Least-squares estimation of an unknown number of shifts in a time series (Q2742772)

The authors generalize the Hájek-Rényi inequality. They use this inequality to obtain general results on the off-line least-squares estimate of changes in the mean of a random processes. Let \(\{ \varepsilon_t\}_{t \geq 0}\) be a sequence of random variables, \(S_{i.j.} =\sum_{t=i}^j \varepsilon_t\), \(1 \leq i < j\leq n.\) The authors suppose that there exists \(C(\varepsilon) <\infty\) such that \(E( S_{i.j.})^2 \leq C(\varepsilon) |j-i+1|^{\phi}\) for all \(i, j\) (condition H1\((\phi)\)). Condition H1\((\phi)\) is fulfilled for a wide family of zero-mean processes, for example, for linear processes of the form \(\varepsilon_t = \sum_{j=0}^\infty f_j \psi_{t-j}\), where \(\{ \psi_t\}_{j\in Z}\) is a sequence of independent variables with zero-mean and finite variance such that \(\sum_{j=0}^\infty j |f_j|< \infty\). The following theorem is proved.NEWLINENEWLINENEWLINELet \(\{ \varepsilon_t\}_{t \in Z}\) satisfy the condition H1\((\phi)\) for some \(1< \phi <2.\) Then a constant \(A(\phi)\geq 1\) exists such that for any \(n \geq 1\), \(1\leq m \leq n\), for any \(\delta >0\), and for any positive sequence \(b_1 \geq b_2 \geq \ldots \geq b_n >0\), the following inequality holds true: NEWLINE\[NEWLINEP(\max_{m \leq k \leq n} b_k |S_{i.k}|> \delta) \leq 4C(\varepsilon)m^{\phi}bm^2\delta^{-2}+4C(\varepsilon)A{\phi}\delta^{-2}(n-m)^{\phi -1}\sum{t=m+1}^n b_t^2.NEWLINE\]NEWLINE The authors consider the model \(y_t = \mu_k^{\ast} + \varepsilon_t\), \(t_{k-1}^{\ast} +1 \leq t_k^{\ast}\), \(1\leq k _\leq r.\) The indices of the break points and the mean values \(\mu_1^{\ast}, \ldots, \mu_r^{\ast}\) are treated as unknown. The purpose is to estimate the unknown means together with the break points when \(n\) observations \((Y_1, \ldots, Y_n)\) are available, and these estimates are presented. They derive the consistency and the rate of convergence of these estimators in the situation where the number of changes is known. Strong consistency is obtained under some mixing conditions. The limiting distribution is also computed under an invariance principle. The case where the number of changes is unknown is addressed.