In algorithmic trading, strategies are typically based on indicators computed from price curves. These indicators, e.g. leading or lagging moving averages or RSI, can be viewed as features that were extracted from the curve. Typically a trader formulates a strategy by translating these features into buying or selling decisions. This is difficult to get right in order to maximise the return, especially when the number of indicators is large.
This white paper describes how machine learning, in particular random forests, can be utilised to automatically find the best algorithmic trading strategy based on a (potentially large) set of indicators. We further demonstrate how the system can be trained efficiently, using both CPUs and GPUs.
- Optimised algorithmic trading strategies
- Using machine learning: random forests
- Maximising trading returns
- Handles large number of indicators
- Fast training on CPUs and GPUs