You tested a strategy on a coin, it lost 10 % a year, and it looks like the strategy is no good. Or you saw a +175 % result in the Winner's List and it looks like that one works. Both readings miss the same line: the one right under the strategy's return, which shows what the coin did with no strategy at all. Across 197,435 backtests on 653 pairs, that second line is the whole story.
The market you are actually in
Take buy and hold, one entry, held to today, on every pair the Arena has run it on since 2018 or the pair's listing. Across 405 pairs the median result is a loss of 47.6 % per year. Only 17 % of pairs made money at all. The bottom tenth lost 90 % a year or more. Averaged over 360 pairs, the deepest drawdown a holder sat through was 92 %. That is the base rate for holding a coin picked at random.
Bitcoin sits at the other end of that distribution: 28.5 % a year over eight and a half years, with a 59 % drawdown on the way. That puts the largest coin in the top tenth of its own market.
Most pairs have one run, and then the long way down. Picking the coin is the first risk, before any indicator is chosen. If you hold only Bitcoin, the rest of this article is about a problem you have mostly avoided. If you hold anything else, it is about the problem you have.
The first job is the exit
Now the same universe with a strategy on it. Each row compares a strategy with buy and hold on the same pairs, same window, same interval.
| Strategy, daily | Pairs | Strategy CAGR | Buy & hold CAGR | Strategy drawdown |
|---|---|---|---|---|
| EMA Cross | 177 | +11.9 % | −30.6 % | −64 % |
| Supertrend | 176 | +10.9 % | −30.3 % | −69 % |
| Williams Alligator | 156 | +6.6 % | −35.6 % | −58 % |
| Smoothed Heiken Ashi | 405 | +0.7 % | −38.3 % | −66 % |
| Capitulation Finder | 308 | −4.2 % | −45.9 % | −71 % |
| RSI / SMA Cross, weekly | 459 | −10.0 % | −44.8 % | −66 % |
Averages across pairs, so a coin that went to zero counts fully. The strategy that lost 10 % a year did so on a universe that lost 45 %. That is 35 percentage points a year of money not lost, and a drawdown of 66 % instead of 92 %. That is what the exit is worth.
Beating buy and hold on a coin that went to zero is survival, not income. It is the first job of a strategy in this market, and every strategy in the table did it. It is not the second job.
The second job is decided by the regime
The second job is earning, and the Edge Library measures what decides it. Each entry filter is tested against the unfiltered version of the same strategy on the same pairs; the number is the median change in annual return across matched pairs, with a 90 % interval, gross of fees.
| Strategy | Unfiltered median CAGR | With bull-market gate | Change |
|---|---|---|---|
| Fear & Greed Cadence | −20.0 % | +7.0 % (strict) | +17.2 pp |
| Supertrend | −22.1 % | +6.4 % (strict) | +16.1 pp |
| MACD Cross | −23.3 % | +7.0 % (strict) | +13.2 pp |
| RSI / SMA Cross | −23.2 % | +7.9 % (strict) | +12.9 pp |
| WMA Trend | −16.3 % | +11.4 % (strict) | +11.7 pp |
| EMA Cross | −14.4 % | +15.0 % (early) | +10.6 pp |
| Ichimoku | −8.5 % | +20.0 % (strict) | +6.9 pp |
Fourteen strategies, 69 filter cells with a whole interval above +1 pp. The gate is the Arena's bull-market stage, read from the 20- and 50-week averages, known on the day. After per-side fees the filtered numbers lose about half a point and keep their sign.
When you trade decides the sign. What you trade with decides the size. The same EMA cross loses 14 % a year unfiltered and makes 15 % when it only trades in a confirmed uptrend. No indicator in the table changes sign on its own; the regime does it for all of them.
Volatility does the same at a smaller scale, and it sorts by strategy family. Mean reversion wants a loud market: RSI / SMA Cross gains 13.7 pp when the ATR is high. Trend following wants a quiet one: Golden Cross gains 14.7 pp when the ATR is low. Match the family to the phase before tuning a parameter.
The bar for an altcoin is Bitcoin
A strategy on an altcoin has two benchmarks, and the honest one is not the coin's own buy and hold. It is Bitcoin held over the same years: 28.5 % a year, 59 % drawdown.
Against that bar the first table reads differently. EMA Cross at +11.9 % across 177 pairs beats its coins and trails Bitcoin by 17 points a year. Supertrend at +10.9 % does the same. On Bitcoin itself, EMA Cross returns 40.5 % a year with a 44 % drawdown across 21 runs, beating buy and hold by 15.8 points; the house ensemble on daily data averages 14.3 % across 127 pairs.
If a strategy on an altcoin cannot beat holding Bitcoin, the altcoin was the mistake, not the strategy. The money in this dataset sits in two places: Bitcoin with a trend strategy, and altcoins with a regime gate. Altcoins without a gate are where the first job gets done and the second does not.
How to read one result
The top entry in the Winner's List is RSI / SMA Cross on Solana, three-day candles, +175 % a year over 97 trades. The robustness field runs the same strategy on the 402 other pairs the factory covered: median −12 %, 37 % of pairs positive, the Solana cell in the 99th percentile. Verdict: spike. A single cell is a claim; the neighbourhood is the evidence.
Four more checks, each with its number from the dataset:
- Win rate is not edge. RSI(2) Mean Reversion wins 64 % of trades and returns −1.7 % a year. EMA Cross wins 34 % and returns +11.9 %.
- Turnover eats the edge. Stoch-RSI / SMA Cross averaged +9.2 % a year before fees and +1.1 % after, at 558 round trips per run; 16 of 28 cells turned negative net, and none beat buy and hold net.
- Daily, two-day and three-day candles are one block. Shifting the alignment of a multi-day candle alone moves CAGR by 6.7 points on average, up to 12.3. A difference smaller than that between intervals is not a ranking.
- The shiniest cells have the fewest observations. Pi Cycle: eight runs on one pair. The verdict threshold is 30 completed trades.
What to do with it
- Decide the universe first. Bitcoin only, or altcoins with a strategy. Altcoins without one are the −47 % row.
- Gate by regime before choosing an indicator. The gate moves the return by 10 to 17 points; the choice of indicator moves it less.
- Benchmark every altcoin result against Bitcoin held over the same years, not against the coin's own buy and hold.
- Before trusting one cell: read its neighbourhood, multiply trades by fees, and look at the drawdown as the number you will actually live through.
What is not claimed: that any of these strategies will earn next year. The regime effect is measured on 2018 to 2026 and the Edge Library keeps re-measuring it; whether it holds through the next full cycle is the open question, and the only one worth a new backtest.
FAQ
Why are the buy-and-hold averages so negative? Because the universe includes every pair the Arena ran, including coins that were delisted or fell 99 %. That is the point: a backtest universe that only keeps survivors would report a market that does not exist.
What does "buy & hold CAGR" mean here? The median single-entry return across all possible entry dates in the window, per pair, then averaged across pairs. It is not the return from one lucky entry date.
What does "spike" mean? The result depends on the pair you picked: the same strategy on the neighbouring pairs is centred far below it. A "plateau" holds across pairs. The field counts the neighbours actually tested rather than estimating them.
Which fees? 0.10 % per side for crypto pairs in the net figures.
Strategies and filters were fixed before the runs; every figure carries its pair count, window and benchmark in the Arena.
Not investment advice, not a recommendation, not a forecast. Historical patterns are not a promise.
Sources: Backtesting Arena, retrieved 21 September 2026: strategy insights matrix (197,435 raw runs, 107,294 scored, 653 pairs), Edge Library crypto (computed 21 September 2026, 14 strategies, 69 helping cells), robustness field for bnh_fixed and rsi_sma, strategy performance for BTCUSDT, Winner's List, strategy catalogue deprecation notes (23 June 2026)
Study the Past — Improve your Future 🥋