When is the market actually awake?

§ 1Data and method

Every 1-minute bar of every crypto perpetual with unbroken coverage is converted to IST wall-clock time, bucketed into 15-minute bins (96 per day) and aggregated to (symbol, IST date, bucket). IST is UTC+05:30 with no daylight saving, so the time-of-day map is a fixed offset of UTC and directly comparable across seasons — which is the entire reason to do this study on the Indian clock rather than on UTC.

From that panel, five independent facets are measured:

• Liquidity — share of the day's quote volume and trade count.
• Movement — mean absolute 1-minute return, mean 1-minute range, total path length.
• Breadth — share of the universe whose 15-minute move exceeds 12 bp, the lab's own round trip (4 bp taker + 2 bp slippage per side). This is “probability a random symbol gives you a tradable move here”.
• Deadness — share of 1-minute bars that move less than 1 bp.
• Thinness — price movement per unit of relative volume, computed within each symbol then medianed. Above 100 means a thin book: more price impact per unit of flow.

The composite activity index is the geometric mean of the five liquidity/movement components, each normalised to its own cross-bucket mean, so 100 is the average 15-minute bucket of the day.

§ 2The shape of the IST day

Figure 1. The IST day. Thin lines are raw 15-minute buckets; heavy lines are a 1-hour centred mean. Shaded red bands are windows whose smoothed composite index falls in the bottom quartile; green bands the top quartile. Volume peaks exactly where the US cash session opens; the deepest trough sits over the Asian afternoon.

§ 3Three regimes, not one curve

Read it as three overlapping populations:

• US (18:30–01:00 IST). 6.5 hours, ~39% of volume. Real directional flow, the deepest books of the day (thinness 82–88 through 22:30, the lowest anywhere), highest breadth.
• Asia (05:30–08:30 IST). A genuine second pulse — Tokyo/HK open — but shallower: 0.89× volume with thinness ~109. The 06:00 IST hour beats every European hour on movement.
• Europe + Asia afternoon (09:00–18:30 IST). Nine and a half hours that never exceed 0.90× volume and mostly sit at 0.74–0.85×. Inside it, 09:00–12:00 is where volume and the composite index bottom together, and 15:00–17:00 runs it close on movement and breadth.

Everything else sits at roughly average. The lull is not a single block you can switch off — it is three separate ones, and none of them is empty.

§ 4The hour the eye does not see: :00 and :30

Figure 2. Each row is an IST hour, normalised to that hour's own average minute, so the within-hour structure is visible rather than the time-of-day level. Bright columns at :00 and :30 run the full height of the chart. The right panel is the whole-day quarter-hour profile.

Volume at :30 is 1.6× volume at :29. In 20 of the 24 IST hours the single busiest minute is :30; in the US-open hours it is :00 — at 19:00 IST the :00 minute carries 3.08× the average minute, and 18:00 IST carries 2.83×. The spikes are one minute wide and visible in every hour of the day.

This is execution and settlement plumbing — 30-minute kline closes, funding marks (05:30, 13:30 and 21:30 IST, all on :30), TWAP/VWAP benchmark prints — not information. The 05:30 IST mark is the clearest: 1.56× the average minute against 0.81× at 05:00, a 1.9× step at the daily candle and funding settlement.

Practically: a market order at :29 pays a thinner book than the same order at :31, and a maker quote parked at :30 gets filled by the burst rather than by flow. Anyone backtesting on 1-minute bars with “volume at signal time” as a liquidity proxy is measuring this artifact.

§ 5Weekends

Figure 3. Left: each weekday scaled to its own average hour, so the rows show shape rather than level. Right: share of weekly volume and mean absolute 1-minute return. Saturday and Sunday keep volume but flatten the curve.

Saturday and Sunday keep about two-thirds of weekday volume and four-fifths of the movement, but they redistribute it: the 19:00–21:00 IST window is 0.66–0.87× its weekday weight, while 00:00–03:00 IST is busier on weekends than on weekdays. The weekend is not a quieter copy of the weekday — it is a flatter curve with the US peak blunted. Volume concentration is lowest on Saturday and Sunday.

§ 6Stability, and the daylight-saving tell

Figure 4. Left: mean absolute 1-minute return by IST clock, 2025 vs 2026 over the common Jan–Sep window. The shape is the same; only its level differs. Right: busiest/quietest bucket volume ratio by quarter — concentration never falls below 3.7×.

Per symbol: the median correlation of an individual symbol's hourly volume profile with the universe is 0.966, and the median symbol's busiest hour carries 2.1× its quietest hour's volume. 34 of 49 symbols peak at 20:00 IST, 9 at 19:00, 3 at 21:00. The pattern is not an average concealing disagreement — it is nearly universal.

The DST tell

The 09:30 ET equity open is 14:30 UTC under EST (20:00 IST) and 13:30 UTC under EDT (19:00 IST). Since IST is fixed, any crypto intraday study that pins a single fixed hour is blending two different events across the year. A “US-open” filter defined in UTC catches the cash open for only part of the year.

§ 7The quiet window is quieter, not dead

The deadness column is the honest one. In the 09:00–12:00 trough, 18.3% of 1-minute bars move less than 1 bp, versus 13.3% during the US peak — a five-point difference on a ~16% base. The absolute floor is high:

• Quietest 15-minute bucket of the day (05:15 IST): mean |15m move| 26 bp, breadth 62%.
• The 12 bp round-trip hurdle is cleared by 68% of symbol-windows in the worst hour and 78% in the best. There is no hour where a typical symbol fails to move enough to pay costs.
• Tail frequency (>1% in 15 minutes, per 1,000 symbol-windows): 37 at 16:00 IST, 41 at 09:00–11:00, 109 at 20:00, 95 at 19:00. Tails do concentrate in the US session — by 2.6× — but the quiet hours still carry meaningful jump risk.

And the single largest 15-minute move of the entire sample happened in the quietest part of the clock: 2025-10-10 21:19 UTC = 02:49 IST on 11 October, where BTC fell 5.1% in one minute and alts (MINA, OGN, CTSI, TIA…) lost 45–60% within 15 minutes. That single 02:00–04:00 IST window supplies 14.8% of the 2,000 largest 15-minute moves in the whole sample — and every one of those is the same cascade. An IST “avoid the dead hours” filter would have sat out the calm days while leaving you fully exposed to that one.

§ 8A control: instruments that do close

Figure 5. The identical pipeline on Binance's tokenised equities and metals (AAPL, NVDA, TSLA, SPY, QQQ, XAU; Jun–Sep 2026). A single IST hour carries 12.7% of the day's volume and the curve collapses outside US hours — the method can see a real session structure when one exists.

Tokenised equities put 12.7% of daily volume into a single IST hour and collapse to index 45 outside US hours — a 6.4× swing, with 60–68% of minutes completely flat. Crypto perps swing 1.7× in activity and never exceed 18.5% flat minutes at hourly resolution. The crypto lull is real but an order of magnitude shallower than an actual market close.

Figure 6. BTC + ETH alone sit between the two: activity index 76 at 09:00–11:00 versus 168 at 20:00, a 2.2× swing, with 24% flat minutes at the trough against 13% at the peak. The diurnal effect is strongest exactly where liquidity is deepest.

§ 9What this implies for the lab

• Backtest windows matter. An IST 09:00–17:00 research window covers 27% of volume and misses every hour where breadth exceeds 75%. A 24/7 backtest averages a 2.5× liquidity difference into a single number.
• Session gating should use IST, and should not be a hard on/off. 19:00–22:00 IST holds 22% of volume; the other 21 hours hold the other 78%, and 68–70% of symbol-windows there still clear the cost hurdle. Deleting them deletes most of the sample, not most of the noise.
• Slippage is not constant. Thinness is 13–23% higher at 02:00–04:30 IST and 17% lower during the US peak than the daily average. A single 2 bp assumption is optimistic in the Asian afternoon and pessimistic at 20:00 IST.
• The quarter-hour spike is a fill-model hazard. Fills at the bar close after a signal land in a 1.26–1.36× volume burst at :00/:30 — good for market orders, bad for makers.
• Re-cut trend and mean-reversion results by session. Path efficiency (|net move| / total path) is flat at 0.234–0.266 all day, which suggests the shape of the path does not change much — only its scale. That is a testable claim for the existing strategies, and the natural next step.

§ 10Caveats

• Futures only. Spot may differ, especially in the Asian session where regional spot flow is larger.
• Volume is a proxy for depth. Binance bulk archives carry no book data; the thinness column is a movement-per-volume proxy, not a measured spread.
• Two years is one macro cycle. The sample runs from a 2025 rally (BTC ~95k → ~119k) through the 10 October 2025 cascade and a 2026 drawdown to ~60k, then a recovery to ~86k. Whether the Asian trough deepens or shallows in a different regime is not identifiable from 21 months.
• Survivorship. The universe is “symbols with continuous coverage”, which excludes everything delisted during the window. The effect on a time-of-day profile should be second-order, but it is not zero.
• Cost hurdle is a lab default (4 bp taker + 2 bp slippage per side), not a measured execution cost.

§ 11Reproducibility

# Primary: 49 crypto perps, 2025-01-01 -> 2026-09-30, 15-minute IST buckets
python scripts/intraday_activity_study.py

# Majors only (BTC, ETH)
python scripts/intraday_activity_study.py --symbols BTCUSDT,ETHUSDT \
    --prefix ist_activity_majors

# Control: tokenised equities / metals
python scripts/intraday_activity_study.py \
    --symbols AAPLUSDT,NVDAUSDT,TSLAUSDT,SPYUSDT,QQQUSDT,XAUUSDT \
    --start 2026-06-01 --end 2026-10-01 --prefix ist_activity_tradfi

# Any bucket size, e.g. 5-minute resolution
python scripts/intraday_activity_study.py --bucket 5 --prefix ist_activity_5m

Each run prints the profile, session table, dead zones, weekday split, stability diagnostics and per-symbol agreement, writes reports/ist_activity_*_profile.csv and reports/ist_activity_*_sessions.csv, and renders the figures.