scientific article; zbMATH DE number 1124615

The problem of detection and identification of changes in statistical models is known as the change point problem (mostly for case of changes in location models), disorder problem or testing the constancy of regression relationship over time. These problems arise in a number of applications (economic modelling, quality control, biology, medicine, meteorology and ecology among others). We consider here the following regression model with possible change after an unknown time point \(m\): \[ Y_i=x^T_i\beta+x_i\delta_nI\{i>m\}+E_i,\quad i=1,\dots,n, \] where \(x_i=(x_{i1},\dots,x_{ip})^T\), \(x_{i1}=1\), \(i=1,\dots,n\), are known regression vectors, \(m (<n)\), \(\beta=(\beta_1,\dots,\beta_p)^T\), \(\delta_n=(\delta_{n1},\dots,\delta_{np})^T\) are unknown parameters, \(E_1,\dots,E_n\) are i.i.d. random variables with common distribution function \(F\). \(I\{A\}\) denotes the indicator of the set \(A\). The model corresponds to the situation when up to an unknown \(m\) the observations follow the regression model with the regression parameter \(\beta\) and then the model changes to the regression model with the regression parameter \(\beta+\delta_n\). The parameter \(m\) is called the change point. The problem of our interest is to test \[ H_0:m=n\text{ against }H_1:m<n. \] Authors usually apply either the likelihood ratio principle or the Bayesian approach. The first principle leads to max-type procedures, the other gives sum-type procedures. We describe likelihood ratio and related procedures when the distribution of the error terms \(F\) is \(N(0,\sigma^2)\), \(\sigma^2>0\) known. This will give motivation how to develop \(L_1\)-procedures. For the entire collection see [Zbl 0882.00044].

Recommendations

Cited in

(11)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4378644)