Machine Learning for Algorithmic Trading: Predictive models to extract signals from market and alternative data for systematic trading strategies with Python

What are the key takeaways from your book?

Using machine learning for trading poses several unique challenges: first, fierce competition due to potentially high rewards in highly efficient market limits the predictive signal in historical market data. Therefore, data becomes the single most important ingredient for a predictive model and requires careful sourcing and handling. In addition, domain expertise is key to realizing the value contained in data through smart feature engineering while avoiding some of the pitfalls of using ML.

Furthermore, ML for trading requires a workflow that integrates predictive modeling with decision making. Many books on ML show how to make good predictions, but to succeed in trading we need to translate predictions into a profitable strategy of buying and selling assets. While we should always keep the ultimate use case of an ML application in mind during development, the opportunities and methodological challenges of backtesting are fairly unique to the trading domain.