Random Walks and Unit Roots; Distributed Lag Models随机游动与单位根的分布滞后模型；.docVIP

下载本文档

7
0
约8.9千字
约 8页
2018-06-27 发布于福建
举报
版权申诉

Random Walks and Unit Roots; Distributed Lag Models随机游动与单位根的分布滞后模型；.doc

1、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Random Walks and Unit Roots; Distributed Lag Models随机游动与单位根的分布滞后模型；

Random Walks and Unit Roots I. Evolutionary Economic Time Series Many economic time series are not stationary, trending upwards such as national income, or following approximate random walks, such as the price of gold. An exception is housing starts. From visual inspection, there is no apparent trend in mean or variance, and no apparent seasonal pattern ( in the deseasonalized series). Most of the variance is attributable to variations with the business cycle. II. Random Walks How do we distinguish between a first order autoregressive time series with parameter close to one, y(t) = b y(t-1) + WN(t), and a random walk, RW(t): RW(t) = RW(t-1) + WN(t) ? Note that the random walk is the infinite sum of past white noise shocks, RW(t) = [1- z]-1 WN(t) = WN(t) + WN(t-1) + WN(t-2) + ....., where the inverse of the difference operator, ?-1 = [1- z]-1 , is the summation operator. Distant shocks are weighed no less than the current shock and the random walk has infinite variance. For a finite sample beginning at time zero, at value RW(0), which could be zero, RW(t) = RW(0) + WN(1) + WN(2) + .... + WN(t), and the random walk will have a finite variance. Its estimated autocorrelation function will damp out slowly, comparable to an autoregressive process with parameter close to one, and its partial autocorrelation function will have a peak at lag one, with value near one. In the regression, y(t) = b y(t-1) + WN(t), the OLS estimator of b is biased downward if b is close to one. If b equals one, the difference between the OLS estimator and one vanishes quickly, shrinking like 1/T, and the OLS estimator of the unit root is called superconsistent. So bias is a problem in small samples. Dickey and Fuller have studied the bias and developed Dickey-Fuller tables for a statistic analogous to the t-statistic, but with a different distribution tabulated from simulations. If b is less than one, then the difference between b and one shrinks like 1/√T. Since we are interested in t