Estimating Dynamic Panel Data Models.

Prior studies on dynamic panel estimation have concentrated on small time dimensions and large individual dimensions. This paper uses the Monte Carlo approach to examine the performance of different approaches designed to decrease the bias of the estimated coefficients for the longer, narrower panels generally found for macro data. The authors find there is considerable bias of the least squares dummy variable approach even when the time dimension of the panel is as large as 30. For panels with small time dimensions, they find a corrected least squares dummy variable estimator to be the best choice. The authors also find that the Anderson-Hsiao estimator performs just as well. The authors apply their recommendations to a panel of countries to show that increases in income growth appear before increases in savings rates and increases in savings rates appear before income growth. This working paper is available at the US Federal Reserve Board. You can access this site by going to www.federalreserve.gov/pubs/workingpapers.htm.