Quantitative Aids:Forecasting

Introduction:

A forecast is necessary to every decision or policy position. Essentially, the policy or decision one advocates assumes a forecasts of what will happen as a result of the action being recommended. Consequently forecasting as an activity permeates all aspects of public management and public policy making.

Forecasts can be generates in many different ways using many different approaches. Some forecasts are purely based on intuition and human judgement, others require complex mathematical and computer based models. Some forecasts rely heavily on social and physical science theory while others do not.

Though approaches and techniques of forecasting come from many different disciplines including economics, mathematics, engineering, psychology and statistics, it has only been over the last 15 years that forecasting has become an identifiable and serious area for study. Most forecasting research up through 1979 was primarily aimed as developing ‘new’ techniques and demonstrating, often on the basis of one or two examples, their superiority to existing forecasting tools. In 1979 the first major evaluation of multiple forecasting methods over many (100) cases was published, launching a new interest in forecasting with a shifted emphasis on evaluation of forecasting methods. This new orientation has attempted to identify which contingencies or situations favor use of which forecasting method.

In the mid 1980s an additional orientation emerged in forecasting research. This approach focused on forecasting as human activity and began to study forecasting as a human and social activity. Thus a social science orientation to study forecasting, particularly in government emerged in the research literature.

Objectives:

Through a focus on forecasting this course aims to:

• have students learn how to appropriately use advanced statistical tools such as multiple regression modeling,
• have students learn the basic approaches to quantitative and qualitative forecasting at a level which will permit them to generate ‘professional’ quality forecasts,
• have students learn the basics of evaluation at a level that will permit them to design evaluation studies and carry them out, and
• have students learn how forecasting processes operate in real public organizations with a particular focus on the politics of the forecasting process

Text Materials:

• Armstrong, J. Scott,Long-Range Forecasting; From Crystal Ball to Computer, Second Edition, New York: John Wiley and Son, 1985. (Reprint)
• Hanke, John, and Reitsch, Arthur, Business Forecasting, Fifth Edition, Boston: Allyn and Bacon, 1995.
• Lecture Notes
• Bretschneider, Stuart, Forecasting: Some New Realities Occasional Paper No. 99, Metropolitan Studies Program, The Maxwell School, Syracuse, New York, December, 1985.
• Forecast Bibliography

Grading:

Evaluation of student performance in this class will be based on 4 projects (see section on projects for details). Each project will be evaluated by the instructor and given a grade. Any unacceptable evaluation will include comments and guidelines on how to revise the project to obtain an acceptable grade. Students may then resubmitup to two projects during the semester within a specific time frame once in order to receive an acceptable grade.

Students receiving acceptable evaluations on either their initial or revised version of all four projects will receive an A for the course. Students who decide not to resubmit any unacceptable project or who fail to submit or resubmit work by specified deadlines will receive an unacceptable grade on that project. One unacceptable grade and three acceptable grades will result in a B+ for the course, two unacceptable and two acceptable grades will result in a B-, three unacceptable grades will result in a C. Homework problems will also be assigned. Student who complete homework and show mastery of the material will be eligible for an increase in their final grade. For example, a student with one unacceptable project and significant efforts in completing homework, will be eligible for an A-. Students who decide not to complete any homework will not be penalized in any way with regard to grading. Similarly, though, they will not be eligible for a final grade enhancement.

Using Computers:

In order to effectively learn the theoretical concepts, many of the homework problems and student projects will require extensive interaction with the computer. The computer is a tool that facilitates the statistical and analytic work. It is therefore important that students develop sufficient skills to make use of statistical packages.

Students are free to work with any software package they wish, though class examples and handouts will make use of Excell. The SAS system is also available on the Maxwell Clusters for those who already know how to use it.

Projects:

Each project will be assigned separately with text description of requirements, deliverables and due dates. In general there will be one every two to three weeks. Two projects, the forecasting expert presentation and the individual data analysis and forecasting project will be assigned earlier in order to provide students enough time to adequately complete them. Below are short descriptions of each of the four projects.

An Evaluation Design

Students will be presented a decision problem which will require evaluation of several alternative approaches. The project will require a design for collecting data that when analyzed will lead to a decision.

Time Series Forecasting Tournament

Students will be organized into teams and given 3 to 5 time series. The teams will then apply as many univariate time series forecasting approaches to them as they deem necessary and generate one set of forecasts for each time series. Results will be compared with actual data from the forecast period to determine which student team did the best job on each series.

Forecasting Expert Presentation

Each student will select a topic from the list below. Other topics can be selected provided they are cleared with the instructor. With guidance from the instructor, students will read three to four articles on the selected topic. The student will then prepare a 10 pages review paper which will also be the basis for a 5 to 10 minute oral presentation to the class.

Individual Data Analysis and Forecasting Project

Student will identify a time series they wish to forecast. The data will be collected and a copy provided to the instructor by spring break. Students are then free to analyze and forecast these data using any univariate technique they desire plus any one multivariate approach (usually multiple regression analysis). The final analysis should be presented in a 10 page paper.

Syllabus

1. Introduction to Forecasting (Armstrong 1, 2, 3, 4, 5; Hanke 1; Bretschneider)
2. 1. The importance of forecasting
   2. Types of forecasting uses
   3. Typologies of forecasting methods
   4. What we know and how we know it
   5. General laws of forecasting
3. Forecasting Process and Evaluation (Armstrong 11, 12, 13; Hanke 4 pp.119-122)
4. 1. Forecasting process
   2. Public vs. private sector
   3. The role of politics
   4. Evaluation designs
   5. 1. Criteria
      2. Data collection
      3. Experimental designs
      4. Quasi-experimental designs
   6. Evaluation of inputs
   7. Evaluation of outputs
   8. 1. Error measurement and accuracy
      2. Beyond accuracy
5. Forecasting Methods I - Judgmental (Armstrong 6; Hanke 11)
6. 1. Types of judgement
   2. Errors in judgement
   3. Implementation issues
   4. 1. Selecting Judges
      2. Posing Questions
   5. Methods
   6. 1. Surveys
      2. Delphi
      3. Meetings
      4. Interviewing
   7. Assessment of uncertainty
7. Review of Statistics (Hanke 2, Freeman et al.)
8. Forecasting Methods II - Univariate Time Series Approaches (Armstrong 7; Hanke 4, 5, 8, 10)
9. 1. Simple smoothing and moving averages
   2. Smoothing models with trend
   3. Smoothing models with seasonality
   4. Adaptive smoothing models
   5. Decomposition Methods
   6. Box-Jenkins ARIMA methods
   7. Interventions and special events
   8. 1. ARIMA Intervention models
      2. Special event analysis
10. Forecasting Methods III - Multivariate Methods (Armstrong 8; Hanke 6, 7, 9)
11. 1. Simple Regression
    2. Multiple Regression
    3. Topics in Regression
    4. 1. Heteroscedasticity
       2. Autocorrelation
    5. Transfer Function Models
    6. Other Multivariate Models
12. Forecasting Methods IV - Dividing and Combining (Armstrong 9, 10)
13. 1. Segmentation Combining methods
14. Forecasting Methods Summarized (Armstrong 14, 15)
15. A Behavioral View of Forecasting in the Public Sector