Nonlinear Serial Dependence in Share Returns on the Johannesburg Stock Exchange

It has been suggested that emerging markets may exhibit nonlinear dependencies due to their propensity for thin trading, high transaction costs and regulatory constraints. Prior research has found evidence of linear serial correlation in South African share returns but tests of nonlinear serial dependence have been limited in this market. This study examines nonlinear serial dependence on a sample of 109 shares from the Johannesburg Stock Exchange (JSE) using a battery of tests and finds evidence of significant nonlinear serial dependence for all shares examined. A windowed test procedure finds, however, that these occurrences are episodic in nature rendering return prediction on the basis of nonlinear serial dependence a potentially complex process.