We tested 2,928 crypto strategy settings. Here is what survived.
Real exchange fees. Two separate halves of history. Most popular strategies failed.
2,928settings tested
0fast strategies passed
3approaches held up
The faster it trades, the worse it does
Minutes
0of 114 passed
Hours
0of 900 passed
Days and weeks
3of 422 passed
failed close, retesting passed
What held up
Slow, spread over many positions, and still with bad years.
Paper tradingFutures portfolio, rebalanced daily
Lite 12 with the market filter
Earned in 3 of 4 periods. Flat in the 2022 to 2023 bear.
Sharpe
2.442020 to 21
0.162022 to 23 bear
2.042024 to mid 25
1.80mid 25 to now
Worst drop 6 to 11%
How it works and the fine print
Holds a basket of up to 12 liquid perpetual futures chosen by slow trend and momentum rules. It halves its size when Bitcoin is under its 100 day average and caps the basket's volatility at 20% a year.
Period
Sharpe
Worst drop
Jul 2020 to Dec 2021
2.44
8.2%
2022 to 2023, bear market
0.16
11.3%
Jan 2024 to Jun 2025
2.04
7.1%
Jul 2025 to Oct 2026
1.80
6.0%
Tested at 250 USDT on every perpetual that existed at the time, delisted coins included. It earned in three of four periods and went flat in the two year bear. It runs in paper mode on Treo; we have not put real money behind it yet.
Research, not offeredFutures, rebalanced weekly
Funding rank, long low funding and short high funding
Strong in both halves, but lost 26.7% in 2020.
Sharpe
1.052019 to 2023
2.062024 to now
Worst drop 13 to 20%
How it works and the fine print
Each week, buys the perpetuals where longs pay the least funding and shorts the ones where they pay the most, in equal amounts.
Period
Sharpe
Worst drop
2019 to 2023
1.05
20.1%
Jan 2024 to Oct 2026
2.06
13.1%
Lumpy: it lost 26.7% in 2020 and 14.2% in 2023. Small coins can squeeze the short side. The coin list is today's top 30, which flatters it.
Research, not offeredMarket neutral, held for weeks
Funding carry, hold spot and short the same perpetual
Steady every month, but shrinking every year.
BTC carry, % a year
+25.8%2021
+9.4%2024
+3.9%2025
+1.7%2026
Worst month about minus 0.2%
How it works and the fine print
Buys a coin on spot and shorts the same amount of its perpetual, so price moves cancel and the position collects the funding that leveraged longs pay.
Year
Carry on BTC
Note
2021
+25.8%
full year
2024
+9.4%
full year
2025
+3.9%
full year
2026
+1.7%
to October
Steady (91% of test months positive on BTC and ETH, worst month about minus 0.2%) but fading every year. Needs spot and futures on one exchange, margin top ups when price spikes, and it carries exchange risk.
What failed
0/216passed
Support and resistance scalping
1 to 15 minute charts
Wins 70 to 80% of trades. One loss wipes out 5 to 12 wins.
Only very deep oversold buys held about 3 days showed a small average edge (0.4 to 1.0% a trade). Added to a real bot, it made results worse.
0held up of 765
Moving average crossovers (SMA, EMA, DEMA)
1 hour charts
Popular settings stopped working in the second half.
Details
Tested: 765 settings
Result: Fail, one ETH variant in retest
26 of 765 passed in one test, but the common window lengths failed in the second half and walk forward testing failed.
0/48passed
Opening range breakout
1 hour charts
No setting made money in both halves.
Details
Tested: 48 settings on BTC, ETH and 20 coins, 2024 to Oct 2026
Result: 0 of 48 passed
44 of 48 settings lost money in both halves after fees, and none made money in both.
0/36passed
VWAP band fade and follow
1 hour charts
The moves are too small to pay the fees.
Details
Tested: 36 settings, 2024 to Oct 2026
Result: 0 of 36 passed
Fading and following the bands both lost in most settings; the moves are a few basis points against a 14 basis point round trip in fees.
0/39passed
Bollinger squeeze breakout
1 hour charts
Worked in one half, lost in the other.
Details
Tested: 39 settings, 2024 to Oct 2026
Result: 0 of 39 passed
31 of 39 settings lost in both halves; the rest worked in one half and lost in the other.
0/49passed
Best day of the week, best hour of the day
1 hour charts
Last year's best day is not next year's.
Details
Tested: About 50 settings, plus the 358 strategy tournament
Result: Fail
A day picked in one period repeated weakly at best (Sharpe under 1). Trading the hour before the 08:00 funding settlement lost in every setting.
0/3passed
Fear and Greed rules
Days
Did not cut drops or add return.
Details
Tested: 3 rule families
Result: Fail
No setting improved both return per unit of risk and worst drop in both halves.
✕failed
Buy last month's top 3 coins
Weekly
Huge in 2020 and 2021, gone since.
Details
Tested: Spot, top 20 coins, 2020 to Oct 2026
Result: Fail
Huge in 2020 to 2023, then lost 30% from 2024 on, with a worst drop of 84%. Classic momentum worked in 2020 and 2021 and has been gone since.
Why winning most trades can still lose money
=
12 small wins of about 0.05% are erased by one loss of 0.6%, before fees.
How we test
2 halvesMust make money in both
Real feesPlus funding on futures
50 to 1,000 USDTReal account sizes
Luck checkMust beat random rules
Sharpe is return divided by how much it swings: above 1 is good, above 2 is rare. Worst drop is the largest fall from a high. A backtest describes the past, not the future.
Questions
What is the best crypto trading strategy?
In our tests nothing fast survived fees. What held up was slow and spread across many coins: a daily rebalanced futures basket with a market filter, and funding based books. Even those have losing years, and past results do not predict future ones.
Does RSI work for crypto?
Not on its own in our tests. RSI under 30 and similar oversold rules had no edge after fees on 15 minute to 4 hour charts. Only very deep oversold buys held for about 3 days showed a small average edge, and adding it to a real bot made results worse.
Do grid bots make money?
0 of 144 grid settings on BTC and ETH passed our test. Grids collect small amounts while price moves sideways and give it back when a strong trend arrives.
Why does scalping lose even when most trades win?
Fees and the size of the losses. In our support and resistance test, 70 to 80% of trades won, but one average loss erased 5 to 12 average wins.
How do you test a strategy?
We split history into two halves and test each separately with exchange fees and funding included, at real account sizes. A strategy passes only if it is positive in both halves with a Sharpe of at least 1 in both, and beats random rules of the same kind.
See it on a demo
10,000 practice USDT on live prices. No card, no exchange key.