Skip to main content

ETS

This document describes the usage of error, trend, and seasonal (ETS) models for forecasting.

Description​

ETS uses state-space models to describe the error, trend, and seasonal components of a time series. TDgpt uses automated ETS modeling to select an appropriate component combination for the input time series and forecast future data.

ETS provides confidence intervals for forecast results.

Parameters​

ParameterDescriptionRequired?
periodThe number of data points included in each period. If not specified or set to 0, a non-seasonal model is used.No
modelThree-character ETS model. The characters specify the error, trend, and seasonal components. The default value is ZZZ, which automatically selects each component.No
dampedWhether to use a damped trend. Enter 1 to enable or 0 to disable it. If not specified, the algorithm determines the setting based on the model.No

In the model string, A specifies an additive component, M specifies a multiplicative component, N omits the component, and Z automatically selects it. For example, ANN specifies additive errors without trend or seasonality.

Input data must contain only finite numeric values. A non-seasonal forecast requires at least two input values. When period is greater than 0, the input data must contain at least two complete periods. Multiplicative components require positive input data.

Example​

In this example, forecasting is performed on the i32 column. Each 12 data points form a period. ETS automatically selects the error, trend, and seasonal components, and uses a damped trend.

FORECAST(i32, "algo=ets,period=12,model=ZZZ,damped=1")

The complete SQL statement is shown as follows:

SELECT _frowts, FORECAST(i32, "algo=ets,period=12,model=ZZZ,damped=1") from foo
{
"rows": rows,       // Rows returned
"period": period,   // Period of results
"algo": "ets",     // Algorithm
"mse": mse,         // Mean square error (MSE)
"res": res          // Forecast results and confidence intervals in column format
}

References​