Research · test #4
Does the bitcoin rainbow chart work? No edge
The rainbow chart looks almost magical: bitcoin's tops touch the red bands, its bottoms the blue ones. We wrote the rules down first, then asked a simple question: what if the bands were drawn only with the prices known at the time? Then the rainbow turned $1 into $22 over ten years. Buying and holding turned it into $194.
In short: buy in bands 1-3, sell in bands 7-9, checked once a month, January 2016 - September 2026. Bands fitted on all data (as on popular charts): ×398. Bands fitted only on data known at each date: ×22. Buy-and-hold: ×194. The rainbow's track record is hindsight.
The rules (fixed before the run)
- Data: Bitstamp BTC/USD daily closes, December 2012 - October 2026.
- Model: log10(price) = a + b · log10(days since 3 January 2009), nine bands by distance from the line in standard deviations (band 5 = ±0.25 sd; bands 1 and 9 beyond ±1.75 sd). The same model drives our live rainbow chart.
- Honest version: on the first day of each month the line and bands are refitted using only prices before that day.
- Rule: at each month-end, band 1-3 = hold BTC, band 7-9 = hold cash, band 4-6 = keep the position. Start in BTC on 1 January 2016. 0.2% cost per switch.
- Edge requires: more growth than buy-and-hold, more than 95% of random strategies with the same number of trades and time in the market, and a better Sharpe ratio than buy-and-hold.
Result
| $1 became | Sharpe | Worst drawdown | Trades | |
|---|---|---|---|---|
| Buy and hold | $194 | 1.02 | -75% | - |
| Rainbow, honest bands | $22 | 0.87 | -49% | 2 |
| Rainbow, hindsight bands | $398 | 1.24 | -49% | 4 |
| Random timing, same trades (median / top 5%) | $20 / $123 | top 5%: 1.05 | 2 |
The honest rainbow lands right at the median of random strategies with the same number of trades. It did cut the worst drawdown - by sitting in cash for five years.
What actually happened
| Honest bands | Hindsight bands |
|---|---|
| Sell May 2017 at $2,298 (band 8) | Sell Aug 2017 at $4,734 (band 7) |
| - stays in cash through the 2017 peak, the 2018 crash and the 2020-21 rally - | Buy Jan 2019 at $3,413 (band 3) |
| Sell Dec 2020 at $28,993 (band 7) | |
| Buy Jun 2022 at $19,925 (band 3) | Buy Sep 2022 at $19,425 (band 3) |
In spring 2017 the line fitted to 2013-2016 prices was much lower than today's line, so a $2,300 bitcoin already looked like a bubble. Every later rally pulled the line up - which is exactly why, looking back, the bands seem to fit the cycles so well.
Do lower bands lead to better returns?
A little. Average return over the next 12 months by honest band at month-end (overlapping windows, so no significance claimed): band 3 +91%, band 4 +166%, band 5 +239%, band 6 +163%, band 7 -19%, band 8 +44%, band 9 -29%. The upper bands were weaker - but the rule spent so long out of the market that it missed far more than it saved.
What this test does not say
- Other band widths or thresholds might have done better on this history - we did not search for them on purpose: picking the best after the fact is the same hindsight problem.
- One market path, two trades: the result rests heavily on 2017.
- Prices before December 2012 were not available from our source; popular charts start in 2010.
Other tests: research. Not investment advice.
FAQ
Does the bitcoin rainbow chart work?
Not as a trading rule. With bands fitted only on data known at the time, buy-blue, sell-red turned $1 into $22 from 2016 to 2026 against $194 for buy-and-hold.
Why does it look so accurate?
Because the line is fitted to all prices, including the tops and bottoms it appears to predict.