Tid: 8 september, 2017, kl 14.30-15.00.

Föredragshållare: Dennis Sundström

Titel: Automatized GARCH parameter estimation(Master's thesis)

Abstract This paper is about automatizing parameter estimation of GARCH type conditional volatil- ity models for the sake of using it in an automated risk monitoring system. Many challenges arise with this task such as guaranteeing convergence, being able to yield reasonable results regardless of the quality of the data, accuracy versus speed of the algorithm to name a few. These problems are investigated and a robust framework for an algorithm is proposed, containing dimension reducing and constraint relaxing parameter space transformations with robust initial values. The algorithm is implemented in java with two models, namely the GARCH and gjr-GARCH model. By using real market data, performance of the algorithm are tested with various in-sample and out-of-sample measures, including backtesting of the widely used risk measure Value-at-Risk. The empirical studies conclude that the more complex gjr-sGARCH model with the conditional student?s t distribution was found to yield the most accurate results. However for the purpose of this paper the GARCH or gjr-GARCH seems more appropriate.

The full report (pdf)

Till seminarielistan
To the list of seminars