The Future of Causal Reasoning in Investment Management: Harnessing the Power of Large Language Models


Joseph Byrum

May 1, 2024

Share on:

In the complex, dynamic, and forward-looking world of investment management, the ability to understand and leverage causal relationships has long been recognized as a crucial factor in making informed decisions and driving better outcomes. Just as Isaac Newton’s groundbreaking work on the laws of motion was inspired by observing an apple falling from a tree, investment professionals have often sought inspiration from the natural world to gain insights into the intricate workings of financial markets.

For instance, Lord May of Oxford, an ecologist and former Royal Society president, teamed up with Andy Haldane, the Bank of England’s chief economist, to develop a complex financial model based on observations from nature. Their work explored how understanding the effects of a “super spreader” contagion in nature could assist in developing public policies to guard against systemic flaws that spread financial contagion. Though written over a decade ago, the paper has extra relevance today, as the authors wondered whether an understanding of the effects of a “super spreader” contagion in nature could assist in developing public policies that would guard against systemic flaws that spread financial contagion.

However, traditional approaches to causal reasoning in this domain have often been limited by the inherent challenges of uncovering and exploiting causal relationships in fast-paced, informationally efficient markets. Now, with the emergence of large language models (LLMs) and their impressive performance on various causal reasoning tasks, we stand at the precipice of a paradigm shift that promises to revolutionize the way investment managers analyze companies and portfolios, how rating agencies apply credit ratings, and the tools available to assist individual investors.

The Dominance of Quantitative Assessment Techniques

Currently, there is a dominant preference for quantitative assessment techniques, particularly correlation-based methods such as the Granger causality test. The Granger causality test is a statistical hypothesis test used to determine whether one time series can be useful in forecasting another. It is based on the idea that if a signal X “Granger-causes” a signal Y, then past values of X should contain information that helps predict Y above and beyond the information contained in past values of Y alone.

While these methods have their merits, they also expose a weakness in current research: the lack of qualitative, heuristics-based techniques that could greatly support investment professionals in managing uncertainty and navigating unknown unknowns. Much like how the banking system relies on a complex series of interactions between financial institutions, with the potential for shocks to propagate through the system like a virus, the investment world is a phenomenally complex adaptive system. The actions of a single investor or a small group, such as the Reddit crowd and GameStop, can create ripple effects that help or hurt multiple parties. No investment takes place in a vacuum.

The Emergence of Large Language Models

Enter large language models. Recent research has demonstrated that LLMs can achieve remarkable performance on a wide range of causal reasoning tasks, such as counterfactual reasoning, actual causality, and causal discovery, often outperforming existing algorithms. By leveraging the vast amounts of knowledge acquired through pretraining on diverse datasets, LLMs can infer causal relationships in a way that mirrors how human domain experts construct causal graphs based on their knowledge of physics, common sense, and specialized domains.

This knowledge-based approach to causal discovery complements and augments traditional data-driven methods, opening up new avenues for causal inference in investment management. For example, LLMs could be used to analyze company reports, news articles, and other unstructured data sources to identify potential causal relationships between various factors, such as management decisions, market trends, and geopolitical events, and their impact on stock prices or portfolio performance.

Imagine how much more effective financial analysts and portfolio managers could be if they could enhance their understanding by surveying social media and other rich sources of business information in real-time, drawing out the key developments and trends that would inform their choices. AI can also prioritize information flow so that it does not become overwhelming for the investment manager.

Integrating Language-Based and Data-Based Analyses

Moreover, the natural language capabilities of LLMs enable them to integrate different types of causal reasoning, understanding and formalizing causal scenarios described in natural language, generating relevant formal premises based on background knowledge, and identifying and framing challenging causal constraints and validations. This ability to integrate language-based and data-based analyses provides investment professionals with complementary approaches to improve decision-making and outcomes, unifying the disparate subfields of causality research and providing powerful new signals for navigating the complexities of financial markets.

The Mathematics of Nature and Finance

Mathematics describes the complex patterns of nature and finance alike. Some traders even use Fibonacci numbers to make money, believing they can identify points at which a stock price is about to reverse. While the effectiveness of such techniques may be debatable, it underscores the fact that the market is moved by the individual choices of its participants. Each move is made at points that feel “natural” without necessarily knowing the underlying mathematics.

The Path Forward

Looking ahead, fully realizing the potential of LLMs in causal reasoning for investment management will require a multifaceted approach. Researchers must continue to develop and refine both quantitative and qualitative assessment techniques, with a focus on methods that can handle the complexity and fast-paced nature of financial markets. Investment professionals, in turn, must be open to incorporating these techniques into their analytical toolkits and developing the skills necessary to interpret and act upon causal insights.

Moreover, there is a need for greater collaboration between academia and industry to bridge the gap between theoretical advances and practical applications, fostering a virtuous cycle of knowledge creation and dissemination. To succeed, the investment management community must be willing to embrace the disruption, which can mean acknowledging when the existing talent pool lacks the prerequisite skills for the future, and that new talent needs to be brought on board.

A New Paradigm of Causal Reasoning

The emergence of large language models and their impressive performance on causal reasoning tasks, coupled with the insights gleaned from the systematic literature review on causality testing in equity markets, serves as a clarion call for the investment management community to embrace a new paradigm of causal reasoning. By harnessing the power of LLMs and integrating them with established causal methods and human expertise, we can move towards a more sophisticated, evidence-based approach to investment decision-making, ultimately achieving better outcomes for investors and navigating the complexities of financial markets with greater confidence and success.

As we stand on the cusp of this transformative change, it is clear that the future of investment management lies in the fusion of human ingenuity and artificial intelligence, working together to uncover the hidden patterns and causal relationships that drive the ever-changing landscape of global finance. Just as the spreadsheet revolutionized business analytics in the 1970s, enabling managers to quickly test scenarios and find answers to complex questions, LLMs are poised to transform the field of investment management. The change might happen within the span of a few years, given the rapid pace of advancement in AI, and if it all comes together, the transformation promised by this new paradigm of causal reasoning will be here before we know it.

Related Articles

The Rise of Small Language Models: Efficiency Meets Specialization

In the rapidly evolving landscape of artificial intelligence (AI), language models have emerged as a powerful tool for understanding and generating human-like text. While large language models (LLMs) have garnered significant attention for their impressive capabilities, a new contender has entered the arena: small language models (SLMs). These compact yet powerful models are challenging the notion that bigger is always better, offering a more efficient and specialized approach to natural language processing (NLP) tasks.


Joseph Byrum

May 31, 2024

The Perils of Complacency: Lessons from the Fallen Tech Titans

For investment managers, the message is clear: adapt or risk being left behind. The firms that will emerge as the new leaders in the industry will be those that are fast, cheaper, and better. They will harness the power of AI and other emerging technologies to streamline processes, enhance decision-making, and deliver superior results for their clients.


Joseph Byrum

May 28, 2024

Organizational Understanding of AI: Bridging the Gap for Effective Implementation

To fully capitalize on the transformative potential of AI, organizations must approach its implementation not as a singular, monolithic endeavor, but as a parallel-path operator — a multifaceted, iterative process that weaves the technology into the very fabric of the business.


Joseph Byrum

May 22, 2024