Nonlinearities and Nonstationarities in Stock Returns
探讨股票收益的非线性发现是否受分布变化影响，提出区分非线性和结构变化的检验方法，发现方差变化能解释数据，而ARCH模型表现不佳。
This paper addresses the question of whether recent findings of nonlinearities qhave been contaminated by possible shifts in the distribution of the first differences of the logarithms of stock prices indexes The paper develops a testing methodology that formally attempts to discriminate between the two types of rejections of the null of linearity It is shown that structural shifts play an important role in the evolution of financial time series: linear processes with shifts in variance are able to replicate the behavior of the tests introduced in the paper whereas stationary ARCH-type filters show little consistency with the data Moreover it is shown that ARCH models fitted to data generated by a simple one-break linear process exhibit levels of persistence similar to the ones usually reported for high-frequency applications Key words: BDS test Nonlinearity Nonstationarity