Econ Papers Daily / 期刊论文
生产函数估计的统一方法：代理变量与动态面板数据
A Unified Approach to Estimating Production Functions: Proxy Variables and Dynamic Panel Data
摘要
We propose a new approach to production function estimation that integrates the strengths of the proxy‐variable (PV) and dynamic panel data (DPD) methods. Our framework augments the set of instruments for the level equation in Blundell and Bond [8] with a Berkson‐type instrument motivated by economic theory, following Olley and Pakes [28], Levinsohn and Petrin [24], and Ackerberg et al. [4]. This modification allows unobserved productivity to include both a time‐invariant (“fixed‐effect”) component and a time‐varying component that follows a potentially nonlinear Markov process. Whereas the PV approach accommodates nonlinear Markov dynamics but not fixed effects, and the DPD approach accounts for fixed effects but only with linear Markov dynamics, our method relaxes both restrictions. Our estimator is straightforward to implement using GMM. Monte Carlo simulations demonstrate that it outperforms the canonical PV and DPD estimators when productivity persistence is low and fixed‐effect heterogeneity is substantial.
中文摘要
本文提出一种新的生产函数估计方法，融合了代理变量法与动态面板数据法的优势。我们的框架在Blundell和Bond [8]的水平方程工具变量集中，引入了一个基于经济理论的Berkson型工具变量，遵循Olley和Pakes [28]、Levinsohn和Petrin [24]以及Ackerberg等人[4]的思路。这一改进允许不可观测的生产率同时包含时不变（“固定效应”）成分和遵循潜在非线性马尔可夫过程的时变成分。代理变量法虽能处理非线性马尔可夫动态但无法应对固定效应，动态面板数据法则能处理固定效应但仅适用于线性马尔可夫动态，而我们的方法同时放宽了这两项限制。该估计量可通过广义矩估计法简便实现。蒙特卡洛模拟表明，当生产率持续性较低且固定效应异质性显著时，本方法优于经典代理变量法和动态面板数据法估计量。