Towards the "Singularity" – AI Reshaping the Asset Management Industry

UBS believes that artificial intelligence is triggering a profound revolution in asset management. The core of this revolution is not about machines replacing humans, but rather the "singularity" brought about by human-machine collaboration—a new investment paradigm that combines human deep insights with machine extraordinary computing power.

According to the trading desk news on August 27, UBS's research report suggests that the most successful investors in the next decade will no longer be purely quantitative or traditional stock pickers, but rather hybrid talents who can master both approaches and use AI as a force multiplier.

UBS points out that analysts have an advantage in identifying significant risks, while in areas where analysts do not have a clear opinion, machine learning models perform better. A hybrid model that combines artificial intelligence and human insights can generate significant returns across a broader stock pool (over 3,860 stocks).

Three Major Tools of AI

UBS notes that AI is no longer a distant concept but a toolbox composed of a series of data-driven technologies that are deeply embedded in the investment process. Its rise is attributed to data explosion, advancements in computing power, and the popularization of AI tools.

The report emphasizes that the three technologies currently having the greatest impact on the asset management industry are:

• Machine Learning: As the core of AI, machine learning models make predictions by learning patterns in data. It is widely used in signal generation, risk modeling, and even finding trading liquidity. Machine learning excels at identifying nonlinear relationships that traditional linear models cannot capture, thereby improving prediction accuracy.

• Neural Networks: Especially deep learning architectures, which perform exceptionally well in handling high-dimensional, unstructured data, such as recognizing patterns in volatility surfaces or capturing temporal dependencies in time series data. Its flexibility is a significant advantage, but the downsides include poor interpretability, high training costs, and the risk of overfitting.

• Large Language Models: Large language models represent the most revolutionary breakthrough in this field, bringing natural language processing into the mainstream. Asset management firms can now extract insights at scale from earnings call transcripts, regulatory documents, and research reports, transforming qualitative text into structured data.

• For example, research by Lopez-Lira et al. shows that the topic scores extracted by large language models from earnings call transcripts have a significant impact on predicting stock returns. However, the report emphasizes that large language models cannot replace domain expertise; their core role is to enhance and extend fundamental analysis.

Comparative Analysis of Human-Machine Advantages

UBS believes that viewing AI as a competitor to replace humans is a mistake. Understanding the respective boundaries of human and machine advantages is key for any investment institution to contemplate future evolution.

Machine Advantages—Speed, Breadth, and Consistency:

• Speed and Breadth: The data processing speed and scale of machines far exceed any human team. An analyst may be able to deeply cover 20 stocks, but a system based on large language models can scan 10,000 earnings call transcripts daily and flag anomalies or common shifts in market sentiment.

• Consistency: As long as the model is robust and the data is clean, machines can tirelessly and impartially execute tasks repeatedly, yielding highly reproducible results (excluding the "hallucination" phenomenon of large language models, which highlights the necessity of human oversight).

Human Advantages - Context, Complexity, and Causal Inference:

• Interpreting Incidents: Financial markets are filled with non-repetitive events that models struggle to learn, such as regulatory changes, management turnover, or black swan events. Humans can combine environmental, industry knowledge, and complex models to understand the far-reaching impacts of these events.

• Understanding Complexity: Constructing investment logic, understanding the interactions of multiple driving factors, and assessing the impact of intangible assets such as corporate culture and management quality are areas where current AI models fail after reaching a certain level of complexity.

• Dealing with the Unknown: AI learns from historical data, and when the market enters a new paradigm (regime shift), historical experience becomes ineffective. While humans are not perfect, they can adapt to new environments more quickly through analogical reasoning and qualitative judgment.

Moreover, in terms of social responsibility and governance, human ethical and value judgments are irreplaceable by machines. A decision-making process involving human oversight is crucial for managing reputation, regulatory, and operational risks.

Investment Integration in the "Singularity" Era

The traditional barriers between quantitative investing and fundamental investing are being broken down by AI, leading both towards a convergence point known as "The Singularity."

Quantitative investors are delving into fundamentals.

With tools like large language models, quantitative models can now process not only traditional financial and price data but also large-scale unstructured data, such as earnings call transcripts, press releases, and alternative data (like satellite images and web traffic), capturing signals that were previously identifiable only by fundamental analysts.

Fundamental managers are embracing scalability.

AI tools have greatly expanded the research scope for fundamental teams. Machine learning models can screen investment targets, AI assistants can read reports and flag anomalous data, and valuation models can automatically generate benchmark cases for discounted cash flow (DCF).

This liberates analysts from tedious data processing tasks, allowing them to focus on high-value activities such as channel research and management interviews.

A New Investment Paradigm in Dance with AI

To validate the "Singularity" theory, the UBS quantitative research team conducted an innovative experiment.

They first identified several functions that clients value most in researchers (company modeling, industry expertise, theme identification, etc.) and quantified them.

The research team compared the views of analysts with the predictions of UBS's internal GBM strategy model. The results showed:

• Human analysts outperformed machines on their top three favored stocks and bottom three disliked stocks. This indicates that in the areas where analysts invest the most effort and have the strongest beliefs, their insights possess genuine alpha.
• For the remaining stocks with medium attention, the GBM model's predictive performance is better. This may be because these stocks perform steadily, lack catalysts, and have lower analyst attention.

Based on this finding, the team constructed a "Singularity" portfolio: For stocks covered by analysts, adopt their top three favorites and least favorites; for the middle stocks, use the predictions from the GBM model for ranking and selection.

Backtesting results (since 2010) show that this hybrid model, which combines human insights and machine predictions, demonstrates strong return generation capabilities across all 3,860+ stocks it covers.

In summary, the report emphasizes that human-machine collaboration will become a key competitive advantage in future investments. Successful investment management firms will build teams that combine human contextual understanding with machine capabilities. In the AI era, companies will achieve differentiated competition through proprietary data, exclusive knowledge bases, and customized models