Ichimoku Kinko Hyo is one of the most visually sophisticated indicators in technical analysis. Five components, each measuring a different dimension of price action: Tenkan-sen (conversion line), Kijun-sen (base line), Senkou Span A and B (the cloud boundaries), and Chikou Span (lagging span). The cloud itself provides support and resistance, the crossovers generate signals, and the lagging span confirms trend direction. It is a complete trading system in one indicator.
We built it. We tested it. We archived it. Here is why.
The Tournament Promise
In our initial tournament screening, Ichimoku Cloud produced encouraging results on several altcoins. Using default parameters adapted for crypto (tenkan_period=9, kijun_period=26, senkou_b_period=52, displacement=26), the strategy generated signals that appeared to capture trend transitions.
On 15-minute candles, the win rate hovered around 45%, which is typical for trend-following strategies. The average winner was roughly 2x the average loser, producing positive expectancy on paper. Several symbols showed Sharpe ratios between 1.0 and 2.5 in the tournament. Not spectacular, but enough to warrant further investigation.
The visual appeal was part of the problem. Ichimoku charts look sophisticated. The cloud shifts color, the spans intersect, and the lagging span confirms or denies. Every trade has a narrative. But narratives are not the same as edges.
The Sweep: Searching for Better Parameters
We ran a parameter sweep across the three core period settings and the displacement. The grid was modest, roughly 200 combinations per symbol, testing shorter and longer periods to find the crypto-optimal configuration.
The sweep produced what looked like improvement. On SOL, the best configuration reached Sharpe 2.8. On AVAX, it hit 2.3. Several other symbols showed their best results at non-default parameters, typically with shorter periods (tenkan_period=7, kijun_period=20) that would theoretically capture crypto's faster cycles.
But the warning signs were already visible in the sweep data. The distribution of Sharpe ratios across the parameter grid was flat. There was no clear peak, no neighborhood of consistently good parameters. The best configuration was only marginally better than dozens of mediocre configurations, which is a hallmark of the multiple comparisons problem generating false positives.
The Validation Collapse
We tested the sweep winners across five market regime periods: 2021-2022 (bull to crash), 2022-2023 (bear and recovery), 2023-2024 (recovery to new highs), 2024-2025 (consolidation), and 2025-2026 (recent).
A strategy must be profitable in at least 4 of 5 periods to earn a ROBUST verdict. Ichimoku Cloud achieved this on zero symbols.
The results were not uniformly terrible. Some symbol-period combinations were modestly profitable. But the pattern was random. A symbol that was profitable in period 1 and 2 would be negative in periods 3 and 4. A different symbol showed the reverse pattern. There was no consistency across regimes, which is the definition of a strategy that has been fitted to specific market conditions rather than capturing a durable edge.
Trend alignment, a related multi-timeframe strategy using SuperTrend and ADX, showed the same pattern. Promising in screening, incoherent in validation. Both strategies were archived on the same day.
Why Complexity Failed
Ichimoku Cloud has five components. Each adds information, but each also adds a degree of freedom. Three core period parameters, plus the displacement, create a four-dimensional optimization space. In that space, you are virtually guaranteed to find configurations that look good on any specific historical period.
The deeper problem is that Ichimoku was designed for Japanese rice futures in the 1930s. Its component periods (9, 26, 52) were calibrated to the Japanese trading week and half-year. The indicator assumes a specific market structure: regular trading hours, closing auctions, and price behavior driven by institutional flow.
Crypto markets violate every one of these assumptions. They trade 24/7, there are no closing auctions, and price behavior is driven by retail flow and liquidation cascades. The periods that make sense in traditional markets have no inherent meaning in crypto. You can optimize them for crypto, but then you are fitting to historical data rather than capturing the structural insight that made Ichimoku work on rice futures.
The Over-Parameterization Trap
A useful mental model: the more parameters a strategy has, the more data it needs to validate. Each parameter adds a dimension to the optimization space, and the amount of data required to distinguish signal from noise grows exponentially with dimensionality.
Our mean reversion strategy has three effective parameters: bb_period, bb_std, and min_confidence. That is a three-dimensional space. With 13 symbols and 5 regime periods, we have 65 independent tests, more than enough to distinguish a genuine edge.
Ichimoku has four parameters in a four-dimensional space. But the issue is worse than it appears because the five components create complex nonlinear interactions. The cloud width depends on the interaction between senkou_span_a and senkou_span_b, which depends on tenkan_period, kijun_period, and senkou_b_period simultaneously. The effective dimensionality of the optimization space is higher than four.
With higher effective dimensionality, the 65 independent tests are no longer sufficient. You need hundreds of independent tests to achieve the same statistical confidence. We do not have hundreds of independent regime periods. Five periods across five years is what reality provides.
What We Learned
Lesson one: complexity is not a proxy for sophistication. A two-parameter strategy (bb_period and bb_std) that works across 13 symbols and 5 regimes is more sophisticated than a four-parameter strategy that works on none. Sophistication is measured by consistency, not by indicator count.
Lesson two: tournament screening is a necessary but insufficient filter. Ichimoku cleared the tournament with Sharpe ratios between 1.0 and 2.5. So did OU mean reversion (Sharpe 1.98) and wavelet decomposition (Sharpe 3.19). All three collapsed in validation. Tournament screening only tells you that a strategy can produce positive results on some data. Validation tells you whether those results persist.
Lesson three: visual complexity creates false confidence. Ichimoku charts are beautiful. The cloud, the crossovers, the lagging confirmation all create a sense of information density. But information density is not the same as predictive power. A simple Bollinger Band chart is less visually impressive and more predictive.
Lesson four: strategies designed for one market structure do not transfer automatically to another. Ichimoku was designed for a specific market. Bollinger Bands are market-agnostic (they adapt to volatility by construction). The more assumptions a strategy bakes in about market structure, the more brittle it becomes when deployed in a different structure.
The Archive Decision
Archiving Ichimoku Cloud was not painful because our pipeline made the decision data-driven. We did not agonize over whether to give it one more chance with different parameters. The validation matrix was clear: zero symbols achieved ROBUST status. The strategy does not have a durable edge in crypto.
We maintain the implementation in our codebase because it may be useful for future research or as a component in ensemble strategies. But it will not be deployed to paper trading or live execution unless future validation provides strong evidence of regime-specific edge that justifies selective deployment.
Our 40-strategy catalog exists precisely for this reason. Most strategies do not work. The value of the catalog is not in its size but in the systematic process for determining which strategies belong in the small subset that actually makes money. Ichimoku Cloud contributed to our understanding of what does not work, which is itself a valuable data point in the search for what does.