Evidence on the Stability of Serial Dependencies within Taiwan Stock Returns

A maintained assumption of all the tests used in the previous sections of this chapter is that whatever stochastic process underlies the data is a stable one, i.e., that the form of any serial dependencies within a data series remains stable. In other words, there is one correct model that describes the stochastic process, and the parameters of this correct model remain constant throughout the time period from which the data are drawn. For example, if the underlying process is a simple linear, autoregressive process, such as an AR(1) process, then the autocorrelation function (ACF), along with the model parameters for the process whose values are functions of the ACF, do not vary over time but remain constant throughout the entire time series. Due to the effects of random variation in the observed data, the estimated ACF will be an imperfect reflection of this stable underlying ACF.