Evaluation of Policy Impacts

这笔赠款中涵盖的第一个项目是与劳动力市场一般均衡模型估计的工作的继续(由国家科学基金会-SBR 9730657提供支持)。人们不能在“局部均衡环境”中恰当地评估国家政策，这是众所周知的，但应用微观经济学家很少谈到这一点。对于解决技能获取问题的政策来说，情况尤其如此。在使用数据进行政策分析时，劳动经济学家通常只关心技能的供应，而忽略了需求。问题是，随着拥有特定技能的个人数量的增加，该技能在劳动力市场上的价值会下降。这种效果阻碍了技能的形成。我们之前的研究表明，在进行政策评估时考虑到一般均衡效应，可以将结果改变多达10倍。到目前为止，我们工作的主要弱点是经验方法。我们已经使用标准方法估计了单个工人的参数，这些方法隐含地假设劳动力市场不会随着时间的推移而改变。这种“稳定状态”的假设显然是有问题的，因为已经有大量文献证明了工资结构的变化。最大的问题是，我们估计的模型与我们在模拟时使用的一般均衡模型不一致。这项工作的主要目标是估计美国劳动力市场过去30年的动态一般均衡模型。这涉及到同时估计和模拟考虑到不断变化的美国劳动力市场的模型。这一过程确保了模拟模型与美国经济之间的相似性。我们正在使用估计模型来审查旨在解决劳动力日益不平等问题的几项不同政策。这些计划的一个主要目标是降低收入不平等。这一版本的模型将提供对收入在经济中分配的一致估计，并将使我们能够比以前的工作更好地估计政策对收入不平等的影响。我们还在扩展模型，以考虑针对低薪工人的福利政策和工资补贴的影响，以及旨在解决收入不平等问题的其他一些政策。第二个项目的目标是解决一个特别的方面，这个方面在大多数差额估计器的实施中可能非常重要。当一个单位“改变”某些特定政策时，往往会出现关键参数的识别。当研究人员在他们的模型中进行推理时，他们通常假设观察次数和时间段的乘积很大。然而，即使单位或时间段的数量很大，在数据中观察到的实际政策变化的数量通常也很少。在这种情况下，研究人员在这些模型中用于推断的这些标准方法是不合适的，可能具有极大的误导性。我们正在开发两种不同的方法，允许研究人员在这些模型中进行推理。对于第一种方法，我们假设数据中有有限数量的政策变化，但使用渐近近似作为观察数量和时间周期的乘积变大。在第二种方法中，我们通过使用精确检验完全避免了大样本近似。

The first project covered in this grant is a continuation of work with on the estimation of general equilibrium models of the labor market (supported by NSF-SBR 9730657). It is well known but rarely addressed by applied microeconomists that one can not properly evaluate national policies in a "partial equilibrium environment." This is particularly true for policies that address skill acquisition. When using data to perform policy analysis, labor economists typically only concern themselves with the supply of skill and ignore demand. The problem is that as the number of individuals possessing a particular skill increases, the value the skill in the labor market falls. This effect discourages skill formation. Our previous research shows that accounting for general equilibrium effects when performing policy evaluations can alter the results by as much as a factor of ten. The major weakness of our work to date is the empirical approach. We have estimated the parameters of the individual worker using standard methods which implicitly assume that the labor market is not changing over time. This "steady state" assumption is clearly questionable as there has been a large literature demonstrating the changes in the wage structure. The biggest problem is that the model we estimated is not consistent with the general equilibrium model that we used when we simulated it. The main goal of this work is to estimate a dynamic general equilibrium model of the U.S. labor market over the last thirty years. This involves simultaneously estimating and simulating the model accounting for the changing U.S. labor market. This procedure guarantees similarity between the simulated model and the U.S. economy. We are using the estimated model to examine several different policies aimed to address the increasing inequality in the workforce. A major goal of these programs is to lower earnings inequality. This version of the model will provide a consistent estimate of the distribution of earnings in the economy and will allow us to do a much better job of estimating the effects of the policies on earnings inequality than our previous work. We are also extending the model to consider the effects of welfare policies and wage subsidies aimed a low wage workers as well as a number of other policies aimed at addressing earnings inequality.The goal of the second project is to address one particular aspect that is likely to be very important in most implementations of difference-in-differences estimators. Identification of the key parameter often arises when a unit "changes" some particular policy. Researchers typically assume that the product of the number of observations and time period is large when they perform inference in their models. However, even when the number of units or time periods is large, the number of actual policy changes observed in the data is typically small. In this case these standard methods that researchers use for inference in these models are not appropriate and may be wildly misleading. We are developing two different approaches which allow researchers to perform inference in these models. For the first approach we assume that there are a finite number of policy changes in the data, but use asymptotic approximations as the product of the number of observations and time period gets large. In the second, we avoid large sample approximations altogether by using exact tests.