Let me show you a few instance of how trading has learned from AI research in other fields.
Long short-term memory network, a specific recurrent neural network architecture, had been developed for speech recognition and semantic analysis. It is breaking new ground in time series prediction in finance. Use of unspecified auto-encoders was the reason for the initial success in deep learning. Prior to that prediction networks used to be shallow.
We are now finding auto-encoders much more effective than factor models were in StatArb. This global ecosystem around AI is making it a tool that is growing very fast in its capabilities and it is becoming cheap to apply AI at scale. If I were to compare the merits of the AI-based portfolio management to hiring and training whiz kids to find alpha, I think:
- AI will be cheaper!
- There will be less specification risk with AI. By that, I mean the risk that the model works today but with minor changes in the trading environment it stops working.
- An AI based approach is somewhat self-learning and the ongoing costs will be lower.
This brings us to: why AI; why science; why the shift to method away from experts. A scientific approach to the problem of investing In 1952, Benjamin Graham wrote a paper “Towards a Science of Security Analysis”. In 1952, computers were scarce, but Graham wrote about “a trustworthy tool”. The last lines of the paper are “security analysis may begin – modesty, but hopefully – to refer to itself as a scientific discipline.”
This is the problem we are trying to solve. Globally, the asset management industry manages more than $164 trillion. Investors spend more than a trillion dollars, just in fees, but we aren’t getting close to solving the problem. The only way we can deliver high quality investing at low recurring cost is by using AI. In summary:
- Instead of championing traders, we need to make better machines.
- Instead of growing expert networks, we need to appreciate collaborative research.
- Instead of seeking access to top funds, we need to seek scientific methods.
(Links to the references made in this article will be in the online version on the GlobalTrading website)
We’d love to hear your feedback on this article. Please click here
