Algorithmic trading on crypto exchanges means software that reads market data through an exchange's API and places, amends and cancels orders according to rules, without a human clicking. It ranges from simple rebalancing bots to market makers quoting thousands of orders a minute. The same knowledge matters to two audiences: people building trading systems, and exchanges that want to attract them.
How bots connect to exchanges
- REST APIs for placing and cancelling orders, querying balances and fetching history. Requests are signed with an API key and secret, usually via HMAC, and are rate-limited.
- WebSocket streams for real-time order book updates, trades and private order and balance events. Serious strategies run on streams, not polling.
- FIX is offered by some exchanges for institutional clients, mirroring traditional markets.
- On-chain access for DEXs: bots call router contracts or submit signed intents, and must manage gas, nonces and MEV exposure. Aggregation and routing logic is covered in the DEX aggregator guide.
Open-source libraries such as CCXT normalize many exchanges' APIs into one interface, which is handy for research and multi-venue tools, though latency-sensitive strategies usually talk to each venue's native API directly.
Common strategy families
| Strategy | What it does | Main risk |
|---|---|---|
| Market making | Quotes both sides and earns the spread and maker rebates | Inventory risk when price trends; adverse selection |
| Cross-exchange arbitrage | Buys where cheaper, sells where dearer | Transfer delays, withdrawal limits, fees eating the edge |
| Funding / basis trades | Holds spot against perpetual futures to collect funding | Liquidation, exchange counterparty risk |
| Execution algorithms (TWAP, VWAP) | Splits large orders to reduce market impact | Predictability if poorly randomized |
| Trend and momentum | Follows price signals over minutes to weeks | Overfitting, regime changes |
| Grid and DCA bots | Places a ladder of orders or buys at intervals | Large losses in strong one-way moves |
Architecture of a trading system
- Market-data handler. Maintains a local order book from snapshots plus incremental updates, validates sequence numbers, and resyncs on gaps.
- Strategy engine. Turns market state and positions into desired orders. Keep it pure and testable.
- Order management system. Tracks every order's life cycle, including partial fills, rejections and cancels that arrive after fills.
- Risk layer. Hard limits on position size, order rate, notional per minute and loss per day, plus a kill switch. It must sit between strategy and exchange and be able to veto anything.
- Persistence and monitoring. Trade records for reconciliation and tax, metrics and alerts for latency, rejects and P&L drift.
Backtesting traps
Most strategies that look great in backtests fail live. Typical reasons: look-ahead bias, ignoring fees and spread, assuming fills at the touch that would not really have happened, ignoring market impact, survivorship bias from only testing coins that still exist, and overfitting parameters to one period. Use tick or order-book data where possible, model queue position for limit orders, and keep an untouched out-of-sample period.
Exchange quirks that break bots
- Rate limits that differ per endpoint and per account tier, with temporary bans when exceeded.
- Minimum order sizes, tick sizes and lot sizes that differ per pair and change over time.
- Scheduled and unscheduled maintenance, delistings, and pairs going into cancel-only mode.
- WebSocket disconnects that silently drop messages unless you check sequence numbers.
- Clock skew causing signed requests to be rejected.
Security of API keys
Create trading keys without withdrawal permission, restrict them to fixed IP addresses, store secrets in a secrets manager rather than code, and rotate them. Many retail losses come from third-party bot services or browser extensions that were given keys with withdrawal rights.
What exchanges should offer algo traders
If you operate a venue, professional flow depends on stable, well-documented APIs, a matching engine that behaves predictably under load, a sandbox, sensible rate limits, maker-taker fee tiers, and a formal market-maker program. These are worth weighing when choosing a white-label exchange vendor or designing your own exchange software.
Increasingly, teams also experiment with AI agents that trade on-chain; the risks of giving autonomous software spending power are discussed in this piece on AI agents and smart contracts.
Frequently asked questions
Do I need low latency to trade crypto algorithmically?
Only for market making and short-term arbitrage. Execution algorithms, rebalancing and trend strategies work fine on ordinary cloud servers.
Is algorithmic crypto trading legal?
Generally yes, but manipulative practices are not, and some jurisdictions regulate those who manage others' money with algorithms. Check the rules where you operate.
Can I use one bot across many exchanges?
Yes, with an abstraction layer such as CCXT or your own adapters. Each venue's rules, fees and failure modes still need specific handling.
What is the most common way bots lose money?
Strategies overfit to past data, and missing risk limits that let a bug or a market shock run unchecked.