Title: Quantile regression based seasonal adjustment
Authors: Mohammed Elseidi - University of Padova (Italy) [presenting]
Massimiliano Caporin - University of Padova (Italy)
Abstract: Different types of time series are affected by both deterministic and stochastic seasonal patterns. The usual assumption is that there is a unique seasonal pattern that affects the location and/or the scale of the variable of interest. However, there are cases where we observe, in a given time series, different seasonal patterns affecting the mean and the variance. Furthermore, seasonal patterns might affect higher order moments. Using traditional approaches for seasonal adjustment might not be efficient and does not ensure the adjusted data are free from periodic behaviors in higher order moments. We introduce a seasonal adjustment method based on quantile regression approach that is capable of capturing different forms of seasonal patterns. By describing the seasonal behavior over an approximation of the entire conditional distribution of a variable of interest, we might remove seasonal patterns affecting the mean and/or the variance only, or remove seasonal patterns varying over the distribution of the variable of interest. The results, focusing both on real and simulated data show the flexibility of the approach and the significant improvement over the traditional methods when the periodic behaviors impact on the distribution and/or on higher order moments.