Testing for unit roots in heterogeneous panels

Any series of observations ordered along a single dimension, such as time, may be thought of as a time series. The emphasis in time series analysis is on studying the dependence among observations at different points in time. What distinguishes time series analysis from general multivariate analysis is precisely the temporal order imposed on the observations. Many economic variables, such as GNP and its components, price indices, sales, and stock returns are observed over time. In addition to being interested in the contemporaneous relationships among such variables, we are often concerned with relationships between their current and past values, that is, relationships over time.

Time Series Analysis

The panel data unit root test suggested by Levin and Lin (LL) has been widely used in several applications, notably in papers on tests of the purchasing power parity hypothesis. This test is based on a very restrictive hypothesis which is rarely ever of interest in practice. The Im–Pesaran–Shin (IPS) test relaxes the restrictive assumption of the LL test. This paper argues that although the IPS test has been offered as a generalization of the LL test, it is best viewed as a test for summarizing the evidence from a number of independent tests of the sample hypothesis. This problem has a long statistical history going back to R. A. Fisher. This paper suggests the Fisher test as a panel data unit root test, compares it with the LL and IPS tests, and the Bonferroni bounds test which is valid for correlated tests. Overall, the evidence points to the Fisher test with bootstrap-based critical values as the preferred choice. We also suggest the use of the Fisher test for testing stationarity as the null and also in testing for cointegration in panel data.

/pdf/a-comparative-study-of-unit-root-tests-with-panel-data-and-a-3mrtl7499l.pdf

A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test

A number of panel unit root tests that allow for cross section dependence have been proposed in the literature, notably by Bai and Ng (2002), Moon and Perron (2003), and Phillips and Sul (2002) who use orthogonalization type procedures to asymptotically eliminate the cross dependence of the series before standard panel unit root tests are applied to the transformed series. In this paper we propose a simple alternative test where the standard DF (or ADF) regressions are augmented with the cross section averages of lagged levels and first-differences of the individual series. A truncated version of the CADF statistics is also considered. New asymptotic results are obtained both for the individual CADF statistics, and their simple averages. It is shown that the CADF_i statistics are asymptotically similar and do not depend on the factor loadings under joint asymptotics where N (cross section dimension) and T (time series dimension) tends to infinity, such that N/T tends to k, where k is a fixed finite non-zero constant. But they are asymptotically correlated due to their dependence on the common factor. Despite this it is shown that the limit distribution of the average CADF statistic exists and its critical values are tabulated. The small sample properties of the proposed tests are investigated by Monte Carlo experiments, for a variety of models. It is shown that the cross sectionally augmented panel unit root tests have satisfactory size and power even for relatively small values of N and T. This is particularly true of cross sectionally augmented and truncated versions of the simple average t-test of Im, Pesaran and Shin, and Choi's inverse normal combination test.

A Simple Panel Unit Root Test in the Presence of Cross Section Dependence

A number of panel unit root tests that allow for cross section dependence have been proposed in the literature that use orthogonalization type procedures to asymptotically eliminate the cross dependence of the series before standard panel unit root tests are applied to the transformed series. In this paper we propose a simple alternative where the standard ADF regressions are augmented with the cross section averages of lagged levels and first-differences of the individual series. New asymptotic results are obtained both for the individual CADF statistics, and their simple averages. It is shown that the individual CADF statistics are asymptotically similar and do not depend on the factor loadings. The limit distribution of the average CADF statistic is shown to exist and its critical values are tabulated. Small sample properties of the proposed test are investigated by Monte Carlo experiments. The proposed test is applied to a panel of 17 OECD real exchange rate series as well as to log real earnings of households in the PSID data.

https://onlinelibrary.wiley.com/doi/pdf/10.1002/jae.951

A simple panel unit root test in the presence of cross-section dependence

When there is so much data that they become a computation burden, it is not uncommon to compute quantities of interest using a sketch of data of size $m$ instead of the full sample of size $n$. This paper investigates the implications for two-stage least squares (2SLS) estimation when the sketches are obtained by a computationally efficient method known as CountSketch. We obtain three results. First, we establish conditions under which given the full sample, a sketched 2SLS estimate can be arbitrarily close to the full-sample 2SLS estimate with high probability. Second, we give conditions under which the sketched 2SLS estimator converges in probability to the true parameter at a rate of $m^{-1/2}$ and is asymptotically normal. Third, we show that the asymptotic variance can be consistently estimated using the sketched sample and suggest methods for determining an inference-conscious sketch size $m$. The sketched 2SLS estimator is used to estimate returns to education.

Sketching for Two-Stage Least Squares Estimation.

We consider issues related to the order of an autoregression selected using information criteria. We study the sensitivity of the estimated order to i) whether the effective number of observations is held fixed when estimating models of different order, ii) whether the estimate of the variance is adjusted for degrees of freedom, and iii) how the penalty for overfitting is defined in relation to the total sample size. Simulations show that the lag length selected by both the Akaike and the Schwarz information criteria are sensitive to these parameters in finite samples. The methods that give the most precise estimates are those that hold the effective sample size fixed across models to be compared. Theoretical considerations reveal that this is indeed necessary for valid model comparisons. Guides to robust model selection are provided.

A Note on the Selection of Time Series Models

Datasets that are terabytes in size are increasingly common, but computer bottlenecks often frustrate a complete analysis of the data. While more data are better than less, diminishing returns suggest that we may not need terabytes of data to estimate a parameter or test a hypothesis. But which rows of data should we analyze, and might an arbitrary subset of rows preserve the features of the original data? This paper reviews a line of work that is grounded in theoretical computer science and numerical linear algebra, and which finds that an algorithmically desirable sketch, which is a randomly chosen subset of the data, must preserve the eigenstructure of the data, a property known as a subspace embedding. Building on this work, we study how prediction and inference can be affected by data sketching within a linear regression setup. We show that the sketching error is small compared to the sample size effect which a researcher can control. As a sketch size that is algorithmically optimal may not be suitable for prediction and inference, we use statistical arguments to provide 'inference conscious' guides to the sketch size. When appropriately implemented, an estimator that pools over different sketches can be nearly as efficient as the infeasible one using the full sample.

An Econometric Perspective on Algorithmic Subsampling

It is well known that the covariance structure of the data alone is not enough to identify an SVAR, and the conventional approach is to impose restrictions on the parameters of the model based on a priori theoretical considerations. This paper suggests that much can be gained by requiring the properties of the identified shocks to agree with major economic events that have been realized. We first show that even without additional restrictions, the data alone are often quite informative about the quantitatively important shocks that have occurred in the sample. We propose shrinking the set of solutions by imposing two types of inequality constraints on the shocks. The first restricts the sign and possibly magnitude of the shocks during unusual episodes in history. The second restricts the correlation between the shocks and variables external to the SVAR. The methodology provides a way to assess the validity of assumptions imposed as equality constraints. The effectiveness and limitations of this approach are exemplified with three applications.

Serena Ng

Papers

Sketching for Two-Stage Least Squares Estimation.

A Note on the Selection of Time Series Models

An Econometric Perspective on Algorithmic Subsampling

Shock Restricted Structural Vector-Autoregressions

An Econometric Perspective on Algorithmic Subsampling