What a trading bot actually is
A bot is software that watches prices, applies rules, and can place or cancel orders without sitting on the screen all day. The intelligence is in the rules — not in a black box that “knows the market.”
- Signal: decide when a setup is valid (indicator, price level, time, or news filter).
- Risk: size the order, set a stop, cap daily loss, and refuse to trade when data is stale.
- Execution: send the order to a broker or exchange API, then log fills, slippage, and errors.
- Supervision: a person still owns the account, the keys, and the off switch.
Trend-following bots
These bots assume that a move already in progress can continue. They buy strength or sell weakness after a moving average, breakout, or higher-high structure confirms.
- Typical inputs: dual moving averages, Donchian channels, or ADX to skip choppy days.
- Strength: they can stay with a long run and avoid guessing bottoms.
- Weakness: they give back profit in ranges and need wide stops.
Mean-reversion and grid bots
Mean reversion bets that price returns toward an average. Grid bots ladder buy and sell orders around a range. Both shine when the tape oscillates — and suffer when a trend never looks back.
- RSI or Bollinger bands often flag “stretched” prices.
- Grid spacing must match volatility; too tight and fees eat the account.
- A hard stop or hedge is required when the range breaks.
DCA, market-making, and event bots
Dollar-cost averaging bots buy on a schedule. Market-making bots quote both sides for a spread. Event bots react to listings, funding, or scheduled data — and they fail when the feed is late.
- DCA reduces timing stress but does not remove drawdown.
- Market making needs inventory limits and a plan for one-way markets.
- News bots are only as good as latency, parsing, and the kill switch.
Backtest, paper, then live — with humility
A pretty equity curve on history is not a live edge. Overfit bots memorise the past. Fees, gaps, partial fills, and API outages never appear in a naive spreadsheet.
- Walk-forward or out-of-sample tests beat a single perfect curve.
- Paper trade the same venue you will use live.
- Start small. Log every override. Retire a bot that cannot survive a month of real noise.
