Automated trading system

An automated trading system (ATS), a subset of algorithmic trading, uses a computer program to create buy and sell orders and automatically submits the orders to a market center or exchange.¹ The computer program generates orders based on predefined rules using a trading strategy derived from technical analysis, statistical computation, or other electronic inputs.² Such systems are often used to implement algorithmic strategies that operate at high speed and frequency.

These systems are mostly employed by investment banks or hedge funds, but simple online tools have extended access to private investors. An estimated 70–80% of all market transactions are carried out through automated trading software, in contrast to manual trades.³⁴

Automated trading systems are often used with electronic trading in automated market centers, including electronic communication networks, "dark pools", and automated exchanges.⁵ They can execute repetitive tasks at speeds orders of magnitude greater than any human equivalent. Traditional risk controls that relied on human judgment are not appropriate for automated trading, contributing to events such as the 2010 Flash Crash. Regulators have introduced controls including trading curbs and circuit breakers in some electronic markets.⁶ High-frequency trading strategies have drawn regulatory scrutiny for their role in market disruption, including spoofing and layering, and the Financial Industry Regulatory Authority (FINRA) has established surveillance programs targeting abusive algorithmic activity.

The automated trading system determines whether an order should be submitted based on, for example, the current market price of an option and theoretical buy and sell prices.⁷ The theoretical buy and sell prices are derived from, among other things, the current market price of the security underlying the option. A look-up table stores a range of theoretical buy and sell prices for a given range of current market price of the underlying security. Accordingly, as the price of the underlying security changes, a new theoretical price may be indexed in the look-up table, thereby avoiding calculations that would otherwise slow automated trading decisions.⁸ A distributed processing on-line automated trading system uses structured messages to represent each stage in the negotiation between a market maker (quoter) and a potential buyer or seller (requestor).⁹

• Trend following¹⁰

Trend following is a trading strategy that bases buying and selling decisions on observable market trends. Various forms have emerged over decades, including the Turtle Trader software program. Unlike financial forecasting, this strategy does not predict market movements; it identifies a trend early in the day and trades automatically according to predefined rules regardless of directional shifts. Trend following gained popularity among speculators, though it requires judgment to configure initial trading rules and entry/exit conditions. Performance depends on market volatility and the difficulty of identifying trends early enough to profit from them.¹¹

One formulation models the stock price ${\displaystyle S_{r}}$ at time ${\displaystyle r}$ as a regime-switching diffusion process:

$${\displaystyle dS_{r}=S_{r}[\mu (\alpha _{r})\,dr+\sigma \,dB_{r}],\quad S_{t}=X,\quad t\leq r\leq T<\infty ,}$$

where ${\displaystyle \alpha _{r}\in \{1,2\}}$ is a two-state Markov chain, ${\displaystyle \mu (i)\equiv \mu _{i}}$ is the expected return rate in regime ${\displaystyle i}$, ${\displaystyle \sigma >0}$ is constant volatility, and ${\displaystyle B_{r}}$ is a standard Brownian motion.¹²

• Volume-weighted average price

The volume-weighted average price (VWAP) weights each trade price by its quantity over a defined period:

$${\displaystyle P_{\mathrm {VWAP} }={\frac {\sum _{j}{P_{j}\cdot Q_{j}}}{\sum _{j}{Q_{j}}}},}$$

where ${\displaystyle P_{j}}$ is the price of trade ${\displaystyle j}$, ${\displaystyle Q_{j}}$ is its quantity, and the sum runs over all individual trades in the defined period, excluding cross trades and basket cross trades.

• Mean reversion (finance)

A continuous mean-reverting time series can be represented by an Ornstein–Uhlenbeck stochastic differential equation, commonly used to model price return to a long-run mean:

$${\displaystyle dx_{t}=\theta (\mu -x_{t})\,dt+\sigma \,dW_{t},}$$

where ${\displaystyle \theta }$ is the rate of reversion to the mean, ${\displaystyle \mu }$ is the mean value of the process, ${\displaystyle \sigma }$ is the variance of the process, and ${\displaystyle W_{t}}$ is a Wiener process.¹³¹⁴

Richard Donchian introduced rule-based trading in 1949, applying a set of mechanical buy and sell signals to futures funds.¹⁵ Because no automated execution technology existed at the time, staff monitored charts manually and assessed whether conditions met the rules before placing orders, a process prone to human error. It nonetheless established the principle of systematic, condition-triggered trading.¹⁶

In the 1980s, rule-based trend-following gained wider adoption among commodity traders, including John Henry, who used systematic strategies to manage futures portfolios. By the mid-1990s, several commercial strategy packages were available to institutional buyers, and declining hardware costs opened access to smaller firms.¹⁷

Early automated systems ran as portfolio-management software operated by brokers on behalf of clients. The first direct-to-consumer automated investing service launched publicly in 2010 with Betterment, founded by Jon Stein. Around 2005, copy trading and mirror trading emerged as forms of automated algorithmic trading. These systems allowed traders to share their trading histories and strategies, which other traders could replicate in their accounts. One of the first companies to offer an auto-trading platform was Tradency in 2005 with its "Mirror Trader" software.¹⁸¹⁹²⁰ This feature enabled traders to submit their strategies, allowing other users to replicate any trades produced by those strategies in their accounts. Later platforms allowed traders to connect their accounts directly to replicate trades automatically, without coding trading strategies themselves. Since 2010, numerous online brokers have incorporated copy trading into their internet platforms, such as eToro, ZuluTrade, Ayondo, and Tradeo.²¹²² Copy trading allows less experienced traders to mirror positions taken by other investors without performing their own analysis.

By 2014, more than 75% of shares traded on United States exchanges, including the New York Stock Exchange and NASDAQ, originated from automated trading system orders.²³

Automated trading, or high-frequency trading, causes regulatory concerns as a contributor to market fragility.²⁴ United States regulators have published releases²⁵²⁶ discussing risk controls to limit disruptions, including financial and regulatory controls to prevent erroneous orders caused by computer malfunction or human error, regulatory breaches, and credit or capital limit overruns.

The use of high-frequency trading (HFT) strategies has grown and by the mid-2010s accounted for a majority of order flow on U.S. equity markets. Although many HFT strategies are legitimate, some are used for manipulative trading. A strategy is illegitimate or illegal if it causes deliberate disruption or attempts to manipulate the market. Such strategies include "momentum ignition strategies": spoofing and layering, where a participant places a non-bona fide order on one side of the market to bait other participants into reacting, then trades on the other side. The Financial Industry Regulatory Authority (FINRA) has reminded firms using HFT strategies of their obligation to test these strategies pre- and post-launch to prevent abusive trading.

FINRA also focuses on the entry of problematic HFT and algorithmic activity through sponsored participants who initiate their activity from outside of the United States. In this regard, FINRA reminds firms of their surveillance and control obligations under the SEC's Market Access Rule and Notice to Members 04-66,²⁷ as well as potential issues related to treating such accounts as customer accounts, anti-money laundering, and margin levels as highlighted in Regulatory Notice 10-18²⁸ and the SEC's Office of Compliance Inspections and Examination's National Exam Risk Alert dated September 29, 2011.²⁹

FINRA conducts surveillance to identify cross-market and cross-product manipulation of the price of underlying equity securities. Such manipulations are typically done through abusive trading algorithms or strategies that close out pre-existing option positions at favorable prices or establish new option positions at advantageous prices.

Several algorithmic trading malfunctions have caused market disruptions large enough to draw regulatory action. These raise concern about firms' ability to develop, implement, and supervise automated systems. FINRA has stated that it will assess whether firms' testing and controls related to algorithmic trading are adequate in light of SEC and firms' supervisory obligations. This assessment may take the form of examinations and targeted investigations. Firms will be required to address whether they conduct separate, independent pre-implementation testing of algorithms and trading systems. FINRA will review whether a firm actively monitors and reviews algorithms once they are placed into production systems, including procedures to detect potential trading abuses such as wash sales, marking, layering, and momentum ignition strategies. Firms will also need to describe their approach to firm-wide disconnect or "kill" switches, as well as procedures for responding to catastrophic system malfunctions.³⁰³¹³²

Two widely cited incidents illustrate the risks:

• On May 6, 2010, the Dow Jones Industrial Average declined about 1,000 points (about 9%) and recovered those losses within minutes. It was the second-largest point swing (1,010.14 points) and the largest one-day point decline (998.5 points) on an intraday basis in the Average's history. This market disruption, known as the Flash Crash, resulted in U.S. regulators issuing new regulations governing automated trading market access.
• On August 1, 2012, between 9:30 a.m. and 10:00 a.m. EDT, Knight Capital Group lost four times its 2011 net income.³³ A bug in one of Knight's trading algorithms submitted erroneous orders to exchanges for nearly 150 different stocks. Trading volumes soared across so many issues that the SPDR S&P 500 ETF (SPY), which is generally the most heavily traded U.S. security, became the 52nd-most traded stock on that day, according to Eric Hunsader, CEO of market data service Nanex. Knight shares closed down 62% as a result, and the firm ultimately merged with Getco.³⁴³⁵