | -0.250 | 256 |
| -0.250 | 0.000 | 82 |
| 0.000 | 0.250 | 134 |
| 0.250 | 0.500 | 229 |
| 0.500 | 0.750 | 307 |
| 0.750 | 1.000 | 80 |
| 1.000 | 1.250 | 120 |
| 1.250 | 1.500 | 69 |
| 1.500 | 1.750 | 20 |
Worked example
Take the 2s10s spread record as of 2026-09-15. Its history runs from 2002-01-02 and holds 6,180 trading days; the lowest level recorded is -1.08 pp, the highest 2.91 pp, and the median 1.05 pp. The steps read today’s level, rank it against the whole history and against the last five years, and turn the rank back into a count of days.
Record: 2s10s Spread (10-Year minus 2-Year) · as of · Computed by Kitalpha from U.S. Department of the Treasury par yields
- Today's 2s10s spread, from the record 0.33 pp The 10-year yield minus the 2-year yield, in percentage points.
- Trading days in the record's history 6,180
- All-history percentile of today's level (0 to 100) 27.5 The share of those days on which the spread sat below today's level, ties counted half.
- About how many of those days sat below today's level 1,700 Day count multiplied by the percentile, divided by 100.
- Share of the history above today's level, in percent 72.5
- Five-year percentile (the most recent 1,260 trading days) 59.5 The same level ranked inside a shorter window.
- Today's level minus the all-history median, in basis points -72 bp The median is the level half the recorded days sat below; the sign says which side of it today sits.
Read the steps in order. The percentile in step three is a share of the day count in step two, so multiplying the two and dividing by 100 gives about how many recorded days sat below today’s level, and 100 minus the percentile gives the share that sat above it. Those two figures answer “has this been seen before”: a level with recorded days on both sides of it has. The five-year percentile ranks the same level against a shorter window and can differ from the all-history figure; the difference is information about the window, not an error. The last step measures distance from the median in basis points, a separate figure the percentile does not contain.
Faded example
Now the second record: the 3m10y spread, the 10-year yield minus the 3-month yield, as of 2026-09-15, with a history of its own that starts on 2002-01-02. Today’s level, the day count and the all-history percentile are given below. Complete the last step: turn the percentile into a count of days.
Second record: 3m10y Spread (10-Year minus 3-Month) · as of
- Today's 3m10y spread, from the record0.89 pp
- Trading days in the record's history6,177
- All-history percentile of today's level (0 to 100)36.4
- days Tolerance ±10 days
Reveal the answer and the explanation
2,248 — The percentile is a share of the history: the fraction of recorded days, ties counted half, on which the spread sat below today's level. Multiplying that share by the number of days turns the rank back into a count of actual trading days. Because the percentile is printed to one decimal place, the count is approximate, which is why the tolerance is ten days. A level with recorded days below it and recorded days above it has company in the record, whatever a headline calls it.
Stored on this device only; not graded.
Retrieval check
Mark your confidence before each answer. Every option carries an explanation; read the ones you rejected too.
-
1. The 2s10s record's all-history percentile places today's level in its own history. The figure describes:
Choose your confidence first. -
2. Using the 2s10s record, what share of its recorded trading days had a spread above today's level? Enter 100 minus the all-history percentile, to one decimal place (percent of days; tolerance ±0.1).
Source record: 2s10s Spread (10-Year minus 2-Year) (as of 2026-09-15)
Tolerance ±0.1 %Choose your confidence first. -
3. Suppose a spread record covers 2,000 trading days and today's level has an all-history percentile of 97.0. Which reading is correct?
Choose your confidence first. -
4. The worked example reads two percentiles for the same 2s10s level: one over the whole record and one over the most recent 1,260 trading days. The two figures can differ because:
Choose your confidence first.