Inference About Predictive Accuracy; Exchange Rate Volatility: Inventory Models

9309927 West数据挖掘或过度拟合，可能是由于专业范围内重复使用有限的数据体，有时会导致样本拟合良好。一些经济学家认为，样本外预测的准确性是衡量模型质量的关键指标。不幸的是，经济学中的大多数实证工作主要依赖于样本内证据。许多预测和预测准确性的样本外测试是非正式的和定性的，部分原因是应用研究人员发现迄今为止开发的程序不足。在样品中工作良好的程序在样品外似乎也不能很好地工作。该项目的一个贡献来自于开发了新的程序，用于对样本外预测和预测误差进行计量经济学推断，这些预测和预测误差一般足以在广泛的应用中使用，最终包括全球环境变化的社会经济影响模型。早期的一项研究发现，在样本外预测中，具有良好样本内拟合的复杂汇率模型并不比随机游走模型更好。本项目使用这些程序来研究汇率波动的竞争模型的预测能力，并开发替代模型，包括将汇率与利率联系起来的模型。本项目的最后一个研究领域是实证库存模型。在研究的一部分，渐近理论和模拟方法将用于评估替代方法估计库存模型。问题包括对归一化的敏感性，有效GMM估计器的有限样本行为，以及在选择工具时所谓的“外生性”和“相关性”之间的权衡。其结果不仅适用于库存的估计，也适用于其他时间序列模型的估计。在第二部分，库存模型将使用第一部分的结果，根据美国、英国、加拿大和日本的两位数数据进行估算。要研究的问题包括成本函数的凸性，对冲击的动态响应，以及特殊冲击与总体冲击的重要性。

9309927 West Data-mining or over-fitting, perhaps as a result of repeated profession-wide use of a limited body of data, leads at times to spuriously good in sample fits. Some economists have argued that out of sample predictive accuracy is the critical measure of model quality. Unfortunately most empirical work in economics relies mainly on in-sample evidence. Many of the out of sample tests of predictions and predictive accuracy are informal and qualitative in part because applied researchers find procedures developed to date inadequate. Procedures that have worked well in sample also do not appear to work well out of sample. One contribution of this project comes from developing new procedures for econometric inference about out of sample predictions and prediction errors general enough to be useful in a wide range of applications, including eventually models of the socioeconomic impacts of global environmental change. An earlier study found that complicated models of exchange rates with good in-sample fit do no better than a random walk model in out of sample predictions. This project uses these procedures to study the predictive ability of competing models of exchange rate volatility and develops alternative models, including one that links exchange rates to interest rates. The final area of research for this project is on empirical inventory models. In one part of the research, asymptotic theory and simulation methods will be used to evaluate alternative methods for estimating inventory models. Issues include sensitivity to normalization, finite sample behavior of efficient GMM estimators, and the tradeoff between what has been called "exogeneity" and "relevance" in choice of instruments. The results will be relevant in estimation of not just inventory but other time series models as well. In a second part, inventory models will be estimated on two digit U.S., U.K., Canadian and Japanese data, using the results of the first part. Questi ons to be studied include convexity of cost functions, dynamic responses to shocks, and the importance of idiosyncratic versus aggregate shocks.