Algorithmic trading did not begin with artificial intelligence or computers racing through orders in microseconds. Its history is better understood as a transfer of tasks: displaying prices, transmitting orders, deciding how to execute them, and eventually making trading decisions with little human intervention.
Those tasks are not interchangeable. An electronic exchange can match orders submitted by people. An execution algorithm can spread a pension fund’s purchase across an afternoon. A high-frequency trading firm can respond to a price change before a human trader notices it. Each uses computers, but for different purposes.
Within the broader history of trading technology, algorithmic trading marks the point where machines became more than communication tools. They became participants in the decision process. The resulting history combines cheaper execution and faster price adjustment with a recurring problem: individually sensible instructions can interact badly.
Algorithmic Trading and High-Frequency Trading Are Different
Algorithmic trading uses computer instructions to determine aspects of an order, such as its timing, size, price or execution venue. The instructions can be simple. A system that buys a fixed quantity every ten minutes does not need to predict the economy, interpret a speech or possess anything resembling intelligence.
High-frequency trading, usually shortened to HFT, is a narrower category. Speed is central to its operation, rather than just a convenience. The distinction matters because a slow execution schedule and a rapid market-making strategy can have very different effects.
| Activity | Decision being automated | Illustrative use |
|---|---|---|
| Electronic order entry | None necessarily; a person can choose the trade | A trader submits a limit order through a screen |
| Algorithmic execution | How to complete an intended trade | A fund divides a large purchase into smaller orders |
| Systematic trading | Whether and what to trade | A model trades when a price relationship crosses a threshold |
| High-frequency trading | Rapid trading and order management | A firm updates quotations as market prices change |
These categories can overlap. A systematic investment strategy might hold assets for months while using an execution algorithm to establish its positions. Calling both that strategy and HFT “computer trading” is accurate, but not especially useful.
The 1970s: Computerised Quotes Before Computerised Decisions
Nasdaq launched its automated quotation system on February 8, 1971. It connected competing market makers and distributed their buying and selling prices electronically. That was an important foundation, but the original quotation system did not automatically complete the entire trading process. Dealers still arranged trades by telephone.
The distinction between seeing a price and executing against it is central to this period. A screen could improve access to information without removing the human negotiation between counterparties. During the 1980s, systems including Nasdaq’s Small Order Execution System and SelectNet moved more of that interaction onto computers. These stages appear in Nasdaq’s history of quotation and execution technology.
The conceptual breakthrough was not that a computer could display a number. It was that market information could become an input to a repeatable process. Once prices and orders were available electronically, software could compare them and act on predefined conditions.
The 1980s: Program Trading and Portfolio Insurance
By the 1980s, program trading allowed institutions to coordinate transactions across baskets of stocks. The New York Stock Exchange’s Designated Order Turnaround system helped route large numbers of orders to trading posts. This was an ancestor of automated execution, not a modern HFT platform.
Two strategies became particularly associated with the period. Index arbitrage traded differences between stock baskets and related futures. Portfolio insurance used models to adjust exposure as markets moved, often selling stock index futures when prices fell.
During the crash of October 19, 1987, these strategies attracted scrutiny. Portfolio insurance could generate additional selling into a falling market, but its contribution remains debated. Other sellers, funding pressures and unreliable price information also mattered. The Federal Reserve’s study of the 1987 crash documents both the feedback mechanism and disagreements over its importance.
The lesson for automation was uncomfortable: a rule intended to protect one portfolio could contribute to pressure when many institutions followed it together. A strategy’s behaviour in isolation was not enough to establish how it would behave in a crowded market.
The 1990s and Early 2000s: Conditions for Automated Markets
Electronic communication networks, or ECNs, expanded the possibilities for matching orders outside traditional exchange arrangements. Meanwhile, smaller price increments changed the economics of quoting. The US move to decimal stock pricing was completed in 2001, replacing fractional quotations with much finer increments.
These developments helped create the conditions in which HFT grew. Smaller price increments reduced the room between adjacent quotations, while competing electronic venues created more prices and order books to monitor. The Congressional Research Service history of high-frequency trading examines decimalisation, electronic markets and regulatory changes as contributors rather than identifying a single invention date.
US Regulation NMS added another structural change. Adopted in 2005, its Order Protection Rule required trading centres to maintain procedures designed to prevent executions at prices worse than protected quotations elsewhere, subject to exceptions. The SEC’s original Regulation NMS release sets out that framework.
The practical implication was that execution involved more than choosing when to trade. It could also require comparing venues, assessing available quantities and deciding where to send each order. Automation suited that repeated comparison task.
This institutional development was related to, but different from, the arrival of electronic exchanges and online stockbroking. Giving an investor a trading screen did not give that investor the infrastructure or access arrangements of a professional automated trading operation.
Execution Algorithms: Automation Without a Speed Race
A useful way to understand execution algorithms is to separate the investment decision from its implementation. Suppose a fund decides to buy 120,000 shares. That decision says nothing about whether it should purchase them immediately, wait for sellers or divide the order throughout the session.
In a simplified example, a time-weighted schedule could divide the purchase into twelve equal portions. Each portion would contain 10,000 shares. The algorithm’s purpose would be orderly execution, not beating competitors by a fraction of a second.
A volume-based approach would behave differently. It could allocate more buying to periods when the market trades more actively. A participation strategy might aim to represent a chosen proportion of observed volume.
Each choice involves a trade-off. Executing quickly reduces the time during which the market can move away, but may consume available offers and push the purchase price higher. Executing slowly reduces immediate pressure but leaves more of the order exposed to later price changes.
There is no universal best schedule. The relevant question is whether the instructions fit the order’s urgency, the available liquidity and the investor’s tolerance for unfinished execution.
The 2000s: Speed Became a Business Model
HFT developed around opportunities that could disappear before slower participants reacted. Its strategies included rapid arbitrage and liquidity provision, rather than one uniform method of predicting prices.
By 2011, the Bank for International Settlements documented HFT’s growing presence in foreign exchange. Electronic interdealer trading in the 1990s, followed by broader customer access in the 2000s, had opened routes for new participants. The same report described firms locating servers close to trading venues to reduce communication delays: the practice known as co-location.
The BIS study of HFT in foreign exchange also distinguished automated execution from automated trading decisions. That distinction prevents an important historical mistake: treating every increase in algorithm use as an increase in speed-based speculation.
Consider two related instruments whose prices normally move together. If one changes first, a trader may try to transact in the other before its quotation adjusts. In that example, processing speed matters because the opportunity depends on an outdated price remaining available.
Fast equipment does not make that transaction risk-free. The quotation may disappear, only one side of the intended trade may execute, or the assumed relationship may fail. Faster mistakes are still mistakes.
May 6, 2010: The Flash Crash Exposed Interactions
The Flash Crash made the interaction between execution algorithms, HFT and market liquidity a public issue. On May 6, 2010, US equity and futures markets experienced a sharp disruption followed by a rapid recovery in many prices.
The joint regulatory investigation examined a large seller’s program to sell 75,000 E-mini S&P 500 futures contracts. Its algorithm targeted 9% of recently observed trading volume without taking price or time into account. In already stressed conditions, it completed the program in about twenty minutes.
The joint SEC and CFTC report on the Flash Crash describes how automated selling, rapid turnover among intermediaries and declining liquidity interacted. High trading volume did not necessarily indicate a large supply of investors willing to hold the positions being sold.
This was not a case of one all-powerful computer deciding that stocks were worth less. It showed how different systems, each responding to its own inputs and constraints, could collectively produce disorderly prices. Volume was a particularly dangerous substitute for measuring the market’s capacity to absorb risk.
The 2010s and 2020s: From Faster Systems to Better Controls
European supervision moved from early guidance into a broader statutory framework. ESMA adopted automated trading systems and controls guidelines in December 2011. After MiFID II began applying on January 3, 2018, their subject matter was incorporated into the newer framework. ESMA subsequently announced the withdrawal of those earlier automated trading guidelines.
The distinction between strategy risk and operational risk became increasingly useful. Strategy risk asks whether the trading idea will lose money. Operational risk asks whether the system will send the orders its designers intended. A profitable model cannot compensate for a deployment error or an ineffective order limit.
On February 26, 2026, ESMA issued further supervisory guidance covering governance, testing, outsourcing and controls applied before orders enter markets. The 2026 ESMA supervisory briefing on algorithmic trading is non-binding guidance, not a new law. Its focus shows how supervision has extended beyond the trading model to the organisation responsible for operating it.
What the Evidence Says About the Speed Race
The benefits of automating execution should not be confused with the benefits of every speed advantage. Software can reduce the work required to compare quotations and manage orders. Whether another tiny reduction in reaction time improves the market is a separate question.
Research published in January 2020 examined exchange messages, including unsuccessful attempts to trade or cancel orders. It found frequent races to transact against, or withdraw, quotations before competitors could respond. Individual races were small, but their aggregate effects mattered. The FCA-hosted research on latency arbitrage concerns that particular activity, not a verdict on all algorithmic trading.
A separate BIS study published in 2020 found that foreign exchange execution algorithms could improve matching between liquidity providers and users. It also identified a transfer of execution risk from dealers to users, alongside concerns about visibility and feedback effects. Those trade-offs are detailed in the BIS report on FX execution algorithms and market functioning.
The Lasting Change: Instructions Became Part of Market Structure
The most useful historical distinction is not between human and machine trading. It is between different decisions delegated to machines: distributing information, routing orders, managing execution, providing quotations and taking investment risk.
For an investor assessing an automated service, the practical questions follow directly. What does the system decide? What happens when liquidity disappears? Who can stop it? How are incomplete trades and rejected orders handled?
Algorithmic trading changed the speed and scale at which instructions could influence prices. It did not remove the need to choose those instructions carefully, or the need for someone to remain responsible when they produce an unwanted result.