Abstract

We compare four methods for the decomposition and seasonal adjustment of daily time series: TBATS (De Livera et al., 2011), STR (Dokumentov and Hyndman, 2015), DSA (Ollech, 2021) and JDemetra+ (Smyk et al., 2023). They cover a broad range of methods, ranging from explicit time series models to semi-parametric, algorithmic approaches. The comparison is made by means of a Monte Carlo experiment based on a standard structural time series model parameterized in different ways. The stochastic setting is later expanded to include two alternative exogenous variables: a deterministic end of month effect and an exogenous factor that mimics the COVID-19 shock. In this way, we can assess the relative performance of the four methods under controlled conditions as well as their sensitivity to deviations from the standard model.