Embedding R in Windows applications, and executing R remotely
Embedding R in Windows applications, and executing R remotely
Embedding R in Windows applications, and executing R remotely
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Fitt<strong>in</strong>g the<br />
Generalized Pareto Distribution<br />
to Threshold Exceedances of<br />
Stock Index Returns<br />
In Bull <strong>and</strong> Bear Periods<br />
Angi Rösch ∗<br />
FOM University of Applied Sciences<br />
Munich, Germany<br />
Abstract<br />
The <strong>in</strong>vestigation of phenomena of jo<strong>in</strong>t extreme returns on f<strong>in</strong>ancial assets<br />
is crucial to f<strong>in</strong>ancial risk assessment. We focussed on jo<strong>in</strong>t threshold<br />
exceedances of daily returns on stock <strong>in</strong>dices, threshold mean<strong>in</strong>g the lower,<br />
respectively upper 10% quantile of the return distribution. Exceedances <strong>in</strong><br />
bull <strong>and</strong> bear periods were treated separately.<br />
In a first step, we fitted the generalized Pareto distribution to the exceedances<br />
of several stock <strong>in</strong>dices, apply<strong>in</strong>g the elemental percentile method<br />
of Castillo <strong>and</strong> Hadi. This provided the basis for the second step: Pairs of<br />
marg<strong>in</strong>al distributions of threshold exceedances were coupled, assum<strong>in</strong>g a logistic<br />
dependence structure for returns, <strong>in</strong> which the degree of dependency is<br />
reflected by a s<strong>in</strong>gle correlation parameter. This led to bivariate distributions<br />
of threshold exceedances of returns.<br />
Compar<strong>in</strong>g this method to one which ignores the dependency among return<br />
exceedances, we found that f<strong>in</strong>ancial risk would then be seriously underestimated.<br />
The degree of dependency tended to be stronger dur<strong>in</strong>g bear<br />
periods.<br />
All computations were carried out <strong>in</strong> the statistical comput<strong>in</strong>g environment<br />
R.<br />
∗ e-mail: angi.r@t-onl<strong>in</strong>e.de<br />
1