// research
Which moving-average crossover works best on crypto?
None of them reliably. Of six common crossovers tested on six large coins, five beat simply holding the coin on only 1 or 2 of 6. The exception, EMA 9/21 on daily candles, beat holding on 5 of 6 — and on just 2 of 6 other coins it had not been picked on. On 4h candles the fast pairs traded more than 300 times per coin and paid for it in fees.
8 min read
A moving-average crossover is the most quoted trend rule there is, and every pair of numbers has its fans. Here six of them — four commonly quoted pairs, two of them on two timeframes — run on the same six coins, with the same costs, over the same 2020–2026 window, and are set against simply buying the coin and holding it. All of them made money in a market that rose; almost none of them made more than holding.
Setup
- Rule: long while the fast average is above the slow one, flat when it falls below. The position opens on the first bar the fast average is above the slow — at a cross, or on the first bar of the window if it already is — and closes at the next open after the cross back. One position, 95% of equity, fills at the next bar's open, no stop.
- Pairs and timeframes: EMA 9/21 and EMA 20/50 on 4h and on daily candles, SMA 20/100 on daily candles, and the SMA 50/200 golden cross on daily averages, read on 4h bars exactly as in the golden cross study. Six variants, nothing tuned.
- Coins and data: BTC, ETH, BNB, XRP, ADA and DOGE on Binance spot, 2020-01-01 → 2026-10-01.
- Costs: 0.1% fee and 0.05% slippage on each side, 10,000 USDT per run.
- Runs: strategies #78–#80, runs #751–#765 — EMA 9/21 on all six coins on both timeframes, the other pairs on BTC. The rest of the 36 variant × coin results are quota-free re-runs of the same configs and are counted in the medians below, not stored. The golden cross is shown by its published run #159 from the earlier study, which covers a different window — see below.
What each metric means: backtest metrics explained.
Results
| Pair | Timeframe | Trades, 6 coins | Made money | Beat holding | Median return | Median max drawdown | Lost to costs alone |
|---|---|---|---|---|---|---|---|
| EMA 9/21 | 4h | 1,952 | 6 of 6 | 2 of 6 | +939.05% | 75.16% | ≈ 60% |
| EMA 20/50 | 4h | 830 | 6 of 6 | 2 of 6 | +1,043.16% | 66.93% | ≈ 33% |
| EMA 9/21 | 1d | 299 | 6 of 6 | 5 of 6 | +1,737.45% | 68.20% | ≈ 14% |
| EMA 20/50 | 1d | 127 | 6 of 6 | 2 of 6 | +2,589.11% | 68.01% | ≈ 6% |
| SMA 20/100 | 1d | 93 | 5 of 6 | 1 of 6 | +1,354.97% | 71.80% | ≈ 4% |
| SMA 50/200 | 1d averages, 4h bars | 50 | 5 of 6 | 2 of 6 | +475.19% | 76.63% | ≈ 2% |
Buy & hold made a median +1,514.97% in the 4h runs and +1,524.18% in the daily runs. "Beat holding" compares each run with holding the same coin over the same run. "Lost to costs alone" is an estimate: about 0.3% per round trip (0.1% fee and 0.05% slippage on each side, on 95% of equity), compounded over the variant's median number of trades.
Three things stand out. Every pair made money on at least 5 of 6 coins — that is what a market that rose ten-fold does to any long-only rule. Beating the coin itself was rare: in 14 of 36 variant × coin pairs, and 5 of those 14 belong to one variant. And the drawdowns stayed large whatever the averages: the medians run from 66.93% to 76.63%, so the crossover did not keep the equity out of the deep falls either.
Backtest
EMA 9/21 crossover, 4h — MA crossover studyBTCUSDT · 1d · Spot · 2020-01-01 → 2026-10-01
- Return
- +1,272.80%
- Max drawdown
- 59.51%
- Sharpe
- 1.17
- Win rate
- 40.8%
- Trades
- 49
Fast pairs pay for their speed
The same EMA 9/21 rule traded 318 times on BTC on 4h candles and 49 times on daily ones. On BTC alone the 4h run paid 23,764 USDT in fees and 11,882 USDT in slippage on a 10,000 USDT start — summed over its trades — and ended at +351.63%; the daily run paid 7,224 and 3,612 and ended at +1,272.80%. Both EMA pairs did better on daily candles than on 4h by the median over the six coins, and coin by coin on 4 of 6 each.
Backtest
EMA 9/21 crossover, 4h — MA crossover studyBTCUSDT · 4h · Spot · 2020-01-01 → 2026-10-01
- Return
- +351.63%
- Max drawdown
- 64.42%
- Sharpe
- 0.76
- Win rate
- 30.2%
- Trades
- 318
| BTC, pair and timeframe | Return | Max drawdown | Trades | Fees + slippage, USDT |
|---|---|---|---|---|
| EMA 9/21, 4h | +351.63% | 64.42% | 318 | 35,647 |
| EMA 20/50, 4h | +387.97% | 59.50% | 141 | 11,710 |
| EMA 9/21, 1d | +1,272.80% | 59.51% | 49 | 10,836 |
| EMA 20/50, 1d | +998.01% | 50.00% | 20 | 3,967 |
| SMA 20/100, 1d | +417.10% | 55.13% | 14 | 1,520 |
| SMA 50/200, golden cross study* | +529.12% | 57.11% | 6 | 540 |
| Buy & hold | +1,059.22% (4h) / +1,078.75% (1d) | — | 1 | — |
* Run #159 covers 2020-01-01 → 2026-01-01 and was charged a 0.05% fee instead of 0.1%.
Fees in USDT grow with the account, so they are not a clean comparison between runs — but the trade count is: a rule that trades 300 times pays 0.3% three hundred times, whatever it earns in between.
Backtest
EMA 20/50 crossover, 4h — MA crossover studyBTCUSDT · 1d · Spot · 2020-01-01 → 2026-10-01
- Return
- +998.01%
- Max drawdown
- 50.00%
- Sharpe
- 1.07
- Win rate
- 45.0%
- Trades
- 20
Backtest
SMA 20/100 crossover, 1d — MA crossover studyBTCUSDT · 1d · Spot · 2020-01-01 → 2026-10-01
- Return
- +417.10%
- Max drawdown
- 55.13%
- Sharpe
- 0.78
- Win rate
- 50.0%
- Trades
- 14
The golden cross
Backtest
Golden Cross 50/200 daily — top-20 studyBTCUSDT · 4h · Spot · 2020-01-01 → 2026-01-01
- Return
- +529.12%
- Max drawdown
- 57.11%
- Sharpe
- 0.96
- Win rate
- 66.7%
- Trades
- 6
The golden cross was not re-run for this article. Its published runs come from the golden cross study: the same rule and the same fill model, but the window ends on 2026-01-01 and the fee is 0.05%. Re-run on this article's window and costs without storing the runs, it made money on 5 of 6 coins, beat holding on 2 (ADA and DOGE) and made +356.55% on BTC. The two BTC numbers differ for three reasons: the window, the fee, and the warm-up — since 2026-10-16 the platform loads an indicator's history before the window, so a new run can trade from January 2020, while the published one waited until its 200-day average was ready in July 2020.
The golden cross traded least — 50 times on six coins — and had the deepest median drawdown, 76.63%: a 200-day average turns after the fall, not before it.
Is the best pair an edge?
No. Six variants were tried on six coins, and in this window the leader depends on what is counted: EMA 20/50 on daily candles had the highest median return, EMA 9/21 on daily candles beat holding most often. Coin by coin the best variant changes too — EMA 9/21 daily on BTC, ETH and BNB, EMA 20/50 on 4h on XRP and DOGE, EMA 20/50 daily on ADA.
To see whether the leaders carry over, the two daily EMA pairs were re-run on six coins they had not been picked on — LTC, LINK, TRX, BCH, ETC and XLM — over the same window and costs. EMA 9/21 made money on 4 of 6 and beat holding on 2 (ETC and XLM), with a median of +131.91% against +133.63% for holding. EMA 20/50 made money on 4 of 6 and beat holding on 2 (BCH and ETC), median +97.90% against +106.59%. The 5-of-6 record did not travel.
EMA 9/21 on all six coins
| Coin | EMA 9/21, 4h | Trades | Buy & Hold (4h) | EMA 9/21, 1d | Trades | Buy & Hold (1d) |
|---|---|---|---|---|---|---|
| BTC | +351.63% | 318 | +1,059.22% | +1,272.80% | 49 | +1,078.75% |
| ETH | +182.76% | 342 | +1,970.71% | +2,202.11% | 46 | +1,969.60% |
| BNB | +2,194.70% | 314 | +5,469.44% | +6,666.59% | 54 | +5,525.66% |
| XRP | +357.38% | 334 | +673.83% | +185.59% | 53 | +673.84% |
| ADA | +1,520.72% | 311 | +639.18% | +785.92% | 53 | +435.42% |
| DOGE | +7,917.73% | 333 | +4,560.15% | +12,485.13% | 44 | +3,886.40% |
Each return links to its run. Buy & hold is the run's own benchmark; for ADA and DOGE it starts once the averages are warm, which is later for the daily runs.
Conclusion
- Every crossover made money on most coins in 2020–2026, and five of the six beat simply holding on only 1 or 2 of 6.
- Speed costs: on 4h candles EMA 9/21 traded more than 300 times per coin, and costs alone would have taken about 60% of the equity; the same pair on daily candles traded about 50 times.
- The one variant that beat holding on 5 of 6 coins beat it on 2 of 6 coins it had not been chosen on. In this window the "best" crossover is a description of the past, not an edge.
More on the slowest of these rules: the golden cross on the top-20 coins. What adding a stop-loss does to a trend rule like these: does a stop-loss help?
Test your own idea
Create a free account, start from a template or from your own rules, and read your first report.
More articles
- Does high funding predict crypto crashes? 8 coins, 2020–2026BIS economists found that high crypto carry predicts crashes. On 8 Binance perpetuals, 2020–2026, crashes came more often — but after bigger gains.
- Crypto momentum after the paper: a weekly test on 8 coinsLiu and Tsyvinski found weekly momentum in bitcoin up to 2018. On 8 coins in 2020–2026 the one-week signal lost to holding; four weeks cut drawdowns.
- Does price move the hour before funding? 8 coins, 2020–2026Do traders sell before paying funding? The hour before each settlement on 8 Binance perpetuals, 2020–2026: no reliable move, and fees took what was left.
- Bitcoin after the halving: 2020 and 2024 backtestedDid buying BTC at the halving pay? A year after 2020: +554.04%, top 2.5% of start days. After 2024: +30.45%, below the median. Two cases, no pattern.
- Weekends and time of day: when does bitcoin move?Do bitcoin gains come on weekends or outside US trading hours? Hourly BTC and ETH data since 2020, and session backtests: with fees, -99.87%.