Trading bots: what the data says
Exchanges list every bot type with zero performance numbers. We replayed them out-of-sample with real fees and slippage. Below: each type, where it exists, and our verdict — including the negative ones and an honest “untested” where we have no study yet.
Grid bot
OOS: negativeSet a price range, let the bot buy low and sell high on every oscillation — marketed as steady income from sideways markets.
DCA / Martingale bot
96% wins, net-negativeAverage down automatically and exit the whole position on a small bounce — marketed with win rates above 90%.
Signal bot
−99.7% replayed honestlyAuto-execute trades from signal channels or TradingView alerts — marketed as copying the callers' skill without the screen time.
Funding-rate arbitrage bot
no proven edge (−14% OOS)Hold spot, short the perpetual, collect the funding payments — marketed as delta-neutral, market-independent yield.
Recurring buy (DCA savings)
fine as savings — measure vs holdingBuy a fixed amount on schedule regardless of price — marketed as the stress-free way to accumulate.
Iceberg / TWAP execution
execution tool, not alphaSlice large orders into small hidden chunks over time — marketed to whales who move markets by showing size.
Smart portfolio / rebalancer
not yet tested by usHold a basket, auto-rebalance to target weights — marketed as diversification on autopilot.
Method in one line: in-sample optimization is easy, so we only trust out-of-sample results with real fee and slippage models — and we publish the failures. The full pipeline is documented in the research journey. Nothing here is investment advice.