Nonparametric Estimation and Testing in Panels of Intercorrelated Time Series (Q4677045)

The authors consider the model for a panel of time series \[ X_{(i)t}=f(X_{(i)t-1},\dots, X_{(i)t-p})+\eta_{t}+\varepsilon_{(i)t}, \] where \(\{X_{(i)t}, i=1,\dots,n;\;t=1,\dots,T\}\) are observations from \(n\) individual time series with \(T\) observations taken from each. The process \(\{\eta_{t}\}\) describes the intercorrelation (or interdependence) in the panel, \(\{\epsilon_{(i)t}\}\) are the error terms assumed to be i.i.d., \(\eta_{t}\) is assumed to be independent of \(\{\epsilon_{(i)t}\}\) and it could be thought of as an external synchronous agent influencing all the series. For the considered model the authors concentrate on two problems: estimating \(f\) nonparametrically and testing whether \(X_{(i)t}\) is linearly generated. Tests of linearity are described and nonparametric estimation of \(f\) is proposed. Simulation experiments on real data sets are presented.