Georgi Boshnakov - School of Mathematics
Table of Contents
Dr. Georgi Boshnakov
Lecturer in Statistics
Tel.:+44 (0)161 306 3684
Fax:+44 (0)161 275 5819
Alan Turing Building/1.125
School of Mathematics
University of Manchester
Oxford Road, M13 9PL Manchester, UK
Time series analysis and forecasting, Data analysis, Statistical computing, Matrices in statistics, Symbolic computation.
Packages and other resources created by me (some with collaborators).
(this is for my R packages, for general R resources see section R - information and resources).
Contributions to TeX/LaTeX.
Support for typesetting and Bulgarian and other information about Emacs.
You may be asked for identification to get access to some of the pages below.
A collection of resources about R. The selection is somewhat taylored to the courses I teach.
I offer projects on time series topics and, in particular, financial time series. The projects require background in time series and in most cases this means that students should do (or have done) a time series or similar course.
MMath students wishing to do such projects in their 4th year will improve their prospects by planning ahead and taking the ATS course in their 3rd year.
1st and 2nd year students reading this section and planning their options should notice that the Statistical methods course in 2nd Sem/2nd year is a prerequisite for (essentially) all subsequent statistics courses. Practical statistics I is a big plus too.
I mainly offer projects which fall under the umbrella of "time series analysis". For other topics in statistics and probability (usually motivated by my research on time series) see my recent publications and research reports.
Here is an example project.
Time series data, such as economic indicators and stock prices, are now routinely collected in vast quantities. This has opened a number of new problems of interest to financial institutions, governments and academy. One such problem is clustering of time series. For example, splitting industries or stock market companies into groups (clusters) of entities, similarly performing over a period of time, according to some indicator, may be helpful for decision and policy makers.
Time series clustering is an area of active research. We will be interested mainly in developing methodology and algorithms for dimensionality reduction, e.g. extracting a vector of features from a time series which may then be fed to appropriate clustering or classification algorithms. Such features may include, among others, trends, seasonality, autocovariance/spectral structure, long memory indices, and even whole models, such as ARIMA. Also, there is no universal set of features for this clustering problem.
Note: Do not tell me that you are fond of time series if you have not taken such courses during your previous studies. Good mathematical and/or computational background is top priority.