Financial machine learning
Labelling, meta-labelling, fractional differentiation and the research tools that grew up around machine learning for markets.
- A laboratory for machine learning in finance: bars, labels and honest validation
A practical tour of the research pipeline for machine learning on market data: information-driven bars, the triple-barrier labelling method, sample weights for overlapping labels, purged cross-validation and feature importance, with the reasoning behind each step.
- Advances in Financial Machine Learning: a map of the techniques and the tools that implement them
A chapter-by-chapter map of Marcos López de Prado's Advances in Financial Machine Learning, from financial data structures to backtest statistics and hierarchical risk parity, with notes on which ideas are most used and how open-source implementations evolved.
- Does meta-labelling add to signal efficacy? How to test it honestly
A framework for testing whether meta-labelling improves a trading signal out of sample: choosing the right baseline, metrics beyond accuracy, purged validation, controls for volatility filtering, multiple testing and what the published evidence suggests.
- Dreaming, thinking machines: from automata to algorithms in the markets
An essay on the long history of the idea of thinking machines, from mechanical automata and Descartes to Turing's imitation game, ELIZA, the Chinese room and modern machine learning, and what that history suggests about the limits of algorithms in financial markets.
- Fractional differentiation: stationary features that keep their memory
Why the usual choice between prices and returns forces a trade-off between stationarity and memory, how fractional differencing with a non-integer order works, the binomial weights, the fixed-window method, choosing d with an ADF test, and pitfalls.
- Meta-labelling: a toy example to build intuition
A step-by-step toy example of meta-labelling: a primary model that picks the side, a secondary model that decides whether to act and how much, and how precision, recall and the F1 score change, using clearly hypothetical numbers.
