[经济学]Chapter2Univariatelinearstochasticmodels-basicconcepts.pptVIP

Chapter 2 Univariate linear stochastic models: basic concepts 2.1 Stochastic processes, ergodicity and stationarity 2.1.1. Stochastic processes, realisations and ergodicity 2.1.2 Stationarity One important simplifying assumption is that of stati-onarity, which requires the process to be in a particu-lar state of ‘statistical equilibrium’(Box and Jenkins, 1976, p. 26). 2.2 Stochastic difference equations White noise Wold’s decomposition (Wold, 1938: see Hamilton, 1994, chapter 4.8) states that every weakly stationary, purely non-deterministic, stochastic process can be written as a linear combination (or linear filter) of a sequence of uncorrelated random variables. By purely non-deterministic we mean that any linearly deterministic components have been subtracted from . This linear filter representation is given by The are a sequence of uncorrelated random variables, often known as innovations, drawn from a fixed distribution with and We will refer to such a sequence as a white-noise proc-ess, often denoting it as . The coefficie-nts (possibly infinite in number) in the linear filter are known as . 2.3 ARMA processes 2.3.1 Autoregressive processes Many realistic models result from particular choices of the weights. Taking with-out loss of generality, choosing allows to be writ-ten or This is known as a first-order autoregressive process, often given the acronym . The backshift (or lag) operator B is now introduced for notational convenience. This shifts time one step back, so that And, in general nothing that . The lag operator allows (possibly infinite) distributed lags to be written in a very concise way. For example, by using this notation the model can be written as so that This linear filter representation will converge as long as