Forecasting is Snow Joke

  • 24 May 2022
  • 3 replies
  • 82 views

Userlevel 3
Badge +1

❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️

Research Collection Thread

Thread of the coolest papers in forecasting ⛄️🌨️️️️🌨️️️️

Add any hot forecasting research you come across! 🔥🔥🔥

Please add a TLDR, a link to the paper, and ideally any relevant code base.

❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️❄️

 


3 replies

Userlevel 3
Badge +1

Distributed ARIMA Models for Ultra-long Time Series (2022)

 

TLDR

Distributed modelling of long time series processes based on minimising a global loss function using MapReduce framework.

Challenges the assumption that the data generating process (DGP) of an ultra-long time series stays invariant, instead assumes that only the DGP of subseries spanning shorter time periods stays invariant.

Links

Link to paper: https://arxiv.org/abs/2007.09577

Link to github repository: https://github.com/xqnwang/darima

Userlevel 3
Badge +1

Feature-based intermittent demand forecast combinations: bias, accuracy and inventory implications (2022)

 

TLDR

Combination approach to modelling & forecasting intermittent demand. Significant improvement in forecasting accuracy versus many of the common methods used for forecasting intermittent demand such as Cronstons and MAPA https://kourentzes.com/forecasting/2016/11/17/mapax-available-for-r-new-mapa-package-on-cran/

 

The approach in this work uses an XGBoost model to compute features tailored for intermittent demand and learn the relationship between the features and combination weights of the pool of intermittent demand forecasting models.

 

It builds upon the M4metalearning work using features of the time series, https://github.com/robjhyndman/M4metalearning 

 

The code base also has a nice collection of functions for multiple time series models and intermittent time series features, even if you weren’t to use them using the proposed. combination approach.

Links

Link to paper: https://arxiv.org/abs/2204.08283

Link to github repository: https://github.com/lily940703/fide

Userlevel 3
Badge +1

Forecasting: theory and practice (2022)

 

TLDR

A forecasting encyclopedia

Should be a go to reference for all things forecasting, more a book than a paper, this work provides reviews of theory and practice of forecasting models, methods, principles, and approaches to prepare, produce, organise, and evaluate forecasts.

An extensive review of the application of these methods within different fields is then provided.

 

Also a really useful list of all the existing packages across R, python & others for applying these methods

Links

Link to paper: https://www.sciencedirect.com/science/article/pii/S0169207021001758

Link to free arxiv version: https://arxiv.org/abs/2012.03854

Reply