Definition
Algorithmic trading is the practice of expressing a market hypothesis as an explicit set of rules and delegating their execution to software that monitors data feeds, generates signals, and routes orders without discretionary human input on each trade. The discipline spans everything from a single retail trader running a moving-average crossover on a weekly bar to global market-makers operating thousands of co-located strategies with sub-millisecond latencies.
The modern formulation emerged from three converging shifts: the electronification of exchanges in the 1990s and 2000s, the commoditisation of historical data and compute, and the migration of financial research from spreadsheets into Python, pandas, and open-source scientific stacks. What was once the domain of bank prop desks is now reachable from a laptop running a broker API.
Why it matters
How it works
A production algorithmic trading system has four loosely-coupled layers. The data layer ingests, cleans, and stores market and reference data — bars, ticks, fundamentals, alternative datasets. The research layer is where alpha factors are constructed, combined, and tested historically against that data. The execution layer translates target positions into orders, manages slippage, and handles partial fills, retries, and venue selection. The risk layer sits across all of them, enforcing position limits, drawdown stops, and exposure caps in real time.
The hardest engineering challenge is not writing a profitable signal but preventing the gap between research and live deployment from corrupting the result. Look-ahead bias, survivorship bias in the data, slippage that was not modelled, and infrastructure latency that was assumed to be zero are the standard ways a backtested edge evaporates in production. The discipline of algorithmic trading is largely the discipline of closing those gaps — through event-based backtesters that replay the past one tick at a time, through walk-forward validation, through paper trading before capital is risked, and through ruthless post-mortems on every divergence between expected and realised P&L.