Sara Berger  and Johan Claesson defend their master thesis 

Conditional Methods for Predicting Scenarios for Market Risk Factors

Abstract

The objective of this thesis is to predict different scenarios for
market risk factors and based on these scenarios calculate and compare
different risk measures for day-to-day risk control. In addition we
analyze the risk factor reduction techniques Observable Factor models
and latent Principal Component Analysis, and determine if they can be
used to reduce the number of significan t parameters in a model and in
that way simplify the calculations without any considerable influence
on the result. We use the risk measures Value at Risk and Expected
Shortfall, which can be calculated with several approaches. The
approaches used in this thesis are the standard methods
Variance-Covariance and Historical Simulation and also the more
complex method Conditional Extreme Value theory . To predict
conditional values of the risk measures, time series prediction models
such as GARCH, GARCH-t and EWMA are implemented. Based on proper
backtests it was concluded that to achieve the most accurate
predictions and calculations of risk measures the conditional methods
based on the standard GARCH model should be used despite the fact that
they are more time-consuming than the unconditional methods. The risk
factor reduction techniques should be used only if absolutely
necessary.