STAHL: Seasonal, Trend, and Holiday Decomposition with Loess

We introduce a novel seasonal adjustment methodology developed by Quantcube Technology, termed Seasonal, Trend, and Holiday Decomposition with Loess (STAHL). This method builds upon the STL procedure outlined by Cleveland et al. and is tailored for application to time series across varying frequencies. Initially, we detail the innovations introduced by STAHL, which include the spectral identification of multiple seasonal frequencies and the systematized preprocessing different seasonal series. This preprocessing encompasses resampling, real-time handling of missing values, and other necessary adjustments. Additionally, we illustrate how the holiday decomposition component effectively manages the variability of moving holiday events. In addition to this, we introduce the Barnacle procedure, to deal with before and after effects of holiday. Subsequently, we provide empirical demonstrations using high frequency data. Specifically, we utilize US Weekly initial claims data to exemplify real-time adjustments during the unprecedented COVID-19 period, with comparisons to the series adjusted by the US Department of Labor. Furthermore, daily Google search data for the term “Easter bunny” illustrate our method’s capability in handling moving holiday effects. We also perform a simulation study to compare the performance of STAHL with existing high frequency seasonal adjustment methodologies.