Statistical tests provide a mechanism for making quantitative decisions about models. Tests determine whether there is enough evidence to reject a conjecture or hypothesis about a model. A conjecture is referred to as a null hypothesis. These tests are used to determine the presence of serial correlation, periodicities, or heteroscedastic variance relative to the series under test. These tests are run against residual series produced by any function that produces an ARTFITRESIDUALS layer. See TD_EXTRACT_RESULTS.
Functions:
- TD_BREUSCH_GODFREY
- Analyzes the residuals under consideration using a Breusch–Godfrey test. This is a test for the presence of serial correlation.
- TD_BREUSCH_PAGAN_GODFREY
- Determines if variance associated with a residual series is homoscedastic or heteroscedastic.
- TD_CUMUL_PERIODOGRAM
- Analyzes the residual series under consideration to determine if periodicities (cyclic patterns) are present within the series.
- TD_DICKEY_FULLER
- Tests for the presence of one or more unit roots in the series under consideration. Series can be either a residual series or any series.
- TD_DURBIN_WATSON
- Analyzes a residual series using a Durbin-Watson test. This is a test for the presence of serial correlation.
- TD_FITMETRICS
- Uses the original series, the predicted series, the original series mean, and the residuals from a modeling fit exercise to determine goodness-of-fit associated with a modeling fit exercise. Produces the same output as that found in an ARTFITMETADATA layer.
- TD_GOLDFELD_QUANDT
- Runs on a residual series to determine if the series has heteroscedastic or homoscedastic variance. The test divides the residual series sample points into two collections of sample points to determine if the variance of the series is homoscedastic or heteroscedastic.
- TD_PORTMAN
- Determines if the residual series under consideration is white noise. White noise is a series which has zero mean, no evidence of serial correlation, and homoscedastic variance.
- TD_SELECTION_CRITERIA
- Calculates metrics against the residuals left from a model validation exercise. The produced metrics are used by the data scientist to determine which forecast model candidates to use for forecasting.
- TD_SIGNIF_PERIODICITIES
- Analyzes the residual series under consideration to determine if periodicities (cyclic patterns) are present in the series.
- TD_SIGNIF_RESIDMEAN
- Analyzes and determines if the residual series under consideration is white noise with a zero mean. White noise is a series which has zero mean, no evidence of serial correlation, and homoscedastic variance.
- TD_WHITES_GENERAL
- Determines if the residual series under test has heteroscedastic or homoscedastic variance.