Statistical methods for analyzing economic data need to be timely, accurate and easy to compute. To accomplish this, parametric models are often assumed, but they are at best approximate, and often lack a good fit in ...Statistical methods for analyzing economic data need to be timely, accurate and easy to compute. To accomplish this, parametric models are often assumed, but they are at best approximate, and often lack a good fit in the tails of the distribution where much of the interesting data are concentrated. Therefore, nonparametric methods have been extensively examined as alternatives to the constrictive assumptions of parametric models. This paper examines the use of Sequential Normal Scores (SNS) for transforming time series data with unknown distributions into time series data that are approximately standard-normally distributed. Particular attention is directed toward detecting outliers (out-of-control values), and applying subsequent analytic methods such as CUSUMs and Exponentially Weighted Moving Average (EWMA) schemes. Two examples of stock market data are presented for illustration.展开更多
文摘Statistical methods for analyzing economic data need to be timely, accurate and easy to compute. To accomplish this, parametric models are often assumed, but they are at best approximate, and often lack a good fit in the tails of the distribution where much of the interesting data are concentrated. Therefore, nonparametric methods have been extensively examined as alternatives to the constrictive assumptions of parametric models. This paper examines the use of Sequential Normal Scores (SNS) for transforming time series data with unknown distributions into time series data that are approximately standard-normally distributed. Particular attention is directed toward detecting outliers (out-of-control values), and applying subsequent analytic methods such as CUSUMs and Exponentially Weighted Moving Average (EWMA) schemes. Two examples of stock market data are presented for illustration.
