Business Knowledge Series course

Presented by Terry Woodfield, Ph.D., Analytical Training Consultant, Education Division, SAS

This class teaches how to model, interpret, and predict time series data using UCMs. The UCM procedure analyzes and forecasts equally spaced univariate time series data using the Unobserved Components Models (UCM).

Learn how to

• Analyze time series data using a novel class of models called the Unobserved Component Models (UCM). The UCMs decompose the response series into components such as trend, seasonals, cycles, and the regression effects due to predictor series.
• Use the UCM procedure to find a suitable model for the series of interest, to obtain extensive model diagnostics, and to generate series forecasts and the forecasts of the constituent components.
• Get detailed understanding of the series dynamics by analyzing the plots of the estimated components.

Who should attend

Those who want to analyze time series data to uncover patterns such as trend, seasonal effects, and cycles using the latest techniques

Formats available	Standard duration
Classroom:	1 day

Before attending this course, you should:

• Have experience with regression modeling.
• Be familiar with at least one time series modeling technique, such as Box and Jenkins or exponential smoothing.
• Have some familiarity with SAS software.

This course addresses SAS/ETS software.

Introduction to Unobserved Components Models

• Overview of different models used for time series analysis.
• Some examples to motivate the decomposition of time series into components.

Stochastic Models for Trend, Season, and Cycle Components

• Random walk and local linear trend.
• Stochastic cycle.
• Saturated and unsaturated trigonometric season.

Handling of the Regression Effects

• Linear time-invariant regression.
• Nonlinear time-invariant regression.
• Time-varying regression coefficient model.
• Lagged response values.

Irregular Component

• White noise.
• ARMA noise.

UCM Procedure Syntax

• Component specification.
• Specification of different types of regression effects.
• Controlling observation span.
• Requesting variety of tabular and graphical output.

Estimation Phase Output

• Parameter estimates.
• Fit summary.
• Residual diagnostics.
• Outlier summary.

Forecast Phase Output

• Forecasts of the response variable and the model components.
• Interpolation of missing response values and smoothed estimates of the model components.

Examples

• Several illustrative examples.
• Modeling tips.

Statistical Framework Underlying the UC Models

• UCMs as state space models.
• One-step-ahead forecasts, likelihood, and state smoothing.

Numerical Issues

• How to determine the problem size.
• Trouble shooting parameter estimation and other problems.