Volatility clustering in crypto: does a big day mean another?
After a big Binance Bitcoin day, the next day was also big 15.84% of the time vs 9.05% normally. See 10 coins, each year, and what this cannot show.
Yes, big days tend to come in bunches. On Binance, a Bitcoin day that landed in its top 10% of move sizes was followed by another top 10% day 15.84% of the time (32 of 202 big days, September 2020 to October 2026), against 9.05% for all days. The direction did not carry over: a big move today said almost nothing about whether tomorrow would rise or fall.
This pattern has a name, volatility clustering. Volatility means how much price moves. "Clustering" means big moves land close together. These are past cases, not a forecast.
What is volatility clustering?
Picture a week of weather where stormy days come in a row. Knowing a storm came does not tell you which way the wind will blow. In our data, crypto moves looked like that: the size of moves bunched up, while the direction did not.
To measure it, we need two ideas.
- Absolute move. The size of a move with the plus or minus sign removed. A 3% rise and a 3% fall both count as 3%.
- Autocorrelation. A score from −1 to +1 that shows how much one day's number lines up with the next day's. Near 0 means almost no link. A positive score means a high number tends to be followed by another high number.
We also used a simple rule for a "big day". A day is big if its absolute move is larger than that of 90% of the days in the previous 365 days. In other words, it is in the top 10% compared with the last year.
Does a big day make another big day more likely?
Yes, for every coin we measured. The table covers 10 Binance perps, which are futures contracts that never expire, priced in USDT. USDT is a token designed to track the US dollar. We used daily candles, which run from midnight to midnight UTC. UTC is the shared world clock.
The first two number columns use all days. The last three use only big days, as defined above.
| Binance perp | Link between moves, with direction | Link between move sizes | Next day also big | Normal rate of big days | Big days counted |
|---|---|---|---|---|---|
| BTCUSDT | -0.07 | 0.17 | 15.84% | 9.05% | 202 |
| ETHUSDT | -0.07 | 0.18 | 14.83% | 9.71% | 209 |
| SOLUSDT | -0.04 | 0.29 | 18.12% | 7.42% | 138 |
| XRPUSDT | -0.03 | 0.27 | 24.74% | 8.99% | 190 |
| DOGEUSDT | 0.03 | 0.17 | 22.41% | 9.03% | 174 |
| ADAUSDT | -0.05 | 0.22 | 19.21% | 9.72% | 203 |
| AVAXUSDT | 0.02 | 0.19 | 18.59% | 8.42% | 156 |
| LINKUSDT | -0.08 | 0.19 | 15.96% | 8.94% | 188 |
| BNBUSDT | -0.07 | 0.34 | 23.12% | 8.95% | 186 |
| TRXUSDT | -0.14 | 0.32 | 27.01% | 8.27% | 174 |
Every coin has a higher "next day also big" share than its normal rate. TRX had the largest gap, at 27.01% against 8.27%. Bitcoin had one of the smallest.
The "normal rate" is about 9%, not 10%, because moves got smaller over time and the cut-off comes from the year before.
Coinbase spot markets, where you buy the coin itself with US dollars, gave the same shape. BTC-USD had 19.67% against 9.64% across 361 big days. ETH-USD had 15.20% against 9.58% across 329 big days.
Does a big move say which way tomorrow goes?
No. Look at the "link between moves, with direction" column. The scores sit between −0.14 and +0.03. That is very close to 0, which means one day's move told us almost nothing about the next day's direction.
The "link between move sizes" column is different. All 10 coins scored between 0.17 and 0.34, clearly above 0.
The same gap shows up on shorter bars. Here is Binance BTCUSDT on three bar sizes, with the same two scores:
| Bar size | Link between moves, with direction | Link between move sizes |
|---|---|---|
| 1 hour | -0.02 | 0.30 |
| 4 hours | -0.02 | 0.23 |
| 1 day | -0.07 | 0.17 |
The size link is strongest on hourly bars and weakest on daily bars. The direction link stays close to 0 on all three.
Did it happen in every year?
No. Here is Binance BTCUSDT by year. The first year is left out because we need 365 prior days to set the cut-off, so 2020 starts in September.
| Year | Big days counted | Next day also big | Normal rate that year |
|---|---|---|---|
| 2020 | 9 | 11.11% | 7.83% |
| 2021 | 47 | 17.02% | 12.88% |
| 2022 | 25 | 16.00% | 6.85% |
| 2023 | 26 | 26.92% | 7.12% |
| 2024 | 40 | 15.00% | 10.93% |
| 2025 | 27 | 7.41% | 7.40% |
| 2026 | 28 | 14.29% | 9.59% |
In six of the seven rows the next-day share was above the normal rate. In 2025 it was not: 7.41% against 7.40%, so no extra bunching that year.
Each row has few big days, so single years are noisy. In 2023, one more or one fewer repeat would move the share by about 4 points.
Do volatile and calm periods really look different?
Yes. Bitcoin's median absolute daily move was 2.35% in 2021, then 1.00% in 2023 and 1.17% in 2026 (through October 19). The median is the middle value after sorting the days from smallest to largest.
The share of days with a move over 5% shows how much the years differed:
- 2021: 20.27% of days
- 2023: 5.75% of days
- 2025: 3.84% of days
Over the whole history, 9.24% of BTCUSDT days moved more than 5%. For SOLUSDT it was 27.04%. The biggest daily fall was −39.98% on March 12, 2020.
Years like 2021 hold many big days, and years like 2025 hold few. For the rhythm inside a day, see which hours move crypto the most.
What does this not show?
- Direction. Move size is not up or down. A big day can be a fall or a rise.
- A forecast. It says nothing about how big the next move will be, or when a calm stretch ends.
- Independent tests. Big days cluster in a few crashes, so 202 big days is less evidence than it sounds. Our count has fewer truly separate periods than days.
- Cause. We did not test why moves bunch up.
- Trading results. We tested no trading rule, so there are no fees or profits here.
- Every coin. We chose large coins that still trade. A random set of coins could behave differently.
If you test a rule that reacts to "big" days, do not treat each day as separate. Judge "big" against the recent past for that coin, as we did, and expect bunches. That is a check on your own research, not a reason to trade. For what came after one kind of big move, see Bitcoin after big one-hour drops.
How we measured this
We used public daily and hourly price bars for 10 Binance USDT perps and Coinbase BTC-USD and ETH-USD, through October 19, 2026. For each daily candle we took the close-to-close percent change and removed its sign. A day was "big" if its size was above the 90th percentile of the previous 365 days. We then counted how often the next day was also big, and compared that with the share of all eligible days that were big.
For the link scores, we paired each bar with the next bar only when no bar was missing between them. Each market's first year is left out. These markets still trade, so the picture can change.
Try this in Comparand
To check clustering on your own dates, ask your connected AI agent: "Get completed daily bars for Binance BTCUSDT from my chosen start date. Work out each day's move size. Mark days above the 90th percentile of the previous 365 days. Tell me how often the next day is also marked, with the counts."
- venue_bars returns open, high, low and close prices plus volume for completed bars only, including 1-day bars. It is free on every key. Your agent does the counting.
For Hyperliquid, ask: "Using hl_stats, show realized volatility and the volatility regime for BTC, ETH and SOL on [one day], then for the next day."
- hl_stats returns realized volatility, meaning movement measured from past price changes, a volatility regime label and a correlation matrix, for up to 20 coins on one day. It is $0.01 per call on pay-as-you-go.
To look inside a busy day, ask: "Show Bitcoin's per-minute labels on Hyperliquid for my chosen start and end time."
- hl_labels returns per-minute labels for one coin over your window. Labels describe the past. It is $0.005 per call on pay-as-you-go.
These are Hyperliquid data, separate from the Binance numbers above. Plans cover data tools under fair-use limits. See the docs to connect, or try 3 coins free. Those are past cases, not a forecast.
The short version
- After a top 10% day, Binance Bitcoin's next day was also top 10% in 15.84% of cases, against 9.05% normally.
- All 10 Binance coins and both Coinbase markets showed the same bunching in move size.
- The link between one day's move and the next day's direction was close to 0.
- The effect was missing in 2025 and rests on few big days in each year.
- Bunching describes past move sizes. It does not give direction or timing.
Questions people ask
What is volatility clustering in crypto?
Volatility clustering means big moves tend to come in bunches. A big day is more likely to be followed by another big day than a normal day is. It describes the size of moves, not their direction.
How likely is another big day after a big crypto day?
On Binance BTCUSDT, a day in the top 10% of moves was followed by another top 10% day 15.84% of the time, against 9.05% for all eligible days. That is 202 big days from September 2020 to October 2026. It is a past count, not a forecast.
Does a big move today tell you which way crypto goes tomorrow?
Almost nothing. For Binance BTCUSDT daily moves, the link between one day's move and the next day's, with direction kept, scored -0.07 on a scale where 0 means no link. The link between move sizes scored 0.17.
Is volatility clustering stronger on hourly charts?
For Binance BTCUSDT, yes. The link between one bar's move size and the next bar's was 0.30 on 1-hour bars, 0.23 on 4-hour bars and 0.17 on daily bars. The link with direction kept stayed close to 0 on all three.
Did volatility clustering happen in every year?
No. For Binance BTCUSDT, 2025 showed no extra bunching: 7.41% against a 7.40% normal rate. In the other six yearly rows the next-day share was higher, but each row has only 9 to 47 big days.