Algorithmic Trading Platform

An algorithmic trading platform is software that allows a trader or firm to write trading rules as a program, test them on past data, and run them so that orders are created and sent to a market without manual entry. It is the working environment for algorithmic trading, and it joins three things that were once separate: market data, strategy logic and order handling.

Platforms range from retail charting packages with a built-in scripting language to systems built in-house by banks, hedge funds and proprietary trading firms. They sit between the trader and the broker or exchange. A platform does not supply a profitable strategy; it supplies the means to express one, measure it and operate it under control.

Most platforms share the same building blocks, whatever their size.

• Market data handling. The platform receives live prices and stores history. A market data adapter converts each venue's feed into one internal format, and a tick database keeps the record for research.
• Strategy engine. This runs the user's rules against incoming data and decides when to buy, sell or do nothing. The rules are a trading strategy written in a language the platform supports.
• Order management. Orders are created, sent, amended and cancelled, and fills are tracked against positions. The connection to the venue is commonly an application programming interface offered by a broker or a standard messaging protocol.
• Risk checks. A risk management system tests every order against limits on size, price, position and loss before it leaves.
• Monitoring. Dashboards, logs and alerts show what the strategy is doing and allow it to be stopped.

Retail platforms bundle charts, a scripting language and a link to one or more brokers. MetaTrader 5, used mainly for foreign exchange and contracts for difference, runs automated programs written in its own language, while Pine Script is the language of the TradingView charting service. Many brokers also publish interfaces that let customers connect programs written in general-purpose languages.

Open-source frameworks give more freedom and require more programming. Institutional systems are larger: they connect to many venues, support direct market access, and contain libraries of execution algorithms. Firms that depend on speed build their own platforms and place them in exchange data centers through colocation, because general-purpose software is too slow for low latency strategies.

Before a strategy trades, the platform is used to simulate it on historical data, a process called backtesting. A sound simulator charges realistic commissions and slippage, fills orders only at prices that were available, and shows the strategy nothing it could not have known at the time. Failing the last test is known as look-ahead bias.

The same code should run in testing and in production. When the two differ, results seen in research may not carry over. Many platforms offer a middle step, paper trading, in which the strategy runs on live prices with simulated money. Because a platform makes it easy to try thousands of rule variations, it also makes overfitting easy, and good practice keeps part of the data untouched for a final check.

An automated strategy can send faulty orders far faster than a person can react. Platforms therefore include limits on order size and price, caps on the number of orders per second, checks for repeated orders, and a switch that cancels everything and halts trading. Brokers and exchanges apply their own controls on top.

Regulators set expectations for these algorithmic trading controls. In the European Union, MiFID II requires firms that trade by algorithm to test their systems, keep records of orders, and maintain effective controls and business continuity arrangements. In the United States the SEC's market access rule requires brokers to apply pre-trade risk checks to orders sent under their name. Exchanges in several countries, including India, require algorithms to be approved or registered before use.

A platform lowers the cost of automation but does not remove the difficulties of trading. Data may contain errors or gaps. Simulated fills are an estimate, and the gap is widest for large orders and thinly traded instruments. Software faults, lost connections and broker outages can leave a strategy holding positions it cannot manage, so unattended operation carries real operational risk.

There is also a dependence on the vendor. A strategy written in a platform's own language is hard to move elsewhere, and a retail platform's speed and order types are set by the broker behind it. For these reasons larger firms treat the platform as core infrastructure and build or heavily customize their own.