Influence Function Method Applied to Energy Time Series Data

摘要

The influence function for bivariate correlation has been applied by Gnanadesikan (1977), and recently by Chernick (1979). We propose the use of the influence function matrix for the auto correlation function of a time series as a method for detecting outliers in energy time series data. The method has proven to be effective in detecting errors in data from the Energy Information Administration's monthly power plant report. The technique also appears to be applicable to the safeguards problem of detecting losses of nuclear materials. A summary statistic, the average square influence function, is proposed for a hypothesis test for outliers. Asymptotic distribution theory for this summary statistic is presented. (ERA citation 06:003295)