Title: TM 745 Forecasting for Business
1TM 745 Forecasting for Business TechnologyDr.
Frank Joseph Matejcik
2nd Session 2/04/08 Chapter 2 The Forecast
Process, Data Considerations, and Model Selection
- South Dakota School of Mines and Technology,
Rapid City
2Agenda New Assignment
- Chapter 2 problems 4, 8, 9
- Tentative Schedule
- About the mailing
- Chapter 2 WK (with odd diversions)
- Try to use ForecastX for Autocorrelation
- Business Forecasting 5th Edition J. Holton
Wilson Barry KeatingMcGraw-Hill
3Tentative Schedule
Chapters Assigned 28-Jan 1 problems
1,4,8 e-mail, contact 4-Feb 2 problems 4,
8, 9 11-Feb 3 problems 1,5,8,11 18-Feb
Presidents Day 25-Feb 4 problems 6,10 3-Mar
5 problems 5,8 10-Mar Exam 1 Ch 1-4
Revised 17-Mar Break 24-Mar Easter 31-Mar
6 problems 4, 7
Chapters Assigned 7-Apr 7 3,4,5(series
A) 7B 21-Apr 8 Problem 6 28-Apr
9 05-May Final
4Web Resources
- Class Web site on the HPCnet system
- http//sdmines.sdsmt.edu/sdsmt/directory/courses/2
008sp/tm745M021 - Streaming video http//its.sdsmt.edu/Distance/
- Answers will be online. Linked from
- The same class session that is on the DVD is on
the stream in lower quality. http//www.flashget.c
om/ will allow you to capture the stream more
readily and review the lecture, anywhere you can
get your computer to run.
5Introduction
- Outline the forecasting process
- Guide to establishment of a successful
forecasting system - Selection of applicable techniques
- Evaluation for trend, cyclical, seasonal
components - Gap, Burnett, GDP, car sales, housing starts
- Statistics review, correlogram
6The Forecast Process
- Managers that use forecasts should know how they
were developed. - Forecasters should know the needs of decision
makers. - Good communication is paramount
7The Forecast Process
- 1. Specify ObjectivesClear Statement of
Objectives - Discussion between "forecasters" and users
- Good communication important at this phase
- 2. Determine what to forecastSales in units or
dollars? - by region? by product line? export?
- Hospital admissions, discharges, patient-days,
acuity days
8The Forecast Process
- 3. Identify the time dimensionslength how far to
forecast? major construction or staffing - periodicity hourly(power), daily, monthly,
annually, etc. - Urgency exam or next year
9The Forecast Process
- 4. Data considerations Internal or External
- Internal data periods available form(units or
dollars) - 5. Model Selection
- a. the pattern exhibited by the data
- b. the quantity of historic data available
- c. the length of the forecast horizon
10(No Transcript)
11The Forecast Process
- 6. Model Evaluation
- Use RMSE and other measures
- Accuracy how it works in the forecast horizon
- Fit how it works in retrospectively
- WE can holdout data to check accuracy
- 7. Forecast Preparation
- Naturally, use the technique
- Use more than one with different databases
12The Forecast Process
- 8. Forecast Presentation
- Presentation is for the users
- Graphics do help
- 9. Tracking Results
- Models do deteriorate need replacement
- Forecasters can learn from mistakes
13Trend, Seasonal, and Cyclical Data Patterns
- Trend long term change in the level of the data.
Stationary no trend (KW) - Seasonal think seasons like summer at schools,
resorts. Weekends, weekdays.Seasonally
adjusted, deseasonalized - Cyclical business cycles, inflation,
recession - Irregular random think wars
14(No Transcript)
15(No Transcript)
16(No Transcript)
17Data Patterns and Model Selection
- Lets apply Table 2-1 to Figures 2-1 and through
2-3 - POP has trend, no cycle, no seasonality
- Holts ES method
- Linear regression trend
18Data Patterns and Model Selection
- THS has trend seasonality and a cycle
- Winters ES
- Linear regression with seasonal adjustment
- Causal regression
- Time-series decomposition
- Cycle may make last two better,again
19Data Patterns and Model Selection
- DPI has a nonlinear trend, no seasonality and no
cycle - Nonlinear regression
- Causal regression
- Holts ES
- When you finish the text you willbe able to make
good forecastsfor a wide variety of patterns.
20A Statistical Review
- Read it as fast as you should. Quickly for many.
- Confidence level is usually used for
intervalsSignificance level for tests. - We will go through the book review slides
quickly, if there is interest we will slow it
down. - Well do a little on the correlation coefficient,
my weird way.
21(No Transcript)
22(No Transcript)
23(No Transcript)
24(No Transcript)
25(No Transcript)
26(No Transcript)
27(No Transcript)
28(No Transcript)
29(No Transcript)
30(No Transcript)
31(No Transcript)
32(No Transcript)
33(No Transcript)
34(No Transcript)
35(No Transcript)
36(No Transcript)
37(No Transcript)
38Katyas Triangle and the correlation coefficient r
39(No Transcript)
40Anscombe Appropriate
Look at horizontal / long side in the central
triangle (Katyas) to get r. r has the same
sign as the scatter slope.
41Anscombe Appropriate
Look at horizontal / long side in the central
triangle (Katyas) to get r. r has the same
sign as the scatter slope.
42Anscombe Outlier
Look at horizontal / long side in The central
triangle (Katyas) to get r. r has the same
sign as the scatter slope.
43Anscombe Outlier
Look at horizontal / long side in The central
triangle (Katyas) to get r. r has the same
sign as the scatter slope.
44Anscombe Unbalanced
Look at horizontal / long side in The central
triangle (Katyas) to get r. r has the same
sign as the scatter slope.
45Anscombe Unbalanced
46Anscombe Unbalanced
47Anscombe Parabola
48Anscombe Parabola
49R-.62
50R.41
51R -.95
52Correlograms An Alternative Method of Data
Exploration
53Disposable Personal Income Exploratory Data
Analysis Model Selection pics after
- Monthly data, so more variable than Figure 2.6.
54(No Transcript)
55(No Transcript)
56Total Houses Sold Exploratory Data Analysis
Model Selection pics after
- Monthly data seasonality, trend
- Winters exponential smoothing
- Regression with trend and seasonality
- Causal regression
- Time series decomposition
- ARIMA
57(No Transcript)
58(No Transcript)
59Integrative Case The Gap
- Should we talk through it with scans or WK slides?
60Using ForecastX to Find Autocorrelation Functions
- Lets see if we can get it to work