CHAPTER 4Artificial Intelligence Techniques

Most of the trading strategies in this book are developed based on the process a theoretical physicist would use. Where a theoretical physicist develops a hypothesis, designs an experiment to test that hypothesis, and confirms the hypothesis based on the test results, we quantitative traders develop a hunch about a possible inefficiency in the market (e.g., retail investors' herd‐like behavior leading to stock momentum), devise a strategy to exploit that inefficiency, and use data to confirm whether that strategy actually works. The use of artificial intelligence (AI) or machine learning techniques is closer to the approach experimental physicists might take in their work: we don't have a preconceived theory of what the most important factors affecting the market are, and therefore, we need to explore as many factors and trading rules as possible with the help of efficient algorithms. Finance practitioners often derisively refer to this methodology as data mining. There is some justification of their derision: financial data are not only quite limited (unless we use tick data), they are also not very stationary in the statistical sense. That is, the probability distribution of returns does not stay constant forever. If we just turn our machine learning algorithms loose on these data, it is very easy to come up with trading rules that worked extremely well in certain past periods, but fail terribly going forward. Of course, even handcrafted ...

Get Machine Trading now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.