TD_POWERTRANSFORM takes a time series or spatial series and applies a power transform equation (log, invert, and so on) to the series to produce a transformed result series. The single input source contains univariate or multivariate series instances. A typical use cases for this function are following situations:
- Heteroscedasticity: Variance of the residuals of a regression model is not constant over time. Use TD_POWERTRANSFORM to stabilize the variance of the data and make it more homoscedastic, which can improve the performance of regression models.
- Nonlinear trends: Time series has trends, such as exponential or quadratic growth or decay. Use TD_POWERTRANSFORM to linearize the trend and improve the accuracy of the forecasting models.
- Skewed distributions: Time series data exhibits skewed distributions that result in biased forecasts. Use TD_POWERTRANSFORM to make the distribution more symmetric and reduce skewness.
You then use the transformed result to build an ARIMA forecasting model.
The following procedure is an example of how to process and then forecast heteroscedastic time series using TD_POWERTRANSFORM:
- Apply TD_POWERTRANSFORM to the heteroscedastic time series to transform it into a homoscedastic time series.
- Use the resulting homoscedastic time series to build an ARIMA forecasting model.
- Use the model to produce the initial forecast of the homoscedastic time series.
- Use the backward transform on the initial forecast to extract the forecast values for the heteroscedastic time series.