SCA 7 时间数列预测及模型设定的专家系统
Effective Forecasting and Time Series Analysis Using the SCA Expert System

• 自动寻找最佳化多变量Transfer Function模型
• 自动侦测并调整离异数据（EXTENDED UTS产品功能）
• 借着降低参数做计值偏误，大幅改善Intervention Analysis的能力

SCA 软件操作容易，适合各样背景的模型使用者。内建的人工智能演算功能，可自动有效地办认出时间数据内错综复杂季节性或非季节性的ARIMA模型，对时间数列进行分析或预测。具备SCA软件，您可克服时间数列分析理论及实务的操作隔阂，减少花在辨认ARIMA模型上的工夫，而将宝贵的时间用在时间数列结果的分析及预测上。

SCA 软件有效地解决了时间数据分析多种可能模型选择上的困扰，并且对时间数据中常见的多个外部干预及离异数据，自动作侦测及修正并估计出修正后的预测模型。此法大幅提高了模型解释能力及预测精确，并能提供干预或离异数据的类型、影响程度及持续长短。让您充分了解过去正常时间数据的型态，及未来正常，或干预下可能发生的预测状况。SCA软件非常适合商务、管理、营销、财务、投资、股票、交通、旅游、经济、生产、医疗、环保、仓储、公用事业及教学、研究…等实务及学术上的应用。

SCA 美国总公司并可就 （１）专家统计咨询，（２）高级预测机制，及（３）现有数据库系统整合，提供全方位的解决或咨询方案。

SCA Software Products


Scientific Computing Associates provides an integrated suite of software products for applications in forecasting, time series analysis, intervention/impact analysis, econometric modeling, data mining, database marketing, design of experiments, quality and productivity improvement, general statistical analysis, decision support, and operations research.

SCA's suite of available software products are modularly designed and available across a wide range of computing platforms that encompass more than twenty-five operating systems on PCs, workstations, and mainframes.

SCA software may be integrated into corporate-specific applications such as demand forecasting, inventory control, marketing analysis, and other decision support systems. It may also be used as a standalone statistical application package, providing users with a convenient programmable command language and convenent graphical user interface.

The SCA Statistical System was developed under the advisory direction of distinguished individuals such as George Box, George Tiao, Lon-Mu Liu, and Mervin Muller. Many of the state-of-the art capabilities of the SCA System are made possible through the contributions and counsel of a number of leading researchers in the fields of statistics, econometrics, decision analysis, engineering, and finance.

Forecasting

The SCA Statistical System is a methodologically advanced software system that provides comprehensive capabilities for business and industrial forecasting applications.

It includes a sophisticated expert system modeling environment for integrated applications that require a high degree of automation. It also includes a complete set of identification and diagnostic tools to facilitate user-directed modeling.

SCA's expert system addresses single series univariate models as well as multivariate models. Using a sophisticated and highly robust expert system approach, the SCA System derives parsimonious models that facilitate reliable and accurate forecasting.

The SCA System also provides advanced outlier detection and adjustment capabilities that allows for the joint estimation of outlier effects and model parameters. Using this advanced feature of the SCA System, the model parameters are automatically desensitized from structural changes in the data (e.g., abnormal pulses, temporary changes, and level shifts).

In addition, SCA's outlier handling capabilities are extended into forecasting. Through this state-of-the art procedure, forecasts are desensitized from the effect of outliers. If outliers are not handled appropriately, especially near the forecasting origin, the impact of anomolous data can be severly detrimental to forecasting accuracy and reliability.

It is recognized that the forecasting process is more than the calculation of forecast values. In a era of change and uncertainty, knowledge of system structure and the interplay of variables is important. Quantitative forecasting methods, such as those provided by SCA, are invaluable in providing such knowledge.

To facilitate quantitative forecasting, SCA provides a myriad of forecasting and modeling methods to address any business application where forecasting accuracy is of paramount importance. A representative sample of the methods addressed by SCA are categorized below.

Univariate methods:

• Box-Jenkins ARIMA models
• Intervention/impact adjustment models
• General exponential smoothing methods
• Simple methods (e.g., moving average)

Multiple-input methods:

• Multiple-input transfer function models
• General regression methods
• Traditional econometric models (2SLS, 3SLS, etc.)
• Multivariate Adaptive Regression Splines

Multivariate methods:

• Simultaneous transfer function models
• Vector autoregressive models
• Vector ARMA models
• State Space models/Kalman Filtering