Bias Reduction by Recursive Mean Adjustment in Dynamic Panel Data Models

Accurate estimation of the dominant root of a stationary but persistent time series are required to determine the speed at which economic time series, such as real exchange rates or interest rates, adjust towards their mean values. In practice, accuracy is hampered by downward small- sample bias. Recursive mean adjustment has been found to be a useful bias reduction strategy in the regression context. In this paper, we study recursive mean adjustment in dynamic panel data models. When there exists cross-sectional heterogeneity in the dominant root, the recursive mean adjusted SUR estimator is appropriate. When homogeneity restrictions can be imposed, a pooled recursive mean adjusted GLS estimator with fixed e¤ects is the desired estimator. Application of these techniques to a small panel of five eurocurrency rates finds that these interest rates are unit root nonstationary as the bias-corrected autoregressive coefficient exceeds 1.