Find some interesting financial time series, preferably one of your own, and estimate VaR for it, for three different time periods, and with at least thee methods for each period (by assuming normality, by the block maxima method and by the Peaks over Thresholds method). One period should contain the present year, and the second should include some earlier financial crisis, and the third period should be crisis-free . Comment on the following: a) differences and similarities of the results obtained with the different methods, b) doesVaR from the normal period predict the risk in the crisis period in a reasonable way, and c) differences between VaR in the three different crisis periods.
An uptodate review of software for extreme valuel statisitics is provided by the Extremes paper "A software review for extreme value analysis" by Eric Gilleland, Mathieu Ribatet, andAlec G. Stephenson. It can be viewed here. There are many R-courses available on the internet.
Many financial time series can be found at the federal reserve releases page and at YAHOO! Finance.
Install R and the evd package in windows:
Go to http://www.r-project.org/, click "download R" and choose your preferred mirror (e.g., http://ftp.sunet.se/pub/lang/CRAN/).
Click "Windows", then "base" and "Download R --version number-- for Windows". After downloading, run the .exe-file to install R.
To get the evd package (or the extRemes package), which contain functions for extreme value statistics, go to the mirror site (e.g., http://ftp.sunet.se/pub/lang/CRAN/). Click "Packages" in the menu to the left, scroll down to find "evd" or "extRemes" in the list of available packages. Click "evd" or "extRemes" and download the Windows binary. It may be a good idea to also download the reference manuals.
To be able to use the evd and extRemes packages, you must install and load it in R. Open R, then click "Install package(s) from local zip files" in the "Packages menu" and select your downloaded zip file. To load the package, type "library(evd)" in the R console (or click "Load package" in the "Packages" menu and select "evd")
Comments from Anders Sjögren on better
user interfaces for R: I
have only used RStudio briefly, but it does indeed seem to be very nifty and
makes R programming easier, e.g. by showing the variables in the workspace. It
definitely seems like a big improvement compared to vanilla R on the
command-line.
There
is also a commercial alternative, called Revolution R ( http://www.revolutionanalytics.com/
), which is free for university use. It is stated to only work on Windows and
RedHat Linux (I don't know about our Linux boxes). It has much better support
for running statistical analyses on datasets larger than RAM.
If you want to, your group can midway into the project book time with the teachers for a brief discussion of the project plan. A short written outline of the plan should be handed in at least two days before such a meeting.
The project results should presented in a short report, which should be intended for the eyes of an imaginary risk manager who knows little about VaR and less about statistics. In particular it should contain a popular "Executive summary" and a few carefully selected graphs with well thought out and selfcontained headings. For Project I, a few model graphs for model checking should be included in an appendix.
The two project reports should be sent to me separately and electronically. The file names should include the group number, the names of the group members, and the title of the project, abbreviated as much as possible. The front page of the report should include the full title of the project, full names of the group members, and a brief summary of who has done what. Pdf files are much preferred to other formats. A signed paper copy of the front page should be given to me.