Review of AI trading
1. Introduction
The definition of Artificial intelligence (AI) is the intelligence demonstrated by machines, or more precisely by AI algorithms[1]. The process of artificial intelligence includes learning, reasoning, and perception, which eventually enables machines to act with a degree of autonomy, and execute iterative tasks. The advantage of an AI trading system over human traders is that it can collect and analyze large amounts of data and compute a strategy with short delay. AI has greatly changed the aspect of everyone’s daily lives, including various financial sectors, eg: in retail and corporate banking (tailored products, chat-bots for client service, credit scoring and credit underwriting decision-making, credit loss forecasting, anti-money laundering (AML), fraud monitoring and detection, customer service, natural language processing (NLP) for sentiment analysis); asset management (advice, management of portfolio strategies, risk management); insurance (advice, claims management); and trading (AI-driven algorithmic trading, automated execution, process optimisation, back-office). In this article, we will mostly focus on AI assisted trading. With the help of AI, financial advisers analyze millions of data points and try to execute trades at the optimal price while mitigating risk to provide higher returns. We will firstly review the applications of AI in financial trading and its advantages; then we will discuss the components of an AI trading system, which will summarize the general process of building a AI trading system; next we will also discuss the limitations of AI in trading and how to solve these limitations; in the end of the article we will point out several AI platforms, which may give some inspiration for readers who want to develop their own AI trading system.
2. AI in financial trading
AI are widely used in all types of investments, for example: stocks, bonds, certificates of deposits (CDs), mutual funds, exchange-traded-funds (ETFs), options, annuities, retirement plans, commodities, cryptocurrencies, precious metals, etc[2]
AI can be used to design and implement intelligent trading and investment decision strategies to support quantitative investment, investment portfolio formation and selection, optimization of investment strategies and portfolio, and high-frequency algorithmic trading. The advantage of AI trading compared to traditional traders are obvious:
• No emotional decisions: human reactions offer room for creativity and eliminating it with machine language and AI may not always be the best choice. People can also make mistakes because of emotional distress. Bad moods and health issues can result in bad performance and increase the chance of human error. AI eliminates such issues since it runs purely on logic and built-in algorithms. This emotionless approach can help make data-driven decisions in trading.
• Reduction in operational costs: The investment process is fully equipped with intermediate operations. Though the entire process is quite digital, it still employs several human based processes. It involves paperwork, due to which operational costs shoot high. But AI and robotic process automation (RPA) help in reducing these logistics and operational costs.
• Assist the manpower: AI and its technology assist the manpower employed in the network. This helps in making the entire investment process quite simple and easy. From gathering the customer’s data from different contracts, forms and other resources to calculating the interest rates and profit margins, AI has improved and assisted a lot of processes.
• Automate decision making: AI helps in the automation of investing decisions. Some investing firms like Switzerland-based UBS have already employed their algorithmic trading systems. These smart systems help in making investment research and decisions.
Traders can not only use AI to identify and define trading strategies; make decisions based on predictions provided by AI-driven models; execute transactions without human intervention; but also manage liquidity, enhance risk management, better organize order flows and streamline execution. When used for risk management purposes, AI tools allow traders to track their risk exposure and adjust or exit positions depending on predefined objectives and environmental parameters, with minimal human intervention. In terms of order flow management, traders can better control fees and/or liquidity allocation to different pockets of brokers (e.g. regional market-preferences, currency determinations or other parameters of an order handling). Here we illustrate several application scenarios of AI trading[3]. .
• For quantitative investment, designing quantitative models such as multifactor models by techniques e.g. evolutionary algorithms (EA) to learn optimal alpha factors for quantitative investment to create investment portfolios[4];
• Predicting trends of stock-wise fundamental and technical factors associated with each security for stock movement and price trend forecasting, modeling stock movement trend in terms of multi-scale stock data, e.g., trading transactions, market data, and external economic and fundamental data;
• Predicting price and market trend of a financial asset by time series analysis techniques such as autoregressive moving average (ARMA), optimization algorithms such as swarm intelligence-based PSO and sequence modeling techniques such as recurrent neural models(RNN), long-short term memory(LSTM) and recommending trading signals and positions;
• Predicting stock price and movement at specific scenarios, e.g., on ex-dividend day by specific events on that day and ex-dividend period, and forecasting overnight movements before the dividend day;
• Providing trading strategies to maximize trading profit on the given market, e.g, Reinforcement learning(RL) can explore a given market and make an optimal decision by trial and error method. By this self-learning, it can achieve or even outperform human-level accuracy[5].
Based on different objectives, different algorithms can be used, eg. forecasting, portfolio optimization, learn to rank, sentiment analysis, reinforcement learning, recommender systems, behavior analysis, deep models, game theories, optimization methods. For more advanced systems, multi-algorithms are used, however, all the AI trading systems bear similar components, which is summarized in the next sectors.
3. Components of AI trading
General trading process may be divided into five components[6], as summarized in Fig.2 :
Fig.2 Components of an algorithm trading system[6].
•˲ Data access/cleaning:
The data includes but is not limited to:
- Financial data: price data on financial instruments from exchanges and electronic communication networks (ECNs), and also financial information from services, such as Bloomberg and Thomson Reuters.
- Economic data: fundamental economic data, such as the overall state of the countries’ economies (for example, unemployment figures), interest rates, gross domestic product, and national policy. ˲
- Social/news data: Sentiment data “scraped” from social media (Twitter, Facebook), RSS feeds, and news services, etc.
- Real time: Live data feeds from exchanges, ECNs, news services, or streamed social media.
- Historic: previously accumulated and stored financial, economic, and social/news data.
After collecting the needed data for a specific task, the raw data (unprocessed computer data straight from source that may contain errors or may be unanalyzed caused by contradictions, disparities, keying mistakes, missing bits, and so on) should be cleaned by removing erroneous (or dirty) data, and analyzed by creating data processing pipelines.
•˲ Pre-trade analysis:
In this step, AI analyzes properties of assets to identify trading opportunities using market data or financial news. The pre-trade analysis can include three parts:
- Alpha model: the Alpha model analyzes real-time and historic data to identify potential trade opportunities.
- Risk model: the risk model evaluates the levels of exposure/risk associated with the financial portfolio.
- Transaction cost model: the transaction cost model calculates all possible transaction costs.
•˲ Execution Models:
After the pre-trade analysis, the AI can generate a trading signal, and the execution model will trigger orders of the selected model from the previous pre-trade analysis.
•˲Post-trade analysis: Analyzes the results of the trading activity, such as the difference between the price when a buy/sell decision was made and the final profit.
• Model˲optimization: After the post-trade analysis, the model should be constantly updated to include new data, new strategies, to increase profit and reduce risk.
4. Some well-known trading platforms
There are already many platforms that support AI assisted trading, below are several popular platforms:
• Trality[7]: Trality is a platform for anybody who wants to create highly intricate, super-creative algorithms within an educational, community-driven infrastructure that promotes learning and development as a trader.
• Pionex[8]: Pionex is a massive crypto trading platform that offers 18 unique bots and free registration along with the greatest variety of strategies and results. Each bot takes a different approach to your investment, allowing you to plan for the future, hold over a shorter period of time or turn a quick profit.
• Coinrule[9]: Coinrule offers the widest range of preset trading strategies, and the crypto trading bot currently allows users to customize investing with more than 150 trading templates automatically executed when market conditions meet predefined parameters.
• Streak[10]: One of the most efficient trading platforms with Algorithmic Trading in India. The biggest benefit of Streak is that it lets the users perform algo trade without coding. The algos can be created even without the technical knowledge of programming.
• Omnesys Nest[11]: It is one of the best algo trading platforms, provided by Thomson Reuters. It has all the excellent features of a state-of-the-art trading platform, including low latency rates and high levels of performance. “ Algonomics: It is a trading platform offered by NSEIT and is one of the best algo trading platforms. The differentiating feature of the platform is its ultra-low latency levels which are beneficial for high volume trades by the investment banks, fund managers and individual algo traders.
5. Drawbacks of AI
Even though AI technology and data are widely and easily available, getting to the point where profits are generated by AI is still a challenging endeavor.
• Lack of usable financial data
The data collected may be erroneous, or contain interference information (noisy data). This will greatly deteriorate the accuracy of AI trading models. Human intervention is needed to correct or get rid of this information, which will waste enormous time.
One illustration of the consequences of the lack of usable data can be shown with the difficulties in financial modeling. When creating functions that especially depict non-linear relationships, the main problem that arises is the problem of overfitting or underfitting.
Overfitting refers to the fact that relationships and correlations made with existing data (training data) may not hold when new data points (testing data) are introduced. It suggests low bias (because of its success with training data) but a high variance in results (as it has difficulty with testing data). Underfitting occurs when no verifiable clear relationship can be found with existing data. It signifies high bias and high variance in results.
•The difficulty in AI transparency and explainability (XAI)[12].
Another major challenge faced for the deployment of AI in finance. It highlights the gap in technical literacy between many of those working in finance in being able to understand and explain to consumers, as well as the gap that exists for regulators and lawmakers that need to create the guidelines to make sure AI is ethically used. A complete black box will not be acceptable to many stakeholders, giving way to the necessity for more transparency. This is true to an extent, but different, as some AI models cannot be explained at all. For AI trading, lack of transparency means that it’s hard to find out the reason for bad decisions made by AI, and hard to improve their performance.
Despite these drawbacks, technologies have been able to develop solutions that can fully take advantage of what AI can offer despite the challenges that exist.
6. Conclusion
AI trading is an extremely fast-moving field and the state of the art today may look antiquated five years from now. Many companies have already achieved great success with the help of AI. The strong side of an AI system is to transform large amounts of data into accurate predictions on a highly repetitive pattern, but it also uncovers its weak side: An AI system is (at least today) not capable of interpreting singular events, so human knowledge and experience are very important success factors for the application of AI models. AI is the future of investing systems. It is very important for financial institutions to closely follow these developments and to invest to maintain their competitive edge. The slow-movers in this development might have to deal with some leftover deals on bad margins and won’t get the chance to fight their way back to the top.
[1] https://en.wikipedia.org/wiki/Artificial_intelligence
[2]Ellaji, C.H., Jayasri, P., Pradeepthi, C. and Sreehitha, G., 2021. AI-based approaches for profitable investment and trading in stock market. Materials Today: Proceedings.
[3] Cao, L., 2020. AI in finance: A review. Available at SSRN 3647625.
[4]Ucar, I., Ozbayoglu, A.M. and Ucar, M., 2015, May. Developing a two level options trading strategy based on option pair optimization of spread strategies with evolutionary algorithms. In 2015 IEEE Congress on Evolutionary Computation (CEC) (pp. 2526–2531). IEEE.
[5]Azhikodan, A.R., Bhat, A.G. and Jadhav, M.V., 2019. Stock trading bot using deep reinforcement learning. In Innovations in Computer Science and Engineering (pp. 41–49). Springer, Singapore.
[6]Treleaven, P., Galas, M. and Lalchand, V., 2013. Algorithmic trading review. Communications of the ACM, 56(11), pp.76–85.
[10]https://www.streak.tech/home
[12]Shin, D., 2020. User perceptions of algorithmic decisions in the personalized AI system: perceptual evaluation of fairness, accountability, transparency, and explainability. Journal of Broadcasting & Electronic Media, 64(4), pp.541–565.
