Thesis (M.Com. (Finance))--University of the Witwatersrand, Faculty of Commerce, Law and Management, School of Economic and Business Sciences, 2016 / The purpose of the study is to investigate the effects of relaxing the assumption of multivariate normality typically utilised within the traditional asset pricing framework. This is achieved in two ways. The first involves the introduction of higher moments into the linear Capital Asset Pricing Model while the second involves a Monte Carlo experiment to determine the impact of skewness and kurtosis on test statistics traditionally employed to assess the validity of asset pricing models. We commence by establishing non-normality for the majority of sample portfolios. A cross-sectional regression approach is employed to estimate factor risk premia and test higher moment Capital Asset Pricing Models. Unconditional coskewness and unconditional cokurtosis are found to be priced within the market equity (size) sorted and book equity/market equity (value) sorted portfolio sets over the period January 1993 to December 2013. Conditional coskewness and conditional cokurtosis are found to be priced for only the size sorted portfolios over the period January 1997 to December 2013. Factor risk premia estimated for coskewness are generally positive while risk premia estimated for cokurtosis are negative. This suggests a positive relationship between coskewness and expected return and a negative relationship between cokurtosis and expected return. The results of the asset pricing model tests are mixed. The pricing errors for higher moment Capital Asset Pricing Models are shown to be significantly different from zero for size sorted portfolios while pricing errors on the value sorted, dual size-value sorted and industry portfolios are found to be statistically insignificant. This suggest that none of the asset pricing models tested are the true model as it would explain variation in expected returns regardless of the data generating process. Finally we show that the Ordinary Least Square Wald test statistic has the most desirable size characteristics while the Generalised Least Squares J-test statistic has the most desirable power characteristics when dealing with non-normal data.
Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:wits/oai:wiredspace.wits.ac.za:10539/20495 |
Date | January 2016 |
Creators | Bester, Johan |
Source Sets | South African National ETD Portal |
Language | English |
Detected Language | English |
Type | Thesis |
Format | application/pdf |
Page generated in 0.0014 seconds