This paper considers estimation of short-run dynamics in time series that contain a nonstationary component. We assume that appropriate preliminary methods can be applied to the observed time series to separate short-run elements from long-run slowly evolving secular components, and focus on estimation of the short-run dynamics based on the filtered data. We use a flexible copula-generated Markov model to capture the nonlinear temporal dependence in the short-run component and study estimation of the copula model. Using the rescaled empirical distribution of the filtered data as an estimator of the marginal distribution, Chen et al. (2022) proposed a simple, yet flexible, two-step estimation procedure for the copula model. The two-step estimator works well when the tail dependence is small. However, simulations reveal that the two-step estimator may be biased in finite samples in the presence of tail dependence. To improve the performance of short-term dynamic analysis in the presence of tail dependence, we propose in this paper a pseudo sieve maximum likelihood (PSML) procedure to jointly estimate the residual copula parameter and the invariant density of the filtered residuals. We establish the root-consistency and asymptotic distribution of the PSML estimator of any smooth functional of the residual copula parameter and invariant residual density. We further show that the PSML estimator of the residual copula parameter is asymptotically normal, with the limiting distribution independent of the filtration. Simulations reveal that in the presence of strong tail dependence, compared to the two-step estimates of Chen et al. (2022), the proposed PSML estimates have smaller biases and smaller mean squared errors even in small samples. Applications to nonstationary macro-finance and climate time series are presented.
We propose a modification of kernel time series regression estimators that improves efficiency when the innovation process is autocorrelated. The procedure is based on a pre-whitening transformation of the dependent variable that has to be estimated from the data. We establish the asymptotic distribution of our estimator under weak dependence conditions. It is shown that the proposed estimation procedure is more efficient than the conventional kernel method. We also provide simulation evidence to suggest that gains can be achieved in moderate sized samples.
We show that the conventional CUSUM test for structural change can be applied to cointegrating regression residuals leading to a consistent residual based test for the null hypothesis of cointegration. The proposed tests are semiparametric and utilize fully modified residuals to correct for endogeneity and serial correlation and to scale out nuisance parameters. The limit distribution of the test is derived under both the null and the alternative hypothesis. The tests are easy to use and are found to perform quite well in a Monte Carlo experiment.
JEL Classification: C22
Keywords: Bandwidth, CUSUM test, Fully modified regression, Null of cointegration, Residual based test, Semiparametric method
Asymptotic expansions are developed for Wald test statistics in cointegrating regression models. These expansions provide an opportunity to reduce size distortion in testing by suitable bandwidth selection, and automated rules for doing so are calculated. Band spectral regression methods and tests are also considered. In such cases, it is shown how the effects of nonstationarity that dominate low frequency limit behaviour also carry over to high frequency asymptotics, with consequential effects on bandwidth rules.
This paper proposes an ADF coefficient test for detecting the presence of a unit root in ARMA models of unknown order. Our approach is fully parametric. When the time series has an unknown deterministic trend, we propose a modified version of the ADF coefficient test based on quasi-differencing in the construction of the detrending regression as in Elliot, Rothenberg and Stock (1996). The limit distributions of these test statistics are derived. Empirical applications of these tests for common macroeconomic time series in the US economy are reported and compared with the usual ADF t-test.