Search for the best day to buy bitcoin and you'll find confident answers: Monday, Thursday, the weekend. We tested them on Binance's price history from August 2017 to September 2026: a weekly buy on each day of the week, and a daily buy at each hour of the day.
- No weekday is reliably cheaper. Over 475 complete weeks, buying every Thursday (the cheapest day) cost 0.60% less on average than buying every Saturday (the dearest). Shuffle the days at random and a gap that big turns up 81% of the time.
- The "cheapest day" changes every year: Monday in three years, Tuesday, Thursday and Friday in two each, and Sunday in one.
- The hour matters even less. The cheapest and dearest hour of the day differ by 0.22% for bitcoin and 0.23% for ether, about two standard 0.1% trading fees.
- A counting slip makes Thursday a clear winner. Count every day in the data, partial weeks included, and Thursday "wins" by 0.63% with a 1.27% spread. Count the same weeks for every day and the effect halves, into noise.
Back when I was trading, I generally avoided buying from Friday to Monday, because that's when most of the big moves seemed to happen. This test looks at price, not at how wild each day is, but on price that habit made no real difference. One thing I did notice: the gap between the cheapest and dearest day was much wider in 2017 (3.6%) than in each of the last three years (about 1%). My guess is that the market was smaller then, so it took less to move the price, though 2017 here is only that year's last 19 weeks.
| Weekly buy on | Mon | Tue | Wed | Thu | Fri | Sat | Sun |
|---|---|---|---|---|---|---|---|
| Average cost vs buying every day, Aug 2017 to Sep 2026 | -0.21% | -0.14% | +0.22% | -0.35% | +0.06% | +0.25% | +0.18% |
| Years it was the cheapest day (of 10) | 3 | 2 | 0 | 2 | 2 | 0 | 1 |
Negative means cheaper. Permutation test over 10,000 shuffles: p = 0.81, no evidence that any weekday is cheaper.
The cheapest day changes every year
Here is the same test year by year. Each cell is how much a weekly buy on that day cost compared with that year's average.
Monday was the cheapest day in 2017, 2020 and 2023, and the dearest in 2018, 2022 and 2026. Sunday was the dearest day in 2017, 2021, 2023 and 2024, and the cheapest in 2022. In any single year the gap between the best and the worst day looks big: 3.6% in 2017 (Monday -1.95%, Sunday +1.66%), 1.8% in 2023. But a gap that size within one year says nothing about the next one.
Over the whole period, Thursday comes out cheapest at -0.35% and Saturday dearest at +0.25%. A 0.60% spread sounds like something. On $100 a week for nine years, $47,500 in all, it's about $285 more bitcoin, and only if you'd known in 2017 which day would win.
Is 0.60% real, or chance?
If the day of the week didn't matter at all, how big a gap would you expect anyway? To find out, we shuffled the seven prices inside each week, so every day of the week got random prices from its own week, and measured the gap between the best and the worst day. We did that 10,000 times. 8,132 of the shuffles produced a gap at least as large as the real 0.60%: p = 0.81. Seven numbers that bounce around will always have a best and a worst; the spread between them is what noise looks like.
The counting slip that makes Thursday a winner
Our first count included every day in the data. That gave Thursday as the clear winner, 0.63% cheaper than average, with a 1.27% spread between the best and worst day. Same prices, twice the effect.
| Count | Cheapest day | Its edge | Best vs worst |
|---|---|---|---|
| Every day in the data | Thursday | -0.63% | 1.27% |
| Complete weeks only (same weeks for every day) | Thursday | -0.35% | 0.60% |
The reason is where the data starts and stops. Binance's archive begins on Thursday 17 August 2017 and our download ends on Wednesday 30 September 2026. So Thursday to Sunday each get one extra buy in August 2017, at $4,100 to $4,300, and Monday to Wednesday one extra buy in September 2026, at about $83,600.
That shouldn't matter much, but average cost doesn't work like an average price. A fixed $100 at $4,100 buys twenty times as much bitcoin as $100 at $83,600, so a single cheap buy pulls the average cost of a weekday down far more than an expensive one pushes it up. Four extra cheap buys made Thursday to Sunday look better; three expensive ones made Monday to Wednesday look worse. Counting the same 475 complete weeks for every day removes the edge the calendar gave them.
We can't see how other "best day to buy bitcoin" studies were done, but any count that starts and ends on different weekdays, or uses a period that started cheap, can find a winner this way. It's the same effect that makes regular buying work at all: see why your average cost beats the average price.
The hour of the day: under 0.15% either way
The same question at the hour level: buy $1 of bitcoin every day at the same hour (UTC), for 3,302 complete days.
| Bitcoin | Ether | |
|---|---|---|
| Days tested | 3,302 | 3,240 |
| Cheapest hour over all years | 10:00 UTC (-0.08%) | 10:00 UTC (-0.10%) |
| Dearest hour over all years | 22:00 UTC (+0.14%) | 22:00 UTC (+0.13%) |
| Best vs worst hour | 0.22% | 0.23% |
| Chance of a gap this size at random | p = 0.060 | p = 0.512 |
The cheapest hour moves around: nine different hours were the cheapest in ten years of bitcoin data. The whole spread is about the size of two 0.1% trading fees, so a cheaper exchange or order type is worth more than the right hour.
One pattern did show up in both coins: the dearest hour sits late in the UTC evening. For bitcoin it was the 22:00, 23:00 or 00:00 UTC hour in 8 of 10 years; for ether, one of the 21:00 to 00:00 hours in 9 of 10. We noticed it in bitcoin after looking, so we wrote down a test before running ether. Ether repeated the location, but not the strength: at p = 0.51, a gap that size turned up in about half of the random shuffles, and bitcoin's p = 0.06 misses the usual 0.05 bar too. It's a faint lean, not a rule.
The effect also shrank. Bitcoin's best-vs-worst hour gap was 1.00% in 2017 and 0.10% in both 2025 and 2026. (2017 is only August to December, and it is the year with the biggest weekday and hour gaps.)
Where TabChart fits
TabChart is a free Windows app for live crypto charts, on Binance and six other exchanges, with no account. To look at the hourly candles behind this test, open BTCUSDT on Binance in TabChart and switch to the 1-hour timeframe. If you buy on a schedule, a time-interval alert (every 7 days, say) sends you the current price by desktop popup, email or Telegram while the app is running.
This is a backtest of what happened, not a forecast or advice. Past prices don't tell you future ones.
How we tested
| When | Computed on 8 October 2026, on prices up to the close of 30 September 2026 |
| Prices | Binance's public data archive, BTCUSDT and ETHUSDT, daily and 1-hour candles; each file checked against Binance's published SHA-256 checksum. Two months of ETHUSDT 1-hour files (July and August 2024) are missing from the archive, so the ether hour test leaves those days out |
| Period | 17 August 2017 (the archive's first day) to 30 September 2026; complete weeks from 21 August 2017; ether hours without July and August 2024 |
| Weekday test | $1 at the close of every Monday, Tuesday… in each ISO year (Monday-to-Sunday weeks); average cost = dollars spent / bitcoin bought. Complete weeks only (475), so every weekday covers the same weeks. "Monday" is the daily close at 00:00 UTC on Tuesday |
| Hour test | $1 every day at the close of the hourly candle that opens at H:00 UTC; complete days only (all 24 hours present) |
| Is it real? | Permutation tests: the prices shuffled within each week (10,000 times) or within each day (2,000 times); p = the share of shuffles with a best-vs-worst gap at least as large as the real one |
| Ether check | The late-evening pattern was noticed in bitcoin first, so ether was run after writing down what it should show (dearest hour within 21:00-00:59 UTC, p < 0.10). The location held, the p-value didn't |
| Not included | Trading fees (the same for every day and hour), slippage, other exchanges |
Download the data: weekday_by_year.csv (every year, every weekday) · hour_by_year.csv and hour_by_year_ethusdt.csv · the CSV behind each chart : t3-hour.csv, t3-weekday-by-year.csv.
FAQ
What is the best day of the week to buy bitcoin? None reliably. From August 2017 to September 2026, Thursday was 0.35% cheaper than average and Saturday 0.25% dearer, a gap that random shuffles matched or beat 81% of the time. The cheapest day changed from year to year.
Is bitcoin cheaper on weekends? Not reliably. Over the whole period Saturday and Sunday were slightly dearer (+0.25% and +0.18%), but in 2018, 2022, 2025 and 2026 they were among the three cheapest days.
What is the best time of day to buy crypto? It barely matters: the best and worst UTC hour differ by 0.22% for bitcoin and 0.23% for ether. Late evening UTC was slightly dearer in both, but not by more than chance would give.
Does it matter which day I set my recurring buy on? Not by more than a fraction of a percent. Whether you buy at all, and what you buy, matters far more: see DCA vs lump sum in crypto.
Why do some studies say Monday (or Thursday) is the best day? One year's data always has a winner, and counting the first and last partial weeks can double the edge. Over 475 complete weeks, no day stood out.
