Machine Learning Algorithms for Trading Bootcamp

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Week 1: Overview of Algorithmic Trading Systems

Overview of investment/trading process and various trading models (60 minutes)

• Presentation: Financial/economic model concepts including the capital asset pricing model, arbitrage pricing model, and multifactor models; technical analysis and common trading strategies, including trending and mean reverting strategies
• Group discussion: Financial concepts and issues
• Q&A
• Break

Sourcing and analyzing market, fundamental, and alternative financial data (60 minutes)

• Hands-on exercise: Import data from freely available public sources such as FRED (economic), IEX (equity), Alpha Vantage (various), Quandl (various)
• Group discussion: Pros and cons of various data sources; issues of data extraction, cleaning, and storage
• Hands-on exercises: Use pandas dataframes and plotting functions to analyze and visualize data
• Q&A
• Break

Developing and backtesting algorithmic trading strategies (60 minutes)

• Hands-on exercises: Create a simple moving average crossover strategy; backtest the trading strategy
• Group discussion: Analysis of the implemented algorithmic strategy; analysis of the backtest results
• Q&A
• Break

Evaluating risk-adjusted performance (60 minutes)

• Presentation: Pitfalls of backtesting, especially overfitting and hindsight bias; overview of risk-adjusted performance metrics including Sharpe/information and Sortino ratios; the Kelly criterion for money management
• Group discussion: Paramount importance of risk management and position sizing in trading; next steps, including forward testing/paper trading; issues and risks with going live with algorithmic trading
• Q&A

Week 2: Prototyping Linear and Nonlinear ML Trading Models

Using linear models for trading (60 minutes)

• Presentation: Training ordinary least squares linear regression model to forecast stock price returns; using lasso and ridge regression to prevent overfitting to noisy financial data; training logistic regression model to predict an economic recession; using the lasso and ridge regularization to prevent overfitting of noisy financial data; understanding how the classifier assigns probabilities
• Hands-on exercises: Use scikit-learn module to train and test three types of linear regression models to predict stock price returns; use scikit-learn module to train and test the logistic classification models to predict an economic recession
• Q&A
• Break

Using DT classification and regression models to forecast market direction and returns (60 minutes)

• Presentation: Building, training, and visualizing decision tree (DT) classification models; using prepruning and postpruning to prevent overfitting of noise in financial data; understanding how the classification metrics are used
• Hands-on exercise: Use scikit-learn library to train and test DT classification models to predict market direction
• Q&A
• Break

Using RF classification and regression models to predict an economic recession and percentage loss in GDP (60 minutes)

• Presentation: Building, training, and visualizing random forest (RF) models; using bagging and feature engineering to prevent overfitting of noisy financial data; understanding how the classifier assigns probabilities and issues with imbalanced classes
• Hands-on exercise: Use scikit-learn library to train and test the RF classification and regression models to predict an economic recession and percentage loss in GDP
• Q&A
• Break

Evaluating and improving DT and RF models using GBM (60 minutes)

• Presentation: Building, training, and testing gradient boosting models (GBM); using out-of-bag errors to measure performance; different types of boosting algorithms and learning rates to improve performance and reduce overfitting of noisy financial data
• Hands-on exercises: Use scikit-learn to train and test GBM models and improve on the above DT and RT models
• Q&A

Week 3: Prototyping Probabilistic ML Models for Trading

Overview of probability and epistemic inference (60 minutes)

• Hands-on exercise: Run simulation of the Monty Hall problem
• Presentation: Types of uncertainty and meaning of probability; inverse probability rule, the fundamental algorithm of all probabilistic programming languages
• Q&A
• Break

Estimating the probability of default (60 minutes)

• Hands-on exercise: Compute the probability of credit default using the inverse probability rule in Python
• Presentation: Overview of the PyMC3 probabilistic programming library, including the basic concepts behind MCMC, NUTS, and ADVI algorithms
• Q&A
• Break

Developing probabilistic linear regression models for quantifying the uncertainty of expected stock price returns (60 minutes)

• Hands-on exercise: Compute alpha, beta, and sample error, including the 95% confidence intervals using a linear market model (MM) of a company’s stock
• Presentation: Sources of errors in financial models and the need to quantify uncertainty in point estimates; overview of issues with confidence intervals and need for using credible intervals instead
• Hands-on exercise: Recode the MM model in PyMC3 to produce credible intervals for all model parameters
• Q&A
• Break

Developing probabilistic classification models for quantifying the uncertainty of an economic recession (60 minutes)

• Presentation: Inverse probability in odds format; simple derivation of the logit function from the inverse probability rule; how it quantifies total uncertainty
• Hands-on exercise: Predict economic recession and quantify total uncertainty in the predicted values
• Q&A

Week 4: Prototyping Unsupervised ML Models to Discover Trading Opportunities

Unsupervised machine learning in algorithmic trading and investing (60 minutes)

• Presentation: Conceptual overview of unsupervised data exploration and transformation machine learning algorithms; importance of feature engineering in finance and how these algorithms help; differences in theoretical and data-driven model building in trading and investing; how algorithms improve on theoretical asset pricing models, portfolio management techniques, and investment style classifications
• Hands-on exercises: Create pandas dataframes to concatenate data; preprocess and standardize data for unsupervised learning algorithms
• Q&A
• Break

Using clustering algorithms to explore equity data for pairs trading (60 minutes)

• Presentation: Overview of pairs trading and clustering algorithms including k-means, hierarchical, and affinity propagation; enabling clustering algorithms to explore and select equities for pairs trading; visualizing clustering hierarchies using dendrograms
• Hands-on exercise: Use scikit-learn’s clustering algorithms to select equity pairs for trading
• Q&A
• Break

Using principal component analysis algorithm to build data-driven asset pricing models (60 minutes)

• Presentation: Overview of various theoretical asset pricing models; how the principal component analysis (PCA) algorithm builds data-driven factor models for asset pricing without requiring any preconceived economic theory
• Hands-on exercises: Use scikit-learn’s PCA algorithm to build factor models for equity prices; visualize factors using heatmaps
• Q&A
• Break

Using PCA algorithm to increase diversification in investment portfolios (60 minutes)

• Presentation: Overview of Markowitz’s mean variance portfolio management theory; how the PCA algorithm improves on it by reducing correlations among asset price returns
• Hands-on exercise: Use scikit-learn’s PCA algorithm to design investment portfolios with reduced correlations among assets; derive portfolio weights with principal components
• Q&A

Week 5: Prototyping Deep Learning Neural Networks for Trading

Deep learning models in algorithmic trading and investing (60 minutes)

• Presentation: Conceptual overview of artificial neural network (ANN) architectures and DL models; how these algorithms help with feature engineering; differences in theoretical and data-driven model building in trading and investing; how DL algorithms improve on regression and classifications problems
• Hands-on exercises: Set up Colab notebook; create pandas dataframes to concatenate data from freely available public sources such as FRED (economic), Yahoo (equity), and Quandl (various); preprocess and standardize data for DL algorithms
• Q&A
• Break

Using multilayer perceptrons to forecast economic recessions (60 minutes)

• Presentation: Overview of multilayer perceptron networks and design choices available in Keras for building any type of ANN
• Hands-on exercise: Use Keras to develop multilayer perceptron networks to classify economic recessions
• Q&A
• Break

Using convolutional neural networks to price equities (60 minutes)

• Presentation: Overview of capital asset pricing model and convolutional neural networks (ConvNet)
• Hands-on exercise: Use Keras to build ConvNets to price Apple stock
• Q&A
• Break

Using recurrent neural network to price equities (60 minutes)

• Presentation: Overview of arbitrage pricing theory and recurrent neural networks (RNN)
• Hands-on exercise: Use Keras to build RNNs to price Apple stock
• Q&A

Week 6: Prototyping Reinforcement Learning Trading Agents

Understanding the reinforcement learning modeling framework (60 mins)

• Presentation: Reinforcement learning (RL) components, modeling framework, equations, and algorithms
• Hands-on exercise: Apply RL components to a trading agent
• Q&A
• Break

Applying the Markov decision process to trading (60 minutes)

• Presentation: Overview of the Markov decision process and its properties; applying the Bellman equations to compute the optimal trading strategy
• Hands-on exercise: Evaluate the optimal trading strategy in a simulated market environment using dynamic programming
• Q&A
• Break

Building a deep Q-network for approximating market environment (60 minutes)

• Presentation: Overview of model-based and model-free RL algorithms, temporal difference learning, and Q-learning; using DL to approximate Q-values
• Hands-on exercise: Prototype a deep Q-network to approximate Q-values using S&P 500 prices
• Q&A
• Break

Prototype RL-based trading strategy (60 minutes)

• Hands-on exercises: Train and test RL-based equity-based trading agent on the deep Q-network
• Presentation: Review and summary of RL, including its advantages and disadvantages
• Q&A