Menu
At Deep Market Making our mission is build AI analysts that help all credit professionals gain insight, confidence, and empowerment.
The text included below is a sample excerpt from the Deep Market Making White Paper – Fair Market Value & Chance to Fill Models for OTC Fixed Income Markets
For a copy of the full white paper please contact Nathaniel Powell sales@deepmm.com Founder & CEO, Deep Market Making
Everyone on Earth who wants to will be able to raise their mental understanding of their world, relationships, art, science, emotions, personal fulfillment, with some effort,
but standing on the shoulders of various AI systems. How this works is rooted in how our human mind works, and likewise, how modern artificial Neural Networks (NN) function.
For traders, we want to help elevate their ability to think strategically, better seeing the forest from the trees. In other words, spending less time analyzing raw market data,
and more time thinking strategically, talking to clients, and ultimately profiting as a trader because the AI is helping perform the low-level processing of the market data,
surfacing the relevant information when and where it’s needed. Out of the machine learning paradigms available today, only deep learning has the capability to learn from
traders how to surface the right information, as all other ML algorithms are just too simple to even begin to approach the mental complexity of the mind of a trader.
Different tissues in the brain developed during different stages of evolution. The parts of our brain we share in common with reptiles include, among others, the emotional centers like the amygdala, and the cerebellum, which helps us to subconsciously plan our movement by solving differential equations , and these parts originally evolved hundreds of millions of years ago. The mammalian tissue, the neocortex, is the largest portion of our human brains, and is much larger than in other great apes. The reason the size of the neocortex makes such a large difference in our intelligence versus other animals is that it allows much deeper layers, which allow for much more abstract reasoning with high-level concepts than other animals can achieve.
The neocortex is made up of about 300 MM pattern recognizing units, arranged in a vast network of parallel hierarchies. Each pattern recognizing unit has about 100 output neurons that light up in different codes depending on which pattern has been recognized (each unit can recognize many different patterns). Different parts of your brain may be responsible for processing different types of information, but the remarkable thing is that every pattern recognizer is functionally equivalent. The reason the neocortex ends up being organized in certain ways is simply because of how inputs and other brain anatomy is wired to it. For example, the visual cortex is in the back of your head typically because your eyes are connected to the back of the neocortex.
Pattern recognizers which happen to be near sensory inputs like our optic nerves recognize low-level patterns – such as lines. The output of the these pattern recognizers are fed to the next layer in the hierarchy to recognize a slightly more abstract concept – such as shapes. Then parts of faces, for example, then whole faces, then expressions, then a high level concept like ”this person is happy.” Then this might feed into your understanding of a social situation, such as ”Mike pulled a funny prank on his buddy and is giddy.” Since there are many high level concepts that you can recognize, there are many such hierarchies that all share lower level pattern recognizers with each other.
As alluded to earlier, the limit to how abstract, high level our thinking can become is limited to the number of layers in our concept hierarchies our brains are capable of wiring. Therefore, if we want to elevate our thinking even higher, we need mental prosthetics of one kind or another. In the past those prosthetics often took the form of other people; in the financial world you might have a team of analysts, associates, junior traders, etc. who can help you process low level information so that you can think of the big picture. The problem with this model is that you can only scale it so far before it becomes prohibitively expensive.
You might be thinking – what about all of the quant finance models, can’t they be a mental prosthetic as well? Yes – they can be and they are to a very small degree. The problem lies in the lack of complexity of these models, and the profound and untrue assumptions they make about the markets. For example, simple economic models often assume that there is perfect competition, or that the markets have already reached equilibrium, because those assumptions allow a simple equation to function. But there is a reason that we don’t hand high-notional OTC trading over to simple formulas and heuristics – they are just too simplistic and would too easily be taken advantage of, unlike the typical human trader who is constantly learning and evolving, with a truly giant 10 to 100 trillion synapse neural network in his head helping him understand the market complexity.
However, now we have a new option for scaling our teams’ information processing capabilities, with the advent of robust deep learning models. Similarly to biological neural networks, the most effective neural networks are very deep, with many layers, and their processing capability is increasing exponentially year after year through improving hardware efficiency, software efficiency, and larger amounts of capital being devoted to training the largest networks. So now every member of your credit team, including the analysts, can now stand on top of many more layers of analysis performed by deep neural networks, which will allow you as a leader to see the forest from the trees at a true 10,000 foot perspective, whereas now you may be stuck below the ridge under a canopy thinking about raw incoming, publicly available data points, rather than focusing on the unique information you have in your possession which will allow you and your team to outperform your competitors.
Just as the evolution of our larger brains allowed us to develop abstract reasoning, mathematics, science, art, and music, all of which our great ape cousins are incapable of, think of the abilities you will be able to invest in as a team if you have a much deeper understanding of your market environment at all times, as Ray Kurzweil, the futurist, explains in his many speeches regarding his (accurate) predictions of the future of technology. At Deep MM our machine learning algorithms primarily learn from the traders via the market data they generate, and although these models are still tiny compared to the mind of a human trader, they are at least orders of magnitude more complex than the traditional ML models employed by our competitors as of this writing.
…
If you see the world the way we do, and want to share in our mission, please reach out and set up a time to chat!