FIE458 Deep Learning with Applications to Finance

• Topics

  Deep learning is the latest frontier in using computers to analyze data.  Most of the most glamorous advances in AI, such as voice recognition and automatic translation, are made possible by deep learning.

  In this class, we will explore whether deep learning can be used to predict financial markets.  Deep learning requires immense amounts of data, so we choose a market where such data is readily available: cryptocurrency markets, such as Bitcoin.

  Cryptocurrency markets are a very promising area of application, because data is readily available and because the market is dominated by small retail investors who are more likely to engage in predictable patterns.

   

  Models

  • Multilayer perceptrons
  • Convolutional neural networks
  • LSTM networks
  • Autoencoders

   

  Data

  • Cryptocurrency data.
  • Other data as time permits, such as Twitter data.

• Learning outcome

  Knowledge

  • Students will understand the principles of deep learning.
  • Students will understand the dangers of overfitting and how to avoid them.
  • Students will understand the structure of the cryptocurrency market and the investment possibilities.

  Skills

  • Students will be able to develop a deep learning model, and the trade-offs between different models.
  • Students will be able to use financial data to fit and evaluate a deep learning model.

  General competencies

  • Students will be able to analyze financial data using a deep learning library.
  • Students will be to able to communicate their findings to investment professionals.
  • Students will be able to form an opinion about issues in both deep learning and the cryptocurrency markets.

• Teaching

  Teaching will be carried out using: 

  • Lectures
  • Sessions where students will practice techniques on their laptop (programming in R)
  • Class student group presentations

  The class will conclude with a group final project using real data. The purpose of the class presentations will be for students to present preliminary work leading up to the final project. Examples of topics for such presentations include: describe the data, describe steps taken to clean the data, preliminary efforts to analyze the data using techniques taught in lectures.

• Recommended prerequisites

  Prior experience with R or Python recommended.

  Prior experience with linear regression is helpful.

• Compulsory Activity

  None

• Assessment

  The class is graded on a portfolio of group class presentations, a group final paper and peer review. One grade is given for the entire portfolio. The group final paper is the most important component.  Students will also submit a reflection note that summarizes their learning and development. All material must be in English.

  For class presentations, students will be expected to present multiple times through the semester, either singly or in groups. Group size must be approved by the lecturer. The final peer review is for all group work. At the end of the term, students will evaluate the contribution of his/her team members. The peer review requires you to behave in a responsible and respectful manner. If I deem a student to deviate from such behavior, I may overrule the peer review.

  The grade awarded may not be appealed due to the nature of the assessments.

• Grading Scale

  A-F.

• Computer tools

  A laptop with R or Python, and the Keras deep learning library.

• Literature

  Ian Goodfellow, Yoshua Bengio and Aaron Courville (2016).  Deep Learning.  MIT Press

  The book is available electronically.