What actually changes when a market never closes?
You already know how a market works: an open, a close, an auction that fixes an official price, a weekend where nothing happens and a Monday that absorbs it. Crypto deletes all of that. Not as a feature or a slogan — as a structural fact with consequences that reach into every number you will read for the rest of this course.
The clock, in hours
The NYSE trades 6.5 hours a day, roughly 252 days a year: about 1,638 hours of continuous price discovery. A crypto spot market trades 24 × 365 = 8,760 hours. That is 5.3 times as much clock, and every one of those hours is an hour in which your exposure exists, moves, and can be liquidated if it is leveraged.
Nothing in that sentence is a value judgement. It is a description of where the risk sits: in a market with an open and a close, the market structure holds part of your risk overnight. In a market with neither, all of it is yours, all of the time.
There is no official close, so "the daily close" is a convention
An equity closing price is struck — an auction with real orders producing one number everybody agrees on. Crypto has no auction and no closing bell, so a daily bar is manufactured by whoever built the dataset, by picking a cut-off. Most use 00:00 UTC. Not all do.
Watch what that does to identical tape. Suppose bitcoin sits at $60,000 at 00:00 UTC, falls to $58,000 by 04:00, and is back at $60,000 by 08:00.
- A provider cutting at 00:00 UTC records one day: open $60,000, close $60,000. Daily return 0.0%. The move is inside the bar and vanishes from a close-to-close series.
- A provider cutting at 05:00 UTC ends its day while the price is near $58,000, then closes the following day at $60,000. Its series shows −3.3% followed by +3.4%.
Same market, same trades, two different return series — and therefore two different volatility numbers, two different correlations against equities, two different backtests. Before you compare any crypto daily series to any other, you need to know its cut-off.
Weekends exist in the data, not in the market
Crypto prices on Saturday. Equities do not. So any joint calculation — a correlation, a beta, a hedge ratio — has to decide what to do with the two extra observations per week. Drop them, and you throw away real price changes. Keep them, and you are comparing a series with 365 observations a year to one with 252. Both choices are defensible; making the choice silently is not.
The same arithmetic reaches annualisation. Scaling a daily standard deviation to a year multiplies by the square root of the number of periods: √252 for equities, √365 for crypto. Use the wrong one and you are off by √(365/252) = 1.20 — a 20% error in a headline volatility figure, with no bug anywhere in the code.
The habit this unit installs
Every crypto number carries three questions before it means anything: which venue, which cut-off, which day count? That is the whole discipline, and it costs nothing but the asking.
In the data
A year of daily bars for bitcoin, followed by the convention that produced them, as the data provider describes its crypto "exchange":
Every day of the week, in UTC, closing at 23:59, with no holidays at all. So each daily bar is cut at UTC midnight by whoever built the dataset, not struck by a closing auction, and every volatility, correlation and backtest number built on it inherits that choice.
Try it now
- Read the trading days and the closing time in the table above, and say whose clock decided where one day ends and the next begins.
- The table below is the same asset by the hour, read at four moments: 00:00 and 05:00 UTC on 2 September 2026, and the same two times a day later. Rebuild two daily bars by hand, one cutting at 00:00 UTC and one at 05:00 UTC: each runs from the open at its cut to the open at the next day's cut. Write down both daily returns.
- Compare the two daily returns you just produced. If they differ, you have proved to yourself that "the daily close" is a decision somebody made, not a fact the market handed you.