Harnessing Generative AI for Smarter Investment Strategies

The idea of automation and “thinking machines” dates back centuries, with ancient civilizations like the Greeks and Egyptians imagining self-operating machines. 17th-century philosophers such as Descartes and Leibniz speculated about machines mimicking human thought, laying the groundwork for AI. The official history of AI research began with the Dartmouth Conference in 1956, where the term “Artificial Intelligence” was coined, and concepts like machine learning and neural networks were discussed.

Generative AI, a subset of AI, creates new data based on patterns learned from training. Examples include OpenAI’s GPT and DALL-E. Modern neural networks, inspired by the brain, form the backbone of Generative AI, enabling better pattern recognition within large datasets.

Large Language Models (LLMs) are a type of Generative AI trained on vast amounts of text to understand and generate coherent language. LLMs, like GPT, leverage their understanding of language and concepts to produce contextually relevant responses. OpenAI’s GPT-1 marked a significant milestone in transformer-based models, setting the stage for advancements like GPT-4, which powers ChatGPT today.

How does Windmill Capital use generative AI?

1. Sentiment analysis 

Sentiment analysis entails determining whether a specific news article or earnings call transcript expresses positive or negative sentiment regarding a particular company. By examining news articles, conference call transcripts, and earnings reports, sentiment analysis can help uncover emerging trends before they impact market prices.

Windmill Capital focuses on the top 750 stocks by market capitalization. After filtering out stocks due to red flags, liquidity issues, and other concerns, the remaining universe typically includes 450-500 stocks. Manually tracking news and company-specific reports can be quite challenging.

This is where Large Language Models (LLMs) become essential. We leverage OpenAI’s capabilities and enhance them with AI tools like fine-tuning and prompt engineering to tailor these models specifically for finance-related adjustments and optimizations.

This approach enables us to generate a numerical score that reflects the severity of positive / negative sentiment. Analysts can then focus solely on articles and reports indicating negative sentiment to better understand the risks associated with specific stocks.

2. Summarisation of con-calls and other reports

In financial text analysis, summarizing and extracting key information from documents is essential for swiftly understanding and processing important data within lengthy and intricate texts. Once the relevant information is extracted, the next step is to apply this data to address downstream financial tasks.

Our models excel at summarizing and extracting vital information from extensive financial documents, transforming complex financial narratives into concise summaries, and facilitating more efficient information processing. Furthermore, our models can effectively retrieve information about specific aspects, such as financial performance or mergers & acquisitions. Additionally, we employ unsupervised clustering, a machine learning technique, for efficient summarization of conference calls and news.

3. Identification of similar businesses

Our LLM models also enable us to efficiently construct thematic portfolios by identifying how closely a company aligns with a particular ‘theme.’ By analyzing company descriptions and business activities, the models detect clusters of companies with similar characteristics or that belong to the same sector. This enables us to categorize stocks into established themes, such as consumption or electric vehicles, and update the universe of pure-play stocks accordingly. This is a huge time-saver as it does not require us to scan more than 5000 stocks which are listed and focus only on the relevant ones. 

Future Applications of Generative AI in Investment Management

The aforementioned pointers are few building blocks in the larger scheme of things at Windmill Capital. This evolution has helped us immensely in constructing and managing efficient portfolios at scale.

Emerging applications of generative AI and LLMs in investment management include analyzing time series data to generate forecasts, identifying anomalies in trading volumes and price patterns, applying financial reasoning for strategic planning, generating investment recommendations, providing advisory services, and supporting financial decision-making. As these technologies evolve, Windmill Capital will continuously enhance the research process by integrating cutting-edge technologies like LLMs into our workflow.

_{Disclaimer: Investment in securities market are subject to market risks. Read all the related documents carefully before investing. Registration granted by SEBI, membership of a SEBI recognized supervisory body (if any) and certification from NISM in no way guarantee performance of the intermediary or provide any assurance of returns to investors.}

_{The content in these posts/articles is for informational and educational purposes only and should not be construed as professional financial advice and nor to be construed as an offer to buy /sell or the solicitation of an offer to buy/sell any security or financial products.Users must make their own investment decisions based on their specific investment objective and financial position and using such independent advisors as they believe necessary.}

_{Windmill Capital Team: Windmill Capital Private Limited is a SEBI registered research analyst (Regn. No. INH200007645) based in Bengaluru at No 51 Le Parc Richmonde, Richmond Road, Shanthala Nagar, Bangalore, Karnataka – 560025 creating Thematic & Quantamental curated stock/ETF portfolios. Data analysis is the heart and soul behind our portfolio construction & with 50+ offerings, we have something for everyone. CIN of the company is U74999KA2020PTC132398. For more information and disclosures, visit our disclosures page here.}