RealPnL
DEMO Binance account / since start -0.46% / 4 open / last trade: TRIM SOL long · -0.10% 8 h ago / updated 08:55 UTC

Research ·

We could cut 39% of our orders. We decided not to.

Resizing less often cut orders from 687 to 420 a year with the same long-run return, but recent years were weaker. Wider still looked great, but only because of 2021.

Half of the demo account's first twelve trades were small trims: the bot shrinking a position to make room for a new coin. Each one is an order a copier has to take too. So we asked: how much of this resizing is actually useful?

The bot only resizes a position when its target differs from what it holds by more than a threshold, currently 30%. We tested 30%, 50% and 75%.

Results, January 2021 to September 2026

Resize when the target differs byReturn / yearSharpeWorst dropOrders / yearTrims / year2024–26 return
30% (live)47.0%1.71−18.5%68723027.6%
50%47.2%1.64−19.8%4207525.4%
75%55.8%1.70−21.8%3131125.0%
2021: +121%+121%2021: +138%+138%2021: +202%+202%20212022: +36%+36%2022: +30%+30%2022: +42%+42%20222023: +56%+56%2023: +60%+60%2023: +62%+62%20232024: +43%+43%2024: +38%+38%2024: +53%+53%20242025: +11%+11%2025: +8%+8%2025: -4%-4%20252026*: +23%+23%2026*: +25%+25%2026*: +26%+26%2026*
Resize at 30% (live)Resize at 50%Resize at 75%

* 2026 runs to 27 September.

The tempting option

75% looks best: 9 more points a year and almost no trims. But look at the years. The whole gain is 2021: +202% against +121%. In 2025 it lost money (−4.5%) while the live setting made +10.6%. And because positions grow unchecked, single coins reached 35–45% of the account, beyond the 25% cap we set so one stop-loss can't do serious damage. We rejected it.

The reasonable option

50% is a genuine trade-off: 39% fewer orders and two-thirds fewer trims, with the same long-run return, but about 2 points a year less from 2024 on and a slightly deeper worst drop. We weighed it and kept 30%. For now we'd rather keep the stronger recent record than cut orders.

We'll revisit this with live data. If real trims turn out to cost copiers more than the backtest assumes, the answer may change, and if it does, it will be written up here.