AUTOBOX
AUTOBOX was developed in 1975 and was originally distributed to both mainframe users and time-sharing network systems such as IDC, CSC and Compuserv. The new generation of AUTOBOX emphasizes the inferential aspects of time series and provides a computational aid to modeling. Modeling is the process of comprehensive statistical analysis often referred to as time series analysis. Rather than dumb-down the approach to modeling and analyzing time series data , AUTOBOX has taken the high road in making sophisticated approaches easy to use. Autobox marries the concepts of time series (Box-Jenkins) , multiple regression and outlier detection and creates a symphony.;AUTOBOX was developed in 1975 and was originally distributed to both mainframe users and time-sharing network systems such as IDC, CSC and Compuserv. The new generation of AUTOBOX emphasizes the inferential aspects of time series and provides a computational aid to modeling. Modeling is the process of comprehensive statistical analysis often referred to as time series analysis. Rather than dumb-down the approach to modeling and analyzing time series data , AUTOBOX has taken the high road in making sophisticated approaches easy to use. Autobox marries the concepts of time series (Box-Jenkins) , multiple regression and outlier detection and creates a symphony.;AUTOBOX was developed in 1975 and was originally distributed to both mainframe users and time-sharing network systems such as IDC, CSC and Compuserv. The new generation of AUTOBOX emphasizes the inferential aspects of time series and provides a computational aid to modeling. Modeling is the process of comprehensive statistical analysis often referred to as time series analysis. Rather than dumb-down the approach to modeling and analyzing time series data , AUTOBOX has taken the high road in making sophisticated approaches easy to use. Autobox marries the concepts of time series (Box-Jenkins) , multiple regression and outlier detection and creates a symphony.
Supported Technologies
AIX,
HP/UX,
Linux,
Windows 95/98/ME,
Windows XP/2000/NT
Software
Click on a technology to view similar products within this category.
Pricing
- Unspecified -
$395 to $5,000
sales@autobox.com
215-675-0652
Resources
Additional Product Information
AUTOBOX is limited to 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 paramaters the program can either be:
1. Purely empirical or
2. 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, 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 been tried. Each dataset speaks for itself and suggests the iterative process. The Final Model could be as simple as:
1. A Holt-Winters model either constant, trend or quadratic with or without additive or multiplicative seasonality.
2. A simple trend model or a simple ordinary least squares model.
3. An exponential smoothing model.
4. A simple weighted average where the weights are either equal or unequal.
5. A Cochrane-Orcutt or ordinary least squares with a first order fixup.
6. A simple ordinary least squares model in differences containing some needed lags.
7. 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.
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.
The Final Model will satisfy both:
1. Necessity tests that guarantee the estimated coefficient is statistically significant.
2. Sufficiency tests that guarantee that the error process is:
-unpredictable on itself.
-not predictable from the set of causals.
-has a constant mean of zero.
The Final model will contain one or more of the following structures:
1. CAUSAL with correct lead/lag specification.
2. MEMORY with correct "autoregressive memory".
3. DUMMY with correct pulses, level shifts or spline time trends.