Machine Learning for Algorithmic Trading: Predictive models to extract signals from market and alternative data for systematic trading strategies with Python, 2nd Edition

Cover
Copyright
Packt Page
Contributors
Table of Contents
Preface
Chapter 1: Machine Learning for Trading – From Idea to Execution
	The rise of ML in the investment industry
		From electronic to high-frequency trading
		Factor investing and smart beta funds
		Algorithmic pioneers outperform humans
		ML and alternative data
		Crowdsourcing trading algorithms
	Designing and executing an ML-driven strategy
		Sourcing and managing data
		From alpha factor research to portfolio management
		Strategy backtesting
	ML for trading – strategies and use cases
		The evolution of algorithmic strategies
		Use cases of ML for trading
	Summary
Chapter 2: Market and Fundamental Data – Sources and Techniques
	Market data reflects its environment
		Market microstructure – the nuts and bolts
		How to trade – different types of orders
		Where to trade – from exchanges to dark pools
	Working with high-frequency data
		How to work with Nasdaq order book data
		Communicating trades with the FIX protocol
		The Nasdaq TotalView-ITCH data feed
		From ticks to bars – how to regularize market data
		AlgoSeek minute bars – equity quote and trade data
	API access to market data
		Remote data access using pandas
		yfinance – scraping data from Yahoo! Finance
		Quantopian
		Zipline
		Quandl
		Other market data providers
	How to work with fundamental data
		Financial statement data
		Other fundamental data sources
	Efficient data storage with pandas
	Summary
Chapter 3: Alternative Data for Finance – Categories and Use Cases
	The alternative data revolution
	Sources of alternative data
		Individuals
		Business processes
		Sensors
	Criteria for evaluating alternative data
		Quality of the signal content
		Quality of the data
		Technical aspects
	The market for alternative data
		Data providers and use cases
	Working with alternative data
		Scraping OpenTable data
		Scraping and parsing earnings call transcripts
	Summary
Chapter 4: Financial Feature Engineering – How to Research Alpha Factors
	Alpha factors in practice – from data to signals
	Building on decades of factor research
		Momentum and sentiment – the trend is your friend
		Value factors – hunting fundamental bargains
		Volatility and size anomalies
		Quality factors for quantitative investing
	Engineering alpha factors that predict returns
		How to engineer factors using pandas and NumPy
		How to use TA-Lib to create technical alpha factors
		Denoising alpha factors with the Kalman filter
		How to preprocess your noisy signals using wavelets
	From signals to trades – Zipline for backtests
		How to backtest a single-factor strategy
		Combining factors from diverse data sources
	Separating signal from noise with Alphalens
		Creating forward returns and factor quantiles
		Predictive performance by factor quantiles
		The information coefficient
		Factor turnover
	Alpha factor resources
		Alternative algorithmic trading libraries
	Summary
Chapter 5: Portfolio Optimization and Performance Evaluation
	How to measure portfolio performance
		Capturing risk-return trade-offs in a single number
		The fundamental law of active management
	How to manage portfolio risk and return
		The evolution of modern portfolio management
		Mean-variance optimization
		Alternatives to mean-variance optimization
		Risk parity
		Risk factor investment
		Hierarchical risk parity
	Trading and managing portfolios with Zipline
		Scheduling signal generation and trade execution
		Implementing mean-variance portfolio optimization
	Measuring backtest performance with pyfolio
		Creating the returns and benchmark inputs
		Walk-forward testing – out-of-sample returns
	Summary
Chapter 6: The Machine Learning Process
	How machine learning from data works
		The challenge – matching the algorithm to the task
		Supervised learning – teaching by example
		Unsupervised learning – uncovering useful patterns
		Reinforcement learning – learning by trial and error
	The machine learning workflow
		Basic walkthrough – k-nearest neighbors
		Framing the problem – from goals to metrics
		Collecting and preparing the data
		Exploring, extracting, and engineering features
		Selecting an ML algorithm
		Design and tune the model
		How to select a model using cross-validation
		How to implement cross-validation in Python
		Challenges with cross-validation in finance
		Parameter tuning with scikit-learn and Yellowbrick
	Summary
Chapter 7: Linear Models – From Risk Factors to Return Forecasts
	From inference to prediction
	The baseline model – multiple linear regression
		How to formulate the model
		How to train the model
		The Gauss–Markov theorem
		How to conduct statistical inference
		How to diagnose and remedy problems
	How to run linear regression in practice
		OLS with statsmodels
		Stochastic gradient descent with sklearn
	How to build a linear factor model
		From the CAPM to the Fama–French factor models
		Obtaining the risk factors
		Fama–Macbeth regression
	Regularizing linear regression using shrinkage
		How to hedge against overfitting
		How ridge regression works
		How lasso regression works
	How to predict returns with linear regression
		Preparing model features and forward returns
		Linear OLS regression using statsmodels
		Linear regression using scikit-learn
		Ridge regression using scikit-learn
		Lasso regression using sklearn
		Comparing the quality of the predictive signals
	Linear classification
		The logistic regression model
		How to conduct inference with statsmodels
		Predicting price movements with logistic regression
	Summary
Chapter 8: The ML4T Workflow – From ML Model to Strategy Backtest
	How to backtest an ML-driven strategy
	Backtesting pitfalls and how to avoid them
		Getting the data right
		Getting the simulation right
		Getting the statistics right
	How a backtesting engine works
		Vectorized versus event-driven backtesting
		Key implementation aspects
	backtrader – a flexible tool for local backtests
		Key concepts of backtrader's Cerebro architecture
		How to use backtrader in practice
		backtrader summary and next steps
	Zipline – scalable backtesting by Quantopian
		Calendars and the Pipeline for robust simulations
		Ingesting your own bundles with minute data
		The Pipeline API – backtesting an ML signal
		How to train a model during the backtest
		Instead of How to use
	Summary
Chapter 9: Time-Series Models for Volatility Forecasts and Statistical Arbitrage
	Tools for diagnostics and feature extraction
		How to decompose time-series patterns
		Rolling window statistics and moving averages
		How to measure autocorrelation
	How to diagnose and achieve stationarity
		Transforming a time series to achieve stationarity
		Handling instead of How to handle
		Time-series transformations in practice
	Univariate time-series models
		How to build autoregressive models
		How to build moving-average models
		How to build ARIMA models and extensions
		How to forecast macro fundamentals
		How to use time-series models to forecast volatility
	Multivariate time-series models
		Systems of equations
		The vector autoregressive (VAR) model
		Using the VAR model for macro forecasts
	Cointegration – time series with a shared trend
		The Engle-Granger two-step method
		The Johansen likelihood-ratio test
	Statistical arbitrage with cointegration
		How to select and trade comoving asset pairs
		Pairs trading in practice
		Preparing the strategy backtest
		Backtesting the strategy using backtrader
		Extensions – how to do better
	Summary
Chapter 10: Bayesian ML – Dynamic Sharpe Ratios and Pairs Trading
	How Bayesian machine learning works
		How to update assumptions from empirical evidence
		Exact inference – maximum a posteriori estimation
		Deterministic and stochastic approximate inference
	Probabilistic programming with PyMC3
		Bayesian machine learning with Theano
		The PyMC3 workflow: predicting a recession
	Bayesian ML for trading
		Bayesian Sharpe ratio for performance comparison
		Bayesian rolling regression for pairs trading
		Stochastic volatility models
	Summary
Chapter 11: Random Forests – A Long-Short Strategy for Japanese Stocks
	Decision trees – learning rules from data
		How trees learn and apply decision rules
		Decision trees in practice
		Overfitting and regularization
		Hyperparameter tuning
	Random forests – making trees more reliable
		Why ensemble models perform better
		Boostrap aggregation
		How to build a random forest
		How to train and tune a random forest
		Feature importance for random forests
		Out-of-bag testing
		Pros and cons of random forests
	Long-short signals for Japanese stocks
		The data – Japanese equities
		The ML4T workflow with LightGBM
		The strategy – backtest with Zipline
	Summary
Chapter 12: Boosting Your Trading Strategy
	Getting started – adaptive boosting
		The AdaBoost algorithm
		Using AdaBoost to predict monthly price moves
	Gradient boosting – ensembles for most tasks
		How to train and tune GBM models
		How to use gradient boosting with sklearn
	Using XGBoost, LightGBM, and CatBoost
		How algorithmic innovations boost performance
	A long-short trading strategy with boosting
		Generating signals with LightGBM and CatBoost
		Inside the black box - interpreting GBM results
		Backtesting a strategy based on a boosting ensemble
		Lessons learned and next steps
	Boosting for an intraday strategy
		Engineering features for high-frequency data
		Minute-frequency signals with LightGBM
		Evaluating the trading signal quality
	Summary
Chapter 13: Data-Driven Risk Factors and Asset Allocation with Unsupervised Learning
	Dimensionality reduction
		The curse of dimensionality
		Linear dimensionality reduction
		Manifold learning – nonlinear dimensionality reduction
	PCA for trading
		Data-driven risk factors
		Eigenportfolios
	Clustering
		k-means clustering
		Hierarchical clustering
		Density-based clustering
		Gaussian mixture models
	Hierarchical clustering for optimal portfolios
		How hierarchical risk parity works
		Backtesting HRP using an ML trading strategy
	Summary
Chapter 14: Text Data for Trading – Sentiment Analysis
	ML with text data – from language to features
		Key challenges of working with text data
		The NLP workflow
		Applications
	From text to tokens – the NLP pipeline
		NLP pipeline with spaCy and textacy
		NLP with TextBlob
	Counting tokens – the document-term matrix
		The bag-of-words model
		Document-term matrix with scikit-klearn
		Key lessons instead of lessons learned
	NLP for trading
		The naive Bayes classifier
		Classifying news articles
		Sentiment analysis with Twitter and Yelp data
	Summary
Chapter 15: Topic Modeling – Summarizing Financial News
	Learning latent topics – Goals and approaches
		How to implement LSI using sklearn
		Strengths and limitations
	Probabilistic latent semantic analysis
		How to implement pLSA using sklearn
		Strengths and limitations
	Latent Dirichlet allocation
		How LDA works
		How to evaluate LDA topics
		How to implement LDA using sklearn
		How to visualize LDA results using pyLDAvis
		How to implement LDA using Gensim
	Modeling topics discussed in earnings calls
		Data preprocessing
		Model training and evaluation
		Running experiments
	Topic modeling for with financial news
	Summary
Chapter 16: Word Embeddings for Earnings Calls and SEC Filings
	How word embeddings encode semantics
		How neural language models learn usage in context
		word2vec – scalable word and phrase embeddings
		Evaluating embeddings using semantic arithmetic
	How to use pretrained word vectors
		GloVe – Global vectors for word representation
	Custom embeddings for financial news
		Preprocessing – sentence detection and n-grams
		The skip-gram architecture in TensorFlow 2
		Visualizing embeddings using TensorBoard
		How to train embeddings faster with Gensim
	word2vec for trading with SEC filings
		Preprocessing – sentence detection and n-grams
		Model training
	Sentiment analysis using doc2vec embeddings
		Creating doc2vec input from Yelp sentiment data
		Training a doc2vec model
		Training a classifier with document vectors
		Lessons learned and next steps
	New frontiers – pretrained transformer models
		Attention is all you need
		BERT – towards a more universal language model
		Trading on text data – lessons learned and next steps
	Summary
Chapter 17: Deep Learning for Trading
	Deep learning – what's new and why it matters
		Hierarchical features tame high-dimensional data
		DL as representation learning
		How DL relates to ML and AI
	Designing an NN
		A simple feedforward neural network architecture
		Key design choices
		How to regularize deep NNs
		Training faster – optimizations for deep learning
		Summary – how to tune key hyperparameters
	A neural network from scratch in Python
		The input layer
		The hidden layer
		The output layer
		Forward propagation
		The cross-entropy cost function
		How to implement backprop using Python
	Popular deep learning libraries
		Leveraging GPU acceleration
		How to use TensorFlow 2
		How to use TensorBoard
		How to use PyTorch 1.4
		Alternative options
	Optimizing an NN for a long-short strategy
		Engineering features to predict daily stock returns
		Defining an NN architecture framework
		Cross-validating design options to tune the NN
		Evaluating the predictive performance
		Backtesting a strategy based on ensembled signals
		How to further improve the results
	Summary
Chapter 18: CNNs for Financial Time Series and Satellite Images
	How CNNs learn to model grid-like data
		From hand-coding to learning filters from data
		How the elements of a convolutional layer operate
		The evolution of CNN architectures: key innovations
	CNNs for satellite images and object detection
		LeNet5 – The first CNN with industrial applications
		AlexNet – reigniting deep learning research
		Transfer learning – faster training with less data
		Object detection and segmentation
		Object detection in practice
	CNNs for time-series data – predicting returns
		An autoregressive CNN with 1D convolutions
		CNN-TA – clustering time series in 2D format
	Summary
Chapter 19: RNNs for Multivariate Time Series and Sentiment Analysis
	How recurrent neural nets work
		Unfolding a computational graph with cycles
		Backpropagation through time
		Alternative RNN architectures
		How to design deep RNNs
		The challenge of learning long-range dependencies
		Gated recurrent units
	RNNs for time series with TensorFlow 2
		Univariate regression – predicting the S&P 500
		How to get time series data into shape for an RNN
		Stacked LSTM – predicting price moves and returns
		Multivariate time-series regression for macro data
	RNNs for text data
		LSTM with embeddings for sentiment classification
		Sentiment analysis with pretrained word vectors
		Predicting returns from SEC filing embeddings
	Summary
Chapter 20: Autoencoders for Conditional Risk Factors and Asset Pricing
	Autoencoders for nonlinear feature extraction
		Generalizing linear dimensionality reduction
		Convolutional autoencoders for image compression
		Managing overfitting with regularized autoencoders
		Fixing corrupted data with denoising autoencoders
		Seq2seq autoencoders for time series features
		Generative modeling with variational autoencoders
	Implementing autoencoders with TensorFlow 2
		How to prepare the data
		One-layer feedforward autoencoder
		Feedforward autoencoder with sparsity constraints
		Deep feedforward autoencoder
		Convolutional autoencoders
		Denoising autoencoders
	A conditional autoencoder for trading
		Sourcing stock prices and metadata information
		Computing predictive asset characteristics
		Creating the conditional autoencoder architecture
		Lessons learned and next steps
	Summary
Chapter 21: Generative Adversarial Nets for Synthetic Time-Series Data
	Creating synthetic data with GANs
		Comparing generative and discriminative models
		Adversarial training – a zero-sum game of trickery
		The rapid evolution of the GAN architecture zoo
		GAN applications to images and time-series data
	How to build a GAN using TensorFlow 2
		Building the generator network
		Creating the discriminator network
		Setting up the adversarial training process
		Evaluating the results
	TimeGAN for synthetic financial data
		Learning to generate data across features and time
		Implementing TimeGAN using TensorFlow 2
		Evaluating the quality of synthetic time-series data
		Lessons learned and next steps
	Summary
Chapter 22: Deep Reinforcement Learning – Building a Trading Agent
	Elements of a reinforcement learning system
		The policy – translating states into actions
		Rewards – learning from actions
		The value function – optimal choice for the long run
		With or without a model – look before you leap?
	How to solve reinforcement learning problems
		Key challenges in solving RL problems
		Fundamental approaches to solving RL problems
	Solving dynamic programming problems
		Finite Markov decision problems
		Policy iteration
		Value iteration
		Generalized policy iteration
		Dynamic programming in Python
	Q-learning – finding an optimal policy on the go
		Exploration versus exploitation – -greedy policy
		The Q-learning algorithm
		How to train a Q-learning agent using Python
	Deep RL for trading with the OpenAI Gym
		Value function approximation with neural networks
		The Deep Q-learning algorithm and extensions
		Introducing the OpenAI Gym
		How to implement DDQN using TensorFlow 2
		Creating a simple trading agent
		How to design a custom OpenAI trading environment
		Deep Q-learning on the stock market
		Lessons learned
	Summary
Chapter 23: Conclusions and Next Steps
	Key takeaways and lessons learned
		Data is the single most important ingredient
		Domain expertise: telling the signal from the noise
		ML is a toolkit for solving problems with data
		Beware of backtest overfitting
		How to gain insights from black-box models
	ML for trading in practice
		Data management technologies
		ML tools
		Online trading platforms
	Conclusion
Alpha Factor Library
	Common alpha factors implemented in TA-Lib
		A key building block – moving averages
		Overlap studies – price and volatility trends
		Momentum indicators
		Volume and liquidity indicators
		Volatility indicators
		Fundamental risk factors
	WorldQuant's quest for formulaic alphas
		Cross-sectional and time-series functions
		Formulaic alpha expressions
	Bivariate and multivariate factor evaluation
		Information coefficient and mutual information
		Feature importance and SHAP values
		Comparison – the top 25 features for each metric
		Financial performance – Alphalens
References
Index