We propose a novel dynamic panel estimator. Different from the commonly used difference and system GMM, our proposed estimator requires only one of the crosssectional dimension (N) or the time dimension (T) to grow large to be asymptotically unbiased. This improves reliability in panels with long time spans, where GMM suffers from weak instrument problems, and more generally in finite samples where results can be sensitive to instrument selection and implementation choices. Computationally simple, it extends readily to higher-order autoregressive and vector autoregressive settings. Monte Carlo simulations show that the estimator exhibits lower finite-sample bias than GMM in shorter panels, including for roots at and near unity. In three applications from political economy and macroeconomics—spanning diverse panels, outcomes, and persistence levels—our estimator yields stable, economically meaningful estimates robust to specification choices. By contrast, standard GMM methods display considerable sensitivity to instrument lags, collapsing, and the choice between difference and system variant, often producing substantively different results under comparable setups.