Crypto algorithmic trading, defined
Crypto algorithmic trading uses explicit rules or models to generate and execute trading decisions in digital-asset markets. It is not the same topic as an “algorithmic cryptocurrency” or algorithmic stablecoin; here, the algorithm controls a trading process.
The algorithm may be as simple as periodic rebalancing or as complex as a portfolio model combining multiple features. In every case, the live result depends on market structure and the surrounding automated trading system.
Spot and perpetual markets behave differently
| Dimension | Spot algorithm | Perpetual/futures algorithm |
|---|---|---|
| Position | Owns or sells the asset. | Controls contract exposure. |
| Leverage | Usually none unless separately borrowed. | Often available; increases liquidation risk. |
| Carrying cost | Custody, borrow or opportunity cost. | Funding, basis and margin costs. |
| Primary failure | Slippage, inventory and custody. | Liquidation, funding and basis dislocation. |
| Exit constraint | Available market liquidity. | Liquidity plus margin state. |
Strategy families
- Scheduled allocation: rebalance or accumulate according to time and target weights.
- Trend and momentum: increase exposure when a measured direction persists.
- Mean reversion: trade the hypothesis that a relative displacement will normalize.
- Market making: quote both sides while controlling inventory and adverse selection.
- Arbitrage and basis: trade linked instruments when the net spread exceeds costs.
- Portfolio models: allocate across strategies or assets under a shared risk budget.
Each family requires a different cost model. Continue to trend following, AI scalping, our arbitrage-bot guide for spread-based strategies, or the futures-bot guide for leveraged derivatives.
A worked net-return example
Consider an illustrative strategy that reports a 1.20% gross gain over a set of trades. Suppose entry and exit trading fees total 0.20%, modeled spread and slippage total 0.28%, funding or borrow costs are 0.12%, and execution failures cost another 0.08%.
1.20% − 0.20% − 0.28% − 0.12% − 0.08% = 0.52% illustrative net result This is not a forecast. It demonstrates why a gross strategy edge must be materially larger than the uncertainty in the cost assumptions. If slippage doubles during volatility, the apparent edge can disappear.
Where a backtest usually lies
- It uses information before that information would have been available.
- It selects assets that survived while omitting those that disappeared.
- It assumes full fills at a convenient candle price.
- It ignores market impact, latency, funding, borrow and exchange limits.
- It tunes repeatedly on the same period until noise resembles a strategy.
- It omits outages, liquidations, key failures and operator mistakes.
Research-to-production gate
What to ask before using one
Ask for the exact asset universe, venue, timeframe, leverage, benchmark, fee model, maximum drawdown, test interval and live status. Ask who can change the strategy and whether position limits are enforced outside the model. A polished equity curve without those conditions is incomplete evidence.
Sources and scope
- CFTC — Trading bot customer advisory
- Investor.gov — AI and investment decisions
- Onchain Off Emotion — Backtests vs live results
The calculation is illustrative and excludes taxes and user-specific circumstances.