Threshold models have a wide variety of applications in economics. Direct a
pplications include models of separating and multiple equilibria. Other app
lications include empirical sample splitting when the sample split is based
on a continuously-distributed variable such as firm size. In addition, thr
eshold models may be used as a parsimonious strategy for nonparametric func
tion estimation. For example, the threshold autoregressive model (TAR) is p
opular in the nonlinear time series literature.
Threshold models also emerge as special cases of more complex statistical f
rameworks, such as mixture models, switching models, Markov switching model
s, and smooth transition threshold models. It may be important to understan
d the statistical properties of threshold models as a preliminary step in t
he development of statistical tools to handle these more complicated struct
ures.
Despite the large number of potential applications, the statistical theory
of threshold estimation is undeveloped. It is known that threshold estimate
s are super-consistent, but a distribution theory useful. for testing and i
nference has yet to be provided.
This paper develops a statistical theory for threshold estimation in the re
gression context. We allow for either cross-section or time series observat
ions. Least squares estimation of the regression parameters is considered.
An asymptotic distribution theory for the regression estimates (the thresho
ld and the regression slopes) is developed. It is found that the distributi
on of the threshold estimate is nonstandard. A method to construct asymptot
ic confidence intervals is developed by inverting the likelihood ratio stat
istic. It is shown that this yields asymptotically conservative confidence
regions. Monte Carlo simulations are presented to assess the accuracy of th
e asymptotic approximations. The empirical relevance of the theory is illus
trated through an application to the multiple equilibria growth model of Du
rlauf and Johnson (1995).