Statistical Methods

for Forecasting
BOVAS ABRAHAM
University of Waterloo
JOHANNES LEDOLTER
University of Iowa

John Wiley & Sons

New York • Chichester • Brisbane • Toronto • Singapore
 Contents

1 INTRODUCTION AND SUMMARY 1

1.1 Importance of Good Forecasts 1

1.2 Classification of Forecast Methods 2
1.3 Conceptual Framework of a Forecast System 3
1.4 Choice of a Particular Forecast Model 4
1.5 Forecast Criteria 5
1.6 Outline of the Book 5

2 THE REGRESSION MODEL AND ITS APPLICATION IN FORECASTING 8

2.1 The Regression Model 9

2.1.1 Linear and Nonlinear Models, 10
2.2 Prediction from Regression Models with Known
Coefficients 12
2.3 Least Squares Estimates of Unknown Coefficients 13
2.3.1 Some Examples, 13
2.3.2 Estimation in the General Linear
Regression Model, 16
2.4 Properties of Least Squares Estimators 20
2.5 Confidence Intervals and Hypothesis Testing 25
2.5.1 Confidence Intervals, 25
2.5.2 Hypothesis Tests for Individual Coefficients, 26
2.5.3 A Simultaneous Test for Regression Coefficients, 26
2.5.4 General Hypothesis Tests: The Extra Sum of
Squares Principle, 27
2.5.5 Partial and Sequential F Tests, 29
ix
2.6 Prediction from Regression Models with
Estimated Coefficients 30
2.6.1 Examples, 31
2.7 Examples 33
2.8 Model Selection Techniques 41
2.9 Multicollinearity 45
2.10 Indicator Variables 49
2.11 General Principles of Statistical Model Building 52
2.11.1 Model Specification, 53
2.11.2 Model Estimation, 54
2.11.3 Diagnostic Checking, 54
2.11.4 Lack of Fit Tests, 56
2.11.5 Nonconstant Variance and Variance-Stabilizing
Transformations, 58
2.12 Serial Correlation among the Errors 60
2.12.1 Serial Correlation in a Time Series, 61
2.12.2 Detection of Serial Correlation among the Errors
in the Regression Model, 63
2.12.3 Regression Models with Correlated Errors, 66

2.13 Weighted Least Squares 74

Appendix 2 Summary of Distribution Theory Results 77

REGRESSION AND EXPONENTIAL SMOOTHING METHODS TO

FORECAST NONSEASONAL TIME SERIES 79

3.1 Forecasting a Single Time Series 79

3.2 Constant Mean Model 81
3.2.1 Updating Forecasts, 82
3.2.2 Checking the Adequacy of the Model, 83
3.3 Locally Constant Mean Model and Simple
Exponential Smoothing 85
3.3.1 Updating Forecasts, 86
3.3.2 Actual Implementation of Simple
Exponential Smoothing, 87
3.3.3 Additional Comments and Example, 89
 XI

3.4 Regression Models with Time as Independent Variable 95

3.4.1 Examples, 96
3.4.2 Forecasts, 98
3.4.3 Updating Parameter Estimates and Forecasts, 100
3.5 Discounted Least Squares and General
Exponential Smoothing 101
3.5.1 Updating Parameter Estimates and Forecasts, 102
3.6 Locally Constant Linear Trend Model and Double
Exponential Smoothing 104
3.6.1 Updating Coefficient Estimates, 107
3.6.2 Another Interpretation of Double
Exponential Smoothing, 107
3.6.3 Actual Implementation of Double
Exponential Smoothing, 108
3.6.4 Examples, 110
3.7 Locally Quadratic Trend Model and Triple
Exponential Smoothing 120
3.7.1 Implementation of Triple Exponential Smoothing, 123
3.7.2 Extension to the General Polynomial Model and
Higher Order Exponential Smoothing, 124
3.8 Prediction Intervals for Future Values 125
3.8.1 Prediction Intervals for Sums of Future
Observations, 127
3.8.2 Examples, 127
3.8.3 Estimation of the Variance, 129
3.8.4 An Alternative Variance Estimate, 132
3.9 Further Comments 133

REGRESSION AND EXPONENTIAL SMOOTHING METHODS TO

FORECAST SEASONAL TIME SERIES 135

4.1 Seasonal Series 135

4.2 Globally Constant Seasonal Models 139
4.2.1 Modeling the Seasonality with Seasonal
Indicators, 140
4.2.2 Modeling the Seasonality with Trigonometrie
Functions, 149
xii CONTENTS

4.3 Locally Constant Seasonal Models 155

4.3.1 Locally Constant Seasonal Models Using
Seasonal Indicators, 158
4.3.2 Locally Constant Seasonal Models Using
Trigonometrie Functions, 164
4.4 Winters' Seasonal Forecast Procedures 167
4.4.1 Winters' Additive Seasonal Forecast Procedure, 167
4.4.2 Winters' Multiplicative Seasonal
Forecast Procedure, 170
4.5 Seasonal Adjustment 173
4.5.1 Regression Approach, 174
4.5.2 Smoothing Approach, 174
4.5.3 Seasonal Adjustment Procedures, 179
Appendix 4 Computer Programs for Seasonal Exponential
Smoothing 182
EXPSIND. General Exponential Smoothing with
Seasonal Indicators, 182
EXPHARM. General Exponential Smoothing
with Trigonometrie Forecast Functions, 185
WINTERS1. Winters' Additive Forecast
Procedure, 188
WINTERS2. Winters' Multiplicative Forecast
Procedure, 190

5 STOCHASTIC TIME SERIES MODELS 192

5.1 Stochastic Processes 192

5.1.1 Stationary Stochastic Processes, 194
5.2 Stochastic Difference Equation Models 197
5.2.1 Autoregressive Processes, 199
5.2.2 Partial Autocorrelations, 209
5.2.3 Moving Average Processes, 213
5.2.4 Autoregressive Moving Average
(ARMA) Processes, 219
5.3 Nonstationary Processes 225
5.3.1 Nonstationarity, Differencing, and
Transformations, 225
CONTENTS Xlll

5.3.2 Autoregressive Integrated Moving Average

(ARIMA) Models, 231
5.3.3 Regression and ARIMA Models, 237
5.4 Forecasting 238
5.4.1 Examples, 240
5.4.2 Prediction Limits, 246
5.4.3 Forecast Updating, 247
5.5 Model Specification 248
5.6 Model Estimation 250
5.6.1 Maximum Likeühood Estimates, 250
5.6.2 Unconditional Least Squares Estimates, 253
5.6.3 Conditional Least Squares Estimates, 257
5.6.4 Nonlinear Estimation, 258
5.7 Model Checking 261
5.7.1 An Improved Approximation of the
Standard Error, 262
5.7.2 Portmanteau Test, 263
5.8 Examples 263
5.8.1 Yield Data, 264
5.8.2 Growth Rates, 267
5.8.3 Demand for Repair Parts, 270
Appendix 5 Exact Likelihood Functions for Three
Special Models 273
I. Exact Likelihood Function for an ARMA(1,1)
Process, 273
IL Exact Likelihood Function for an AR(1) Process, 278
III. Exact Likelihood Function for an MA(1) Process, 279

6 SEASONAL AUTOREGRESSIVE INTEGRATED MOVING AVERAGE MODELS 281

6.1 Multiplicative Seasonal Models 283

6.2 Autocorrelation and Partial Autocorrelation Functions
of Multipücative Seasonal Models 285
6.2.1 Autocorrelation Function, 286
6.2.2 Partial Autocorrelation Function, 291
6.3 Nonmultiplicative Models 291
 CONTENTS

6.4 Model Building 293

6.4.1 Model Specification, 293
6.4.2 Model Estimation, 299
6.4.3 Diagnostic Checking, 299
6.5 Regression and Seasonal ARIMA Models 299
6.6 Forecasting 302
6.7 Examples 306
6.7.1 Electricity Usage, 306
6.7.2 Gas Usage, 308
6.7.3 Housing Starts, 310
6.7.4 Car Sales, 310
6.7.5 Demand for Repair Parts, 313
6.8 Seasonal Adjustment Using Seasonal ARIMA Models 317
6.8.1 X-11-ARIMA, 317
6.8.2 Signal Extraction or Model-Based Seasonal
Adjustment Methods, 318
Appendix 6 Autocorrelations of the Multiplicative
(0, d,l)(l, Al)i2 Model 319

RELATIONSHIPS BETWEEN FORECASTS FROM GENERAL

EXPONENTIAL SMOOTHING AND FORECASTS FROM ARIMA
TIME SERIES MODELS 322

7.1 Preliminaries 322

7.1.1 General Exponential Smoothing, 322
7.1.2 ARIMA Time Series Models, 323
7.2 Relationships and Equivalence Results 327
7.2.1 Illustrative Examples, 329
7.3 Interpretation of the Results 330
Appendix 7 Proof of the Equivalence Theorem 331

SPECIAL TOPICS 336

8.1 Transfer Function Analysis 336

8.1.1 Construction of Transfer Function-Noise
Models, 338
CONTENTS XV

8.1.2 Forecasting, 344

8.1.3 Related Models, 348
8.1.4 Example, 348
8.2 Intervention Analysis and Outliers 355
8.2.1 Intervention Analysis, 355
8.2.2 Outliers, 356
8.3 The State Space Forecasting Approach, Kaiman
Filtering, and Related Topics 359
8.3.1 Recursive Estimation and Kaiman Filtering, 361
8.3.2 Bayesian Forecasting, 363
8.3.3 Models with Time-Varying Coefficients, 364
8.4 Adaptive Filtering 368
8.5 Forecast Evaluation, Comparison, and Control 370
8.5.1 Forecast Evaluation, 372
8.5.2 Forecast Comparison, 373
8.5.3 Forecast Control, 374
8.5.4 Adaptive Exponential Smoothing, 377

REFERENCES 379

EXERCISES 386

DATA APPENDIX 418

TABLE APPENDIX 426

AUTHOR INDEX 435

SUBJECT INDEX 437