Algorithmic trading
Algorithmic trading is the use of software to decide and place trades in financial markets, replacing discretionary human judgment with codified rules, models, or learned policies algorithmic-trading-wikipedia. It is already the norm in established markets — roughly half of US equity volume and ~80%+ of FX orders are algorithmic algorithmic-trading-wikipedia — so this wiki’s crypto-centric founding sources are a recent entrant to a long-running field. A trading system spans three layers:
- Predict — forecast where the market is going. A model like Kronos consumes price history (OHLCV) and outputs a forecast.
- Decide (strategy) — turn information into a position. This is either a hand-coded rule set (the strategies a bot like banbot runs) or a learned policy (a reinforcement-learning agent like TensorTrade).
- Execute — place and manage the orders against a live exchange, sizing positions and respecting account state. An event-driven engine like banbot owns this layer.
The same software is validated before it risks money through backtesting — replaying historical data — and the hardest correctness problem in that replay is avoiding lookahead bias (the event-driven design is the standard guard).
Tuning and validation
Strategies and models carry parameters, so algorithmic trading leans on
hyperparameter optimization — banbot ships Bayesian/TPE
(model-based) and CMA-ES (model-free)
search to tune a strategy. Those optimizers are documented as subjects in
../optimization-algorithms-wiki; here they are a means, which marks this wiki’s scope boundary: the
trading system is the subject, the optimizer is a tool it uses.
The supporting layers the three-layer stack leaves out
A field catalog of ~97 libraries awesome-systematic-trading sorts the software into more categories than predict / decide / execute: alongside backtest engines and bots sit portfolio optimization (PyPortfolioOpt, Riskfolio-Lib), derivative pricing (tf-quant-finance, FinancePy, QuantLib wrappers), risk analytics (pyfolio), indicator and metrics libraries, broker APIs, and market-data plumbing. The three layers are what a strategy does; these are what a desk needs around it, and they mark where this wiki’s coverage stops.
The strategy layer’s playbook
The decision layer implements one of a known set of strategy families — arbitrage, pairs / statistical arbitrage, pairs / statistical arbitrage, mean reversion, scalping, market-making, momentum / trend following, and event arbitrage algorithmic-trading-wikipedia — whether hand-coded in a bot or learned by an RL agent. market-making is the odd one out: it is a standing obligation to quote both sides rather than a directional view, and it is the strategy the others trade against — a bot sending a market order is buying from a market maker.
Who runs it
The systems this wiki pages are open-source projects, a university lab, and individuals. The firms with the real books — prop shops, quant funds, bank desks — publish nothing, and reach this corpus only through their recruiting: quant-trading-firm and the quant-interview that filters for it.
chan-algorithmic-trading sorts the same space into two camps rather than seven families — mean reversion and momentum — and treats the rest as instances. The taxonomies aren’t in conflict; one lists what desks run, the other names the bet underneath each.
Related
algorithmic-trading-wikipedia · mean-reversion · chan-algorithmic-trading · market-making · quant-trading-firm · quant-interview · backtesting · event-driven-trading · vectorized-backtesting · awesome-systematic-trading · reinforcement-learning-trading · financial-time-series-foundation-model · banbot · tensortrade · kronos-financial-foundation-model · synthesis