Theoretical and Practical Limits of Kolmogorov-Zurbenko Periodograms with DiRienzo-Zurbenko Algorithm Smoothing in the Spectral Analysis of Time Series Data
share
The Kolomogorov-Zurbenko periodogram with DiRienzo-Zurbenko algorithm smoothing is the state-of-the-art method for spectral analysis of time series data. Because this approach assumes that a sinusoidal model underlies time-series data and because its algorithms are adaptive in nature, it is superior to traditional use of autoregressive integral moving average (ARIMA) algorithms. This article begins with a presentation of its statistical derivation and development followed by instructions for accessing and utilizing this approach within the R statistical program platform. The discussion then turns to a presentation of its theoretical and practical limits with regard to sensitivity (i.e., ability to detect weak signals), accuracy (i.e., ability to correctly identify signal frequencies), resolution (i.e., ability to resolve signals with close frequencies), and robustness with respect to missing data (i.e., sensitivity and accuracy despite high levels of missingness). Next using a simulated time series in which two signals close in frequency are embedded in significant amounts of random noise, the predictive power of this approach is compared to the traditional ARIMA approach, with support also garnered for its being robust even in the face of significant levels of missing data. The article concludes with brief descriptions of studies across a range of scientific disciplines that have capitalized on the power of the Kolmogorov-Zurbenko periodogram with DiRienzo-Zurbenko algorithm smoothing.
READ FULL TEXT
