1: \begin{abstract}

2: \noindent This paper develops a threshold regression model where an unknown

3: relationship between two variables nonparametrically determines the

4: threshold. We allow the observations to be cross-sectionally dependent so

5: that the model can be applied to determine an unknown spatial border for

6: sample splitting over a random field. We derive the uniform rate of

7: convergence and the nonstandard limiting distribution of the nonparametric

8: threshold estimator. We also obtain the root-n consistency and the

9: asymptotic normality of the regression coefficient estimator. Our model has

10: broad empirical relevance as illustrated by estimating the tipping point in

11: social segregation problems as a function of demographic characteristics;

12: and determining metropolitan area boundaries using nighttime light intensity

13: collected from satellite imagery. We find that the new empirical results are

14: substantially different from those in the existing studies.

15: 

16: \vspace{0.15in}

17: 

18: \noindent \noindent \textit{\noindent Keywords}: threshold regression,

19: sample splitting, nonparametric, random field, tipping point, metropolitan

20: area boundary.

21: 

22: \noindent \textit{\noindent JEL Classifications}: C14, C21, C24, R1

23: \end{abstract}

24: