Ep. 375: Bilal on How to Use AI to Turbocharge Your Market Analysis

Bilal Hafeez (0:03)

Welcome to the Macro Hive Conversations Podcast. I'm Bilal Hafeez, Head of Strategy at Macro Hive. Before we dive in, a quick update. Following Macro Hive's acquisition by the BGC Group, BGC has launched a new regulated business called Macro Hive Markets. For the first time, onboarded clients can also access execution services via Macro Hive Markets. This new workflow will allow macro institutional investors to benefit from strategy and quant insights, alongside access to global liquidity. Please contact Andrew Simon or Gail Motibrian on Bloomberg, or click the link in the description below for more details on Macro Hive Markets and its service offering.

On this podcast episode, I'm going to flip the script and I'm going to talk about how I use AI to do my research and market analysis. And this is because I've been inundated with questions in recent months about what is the best way to use AI in doing work for financial markets and what are the specific use cases. So I thought I'd take up this opportunity to go through my process. I'll be using slides as well to demonstrate some of the flows I use. So if you're on YouTube, you'll be able to view the slides. Otherwise, you can ask me for the slides and I can email them over to you. Essentially, I'll talk about six broad themes. The first is how AI provides a new superpower, which is that it allows everyone to essentially code. The second is around how to use prompts. The third is using skills. The fourth is the importance of memory within your AI system. The fifth is around how to use co-work and other continuous agent systems. And finally, I'll talk about adding your own knowledge base or rag systems to your overall process. So here we go.

Now, one of the important things to note about large language models and AI, and this is for me is perhaps the most important thing about AI, is that essentially it democratizes coding and programming. So if we go back from historic perspective and you think about movies like Imitation Game with Benedict Cumberbatch, you'll see that in the 40s and 50s, computers were essentially these big machines where you had people poking the machines, inputting wires like transistors on the old radio, and telegraphy systems trying to get the computer to work. Then we moved to machine codes with ones and zeros. Then as time went on, we moved more towards language that looked closer to English when it came to coding. So in the 50s to 70s, we started to see the development of assembly language where you started to have, you could input words like MOV for move or ADD for add and so on. Then later, we had the development of object-orientated languages, which allowed you to compress complex concepts together. And then in the 90s, we had Python, which again allowed you to abstract higher language to do more coding tasks. However, despite all of that, you still needed to have some understanding of how to speak computer language. It's much like if everyone had to be able to speak Latin, to operate something. Not the whole population would be able to do that. So today, you have computer scientists who are trained, who can do the coding, but it does restrict the number of people that can use that. And in some ways, up until recently, the whole approach was to try to get humans to speak like machines. For me, the big revolution with large language models in AI is that finally machines can speak like humans. So what this means is, I, or anyone else, can just type in something in a prompt, like how you would speak naturalistically, and then the large language model will translate that into code. Maybe it's Python, then downstream, it would go into C, and then machine code. So this, for the first time, we're now able to have coding in natural English language. And for me, that is an underappreciated superpower that AI is now giving you. And this should then allow almost everybody to be able to code through using a large language model. At the moment, developers are using this at a very effective level. But I think going forward, almost everybody will start to do this. And as we saw in earlier years, when people were using Excel in financial markets a lot, in recent years, younger people tended to use Python. Now there's no excuse for anyone to now use large language models, which could then be used to create Python code, which in turn would allow you more easily to use machine learning techniques or advanced econometrics and so on. So this is really the first important principle to note, that you have no excuse not to code anymore, because coding is essentially your natural English language. So this is the first principle that I think is very important to understand. Now the second thing I think that is important is the importance of prompting. And what's important here is that there is a technique to good prompting, and there's all sorts of different techniques one can use when you prompt, but I use what's called a clear template, and it's an acronym, you can use other acronyms as well. So essentially what clear stands for is that when you do write a prompt, i.e. what you input into that box that appears when you open up Claude or ChatGBT.