Research review · López de Prado

López de Prado, Reproduction & Review

A from-scratch reproduction and at-scale extension of Marcos López de Prado's methods across crypto, US equities and forex, every empirical claim gated by the Deflated Sharpe Ratio

This is a reproduction-and-extension program built from scratch around the methods in Marcos López de Prado's Advances in Financial Machine Learning and the surrounding papers, information-driven bars, fractional differentiation, financial labeling and cross-validation, ensembles and feature importance, portfolio construction, causal factors, predictive features, and bet sizing. Each study reproduces the original claim on a controlled benchmark, then re-runs it at scale on real, fully-costed, no-look-ahead data spanning crypto, US equities and forex, with every empirical headline gated by the Deflated Sharpe Ratio.

The honest through-line is simple: almost nothing clears deflated significance anywhere , which is precisely López de Prado's thesis, while several of his methodological claims reproduce cleanly: the activity clock really does thin the tails, fractional differencing really does keep the memory, bagging really does generalize where boosting does not, and a confounder really can manufacture a factor out of nothing. These are methodology results, not money results, and they are framed that way throughout.