A New Approach to Testing in the Generalized Method of Moments Framework

Timothy VogelsangSBR-9818695这个项目致力于构建一类新的检验统计量，可用于在广义矩方法(GMM)框架中检验假设。这一新类别的统计量不需要直接估计矩条件的长期方差(频率为零时的谱密度)。新的方法很可能会产生一类比目前可用的测试具有更好的有限样本属性的测试。这将是一项重要贡献，应该会对宏观经济和金融领域的实证工作产生积极影响，原因有两个。首先，GMM框架在实证工作中被广泛使用。应用领域包括经济周期模型的估计、随机波动率模型、资产定价模型、协方差结构的估计等。第二，有大量证据表明，应用于GMM模型的标准检验具有较差的有限样本性质。尺寸通常会膨胀，功率可能会很低。因此，GMM框架中的推理可能非常不精确。新的方法应该提供替代测试，给出更准确的推断。这些测试可能会吸引相对较大的用户群，因为新的测试很容易计算，并且不需要从业者做出诸如核、截断滞后、自动截断滞后近似模型和权重、滞后长度等选择。由于推理可能对这些选择敏感，新方法为从业者提供了一个简单、更健壮的测试框架。

Timothy VogelsangSBR-9818695This project is concerned with constructing a new class of test statistics that can be used to test hypotheses in the generalized method of moments (GMM) framework. The statistics in this new class do not require direct estimates of the long run variance (spectral density at frequency zero) of the moment conditions. It is likely the new approach will result in a class of tests that have better finite sample properties than currently available tests. This will be an important contribution and should positively impact empirical work in macroeconomics and finance for two reasons. First, the GMM framework is widely used in empirical work. Applications include estimation of business cycle models, stochastic volatility models, asset-pricing models, estimation of covariance structures, etc. Second, there is considerable evidence that standard tests applied to GMM models have poor finite sample properties. Size is often inflated, and power can be low. Thus, inference in the GMM framework can be very imprecise. The new approach should provide alternative tests that give more precise inference. The tests are likely to attract a relatively large club of users because the new tests are easy to compute and do not require the practitioner to make such choices as kernel, truncation lag, automatic truncation lag approximating models and weights, lag lengths, etc. Because inference can be sensitive to those choices, the new approach provides practitioners with a simple more robust testing framework.