Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/40185 
Year of Publication: 
2009
Series/Report no.: 
CPQF Working Paper Series No. 19
Publisher: 
Frankfurt School of Finance & Management, Centre for Practical Quantitative Finance (CPQF), Frankfurt a. M.
Abstract: 
Principal Component Analysis (PCA) is a common procedure for the analysis of financial market data, such as implied volatility smiles or interest rate curves. Recently, Pelsser and Lord [11] raised the question whether PCA results may not be 'facts but artefacts'. We extend this line of research by considering an alternative matrix structure which is consistent with foreign exchange option markets. For this matrix structure, PCA effects which are interpreted as shift, skew and curvature can be generated from unstructured random processes. Furthermore, we find that even if a structured system exists, PCA may not be able to distinguish between these three effects. The contribution of the factors explaining the variance in the original system are incorrect. Finally, for a special case, we provide an analytic correction that recovers correct factor variances from those incorrectly estimated by PCA.
Subjects: 
Principal Component Analysis
PCA
Level
Slope
Curvature
Twist
Bisymmetric Matrices
Centro-symmetric Matrices
Document Type: 
Working Paper

Files in This Item:
File
Size
594.99 kB





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.