SE-507 Machine Learning with Applications in Finance (Autumn 2021)

Upon successful completion of this course the student will

• understand the various machine learning (ML) paradigms.
• be able to analyse the merits and limits of the different ML approaches.
• understand the simplifying assumptions under which advanced machine learning techniques are equivalent to well-known statistical models.
• be able to apply machine learning techniques to financial data using R.
• test multiple hypotheses, estimate non-linear times series models, and generate forecasts using neural nets using R.
• communicate the results of the analysis to specialists as well as to non-specialist audiences.

This elective course links a conventional finance program to the machine learning arena.

After introducing dominant paradigms of machine learning, we discuss various neural network models suitable for financial time series analysis. In particular, recurrent neural networks (RNNs) are presented as non-linear time series models generalizing classical linear models of the ARMA type. Moreover, the potential of reinforcement learning for solving problems of dynamic asset management and option pricing will be discussed and demonstrated.  The remainder of the course covers an aspect of statistical learning in finance:  the avoidance of data mining issues in multiple hypothesis testing. It is demonstrated how the bootstrap technology can be applied to provide a statistically valid performance assessment of different investment strategies (trend following, cross-sectional momentum, and volatility-responsive strategies).

The course consists of lectures and seminars (group sessions). Expected total workload: 200 hours.

We use a dual approach to familiarize the student with machine learning techniques suitable for applied work and research work in quantitative finance. Theoretical arguments presented in the lectures are augmented by seminar sessions during which students apply learning approaches (algorithms, machine learning procedures) to solve practical problems of quantitative finance. The projects are designed to help students to understand the potential as well as the limitations of a specific learning paradigm and to train specific data analytical skills using R.