Tid: 21 januari 2008 kl 15.15-16.00
Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!
Föredragshållare: Inga Nkomo
Titel: Tests for changes in correlation in financial data. (Examensarbete)
Sammanfattning: Measuring correlation between log returns for different stock prices is of importance to financial analysts. This thesis focuses on methods for investigating the properties of bivariate log return distributions, in particular methods for investigating whether the correlation is changing with time and return sizes. Splitting the sample into a part consisting of returns with large absolute values and a part with small absolute values can lead to what appears to be a correlation breakdown in volatile market periods. However, as in shown in this thesis, this may simply be the consequence of computing conditional correlations, even on a sample of independent and identically distributed bivariate return vectors.
Extreme value theory, univariate and bivariate, is used to investigate marginal- and dependence properties of return data. To test whether correlation is changing with return sizes the corresponding conditional correlations for bivariate normal and Student t distributions are computed and used in hypothesis testing with return data.
The conclusion is that the hypothesis that the bivariate log return vectors have a bivariate Student t distribution with four degrees of freedom and a fixed correlation parameter cannot be rejected at any reasonable significance level.
|Sidansvarig: Gunnar Karlsson