Automatic Time Series Forecasting: The forecast Package for R
By Rob J Hyndman, Yeasmin Khandakar. 2008
4,106 citations recorded by the source corpus.
Automatic forecasts of large numbers of univariate time series are often needed in business and other contexts. We describe two automatic forecasting algorithms that have been implemented in the forecast package for R. The first is based on innovations state space models that underly exponential smoothing methods. The second is a step-wise algorithm for forecasting with ARIMA models. The algorithms are applicable to both seasonal and non-seasonal data, and are compared and illustrated using four real time series. We also briefly describe some of the other functionality available in the forecast package.
Subfields: Management Science and Operations Research, Statistics and Probability, Time-Series & Forecasting
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Management Science and Operations Research
- Rob J Hyndman
- Fotios Petropoulos
- Konstantinos Nikolopoulos
- Christoph Bergmeir
- Vassilios Assimakopoulos
- Evangelos Spiliotis
- Spyros Makridakis
- George Athanasopoulos
In Statistics and Probability
- George Athanasopoulos
- Fotios Petropoulos
- Han Lin Shang
- Slawek Smyl
- Konstantinos Nikolopoulos
- Vassilios Assimakopoulos
- Rob J Hyndman
- Christoph Bergmeir
In Time-Series & Forecasting
- Fotios Petropoulos
- Rob J Hyndman
- Jan Gasthaus
- Bernie Wang
- Bryan Lim
- Christoph Bergmeir
- Tim Januschowski
- Konstantinos Nikolopoulos
Citations in the corpus
- DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks (2017, methodology): “ing baselines which represent the state-of-the-art on demand integer datasets to the best of our knowledge: 7 Croston: the Croston method developed for intermittent demand forecasting from R package [12] ETS: the ETS model [11] from R package with automatic model selection. Only additive models are ”
- Time-series forecasting with deep learning: a survey (2020, background): “Indeed, a rich body of literature exists for automated approaches to time-series forecasting—including automatic parametric model selection [18], and traditional machine learning methods such as kernel regression [19] and support vector regression [20].”
- N-BEATS: Neural basis expansion analysis for interpretable time series forecasting (2019, methodology): “The pinnacle of the statistical approach encapsulates ARIMA, auto-ARIMA and in general, the unified state-space modeling approach, that can be used to explain and analyze all of the approaches mentioned above (see Hyndman and Khandakar [14] for an overview).”
- Forecasting Across Time Series Databases using Long Short-Term Memory Networks on Groups of Similar Series (2017, methodology): “ion (see Section 3.6), and pass on the sum of trend and reminder to the next step of data preprocessing. In R, STL is implemented in the stl function from the forecast package (Hyndman et al., 2015a; Khandakar and Hyndman, 2008). To obtain deterministic seasonality, we set the s.window parameter to ”
- Optimal combination forecasts for hierarchical time series (2011, methodology): “For each series, we select an optimal ARIMA model using the automatic algorithm of Hyndman and Khandakar (2008). We then re-estimate the parameters of the model using a rolling window beginning with the model fitted using the first 24 observations (1998:Q1–2003:Q4).”
- Forecasting at Scale (2018, neutral): “We then produce a set of forecasts for this model along with a set of reasonable baselines (including completely automatic techniques, cf. [Hyndman and Khandakar, 2007]) across a variety of simulated forecast dates.”
- The Tourism Forecasting Competition (2011, methodology): “The classification of the exponential smoothing methods in Table 1 originated with Pegels (1969) and was expanded by Gardner (1985), Hyndman et al. (2002) and Taylor (2003). Each of the fifteen methods listed has a trend and a seasonal component.”
- ETSformer: Exponential Smoothing Transformers for Time-series Forecasting (2022, background): “Therefore, neighboring data points have similar values, and recent tokens should be given a higher weight 1 when measuring their similarity [13, 14].”
- Forecasting in social settings: The state of the art (2020, methodology): “arima() and ets() methods (Hyndman et al. 2002; Hyndman and Khandakar 2008) as implemented in Hyndman et al.”
- CoST: Contrastive Learning of Disentangled Seasonal-Trend Representations for Time Series Forecasting (2022, methodology): “A straightforward solution is to optimize this hyper-parameter by grid search on the training or validation loss (Hyndman & Khandakar, 2008), but such an approach is too computationally expensive.”
- Predictive Off-Policy Policy Evaluation for Nonstationary Decision Problems, with Applications to Digital Marketing (2017, methodology): “The time series prediction algorithm that we use in our experiments is the R forecast package for fitting ARIMA models, as described by Hyndman and Khandakar (2008).”
- Forecasting hierarchical and grouped time series through trace minimization (2015, methodology): “Using the training data, 1 to 12-step-ahead base forecasts were generated across the entire hierarchy from both ARIMA and ETS models using the default settings in the automated algorithms of Hyndman and Khandakar (2008) and implemented in the forecast for R (see Hyndman, 2015).”
- Host lifestyle affects human microbiota on daily timescales (2014, background): “1) ‘forecast’ package to fit standard time series models known as autoregressive integrated moving average (ARIMA) models [53].”
- Visualising forecasting algorithm performance using time series instance spaces (2017, neutral): “Hyndman, RJ and Y Khandakar (2008).”