top of page

Algorithmic trading using Artificial intelligence

Keywords: AI, algorithmic trading, trading strategies, financial markets.



The undeniable motivation for algorithmic trading is the remarkable sequence of breakthroughs in quantitative finance that began in the fifties with portfolio optimisation theory. The second relevant milestone was the development of the Capital Asset Pricing Model (CAPM), the third milestone occurred in the 1970s and was entirely statistical and computational. The fourth milestone came in 1973 with the publication of the Black and Scholes (1973) and Merton (1973) articles on the pricing of options and other derivative securities [1].

Algorithmic trading

Algorithmic trading is a valuable topic in the financial market and has been widely discussed in modern artificial intelligence. For both institutional investors and individual investors, there is a strong demand in exploring autonomous trading algorithms that are adaptable to the dynamic trading market. This article will discuss how algorithmic trading works using Artificial intelligence technology and discusses the top five trading strategies adopted in algorithmic trading.

Algorithmic trading refers to the utilization of algorithms to automate one or more aspect of the trading process:

Figure 1: Anatomy of the trading process [2]

  • Pre-trade analysis (data analysis) : this is the most common use of algorithms within a trading environment. It encompasses any system that utilises financial data or news to analyse certain properties of an asset. This first process comprises three main components: the alpha model, designed to predict the future behaviour of the financial instruments that the algorithmic system is intended to trade; the risk model, which evaluates the degree of exposure associated with the traded instruments; and the transaction cost model, which calculates any potential costs of trading[2].

  • Trading signal generation : buy and sell recommendations based on charts, indicators, technical analysis, or stock essentials. The first element of an algorithmic trading system is the Portfolio Construction Model (taking as inputs the results of the Alpha Model, Risk Model and Transaction Cost Model) and ideally selects the top portfolio of instruments they wish to hold/trade. This involves attempting to maximise potential profits, while limiting risk and the costs associated with the trades [2].

  • Trade execution : is further divided into agency/broker execution (when a system optimises the execution of a trade on behalf of a client) and principal/proprietary trading (where an institution trades on its own account).

Figure 2: Components of an Algorithmic Trading System [2]

Each stage of this trading process can be either managed by traders, by algorithms and traders, or fully by algorithms. Higher frequency and lower latency while trading are the difference between algorithmic traders and human traders. Frequency refers to how long it takes the trader to form a decision after the last decision made. Latency refers to the time delay between a trade is placed and its execution. Therefore, the algorithmic traders can both make trade decisions more often and execute their trades faster. Additionally, algorithmic traders lack human judgement [3].

How algorithmic trading works using artificial intelligence?

Financial trading has for an extended time been dominated by highly sophisticated forms of data processing and computation in the dominance of the “quants”. Yet over the last 20 years high-frequency trading (HFT), as a sort of automated, algorithmic trading focused on speed and volume instead of smartness, has dominated the arms race in financial markets [4].

Machine learning and AI are today changing the cognitive parameters of this race, shifting the boundaries between “dumb” algorithms in high-frequency trading (HFT) and “smart” algorithms in other sorts of algorithmic trading [4].

Whereas HFT is essentially focused on data internal and dynamics endemic to financial markets, new sorts of algorithmic trading enabled by AI are expanding the ecology of financial markets through ways in which automated trading draws on an extensive range of data such as social data for analytics (sentiment analysis). A report argues that whereas HFT is “about speed, machine learning is about depth and breadth of insight”, and while speed still matters, “it’s a different kind of speed” than HFT [5].

Among the AI methods utilized in algorithmic trading, allow us to define the more common ones:

  • Fuzzy logic model in algorithmic trading is constructed to solve problems by taking into account all the accessible information and making the best practical decision given the input. In a number of the most advanced algorithmic trading model's fuzzy logic models can help trading analysts create automatic sell and buy signals.

  • Decision tree trading model is a directed acrylic graph with nodes designated by decisions. The decision tree showed good performance in searching for rules hidden in substantial amounts of data. The visible relationships connection between nodes branches and leaves makes it the appropriate approach for feature selection and prediction of investment trading decisions in capital markets [6]. Regression trees and classification are the most frequently used in algo trading.

  • Neural network models are one among of the acclaimed Machine learning (ML) model that is convenient for algorithmic traders (link).

In extension to those models, there are variety of other decision-making models that can be employed in the setting of algorithmic trading to help quantitative traders forecasting changes in the price of a financial instrument. Using multiple different models has improved the accuracy of prediction which will maximise the intricacy of programming performance.

Top trading strategies adopted in algorithmic trading

  • Arbitrage trading strategy : strives profits from short-term market inefficiencies that cause the mis-pricing of the same asset in different markets or related assets in the same financial market.

  • Technical analysis trading strategy : predicts future price activity based on an analysis of preceding price movements.

  • Momentum and trend trading strategy : if a trend (upward or downward) is in place then the market could continue in that direction until there are signals showing it has come to end. Momentum is defined as the increase in the price of the stock that can be due to sentiment, earnings and the news. Both techniques depend on the short term price movements than the long term one.

  • Statistical arbitrage trading strategy : uses short-term correlations among security prices to form short-term price predictions and trade to profit from these predictions.

  • Mean reversion trading strategy : is the effect of a market's price reverting back to its historical average/previous price. This sort of strategy is generally based on a mathematical model that assumes a security's high or low price is temporary and will trade back to its average price over a period of time.


Machine learning models are getting very common in algorithmic trading. It has recently been argued that machine learning will transform how people trade securities and manage investments for generations [7]. With the increase in computing capacity, decrease costs of data storage, and availability of Big Data, applying machine learning techniques to trading and investment management problems has become an increasingly viable and exercised option within the financial industry.

bottom of page