Marcos López de Prado > Quotes

“Supervised learning algorithms typically require stationary features. The reason is that we need to map a previously unseen (unlabeled) observation to a collection of labeled examples, and infer from them the label of that new observation. If the features are not stationary, we cannot map the new observation to a large number of known examples. But stationary does not ensure predictive power. Stationarity is a necessary, non-sufficient condition for the high performance of an ML algorithm. The problem is, there is a trade-off between stationarity and memory. We can always make a series more stationary through differentiation, but it will be at the cost of erasing some memory, which will defeat the forecasting purpose of the ML algorithm.”
― Marcos Lopez de Prado, Advances in Financial Machine Learning

“Econometrics is the application of classical statistical methods to economic and financial series. The essential tool of econometrics is multivariate linear regression, an 18th-century technology that was already mastered by Gauss before 1794. Standard econometric models do not learn. It is hard to believe that something as complex as 21st-century finance could be grasped by something as simple as inverting a covariance matrix.”
― Marcos Lopez de Prado, Advances in Financial Machine Learning

“Many investment managers believe that the secret to riches is to implement an extremely complex ML algorithm. They are setting themselves up for a disappointment. If it was as easy as coding a state-of-the-art classifier, most people in Silicon Valley would be billionaires.”
― Marcos Lopez de Prado, Advances in Financial Machine Learning

“Rather than imposing a functional form, particularly when that form is unknown ex ante, they would allow algorithms to figure out variable dependencies from the data. And rather than making strong assumptions on the data, the algorithms would conduct experiments that evaluate the mathematical properties of out-of-sample predictions. This relaxation in terms of functional form and data assumptions, combined with the use of powerful computers, opened the door to analyzing complex data sets, including highly nonlinear, hierarchical, and noncontinuous interaction effects.”
― Marcos Lopez de Prado, Machine Learning for Asset Managers

“the instability caused by covariance structure can be measured in terms of the magnitude between the two extreme eigenvalues. Accordingly, the condition number of a covariance or correlation (or normal, thus diagonalizable) matrix is defined as the absolute value of the ratio between its maximal and minimal (by moduli) eigenvalues.”
― Marcos Lopez de Prado, Machine Learning for Asset Managers

“Dollar bars are formed by sampling an observation every time a pre-defined market value is exchanged. Of course, the reference to dollars is meant to apply to the currency in which the security is denominated, but nobody refers to euro bars, pound bars, or yen bars (although gold bars would make for a fun pun).”
― Marcos Lopez de Prado, Advances in Financial Machine Learning

“Given the finite and nondeterministic nature of these observations, the estimate of the covariance matrix includes some amount of noise. Empirical covariance matrices derived from estimated factors are also numerically ill-conditioned, because those factors are also estimated from flawed data. Unless we treat this noise, it will impact the calculations we perform with the covariance matrix, sometimes to the point of rendering the analysis useless.”
― Marcos Lopez de Prado, Machine Learning for Asset Managers