\( \newcommand{\bm}[1]{\boldsymbol{#1}} \newcommand{\textm}[1]{\textsf{#1}} \def\T{{\mkern-2mu\raise-1mu\mathsf{T}}} \newcommand{\R}{\mathbb{R}} % real numbers \newcommand{\E}{{\rm I\kern-.2em E}} \newcommand{\w}{\bm{w}} % bold w \newcommand{\bmu}{\bm{\mu}} % bold mu \newcommand{\bSigma}{\bm{\Sigma}} % bold mu \newcommand{\bigO}{O} %\mathcal{O} \renewcommand{\d}[1]{\operatorname{d}\!{#1}} \)

Chapter 16 Deep Learning Portfolios

“Robots took the jobs of factory workers. Artificial intelligence will take the jobs of PMs.”

— Anonymous

Artificial intelligence (AI) is a very broad term that, roughly speaking, refers to the general ability of computers to emulate human thought and perform tasks in real-world environments. Machine learning (ML) is a subset of AI that refers to the technologies and algorithms that enable systems to identify patterns, make decisions, and improve themselves through experience and data. Particularly, neural networks (NN) are flexible architectures that attempt to emulate the structure of the neurons in the human brain and provide very powerful tools for systems to learn automatically from examples. This is in opposition to the traditional approach in computer science where the machine is already programmed beforehand to perform one specific task, such as the portfolio formulations considered in this book.

More specifically, deep neural networks, also broadly referred to as deep learning (DL), have ignited a revolution in many domain-specific areas with an outstanding performance, putting previous traditional methods to shame in terms of performance. Some areas revolutionized by deep neural networks (in the sense of achieving close-to-human or superhuman performance) include:

  • image recognition: superhuman performance (this goes from simple recognition of cats to sophisticated cancer detection from X-rays);
  • natural language processing (NLP): human-level performance;
  • board and video games: superhuman performance (e.g., Chess, Go, and Atari video games);
  • video processing: close to human performance (e.g., real-time video navigation in drones);
  • protein folding: superhuman performance;
  • self-driving cars: not yet superhuman or even human-level, but soon-to-arrive;
  • professional and academic benchmarks: human-level performance (e.g., passing a simulated bar exam with a score around the top 10% of test takers).

The million-dollar question is whether this revolution will extend to financial systems. To start with, there are many ways in which DL can be used in financial systems. In fact, it has already been successfully employed in some specific problems, such as sentiment analysis of news, credit default detection, or satellite image analysis to detect stock levels or crop production related to companies.

Following the theme of this book, we will investigate how DL can be applied in portfolio design. That is, whether one can design a black box that takes as input some financial data and outputs the portfolio to be used. At the time this book was written, the subject was still developing, and the answer to this question remained unclear. In fact, numerous papers were being published, and various open-source software libraries were becoming available.

This material will be published by Cambridge University Press as Portfolio Optimization: Theory and Application by Daniel P. Palomar. This pre-publication version is free to view and download for personal use only; not for re-distribution, re-sale, or use in derivative works. © Daniel P. Palomar 2024.