Reliability of VaR models in extreme circumstances

Tvegana vrednost (VaR) je pogosto uporabljena mera tržnega tveganja. Izračunamo ga lahko po različnih metodah, ki jih delimo v tri skupine: parametrične linearne metode, metode zgodovinske simulacije in Monte Carlo metode. Na njih temelječi VaR modeli so bili na preizkusu v izjemnih okoliščinah, kakršne je predstavljala svetovna finančna kriza. V diplomskem delu so predstavljene osnovne VaR metode. Ustrezni VaR modeli so testirani na podatkih cen delnic portfeljev, sestavljenih iz nekaterih delnic indeksov SBI 20, DAX 30 in XMI. Opisan je način testiranja teh modelov. Rezultati kažejo, da so za dnevne VaR napovedi najprimernejši modeli zgodovinske simulacije, za 10-dnevne napovedi Gumbelov linearni VaR model, za mesečne in kvartalne napovedi pa so se vsi testirani modeli izkazali za nezanesljive.Value-at-Risk (VaR) is a widely used measure of market risk. It can be calculated by different methods, which are divided into three groups: parametric linear methods, historical simulation methods and Monte Carlo methods. VaR models, based on these methods, were on a trial in extreme circumstances such as represented by the global financial crisis. This thesis presents the basic VaR methods. The corresponding VaR models are tested on the data of share prices of portfolios, consisting of certain shares from SBI 20, DAX 30 and XMI indices. A method of testing these models is described. The results show that for daily VaR forecasts historical simulation models exhibit best performance, for 10-day forecasts Gumbel\u27s linear VaR model performs best, while for monthly and quarterly forecasts, all the tested models were proven to be unreliable