Bayesian Data Analysis provides a comprehensive dive to the methods offered by the Bayesian statistical paradigm to analyse economic and financial data. The course opens with a deep look at Bayes' rule; the simple rule for optimally updating statistical information that underpins all of Bayesian statistical methods. The course then studies how simple applications of Bayes' rule can provide new ways to understand probabilities, specifying statistical models, and learning from data by estimating models through the Bayesian lens. The course also covers a wide variety of novel computational techniques used for inferring patterns in large datasets and their connections to generic machine learning methods that can be used for analysing data across economics and all sciences.

Synchronous:

10 x 2-hour lectures on campus/online as appropriate for the course content

10 x 2-hour labs on campus/online as appropriate for the course content

None

None

1. Project output; Group; 1800 words maximum; 25%; ILOs 3-5.

2. Project output; Group; 1800 words maximum; 25%; ILOs 1, 3-4.

3. Degree exam: in-person; Individual; 90 minutes; 50%; ILOs 1-2, 4-5.

It is the default expectation that all courses will offer opportunities for reassessment or deferred assessment. Where it is not possible to offer this in some assessment components, the grade achieved at the first attempt will be counted towards the final course grade, and any exceptions for this course are described below.

[No exceptions]

The main aim of this course is to cover the foundation of the Bayesian approach to probability, a statistical framework that will equip the students with new data analytic skills. While the motivation of the techniques and approaches is generally through the lens of economic and financial data, the skills provided by this course are generable and transferrable such that they are useful in several other disciplines related to machine learning and computer science. The lectures teach the basic theory behind the Bayesian approach to statistical inference and the extensions to account for complex data situations. The computer labs focus on teaching and practicing computational techniques that are used in Bayesian inference, and that are likely not immediately related to computational techniques students have acquired in previous econometrics classes. At the same time the computer labs will allow students to apply theory and computation to numerous datasets across economics, finance, and other fields, with a focus on computation for high-dimensional datasets.

By the end of this course students will be able to:

1. Critically distinguish the fundamental differences between the Bayesian approach to probability and traditional frequentist approaches.

2. Construct and specify flexible Bayesian models by means of likelihood and prior functions adapted to specific modelling situations.

3. Program advanced Markov chain Monte Carlo algorithms for inference to estimate Bayesian models.

4. Demonstrate ability in developing Bayesian machine learning algorithms for inference in problems of high and ultra-high dimensions.

5. Adapt Bayesian inference principles to empirical problems in finance and economics.