Machine Learning for Asset Managers (Elements in Quantitative Finance)

If you have seen one of Lopez de Prado's lectures, you will understand immediately that his work offers advances to standard problems in portfolio management. This small book summarizes the main contributions and cannot be recommended strongly enough.The book is comprised of several very dense chapters of all new material in the field. These include advances in covariance estimation based on random matrix theory, optimal clustering, labelling methods for trend tracking, using ML for variable selection for prediction, hierarchical risk parity (instead of Markowitz portfolio optimization) for portfolio selection, and two alternatives to standard sharpe-ratio back testing of strategies. All of these new methodologies address fundamental problems with the standard techniques that have been used for decades in portfolio theory and in sum represent a big step forward in the theory.This book is worth purchasing just to read the introduction. It gives great insight into how to use machine learning for finance that you won't find in any other book.The author devised a machine learning algorithm based on 'theories' of finance 'discovered' from machine learning, not based on back testing. This made all the difference during the flash crash. Unlike many others, his algorithm made lots of money during this period because of the way he designed it.This book will NOT teach you general machine learning techniques. As the author admits in the introduction, you must learn these things from other books. But the techniques here should be acquired by everyone in finance.

"Machine Learning for Asset Managers" is everything I had hoped.In this concise Element, De Prado succinctly distinguishes the practical uses of ML within Portfolio Management from the hype.Concepts are presented with clarity & relevant code is provided for the audiences’ purposes.This work serves as support for the argument that there are better systematic & objective methods for creating investment portfolios than the simplified heuristics of our past.For that, I am thankful.De Prado’s work will be remembered as the 21st Century’s “The Intelligent Investor” (Benjamin Graham).

In his "Advances in Financial Machine Learning" Marcos Lopez de Prado touches on multiple uses for asset managers, my personal favorite one was his asset allocation process where he compensates for what he calls the "Markowitz's curse".This new book though, it just expands on the best concepts of the other one, particularly for anyone interested in asset management. Just wow... An absolute fountain of useful and actionable information for any asset manager.It also happen to me one of the most beautifully printed books I've ever had.

Not very insightful and looks like a data scientist is presenting his/her day job. Good for people completely new to data science. Would be helpful if it could highlight what’s the difference or so special about asset management applications vs others

Marcos has brought attention to a number of interesting analysis techniques in the recent years.His previous book, Advances in Financial Machine Learning, was much more broad/technical in scope. I enjoyed this text because it is much more concise and lays out a clear entry point into purely data-driven investing.The book begins by outlining a powerful and intuitive technique which can be used to de-noise correlation matrices. Marcos provides strong validation for this methodology with stylized Monte Carlo.Using MC in this way is an innovative idea -- one which the physics community has embraced for decades. In particle physics we rely on Monte Carlo to remove reliance on empirically collected data and to validate our understanding. Applying this approach to finance is sure to bear fruit by bringing more objectivity to finance in the coming years.The book continues on in a similar manner to outline more clear and simple techniques which can yield similarly objective results.

Interesting, not because it contains new mathematical developments or ideas (most of the clustering related content is between 10 to 20 years old; same for the random matrix theory (RMT) part, which is already applied in many hedge funds and other sophisticated asset managers), but because all these existing ideas and methods (prevalent in the industry or in (some part of, e.g. econophysics rather than finance) academia) are, maybe for the first time, exposed clearly and linearly throughout the book. You may find RMT experts, clustering experts, optimizers in chief, and various frameworks to deal with multiple testing biases, but rarely all this knowledge is brought together in one place with a consistent presentation and set of notations. Marcos has achieved this, by clearly exposing all these elements (if not the most recent sophistications).

Eight concise chapters, each covering key ideas in portfolio management using ML techniques.

The topics in this book are interesting, but it is so full of errors, some very material, that it leads one to doubt the conclusions it makes. Results stated in the text often do not correspond to the code snippets they refer to. The code snippets are full of bugs, some just blatantly careless as to suggest they were never tried, and it leads me to think there was minimal editing done. Moreover, the code examples are stated in buggy Python 2. (Who uses Python 2 in 2020 when many cloud platforms are going so far as to discontinue support for it altogether?)