What are the Time Series Tools?
The Time Series module allows you to model time based data and forecast future values with confidence. In this blog I will cover the ARIMA, ETS, and TS Forecasting tools and how they can be used to achieve your predictive modelling goals.
What does the Arima tool do?
ARIMA tool estimates a time series forecasting model, either as a univariate model or one with covariates (predictors). In this example we will stick to the univariate model for simplicity.
What does the ETS tool do?
Estimates a univariate time series forecasting model using an exponential smoothing method.
What is the difference between ETS and ARIMA?
ETS is good at recognizing repeated patterns and seasonality over time while ARIMA is better with volatile "noisy" data. ARIMA is also better with stationary data, one where the mean and variance stays roughly the same over time despite significant dips and jumps in the values.
What does the TS Forecast tool do?
The TS Forecast tool requires a data stream that is the output stream from either an ARIMA or ETS tool
In this example I will show how we can forecast the number of units sold in the 6 weeks of data following on from our final fiscal week in the data.
What do the 3 Output Anchors do?
- O: contains a data stream of the forecast values and confidence bounds, this is the output that is used as an input data stream for other TS tools
- R: consists of a report with the forecast values and confidence bounds, as well as a forecast plot that shows the historical values and forecast values with error bounds
- I: interactive HTML dashboard consisting of plots and metrics
The Outputs:
O:
R:
I:
The ETS Forecast provides a wider range of values:
This is likely because ETS tends to give more weight to recent changes and models the trend and seasonality directly, which can make it more sensitive to fluctuations or trends.