Yaga.sh · BTC research

Do Elliott waves hold up on Bitcoin? We're building the data to check.

A research project on Bitcoin's full price history. The minute data from three exchanges is collected and verified. The wave counting is not built.

Open the chart GitHub

Data until .... Research, not trading advice or signals.

Bitstamp BTC/USD since .... One candle per month, log scale. Open the chart for every bar.

The question

The claims are everywhere. The evidence is thin.

Batchelor & Ramyar, 2006

The Dow, 1914–2002

A few significant Fibonacci ratios appear, but no more than chance predicts given the number of tests.

Kempen, 2016

Dow, Nasdaq, DAX, 1974–2015

Across 17,931 retracements, no Fibonacci level stands out. Retracements follow a log-normal distribution.

Bitcoin

No rigorous study found

Figures like "73% of wave 2s retrace 50–61.8%" come from posts and blogs, with no chance baseline.

Prechter's reply is fair: Elliott claims ratios between specific wave pairs, so testing every swing dilutes the effect. Testing that properly takes clean data first.

The dataset

Three exchanges, stored separately.

...

one-minute rows

...

years of history

...

venues

Spot · BTC/USD

Bitstamp

First minute: ...
Last minute: ...
1m rows: ...
Missing minutes: ...
Last sync: ...

The longest history. Checked against the ff137 snapshot, its own candles and TradingView prices. Before 2013 it is mostly no-trade minutes.

Spot · BTCUSDT

Binance

First minute: ...
Last minute: ...
1m rows: ...
Missing minutes: ...
Last sync: ...

Every archive checksum verified. Checked against its own candles and TradingView prices. The missing minutes are real exchange outages.

Linear perp · BTCUSDT

Bybit

First minute: ...
Last minute: ...
1m rows: ...
Missing minutes: ...
Last sync: ...

Checked against its own candles. Prices and volumes match TradingView's BYBIT:BTCUSDT.P, and this is the chart's default.

Numbers as of the last export, ....

How it's kept

Boring rules, on purpose.

Closed minutes only

The forming minute is never stored. A bar counts as complete only when a successful fetch really covered it.

Nothing filled in

Raw rows are exactly what the exchange returned. Missing minutes are recorded, never invented.

Bars from real trades

Bars use traded minutes only. Flat no-trade minutes would invent opens and extremes.

No quiet edits

If a closed day changes later, its old file is kept and its revision goes up.

Where it stands

The data layer is built. Nothing else is.

Built

A collector for three venues, syncing every 15 minutes.
15m, 1H, 4H and 1D bars derived from the minutes.
A daily audit against the exchanges' data.

Not built

Wave counting.
Statistics on ratios and patterns.
Forecasts of any kind.

What gets built next is not decided.

The chart

Look at the data yourself.

Candlesticks for all three venues, on 1D, 4H, 1H and 15m, with no drawings on top.

Open the chart

Research, not trading advice or signals.