As an early adopter of Artificial Intelligence (AI) investing, asset management firm Differential Capital has used data to uncover what the market is not seeing, to the benefit of investors for the past six years and has now shared their learnings with competitors.
The company has developed a Python package which allows for stock-level extensions to Menchero’s Method for portfolio attributions, which they have shared with the financial services industry.
“Portfolio attribution is a critical tool for investment managers, enabling them to dissect and understand the sources of portfolio performance relative to a benchmark. By breaking down returns into allocation and selection effects, attribution analysis provides transparency into which investment decisions are driving value and where adjustments may be needed,” said Vincent Anthonyrajah, CEO of Differential Capital.
Generative AI – Large Language Models (LLM) in particular – has transformed the way we engage with technology in only a few short years. “Companies with large developer and engineering costs, such as telecommunications companies and banks who are at the mercy of the current high IT cost inflation, will stand to gain from using LLM as capabilities of these systems improve,” Anthonyrajah said.
The development and implementation of reasoning is one of the latest paradigms used to realise performance gains in the LLM space. OpenAI’s GPT-o1 model uses Chain of Thought (CoT) reasoning, designed to emulate human cognitive processes by breaking down complex problems into manageable steps. “While straightforward in principle, building robust and consistent CoT’s requires reinforcement learning.”
The recently-released o3 OpenAI update allows users to see the chain of thought as responses are developed, which ensures transparency and provides understanding of how the model comes to a conclusion. Previously LLM’s could only engage with text, but now, through new technological advances, are able to understand data in images audio and video.
Menchero’s Method provides a way to accurately distribute attributions effects over various points within a time period, but only determines attributions at a sector-level. “Seeing a need for stock-level attributions calculations as well as to incorporate trading profits in attributions, we expanded Menchero’s Method to allow these to be performed in an interpretable and mathematically consistent manner. Using the same mathematical framework as sector level calculations, we adjust interpretations and aggregations to correctly reflect the hierarchy between sectors and stocks.”
Due to a lack of available tools to perform such calculations, Differential Capital has released a Python package to share these techniques. “The more the financial services industry adopts AI investing and machine learning, the stronger this sector becomes.”
ENDS//
