Richard Donchian and the first rules-based fund (1949)
Systematic trading starts with Richard Donchian, who in 1949 launched Futures, Inc., one of the first publicly held commodity funds run on predefined buy and sell rules instead of discretion. Algorithmic strategies are now commonly estimated at around 70% of US equity volume; that took seven decades of steps, beginning with one fund manager who trusted rules over intuition.
Donchian's channel breakout rules are simple: buy when price exceeds its recent high, sell when it drops below its recent low. Modern platforms still ship them as Donchian Channels.
Running a system then meant charting prices by hand from ticker tape: no computers, no spreadsheets, no live feeds, hours of plotting before a call to the broker.
Slow, but the principle held and still holds: take emotion out of the entry and exit. The CMT Association, which awards the Chartered Market Technician designation, credits Donchian as a founding figure of systematic method.
| Year | Step | What changed |
|---|---|---|
| 1949 | Donchian's Futures, Inc. | Rules replace discretion |
| 1983 | Turtle traders | Rules can be taught |
| 1990s | TradeStation on PCs | Retail backtesting and signals at home |
| 1997 | CME E-mini on Globex | Orders reach the exchange without a human |
| 1998 to 2001 | System-assist, Collective2 | Signals sold by subscription |
| 2005 to 2010 | Tradency, eToro CopyTrader | Mirror, copy and social trading |
| 2017 | Telegram channels | Signals without an intermediary |
| 2024 | LLM trading agents | Systems that adapt without new code |
The Turtle traders experiment (1983)
In 1983 commodities trader Richard Dennis bet his partner William Eckhardt that great traders could be made, not born, and set out to prove it. He recruited people with no trading background, gave them a fixed set of trend-following rules and allocated real capital.
Dennis won the bet. The Turtle Traders applied breakout signals, position sizing formulas and strict stop losses to commodity futures and made substantial returns, showing that a well-designed system applied consistently could beat discretionary judgment.
Around the same time, John W. Henry, later principal owner of the Boston Red Sox and Liverpool FC, built a large asset manager on mathematical trend-following models in futures, running billions with purely systematic methods.
Research later backed these methods. Moskowitz, Ooi and Pedersen's time-series momentum paper in the Journal of Financial Economics (2012) found consistent risk-adjusted returns from momentum across asset classes over long periods.
Personal computers bring trading systems to retail (1990s)
Affordable PCs and dedicated software turned systematic trading from an institutional craft into something individuals could run at home. On TradeStation, retail traders coded their own rules, backtested them on historical data and generated signals without a team of quants.
The decade also brought the first commercial systems: developers sold rule sets coded for TradeStation, and buyers ran someone else's method on their own account. The link was still personal; if the software broke, the developer took the call.
Commercial internet access in the late 1990s changed the speed. Live data feeds became affordable, systems ran on real-time prices instead of end-of-day calculations, and the delay between signal and execution, once measured in hours, began to collapse.
Electronic exchanges and CME Globex
The Chicago Mercantile Exchange completed the automation of execution when it launched E-mini futures in 1997, starting with the S&P 500 E-mini, built to trade electronically on Globex instead of the open-outcry floor.
With Globex, a computer could calculate where an order belonged and send it straight to the exchange with no human in the chain: signal, size and transmit faster than anyone could pick up a phone. This is when systematic trading became fully automated in the modern sense.
The effect on markets was measurable. Hendershott, Jones and Menkveld's study in the Journal of Finance (2011) found that algorithmic trading on the New York Stock Exchange came with better liquidity and narrower bid-ask spreads.
Trading signals by subscription (from 1998)
In parallel, the model most retail traders still use took shape: paying a recurring fee to follow someone else's signals instead of building a system.
In futures, the model is partly traced to Attain Capital, where client Jack Telford let other clients follow the signals of a TradeStation system he had built, for a modest fee. That formalised the system-assist model: a developer codes the strategy, subscribers receive the signals, and neither manages the other's infrastructure.
Collective2, launched in 2001, was an early platform that gathered many strategy developers and let subscribers route signals straight to a brokerage account. That middle layer separated signal generation from execution and took the developer out of the subscriber's daily routine.
Mirror, copy and social trading (from 2005)
Signal following turned social when mirror trading, pioneered in forex by firms including Tradency around 2005, let retail traders replicate vetted strategies automatically in real time. The trader picked a method and an allocation; the platform synced it with the brokerage.
eToro took it further in 2010 with CopyTrader, which replicated other individuals' positions rather than institutional strategies. Track records, drawdowns and open positions became visible before you committed capital, and by 2012 eToro reported more than two million registered users and had become synonymous with what the industry called social trading.
An MIT study of eToro by Pan, Altshuler and Pentland (2012) found that moderate social influence improved returns while excessive copying made behaviour uniform and the system more fragile. That tension hasn't gone away; our explainer on what social trading is covers the research since.
Telegram signal channels (from 2017)
From around 2017, Telegram became the default infrastructure for much of the crypto signal business, and later for execution itself through Telegram sniper bots. A provider opens a channel, subscribers join, and calls go out as messages, executed by hand, by bots reading the channel, or through API links to exchange accounts.
There is no Collective2 or eToro in the middle any more, and that cuts both ways. No standard performance verification, no regulated dispute process, no enforced track record: all you have to go on is the provider's reputation.
The model spread fast in crypto precisely because it needed no registration and scaled from one subscriber to a hundred thousand on the same channel. Our guide to Telegram signal group scams shows what that gap allows.
AI trading agents (from 2024)
The current shift replaces the human signal provider with an AI agent that reasons over data, adapts and executes without a fixed rule set. From Donchian's channels to Telegram bots, earlier systems did exactly what a developer coded; agents built on large language models also read news, earnings call transcripts, social sentiment and macro commentary next to price data. We compare the two in AI agents vs trading bots.
In late 2024, Xiao, Sun, Luo and Wang published TradingAgents on arXiv, a small trading desk staffed by models:
- analyst agents gather and interpret data;
- a risk agent watches exposure limits;
- a trader agent decides and executes.
They tested it against buy-and-hold, MACD and SMA strategies on equities from January to November 2024.
Agents are also active on prediction markets: on Polymarket, autonomous agents have executed thousands of trades in a single month, CoinDesk reported. The Olas protocol frames its aim as user-owned agent economies, where individuals deploy software that earns for them across markets.
The market for AI in trading is projected to grow from $21.59 billion in 2024 to $24.53 billion in 2025, 13.6% a year, according to Nasscom. Retail has access too: brokers such as Alpaca let traders connect LLM agents to live accounts with little code, the agent turning plain-language instructions into API calls.
What sets this generation apart is adaptability. A breakout system trades in 2024 exactly as it did when coded; an agent can change its behaviour as conditions change, without a developer.
That brings new failure modes: agents that adapt wrongly, overfit recent data, or interact unpredictably with other automated systems. The SEC and CFTC have both signalled interest in who is accountable when an autonomous agent disrupts a market.
The thread from Donchian's rules to AI agents
Across seventy-five years, every step solved the same problem: decide when to buy and when to sell, and act faster and more consistently than unaided judgment. Donchian's 1949 rules, Dennis's Turtle signals, a TradeStation system sold by subscription, a Telegram channel and a multi-agent LLM reading earnings transcripts differ in medium, not in purpose.
Each step also moved the trust question: from the fund manager, to the system's rules, to the platform's records, to a channel admin's reputation, and now to a model's judgment. Knowing where that trust sits is still the first check before you follow any system, human or machine.