Transformations for Stabilizing Spread In a Time Series Model: a Comparative Study

Abstract

Using a Monte–Carlo simulation, we compare the efficiency of various estimators for estimating the correct power transformation to stabilize spread in a time series model. First, we divide the data into chronological time blocks. One method uses the variance stabilizing transformations based on means and variances from the blocks, and the other method is exploratory, relying on their medians and inter-quartile ranges from the blocks. Modified methods which permit the blocks to move are investigated and are shown to be generally more efficient. The effects of error contamination are also included in this study.

Keywords:

• Contaminated distribution
• homoscedasticity
• interquartile range
• median
• simulation study