In our previous article AI in our daily lives, we talked about the benefits of using artificial intelligence daily. We mentioned that the same is true for investing.
Traditional machine learning methods “represent a significant extension of the quantitative investor’s toolbox, but they are not qualitatively different from traditional statistical methods,” according to an official paper from Acadian Asset Management.
Artificial intelligence, which complements traditional human models and techniques in most cases, is currently seen as an aide to human intelligence. However, the power of advanced artificial intelligence, like deep learning (DL) and deep reinforcement learning (DRL), lies in its ability to directly find patterns in data and make predictions regardless of human intelligence or experience. While some investment managers claim that these algorithms solve incredibly complex problems in medicine, autonomous driving, mechanical engineering, robotics and other fields, they vehemently deny that DL and DRL will solve investment problems and build independent investment strategies.
This denial might become their downfall.
DL and DRL algorithms look at the data, identifying patterns and similarities between the target and the data, and then use this knowledge to predict prices. An instructive example of powerful self-learning algorithms is DeepMind’s AlphaGo Zero, designed initially for the game of Go, an extremely complex Chinese board game in which the number of possible board positions exceeds the number of atoms in the known observable universe.
Unlike IBM Deep Blue – a human-developed hard-coded computer program created in the 1990s for chess, AlphaGo Zero ran the algorithm from scratch with no human data or engineering decisions and no knowledge of the field going beyond the rules of the game. A 2017 DeepMind blog post concludes that AlphaGo Zero used “a new form of reinforcement learning in which AlphaGo Zero becomes its own teacher”. The system starts with a neural network that knows nothing about the game of Go. It then plays games against itself, combining this neural network with a robust search algorithm. During the game, the neural network is tuned and updated to predict the moves and the possible winner of the game. “Throughout millions of self-playable games,” the system gradually learned the game of Go from scratch, accumulating human knowledge over thousands of years in just a few days. AlphaGo Zero also discovered new knowledge, developing unconventional strategies and new creative moves.
Although AlphaGo Zero was explicitly designed (but not programmed) to play Go, DeepMind reported that a later version of the program, AlphaZero, achieved similarly striking success in chess and shogi: “AlphaZero quickly learns each game to become the strongest player in history. Each algorithm starts its training with the basic rules of the game, without any built-in knowledge of the subject area.” The unprecedented success of these and other experiments led the DeepMind team to the general conclusion that reinforcement learning can be used to achieve superhuman results in other areas: “Our results comprehensively demonstrate that a pure approach to reinforcement learning is workable even in the most challenging areas: you can train to a superhuman level, without human examples.”
Models based on human intelligence can explain complexity and randomness better than any artificial intelligence. Still, it must be recognized that complexity and randomness are human concepts derived from observations made by human intelligence. Phenomena that people perceive as complex and random may not be perceived like that by DL and DRL systems. As Tom Simonite writes in Wired: “Artificial intelligence is alien intelligence that perceives and processes the world in ways that are fundamentally different from ours.”
There is no doubt that DL and DRL can process vast amounts of data due to their large capacity, more data usually leads to higher prediction accuracy.
Currently, robots manage $1.6 trillion out of the $80 trillion professionally managed assets globally. More than $9 trillion of assets are expected to be algorithmically executed by 2025. arty is based on integrating Big Data processing technology and artificial intelligence, as well as machine learning, quantitative analysis methods, and high-precision software. As a result of cooperation with arty you can stop looking for ‘hot’ stocks, the elusive unicorns that deliver exceptional returns. Instead you will get professional advice on how to build an optimized portfolio that gives you sustained, long-term capital appreciation with minimum risk.
Adapted from Institutional Investor, Angelo Calvello, PhD, Co-Founder of Rosetta Analytics.