This paper studies unit root testing in large panels with cross-sectional dependence. The authors propose tests for unit roots in panels where cross-sectional units are correlated, modeling this correlation using an unknown number of unobservable common factors. They derive the asymptotic distribution of their tests under the null hypothesis of a unit root and local alternatives. The tests are shown to have significant asymptotic power when there are no incidental trends, but lose power when incidental trends are present and deterministic components must be removed. Monte Carlo simulations are used to evaluate the finite sample properties of these tests. The paper introduces a dynamic panel model with fixed effects, suitable for macroeconomic data without deterministic trends. It considers a near unit root model where the autoregressive parameter is close to one. The model assumes that the error term follows an approximate factor model, with unobservable common factors and idiosyncratic shocks. The authors derive assumptions on the error terms, factors, and other variables, and show that the long-run covariance matrix of the factors does not need to be of full rank. They also show that the variance matrix of the factors must be positive definite. The paper concludes that the proposed tests have power against local alternatives that shrink towards the unit root at a rate of 1/√nT under certain conditions. However, the tests do not have power in the presence of incidental trends. The authors also provide an upper bound on the rate at which the alternative hypothesis can drift towards the null for nontrivial power to exist. The paper is organized into sections introducing the model, proposing tests, analyzing their properties, and comparing finite sample properties. All technical proofs and derivations are in the appendix.This paper studies unit root testing in large panels with cross-sectional dependence. The authors propose tests for unit roots in panels where cross-sectional units are correlated, modeling this correlation using an unknown number of unobservable common factors. They derive the asymptotic distribution of their tests under the null hypothesis of a unit root and local alternatives. The tests are shown to have significant asymptotic power when there are no incidental trends, but lose power when incidental trends are present and deterministic components must be removed. Monte Carlo simulations are used to evaluate the finite sample properties of these tests. The paper introduces a dynamic panel model with fixed effects, suitable for macroeconomic data without deterministic trends. It considers a near unit root model where the autoregressive parameter is close to one. The model assumes that the error term follows an approximate factor model, with unobservable common factors and idiosyncratic shocks. The authors derive assumptions on the error terms, factors, and other variables, and show that the long-run covariance matrix of the factors does not need to be of full rank. They also show that the variance matrix of the factors must be positive definite. The paper concludes that the proposed tests have power against local alternatives that shrink towards the unit root at a rate of 1/√nT under certain conditions. However, the tests do not have power in the presence of incidental trends. The authors also provide an upper bound on the rate at which the alternative hypothesis can drift towards the null for nontrivial power to exist. The paper is organized into sections introducing the model, proposing tests, analyzing their properties, and comparing finite sample properties. All technical proofs and derivations are in the appendix.