CS815 – AI for Finance

General Aims

This class aims to provide an overview of the application of evolutionary computation techniques – those which mimic natural evolutionary processes (genetic algorithms, genetic programming and neural networks in particular) – to a range of financial applications such as forecasting, portfolio optimisation and algorithmic trading.

The course is very practical in its nature: much of the learning is achieved via a number (around 4) of assessed small mini-projects, and students are expected to develop solutions to problems using evolutionary computation techniques, evaluate these on real data sets, and compare them with other more traditional approaches. Consequently , a large amount of self-directed study and learning is expected.

Learning Outcomes

After completing this class participants will be able to:

• Understand the benefits and opportunities for evolutionary computing in the context of financial applications.
• Understand the principles of evolutionary computation: genetic programming and genetic algorithms in particular, and also neural networks (particularly those configurations most suited to time series data).
• Understand how the computational approaches covered in the class may be applied to financial problem solving and understand their limitations.
• Develop and evaluate practical solutions to finance-based problems.

Syllabus

Part 1

• Programming using R (Students will have some familiarity with R, typically from a statistic point of view, but this initial section looks at its use as a programming language)
• Principles of genetic algorithms and genetic programming
• Portfolio Optimisation – This topic looks at how you can use evolutionary algorithms, and GAs in particular, to develop an optimal portfolio – a balanced set of investments that will yield the best return for the least risk
• Principles of basic neural networks
• Forecasting and Predication – This topic looks at how it is possible to use genetic programming and/or neural networks to generate predictive functions for time-series data.

Part 2

• Advanced Neural Network configurations (e.g. RNNs and LSTMs)
• Algorithmic Trading – This topic looks at the application of both Neural Networks potentially combined with GAs or GP to identify automatically at what point a trade should be executed, and evaluate the effectiveness of the strategy through backtesting.

Recommended Text/Reading

The class lecturer will recommend a mixture of specific and generic references for each part of the course. Where possible, free online material will be recommended.