We next consider the performance of the estimation procedure proposed in Section 2.1. Figure 2 presents the kernel densities for the regression slope estimators for selected samples and the frequency k = 1. As predicted by Theorem 1, the consistency of the time-varying threshold estimator implies that the observations can be correctly classified into subsets, and hence the true model slope parameters β1 and β2 can be consistently estimated. Overall, the parameter estimation procedure works well in the finite sample properties. Table 1 presents the summary statistics (i.e., mean, and standard deviation) for the least-square parameter estimates.