This paper develops an asymptotic theory of inference for an unrestricted two-regime Ž . threshold autoregressive TAR model with an autoregressive unit root. We find that the asymptotic null distribution of Wald tests for a threshold are nonstandard and different from the stationary case, and suggest basing inference on a bootstrap approximation. We also study the asymptotic null distributions of tests for an autoregressive unit root, and find that they are nonstandard and dependent on the presence of a threshold effect. We propose both asymptotic and bootstrap-based tests. These tests and distribution theory Ž . Ž allow for the joint consideration of nonlinearity thresholds and nonstationary unit . roots .Our limit theory is based on a new set of tools that combine unit root asymptotics with empirical process methods. We work with a particular two-parameter empirical process that converges weakly to a two-parameter Brownian motion. Our limit distributions involve stochastic integrals with respect to this two-parameter process. This theory is entirely new and may find applications in other contexts.We illustrate the methods with an application to the U.S. monthly unemployment rate. We find strong evidence of a threshold effect. The point estimates suggest that the threshold effect is in the short-run dynamics, rather than in the dominate root. While the conventional ADF test for a unit root is insignificant, our TAR unit root tests are arguably significant. The evidence is quite strong that the unemployment rate is not a unit root process, and there is considerable evidence that the series is a stationary TAR process.

Threshold models~sample splitting models! have wide application in economics+ Existing estimation methods are confined to regression models, which require that all right-hand-side variables are exogenous+ This paper considers a model with endogenous variables but an exogenous threshold variable+ We develop a twostage least squares estimator of the threshold parameter and a generalized method of moments estimator of the slope parameters+ We show that these estimators are consistent, and we derive the asymptotic distribution of the estimators+ The threshold estimate has the same distribution as for the regression case~Hansen, 2000, Econometrica 68, 575-603!, with a different scale+ The slope parameter estimates are asymptotically normal with conventional covariance matrices+ We investigate our distribution theory with a Monte Carlo simulation that indicates the applicability of the methods+ We thank the two referees and co-editor for constructive comments+ Hansen thanks the National Science Foundation for financial support+ Caner thanks

Tests of the null hypothesis of stationarity against the unit root alternative play an increasingly important role in empirical work in macroeconomics and in international finance. We show that the use of conventional asymptotic critical values for stationarity tests may cause extreme size distortions, if the model under the null hypothesis is highly persistent. This fact calls into question the use of these tests in empirical work. We illustrate the practical importance of this point for tests of long-run purchasing power parity under the recent float. We show that the common practice of viewing tests of stationarity as complementary to tests of the unit root null will tend to result in contradictions and in spurious rejections of long-run PPP. While the size distortions may be overcome by the use of finite-sample critical values, the resulting tests tend to have low power under economically plausible assumptions about the half-life of deviations from PPP. Thus, the fact that stationarity is not rejected cannot be interpreted as convincing evidence in favor of mean reversion. Only in the rare case that stationarity is rejected do size-corrected tests shed light on the question of long-run PPP.

This paper proposes the least absolute shrinkage and selection operator–type (Lasso-type) generalized method of moments (GMM) estimator. This Lasso-type estimator is formed by the GMM objective function with the addition of a penalty term. The exponent of the penalty term in the regular Lasso estimator is equal to one. However, the exponent of the penalty term in the Lasso-type estimator is less than one in the analysis here. The magnitude of the exponent is reduced to avoid the asymptotic bias. This estimator selects the correct model and estimates it simultaneously. In other words, this method estimates the redundant parameters as zero in the large samples and provides the standard GMM limit distribution for the estimates of the nonzero parameters in the model. The asymptotic theory for our estimator is nonstandard. We conduct a simulation study that shows that the Lasso-type GMM correctly selects the true model much more often than the Bayesian information Criterion (BIC) and another model selection procedure based on the GMM objective function.

No abstract