12.07.2015 Views

"Frontmatter". In: Analysis of Financial Time Series

"Frontmatter". In: Analysis of Financial Time Series

"Frontmatter". In: Analysis of Financial Time Series

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

360 MULTIVARIATE VOLATILITY MODELSBecause Σ t is positive definite, there exist a lower triangular matrix L t with unitdiagonal elements and a diagonal matrix G t with positive diagonal elements suchthatΣ t = L t G t L ′ t . (9.7)This is the well-known Cholesky decomposition <strong>of</strong> Σ t . A nice feature <strong>of</strong> the decompositionis that the lower <strong>of</strong>f-diagonal elements <strong>of</strong> L t and the diagonal elements <strong>of</strong> G thave nice interpretations. We demonstrate the decomposition by studying carefullythe bivariate and three-dimensional cases. For the bivariate case, we haveΣ t =[ ]σ11,t σ 21,t, Lσ 21,t σ t =22,twhere g ii,t > 0fori = 1 and 2. Using Eq. (9.7), we haveΣ t =[ ] [ ]1 0g11,t 0, Gq 21,t 1 t =,0 g 22,t[ ] [ ]σ11,t σ 12,t g11,t q 21,t g 11,t=σ 12,t σ 22,t q 21,t g 11,t g 22,t + q21,t 2 g .11,tEquating elements <strong>of</strong> the previous matrix equation, we obtainσ 11,t = g 11,t , σ 21,t = q 21,t g 11,t , σ 22,t = g 22,t + q 2 21,t g 11,t. (9.8)Solving the prior equations, we haveg 11,t = σ 11,t , q 21,t = σ 21,tσ 11,t, g 22,t = σ 22,t − σ 2 21,tσ 11,t. (9.9)However, consider the simple conditional linear regressiona 2t = βa 1t + b 2t , (9.10)where b 2t denotes the error term. From the well-known least squares theory, we haveβ = Cov(a 1t, a 2t )Var(a 1t )= σ 21,tσ 11,t,Var(b 2t ) = Var(a 2t ) − β 2 Var(a 1t ) = σ 22,t − σ 2 21,tσ 11,t.Furthermore, the error term b 2t is uncorrelated with the regressor a 1t . Consequently,using Eq. (9.9), we obtaing 11,t = σ 11,t , q 21,t = β, g 22,t = Var(b 2t ), b 2t ⊥a 1t ,

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!