You are here: Predictive Analytics > Time Series > ETS

ETS Tool

The ETS tool estimates a univariate time series forecasting model using an exponential smoothing method. Exponential smoothing is a commonly used forecasting approach that is based on a weighted average of past observations, with the weights declining in size for more distant past values (the weights are said to follow an exponential decay function). The tool is able to account for three time series components: level, trend, and seasonality. The tool can use fully automated methods to model the three components in the "best way" based on statistical criteria, or the user can specify the underlying methods used. An excellent discussion of the methods used can be found in Chapter 7 of Hyndman and Athanasopoulos's online book Forecasting: Principals and Practice*

This tool uses the R tool. Install R and the necessary packages by going to Options > Download Predictive Tools.

Input

An Alteryx data stream.

Configuration Properties

Required parameters

Model name: Each model needs to be given a name so it can later be identified. Model names must start with a letter and may contain letters, numbers, and the special characters period (".") and underscore ("_"). No other special characters are allowed, and R is case sensitive.
Select the target field: Select the field from the data stream you want to forecast. Measurements for this field need to be made at regular time intervals (e.g., daily, monthly, quarterly, etc.).
Target field frequency: Choose the time interval for the observations of the target field.

Model type

Error type: This option controls how the effect of the nearest prior periods is modeled. Choices are Auto (the default), Additive, and Multiplicative. If Auto is used, both additive and multiplicative specification are estimated and a statistical information criteria is used to select between models. The estimated parameter that gives the relative weight between more recent and more distant past values in the output is alpha.
Trend type: This controls how the effect of trend is modeled. The choices are Auto (the default), Additive, Multiplicative, and None. If Auto is used, both additive and multiplicative specifications, along with models with no correction from trend, are considered, and a statistical information criteria is used to select between models. The estimated parameter that gives the relative weight between more recent and more distant trend values in the output is beta.
Trend dampening: This option controls the extent to which the effect of recent trend effects are reduced (dampened). The choices are Auto (the default), Yes, and No. The Auto option considers models both with and without dampening, and selects the best one based on a statistical information criteria. The estimated parameter phi in the output (only for cases where trend dampening is included best model) indicates the extent to which the forecast trend has been dampened.
Seasonal type: This controls how seasonal effects are modeled. The choices are Auto (the default), Additive, Multiplicative and None.

Other options

Information criteria for model selection: The criteria used to compare different models and select the best model. The choices provided are Auto (the default), the Akaike information criterion (AIC), the corrected Akaike information criterion (AICc) or the Bayesian information criteria (BIC). If the Auto option is selected then the AICc is used if there are 48 or fewer observations of the target, otherwise the AIC is used.
Use a Box-Cox transformation: If this option is selected, the user can provide a value of lambda (falling between 0 and 1) for doing a Box-Cox transformation of the target field. In this option is selected, multiplicative specifications of the three time series components are not considered.
Series starting period (optional): This option allows the user to specify the starting period of the time series, which will be reflected in the decomposition and forecast plots.
The number of periods to include in the forecast plot: This option results in a plot that contains the original data and a number of forecast future points (along with 80% and 95% confidence intervals around the forecast points). The user can specify the number of periods that should be forecast into the future for the plot.
Select Week Format: This allows the user to choose a method to specify work weeks. These options relate to what constitutes the first week of the year, and what day of the week a week begins on.

US â Sunday is the first day of the week
UK â Monday is the first day of the week
ISO8601 â Monday is the first day of the week

Graphics Options

Plot size: Select inches or centimeters for the size of the graph.
Graph resolution: Select the resolution of the graph in dots per inch: 1x (96 dpi); 2x (192 dpi); or 3x (288 dpi). Lower resolution creates a smaller file and is best for viewing on a monitor. Higher resolution creates a larger file with better print quality.
Base font size (points): Select the size of the font in the graph.

Outputs

O: Consists of an output stream containing the ETS model object that can be used for both point forecasts and a user specified percentile confidence interval surrounding those forecasts.
R: Consists of the report snippets generated by the ETS tool: a statistical summary, autocorrelation diagnostic plots and forecast plots.
I: An interactive HTML dashboard consisting of plots and metrics. You can interact with the visualizations by clicking on the different graphical elements to reveal more information, values, metrics and analytics.

*Hyndman, R.J. and Athanasopoulos, G. (2012) Forecasting: Principles and Practice.