F71DA - Data Analytics and Time Series Analysis

Course leader(s):

• Fraser A. Daly (Edinburgh, January)

Syllabus

1. Time Series Analysis (1.1 1. Autocorrelation, stationarity and operators, 1.2 2. MA, AR, ARMA and ARIMA processes, 1.3 3. Forecasting, 1.4 4. Estimation and model selection, 1.5 5. Vector time series, 1.6 6. Introduction to ARCH and GARCH models)

2. Machine Learning (2.1 1. Classification, 2.2 2. Unsupervised learning, 2.3 3. Logistic regression and neural networks)

3. Loss Distributions (3.1 1. Properties of loss distributions, 3.2 2. Fitting loss distributions, 3.3 3. Reinsurance)

4. Extreme Value Theory (4.1 1. Block maxima, 4.2 2. Threshold exceedances)

5. Credibility (5.1 1. Bayesian credibility models, 5.2 2. Empirical Bayes credibility theory)

6. Copulas (6.1 1. Properties of a copula, 6.2 2. Sklar's theorem, 6.3 3. Examples of copulas, 6.4 4. Estimation of copulas)

Learning outcomes

By the end of the course, students should be able to do the following:

• establish important properties of models for time series data (including linear and non-linear models) and forecasts from these models.
• select appropriate models for time series data using a range of statistical tools, including autocorrelations, information criteria and model diagnostics.
• compare distributions for individual and aggregate losses in a variety of settings, and how these apply to modelling data.
• analyse extreme value distributions for the modelling and comparison of tails of data.
• evaluate a range of families of copulas as models for multivariate data.
• apply Bayesian techniques in credibility theory and related estimation problems.
• compare several supervised and unsupervised machine learning algorithms using appropriate diagnostics, in the context of model complexity and overfitting.
• apply appropriate software tools in the computational aspects of data analysis and time series modelling.

Further details

Curriculum explorer: Click here

SCQF Level: 11

Credits: 15