Saturday, September 02, 2023
003 - Lessons from Artificial Intelligence (Part II)
In the second episode of this subseries, I recount the beginnings of my interest in artificial intelligence, my early experimentation with artificial neural networks, and how this experimentation came to intersect with finance.
Duration: 00:02:50
Episode Transcript
Good evening. You’re listening to the Reflections in Beige Podcast, hosted by Michael LeSane. The date is the first of September, 2023.
My interest in machine learning, or artificial intelligence, began during my final year of high school. What started with using Markov chains to probabilistically model natural language in chatbots would blossom into a full-fledged obsession with computational linguistics during my college years.
Though I had a passing familiarity with the concept of artificial neural networks during those years, it wasn’t until my final year of undergrad that I actually began to tinker with them, porting a simple library to my language of choice… PHP… which I’d further develop.
My experimentation started with simple things like logic gates, before graduating to areas like handwritten character recognition and simple arithmetic, though I was not too successful with the latter.
It was, however, the application of neural networks to algorithmic trading simulations – or, that is to say, letting them loose on the stock market – that proved to be a very interesting match.
I had previously tinkered with more rigid and statistical approaches to algorithmic trading simulations during the prior summer or two, but neural networks potentially held the key to accessing more novel insights in financial data.
Observing rapid oscillation in after-market prices late into the night, which I presumed to be driven by high-frequency trading systems plugged directly into exchanges on Wall Street and running 24/7, sent the point home that the only profitable opportunity for those without the proper connections, so to speak, is some form of information arbitrage.
I proceeded to query and scrape raw financial data from various public sources, which I then normalized – or converted to a form which could be used as neural network inputs – and used the percentage in price change after fifteen minutes as the output and the target for training.
The neural network model itself was instantiated and trained using the aforementioned library I was developing, and plugged directly into the trading simulation infrastructure I’d developed during the previous summer.
With that, the simulation was ready for trial sessions.
Whether my assessment of the nature of the price patterns was correct or not, information arbitrage has, in my experience, consistently proven to be one of the two most valuable assets to have, professionally speaking. The other is capacity for execution based on that information. This lesson transcends finance, and even business, and is applicable to any sphere, professional or otherwise, with a counterparty or a competitive element.
Let that be the lesson for tonight.
This subseries will resume the episode after next, as I have an interview scheduled for this program tomorrow. I hope you’ll join us for a musical edition of the Reflections in Beige Podcast.
My interest in machine learning, or artificial intelligence, began during my final year of high school. What started with using Markov chains to probabilistically model natural language in chatbots would blossom into a full-fledged obsession with computational linguistics during my college years.
Though I had a passing familiarity with the concept of artificial neural networks during those years, it wasn’t until my final year of undergrad that I actually began to tinker with them, porting a simple library to my language of choice… PHP… which I’d further develop.
My experimentation started with simple things like logic gates, before graduating to areas like handwritten character recognition and simple arithmetic, though I was not too successful with the latter.
It was, however, the application of neural networks to algorithmic trading simulations – or, that is to say, letting them loose on the stock market – that proved to be a very interesting match.
I had previously tinkered with more rigid and statistical approaches to algorithmic trading simulations during the prior summer or two, but neural networks potentially held the key to accessing more novel insights in financial data.
Observing rapid oscillation in after-market prices late into the night, which I presumed to be driven by high-frequency trading systems plugged directly into exchanges on Wall Street and running 24/7, sent the point home that the only profitable opportunity for those without the proper connections, so to speak, is some form of information arbitrage.
I proceeded to query and scrape raw financial data from various public sources, which I then normalized – or converted to a form which could be used as neural network inputs – and used the percentage in price change after fifteen minutes as the output and the target for training.
The neural network model itself was instantiated and trained using the aforementioned library I was developing, and plugged directly into the trading simulation infrastructure I’d developed during the previous summer.
With that, the simulation was ready for trial sessions.
Whether my assessment of the nature of the price patterns was correct or not, information arbitrage has, in my experience, consistently proven to be one of the two most valuable assets to have, professionally speaking. The other is capacity for execution based on that information. This lesson transcends finance, and even business, and is applicable to any sphere, professional or otherwise, with a counterparty or a competitive element.
Let that be the lesson for tonight.
This subseries will resume the episode after next, as I have an interview scheduled for this program tomorrow. I hope you’ll join us for a musical edition of the Reflections in Beige Podcast.