WeLeto[g Introduction to Econometrics Third Edition G.S. Maddala JOHN WILEY & SONS, LTD (Chichester « New York + Weisheim + Brisbane « Toronto « Singaporeep ©2001 by Kanes Mal ‘tidy ie i ft sesamiae 732,43 viet oper —— 519. C626 Tn Gu 18 79T ‘ema or ceder an ease seve eng}: Boole Wistau Home age on ptiewiey oak ‘This bok was peony pb by Paice Ie. Repzed 2001, ay 2002 Alt Rigs Reserved No pat oh polation may be repose ceed it iol se, ot ‘ened in any frm by any means, lec, mecaniel photog cet Seg ot nevis nee under he ems th Copyright, Deis and Pet Act 1585 or und he te of ‘sence aed by he Capp Licensing Agee) 90 Totenbam Court Ras London WIP SHE. {UK witout he eemisen wring of th Publier nd he cope owne, 5. Mats eed ih ner Cop Debra Pe At Kb eed the author ofthis work iene peed er ey Etro fics hn Wiley & Sonn, 603 Third Avene, New ek, MY 101580513, USA Wily-VCH Vecag Gat, epter 3, D684 Wenham, Germany ob Wey & Sons Aut Li 33 Fark Ro Mito, unin 406 Asta on Wiey& Seas (As) Pei, 2 Clement Lap 102-01, Sin Xing Disp, Singapore 125809 ein Wily & Sons (Cans) ti, 2 Wert Ros, Reaile,Ontaa MPW IL, Casas Pris LbearyCeaoguing in Pdteaton Date ‘Acalog ree fo i bok val om he Bish Litary son oan-ora82 “pee in 1/291 Tes by Laver Work, Mads, nn Plated sd bound in Gest Stan by Bde Li Gord and Kings Lyn “his bok ised o espe spony manatee lle Fresy, in Which east (wots ar pled foreach one sed fr paper prodeaon | Contents Foreword ‘Preface to the Second Baition xix Preface to the Thied Edition wil Obiewsry eat PART L INTRODUCTION AND THE LINEAR REGRESSION MODEL 1 1. What is Beonometrics? 3 ‘What is in this Chapter? 3 LL What is Beonometics? 3 12. Beanomie and Boonometric Models 4 13 The Aims and Methodology of Econometrics 6 14 What Constitutes a Tet of an Economic Theory? 9 Summary and an Online ofthe Book 8 2 Statistical Background and Matix Algebra 1 ‘What isin this Chapter? a 21 Inuoduction 2 22° Probebility R ‘Addition Rules of Probability 1B Conditional Probability and the Multiplication Role a Bayes’ Taeorem 15 Summation and Product Operations 5 23° Random Variables and Probably Disuibutions 0 ‘oint, Marginal, and Conditional Distbutions 18 Dlustrative Example 1B 24 The Normal Probability Distibution and Related Distributions 9 “The Normal Distribution 19 Related Distributions 2025 Classical Statistical Inference Point Estimation 26 Propertos of Estimators ‘Unbiacednese ciency Consistency Other Asymptotic Properties 27 Sampling Distbutions for Samples from a Normal Population 28 Interval Estimation 29° Testing of Hypotheses 210 Relationship Between Confidence Interval Procedures and Tests of Hypottiees 2.11 Combining Independent Teste Summary Bxercises ‘Appendix to Chapter 2 Matrix Algebra Exercises on Matrix Algebra ‘Simple Regression What isin this Chapter? 3.1 Tnvoduetion 3.2 Specification of the Relationships 33. The Method of Moments lustrative Example 34 The Method of Least Squares Reverse Regression IMuseatve Example 3.5. Statistical Inference inthe Linear Regression Model Mlustrative Example Confidence Intervals for a, 6, and o ‘Testing of Hypotheses Example of Comparing Test Scores ffom the GRE and GMAT Tests Regression with No Constant Terma 36 Analysis of Variance forthe Simple Regression Model 3.7 Prediction with the Simple Regression Mode! Prediction of Expected Valves lastradve Example 38 Outliers Some Mlustrative Examples 39 Altemative Functional Forms for Regression Equations Tlustrative Example "3.10 Inverse Prediction inthe Least Squares Regression Model “B.L1 Stochestic Regrestors | 1 conrenrs *3.12 The Regression Fallacy “The Bivariate Normal Distcbution Galton’s Result and the Regression Fallacy ‘A’Note on the Tecm: “Regression” Summary Exercises ‘Appendix t0 Chapter 3 4 Multiple Regression What af 42 43 44 45 46 ar as 49 410 ant oad is i this Chapter? Introduction ‘A Model with Two Explanatory Variables ‘The Least Squares Method IMlusvative Example Statistical Infereace in the Multiple Regression Model Mustative Example Formulas for the General Case of & Explanatory Variables Some llustrative Examples Interpretation of the Regression Coefficients ‘iusirative Example Partial Correlations and Multiple Correlation Relationships Among Simple, Parisl, and Muluple Corelation Coeficienss ‘Two Illustrative Examples ‘Prediction in the Multiple Regression Model lustrative Example ‘Analysis of Variance and Tests of Hypotheses Nested and Nonnested Hypotheses: ‘Tests for Linear Functions of Parameters Mlusrative Example ‘Omission of Relevant Variables and Inclusion of Irrelevant ‘Variables Omission of Relevant Variables ‘Example 1: Demand for Food in the United States ‘Exaraple 2: Produetion Functions and Management Bias Inclusion of Irrelevant Variables Degrees of Freedom and R* ‘Tests for Subility ‘The Analysis of Variance Test Example 1: Stability of the Demand for Food Function Example 2: Stability of Production Functions Predictive Tests for Stability istrative Example ‘The LR, W, and LM Tests Mustrative Example 102 62 108 108 105 108 2 ru ra 2 130 2 Be Bs 169 0 43 as “1 Mg 183 153 134 136 137 158 159 10 ist 162 163 164 168 168 169 173 13 116 116vi Summary Exercises ‘Append to Chapter 4 “Thue Multiple Regression Model in Matrix Notation Data Seis PART Il VIOLATION OF THE ASSUMPTIONS OF THE BASIC MODEL 5 Heteroskedasticity ‘What is in this Cnapter? 5.1 Introduction Mlustrative Example 52 Detection of Heteraskedaticity Tlustative Example ‘Some Other Tests lastratve Example ‘An Intuitive Justification for the Breusch~Pepan Test 53. Consequences of Heteroskedasticty ‘Eaiimation ofthe Variance of the OLS Estimator Under Hetaroskeducicity SA Solutions to the Hoteroskedasticity Problem Mustratve Example 55° Heteroskedastcity and the Use of Deftators lustrative Example: The Density Gradient Model *56 Testing the Linear Versus Log-Linear Functional Form “The Box-Cor Test The BM Tost The PE Test Summary Exerciacs ‘Appendix to Chapter S Generalized Least Squares 6 Autocorrdation ‘What isin this Capes? 61 Imoduetion 62 Durbin—Watson Test lustrauve Example 63 Bstimacion in Levels Versus First Differences Some Hlustrative Examples 6A Estimation Procadurs with Autocorrelsted Errors Iterative Procedures Grid-Seareh Procedures Mlustative Example 65 Effect of AR(I) Enors on OLS Estimates conrenrs m7 179 185, 185 192 197 19 2 212 5 27 21 219 219 21 228 nr 27 21 ne 230 2 236 237 238 238 | | CONTENTS 66 Some Funher Comments on the DW Test ‘The von Neamann Ratio The Berenblut- Wed Test 627 Tests for Serial Correlation in Motels with Lagued Dependent Variables Durbia's ATest, Durbin's Akemative Test lustrative Example 68 | A General Test for Higher-Order Serial Conrelation: The LM Test 69 Suategies When the DW Test Statistic ie Significant Errors Not AR(!) ‘Autocorrelation Caused by Omitted Variables Serial Corelation Due to Misspecified Dynamics ‘The Wald Test ustative Example £6.10 Treas and Random Walks Spurious Trends Differencing and Long-Run Effets: The Concept of Coin "611 ARCH Models and Serial Corelation 612. Some Comments on the DW Test and Durti ‘summary Exercises, 5 Test and Test 7 Malticolinesrty ‘What isin this Chapter? 71 Introduction 72. Some ilusuative Examples 73 Some Measures of Multicollnearity TA. Problems with Measuring Mulicellinearity 7.5 Sdivions 10 the Multicollinearity Problem: Ridge Regression 7.6 Principal Component Regression 1.7 Dropping Vasablee 718 Miscellaneous Other Solutions ‘Using Ratios or First Differences Using Fxtraneous Estimates GGeuing More Data Summary Exercises Appendix to Chapter 7 Lincerly Dependent Explanstory Variables ‘8 Dummy Variables and Truncated Variables ‘What isin this Chapter? SL Inreduction 21 a 21 293 23 301 301 30132 83 Ba as 86 37 88 89 Dunmy Variables fr Changes inthe Inereept Tem lative Example wo Morera Examples Dummy Variables for Changes in Stope Coofciens Dummy Variables for Cros Equntion Constants Dama Variables for Testing Sally of Repression Coefficients FA Pee Dommy Vasabes Under Hetrokodastciy and Autcenilaton Dura Dependent Vales The Linear Probability Model andthe Linear Dissminane Ption “he Linea Probability Model “he Linear Diseriina Functon ‘The Poti and Logit Mods sti Example The Problem of Dispropoionae Sampling Fret of tfc of Changes inthe Explanatory Vaibes easing Goodness of Fit 10. Ilusrative Example Truncated Variables: The Tobit Modet Some Examples Method of Estimation Limiations of the Tobit Model ‘The Truncated Regression Model Summary Brews Simultanous Equations Models Wats the Capt 1 eden 212 Enogenes and Exogenous Vai 93° ‘Menton Potion een Gogh Reed Fr trav ample 94 Ness) an Suen Condos for Kenton Tntoaie Enege 915. Methods of satan Te Isrumenl Vasa Mod Neseg sate Example 9.6 Muha af Eman: Th To Sage Lent yar Method Computing Standard Errors Bi Ivete ample 24, Tue Question of Norman 98 The Lint vena Mino Likeload Matos Mlustrative Example ‘conTENTs 302 305 306 307 310 313 316 317 318 318 320 32 34 325 307 327 329 333, 333 334 33s 5336 338 339 363 33 34s 346 351 353 358 356 as7 361 367 368 ‘CONTENTS 1» "9.9. On the Use of OLS in the Estimation of Simultmenus ‘Bauations Models ‘Working's Concept of Identification Recursive Systems Esimation of Cobb—Dougls Production Functions *9.10 Exogencity and Cavsality ‘Weak Bxogencity Supsrexogeneity Strong Exogeneity Granger Causality Granger Causality and Exogenity Tests for Exogencity| 9.11 Some Problems with Instrumental Variable Methods Summary Brecises Appendix to Chaps 9 ‘Nonlinear Regressions, Models of Expectations, and Nounormality ‘What is inthis Chapter? 10.1 Introduction 10.2. The Newton—Raphson Method 10.3 Nonlinear Least Squares "The Gauss-Netion Method 104 Models of Expectations 10.3 Naive Models of Expectations 10.6 The Adaptive Expectations Model 10.7 Estimation withthe Adaptive Expectations Modet Estimation inthe Autoregressive Form Eximation inthe Distributed Lag For 10.8 Two Dhstaive Examples 10°9. Expectional Variables and Adjustent Laps 10-10 Pars Adjustment with Adaptive Expectations 10.11 Allemative Distributed Lag Models: Polynomial Lags Finite Lags: The Polynomial Lag Thstrative Example ‘Choosing the Degree of the Polynomial 10.12 Rational Legs 10.13 Rational Expectations 1.14 Teas for Rationality 0.15 Bsimation of ¢ Demand and Supply Model Under Rational Expectations ‘Case 1 ‘Case 2 usvative Example 10.16 The Sesial Correlation Problem in Rational Expectations Models, 369 an 33 33 315 378 a8 319 380 380 3a) 382 383 386a 10.17 Nonnormality of Errors “Tests for Normality 10.18 Data Transformations Summary Broreises 111. Brrors in Variables ‘Whit is in this Chapter? ILL Isttoduetion 112. The Classical Solution for a Single-Bquation Model with One Explanatory Variable 113 The Single-Equation Model with Two Explanatory Variables ‘Two Bxplanatory Variables: One Measured with Error Blustative Bxample ‘Two Explanatory Variables: Both Measured with Exzor 114 Reverse Regression 115 Inserumental Variable Methods 116 Proxy Variables Coefficient ofthe Proxy Variable 117 Some Other Problem ‘The Case of Multiple Bquations Correlated Errors Summary Exercises PART III SPECIAL TOPICS 12 Diagnostic Checking, Model Selection, and Specification Testing ‘What isin this Chepiec? 121 Introduction 122 Diagnostic Tests Based on Least Squares Residuals ‘Tests for Omitted Varablee ‘Tests for ARCH Effecis 123 Problems with Least Squares Residuals 124 Some Other Types of Residuals ‘Predicted Residuals and Studenized Residuals Dummy Variable Method for Studentized Residuals [BLUS Residuals Recursive Residuals lustratve Example 125 DFFITS and Bounded Influence Eatimation lustative Example 12.6 Model Selection HYypothesis-Testing Search Inexpreive Search 461 463 465 466 467 469 470 470 am an an 474 496 478 479 480 41 ‘CONTENTS fc n8 129 ‘Simplification Search roxy Variable Search ‘Dasa Selection Search oat Data Mode! Construction Hendry’s Approach to Model Selection Selection of Regressrs ‘Taeil’s R? Criterion (Criteria Based on Minimizing the Mean-Squared not of Prediction ‘Akaike's Information Criterion Implied F-Ratios forthe Various Criteria ‘Theorem and Posterior Odds for Model Selection ‘Crose-Validation 12.10 Hausman’s Specification Error Test ‘An Applicaton: Testing for Ecors in Variables or Bxogenety ‘Some Tlusuative Examples ‘An Omitted Variable Interpretation of the Hausman Test 12.11 The Plosser—Sehwert-White Differencing Test 12.12 Tess for Nomested Hypotheses ‘The Davidson and MacKinnon Test ‘The Encompassing Test ‘A Basic Problem in Testing Nonnested Hypotheses Hypothesis Testing Versus Model Selection as Research stateey ‘Summary Exercises “Appendix to Chapter 12 13 Introduction to Time-Series Analysis ‘What is inthis Chapter? BI 2 133 Ba Inzodvction ‘Two Methods of Time-Serios Analysis: Frequency Domain and ‘Time Domain Stationary and Nonstationary ‘Time Series ‘Strict Stationarity ‘Weak Staionarity Propeties of Aulocomelation Function Nonsationatiy Some Useful Models for Time Series ‘Purely Random Process Random Walk ‘Moving Average Process ‘Autoregressive Process ‘Autoregressive Moving Average Process ‘Autoregressive Integrated Moving Average Process ai 481 a1 432 483aw “ ‘conrenrs 135 Estimation of AR, MA, and ARMA Models Estimation of MA Models atimation of ARMA Models Residuals fromm the ARMA Models ‘Teeting. Goodnest of Fit 136 The Box-Jenkins Approsch Forecasting from BoxTJenkins Models Mlustative Example “Trond Elimination: The Traditional Method A Summary Assessment Seasonality in the Box—Jenkins Modeling 13.1 R® Measures in Time-Series Models Summary Exercises Dats Sots Vector Autoregressions, Unit Roots, and Cointegration What isin this Chapter? 14.1 Inttodsetion 142 Vector Autoregressions 143 Problems with VAR Models in Pa 144 Unit Ros 145 Unis Root Tests Dickey-Fuller Test ‘The Serial Concation Problem ‘The Low Power of Urit Root Tests ‘The DF-GLS Test What are the Nall and Alternative Hypotheses in Unit Root Teste? ‘Tests with Stationaity as Nall ‘Confrmatory Anaysie Pane! Data Unit Root Tests Structural Change and Unit Roots 146 Coimesration 147 The Coimegrating Regression 148 Vector Antoregressions and Cointegration 149 Cointegration and Error Correction Models 14.10 Tests for Cointegration 14.11 Coincegration and Testing ofthe REH ard MEH 14.12 A Summary Assessment of Cointegration Summary Exercises 504 525 526 327 529 33 532 535 535 540 S41 343 543 546 sa 548 8 349 550 $50 350 552 353 554 355 337 +360 366 568 510 ‘CONTENTS 435 Panel Data Analysis ‘What isin this Chapter? 15.1 Iatroductio 152. The LSDV or Fixed Bffects Model 153 The Random Effects Model 15.4 Fixed Effects Versus Random Effects ‘aveman Test Breasch and Pagan Test 155° The SUR Model 1566 Dynamic Panel Data Models 15.7 The Random Coefficient Model Summary 16 Large-Sample Theory ‘What isin this Chapter? 16.1 The Maximom Likelihood Method 162 Methods of Solving the Likelihood Equations 16.3 The Cramer-Rao Lower Bound 164 Large-Sample Tests Based on ML. 16.5. GIVE and GMM Summary 17 Small-Sample Inference: Resampling Methods ‘What isin this Chapter? UT.1 Tntoduetion 172. Monte Carlo Methods ‘More Eficient Monte Carlo Methods Response Surfaces 17.3 Resampling Methods: Jackknife and Bootstrap ‘Some Mustrative Examples ‘OdherIsues Relating 10 Bootstrap 17.4 Booisiap Confidence fntervals 17.5. Hypothesis Testing with the Bootstrap 1716 Boostapping Residuals Versus Bootstrapping the Data 177 NonllD Emors and Nonsttionary Models Heteroskedasticity and Autocomelation ‘Unit Root Tests Basod on the Bootstrap Cointgraton Tests 17.8 Misellaneous Other Applications S13 mB om 5 8 578 59 ‘al 383Summary ‘Appendices ‘Appendix A: Data Sets ‘Appendix B: Data Sets on the Wed “Appendix C: Computer Programs Index ‘conrewTs 613 61s or Foreword 1 would lke w thank Professor Kajal Labiti for his help with the publication of this, ‘edition. After my husband G, S. Maddala passed away, Jo Ducey (his assistant) gave m= the manuscript he hed worked on until the last day of bis life. Looking atthe notes on his desk, I remembered how worried he was about being able to complete the revisions bofoce he died. Because I could not allow my husband's final effort to go to waste, I immediately sought the help of Kajal Lahin, my husband’ former student and long-time fiend. He graciously organized the manuscrip, incorporated my husband's revisions and ‘edited the text t provide a smooth transition between the old and new material. Labi also addressed other questions that my husband's passing left unansvtered. For instance, readers wil notice that the last tee capers do not include exercises. Prom my husband’s personal notes and computer fies, Lahiri determined that my husband intentionally omitted ‘exercises from the chapter on panel data. \s for Chaptess 16 (Lazge-Sample Theory) and 17 (Small Sample Inference: Resampling Methods), my husband didnot have thoss ready ‘atthe time of his death. We elected not to gdd new exercises, in order to preserve my hosband’s influence on the entre book. Without Lahin's immense help and advice, this book would not have been published. I am very gratefl for his hard work and dedication to this text {would also like to thank Ms. Samantha Whittaker and Mr. Stove Hardman for their valuable help and encouragement in preparing this raanusript for publication. ‘This volume is dedicated tall of G, S. Maddala's tuden, who meant so mvc to im. Kay Maddala October 2000Preface to the Second Edition ‘There hate been many imporant developments in econometrics during the last two sdecaces, but introductory books inthe fleld sill deal masdy wlth what econometnies was inthe 1960s, The present book is meant wo familiarize students (and researchers) with some ‘of these developments, explaining em with very simple models without cluterng up the ‘exposition with too much algebraic detail. Whece proofs involve complicated expressions they are omiited end appropriate references are given. Ten of the more difficult sections Ihave beon marked with an asterisk indicaing that they ae optional. Beginning stxdents ‘can tkip them and proceed. The book also contains several examples illustrating the techniques at each stage, and wiere illustrations are not given, some datasets have been ‘provided so that students can compute the required results themselves Since the book covers quite a few topics, only a few examples have been given to illustrate eack point. Giving oo maay ilustrations fara single point might be boing for ‘some students aad woud have made the book much foo bulky. The inscucter's manual ‘contains mere illustrative examples and questions and answers, The exereoes given a the ‘end of sach chaptr are somewhat challenging. However, the instructor's manta contans snawers or guidelines, The instructor's manual also gives a “guided tour” of the material in each chapter as well 36 some detiled explanations for some points that are touched brety in the book. ‘Some ofthe questions atthe end ofthe chapters have boen ken from the examinations a several US. and U.K. universities, and from P. C. B. Phillips and M.R. Wickes, Exercises in Econometrics (Cambridge, MA; Ballinges Publishing Co. 1978), Vol (Many of the questions in that book ate from examinations in the U.K. universities) Since quo td ot oped wih nr chang, have Gt bee ge ‘the exact source for each question.Pa PREFACE TO THE SECOND EDTON “There ace many distinguishing features of the book, some of which ar: 1 Aaa Samm i Gee 2 A gig ce Sc i pti Coe 2 Sin ee ew ag ‘nce Sa ee Ee ace es SE aa went ee ces ees a ts Delt oi a care Sea nc a mens 1. SEEN elton ni in i in so te cian ern a ann ate er crac Sie ee oh SESE a att ee Bee ene aa ey a pee area ee 4 Ero tone end nan it ce ain, sss meena tay Cote seen en eee om nec nth nid Byatt na wr aly ae 1 ee ae Sy setae cla as hina ee Fe ci arma ae ck aoe ” ‘These are not new topics for advanos texts but these tops (most of which are never sample theory (maximum likelihood and GMM), and resampling methods (Monte Carlo and bootstrap fo smallsarple inference. In cach ‘ase, the discussion is at an elementary ard easily understandable level. The chapters are also organized into three pars. Part Tand Il are appropriate for undergraduate courses Pare ll can be added for (two-semestr) beginning graduate and MBA courses (with selective deletions if necessary), With the expanded and comprehensive coverage this book should also be of ue to graduate students and empirical researches alike. To make space fo the new material and forthe convenience of students, the datasets inthe second edition (as well as new datasets) have been put on fo a Website, Tn addition tothe people I thanked in my prface tothe second edition, {would like to thank the following. First of al | would ike to thank Min Qi at Kent State University for hor deziled comments on the second edition, suggestions for the topics to be adkied to this ‘hind edition, and constant encouragement inthe preparation ofthis edition. Without her help thie edition would not have been completed, I would also like to thank Hengyi Li at the Chinese University of Hong Kong for his help wit the datasets, and Yong Hoon Koo or several corrections to the second extion, Yasushi Toda is thanked for supplying one ofthe large data sis. Mark N. Haris and Lachlan R. Macquarie from Monash Univecsty ‘provided several useful suggestons, anf references fr ths revisoa, for which Iam very ‘thankful, Others who provided some hepful comments include Badi Baltgi, Kajal Lahin, IieMoo Kim, Tom Means, Mary MeGurvey, Robert Trost, Yong Yin and several others ‘(oo many to list) who used the esrlie: editions of this book. Finally woul like to thank Jo Ducey and Shaowen Wa for typing the aditions to the second edition. GS. MaddalaObituary’ : On June 4, 1999 G. S. Maddala (popalarly and affectionately known as GS) passed awey jn Columbos, OH at the age of 66. A leading figure in the econometrics profession for more than three decades, he held the University Eminent Scholar Professorship in the Department of Esonomics at Ohio State University atthe time of hie death. GS is survived by his wife Kameswari, “Kay.” and sevecal members of his immediate fail i son, Vivek, of San Francisco; and two sisters who ive GS's previous university affliatons include Stanford University (1963—1967), Univer sity of Rochester (1967-1975), and University of Florida (1975-1993). At Pioida, he was a Graduate Reseaich Profesor and the Diector of the Center for Beonometrics ‘and Decision Sciences. He held visiting appointments at Comell, Yale, CORE, Monash, Columbia, the California Institute of Technology (asthe Fairchild Distinguished Scholar, and many other institutions. At the time of his death, GS was an Advisory Editor of Econometric Theory, an Associate Editor of the Journal of Statistical Planning and lfer- fence, and 2 long-time Fellow and active participant of the Econometric Society. He was an Associate Editor of Econometrica during 1970-1979 and an Associate Bditor of the Journal of Applied Econometrics from 1993 to 1996. GS had » passion for wring and research and the gift of a billant expositer —the ability to co through the technical superstructure Lo reveal essential detail only, yet in the process of simplification never to lose the nerve ceater ofthe methods he sought o explain. ‘This skill was apparent in all his waiting and wae the central element ia his textbook ‘expositions. His 1977 econometrics text redefined the boundaries of econometrics that ‘could be integrated ino graduate econometics teaching and became 2 new standard for ‘subsequent economic textbooks. His 1983 Econometric Society monograph, “Limited ‘Dependent and Qualitative Variables in Econometrics," was an immediate bestseller and ‘was declared a citation classic in Current Contents (Vol. 3, July 26, 993). It has fueled "Tis bony ft paren he Jal Exerc Theory, plied by Came Usiveiy Fee,far osiruany ‘much of te innovative applied research in this area during the last 15 years, and it has served asa bile to empirical researchers in applied microeconomics ‘Bom in India to a family of very modest means, GS came 0 the Department of Economics at the University of Chicago as a Fulbright Scholar with @ R.A. in Mathematics fom Andhra University and an M.A. in Stastice from Bombay University. Since completing his doctorate at Chicago in 1963, GS has been a prolific writer, producing fover I10 articles and 12 books covering almost every emerging arcs of economeuics ‘hroughout his carer. ‘According to the Social Science Citation Index, GS has been one of the top five most cited econometricians during each of the years 1988-1994, This is undeniable testimony 10 his influence on contemporary econometric thought. He hes worked on an ‘enormous range of topies in econometrics, coming close to covering the entire field In econometric methodology, be has contibuted to distributed lags, generalized least squares, patel data, sitwltaneous equations, otvors in variables, income distribution, ‘witeiig regressions, disequilibrium models, qualitative and limited dependent variable ‘models, selfselection mes, outliers ad bootstp methods, unit roots and cointegration ‘methods, and Bayesian econometrics, among other topics. In empirical economics, GS thas written papers in euch diverse areas a8 consumption, production and cost functions, ‘money demand, regulation, pueudo-cala, returns to college education, housing market discrimination, eucvey dst on expectations, ad risk premia in futures markets, How rermarkable it Is that he has made significant contributions to such vatied areas of economeircs. Yet, in his often self-deprecating, way, GS used to say that he never made any hard decisions to do anything in life, he just tumbled onto them—tha his whole atitude in these matters was ialluenced by the Indian philosophy that “nothing matters.” ‘Byen though hit mast influential contibutions have been in methodology, he loved to write econometrics in plain English and bad an extrotdinary talent to penetrate to the ‘esseace of problems, conveying basic ideas lucdly in simple words. This exceptional expository capability made bitn revered by applied and theoretical econometictans alike. GS's siyle was often to take a crital but constructive fook at evolving econometric techniques, in particular those that have litle practical application. In doing so, he never hesitated to go against the tide ofthe profession. He was onc of the few econometicians who consianly asked whether the questions being answered are worth asking, abways maintaining a clear perspective on # wide cange of issues in econometrics and theic ‘eladonship to economic problems. In oral presentations at seminars and conferences, he ‘was well known for making points with a characteristic sense of wit and humor, not to snenion bs self-deprecating, oft tated remak that “it wae all in dhe paper,” repretenting a humorous counterculture to present-day demonstrative livedata-show presentations. "Those close to him were constantly awace of GS's pervasive sense of humor. Witheat hi presence, confereaces will never be the same. GS has been a mentor and a source of inspiration to more than 50 doctoral students, ‘whom he hae supervised over the years. They have beneBied from the durability of his insights and she gower of his analytical mid. Until he last day of hs life, GS kept doing ‘what he liked to do the most—tlking to graduate students, worrying about them, and ‘writing papers and books with them. After hearing the news of his death, one of GS's ‘current PILD. students sen the following e-mail message 10 all his Ohio State University onrruaay et rotors and fellow graduate stents,“ stated to work with GS almost a Even hough tina sho amount of time I reazed lfc that anyone who was case 1.65 mt hve alan S's gal pron war on goa an eis am eronometician. Aran sconometcin ne wrote tty boos an pepere at anyone ‘an rad. Hower, his quali an a person was only enjoyed by afow who wee te fim. fal ay lose that God gv mo the oppatnity te cle to him dung tis year” Ther could be no more Sting eg for teacher to whom hit Soden ea £0 much and none tat GS himee? would have pied tore Mehy than hat With hs death, the economics profesion has lost one of is most prolife eerget, anda nda memes, a man wha was loved by so aay of ts Kajal Laird State University of New York, Albany Peter C.B. Philips Yale University June 1999Part | Introduction and the Linear Regression Model ‘This part consists of four chapters. ‘Chapter discusses the changes in the methodology that have taken place since the 1930s and 1960s and gives an outline of the book. ‘Chapter 2 reviews some basic results in states. Most students will have covered this introdvetory material in courses in stalstics. This chapter also provides an inttoduction to matrix algebra. Chapter 3 covers the simple regression model and Chapter 4 the multiple regression model in detail. This forms the basics of linear regression under the assumption of independent anc identically distributed normal eerrs. These assumptions ae relaxed in Part I1 What is Econometrics? Chapter? ‘This chapter explains the scope and methodology of econometics. Beonometrics deals ‘with the application of stastical (ols to economic data. The first tsk an econometiian face is that of formulating an economic relationship, which is necessarily a simplifed ‘model of the real-world process. Estimation and testing of est models with observed ‘data, andthe use of the estimated models for prediction and policy analysis are the cer to major goals of econometrics This chapter ass contains a schematic depiction ofthe various methodological steps {involved in an econometric analysis, 1.1 What is Econometrics? Literally speaking, the word “econometies” means “measuement in economics.” This 100 broad defniton to be of any use because most of economics is concermed With measurement. We measure our gross national product, employment, money supply, ‘exports, imports, price indexes, and so on. What we mean by econometrics is: ‘The application of sasical and matbematcal methods wo the analysis of economic dat, with a purpose of giving empuicel content eccnomic thers aed verifying Wem of ‘efving hers In this respect econometrics is distinguished from mathematical economics, which consists of the applicatioa of matematies only, and the theories derived need not necessarily have an empirical content