The S&P 500, the Nasdaq and the Dow, day by day since 2013, with bitcoin on the same axis of time, and how far each moved over the same 60 days.
7,663 the S&P 500 on Oct-22 2026, the last day in this build, measured and published by askthetape.com.
«Is bitcoin a risk asset?» is usually answered with an opinion. Here are the numbers instead. The S&P 500 closed at 7,663 on 2026-10-22, the Nasdaq Composite at 26,009, the Dow at 52,868, and one bitcoin was worth 10.61 units of the S&P. Over the 4,087 sessions since 2010-07-19, the 60-day change in bitcoin against the 60-day change in the S&P has a correlation of +0.08 — positive, so they lean the same way, and far from the +1.0 that the phrase «bitcoin IS a risk asset» would need. Nasdaq +0.068, Dow +0.094: the three indices say the same thing, which is what you would expect from three indices. 🩸 On LEVELS the same pair reads +0.928, and that number means almost nothing: two things that both rose over twelve years correlate by construction. We print it here so nobody reads the picture as one. 🩸 The stock exchange is shut at weekends and on US holidays. Those days are NOT drawn and nothing is carried forward — 1,849 of our 5,936 bitcoin days have no index close. No forecast is drawn here. 🩸 Three more views answer in numbers what this page used to leave to opinion: how often the S&P closes the same way several sessions running, and what followed. Five sessions up in a row happened 260 times since 1970, counted on the close that made it five. Twenty sessions later the index stood higher 169 times of 260 — it does that 61 times of 100 from any session of this tape. Five sessions down in a row happened 172 times; twenty sessions later, higher 115 times of 172, against the same 61 of 100. Bitcoin over those same UTC days: a median of +0.41 % on the 2,227 sessions the S&P closed up, -0.14 % on the 1,859 it closed down. Every count here is of SESSIONS — a run is a run of closes, not of calendar days — and the session that is still running is in none of them. Medians, not a law.
| leg | public market-data daily closes, one pull a day, same fallback as the ^TNX and Apple legs: ^GSPC 14,324 rows from 1970-01-02 · ^IXIC 14,046 rows from 1971-02-05 · ^DJI 8,764 rows from 1992-01-02. Bitcoin is Bitstamp's daily close, pulled by us. The session stamp is 13:30Z (9:30 in New York), so its UTC date is the trading date. |
|---|---|
| cadence | daily on trading days; about 5 days in 7, freshness allowance 5 days — the widest gap between two prints over the last two years is 4 days (measured on this pull) |
| since | 2010-07-19 — the first session both legs exist (our bitcoin leg starts 2010-07-17; ^GSPC reaches back to 1970-01-02 and is not drawn before bitcoin begins) |
| 🩸 the date rule | the paper leg is NOT carried forward: a day the exchange was shut has no point here, so a weekend cannot move a ratio. 1,849 of our 5,936 bitcoin days since 2010-07-17 have no paper close and are not drawn. The neighbouring page us-10y-yield-vs-btc uses the OTHER rule — it carries the yield up to 7 days so the bitcoin day survives — and that is why a ratio lives here and not there. |
| 🩸 where it is thinner | an index is a price of a BASKET whose members change, and the levels here are not adjusted for dividends — ^GSPC and ^DJI are price indices, so a long-run comparison against bitcoin understates what owning the basket actually paid. The ratio view is therefore a curiosity, not a portfolio: it says how many index units one bitcoin buys, nothing about total return. |
| 🩸 the closes are hours apart | We pair the 4pm New York close with midnight UTC; in summer the stock close is 20:00 UTC. On a day when the tape moved after the New York close, the index point and the bitcoin point of the same date are not the same moment. |
| 🩸 what these correlations are not | 60-day changes: S&P +0.08, Nasdaq +0.068, Dow +0.094 on 4,087 common sessions. Levels: +0.93, +0.93, +0.94. A correlation over a window is not a mechanism and not a rule. We keep the correlation views separate from the historical run-event scores below. |
| policy | no backfill · read it here · export: whole history |
| hourly window | our hourly capture — the hourly closes of the FUNDS that track the indices (SPY, QQQ, DIA), not the index the day bars use; the bitcoin leg and the ratio are ours · since 2025-09-19 13:00Z by data, 1,924 hours with data inside a 9,583-hour span → 2026-10-23 19:00Z (the newest 400 days of an hourly record that starts 2023-06-27; the Tape and Desk windows draw these 400 days) · bitcoin against them by the hour starts 2026-03-31 · captured 2026-10-23T21:21:56Z |
| 🩸 runs · what is counted | every closed session of ^GSPC from 1970-01-05 (the first session with a previous close on this pull; the carrier holds 14,324 rows from 1970-01-02) to 2026-10-22: 14,323 sessions, 7,584 up (52.9 %), 6,710 down, 29 exactly level. A run is a maximal block of consecutive SESSIONS that closed the same way; a level session closed neither way, so it ends a run and joins none. 🩸 Every count of runs on this page is a count of runs that have ENDED — the run standing open right now has no final length (a run of three can still become five), so it is named and not counted. Up-runs 3,515 (median 2, longest 14, ended Apr-15 1971) · down-runs 3,513 (median 1, longest 11, ended Oct-27 1971). Of 5 or more: 260 up and 172 down. |
| 🩸 what a level session does to the count | that rule is a choice and it moves the numbers, so here is the other one measured on the SAME pull: if a session that closed exactly level were allowed both to continue a run and to count as one of its sessions, this tape would show 267 up-runs and 173 down-runs of 5 or more instead of 260 and 172. We print the stricter count, because the view says «sessions up in a row» and a level session did not close up. Either way it is 29 sessions out of 14,323. |
| 🩸 the ladder | 4+: 534 up / 376 down · 5+: 260 up / 172 down · 6+: 140 up / 74 down · 7+: 70 up / 27 down · 8+: 35 up / 13 down · 9+: 17 up / 6 down · 10+: 8 up / 2 down — ended runs only. The threshold for «what followed» is 5 because that is where the sample is still a sample: at 8 or more there are 35 up-runs in 56 years, and a median of that is a story about a handful of months. |
| 🩸 what followed a run · where the clock starts | Measured from the close that made it 5 in a row — the moment a reader can actually know it happened: the entry is the close of the session on which the run REACHED 5, not the close of the session on which it broke. A run that breaks is only knowable AFTER the direction has already changed, so scoring from it lets the future pick the events: on this tape the first of the next sessions was then a session against the run 259 times of 260 after up-runs and 172 of 172 after down-runs — by construction, not by the market. One run is ONE event, taken once on the close that reached 5, however far it runs afterwards. Forward over 5 and 20 SESSIONS (not calendar days). An event counts only once its horizon has PASSED; one that has not is drawn hollow and counted nowhere (0 events at 20 sessions and 0 events at 5 in this build). The run standing open right now is 1 session down and has not reached 5, so it is no event yet. After 5+ up: n = 260 events, next 5 sessions median +0.49 % / mean +0.28 % / higher 162 of 260; next 20 median +1.32 % / mean +0.87 % / higher 169 of 260 (Aug-20 1970 → May-27 2026). After 5+ down: n = 172 events, next 5 median +0.66 % / mean +0.60 % / higher 102 of 172; next 20 median +1.74 % / mean +1.33 % / higher 115 of 172 (Jan-29 1970 → Jun-26 2026). 🩸 The base rate of the same tape, from ANY of its 14,323 sessions: next 5 median +0.30 % and higher 56.5 %, next 20 median +1.07 % and higher 61.4 % — every number above is to be read against it. Every median and share is computed from the raw closes at full precision and rounded once, when printed. |
| 🩸 how independent these events are, and how they sit across the tape | n counts RUNS, not independent observations: 20-session windows overlap when two runs land close together. Keeping only events whose 20-session windows do not overlap leaves 203 of 260 up-events (median +1.45 %) and 146 of 172 down-events (median +1.74 %). Across three stretches of the same tape, same entry and same horizon — 1970-1991: 5+ up n = 104 median +0.95 % higher 54.8 %, 5+ down n = 95 median +0.95 % higher 61.1 % · 1992-2009: 5+ up n = 75 median +1.57 % higher 69.3 %, 5+ down n = 39 median +1.84 % higher 74.4 % · 2010-2026: 5+ up n = 81 median +1.48 % higher 74.1 %, 5+ down n = 38 median +2.76 % higher 73.7 %. No conclusion is drawn from the split; it is here so that one number for 56 years is not read as one number for every regime. |
| 🩸 bitcoin on green and red sessions | 🩸 the two closes are not the same window: the S&P session is 09:30-16:00 New York (13:30-20:00 UTC in summer, about six and a half hours), while bitcoin's «day» here runs midnight to midnight UTC — the SAME UTC DAY, the day that CONTAINS the session, plus the hours before it opened and after it shut. And because the exchange is shut at weekends and on holidays, 1,847 of the 5,934 daily bitcoin moves on this tape (31.1 %) take part in no count on this page at all. Only the 4,087 sessions where BOTH tapes have a close (2010-07-19 → 2026-10-22): bitcoin's own daily close against its own previous daily close, the index's close against its own previous SESSION close. Nothing is carried across a weekend in either leg, which is why a Monday here is a Monday and not a Friday-to-Monday. Whole tape — green n = 2,227, median +0.41 %, mean +0.79 %, bitcoin higher 1,264 of 2,227; red n = 1,859, median -0.14 %, mean -0.02 %, higher 873 of 1,859. Last 252 common sessions (2025-10-22 → 2026-10-22) — green n = 133, median +0.36 %, mean +0.78 %; red n = 119, median -1.01 %, mean -1.09 %. Median AND mean are both printed because here they disagree: the mean of a bitcoin session is pulled by a handful of very large ones, and printing only one of the two would be a claim about the shape that the shape does not support. |
| 🩸 why these are not the numbers we ourselves posted | a chart we published on 2026-09-20 measured bitcoin from the PREVIOUS SESSION's close, so a Monday carried the whole weekend with it. Both rules, measured on this build: over the last 252 common sessions a red session cost bitcoin -1.19 % on the carrying rule against -1.09 % on this page's rule; over the last 60, -0.51 % against -0.48 % on red sessions and +1.58 % against +1.41 % on green ones. Same tape, different question — «what bitcoin did on that session» is not «what bitcoin did since the last session» — and the page keeps the rule it states everywhere else: nothing is carried. |
| 🩸 why only the S&P in these three views | the same carrier holds the Nasdaq (14,046 rows from 1971-02-05) and the Dow (8,764 rows from 1992-01-02), and the same arithmetic would run on both. They are left out on purpose: the Dow's tape here starts in 1992, so «since 1970» would mean three different windows under one caption, and a count of runs is only comparable inside one window. |
| 🩸 where these three views are thinner | a run is a property of CLOSES, and a close is one price at one moment: two sessions that both ended +0.01 % up count exactly like two that ended +2 % up, and nothing here weighs the size of a move. The index is a price index and is not adjusted for dividends, so a long run of small up sessions is not a statement about what owning the basket paid. And the running session is missing by construction — it has not closed, so it is in no run, no count and no chart. |
| 🩸 green for both, then a week · what is counted · [22.09 VIEWS1] | a common day is one where the index has a session close and bitcoin has both that day's close and the previous calendar day's: 4,087 of them, 2010-07-19 → 2026-10-22. Green for both means the index closed above its own previous SESSION and bitcoin above its own previous UTC DAY — the entry is that close, which is the moment a reader can know it happened, and nothing is carried across a weekend. Green for both: 1,264 days (2 episodes still inside the 7-day window, drawn hollow and counted nowhere); bitcoin over the next 7 days median +0.8381 %, mean +2.9825 %, higher 701 of 1262. Red for both: 981 days; median +0.3670 %, mean +1.1148 %, higher 510 of 979. THE BASE of the same tape and horizon, from ANY of its 4,082 common days: median +0.7419 %, higher 54.97 %. The placebo: 2,000 random draws of the same size put the green-for-both median at two-sided p = 0.670 and the red-for-both one at p = 0.064, and because the 7 days windows overlap those p-values are read as a placebo, not as a test — non-overlapping events leave 436 of 1262 green days (median +0.87 %) and 397 of 979 red ones (median +0.45 %). Every median is computed from the raw closes at full precision and rounded once, when printed. |
| this build | data to Oct-22 2026 |
| policy | no backfill · read it here · export: whole history |
The chart itself is drawn in the browser: turn JavaScript on to hover a day, pick a window and draw your line. The numbers above come from the last build, Oct-23 2026 21:24Z.
What the tape says today · what's new · all 111 metrics