As big data grew and computing power increased in the early 2000s, the incorporation of AI into trading systems started to pick up steam ^[1]  . Early research highlighted AI's capacity to handle enormous amounts of financial data, facilitating quick analysis and trade execution (Treleaven, Galas, & Lalchand, 2013). Artificial intelligence (AI) systems, especially those that use machine learning (ML), neural networks, and evolutionary algorithms, have the ability to recognize patterns in both historical and current data, which makes them more flexible than traditional quantitative models.
Predictive accuracy is one of AI's most important effects on algorithmic trading ^[1]  . Long Short-Term Memory (LSTM) networks, a deep learning architecture, have been shown by Fischer and Krauss (2018) to significantly outperform conventional autoregressive models in stock price forecasting. Their analysis of the S&P 500 index showed that AI models could identify intricate non-linear patterns that traditional methods frequently miss. The durability of ensemble learning methods, such as random forests and gradient boosting, in surpassing logistic regression in financial forecasts was further confirmed by the study conducted by Krauss, Do, and Huck (2017) ^[6]  .
A branch of machine learning called reinforcement learning (RL) has drawn interest due to its capacity to discover the best trading strategies via trial and error [4]. A deep reinforcement learning framework for stock trading was created by Deng et al. (2016)^[1]   and it dynamically modified portfolio positions in reaction to shifting market conditions. Their model outperformed fixed-rule systems in terms of returns and volatility. This transition from static, tried-and-true methods to dynamic, real-time learning models highlights AI's primary benefit: its ability to adapt and improve over time.
A study comparing the performance of AI-based trading algorithms to conventional moving average and momentum methods was carried out by Chan and Lo (2020). According to their research, AI-driven tactics produced better Sharpe ratios and were more shock-resistant. Conventional approaches were more vulnerable to erroneous signals during volatile times, although being simpler to understand and apply. The authors came to the conclusion that AI's higher performance was mostly due to its capacity to synthesis a variety of data sources, including unstructured text and historical prices^[1]  .
Another revolutionary tool in AI's toolbox is Natural Language Processing (NLP). Bollen, Mao, and Zeng (2011) demonstrated how sentiment analysis using Twitter feeds could predict market movements with significant accuracy, and their study was a turning point in the use of unstructured data in trading strategies. More recent research has expanded on this innovation by using news analytics, earnings call transcripts, and even geopolitical developments as inputs for algorithmic models^[1]  .
Even with its obvious benefits, AI in trading has drawbacks. The problem of overfitting in AI models, especially in deep learning applications, was brought to light by Zhang et al. (2021). In actual trading scenarios, models based on historical data could perform poorly since they don't generalize well to future market conditions ^[16]  . Furthermore, interpretability and regulatory compliance are hampered by the "black-box" nature of many AI systems. According to the International Organization of Securities Commissions (IOSCO, 2018) ^[6]  , stakeholders and regulators find it challenging to comprehend and oversee AI models' decision-making processes because to their opacity.
Systemic and ethical issues are another area of concern ^[9]  . As demonstrated by the 2010 Flash Crash, AI systems running at high frequencies can cause flash crashes and market instability. Explainable AI (XAI) should be used in trading algorithms to improve accountability and transparency, according to studies like those of Leinweber (2022) ^[12]  . A recurrent subject in recent research is the necessity of strong governance mechanisms to oversee AI-based trading.
Kroll, Barocas, and Felten (2017) state that integrating rule-based logic with machine learning enables the creation of strategies that are both interpretable and adaptive, and that hybrid models that combine AI with traditional statistical techniques are increasingly being proposed to mitigate some of these challenges ^[11]  . This hybrid approach is becoming more and more popular in institutional trading environments where risk management and auditability are crucial.
The use of artificial intelligence in algorithmic trading is reviewed by Hao and Chen^[6]  , who highlight how machine learning, deep learning, and natural language processing enhance predicted accuracy and flexibility. To guarantee dependability in financial markets, they also draw attention to issues like overfitting, interpretability, and regulatory concerns, advocating for hybrid models and explainable AI.
In their analysis of algorithmic trading in African markets, Baba-Yara, John, and Melecky (2020) conclude that although it improves efficiency and liquidity, it also increases the risk of instability, especially in less developed financial systems^[1]  .
Guntuka (2024) talks about how algorithmic trading methods driven by AI are changing financial markets, emphasizing advancements in automation and predictive modeling that speed up and enhance decision-making^[13]  .
Marimuthu (2025) examines AI-powered trading in foreign exchange markets, highlighting how technology can increase liquidity and efficiency while also raising concerns about possible volatility amplification^[13]  .
The impact of algorithmic trading on investor portfolios is examined by Gopakumar, Dananjayan, and Narayanasamy (2023), who highlight both the potential for higher returns and the worries about elevated systemic risks^[6]  .