Advances in artificial intelligence (AI) has opened up a rich vein of new analytical techniques but investment professionals will have to upskill to mine the digital gold, according to a new CFA Institute paper.
Carrying the catchy title ‘Unstructured Data and AI: Fine-Tuning LLMs to Enhance the Investment Process’, the report – authored by Brian Pisaneschi, CFA Institute senior investment data scientist – says the “role of the investment professional is changing rapidly”.
“Staying abreast of technological trends, mastering programming languages for parsing complex datasets, and being keenly aware of the tools that augment our workflow are necessities that will propel us forward in an increasingly technical finance domain,” the paper says.
Pisaneschi says investment specialists can benefit by applying new AI techniques garnered from large language models (LLMs) such as ChatGPT to probe in-house built datasets.
“The investment industry is, by its very nature, a confidential arena. Financial institutions, investment firms, and individual traders often hold close their trading strategies, market analyses, and proprietary algorithms. This guarded approach extends to data, where exclusive data sources and proprietary databases can provide an edge,” he says in the report.
“… Fine-tuning these powerful models on proprietary data can provide more value than what the underlying models provide in isolation. Supervised fine-tuning, or using human-labeled data to train smaller language models, still holds value despite larger frontier models’ capabilities with little to no human-labeled data.”
For example, the paper says AI tools such as machine-learning and natural language processing (NLP) could prove useful in analysing environmental, social and governance (ESG) factors.
Pisaneschi tested the approach in an ESG case study, constructing a model portfolio based on ‘alternative’ data including corporate tweets.
The ESG results “yielded interesting results by pointing to the importance of real-time information advantages from alternative data”, the CFA report says.
However, the test run covered a period of peak ESG hype “and may not persist outside the limited evaluation period”.
“Additionally, such imperfect models as the fine-tuned model in this case study should be used with caution when applied to small datasets,” the paper says. “The value in using such models comes from being able to test a hypothesis on a large dataset—in this case, the Russell 1000 and 2000—providing a basis for further investigation and resource allocation.”
Despite such potential pitfalls, Pisaneschi says AI – fuelled by better algorithms, bigger (and more diverse) data sets and “exponential” leaps in computing power – will inevitably form part of the investment prospecting kit.
“Being able to leverage the tools and techniques to parse these data, particularly with NLP, is an invaluable resource that should not be left unexplored,” he says.