The R Package deseats for Data-Driven Trend and Seasonality Estimation in Time Series

The R package deseats is introduced that allows for the application of a datadriven locally-weighted-regression algorithm for estimating the trend and the seasonality in univariate, equidistant time series with short-memory errors. A corresponding data-driven semiparametric model with autoregressive movingaverage errors and point as well as interval forecasting approaches are established and the main functions of the deseats package are described. deseats is applied under consideration of real-world time series and the seasonal component identification is compared to that of STL, TRAMO-SEATS and X13-ARIMA. Finally, the quality of the bandwidth selection algorithm and the consistency of the automated component estimators are highlighted through a simulation study, while the quality of seasonality estimation is compared to that of well-established and widely-used methods, including aforementioned methods and others, in a second simulation study. The new algorithm captures the error-autocorrelation well and often produces seasonality estimates with mean squared error smaller than or at least similar to other methods, if the number of observations is sufficiently large. This finding also holds in simulated scenarios, where the underlying model assumption of the determinism of trend and seasonal components is violated.