This paper by Blundell and Bond (1998) addresses the estimation of Cobb-Douglas production functions using panel data, focusing on the challenges posed by persistent firm-specific effects. Standard GMM estimators, which eliminate these effects by taking first differences, have been found to produce unsatisfactory results due to weak instruments. The authors attribute this to the high persistence of firm sales, capital, and employment series, leading to weak correlations between lagged levels and subsequent first differences. To address this issue, the authors propose an extended GMM estimator that incorporates additional moment conditions based on reasonable stationarity restrictions on the initial conditions process. This estimator uses lagged first-differences of the series as instruments for the levels equations, in addition to the usual lagged levels as instruments for the first-difference equations. They demonstrate that this approach significantly reduces finite-sample biases and yields more reasonable parameter estimates. Using data from a panel of R&D-performing US manufacturing companies, the authors confirm that the first-differenced GMM estimator produces low and statistically insignificant capital coefficients and suggests decreasing returns to scale. However, the extended GMM estimator provides more accurate estimates, with a higher and significantly positive capital coefficient and no rejection of constant returns to scale. The importance of allowing for an autoregressive component in the productivity shocks is also highlighted. The paper concludes that the system GMM estimator is more effective in this context, providing more reliable parameter estimates and reducing finite-sample biases associated with first-differenced GMM.This paper by Blundell and Bond (1998) addresses the estimation of Cobb-Douglas production functions using panel data, focusing on the challenges posed by persistent firm-specific effects. Standard GMM estimators, which eliminate these effects by taking first differences, have been found to produce unsatisfactory results due to weak instruments. The authors attribute this to the high persistence of firm sales, capital, and employment series, leading to weak correlations between lagged levels and subsequent first differences. To address this issue, the authors propose an extended GMM estimator that incorporates additional moment conditions based on reasonable stationarity restrictions on the initial conditions process. This estimator uses lagged first-differences of the series as instruments for the levels equations, in addition to the usual lagged levels as instruments for the first-difference equations. They demonstrate that this approach significantly reduces finite-sample biases and yields more reasonable parameter estimates. Using data from a panel of R&D-performing US manufacturing companies, the authors confirm that the first-differenced GMM estimator produces low and statistically insignificant capital coefficients and suggests decreasing returns to scale. However, the extended GMM estimator provides more accurate estimates, with a higher and significantly positive capital coefficient and no rejection of constant returns to scale. The importance of allowing for an autoregressive component in the productivity shocks is also highlighted. The paper concludes that the system GMM estimator is more effective in this context, providing more reliable parameter estimates and reducing finite-sample biases associated with first-differenced GMM.