FORECASTING

Methods and Applications

Third Edition

Spyros Makridakis
European Institute of Business
Administration (INSEAD)

Steven C Wheelwright
Harvard University, Graduate
School of Business Administration

Rob J. Hyndman
Monash University, Department of
Mathematics and Statistics
Technische University Darmstadt
Fachbereich 1
Betriebswirtschaftliche Bibiiothek
Inventar-Nr.
Abstell-Nr.:

John Wiley & Sons, Inc.

 CONTENTS
2/5 Prediction intervals 52
1 / THE FORECASTING
2/6 Least SQuares estimates 54
PERSPECTIVE I
2/6/1 Discovering and describing
1/1 Why forecast? 2 relationships 59
1/2 An overview of forecasting 2/7 Transformations and
techniques 6 adjustments 63
1/2/1 Explanatory versus time series 2/7/1 Mathematical transformations 63
forecasting 10 2/7/2 Calendar adjustments 67
1/2/2 Qualitative forecasting 12 2/7/3 Adjustments for inflation and
population changes 70
1/3 The basic steps in a forecasting
task 13 Appendices 71

References and selected 2-A Notation for Quantitative

forecasting 71
bibliography 17
2-B Summation sign X 72
Exercises 19
References and selected
bibliography 74

2 / BASIC FORECASTING Exercises 76

TOOLS 20
3/ TIME SERIES
2/1 Time series and cross-sectional
data 21 DECOMPOSITION 81
.•
3/1 Principles of decomposition 84
2/2 Graphical summaries 23
2/2/1 Time plots and time series 3/1/1 Decomposition models 84
patterns 24 3/1/2 Decomposition graphics 87
2/2/2 Seasonal plots 26 3/1/3 Seasonal adjustment 88
2/2/3 Scatterplots 27 3/2 Moving averages 89
2/3 Numerical summaries 28 3/2/1 Simple moving averages 89
2/3/1 Univariate statistics 29 3/2/2 Centered moving averages 94
2/3/2 Bivariate statistics 34 3/2/3 Double moving averages 98
2/3/3 Autocorrelation 38 3/2/4 Weighted moving averages 98

2/4 Measuring forecast accuracy 41 3/3 Local regression smoothing 101

2/4/1 Standard statistical measures 42 3/3/1 Loess 104
2/4/2 Out-of-sample accuracy 3/4 Classical decomposition 106
measurement 45
3/4/1 Additive decomposition 107
2/4/3 Comparing forecast methods 46
3/4/2 Multiplicative decomposition 109
2/4/4 Theirs U-statistic 48
2/4/5 ACF of forecast error 50

IX
 3/4/3 Variations on classical 4/5/1 Initialization 174
decomposition 112 4/S/2 Optimization 176
4/5/3 Prediction intervals 177
3/5 Census Bureau methods 113
3/5/1 First Iteration 114 References and selected
3/5/2 Later iterations 118 bibliography 179
3/5/3 Extensions to X-I2-AR1MA 119
Exercises 181
3/6 STL decomposition 121
3/6/1 Inner loop 122
3/6/2 Outer loop 123 5/ SIMPLE REGRESSION 185
3/6/3 Choosing the STL parameters 124
3/6/4 Comparing STL with
5/1 Regression methods 186
X-12-ARIMA 124 5/2 Simple regression 187
3/7 Forecasting and 5/2/1 Least souares estimation 188
decomposition 125 5/2/2 The correlation coefficient 193
S/2/3 Cautions In using correlation 196
References and selected 5/2/4 Simple regression and the
bibliography 127 correlation coefficient 198
5/2/5 Residuals, outliers, and
Exercises 130 Influential observations 203
5/2/6 Correlation and causation 208

4 / EXPONENTIAL SMOOTHING 5/3 Inference and forecasting with

simple regression 208
METHODS 135 5/3/1 Regression as statistical
4/1 The forecasting scenario 138 modeling 209
5/3/2 The F-test for overall
4/2 Averaging methods 141 significance 211
4/2/1 The mean 141 5/3/3 Confidence Intervals for individual
4/2/2 Moving averages 142 coefficients 215'
5/3/4 Mests for Individual
4/3 Exponential smoothing coefficients 217
methods 147 5/3/5 Forecasting using the simple
4/3/1 Single exponential regression model 218
smoothing 147 5/4 Non-linear relationships 221
4/3/2 Single exponential smoothing:
an adaptive approach 155 5/4/1 Non-linearity In the
4/3/3 Hot's linear method 158 parameters 222
3 4/3/4 Holt-Winters' trend and 5/4/2 Using logarithms to form linear
seasonally method 161 models 224
•A 4/3/5 Exponential smoothing: Pegels' 5/4/3 Local regression 224
classification 169 Appendixes 228
H
4/4 A comparison of methods 171 5-A Determining the values of o
and b 228
4/5 General aspects of smoothing
- methods 174
S
3
References and selected 6/6 Econometric models 299
bibliography 230 6/6/1 The basis of econometric
modeling 299
Exercises 231
6/6/2 The advantages and drawbacks
of econometric methods 301

6/ MULTIPLE REGRESSION 240 Appendixes 303

6-A The Durbln-Watson statistic 303
6/1 Introduction to multiple linear
regression 241 References and selected
bibliography 305
6/1/1 Multiple regression model:
theory and practice 248 Exercises 306
6//I/2 Solving for the regression
coefficients 250
6/1/3 Multiple regression and the
7/ THE BOX-JENKINS
coefficient of determination 251
6/1/4 The F-test for overall METHODOLOGY FOR
significance 252
6/1/5 Individual coefficients: confidence ARIMA MODELS 311
intervals and f-tests 255
6/1/6 The assumptions behind multiple
linear regression models 259
7/1 Examining correlations in times
series data 313
6/2 Regression with time series 263 7/1/1 The autocorrelation function 313
6/2/1 Checking independence of 7/1/2 A white noise model 317
residuals 265 7/1/3 The sampling distribution of
6/2/2 Time-related explanatory autocorrelations 317
variables 269 7/1/4 Portmanteau tests 318
7/1/5 The partial autocorrelation
6/3 Selecting variables 274-* coefficient 320
6/3/1 The long list 276 7/1/6 Recognizing-seasonality in a
6/3/2 The short list 277 time series 322
6/3/3 Best subsets regression 279 7/1/7 Example: Pigs slaughtered 322
6/3/4 Stepwise regression 285
7/2 Examining stationarity of time
6/4 Multicollinearity 287 series data 324
6/4/1 Multicollinearity when there are 7/2/1 Removing non-statlonarlty in a
two regressors 289 time series 326
6/4/2 Multicollinearity when there are 7/2/2 A random walk model 329
more than two regressors 289 7/2/3 Tests for statationarlty 329
7/2/4 Seasonal differencing 331
6/5 Multiple regression and
7/2/5 Backshlft notion 334
forecasting 291
6/5/1 Example: cross-sectional 7/3 ARIMA models for times series
regression and forecasting 292 data 335
6/5/2 Example: time series regression 7/3/1 An autoregressive model of
and forecasting 294 order one 337
6/5/3 Recapitulation 298

xi
 7/3/2 A moving average model of order
one 339 8 / ADVANCED FORECASTING
7/3/3 Higher-order autoregressive MODELS 388
models 339
7/3/4 Higher-order moving average 8/1 Regression with ARIMA
models 342 errors 390
7/3/5 Mixtures: ARIMA models 344 8/1/1 Modeling procedure 391
7/3/6 Mixtures: ARIMA models 345 8/1/2 Example: Japanese motor vehicle
7/3/7 Seasonally and ARIMA production 393
models 346 8/1/3 Example: Sales of petroleum and
7/4 Identification 347 coal products 396
8/1/4 Forecasting 400
7/4/1 Example I: A non-seasonal time
series 349 8/2 Dynamic regression models 403
7/4/2 Example 2: A seasonal time 8/2/1 Lagged explanatory variables 403
series 352 8/2/2 Kqyck model 406
7/4/3 Example 3: A seasonal time series 8/2/3 The basic forms of the dynamic
needing transformation 354 regression model 407
7/4/4 Recapitulation 357 8/2/4 Selecting the model order 409
7/5 Estimating the parameters 358 8/2/5 Forecasting 413
8/2/6 Example: Housing starts 415
7/6 Identification revisited 360
8/3 Intervention analysis 418
7/6/1 Example 1: Internet usage 362
7/6/2 Example 2: Sales of printing/ 8/3/1 Step-based Interventions 419
writing paper 362 8/3/2 Pulse-based Interventions 421
8/3/3 Further reading 422
7/7 Diagnostic checking 364 8/3/4 Intervention models and
forecasting 423
7/& Forecasting with ARIMA ..
models 366 8/4 Multivariate autoregressive
7/8/1 Point forecasts 366 models 423
7/8/2 Out-of-sample forecasting 370 8/5 State space models 429
7/8/3 The effect of differencing on
forecasts 371 8/5/1 Some forecasting models in state
7/8/4 ARIMA models used in time space form 429
series decomposition 372 8/5/2 State space forecasting 431
7/8/5 Eoulvalances with exponential 8/5/3 The value of state space
3; smoothing models 373 models 433

N References and selected 8/6 Non-linear models 433

bibliography 374 8/7 Neural network forecasting 435
Exercises 377
s

Xli
References and selected 10/1/1 The accuracy of forecasts in
bibliography 440 financial and other markets 484
10/1/2 Non-Investment type
Exercises 444 forecasts 490

10/2 The nature of judgmental biases

9/ FORECASTING THE LONG- and limitations 492
10/2/1 ludgmental biases in
TERM 451 forecasting 493
10/2/2 Dealing with judgmental
9/1 Cycles versus long-term trends:
biases 496
forecasting copper prices 452 10/2/3 Conventional wisdom 502
9/1/1 Forecasting IBM's sales 457
10/3 Combining statistical and
9/2 Long-term mega economic judgmental forecasts 503
trends 459 10/3/1 Arriving at final forecasts during
9/2/1 Cycles of various durations and a budget meeting 503
depths 461
9/2/2 Implications of extrapolating 10/4 Conclusion 508
long-term trends 464
References and selected
9/3 Analogies 466 bibliography 509
9/3/1 The Information versus the Exercises 512
Industrial Revolution 467
9/3/2 Five major inventions of the
Industrial Revolution and their
analogs 469 11 / THE USE OF FORECASTING
9/4 Scenario building 472 METHODS IN PRACTICE
9/4/1 Businesses: gaining and/o*r 514
maintaining competitive
advantages 472 I I/I Surveys among forecasting
9/4/2 Jobs, work, and leisure time 475 users 515
9/4/3 Physical versus tele-Interactions: I I/I/I Familiarity and satisfaction with
extent and speed of major forecasting methods 516
acceptance 476 11/1/2 The use of different forecasting
methods 520
References and selected
bibliography 478 11/2 Post-sample accuracy: empirical
findings 525
Exercises 480
11/3 Factors influencing method
selection 532
10 / JUDGMENTAL FORECASTING
11/4 The combination of forecasts 537
AND ADJUSTMENTS 482 11/4/1 Factors that contribute to making
10/1 The accuracy of judgmental combining work 538
11/4/2 An example of combining 539
forecasts 483

xill
References and selected 1/2 Statistics packages 578
bibliography 543 1/3 Specialty forecasting packages 579
1/4 Selecting a forecasting
Exercises 547 package 582
2 Forecasting associations 583
12/IMPLEMENTING 3 Forecasting conferences 585
FORECASTING: ITS USES, 4 Forecasting journals and
ADVANTAGES, AND newsletters 585

LIMITATIONS 549 5 Forecasting on the Internet 586

12/1 What can and cannot be References and selected

predicted 551 bibliography 588
12/1/1 Short-term predictions 553
12/1/2 Medium-term predictions 554
12/1/3 Long-term predictions 557
APPENDIX II / GLOSSARY OF
12/2 Organizational aspects of
FORECASTING TERMS 589
forecasting 558
12/2/1 Correcting an organization's APPENDIX III / STATISTICAL
forecasting problems 561
12/2/2 Types of forecasting problems TABLES 549
and their solutions 562
A: Normal probabilities 620
12/3 Extrapolative predictions versus B: Critical values for t-statistic 621
creative insights 567
C: Critical values for F-statistic 622
12/3/1 Hindsight versus foresight -S69
D: Inverse normal table 628
12/4 Forecasting in the future 571
E: Critical values for x2 statistic 629
12/4/1 Data, Information, and
forecasts 571 F: Values of the Durbin-Watson
' 12/4/2 Collective knowledge, experience, statistic 630
and forecasting 572
G: Normally distributed observations 632
References and selected
bibliography 575
AUTHOR INDEX 633
Exercises 576
SUBIECT INDEX 636
APPENDIX 1 / FORECASTING
RESOURCES 577
I Forecasting software 578
1/1 Spreadsheets 578

xlv