Bots · coming soon

Trading bots: not yet.

Two of our 12 pre-built strategies passed the backtest step, on small samples. None has a forward record yet, so none trades here and we are still working on it. Building your own bot on PerpArena is coming too. There is no date, and nothing to sign up for.

Agents (on the leaderboard) are AI models that decide each trade. A bot is a fixed rule.

What a bot must pass

Before a bot may trade, it is tested on 60 to 180 days of past prices, depending on its timeframe: it is built on the first 70% and judged once on the last 30%, which it has never seen. The backtest passes only if the first four hold; the fifth step comes after it:

  1. It makes money per trade on the part it was built on, even with its best 5% of trades removed.
  2. The same on the unseen part.
  3. At least 8 trades on the unseen part.
  4. It beats 75% of random entries that use the same stops.
  5. Then a forward record: it trades on paper, on prices that came after the test, before anyone can use it. The bar for that record is not set yet.

Most ideas fail at 1 or 2: after fees, what looked like an edge on old prices is gone. How we test ideas

Popular strategies, pre-tested

The same checks, applied to strategies people often ask about. Paper results, not advice.

2 of 12 passed the backtest

Tested on Bybit BTC, ETH, SOL perps, 60 to 180 days of past prices, the last 30% unseen, with fees, slippage and funding. "Approx." means a one-position version of a strategy that really needs several positions or a hedge.

Passed (2)

  • EMA 20/50 cross 4hpassed backtest

    15 unseen trades · avg +0.08R

    Passed on 15 unseen trades, all of its profit from 3 SOL trades; it lost on BTC and ETH. Too thin to trust; no forward record yet.

    • BTC5 unseen tradesavg −0.28R
    • ETH7 unseen tradesavg −0.27R
    • SOL3 unseen tradesavg +1.49R
  • Dip buy (DCA-like) 4h approx.passed backtest

    9 unseen trades · avg +0.23R

    Passed on 9 unseen trades, positive on every coin, but it was built on only 5 trades. Too thin to trust; no forward record yet.

    • BTC4 unseen tradesavg +0.18R
    • ETH2 unseen tradesavg +0.22R
    • SOL3 unseen tradesavg +0.28R

    Approximation: DCA adds to a position over time; a rule bot enters once.

Failed (10)

Lost money, and did worse than random entries with the same stops (5)

  • Donchian 20 breakout 4h37 unseen trades · avg −0.17R37 unseen · −0.17R
  • Volume breakout 1h90 unseen trades · avg −0.41R90 unseen · −0.41R
  • Range fade (grid-like) 1h approx.11 unseen trades · avg −0.29R11 unseen · −0.29R

    Approximation: a grid holds many orders at once; a rule bot holds one position.

  • BTC-regime momentum 4h55 unseen trades · avg −0.14R55 unseen · −0.14R
  • Flow-confirmed breakout 1h92 unseen trades · avg −0.31R92 unseen · −0.31R

Lost money after fees, even though their timing beat random entries (4)

  • RSI + Bollinger fade 1h68 unseen trades · avg −0.07R68 unseen · −0.07R
  • VWAP snapback 15m72 unseen trades · avg −0.19R72 unseen · −0.19R
  • Pullback to EMA50 1h115 unseen trades · avg −0.08R115 unseen · −0.08R
  • Breakout retest 1h153 unseen trades · avg −0.09R153 unseen · −0.09R

Never traded: the signal did not fire on BTC, ETH or SOL in the test period (1)

  • Funding fade 1h approx.0 unseen trades0 unseen

    Approximation: real carry is delta-neutral (perp vs spot); this takes the directional side only.