AI For Asset Managers

Day One Morning

Introduction

• Alternative data sources – from individual activity (e.g. social media), business processes (e.g. credit card transactions) and sensors (e.g. satellite imaging, smartphone GPS-based geolocation, internet of things)

• AI vs machine learning (ML)

ML Methods In AM – Supervised Learning

• Regression. Advantages of ML regression vs ordinary least squares. Penalised regression techniques - Lasso, Ridge and Elastic Net. Non-parametric regression – K-nearest neighbours and LOESS regression

• Case Study 13 – Using penalised regression techniques to forecast returns for multi-asset trend-following strategy across S&P 500, USTs and gold

• Case Study 14 – Using non-parametric regression to generate optimal risk-premia weighted asset allocations in multi-asset portfolio

• Case Study 15 – Using XGBoost to forecast returns of long-short equity ETF sector strategy based on multiple macro risk factors. Comparison to S&P 500

• Classification. Generation of buy and sell signals. Classification of volatility regimes. Logistic regression, Support Vector Machines (SVM) decision trees, Random Forest and Hidden Markov Models (HHM)

• Case Study 16 – Using logistic regression to improve equity call over-writing strategy

• Case Study 17 – Using SVM to generate implied vol buy and sell signals in FX options

• Case Study 18 – Using Random Forest in stock selection

• Case Study 19 – Using HHM for marketing timing decisions on S&P 500

Day One Afternoon

ML Methods In AM – Unsupervised Learning

• Clustering for regime identification to improve asset allocation (e.g. high or low volatility; rising or falling interest rates). Clustering algorithms - K-Means, Ward, Birch, Affinity, Spectral, MiniBatch, Aggregate, HDSBscan and ICA

• Case Study 20 – Using clustering algorithms to identify inflation regime to support asset allocation

• Factor analysis

• Case Study 21 – Using Principal Component Analysis to explain variation in EURUSD implied volatility surface

ML Methods In AM – Deep Learning

• Deep learning vs classical ML

• Neural networks

• Case Study 22 – Multi-Layer Perceptron Neural Network to construct equity sector long-short strategy and predict its performance

• Long short-term memory (LSTM)

• Case Study 23 – Using LSTM to construct S&P 500 timing strategy

• Case Study 24 – Using Restricted Boltzmann Machine to construct FX trading strategy