Ihsan Saracgil: The potential benefits of NLP and LLMs in financial research
What do you consider your biggest professional achievement to date?
At Visible Alpha, I created the research initiative that shows systematic investors the power of leveraging deeper fundamental datasets. This initiative has helped the company build a solid quant client base, and it ultimately helps our quant clients think differently about the data inputs for their strategies. It’s been quite rewarding to see the impact this research has on our quant clients, and I personally find fulfilment in being able to discuss it with my quantitative research and data science counterparts on the buy side and sell side.
What has been your/your firm’s top 3 priorities for the coming year?
In regards to our quant audience, we are working toward offering a more dynamic data-as-a-service model by building more commercial datasets on changing themes. This requires an extremely thoughtful and detail-oriented approach because these text datasets are hard to structure and derive value. LLMs and generative AI are changing the game in this space, and Visible Alpha is determining how we can leverage new technology to help our clients tap into the value of these datasets.
What do you think are the biggest challenges facing data scientists/AI experts/quantitative practitioners for 2023 and beyond?
There are lots of new tools coming to the market, but they are largely untested. It is important to carefully evaluate which ideas have real promise and are solving real problems versus those that are moon-shot ideas that won’t live up to the hype.
NLP continues to be a big area of interest during our research – is the industry really using it to its full potential? Where else can we go with NLP and have you seen examples in other industries that we can learn from?
I think NLP is largely under-utilized and when it is utilized, it is for frivolous use cases that are not backed by any theory. Finance has an enormous number of analyst reports and reported filings. Except for filings, we don’t do much with published research. So what you get from a typical vendor is a big focus on the numbers extracted from filings, but you don’t have any context behind the numbers. People read a ton of research but they can only consume a tiny fraction of what is put out there – I heard from one sell-side director of research that 95% of reports they published don’t even get opened! NLP and LLMs should allow us to get more value out of the published opinions, beyond just reducing them to a sentiment score that doesn’t offer much value. Through NLP and LLMs we can focus on key topics, and compare and summarize different viewpoints without reading hundreds of pages. It should improve the writing to better serve the needs of clients, too – if we can increase read/open rates on research, reports could become more succinct and to the point. For example, there are areas of academia, open access research, and law that are going in this direction. Very lengthy articles that are time-consuming to read are being quickly summarized and distilled by LLMs, enabling people to find what they are looking for much faster.
At Quant Strats, we always discuss the challenges and opportunities of blending quant and fundamental strategies and this is always a popular topic – why do you think this is? What do you think is the most important questions for quants when considering this strategy?
Blending quant and fundamental strategies is a popular topic for a simple reason – the quant industry is very competitive and therefore requires venturing into new territories for an edge. Firms like Visible Alpha are making fundamental data a lot more accessible and immediately useable for quants and, as the cost of working with more detailed datasets goes down, its consumption and integration into quant strategies will go up.
Subscribe to Our Free Newsletter