RealPnL
DEMO Binance account / since start -0.46% / 4 open / last trade: TRIM SOL long · -0.10% 9 h ago / updated 09:25 UTC

Research ·

Why the bot holds four coins, not ten

Holding only the strongest 4 coins, one new one per cycle, matched all-10 on returns with far fewer orders. Without a per-coin cap, one coin once reached 101.6% of the account.

The first version of the bot held all ten coins at once. On its first day it opened nine positions in one go. That's fine for one account, but it's bad for copy trading: every order the lead opens, every copier opens too, so nine orders at once locks up a copier's margin nine times.

So the question was: can the bot hold only the coins with the clearest trend, and open them one at a time, without giving up the edge?

What we tested

144 variations on six years of 4-hour data: hold at most 2, 3, 4 or 5 coins; enter at a score of 0.5, 0.7 or 0.85; exit below 0.2 or 0.4; open at most 1, 2 or any number of new positions per cycle; and either keep each coin's normal size or scale the held coins up to use the same total exposure.

What came out

2021 – Sep 2026All 10 coins4 coins, 1 new per cycle
Return per year42.5%47.0%
Sharpe ratio1.631.71
Worst drop−19.7%−18.5%
Most new positions in one cycle91
Most orders in one cycle125
Peak margin in use (5×)33.4%20.5%
2021: +96%+96%2021: +121%+121%20212022: +18%+18%2022: +36%+36%20222023: +59%+59%2023: +56%+56%20232024: +45%+45%2024: +43%+43%20242025: +12%+12%2025: +11%+11%20252026*: +28%+28%2026*: +23%+23%2026*
All 10 coins at onceStrongest 4, one new per cycle

* 2026 runs to 27 September.

Most variations did at least as well as holding everything: 112 of 144 had a better risk-adjusted return, and 135 of 144 had a smaller worst drop. Nearby settings gave similar results, so this isn't one lucky combination.

The part that almost went wrong

Scaling the held coins up to use the same total exposure looked best on paper. Then we checked the largest single position it ever held: 101.6% of the account in one coin. A 25% stop-loss on that costs a quarter of the account in one move, and every copier with it.

So every coin is capped at 25% of the account. That cost about 5 points a year in the backtest (52.7% to 47.0%) and it's not negotiable: one bad day for one coin should never be a disaster.

What we did not find

We split the history in two, picked the best rules using only 2021–2023, and checked them on 2024–2026. They did not beat holding all ten in the later period (Sharpe 1.06–1.21 against 1.23). So holding fewer coins doesn't reliably make more money. What it reliably does is make fewer orders, use less margin and keep each position a sensible size. For copy trading, that's the point.

The rule now: at most 4 coins, at most 1 new one per cycle, enter at 0.70, exit below 0.40, never more than 25% in one coin.