AI Series: How Goldman Sachs leverages AI to extract better investment insights

08 November 2023 by National Bank Investments
Photo of Terry Dimock from NBI with Charles Nadim from Laurene Azoulay

Financial articles, earnings call transcripts, analyst reports, and regulatory filings, a lot of the valuable data sets we leverage have become larger, less structured, and more complex in nature. See how Goldman Sachs Asset Management L.P. (GSAM) utilizes its deep resources and AI to extract investment insights from large, unstructured data sets.

Investment processes have evolved closely alongside the latest in data and quantitative techniques. Financial articles, earnings call transcripts, analyst reports, and regulatory filings. More recently, a lot of the valuable data sets we leverage have become larger, less structured, and more complex in nature. 

The Goldman Sachs Asset Management Edge

A key advantage Goldman Sachs Asset Management (GSAM) is its deep resources. As a result, they can fine-tune and deploy AI tools alongside a variety of data inputs from around the globe in their investment process to inform holistic and forward-looking views on companies.

To extract investment insights from these large, unstructured data sets, you need technology and computing power. This is where Goldman Sachs stands out. GASM can leverage the broader investment of the firm in these technologies, and have access to significant computing capacity, GPUs, and cloud technology.

This allows them to capture subtle clues from management about their feelings and the true prospects of that company.

To do so, they implemented signals to capture whether management is exhibiting bullish or bearish sentiment. They can also look at what management says during the question-and-answer period of an earnings call when comments are less scripted and expose more human biases and tendencies.

It is also valuable from a security pricing standpoint, where they can look at how the management of a company views its future growth prospects, compared to what the market thinks will happen. 

The most recent advancement

More recently, their models can add some semantic meaning behind words. Suddenly, an algorithm can understand the context of a word within a sentence and refine how we measure sentiment within the text.

And that really is a game changer in sentiment analysis because we can really make use of those new tools to extract greater insights from those nuances.

« Leveraging artificial intelligence along with acoustic features we seek to identify ambiguity, emotion, or evasiveness in audio files to complement the information we can get from text transcripts. While the text may tell you what a CEO saying, the audio will help understand how they say it. » Laurene Azoulay, Global Head of Client Portfolio Management for Quantitative Equity at GSAM.

Stay tuned for more in the series:

  • Jarislowsky Fraser: How to catch the AI wave by investing in Canadian companies.
  • Manulife: AI, an indispensable tool for managing ESG risks.
  • Nuveen: Spatial finance for responsible investment.

Terry Dimock

Hello everyone, we're here today with Laurene Azoulay from Goldman Sachs Asset Management to talk about artificial intelligence. Welcome Lorraine.

Laurene Azoulay

Thanks, Terry. Glad to be here.

Terry Dimock

So let's start with a simple question. So, what is artificial intelligence and why is the current hype?

Laurene Azoulay

So artificial intelligence is the ability of a machine to synthesize information in order to solve a problem. The level of intelligence is going to be defined by the difficulty of the problem and how quickly it can be solved. To your point, it's not new. It was actually first introduced by a very famous computer scientist in history. His name is Alan Turing. It was all the way back in the 1950s.

But what is different today is that we're seeing a significant breakthrough with the large language models that is based on what we call transformer technology. That's the T in ChatGPT or in Google's BERT. This technology introduces contextual relationships between words and documents in a way that is highly efficient, but also more importantly, resembles how humans process language.

Terry Dimock

So how does your team, the quantitative investment strategy team at Goldman Sachs, benefit from this technology?

Laurene Azoulay

I would say that the recent developments are more evolution rather than a revolution. It's not revolutionizing our approach to investing, rather we're excited to leverage these techniques and really help us continually enhance how we evaluate investment opportunities. My team has been a systematic investor in equities going back to 1989. For what we do, we need to synthesize information from data.

And then construct investment signals. If you go back 20 years ago, it was a relatively simple process. We were generally looking at structured data from a balance sheet or income statement. More recently, a lot of the valuable data sets we leverage became larger, less structured, more complex in nature. Think in terms of financial news articles, earnings course transcripts, analysis research reports, regulatory filings.

And so in order to extract investment insights from all of these large, unstructured data sets, we really need technology and computing power. And this is where being part of Goldman Sachs is a key advantage for our team, as we can leverage the broader investment of the firm in these technologies. You know, just one example, we have access to significant computing capacity, including GPUs, cloud technology, and we can really leverage that in our research process.

Terry Dimock

So with the latest leap in AI technology, you can work with meaning and context. So how does that change things for your team?

Laurene Azoulay

Oh, as you know, human language is very complicated and there are nuances to words and context is just so important to meaning. For us to form a view around a company, we may be looking for subtle clues coming from all the different market participants. For many years now, we have researched and implemented signals that seek to capture whether the management of a company is exhibiting a bullish or bearish sentiment we would look at what the management would say during earning calls, especially during the Q&A part of the call. It's like much less scripted. There is much more biases that can be exposed. So the first step to this sentiment analysis is to parse through earnings call transcript. So again, 10 years ago, we would represent textual data in a pretty simple way. It would just be a succession of words.

And we'd be looking at the frequency of those words. And that would help us overall to get a sense of the topic, but also the overall sentiment being positive or negative. More recently with the transformer models, those models helped us to add semantics behind words. This technology basically will represent a word in multiple dimensions such as a word that would appear in a similar context as another word to have a pretty similar representation. Just to give you a concrete example, the words king and queen will be similar. But even more than that, if you take the representation of the king minus the representation of man plus the representation of woman, you will get the representation of the word queen. Okay, and so, even though again, for a human, that seems obvious. Suddenly a machine, an algorithm, can understand not only the context of a word within a sentence but also the relationship between words. And that really is a game change in sentiment analysis because we can really make use of those new tools to extract greater insights from those nuances.

Terry Dimock

So, Goldman Sachs has been experimenting with this and using this in concrete ways for a while. So what's next?

Laurene Azoulay

So many interesting things to explore. The team spent a lot of time on text for the past decade or so. Some of our newest area of research though has been turning towards leveraging audio files. So, the text will tell you what a CEO is saying. The audio would tell you how the CEO is saying it. And again, leveraging artificial intelligence techniques along with acoustic features.we really seek to identify things that can be greedy or emotional or even evasiveness on your files and that comes and complements the information we can get from text transcripts. So, you know, we've seen many, many advances in the past few years. We're very optimistic about what the future will bring, I think both on the investment side and for the broader society.

Terry Dimock

Well, we're all looking forward to seeing this evolution. Thank you very much, Lorraine, for your time today. And thank you for everyone who is listening online. Have a great day.

Legal notes

The information and the data supplied in the present document, including those supplied by third parties, are considered accurate at the time of their printing and were obtained from sources which we considered reliable. We reserve the right to modify them without advance notice. This information and data are supplied as informative content only. No representation or guarantee, explicit or implicit, is made as for the exactness, the quality and the complete character of this information and these data. The opinions expressed are not to be construed as solicitation or offer to buy or sell shares mentioned herein and should not be considered as recommendations.

® NATIONAL BANK INVESTMENTS is a registered trademark of National Bank of Canada, used under licence by National Bank Investments Inc. 

© 2023 National Bank Investments Inc. All rights reserved. Any reproduction, in whole or in part, is strictly prohibited without the prior written consent of National Bank Investments Inc. 

National Bank Investments is a signatory of the United Nations-supported Principles for Responsible Investment, a member of Canada’s Responsible Investment Association, and a founding participant in the Climate Engagement Canada initiative.