TM 745 Forecasting for Business

1
TM 745 Forecasting for Business TechnologyDr.
Frank Joseph Matejcik
3rd Session 5/31/04 Chapter 3 Moving Averages
and Exponential Smoothing Chapter 4 Introduction
to Forecasting with Regression Methods

• South Dakota School of Mines and Technology,
  Rapid City

2
Agenda New Assignment

• ch3(1,5,8,11) ch4(6,10) Tentative Schedule
• Chapter 3, 4 WK (with odd diversions)
• Try to use ForecastX for Autocorrelation
• Business Forecasting 5th Edition J. Holton
  Wilson Barry KeatingMcGraw-Hill

3
Tentative Schedule
Chapters Assigned 17-May 1 e-mail,
contact problems 1,4, 8 24-May 2
problems 4, 8, 9 31-May 3,4 problems
ch3(1,5,8,11) ch4(6,10) 07-June 5 problems
5,8 14-June Test 6 start 21-June 6 finish,
7 28-June 8 05-July Final 9
Attendance Policy Help me work with you.
4
Web Resources

• Class Web site on the HPCnet system
• http//sdmines.sdsmt.edu/sdsmt/directory/courses/2
  007su/tm745001
• 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.

5
Moving Averages Exponential Smoothing

• All basic methods based on smoothing
• 1. Moving averages
• 2. Simple exponential smoothing
• 3. Holt's exponential smoothing
• 4. Winters' exponential smoothing
• 5. Adaptive-response-rate single exponential
  smoothing

6
Moving Averages

• Ex. Three Quarter Moving Average(1999Q11999Q2
  1999Q3)/3 Forecast for 1999Q4
• Slutsky-Yule effect Any moving average could
  appear to be acycle, because it is a serially
  correlated set of random numbers.

7
(No Transcript)
8
(No Transcript)
9
Simple Exponential Smoothing
10
Simple Exponential Smoothing

• Alternative interpretation

11
Simple Exponential Smoothing

• Why they call it exponential property

12
(No Transcript)
13
Simple Exponential Smoothing

• Advantages
• Simpler than other forms
• Requires limited data
• Disdvantages
• Lags behind actual data
• No trend or seasonality

14
Holt's Exponential Smoothing(Double Holt in
ForecastXTM)
15
ForecastXTM Conventions forSmoothing Constants

• Alpha (a) the simple smoothing constant
• Gamma (g) the trend smoothing constant
• Beta (b) the seasonality smoothing constant

16
(No Transcript)
17
(No Transcript)
18
Holt's Exponential Smoothing

• ForecastX will pick the smoothing constants to
  minimize RMSE
• Some trend, but no seasonality
• Call it linear trend smoothing

19
Winters'
20
(No Transcript)
21
Adaptive-Response-Rate Single Exponential
Smoothing
22
(No Transcript)
23
(No Transcript)
24
Adaptive-Response-Rate Single Exponential
Smoothing

• Adaptive is a clue to how it works
• No direct way of handling seasonality
• Does not handle trends
• ForecastX has different algorithm

25
Using Single, Holt's, or ADRES Smoothing to
Forecast a Seasonal Data Series

• 1. Calculate seasonal indices for the series.
  Done in HOLT WINTERS ForecastX.
• 2. Deseasonalize the original data by dividing
  each value by its corresponding seasonal index.

26
Using Single, Holt's, or ADRES Smoothing to
Forecast a Seasonal Data Series

• 3. Apply a forecasting method (such as ES,
  Holt's, or ADRES) to the deseasonalized series to
  produce an intermediate forecast of the
  deseasonalized data.
• 4. Reseasonalize the series by multiplying each
  deseasonalized forecast by its corresponding
  seasonal index.

27
(No Transcript)
28
New-Product Forecasting(growth curve fitting)
29
Gompertz Curve
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
Logistic Curve
34
(No Transcript)
35
(No Transcript)
36
Bass Model (See Chapter 1,too)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
Event Modeling

• Event Indices Legend
• 0. No event present
• Free-standing inserts (FSIs)
• FSI/radio, television, print campaign
• Load (trade promotion)
• Deload (month after effect of load)
• Thematics (themed adg campaign)
• Instant redeemable coupon (IRC)

42
(No Transcript)
43
(No Transcript)
44
Forecasting Jewelry Sales using Exponential
Smoothing
45
Forecasting Jewelry Sales using Exponential
Smoothing
46
Forecasting Houses Sold Sales using Exponential
Smoothing
47
Forecasting Houses Sold Sales using Exponential
Smoothing
48
Summary

• All basic methods based on smoothing
• 1. Moving averages
• 2. Simple exponential smoothing
• 3. Holt's exponential smoothing
• 4. Winters' exponential smoothing
• 5. Adaptive-response-rate single exponential
  smoothing
• Use of Deseasonalized Series
• techniques not clear winners

49
Integrative Case The Gap
50
Solutions toCase Questions 1
51
Solutions toCase Questions 3
52
Case Questions Solutions to Case Questions

• Skipped the details of this one in lecture, but
  worth a read.
• Holts beats Winters but not by much Lets
  try it live.

Using ForecastX to Make Exponential Smoothing
Forecasts
53
Introduction to Forecasting with Regression
Methods

• Fundamentals
• Jewelry
• Disposable Income
• Gap

54
The Bivariate Regression Model
55
The Bivariate Regression Model
56
Visualizationof DataImportant in
Regression
57
(No Transcript)
58
A Process for Regression Forecasting

• Inspect data (graphically) trends, seasonals,
  cycles, and outliers
• Make forecasts for all the Xs (predictors,
  independent variables)
• estimate coefficients (use a holdout)
• compare various models

59
Forecasting with a Simple Linear Trend
60
Forecasting with a Simple Linear Trend
61
Forecasting with a Simple Linear Trend
62
Forecasting with a Simple Linear Trend
63
Forecasting with a Simple Linear Trend
64
Using a Causal Regression Model to Forecast

• Not using a trend line
• Yf(X) where X is an appropriate explanatory
  variable
• Use knowledgeable people library for Xs
• Logical construct (Jevons sunspottheory of
  business cycles)
• Try to forecast Jewelry sales

65
A Jewelry Sales Based on Disposable Personal
Income
66
(No Transcript)
67
(No Transcript)
68
A Jewelry Sales Based on Disposable Personal
Income
69
A Jewelry Sales Based on Disposable Personal
Income
70
Jewelry Sales Based on fig 4.7 Disposable
Personal Income
71
Jewelry Sales Based on fig 4.7 Disposable
Personal Income
72
Jewelry Sales Based on fig 4.7 Disposable
Personal Income
73
Statistical Evaluation of Regression Models tab
4.5
74
Statistical Evaluation of Regression Models tab
4.5
75
Statistical Evaluation of Regression Models tab
4.5
76
Statistical Evaluation of Regression Models

• 1. Check to see if the sign of the slope makes
  sense
• 2. Check the significance of the slope using a
  t-test.
• 3. How much of the variation is explained by the
  regression using R2

77
Using the Standard Error of the Estimate
78
Serial Correlation
79
Serial Correlation
80
Serial Correlation Fixes

• 1. First differencing the data
• 2. Use multiple regression extra variables
• 3. Use the square of the existing causal variable
  as another variable
• 4. Advanced models includingserial correlation.

81
Heteroscedasticity
82
Heteroscedasticity
83
Heteroscedasticity Fixes

• Transformation
• logarithm
• square root
• others
• Non least squares regression

84
Cross-Sectional Forecasting

• One time period
• Another explanatory variable
• Similar to causal methods, but data is separated.
• The population of cities wasthe predictor
  variable.

85
Forecasting Total Houses Sold Sales w/ Bivariate
Regression
86
Forecasting Total Houses Sold Sales w/ Bivariate
Regression
87
Integrative Case The Gap
88
Solutions toCase Questions 1
89
Solutions toCase Questions 2
90
Solutions toCase Questions 2
91
Solutions toCase Questions 2
92
Solutions toCase Questions 2
93
Using ForecastX to Make Regression Forecasts

• Try it?
• Try it?
• Next week more on regression
• Maybe, that weird plot will beexplained

Further Comments on Regression Models