zbMATH — the first resource for mathematics

Change-point estimation in ARCH models. (English) Zbl 0997.62068
From the paper: This paper studies ARCH models defined by the equations $r_k=\sigma_k \varepsilon_k, \quad\sigma^2_k= a+\sum^\infty_{j=1} b(j)r^2_{k-j}, \tag{1}$ where the $$\varepsilon_k$$ are independent and identically distributed (i.i.d.) errors and $$a$$ and the $$b(j)$$ are non-negative constants. Since their introduction ARCH type models have become perhaps the most popular and extensively studied financial econometric models. We study the estimation of a change-point in (1). Note first that if the $$\varepsilon_k$$ have mean zero and unit variance and the sequence $$\{r_k\}$$ is weakly stationary, then $\text{var} [r_k]= a\left/ \left(1-\sum^\infty_{j=1} b(j) \right.\right). \tag{2}$ Suppose now that the parameters $$a$$ and $$b(j)$$ change at an unknown point $$k^*$$ in such a way that the variance given by (2) changes. We are interested in estimating the change-point $$k^*$$. This problem is different from the problem of testing for structural changes where $$k^*$$ is assumed to be known and one tests whether a change in model structure has taken place at $$k^*$$.
Besides the change-point problem we study the cross-covariance function for ARCH models. Bounds for the cross-covariance function are derived and explicit formulae are obtained in special cases. Consistency of a CUSUM type change-point estimator is proved and its rate of convergence is established. A Hájek-Rényi type inequality is also proved. The results are obtained under weak moment assumptions.

MSC:
 62M10 Time series, auto-correlation, regression, etc. in statistics (GARCH)
Full Text: