🤖 Does a trading bot work?
Check it before the money.
A backtest does not prove a bot will make money. We tested 1,605 variants of popular "buy the drop" strategies on 1-second candles over 90 days — after honest fees and fills, none passed selection. Only paper arbitrage between exchanges holds up, and we show why that is still not money.
🎁 What you get from FinEdg
Protection from pretty backtests
We show rejected strategies too, with the reason — you see what was tested and how it ended.
Risk-free paper trading
Run a strategy on live prices with no real orders and no money before you risk any.
Arbitrage, explained honestly
How cross-exchange gaps work, what fees take, and why the on-screen spread is bigger than the executable one.
A MoonBot-style coin screener
Top 10/30/50, memecoins, new listings; how often a coin drops and rebounds. Closed test for now.
Coin write-ups
Alive or faded, who took its niche, possible scenarios — the AI analyst's hypothesis kept apart from facts.
Plain-language learning
Backtest, forward test, basis points — and why "90% winning trades" can still lose money.
🔬 How to check a bot: 5 steps
- Use fine-grained data. A strategy that catches a 30-second drop "sees" things that never happened on 5-minute candles. We use 1-second candles.
- Fill honestly. A limit order counts as filled only if price traded through it and stayed at least 2 seconds. Fees on every trade.
- Split the history. Tune on the first 70% of history, judge on the last 30% that tuning never saw.
- Paper-trade forward. Frozen settings run on the live market without money. Only trades after the freeze count as evidence.
- Decide the kill rule first. Write down when a strategy is dropped before it starts, and keep the losers visible.
2.5 ÷ (1 + 2.5) ≈ 71%
📊 What failed and what holds up
| Strategy | How we tested | Result | Verdict |
|---|---|---|---|
| Buy a 30-second drop (SCALPEMA) | 90 days, 67 coins, 1-second candles | negative on altcoins in both tuning and check periods | culled |
| Drop grid (MoonBot DropsDetection), with and without a 5% stop | 90 days, 27 grid variants | +0.59% per trade in September, negative in July–August; one cascade day −$2,557 | culled |
| Anti-dump scheme (limit −1%, target +1%, stop −2.5%) | Bybit, 180 days, 86 coins | negative every month: 66% winners vs 71% needed | do not port |
| Cross-exchange perp-perp arbitrage, entry ≥ 0.5% | paper book on real order books | 1,766 trades, +0.58% average after fees | holds on paper |
⚠️ Downsides: when this does NOT work
Paper is not money.
The arbitrage result is paper. Executing both legs live at once is not proven yet.
Order-book depth.
On small coins the book holds about $50 per leg; more size eats the whole spread.
The on-screen spread misleads.
By last prices the gap averages 0.68%, the executable one about 0.08%.
Short history is one regime.
28 good days turned negative over 90 days. Test any "perfect" strategy on long history.
❓ Questions before your next step
Can a backtest prove that a bot makes money?
No. A backtest can reject a bad idea. Only a later period unused for tuning and a forward paper run give evidence, and even that is not a guarantee.
Why test on 1-second data?
Strategies that buy fast drops live on 30-second windows. On 5-minute candles those drops are invisible, so the result is fiction.
Is the arbitrage result real money?
No, it is a paper book with real order-book prices and fees. Live execution of both legs and order-book depth are not proven yet.
Does FinEdg place orders for me?
No. FinEdg is research and paper trading. There is no custody and no order placement from these pages.
🚀 Where to start
Everything below is open without an account.
📖 Glossary
- Backtest
- running a strategy over past prices.
- Forward test
- trading frozen settings on prices that did not exist when they were tuned.
- Basis point
- 0.01%. +58 bps = +0.58% of the trade size.
- Drop
- a sharp price fall within seconds; drop-buyers buy the dip and sell the rebound.
- Perp-perp arbitrage
- buying a perpetual future on one exchange and selling it on another at the same time, earning when prices converge.
Paper arbitrage: FinEdg paper book, closed trades with entry ≥ 0.5%, read 2026-09-30. · 1,605 strategies: Binance spot 1-second candles 2026-06-30…09-27, 67 coins, 0.05% fee per side. · Anti-dump scheme: 1-second bars from Bybit's public trade archive, 2026-04-01…09-27, Bybit VIP0 fees. · On-screen vs executable spread: FinEdg measurement 2026-09-17, 102 pairs.
Not financial advice. Past and paper results do not guarantee future ones. Crypto assets can lose all their value. Risk disclosure