1: \begin{abstract}

2: This paper is concerned with inference in regression models with either a

3: kink or a jump at an unknown threshold, particularly when we do not know

4: whether the kink or jump is the true specification. One of our main results

5: shows that the statistical properties of the estimator of the threshold

6: parameter are substantially different under the two settings, with a slower

7: rate of convergence under the kink design, and more surprisingly slower than

8: if the correct kink specification were employed in the estimation. We thus

9: propose two testing procedures to distinguish between them. Next, we develop

10: a robust inferential procedure that does not require prior knowledge on

11: whether the regression model is kinky or jumpy. Furthermore, we propose to

12: construct confidence intervals for the unknown threshold by the bootstrap

13: test inversion, also known as grid bootstrap. Finite sample performances of

14: the bootstrap tests and the grid bootstrap confidence intervals are examined

15: and compared against tests and confidence intervals based on the asymptotic

16: distribution through Monte Carlo simulations. Finally, we implement our

17: procedure to an economic empirical application.

18: \end{abstract}

19: