Bitcoin fractals: what they are and do they actually work

Bitcoin fractals are past stretches of the BTC chart that look like the current one, used to see what price did next. Do bitcoin fractals predict price? Not reliably: they give you statistics from a handful of similar setups, the sample is tiny, and different lookback windows often disagree. Where fractals help is in showing the range of possible outcomes, not in timing a buy or a sell. Our Fractals block runs this search every day; below is how to read it.

What traders mean by bitcoin fractals

"Fractal" covers three different ideas in trading, and they get mixed up all the time.

When someone on crypto Twitter says "the 2017 fractal is playing out," they almost always mean the second one, an analog. The Williams fractal is a charting indicator, and fractality is a research topic. None of these has anything to do with Fractal Bitcoin, which is a separate blockchain project.

The Williams fractal: five candles and a turn

The indicator comes from trader Bill Williams and his 1995 book Trading Chaos (TradingView help). The rule is mechanical (MetaTrader 5 docs):

1. Take at least five consecutive candles.
2. Up fractal: the middle candle has the highest high, and the two candles on each side have lower highs.
3. Down fractal: the middle candle has the lowest low, and the two candles on each side have higher lows.
4. Breakout: price moves above the last up fractal or below the last down fractal. Williams treated that as an entry signal.

The weakness is built into the definition. A fractal is only confirmed once two more candles close to its right. On a daily bitcoin chart that is a two-day lag; on the 4-hour chart, eight hours. TradingView's own description says the indicator lags by two or more candles, though it does not repaint. Used alone, Williams fractals throw off plenty of false breakouts, so traders usually pair them with a trend filter.

Fractal analogs: "history rhymes"

The second approach looks for past stretches that resemble the last few weeks or months. A trader overlays, say, 2019⁠–⁠2020 on 2023⁠–⁠2024 and draws a conclusion about where price goes next. The trouble with doing it by eye is that people find the resemblance they are looking for and quietly drop the matches that don't fit.

Our Fractals block does the same thing formally, with no cherry-picking (methodology):

1. Take the latest 4-hour BTC/USDT candles on Binance in three windows: 2 weeks, 50 days and 6 months.
2. Compare the shape (log price, normalised) with every stretch of history since August 2017 using correlation.
3. Allow the past stretch to have played out 0.5⁠–⁠2× faster or slower.
4. Keep the six best non-overlapping matches and measure how price moved over the same length of time after each one.

A correlation of 0.9 means two curves look very much alike. It does not mean the causes were alike: one stretch may have been driven by the Fed, another by the halving, a third by an exchange collapse.

What bitcoin fractals show as of Oct 19, 2026

In the Oct 19, 2026 run (latest candle closed at 8 PM ET on Oct 18, BTC at $82,116 on Binance), the three windows returned this (our data):

The best matches have correlations between 0.85 and 0.93, so the shapes really are close. Look at what that gives you:

• The windows disagree. The 2-week window leans slightly up, the 50-day window fell in five of six cases, and the 6-month window leans up again. Fractals alone cannot tell you which one to trust.
• Median and mean diverge. In the 50-day window the median is −8.6% but the mean is +3.7%, because a single 2020 match returned +79.5% and dragged the average up.
• Matches come from different eras. The top 2-week match is Aug 17⁠–⁠30, 2022, a bear market after the Terra and Celsius collapses. Next come January 2024, the first days of US spot ETF trading, October 2025 and spring 2018. Market conditions were different each time.

To be fair about the numbers: "4 of 6 went up" is a weak signal. If price after each match were a coin flip, you would see four or more rises out of six about 34% of the time.

What the research says

The idea that prices are self-similar came to finance from mathematician Benoit Mandelbrot, who coined the word "fractal." In 1963 he used cotton prices to show that large jumps happen far more often than a normal distribution allows, and proposed a different model (The Journal of Business, 1963).

Bariviera and co-authors (Physica A, 2017) estimated the Hurst exponent of bitcoin returns over 2011⁠–⁠2017. They found it shifted a lot in bitcoin's early years and then settled, and that behaviour looked similar across scales, a sign of a self-similar process. Statistical self-similarity, though, is not the same as a specific chart shape repeating.

The closest test of chart "patterns" is Lo, Mamaysky and Wang (Journal of Finance, 2000), who trained an algorithm to detect formations like head-and-shoulders in US stocks from 1962 to 1996. Their conclusion is cautious: some patterns carry incremental information about returns, which is not the same as a profitable strategy. And Sullivan, Timmermann and White (1999) showed why searching through many rules easily produces a strategy that only looks like it works, and offered a statistical correction for that kind of data snooping.

Limitations

• Small samples. Bitcoin has about nine years of reliable Binance history. Similar stretches number in the single digits, and conclusions from six cases are unstable.
• The market changes. In 2018 there were no spot ETFs, no deep options market and few companies holding BTC. The same chart shape inside a different market structure can mean something else entirely.
• Curve fitting. Tweak the window, the scale and the correlation threshold and you can find a fractal for any conclusion you want. That is why we show three fixed windows side by side even when they contradict each other.
• No causes. A fractal describes shape, not what produced it: demand, liquidations, macro.
• Lag. A Williams fractal confirms two candles after the extreme, by which point part of the move is gone.

How to use fractals sensibly

1. Read the range of outcomes, not just the median. It is a more honest picture of how uncertain the next few weeks are.
2. Compare windows. If all three point the same way, at least the picture is consistent; if they argue, the fractals are telling you nothing.
3. Check the backdrop: cycle indicators like MVRV and the Mayer multiple, funding and ETF flows in the daily brief.
4. Treat Williams fractals as reference levels, not a standalone entry system.

FAQ

Do fractals work for bitcoin? Not as a precise forecast. As a way to see how widely outcomes can vary after a similar-looking chart, yes.

What timeframe is best for Williams fractals? There isn't one. Lower timeframes give more signals and more false breakouts; the daily chart gives fewer, with a two-day lag.

Where can I see today's bitcoin fractals? In the Fractals block on the CoinLab homepage. It updates daily and shows each match, its dates and what happened next. More on the coin itself is on the bitcoin page.

Will the 2017 or 2021 cycle repeat? The shape may echo the past, but the participants, volumes and instruments are different. No fractal guarantees a cycle will repeat.

This is not investment advice.