Weekends and time of day: when does bitcoin move?

Two often-repeated claims about bitcoin's calendar: that it does something different on weekends, and that its gains happen outside US trading hours. Both can be checked directly against the candles, and both can be traded — so this study does both: statistics from 1h and 1d candles first, then backtests that hold BTC or ETH only in one window.

The short answer: weekends differ in volatility, not in return. The time of day shows a real-looking pattern before costs — mostly in the two hours after the US close — and switching in and out every day costs far more than any window earned.

Data and method

• Candles: Binance spot BTCUSDT and ETHUSDT. Hourly candles 2020-01-01 → 2026-09-30 (59,129 hours each — a few dozen hours are missing from the exchange history); daily candles over the same period, and again from 2017-08-17, the start of Binance spot history, for a longer sample.
• Returns: each candle's own move, open to close, in UTC. "Weekend" is Saturday and Sunday UTC. Gross — no fees.
• Sessions: US 13:00–21:00 UTC, fixed all year (the platform's time predicates are UTC and do not follow daylight saving time); Asia 00:00–08:00 UTC; Europe 08:00–13:00 UTC; and 21:00–24:00 UTC. A second, DST-aware version of the US session is the NYSE cash session rounded to whole hours: 13:00–20:00 UTC in US summer time, 14:00–21:00 UTC in winter.
• Backtests: five strategies, each holding 99% of the account in one window and nothing outside it, on BTC and ETH, 2020-01-01 → 2026-10-01, 10,000 USDT, fills at the next open, 0.1% fee and 0.05% slippage on every buy and sell. The three session strategies use 1h candles: a TIME condition decides on the close of the hour before the session and fills at its first open, and flatAtHourUtc closes at the session's last close. The weekend and weekday strategies use daily candles and DAYOFWEEK conditions. Ten runs in all; nothing was tuned.

Weekends: calmer, not richer

The clearest difference is size: BTC's weekend days were about 40% calmer than weekdays (a standard deviation ratio of 0.61; on ETH 0.74). On returns the evidence points both ways — a lower mean on weekends but a higher median — and the gap in means is well inside the noise (Welch t = -1.08). ETH had no gap at all: +0.225% a weekend day against +0.210% a weekday (t = 0.09).

The longer sample from 2017 does not change the picture: BTC +0.107% on weekends and +0.170% on weekdays, ETH +0.274% and +0.125%, neither difference significant. Year by year the sign flips: BTC's weekends were behind in 2020 and 2024, level in 2021–2023, ahead in 2025 and behind again in 2026 so far.

By single weekday, Thursday was the weakest day on both coins in both samples (BTC 2020–2026: -0.171% a day against +0.445% on Mondays and +0.458% on Wednesdays). The largest t-statistic of the 28 weekday tests was -2.52 (ETH Thursdays since 2017) — and with seven days tested at once, a single day needs |t| above about 2.7 before chance stops being the easy explanation. BTC and ETH also move together, so seeing it on both is not a second, independent confirmation.

This fits the best weekday to DCA: over 351 weeks the weekday you buy on moved a DCA result by 2–3 percentage points.

Time of day: the hours after the US close

"Compounded" is the gross return of holding only in those hours for the whole period, hour after hour; "share of the gain" is that window's part of the total log return of buy & hold (it adds up to 100% across US + outside, and across Asia + Europe + US + 21–24).

Read with the fixed 13:00–21:00 window, the claim barely holds: the US session had a third of the hours and 27.8% of BTC's gain — and 37.2% of ETH's, slightly more than its share. Read with the DST-aware window, it holds strongly: 5.7% and 11.5% of the gain in 29% of the hours. The difference between the two readings is essentially one hour, 20:00–21:00 UTC, which in summer falls after the US close.

That is where the pattern sits: in the hours right after the US close. 21:00–22:00 UTC alone compounded to +173.3% on BTC and +231.5% on ETH; 22:00–23:00 to +108.4% and +148.9%. The Asian session lost money on both coins. The 21:00 hour is also the only one whose average return stands out from zero at about the bar that testing 24 hours at once requires (t = 3.13 on BTC, 3.03 on ETH) — a pattern worth knowing, not a law.

Year by year the US session was erratic: on BTC it gave +181.7% in 2020 and lost in 2021, 2022, 2024 and 2025, while the hours outside it were ahead in four of the seven calendar years (2026 to the end of September).

Trading it: costs take everything

A daily round trip costs about 0.3% here (0.1% fee and 0.05% slippage, in and out). The average gross move of a session was far smaller: +0.048% a day for BTC's US session, +0.100% for the 16 hours outside it, +0.004% for Asia. Over 2,464 round trips the arithmetic is merciless.

Backtest

Hold only 13:00–21:00 UTC (US session), 1h — article study

BTCUSDT · 1h · Spot · 2020-01-01 → 2026-10-01

Return

-99.87%

Max drawdown

99.88%

Sharpe

-2.41

Win rate

40.1%

Trades

2,464

Open the report →

Backtest

Hold only 21:00–13:00 UTC (outside US session), 1h — article study

BTCUSDT · 1h · Spot · 2020-01-01 → 2026-10-01

Return

-99.62%

Max drawdown

99.68%

Sharpe

-1.45

Win rate

42.7%

Trades

2,464

Open the report →

Weekends are only 352 round trips, so the result survives — but holding the weekend alone lost money on BTC, and holding only weekdays kept a fraction of buy & hold:

Backtest

Hold only on weekends (Sat–Sun UTC), 1d — article study

BTCUSDT · 1d · Spot · 2020-01-01 → 2026-10-01

Return

-53.78%

Max drawdown

58.13%

Sharpe

-0.43

Win rate

46.9%

Trades

352

Open the report →

Backtest

Hold only on weekdays (Mon–Fri UTC), 1d — article study

BTCUSDT · 1d · Spot · 2020-01-01 → 2026-10-01

Return

+204.58%

Max drawdown

78.69%

Sharpe

0.58

Win rate

49.7%

Trades

352

Open the report →

Buy & hold over the same period, as the runs report it: +1,063.56% for BTC and +1,983.46% for ETH on 1h candles, +1,078.75% and +1,969.60% on daily candles.

The same, without costs

For scale only, the same strategies re-run as unsaved dry runs with the fee and slippage set to zero — a cost-free illustration, not something an account could have earned:

Even with free trading, no window beat simply holding: each one is a slice of buy & hold, and the slices multiply back to roughly the whole.

Conclusion

• Weekends: bitcoin moved less on weekends — about 40% smaller days — but did not reliably earn more or less.
• Time of day: with hourly data, the hours right after the US close (21:00–23:00 UTC) carried an outsized share of BTC's and ETH's gains since 2020, and the Asian morning lost money. Whether "the US session" looks bad depends on one boundary hour.
• Trading it: holding only in a session meant a round trip a day, and at 0.1% fees plus 0.05% slippage every session strategy lost more than 99%. Holding only on weekdays ended at +204.58% on BTC, against +1,078.75% for buy & hold.

More research: the best weekday to DCA and DCA or lump sum. How to read the metrics on the cards: backtest metrics explained.