Koryu · 黄龍
Koryu
← Research library
Research · 研究 · 30 · Craft

Everyone can see the average. Almost no one can follow it.

14 Aug 20265 min readMomentum craftKoryu Research

This article discusses a strategy family studied in public research, including its costs and failure modes. It is educational analysis, not a recommendation to run any strategy. Nothing here is investment advice. Decisions are yours.

Strip crypto quant research to the one finding that has survived every cycle, every venue change, and every wave of sophistication, and it is embarrassingly simple: holding BTC when it trades above a long moving average and standing aside when it does not has historically kept most of the upside while dodging the catastrophic middles of the winters. The oldest idea in technical analysis, applied to the newest asset class, remains its most defensible systematic result.

What the strategy family is

Time-series trend following asks one question of one asset: is it above or below its own recent history? The classic implementations use a long simple moving average, 200 days being the canonical line, or a faster stack like the 20/50 combination our regime dial uses. Above: hold. Below: hold cash, or stables. No prediction, no target, no narrative, just a mechanical answer to whether the tide is in. The family's academic pedigree is deep, time-series momentum is among the most replicated effects in empirical finance across two centuries of assets, and its crypto translation has been studied publicly for a decade with the same qualitative result: worse than holding in relentless bull years, dramatically better through the full cycle, because the cycle includes winters.

The evidence, bull and bear

The case rests on asymmetry. In 2022, BTC fell roughly three quarters from its 2021 peak; a disciplined long-average follower was in stables for most of that decline, because price crossed below trend early in the collapse and stayed there for a year. Avoiding even half of a 75% drawdown transforms a track record, and that recovery arithmetic is the whole reason we sell discipline instead of predictions: the buy-and-holder needed a 300% recovery to break even, the trend follower needed a fraction of that. Our own founding experiment adds an independent datapoint from a different angle: in the transplant experiment, the same trend stack, used merely to gate an alt strategy rather than to trade BTC, was the single component that turned a losing system into a winning one. One mechanism, two very different applications, the same conclusion: in a one-factor market, knowing the factor's trend state is most of the risk battle.

The whipsaw ledger nobody advertises

Now the invoice. Trend following pays for its winters in chop: every time price crosses the line and crosses back, the follower sells low and rebuys higher, a guaranteed small loss called a whipsaw, and ranging years produce several. The strategy's equity curve is therefore emotionally brutal in a specific way: it looks worse than holding through every strong bull stretch, it bleeds small cuts through every range, and it justifies itself in one compressed period per cycle, when the winter arrives and the follower is not there. Most people who adopt trend rules abandon them mid-range, having paid three whipsaws and watched a buy-and-hold neighbor gloat, usually months before the rule would have earned its keep. That behavioral difficulty is precisely the strategy's source of return, the reason the effect survives being public, which deserves its own section.

Why it still works after everyone knows

An anomaly this old and this published should have been arbitraged away, and its persistence has two standing explanations. Behavioral: the strategy requires selling into fear after a break and rebuying above your exit after a recovery, actions that feel wrong at the moment of decision, and it requires eating whipsaws without abandoning the rule, a discipline most participants demonstrably lack. Crowds cannot arbitrage away a return that is compensation for doing what crowds cannot do. Structural, and crypto-specific: each cycle imports a fresh cohort of participants with no winter memory and maximal leverage, the bull-market genius effect on schedule, recreating the exact herding-then-cascading dynamics that trends feed on. The edge is not informational, everyone can see the moving average. It is temperamental, and temperament, unlike information, does not diffuse.

Honest limits

Four, stated plainly. Trend following never catches tops or bottoms; it concedes the first leg of every recovery and the last leg of every peak by construction. Its verdict sample is tiny: crypto has produced a handful of full cycles, so every backtest of the family leans on few independent episodes, exactly the condition where overfitting does its worst work, which is why parameter humility, plateaus over precision, matters more here than sophistication. A regime can arrive that punishes the rule for years, extended violent chop is its documented kryptonite. And execution is not free: crossings cluster in volatile tape, and the follower pays spreads and slippage exactly then. None of these limits overturn the family's record. They explain why its return exists and who ends up collecting it.

Where this site stands

Our Regime Dial publishes a trend-stack reading on BTC daily, with breadth attached, as measurement: what the oldest strategy in crypto would currently see, computed from a fixed public formula and attested through a commit-reveal record. We do not tell anyone to act on it, and the majors trend sleeve that would formalize this family into a tested specification is queued in the journal overview behind the stop-geometry work, where it will face the same gates as everything else. The oldest idea in the market deserves the same discipline as the newest: measured in public, validated out-of-sample, or not shipped at all. Decisions are yours.

Related reading
Momentum craftBTC dominance rotation: when majors lead and alts bleed5 min readMomentum craftWhy breakouts fail in crypto chop: the regime problem5 min readMomentum craftStop placement in a 24/7 market: distance, wicks, and cascades5 min read
Frequently asked

Does trend following actually work on bitcoin?

Across full cycles, the public record says yes in a specific sense: worse than holding through relentless bull years, dramatically better through the cycle because it sidesteps most of each winter. In 2022 a disciplined long-average follower was in stables for most of a roughly 75% decline.

What is the catch with trend following?

Whipsaws: every false crossing sells low and rebuys higher, and ranging years produce several. The strategy looks worse than holding in every strong bull stretch and justifies itself in one compressed period per cycle, which is exactly when most adopters have already abandoned it.

Why hasn't the effect been arbitraged away?

Because its return is compensation for doing what crowds cannot: selling into fear, rebuying above your exit, and eating whipsaws without abandoning the rule. The edge is temperamental rather than informational, and temperament does not diffuse. Each crypto cycle also imports a fresh leveraged cohort that recreates the dynamics.

Which moving average is best for BTC?

The honest answer is that broad parameter regions work similarly and precision is overfitting bait: crypto offers few independent cycles to tune on. The 200-day is canonical; faster stacks trade earlier warning for more noise. Plateaus over precision.

How does this relate to the Regime Dial?

The dial publishes a fixed trend-stack reading on BTC plus market breadth daily, as measurement rather than advice, attested. A formalized majors trend sleeve is queued in the derivation program, where it faces the same validation gates as everything else.