The IIF Workshop on Feature-based Time Series Forecasting and Applications

Recently, three related feature-based algorithms have been developed with great success. The FFORMS (feature-based forecast model selection) framework by Talagala, Hyndman & Athanasopoulos (2018) uses a random forest to select the best forecasting method based on time series features. Remarkably, the related FFORMA (feature-based forecast model averaging) method proposed by Montero-Manso et al. (2019) won second position in the recent M4 competition. Third, Talagala, Li & Kang (2019) proposed the FFORMPP (feature-based forecast model performance prediction) framework that uses an efficient Bayesian multivariate surface regression approach to model forecast error as a function of features calculated from the time series.

Feature-based time series forecasting has shown great potential, while some key issues remain to be tackled including automatic feature extraction and selection, feature-based density forecasting, forecasting related time series with features, time series generation based on features, and minimum forecast model pools for forecast model averaging. The proposed workshop aims to bring people together with diverse expertise and facilitate new research collaborations.

The Chinese Academy of Sciences, Beihang University and Central University of Finance and Economics will jointly host the 2023 International Symposium on Forecasting. We believe this workshop will be a useful forerunner for the Chinese forecasting community, and also a good opportunity to bring together international forecasters and Chinese forecasters to form a potential collaborating network.