Can AI predict future stock returns?

Artificial intelligence can outperform human analysts. Used together, humans and AI are particularly difficult to beat.

Machine learning is a field of research focused on developing methods that analyse data and improve performance without direct human intervention. It is a core component of artificial intelligence. Machine learning algorithms build models based on sample data, known as training data, to make predictions or decisions without being explicitly programmed. These algorithms are used in a wide range of applications, including medicine, email filtering, speech recognition, and computer vision, where it is difficult or impractical to design traditional algorithms for the task. Some even try to apply AI techniques to online gambling, using platforms like Richard Casino login, though successful cases remain rare, and it’s generally better to rely on trusted casinos offering bonuses or free spins.

While algorithmic trading has long been established in financial markets, AI can also support decision-making in the early stages of portfolio construction. The potential advantages of AI include:

• superior computing power to analyse large volumes of data in a short time
• avoidance of cognitive biases to which humans are prone, as AI operates more rationally

Can AI replace humans in predicting stock returns, or does it primarily enhance their forecasts?

What are the Best AI stocks to invest in right now?

Check our buy/sell tips

AI vs. human analysts

Sean Cao, Wei Jiang, Junbo Wang, and Baozhong Yang, authors of From Man vs. Machine to Man + Machine: The Art and AI of Stock Analyses, published in the October 2024 issue of the Journal of Financial Economics, examined how AI compares with human analysts in predicting stock returns. Their objective was to determine:

• under which circumstances human analysts retain an advantage over AI
• how combining human analysts and AI affects the accuracy of stock forecasts
• what implications these findings have for the broader application of AI in skilled professions and decision-making processes

The authors developed their own AI model to predict 12-month stock returns, derived from 12-month price targets, and compared these predictions with analyst forecasts made at the same time for the same stocks. The inputs included company-specific, industry, and macroeconomic variables, as well as text data from company announcements, news, and social media, updated shortly before the analyst forecasts. Information from the analyst forecasts themselves was deliberately excluded so that the AI model would not benefit from analysts’ insights.

The analyst forecast sample was compiled from the Thomson Reuters Institutional Brokers’ Estimate System database. After merging IBES with CRSP and Compustat data, the final sample consisted of 1,153,565 twelve-month price target forecasts for 6,315 companies, provided by 11,890 analysts from 861 brokerage firms, as well as 5,885,063 earnings forecasts for the first to fourth quarters for 8,062 companies, submitted by 14,363 analysts from 926 brokerage firms. The model covered the period from 2001 to 2018.

Key findings

1. An AI analyst trained to process company announcements, industry trends, and macroeconomic indicators outperformed most analysts, beating 54.5% of them in predicting stock returns. This advantage likely stems from superior information processing or immunity to predictable human biases linked to incentives and psychological factors.
2. Compared with analysts, the AI model generated superior returns, with alpha ranging from 50 to 72 basis points per month, statistically significant at the 1% level in almost all cases.
3. The AI model significantly outperformed analysts in the lower-skill quantiles and performed almost on par with elite analysts, achieving a 49.3% win rate against analysts who had beaten the market in each of the previous five years. Only 7.3% of all analysts met this benchmark.
4. Macro variables and company returns contributed most to the AI model’s performance, accounting for 27.6% and 24.4% respectively, followed by company-related variables at 22% and text-based information at 9.3%. Earnings information had the smallest contribution, at 2%.
5. Human analysts outperformed AI when institutional knowledge was critical, such as in cases involving intangible assets or financial distress. They also performed better for smaller and less liquid companies, as well as in industries or firms experiencing rapid change, high competitive intensity, elevated risk, or significant financial stress.
6. AI performed best when information was transparent but extensive.
7. When analyst forecasts were added to the information set used by the AI model, the combined model outperformed the pure AI model by 54.8% and significantly reduced extreme errors.
8. The likelihood of extreme errors was similar for analysts and AI, at 9.3% and 7.8% respectively, using the 90th-percentile threshold. The combined model avoided around 90% of the extreme errors made by human analysts and 40% of those made by AI alone.
9. The combined model leveraged the complementary strengths of humans and AI. AI excelled at large-scale data processing and pattern recognition, while human analysts contributed contextual understanding and judgement, resulting in more robust forecasts.
10. Analysts began to catch up with AI once alternative data became available and their employers developed internal AI capabilities.
11. The documented synergies between humans and machines illustrate how professionals can adapt and retain their advantage as AI capabilities continue to advance.

Augmentation by AI

Perhaps the most interesting finding was that while the model outperformed analysts in predicting returns, analysts beat the machine 69.2% of the time in predicting profits. However, the combined analyst-AI model outperformed 55% of analysts’ predictions.

Their findings led the authors to conclude: “Overall, the study supports the hypothesis that analysts’ capabilities can be augmented by AI and, more importantly, that analysts’ work with the help of AI modelling offers additional value and synergies, especially in unusual and rapidly evolving situations.

They added: ‘While the future of AI remains uncertain, the parts of human capabilities that are complementary to AI enable promising collaboration and complementarity between humans and machines.’

A study examined the live performance of AI-driven investment funds. Rui Chen and Jinjuan Ren, authors of Do AI-Powered Mutual Funds Perform Better?, published in the August 2022 issue of Finance Research Letters, analysed the performance of AI-powered investment funds. Their data sample came from the CRSP Survivor Bias-Free US Mutual Fund Database and covered the 26-month period from November 2017 to December 2019.

They defined the fund types as follows:

• AI-powered funds: Use machine learning technologies to actively select stocks in portfolio construction.
• Quantitative funds: Use fixed rules and numerical methods to create computer-driven models and make investment decisions.
• Discretionary funds: Traditional funds where stock selection and investment decisions are primarily made through human judgement.

Key findings

1. The performance of AI-powered investment funds was statistically indistinguishable from the overall market in 25 of the 26 months of the sample period.
2. AI-powered funds did not generate significant risk-adjusted returns, showing only marginally better stock selection abilities (except when using equal weighting) and no market timing abilities.
3. AI-assisted funds outperformed their human-managed competitors mainly due to lower turnover—31% compared to 72%—which led to lower transaction costs and slightly better stock selection.
4. AI-assisted funds held fewer stocks (149 versus 197), resulting in more concentrated portfolios.
5. AI-assisted funds avoided some common behavioural patterns, such as the disposition effect.

Key takeaways for investors

Cao, Jiang, Wang, and Yang demonstrated that stock forecasts can be improved by combining the strengths of AI and human analysts, producing better results than relying on either one alone. Their research showed that AI and human analysts complement each other: AI excels at processing large amounts of data and recognising patterns, while humans provide contextual understanding, intuition, and nuanced insights. Recognising these synergies helps in developing systems that maximise both AI and human input.

However, there is currently no evidence that AI-supported funds consistently perform better on a risk-adjusted basis. Perhaps the most important insight is that AI models help reduce human bias and make forecasts more accurate, which could lead to more efficient markets and limit opportunities to generate alpha through security selection.