Panicca: Panic on Cross‐Section Averages

Abstract

Summary The cross‐section average (CA) augmentation approach of Pesaran (A simple panel unit root test in presence of cross‐section dependence. Journal of Applied Econometrics 2007; 22: 265–312) and Pesaran et al. (Panel unit root test in the presence of a multifactor error structure. Journal of Econometrics 2013; 175: 94–115), and the principal components‐based panel analysis of non‐stationarity in idiosyncratic and common components (PANIC) of Bai and Ng (A PANIC attack on unit roots and cointegration. Econometrica 2004; 72: 1127–1177; Panel unit root tests with cross‐section dependence: a further investigation. Econometric Theory 2010; 26: 1088–1114) are among the most popular ‘second‐generation’ approaches for cross‐section correlated panels. One feature of these approaches is that they have different strengths and weaknesses. The purpose of the current paper is to develop PANICCA, a combined approach that exploits the strengths of both CA and PANIC. Copyright © 2015 John Wiley & Sons, Ltd.

Optimal panel unit root testing with covariates

“…In order to avoid spurious results when z i,t in non-stationary, the estimation is carried out using the first-differenced data, and the estimated first-differenced idiosyncratic errors are then accumulated up to levels. This givesû i,t , which is consistent for u i,t under very general conditions (see Bai andNg, 2004, andReese andWesterlund, 2016). Testingû i,t is therefore asymptotically equivalent to testing u i,t , which motivates our approach.…”

“…Two approaches will be employed, which differ only in how the factors are estimated. Specifically, while the PANIC approach of Bai and Ng (2004) is based on estimated principal component factors, the PANICCA approach of Reese and Westerlund (2016) instead takes the cross-sectional average of the observables as the estimated factors. In order to avoid spurious results when z i,t in non-stationary, the estimation is carried out using the first-differenced data, and the estimated first-differenced idiosyncratic errors are then accumulated up to levels.…”

Parameter Heterogeneity, Persistence and Cross-Sectional Dependence: New Insights on Fiscal Policy Reaction Functions for the Euro Area

“…This outcome can be rationalized in the light of both the specific span of available data and the institutional framework. Indeed, the finding 18 These statistics are the analogs of ta and tb of Moon and Perron (2004) who adopt a different defactoring method and consider the factors to be just nuisance parameters 19 Some minor inconsistency across idiosyncratic components unit-root tests results of Pa, Pb and PMSB can be explained in the light of the Monte Carlo simulations in Reese and Westerlund (2016), who find evidence of Pa and Pb over-rejection and of PMSB under-rejection for N=10 and T=20.…”

“…To test for the nonstationarity of the idiosyncratic components it  , Bai and Ng (2004) and Reese and Westerlund (2016) proceed by pooling individual ADF t statistics obtained on defactored residuals, with the factors proxied either by principal component estimates (Bai and Ng, 2004) or by cross-sectional averages augmentation (Reese and Westerlund, 2016). As the idiosyncratic components it  in a factor model is by construction only weakly correlated across units while factors t f involve strong correlation, the pooled tests based on defactored residuals likely satisfies the crosssectional independence assumption.…”

“…In the present case, given the assumption of a single factor, a simple ADF-type test for nonstationarity of the factors can be used. (Bai and Ng, 2004, and Reese and Westerlund, 2016.…”