Deviation inequalities and Cramér-type moderate deviations for the explosive autoregressive process (Q2676936)

Let \(n\geq1\) be fixed, and consider the first-order Gaussian autoregressive model given by \(y_t=\theta_ny_{t-1}+u_t\) for \(t=1,\ldots,n\), where \(y_0\in\mathbb{R}\), \((u_t)\) is a sequence of mean-zero Gaussian random variables with finite variance, and \(\theta_n=1+\frac{c}{k_n}\) for some \(c>0\). The authors study the explosive case, where \(k_n=1\) for all \(n\), and the mildly explosive case in which \(k_n\to\infty\) with \(k_n=o(n)\). They establish deviation inequalities and Cramér-type moderate deviations for the quadratic functionals \(\sum_{t=1}^ny_{t-1}u_t\) and \(\sum_{t=1}^ny_{t-1}^2\), which are then applied to establish corresponding results for the least-squares estimator \(\hat{\theta}_n\) of \(\theta_n\). As a statistical application of these results, the authors consider the unit root test of the hypotheses \(H_0\mbox{: }\theta_n=1\) versus \(H_1\mbox{: }\theta_n=1+\frac{c}{k_n}\), where \(c\not=0\) and either \(k_n=1\) or \(k_n\to\infty\) with \(k_n=o(n)\). The asymptotic power of the test statistic is analysed in this setting. Numerical simulations are also presented to illustrate results on moderate deviations and the unit root test.