If you trade with Ichimoku, you know the rule: you buy when the fast line (Tenkan) crosses the slow one (Kijun) from below. Ideally, though, only once price is also above the cloud, because the cloud is supposed to confirm the signal. The whole thing is almost always set to 9, 26 and 52.
Two assumptions sit behind that habit: the cloud makes the signals better, and the numbers are proven. Both can be checked on crypto.
The test ran on daily candles from January 2018 to 13 October 2026, on bitcoin, ether and today's ten largest coins. Neither assumption holds.
Ichimoku reads the market from highs and lows, not from closing prices
Ichimoku Kinkō Hyō means roughly "equilibrium chart at a glance". It was devised by Goichi Hosoda, a journalist at the Tokyo newspaper Miyako Shimbun. A first version appeared there in 1935 under a different name. He only published the method as a book in August 1969, under the pen name Ichimoku Sanjin, and by 1981 it had grown to seven volumes.
The chart shows five lines:
| Line | How it is built |
|---|---|
| Tenkan (conversion line) | midpoint of the high and low of the last 9 candles |
| Kijun (base line) | midpoint of the high and low of the last 26 candles |
| Senkou Span A | midpoint of Tenkan and Kijun, shifted 26 candles forward |
| Senkou Span B | midpoint of the high and low of the last 52 candles, shifted 26 candles forward |
| Chikou | today's close, shifted 26 candles back |
The area between Span A and Span B is the cloud, kumo in Japanese.
Ichimoku differs from most indicators in three ways. It works with the midpoint between the highest high and the lowest low, not with the average of closing prices as a moving average does. That is why its lines stay flat as long as no new high or low arrives, and jump as soon as one does.
It also draws the cloud into the future. For the next 26 candles, a zone is already on the chart. That is not a forecast but the past, shifted forward.
And Ichimoku means to be a whole system rather than a single signal: trend, support and resistance are meant to be visible at a glance. Hosoda added a theory of time, a theory of waves and rules for price targets. In the West, mostly just the chart is used.
Ichimoku was meant as a whole system, but mostly two parts of it are traded: the cross of Tenkan and Kijun, and the cloud.
Those two parts are what the test checked.
The cross of the two lines beats holding bitcoin
First, what speaks for Ichimoku. With 9/26/52 and no cloud, the strategy made 31.2 percent a year on bitcoin, 29.0 after fees. Simply holding bitcoin made 23.1 percent. The worst decline was 54 instead of 81 percent.
On today's ten largest coins, the strategy beat holding on seven. It was in the market only a little over half the time, 53 percent of days on bitcoin.
The Tenkan-Kijun cross caught trends on large coins and sat out part of the crashes.
Is that down to 9, 26 and 52, or would other numbers have done the same?
On crypto, 9, 26 and 52 are one pair of numbers among many
Only the first two numbers matter for the buy signal: 9 days for Tenkan, 26 for Kijun. Against them, 30 randomly drawn combinations ran on bitcoin, each between 5 and 80 days, drawn before the first comparison run.
| Bitcoin, January 2018 to October 2026 | Annual return after fees |
|---|---|
| 9/26 (tradition) | 29.0% |
| Median of the 30 random pairs | 28.3% |
| Random pairs better than tradition | 14 of 30 |
| Range of the random pairs, before fees | 8.2 to 43.5% |
| Holding bitcoin | 23.1% |
Binance BTCUSDT, daily candles, Ichimoku TK cross without cloud, platform standard costs.
The tradition sits in the middle of the field, at the 53rd percentile. Ether shows the same picture: 5 of 12 random pairs did better, and the tradition sat at the 58th percentile. Comparing only pairs with a similar number of trades changes nothing on bitcoin (55th percentile).
Where the numbers come from is less clear than it often sounds. In the West, the usual story is that 26 was a trading month in Tokyo's six-day week at the time. There is no evidence for that from the original. Japanese accounts call 9, 17, 26 and further values the basic numbers of Hosoda's time theory, with no clear derivation.
The numbers the six-day explanation would suggest for a seven-day market are 10/30/60. On the ten large coins, they came out ahead on six, behind on three, and level once.
Whether 9/26/52 come from the six-day week or not, on crypto they are one pair of numbers among many, not a tested optimum.
That leaves the cloud itself. It is the part that sets Ichimoku apart from any other line cross.
As a buy condition, the cloud costs return on all ten coins
With the cloud switched on, the strategy only buys while price is above the cloud, and also sells as soon as price falls below its lower edge. Everything else stays the same: the same coins, the same days, 9/26/52.
| Coin | without cloud | with cloud |
|---|---|---|
| BTC | 31.2% | 9.9% |
| ETH | 41.7% | 40.3% |
| BNB | 51.4% | 23.8% |
| XRP | 13.4% | −18.3% |
| SOL | 116.0% | 0.8% |
| TRX | 40.2% | 21.1% |
| ZEC | 94.8% | 2.4% |
| HYPE | 30.9% | −31.6% |
| DOGE | 77.2% | 25.8% |
| LINK | 47.0% | 31.6% |
| Median | 44.4% | 15.5% |
| beats holding | 7 of 10 | 1 of 10 |
Annual return before fees. Today's ten largest coins with a Binance USDT pair, daily candles from 2018 or listing to 13 Oct 2026.
On every one of the ten coins, the cloud variant came in below the one without it. The condition removed two thirds of the trades on average, from 128 to 41 per coin. Judged by the result, more of the removed trades were good ones than bad ones.
The cloud does not confirm the signal. It delays it.
That leaves the feature the cloud is known for: it is drawn 26 candles into the future. Does that shift at least help?
Without the forward shift, the cloud does better
The same cloud rule ran a second time, this time without the shift. The cloud is then built from today's prices instead of those from 26 days earlier.
Without the shift, the result was higher on eight of the ten coins, and the median rose from 15.5 to 20.2 percent. Both variants stayed far below the strategy without any cloud.
The feature that makes Ichimoku unique is only a handicap on these ten coins.
"Today's top ten are the winners, so everything looks good"
True. These ten coins are the ones that stayed on top until October 2026. Across all 103 coins Ichimoku has run on in the Arena so far, the strategy makes −1.0 percent a year after fees on daily candles. The high returns in the table are flattered.
The comparison with and without the cloud is not. Both variants run on the same coins over the same days, so the winners' advantage is in both columns. That the cloud hurts on ten of ten coins does not depend on which coins were picked.
What measurably helps Ichimoku on the platform is not a line but the market phase: a filter that only allows trading in a confirmed bull market lifts the strategy by 6.9 percentage points a year. Why that holds for almost every strategy is in In crypto, a strategy's first job is the exit.
Take the cloud out of your buy rule
- Do not wait to buy until price is above the cloud. On all ten large coins this condition pushed the result down, in the median from 44 to 16 percent a year.
- Treat 9/26/52 as tradition, not as a measured result. On bitcoin and ether they landed in the middle of random numbers.
- But do not pick the best number in hindsight either. On bitcoin the random pairs ranged from 8 to 44 percent. You only know the best of them afterwards, and you cannot trade it.
- Set the market phase first, then the indicator. The bull-market filter moved more than any setting on the lines.
What was checked is daily candles over a period of almost nine years. Weekly candles have not been checked yet; on them, Ichimoku has so far made −4.2 percent a year after fees across 45 coins. Whether the cloud helps there would be the next measurement.
You can test the strategy with all its settings yourself: Ichimoku Cloud in the Arena.
How it was calculated
Strategy: the Arena's Ichimoku TK cross. Buy when Tenkan crosses Kijun from below; sell on the cross downward. With the cloud filter, additionally: buy only above the cloud, sell also below its lower edge. Binance daily candles, 1 January 2018 (or listing) to 13 October 2026, no other filters.
Period test: 30 random combinations for bitcoin and 12 for ether, each three different numbers between 5 and 80, drawn with a fixed seed. Without the cloud, only the first two take effect. The rules and the numbers were fixed before the first comparison run. The ten-coin comparisons use today's top-ten list and report returns before fees.
FAQ
Why does the 52 not matter in the first test? It only sets the slower cloud boundary (Senkou Span B). Without the cloud filter, only the cross of Tenkan and Kijun decides.
Why not just recommend the best combination? Because it would be picked from the same data on which it looks good. The range of 8 to 44 percent shows how much luck a single setting can contain.
Does this apply to stocks or currencies? Only crypto was tested. It trades seven days a week, and coins swing more than the markets Ichimoku was built for.
Not investment advice, not a recommendation, not a forecast — historical patterns are no guarantee.
Study the Past — Improve your Future 🥋