The 200-week moving average (the average of weekly closing prices over the last 200 weeks, about four years) is one of the most-cited long-term filters in crypto. The idea is simple: trade above it, step aside below it. In our backtests this regime filter helps trend strategies almost across the board — it keeps them flat through bear markets.
But it has a quirk that looks, at first glance, like a flaw. This week we put that quirk under the microscope — and the result runs against intuition.
The apparent gap
A 200-week moving average needs 200 weeks of history. Most coins on an exchange like Binance don't have that — they were listed in 2021, 2022 or later. So what happens to the filter when there aren't 200 weeks yet?
In our house strategy — a curated trend ensemble that sits behind the 200-week filter — the filter passes automatically. No reference value, no blocker: the coin is allowed to trade.
That smells like a design bug. Young coins run without the very filter the strategy's edge rests on. And young coins are statistically the riskiest — many fresh alts bleed out after listing. The obvious fix: give young coins a real filter instead of waving them through.
Measure first, build second
Before we ship a strategy change, we test it. Not against a hand-picked chart, but across the cross-section. Concretely: 207 coins, daily candles, with realistic trading costs (0.20% per round-trip). Split into two groups:
- 92 "mature" coins with at least 200 weeks of history
- 115 "young" coins with fewer than 200 weeks
And we tested not one but several corrections against each other:
- Hard block — no trading until 200 weeks of history exist.
- Adaptive — use the longest available weekly average, from week 52 on.
- Soft adaptive — wave through in the first year as before, apply a real filter from week 52.
Each against the current behaviour (auto-pass) as the baseline.
The result: the "bug" is a feature
The hard correction dismantles the strategy. Not just on young coins — on the mature ones too. The reason is obvious once you see it: the early phase of every coin — its first few years, before it's 200 weeks old — often contains the strongest bull market of its life. Bitcoin in 2017, the alt rally of 2020/21: all of that happens in a window where the 200-week average doesn't even exist yet. The auto-pass is what lets the strategy capture that early phase. Block it, and you sit out the best part.
The soft correction is built more honestly and does help young coins — it halves the loss. But "halves" here means: from deep red to less-deep red.
| Group | Variant | Net CAGR (median) | Max drawdown | Trades (median) |
|---|---|---|---|---|
| Mature (≥200 wk) | Current (auto-pass) | +10.8% | 75% | 23 |
| Mature | Soft correction | +6.8% | 70% | 16 |
| Young (<200 wk) | Current (auto-pass) | −35.6% | 69% | 8 |
| Young | Soft correction | −20.7% | 57% | 5 |
| All 207 | Current (auto-pass) | −13.1% | 73% | 15 |
| All 207 | Soft correction | −11.9% | 61% | 7 |
Two things stand out:
First: young coins lose money with this trend strategy — filter correction or not. The soft variant pushes the median loss from −35.6% to −20.7% and cuts drawdown noticeably, but it doesn't turn a losing proposition into a winning one. Trend-following on fresh alts is structurally hard; no filter reverses that.
Second: the correction costs the mature coins about four percentage points of return per year (+10.8% → +6.8%). Exactly where the strategy makes its money, and where most users actually trade — Bitcoin, Ethereum, the big names.
You can't sell the soft correction as "better." It's a trade: less drawdown and lower trading costs in exchange for less return in the core. A risk trade, not a return gain.
The catch in the mature numbers
There's a caveat worth stating out loud: the 92 mature coins are exactly the ones that survived. A coin only has 200 weeks of history if it wasn't delisted and stayed relevant for four years. Their strong early bull market, which the auto-pass captures, is partly a survivorship effect. The young coins — the recent listings — are the more honest look at how the strategy handles what's actually new to the market.
So the decision wasn't pure backtest arithmetic. But even with that caveat: the correction doesn't make the young group profitable, and it weakens the group the proven edge rests on. That isn't enough for a new strategy version.
What we did with it
Nothing. The strategy stays frozen, unchanged.
That's the point. A change that sounded like an obvious bug fix — "young coins skip the filter, surely you fix that" — didn't survive scrutiny. The apparent bug was the vehicle for early bull-market participation. Had we shipped the intuition unchecked, we'd have traded return for a feeling of tidiness.
Backtests aren't predictions, and these numbers are medians across market phases with the stated limits. Not a trading signal. But the lesson travels: an idea that feels like a bug fix has to survive contact with the data. Ours didn't — and knowing that in time is cheaper than learning it later in live trading.
What Backtesting Arena contributes here
We treat strategy changes as experiments: first a hypothesis, then a test across the cross-section, then a decision with visible numbers — even when the decision is "change nothing." The 200-week filter, the mature/young cohort split, the realistic costs and the survivorship warning aren't garnish; they're the core of the method. Anyone backtesting on their own can apply the same discipline: never pin a change to one pretty chart — pin it to the cross-section, and show the cases where the idea fails.
FAQ
What is the 200-week moving average? The average of weekly closing prices over the last 200 weeks (about four years). Used as a regime filter: above the average counts as constructive, below it as defensive.
Why do young coins skip the filter? Because there aren't yet 200 weeks of history to form an average from. In our strategy the filter then passes automatically instead of blocking.
Isn't that a bug? It looked like one. But the test shows this auto-pass captures a coin's early — and often highest-returning — phase. "Fixing" it lowered returns instead of improving them.
Does a filter correction help young coins at all? Yes, but only so far. The soft variant halved the median loss (−35.6% → −20.7%) and cut drawdown. Young coins still ended up negative with this trend strategy.
Then why not change the strategy at least for young coins? Because the same mechanism that helps young coins weakens the mature ones — exactly the coins the proven edge and most real trades sit on. The trade isn't worth it.
Are these numbers reliable? They're medians over 207 coins with realistic costs, not a prediction. The mature group is also survivorship-skewed (only surviving coins have 200 weeks of history). Enough to decide on, not a guarantee.