Local and Global Trend Bayesian Exponential Smoothing Models

• DOI: 10.48550/arxiv.2309.13950
• Publication Date: 2023-01-01
• Authors: Slawek Smyl; Christoph Bergmeir; Alexander Dokumentov; Erwin Wibowo; Daniel F. Schmidt
• Keywords: Exponential Smoothing; Leverage (statistics); Univariate; Additive model

This paper describes a family of seasonal and non-seasonal time series models that can be viewed as generalisations additive multiplicative exponential smoothing models, to model grow faster than linear but slower exponential. Their development is motivated by fast-growing, volatile series. In particular, our have global trend smoothly change from multiplicative, combined with local trend. Seasonality when used in the error always heteroscedastic through parameter sigma. We leverage state-of-the-art Bayesian fitting techniques accurately fit these are more complex flexible standard models. When applied M3 competition data set, outperform best algorithms well other benchmarks, thus achieving knowledge results per-series univariate methods on this dataset literature. An open-source software package method available.