The Arbitrage pricing theory: an assessment of the robustness of empirical techniques employed under conditions of thin trading and in the presence of non-normalities

Page, Michael J

The Arbitrage pricing theory: an assessment of the robustness of empirical techniques employed under conditions of thin trading and in the presence of non-normalities

Thesis / Dissertation

1993

Abstract

Ever since its original development by Ross in 1976, and in spite of the objections of Shanken (1982), proponents of the Arbitrage Pricing Theory have claimed that the theory offers a testable alternative to the Capital Asset Pricing Model. Over the last decade however, empirical research into the APT has not proved to be universally successful, particularly when consideration is given to the fact that the majority of tests of the validity of the theory have used significantly overlapping NYSE datasets. The number of priced factors discovered commonly ranges from one (Tryzinka, 1986) to five or six (Cho, 1984). Research into the identification of the APT factors has also proved to be difficult. While the economic identification of the factors is not a necessary condition for tests of the APT itself, the importance of this aspect as far as future practical application of the theory is concerned cannot be overrated. This research seeks to add to the literature on the testability of the APT by showing that, theoretical objections aside, the current empirical procedures employed in the determination and pricing of the factors lack power. On the basis of an extensive simulation study the statistical procedures of principal components analysis, principal factor analysis and generalised least squares regression are shown to produce results that appear independent of the true number of underlying factors generating risky asset returns. When market microstructure effects and significant levels of thin trading are simulated into the data the power of the procedures is further reduced. Interestingly, in contrast to the suggestions of some prior researchers (Pari and Chen, 1984), when nonnormalities of the order observed on the JSE are simulated into the idiocyncratic risk component of returns they have no noticable effect on the results. This finding also bears testimony to the low power of the procedures. A comparison of the use of APT and CAPM methodologies is presented in the final section of the research and, in spite of the power problems, the APT is found to be preferable. In an analysis of both unit trust performance and price/earnings and size anomalies the APT methodology is found to provide results that are either as good as, or improve on, those obtained using a traditional CAPM approach. This finding applies in spite of the uncertainty as to the appropriate number of factor mimicking portfolios to use.

Reference:

Collections

PhD / Doctoral

Full item page