An add-in to dynamically create a validation column for time series data. Developed by Ian Cox (Ian@JMP) and Julian Parris (julian).
** Note: Model validation is not built into the time series platforms in JMP or JMP Pro. However, this can be done manually:
Partition a time series using this add-in, and then hide and exclude the holdout partition (select the rows, and then select Rows > Hide and Exclude).
Fit the time series model(s) using the Time Series platform (from the Analyze > Modeling menu).
For the chosen model, save the prediction formula using the red triangle for the model. This produces a new data table with predicted values and residuals for the training data.
To calculate residuals for the validation data, copy and paste the holdout values from the original table into the appropriate cells in this new table. Then, create a new column, and use the Formula Editor to calculate the residuals (actual - predicted). Note that forecasts for the validation data are automatically computed when the prediction formula is saved to the data table.
Summarize the residuals for the training and validation data using the Distribution platform (to produce validation statistics), and graph the actual and forecasted time series and the residuals using the Graph Builder.
To forecast future values using both the training and validation data, return to the original data, unhide and unexclude the holdout partition, and refit the model using the Time Series platform. Specify the number of forecast periods (from the red triangle), and save the prediction formula for this model to the data table. Repeat steps 4 and 5 above.
These steps apply to models built using the JMP Time Series platform. Regression-based forecasting models can be built using the Analyze > Fit Model platform. Built-in model validation can be used for these models. Here, we do not hide and exclude the validation data - we create a validation column, and use this column in the validation