This book provides a comprehensive introduction to how ML can add value to the design and execution of trading strategies. It is organized into four parts that cover different aspects of the data sourcing and strategy development process, as well as different solutions to various ML challenges.
Part 1 – Data, alpha factors, and portfolios
The first part covers fundamental aspects relevant across trading strategies that leverage machine learning. It focuses on the data that drives the ML algorithms and strategies discussed in this book, outlines how you can engineer features that capture the data's signal content, and explains how to optimize and evaluate the performance of a portfolio.
Chapter 1, Machine Learning for Trading – From Idea to Execution, summarizes how and why ML became important for trading, describes the investment process, and outlines how ML can add value.
Chapter 2, Market and Fundamental Data – Sources and Techniques, covers how to source and work with market data, including exchange-provided tick data, and reported financials. It also demonstrates access to numerous open source data providers that we will rely on throughout this book.
Chapter 3, Alternative Data for Finance – Categories and Use Cases, explains categories and criteria to assess the exploding number of sources and providers. It also demonstrates how to create alternative datasets by scraping websites, for example, to collect earnings call transcripts for use with natural language processing (NLP) and sentiment analysis, which we cover in the second part of the book.
Chapter 4, Financial Feature Engineering – How to Research Alpha Factors, presents the process of creating and evaluating data transformations that capture the predictive signal and shows how to measure factor performance. It also summarizes insights from research into risk factors that aim to explain alpha in financial markets otherwise deemed to be efficient. Furthermore, it demonstrates how to engineer alpha factors using Python libraries offline and introduces the Zipline and Alphalens libraries to backtest factors and evaluate their predictive power.
Chapter 5, Portfolio Optimization and Performance Evaluation, introduces how to manage, optimize, and evaluate a portfolio resulting from the execution of a strategy. It presents risk metrics and shows how to apply them using the Zipline and pyfolio libraries. It also introduces methods to optimize a strategy from a portfolio risk perspective.
Part 2 – ML for trading – Fundamentals
The second part illustrates how fundamental supervised and unsupervised learning algorithms can inform trading strategies in the context of an end-to-end workflow.
Chapter 6, The Machine Learning Process, sets the stage by outlining how to formulate, train, tune, and evaluate the predictive performance of ML models in a systematic way. It also addresses domain-specific concerns, such as using cross-validation with financial time series to select among alternative ML models.
Chapter 7, Linear Models – From Risk Factors to Return Forecasts, shows how to use linear and logistic regression for inference and prediction and how to use regularization to manage the risk of overfitting. It demonstrates how to predict US equity returns or the direction of their future movements and how to evaluate the signal content of these predictions using Alphalens.
Chapter 8, The ML4T Workflow – From Model to Strategy Backtesting, integrates the various building blocks of the ML4T workflow thus far discussed separately. It presents an end-to-end perspective on the process of designing, simulating, and evaluating a trading strategy driven by an ML algorithm. To this end, it demonstrates how to backtest an ML-driven strategy in a historical market context using the Python libraries backtrader and Zipline.
Chapter 9, Time-Series Models for Volatility Forecasts and Statistical Arbitrage, covers univariate and multivariate time series diagnostics and models, including vector autoregressive models as well as ARCH/GARCH models for volatility forecasts. It also introduces cointegration and shows how to use it for a pairs trading strategy using a diverse set of exchange-traded funds (ETFs).
Chapter 10, Bayesian ML – Dynamic Sharpe Ratios and Pairs Trading, presents probabilistic models and how Markov chain Monte Carlo (MCMC) sampling and variational Bayes facilitate approximate inference. It also illustrates how to use PyMC3 for probabilistic programming to gain deeper insights into parameter and model uncertainty, for example, when evaluating portfolio performance.
Chapter 11, Random Forests – A Long-Short Strategy for Japanese Stocks, shows how to build, train, and tune nonlinear tree-based models for insight and prediction. It introduces tree-based ensembles and shows how random forests use bootstrap aggregation to overcome some of the weaknesses of decision trees. We then proceed to develop and backtest a long-short strategy for Japanese equities.
Chapter 12, Boosting Your Trading Strategy, introduces gradient boosting and demonstrates how to use the libraries XGBoost, LightBGM, and CatBoost for high-performance training and prediction. It reviews how to tune the numerous hyperparameters and interpret the model using SHapley Additive exPlanation (SHAP) values before building and evaluating a strategy that trades US equities based on LightGBM return forecasts.
Chapter 13, Data-Driven Risk Factors and Asset Allocation with Unsupervised Learning, shows how to use dimensionality reduction and clustering for algorithmic trading. It uses principal and independent component analysis to extract data-driven risk factors and generate eigenportfolios. It presents several clustering techniques and demonstrates the use of hierarchical clustering for asset allocation.
Part 3 – Natural language processing
Part 3 focuses on text data and introduces state-of-the-art unsupervised learning techniques to extract high-quality signals from this key source of alternative data.
Chapter 14, Text Data for Trading – Sentiment Analysis, demonstrates how to convert text data into a numerical format and applies the classification algorithms from Part 2 for sentiment analysis to large datasets.
Chapter 15, Topic Modeling – Summarizing Financial News, uses unsupervised learning to extract topics that summarize a large number of documents and offer more effective ways to explore text data or use topics as features for a classification model. It demonstrates how to apply this technique to earnings call transcripts sourced in Chapter 3 and to annual reports filed with the Securities and Exchange Commission (SEC).
Chapter 16, Word Embeddings for Earnings Calls and SEC Filings, uses neural networks to learn state-of-the-art language features in the form of word vectors that capture semantic context much better than traditional text features and represent a very promising avenue for extracting trading signals from text data.
Part 4 – Deep and reinforcement learning
Part 4 introduces deep learning and reinforcement learning.
Chapter 17, Deep Learning for Trading, introduces TensorFlow 2 and PyTorch, the most popular deep learning frameworks, which we will use throughout Part 4. It presents techniques for training and tuning, including regularization. It also builds and evaluates a trading strategy for US equities.
Chapter 18, CNNs for Financial Time Series and Satellite Images, covers convolutional neural networks (CNNs) that are very powerful for classification tasks with unstructured data at scale. We will introduce successful architectural designs, train a CNN on satellite data (for example, to predict economic activity), and use transfer learning to speed up training. We'll also replicate a recent idea to convert financial time series into a two-dimensional image format to leverage the built-in assumptions of CNNs.
Chapter 19, RNNs for Multivariate Time Series and Sentiment Analysis, shows how recurrent neural networks (RNNs) are useful for sequence-to-sequence modeling, including for univariate and multivariate time series to predict. It demonstrates how RNNs capture nonlinear patterns over longer periods using word embeddings introduced in Chapter 16 to predict returns based on the sentiment expressed in SEC filings.
Chapter 20, Autoencoders for Conditional Risk Factors and Asset Pricing, covers autoencoders for the nonlinear compression of high-dimensional data. It implements a recent paper that uses a deep autoencoder to learn both risk factor returns and factor loadings from the data while conditioning the latter on asset characteristics. We'll create a large US equity dataset with metadata and generate predictive signals.
Chapter 21, Generative Adversarial Networks for Synthetic Time-Series Data, presents one of the most exciting advances in deep learning. Generative adversarial networks (GANs) are capable of learning to reproduce synthetic replicas of a target data type, such as images of celebrities. In addition to images, GANs have also been applied to time-series data. This chapter replicates a novel approach to generate synthetic stock price data that could be used to train an ML model or backtest a strategy, and also evaluate its quality.
Chapter 22, Deep Reinforcement Learning – Building a Trading Agent, presents how reinforcement learning (RL) permits the design and training of agents that learn to optimize decisions over time in response to their environment. You will see how to create a custom trading environment and build an agent that responds to market signals using OpenAI Gym.
Chapter 23, Conclusions and Next Steps, summarizes the lessons learned and outlines several steps you can take to continue learning and building your own trading strategies.
Appendix, Alpha Factor Library, lists almost 200 popular financial features, explains their rationale, and shows how to compute them. It also evaluates and compares their performance in predicting daily stock returns.
