ChatGPT Empowering Investment Research: Trends and Use Cases
Investment research is the first area that we are seeing the deployment of ChatGPT-like capabilities both from large incumbents and from startups.
Financial Analysts and Financial Advisors are sensing the potential empowerment or threat. Moreso, Financial Market Intelligence providers who will need to up their existing Machine Learning and Artificial intelligence capabilities to stay competitive.
Do-It-Yourself investors are sensing the potential empowerment.
The half-a-dozen use cases that I have curated below, show the nascent trends in these areas. We see domain-specific LLMs being tested and we see improved smaller language models and existing language models combining generative AI capabilities.
At the same time, we need to be aware of the factual errors at this stage. I personally suggest using these LLMs mainly in domains that you already have some knowledge so that you can exercise your judgment about the plausibility of facts and ask for sources or double-check elsewhere (Don’t Trust, Verify is the new Black).
Bloomberg, Microsoft, AWS
BloombergGPT is training a huge hybrid LLM using its own data and from public sources. This qualifies as a milestone in Investment research.
According to Decoder, the datasets on which the current Bloomberg GPT has been trained include 363B financial-specific tokens that came from its own data assets and 345B generic tokens that came from online text datasets. The domain-specific “BloombergGPT,” is a 50-billion-parameter decoder-only language model optimized for financial tasks.
NEW YORK March 30, 2023 — Bloomberg today released a research paper detailing the development of BloombergGPTTM, a new large-scale generative artificial intelligence (AI) model. This large language model (LLM) has been specifically trained on a wide range of financial data to support a diverse set of natural language processing (NLP) tasks within the financial industry. This model will assist Bloomberg in improving existing financial NLP tasks, such as sentiment analysis, named entity recognition, news classification, and question answering, among others.
The vision (pretty realistic at this stage) is for a Digital assistant who can Answer questions with accurate financial facts but also quickly provide sentiment of news headlines; classify news; create headlines; inquire data to test a hypothesis.
Here is a basic example of how this model generates automatically News Headlines from short texts. A job that humans currently, undertake.
The potential of the BloombergGPT model is to develop new customized applications in a cost-effective way that will empower the B2B clients of the resilient Bloomberg Terminal. Market Intelligence will be elevated to another level.
The other resilient low-code application that is entrenched in financial services, is Microsoft`s Excel. The Copilot integration into Excel, is another milestone in this nascent phase of integrating these LLMs into existing ecosystem apps.
AWS is also proposing the fine-tuning of GPT-J 6B language model for financial text generation. The training would be on a publicly available dataset of SEC filings. GPT-J is a much smaller model (less parameters) but is costs less and does well when tasks are narrowed (e.g. financial text generation).
The trade-off between cost and the objectives of the model training are a crucial business decision. The calibration of these models will remain a human decision. I picked the example that AWS shares and compared to the answers generated from two other LLMs that I like to experiment with.
Here is the example sourced from the AWS article.
- “What drives sales growth at Amazon?”
- “Amazon is the world’s largest online retailer. It is also the world’s largest online marketplace. It is also the world” (response of GPT-J 6B with no fine-tuning)
- “Sales growth at Amazon is driven primarily by increased customer usage, including increased selection, lower prices, and increased convenience, and increased sales by other sellers on our websites.” (response of GPT-J 6B with fine-tuning)
I asked ChatGPT this question and the answer was satisfactory:
- Wide selection: Amazon offers a wide selection of products, including both its own branded products and products from third-party sellers. This enables customers to find almost anything they need on the platform, which drives sales growth.
- Competitive pricing: Amazon is known for offering competitive prices on its products, which attracts price-sensitive customers and drives sales growth.
- Convenience: Amazon offers a convenient shopping experience, with features such as fast delivery, easy returns, and one-click ordering. This makes it easy for customers to make purchases, which drives sales growth.
- Personalization: Amazon uses data and analytics to personalize the shopping experience for each customer, providing recommendations and offers based on their browsing and purchase history. This makes the shopping experience more relevant and engaging, which drives sales growth.
- Prime membership: Amazon’s Prime membership program offers benefits such as free shipping, access to streaming services, and exclusive deals. This incentivizes customers to shop more frequently on the platform, which drives sales growth.
Overall, Amazon’s sales growth is driven by a combination of factors, including its wide selection, competitive pricing, convenience, personalization, and membership programs. By continually improving its offerings in these areas, Amazon is able to attract and retain customers and drive sales growth over time.
I also asked PerplexityAI the same question and the answer included more facts:
Several factors drive sales growth at Amazon. Small- and medium-sized businesses (SMBs) account for roughly 60% of products sold in Amazon stores, which contributes to Amazon’s sales growth
1. Amazon Web Services, the company’s cloud computing division, generated $21.4 billion in sales, a 16% rise from a year earlier
2. Amazon’s revenue growth has decelerated in North American net sales and International net sales segments, where the operating profit margins are 4.7% and 1.2%, respectively
3. Amazon’s third-party sellers are likely to increase the prices of the goods they sell, which would boost revenue for Amazon since the company gets a percentage of third-party sales as commissions
4. Amazon also sells its own products and can take this opportunity to enact modest price increases on those as well
5. These factors, along with others, contribute to Amazon’s sales growth.
Four Fintechs
In Fintech startup land, I spotted 4 companies in the Investment research domain piloting or integrating ChatGPT-like models.
Alphasense is a 12yr B2B market intelligence business that has been using AI to help companies and financial institutions find information on companies, data, and more. You can think of them as a domain-specific search engine and market intelligence platform. They do plan to go public once the markets permit.
Last month, they raised $100 million (Series D) at $1.8 billion valuation partly to deploy new AI capabilities by integrating advanced LLMs. The funding is led by Alphabet’s CapitalG. They are very much aware of the hallucination issue and plan to address it by including footnoted sources (this is what I like about PerplexityAI).
Magnifi, is a 5yr old Fintech part of the TIFIN group, offering investment search assistance for DIY investors of financial advisors. You can think of them as a copilot for a DIY investor and a conversational investing interface. According to CNBC, their Magnifi Personal Account subscription is powered by ChatGPT (Monthly subscriptions cost $14).
CNBC reporter provided an imaginary portfolio, with iShares 20 Plus Year Treasury Bond ETF as one of the holdings and asked how Fed rate hikes could affect this portfolio. Here is the answer.
What I find interesting about Magnifi is that it can leverage in creative / valuable ways the intelligence from their acquisition of SharingAlpha (last year). SharingAlpha is an Isreali Fintech that developed an open-sourced Fund rating platform. It ranks funds based on the average rating provided by Fund Selectors globally based on qualitative parameters. Their mantra was ‘where Morningstar meets TripAdvisor’.
Portrait Analytics, is a Boston-based company that just raised $3m seed round to develop its generative AI research platform for investment analysts. The founder (David Plon) is a young hegde fund analysts that wants to empower analysts. Here is an example of a conversational search from their website asking
- What is the 2023 guidance of net charge-offs and what factors are driving it, for Discover Financial Services ($DFS)?
FinChat.io, is a newly launched product from Stratosphere.io which I heard about from a Linkedin post by Linas Belunas.
Crunchbase has no info on either entity. PerplexityAI helped me find a Tweet from their handle that states they launched in 2021 as `The #1 terminal for financial data, segments, KPIs and investor tools. Completely free to get started.`
I checked out Stratosphere which seems to be a financial data resource for companies (a la yahoo.finance). You type a company name or ticker and it generates all sorts of financial information. I asked for $BABA and $SQ (Jack Ma and Jack Dorsey 😉 ).
Then I tried Finchat.io, the ChatGPT trained app using Stratosphere`s data who aspires to become an AI-powered equity analyst. I was not impressed with the result.
Conclusion
The deployment of large language models (LLMs) like ChatGPT in investment research is a nascent but rapidly evolving trend. I expect to hear soon from incumbents like S&P Global, PitchBook, and Refinitiv, as the markets feels the pressure to position themselves in the B2B space. These businesses will need to decide the trade-offs between costs and their business objectives.
I anticipate experimentation / research to evaluate the appropriate calibrations especially for B2B offerings.
The B2C offerings will soon be faced with the challenge of what a sustainable business model.
I risk repeating myself: use LLMs in domains where you have some prior knowledge and you are able to exercise judgment and verify factual errors.
Use LLMs freely, if you are in creative mode.
📌 Twitter: https://twitter.com/efipm
📌 Subscribe to my YouTube Channel with my insights and industry leader interviews. New video every Wednesday: https://www.youtube.com/EfiPylarinou
📌 Spotify Podcasts. Follow here: https://open.spotify.com/show/5bRkZEYHSwPiGx7vTqylw6?si=Mg3hN5PDQ86K10GjeK52jw
📌 Linkedin: https://www.linkedin.com/in/efipylarinou/
📌 TikTok: https://www.tiktok.com/@efiglobal
📌 Web: https://efipylarinou.com/
Subscribe to DDIntel Here.
Visit our website here: https://www.datadriveninvestor.com
Join our network here: https://datadriveninvestor.com/collaborate
