Efficient Estimation of the Geometric Distributed Lag Model: Some Monte Carlo Results on Small Sample Properties
通过蒙特卡洛模拟，比较了Koyck分布滞后模型中工具变量、两步有效估计和迭代高斯-牛顿估计的小样本性质，发现两步有效估计通常优于工具变量估计，且与迭代估计性质相似。
In this paper we report Monte Carlo results on the small sample properties of instrumental variables, asymptotically efficiënt two-step and iterative Gauss-Newton estimators for a Koyck (1954) distributed lag model with uncorrelated errors (model 1) and with first order autoregressive errors (model 2).We use the technique of control variables to increase the precision of the Monte Carlo results" and summarize the outcome using response functions.Two main questions have been investigated for a sample size T=30 and T=60: (a) are the asymptotically efficiënt estimators to be preferred to a consistent but inefficiënt instrumental variables estimator?, (b) does it pay to iterate an asymptotically efficiënt estimator until convergence is achieved?For the sample sizes considered, we conclude that the efficiënt two-step estimator is usually preferred to the instrumental variables estimator and that it has properties which are very similar to those of the iterative Gauss-Newton estimator.