Is there a best day of the week to buy Bitcoin?

Each weekday against the week around it

For every day from Jul 23, 2015 to Oct 8, 2026 we compare that day's close with the average close of the seven days centred on it, then average by weekday. Positive means the day closed below its week: cheaper.

How to read it. One day's price typically sits 1.31% above or below its own week, so a single week says nothing and it takes 585 of them to pin an average down even this far. The margin of error is two standard errors: if chance alone were at work, an average would land inside it about 19 times in 20. None of the seven averages is larger than its own margin of error.

The last column needs no statistics. It splits the same days into 10 calendar years, 2016 to 2025, which share no days with each other. A weekday with no edge comes out ahead in about half of them.

A check that assumes nothing. Inside every run of seven days we dealt the seven figures to the weekdays at random, 2,000 times over. That removes any real weekday effect and leaves everything else as it was. The best-to-worst gap came out at least as large as the real 0.24% in 71% of those deals. At least 1 weekday came out as one-sided across the calendar years as in the real table in 55% of them.

What it is in money. The whole gap, Thursday against Wednesday, is 0.24% of a purchase: $0.24 on a $100 buy, about $13 over a year of weekly buys. This table describes 585 past weeks. It is not a forecast of the next one.

The seven plans, full price history

$100 per week, 842 purchases each ($84,200 invested), no fees, valued at the Oct 11, 2026 close of $83,678. The last purchases land in the week ending Oct 6, 2026.

Read this table carefully before quoting it. About 64% of the Bitcoin in every row came from that row's first 10 purchases, made between Aug 18, 2010 and Oct 26, 2010 at prices from $0.06 to $0.18. That is $1,000 of the $84,200 invested. Our price series for those weeks is rounded to whole cents. Had each of those prices been one cent higher, a row would hold 8.26% less Bitcoin; one cent lower, 11.26% more. The whole table spans 6.22%. Two early prices in our series look wrong: Sep 15, 2010 at $0.09 and Sep 16, 2010 at $0.15, between days at $0.06. Coin Metrics, a second source, shows no such jump, and the methodology page has both sets of figures. With both priced at $0.06, Wednesday would finish first instead of Friday and the table would span 8.57%.

The plans do not start together. Only one weekday falls on any given date, so seven weekly plans begin on seven days in a row, and every purchase of the plan that starts first lands six days ahead of the matching purchase of the plan that starts last. While the price is climbing, earlier is cheaper. Start this whole comparison one day later and Wednesday moves from 2nd to 7th. Run the same seven plans from Jul 20, 2015, where the Coinbase closes in our series begin, and they finish in exactly the order they started in: Monday, Tuesday, Wednesday, Thursday, Friday, Saturday and Sunday, 1.36% apart from first to last.

Before Jul 20, 2015 the weekday labels are a day out. For those years our prices are blockchain.info's market price, an average across exchanges stamped at 00:00 UTC, which tracks the previous day's close. A purchase labelled Monday in those years was priced at about Sunday's close, and 99.89% of the Bitcoin in this table was bought in those years. That is why the first table uses Coinbase closes only. The value column here is arithmetic on a 16-year schedule. It is published for completeness, not as a target.

How these numbers were computed

Everything uses one price per UTC day. From Jul 20, 2015 that is Coinbase's daily candle close, the last trade of the UTC day. Before it, it is blockchain.info's market price, an average across exchanges stamped at 00:00 UTC. Over the 4,101 days where both exist, that figure sits a median 0.05% from Coinbase's close of the day before and 1.5% from its close of the same day (measured Oct 12, 2026), so it is in effect the previous day's close. All dates are UTC. An automatic buy that fires at another time of day is priced differently from the close used here.

Week-by-week test. Coinbase closes only. For each day, the close is compared with the geometric mean of the seven closes centred on it: three before, the day itself, three after. The figure is 100 times the natural log of that mean over the day's close, which for gaps this small is the percentage the day closed below its week. A plain dollar spread over the week is not used, because a fixed amount spread over seven moving prices buys slightly more than the same amount on the middle day whatever the weekday, which would put every weekday a few hundredths of a percent below zero. Centring the week on the day keeps a rising price out of the answer. Against a fixed Monday-to-Sunday week, later days would look dearer simply for being later. The figure shown is the mean over every Monday, every Tuesday and so on, and the margin of error is two standard errors of that mean. A calendar year is counted only when all of its days were measured. The random deals shuffle the seven figures within each consecutive run of seven measured days, 2,000 times, from a fixed starting seed so the same prices always give the same shares.

Seven plans. For each weekday we find the first UTC date on or after the start of the data that falls on it and run a weekly schedule from there: $100 per purchase, zero fees, using the same engine as the main calculator. The number of purchases is capped at the lowest of the seven, so all seven invest exactly the same total. Every row is valued at the same last daily close, so ranking by value and ranking by BTC are the same thing. Figures refresh once every 24 hours; the data runs from Aug 18, 2010 to Oct 11, 2026. Price sources, fee math and UTC handling are on the methodology page.