Autobox handles a single endogenous equation incorporating either pre-identified causal series or empirically identified dummy series, which are found to be statistically significant. The set of pre-identified series can be either stochastic or deterministic (dummy) in form. In its search for the most appropriate model form and the optimal set of parameters the final model can either be:

• Purely empirical
• A starting model could be used.

A final model may require one or more of the following structures:

1. Power transforms like Log, Square Root and Reciprocal etc.
2. Variance stabilization due to deterministic changes in the background error variance.
3. Data segmentation or splitting as evidenced by a statistically significant change in either model form or parameters.

Enroute to its tour de force Autobox will evaluate numerous possible
models/parameters that have been suggested by the data itself. In practice, a realistic limit is set on the maximum number of model form iterations. The exact specifics of each tentative model is not pre-set thus the power of Autobox emerges. The kind and form of the tentative models may never before have been tried. Each dataset speaks for itself and suggests the iterative process.

The Final Model could be as simple as:

1. A simple trend model or a simple ordinary least squares model.
2. An exponential smoothing model.
3. A simple weighted average where the weights are either equal or unequal.
4. A Cochrane-Orcutt or ordinary least squares with a first order fixup.
5. A simple ordinary least squares model in differences containing some needed lags.
6. A spline-like set of local trends superimposed with an arbitrary ARIMA model and perhaps a pulse or two.

The number of possible final models that Autobox could find is infinite and only discoverable via a true expert system like Autobox. A final model may require one or more of the following

Seasonal Structures:

1. Seasonal ARIMA structure where the prediction depends on some previous
   reading S periods ago.
2. Seasonal structure via a complete set of seasonal dummies reflecting a fixed response based upon the particular period.
3. Seasonal structure via a partial set of seasonal dummies reflecting a fixed response based upon the particular period.