Instant ebooks textbook Time series analysis: nonstationary and noninvertible distribution theory 2n

Page 1


Time Series Analysis: Nonstationary and Noninvertible Distribution Theory

2nd Edition Katsuto Tanaka

Visit to download the full and correct content document: https://textbookfull.com/product/time-series-analysis-nonstationary-and-noninvertibledistribution-theory-2nd-edition-katsuto-tanaka/

More products digital (pdf, epub, mobi) instant download maybe you interests ...

Time series analysis : nonstationary and noninvertible distribution theory Second Edition Tanaka

https://textbookfull.com/product/time-series-analysisnonstationary-and-noninvertible-distribution-theory-secondedition-tanaka/

Indexation and Causation of Financial Markets

Nonstationary time series analysis method 1st Edition Yoko Tanokura

https://textbookfull.com/product/indexation-and-causation-offinancial-markets-nonstationary-time-series-analysis-method-1stedition-yoko-tanokura/

Spectral Analysis for Univariate Time Series 2nd Edition Donald B. Percival

https://textbookfull.com/product/spectral-analysis-forunivariate-time-series-2nd-edition-donald-b-percival/

Time series analysis 1st Edition Palma

https://textbookfull.com/product/time-series-analysis-1stedition-palma/

Mastering Time Series Analysis and Forecasting with Python Bridging Theory and Practice Through Insights Techniques and Tools for Effective Time Series Analysis in Python 1st Edition Sulekha Aloorravi

https://textbookfull.com/product/mastering-time-series-analysisand-forecasting-with-python-bridging-theory-and-practice-throughinsights-techniques-and-tools-for-effective-time-series-analysisin-python-1st-edition-sulekha-aloorravi/

Distribution systems analysis and automation 2nd Edition Juan Manuel Gers

https://textbookfull.com/product/distribution-systems-analysisand-automation-2nd-edition-juan-manuel-gers/

Design and analysis of time series experiments 1st Edition Bartos

https://textbookfull.com/product/design-and-analysis-of-timeseries-experiments-1st-edition-bartos/

Applied time series analysis with R Second Edition

Elliott

https://textbookfull.com/product/applied-time-series-analysiswith-r-second-edition-elliott/

Time Series Analysis Methods and Applications for Flight Data Zhang

https://textbookfull.com/product/time-series-analysis-methodsand-applications-for-flight-data-zhang/

TimeSeriesAnalysis

NonstationaryandNoninvertibleDistributionTheory

ProfessorofStatisticsandEconometrics

GakushuinUniversity,Tokyo

SecondEdition

Thiseditionfirstpublished2017 ©2017JohnWiley&Sons,Inc

LibraryofCongressCataloging-in-PublicationData

Names:Tanaka,Katsuto,1950-

Title:Timeseriesanalysis:nonstationaryandnoninvertibledistribution theory/KatsutoTanaka,professorofstatisticsandeconometrics, GakushuinUniversity,Tokyo.

Description:Secondedition.|Hoboken,NJ:JohnWiley&Sons,Inc.,[2017] |Series:Wileyseriesinprobabilityandstatistics|Includes bibliographicalreferencesandindexes.

Identifiers:LCCN2016052139(print)|LCCN2016052636(ebook)|ISBN 9781119132097(cloth)|ISBN9781119132110(pdf)|ISBN9781119132134 (epub)

Subjects:LCSH:Time-seriesanalysis.

Classification:LCCQA280.T352017(print)|LCCQA280(ebook)|DDC 519.5/5–dc23

LCrecordavailableathttps://lccn.loc.gov/2016052139 9781119132097

Setin10/12ptWarnockbySPiGlobal,Chennai,India PrintedintheUnitedStatesofAmerica

Contents

PrefacetotheSecondEdition xi

PrefacetotheFirstEdition xiii

PartIAnalysisofNonFractionalTimeSeries 1

1ModelsforNonstationarityandNoninvertibility 3

1.1StatisticsfromtheOne-DimensionalRandomWalk 3

1.1.1EigenvalueApproach 4

1.1.2StochasticProcessApproach 11

1.1.3TheFredholmApproach 12

1.1.4AnOverviewoftheThreeApproaches 14

1.2ATestStatisticfromaNoninvertibleMovingAverageModel 16

1.3TheARUnitRootDistribution 23

1.4VariousStatisticsfromtheTwo-DimensionalRandomWalk 29

1.5StatisticsfromtheCointegratedProcess 41

1.6PanelUnitRootTests 47

2BrownianMotionandFunctionalCentralLimit Theorems 51

2.1TheSpace L2 ofStochasticProcesses 51

2.2TheBrownianMotion 55

2.3MeanSquareIntegration 58

2.3.1TheMeanSquareRiemannIntegral 59

2.3.2TheMeanSquareRiemann–StieltjesIntegral 62

2.3.3TheMeanSquareItoIntegral 66

2.4TheItoCalculus 72

2.5WeakConvergenceofStochasticProcesses 77

2.6TheFunctionalCentralLimitTheorem 81

2.7FCLTforLinearProcesses 87

2.8FCLTforMartingaleDifferences 91

2.9WeakConvergencetotheIntegratedBrownianMotion 99

2.10WeakConvergencetotheOrnstein–UhlenbeckProcess 103

2.11WeakConvergenceofVector-ValuedStochasticProcesses 109

2.11.1Space C q 109

2.11.2BasicFCLTforVectorProcesses 110

2.11.3FCLTforMartingaleDifferences 112

2.11.4FCLTfortheVector-ValuedIntegratedBrownianMotion 115

2.12WeakConvergencetotheItoIntegral 118

3TheStochasticProcessApproach 127

3.1Girsanov’sTheorem:O-UProcesses 127

3.2Girsanov’sTheorem:IntegratedBrownianMotion 137

3.3Girsanov’sTheorem:Vector-ValuedBrownianMotion 142

3.4TheCameron–MartinFormula 145

3.5AdvantagesandDisadvantagesofthePresentApproach 147

4TheFredholmApproach 149

4.1MotivatingExamples 149

4.2TheFredholmTheory:TheHomogeneousCase 155

4.3Thec.f.oftheQuadraticBrownianFunctional 161

4.4VariousFredholmDeterminants 171

4.5TheFredholmTheory:TheNonhomogeneousCase 190

4.5.1ComputationoftheResolvent–Case1 192

4.5.2ComputationoftheResolvent–Case2 199

4.6WeakConvergenceofQuadraticForms 203

5NumericalIntegration 213

5.1Introduction 213

5.2NumericalIntegration:TheNonnegativeCase 214

5.3NumericalIntegration:TheOscillatingCase 220

5.4NumericalIntegration:TheGeneralCase 228

5.5ComputationofPercentPoints 236

5.6TheSaddlepointApproximation 240

6EstimationProblemsinNonstationaryAutoregressive Models 245

6.1NonstationaryAutoregressiveModels 245

6.2ConvergenceinDistributionofLSEs 250

6.2.1ModelA 251

6.2.2ModelB 253

6.2.3ModelC 255

6.2.4ModelD 257

6.3Thec.f.sfortheLimitingDistributionsofLSEs 260

6.3.1TheFixedInitialValueCase 261

6.3.2TheStationaryCase 265

6.4TablesandFiguresofLimitingDistributions 267

6.5ApproximationstotheDistributionsoftheLSEs 276

6.6NearlyNonstationarySeasonalARModels 281

6.7ContinuousRecordAsymptotics 289

6.8ComplexRootsontheUnitCircle 292

6.9AutoregressiveModelswithMultipleUnitRoots 300

7EstimationProblemsinNoninvertibleMovingAverage Models 311

7.1NoninvertibleMovingAverageModels 311

7.2TheLocalMLEintheStationaryCase 314

7.3TheLocalMLEintheConditionalCase 325

7.4NoninvertibleSeasonalModels 330

7.4.1TheStationaryCase 331

7.4.2TheConditionalCase 333

7.4.3ContinuousRecordAsymptotics 335

7.5ThePseudolocalMLE 337

7.5.1TheStationaryCase 337

7.5.2TheConditionalCase 339

7.6ProbabilityoftheLocalMLEatUnity 341

7.7TheRelationshipwiththeStateSpaceModel 343

8UnitRootTestsinAutoregressiveModels 349

8.1Introduction 349

8.2OptimalTests 350

8.2.1TheLBITest 352

8.2.2TheLBIUTest 353

8.3EquivalenceoftheLMTestwiththeLBIorLBIUTest 356

8.3.1EquivalencewiththeLBITest 356

8.3.2EquivalencewiththeLBIUTest 358

8.4VariousUnitRootTests 360

8.5IntegralExpressionsfortheLimitingPowers 362

8.5.1ModelA 363

8.5.2ModelB 364

8.5.3ModelC 365

8.5.4ModelD 367

8.6LimitingPowerEnvelopesandPointOptimalTests 369

8.7ComputationoftheLimitingPowers 372

8.8SeasonalUnitRootTests 382

8.9UnitRootTestsintheDependentCase 389

8.10TheUnitRootTestingProblemRevisited 395

8.11UnitRootTestswithStructuralBreaks 398

8.12StochasticTrendsVersusDeterministicTrends 402

8.12.1CaseofIntegratedProcesses 403

8.12.2CaseofNear-IntegratedProcesses 406

8.12.3SomeSimulations 409

9UnitRootTestsinMovingAverageModels 415

9.1Introduction 415

9.2TheLBIandLBIUTests 416

9.2.1TheConditionalCase 417

9.2.2TheStationaryCase 419

9.3TheRelationshipwiththeTestStatisticsinDifferencedForm 424

9.4PerformanceoftheLBIandLBIUTests 427

9.4.1TheConditionalCase 427

9.4.2TheStationaryCase 430

9.5SeasonalUnitRootTests 434

9.5.1TheConditionalCase 434

9.5.2TheStationaryCase 436

9.5.3PowerProperties 438

9.6UnitRootTestsintheDependentCase 444

9.6.1TheConditionalCase 444

9.6.2TheStationaryCase 446

9.7TheRelationshipwithTestingintheStateSpaceModel 447

9.7.1Case(I) 449

9.7.2Case(II) 450

9.7.3Case(III) 452

9.7.4TheCaseoftheInitialValueKnown 454

10AsymptoticPropertiesofNonstationaryPanelUnitRoot Tests 459

10.1Introduction 459

10.2PanelAutoregressiveModels 461

10.2.1TestsBasedontheOLSE 463

10.2.2TestsBasedontheGLSE 471

10.2.3SomeOtherTests 475

10.2.4LimitingPowerEnvelopes 480

10.2.5GraphicalComparison 485

10.3PanelMovingAverageModels 488

10.3.1ConditionalCase 490

10.3.2StationaryCase 494

10.3.3PowerEnvelope 499

10.3.4GraphicalComparison 502

10.4PanelStationarityTests 507

10.4.1LimitingLocalPowers 508

10.4.2PowerEnvelope 512

10.4.3GraphicalComparison 514

10.5ConcludingRemarks 515

11StatisticalAnalysisofCointegration 517

11.1Introduction 517

11.2CaseofNoCointegration 519

11.3CointegrationDistributions:TheIndependentCase 524

11.4CointegrationDistributions:TheDependentCase 532

11.5TheSamplingBehaviorofCointegrationDistributions 537

11.6TestingforCointegration 544

11.6.1TestsfortheNullofNoCointegration 544

11.6.2TestsfortheNullofCointegration 547

11.7DeterminationoftheCointegrationRank 552

11.8HigherOrderCointegration 556

11.8.1CointegrationintheI(d )Case 556

11.8.2SeasonalCointegration 559

PartIIAnalysisofFractionalTimeSeries

567

12ARFIMAModelsandtheFractionalBrownianMotion 569

12.1NonstationaryFractionalTimeSeries 569

12.1.1Caseof d = 1 2 570

12.1.2Caseof d > 1 2 572

12.2TestingfortheFractionalIntegrationOrder 575

12.2.1i.i.d.Case 575

12.2.2DependentCase 581

12.3EstimationfortheFractionalIntegrationOrder 584

12.3.1i.i.d.Case 584

12.3.2DependentCase 586

12.4StationaryLong-MemoryProcesses 591

12.5TheFractionalBrownianMotion 597

12.6FCLTforLong-MemoryProcesses 603

12.7FractionalCointegration 608

12.7.1SpuriousRegressionintheFractionalCase 609

12.7.2CointegratingRegressionintheFractionalCase 610

12.7.3TestingforFractionalCointegration 614

12.8TheWaveletMethodforARFIMAModelsandthefBm 614

12.8.1BasicTheoryoftheWaveletTransform 615

12.8.2SomeAdvantagesoftheWaveletTransform 618

12.8.3SomeApplicationsoftheWaveletAnalysis 625

12.8.3.1Testingfor d inARFIMAModels 625

12.8.3.2TestingfortheExistenceofNoise 626

12.8.3.3TestingforFractionalCointegration 627

12.8.3.4UnitRootTests 627

x Contents

13StatisticalInferenceAssociatedwiththeFractionalBrownian Motion 629

13.1Introduction 629

13.2ASimpleContinuous-TimeModelDrivenbythefBm 632

13.3QuadraticFunctionalsoftheBrownianMotion 641

13.4Derivationofthec.f. 645

13.4.1StochasticProcessApproachviaGirsanov’sTheorem 645

13.4.1.1Caseof H = 1∕2 645

13.4.1.2Caseof H > 1∕2 646

13.4.2FredholmApproachviatheFredholmDeterminant 647

13.4.2.1Caseof H = 1∕2 649

13.4.2.2Caseof H > 1∕2 650

13.5MartingaleApproximationtothefBm 651

13.6TheFractionalUnitRootDistribution 659

13.6.1TheFDAssociatedwiththeApproximateDistribution 659

13.6.2AnInterestingMomentProperty 664

13.7TheUnitRootTestUnderthefBmError 669

14MaximumLikelihoodEstimationfortheFractional Ornstein–UhlenbeckProcess 673

14.1Introduction 673

14.2EstimationoftheDrift:ErgodicCase 677

14.2.1AsymptoticPropertiesoftheOLSEs 677

14.2.2TheMLEandMCE 679

14.3EstimationoftheDrift:Non-ergodicCase 687

14.3.1AsymptoticPropertiesoftheOLSE 687

14.3.2TheMLE 687

14.4EstimationoftheDrift:BoundaryCase 692

14.4.1AsymptoticPropertiesoftheOLSEs 692

14.4.2TheMLEandMCE 693

14.5ComputationofDistributionsandMomentsoftheMLEand MCE 695

14.6TheMLE-basedUnitRootTestUnderthefBmError 703

14.7ConcludingRemarks 707

15SolutionstoProblems 709

References 865

AuthorIndex 879

SubjectIndex 883

PrefacetotheSecondEdition

Thefirsteditionofthisbookwaspublishedin1996.Thebookwaswrittenfrom atheoreticalviewpointoftimeserieseconometrics,wherethemainthemewas todescribenonstandardtheoryforlineartimeseriesmodelsthatarenonstationaryand/ornoninvertible.Ialsoproposedmethodsforcomputingnumericallythedistributionsofnonstandardstatisticsarisingfromsuchprocesses.

Themainthemeofthepresenteditionremainsthesameandreflects thedevelopmentsandnewdirectionsinthefieldsincethepublicationof thefirstedition.Inparticular,thediscussiononnonstationarypaneldata analysishasbeenaddedandnewchaptersonlong-memorydiscrete-timeand continuous-timeprocesseshavebeencreated,whereassomechaptershave beenmergedandsomesectionsdeleted.

Thiseditionisdividedintotwoparts:PartI:AnalysisofNonFractionalTime SeriesandPartII:AnalysisofFractionalTimeSeries,wherePartIconsistsof Chapters1through11whilePartIIconsistsofChapters12through14.The distinctionbetweennonfractionalandfractionaltimeseriesisconcernedwith theintegrationorderofnonstationarytimeseries.PartIassumestheintegrationordertobeapositiveinteger,whereasPartIIrelaxesthatassumptionto allowtheintegrationordertobeanypositiverealnumber.

Chapter1isessentiallythesameasthefirstedition,exceptfortheadditionof anintroductorydescriptiononnonstationarypanels,andisapreludetosubsequentchapters.Thethreeapproaches,whichIcalltheeigenvalue,stochastic process,andFredholmapproaches,totheanalysisofnonfractionaltimeseries areintroducedthroughsimpleexamples.Chapter2mergedChapters2and 3ofthefirsteditionanddiscussestheBrownianmotion,theItointegral,the functionalcentrallimittheorem,andsoon.

Chapters3and4discussfullythestochasticprocessapproachandthe Fredholmapproach,respectively.Theseapproachesareusedtoderivelimiting characteristicfunctionsofnonstandardstatisticsthatarequadraticfunctionals oftheBrownianmotionoritsratio.Chapter5isconcernedwithnumerical integrationforcomputingdistributionfunctionsviainversionofcharacteristic functionsderivedfromthestochasticprocessapproachortheFredholm approach.Chapters6through11dealwithunitrootandcointegration

problems.Chapters1through11exceptChapter10weremainchaptersof thefirstedition.Newtopicssuchasunitroottestsunderstructuralbreaks, differencesbetweenstochasticanddeterministictrends,havebeenadded. Chapter10hasbeenaddedtodiscussnonstationarypaneldatamodels,where ourmainconcernistocomputelimitinglocalpowersofvariouspanelunit roottests.Forthatpurposethemovingaveragemodelsarealsoconsideredin additiontoautoregressivemodels.

Chapters12through14havebeenwrittennewlyunderPartII:Analysisof FractionalTimeSeries.Chapter12discussesthebasictheoryoflong-memory processesbyintroducingARFIMAmodelsandthefractionalBrownianmotion (fBm).ThewaveletmethodisalsointroducedtodealwithARFIMAmodels andthefBm.Chapter13isconcernedwiththecomputationofdistributions ofquadraticfunctionalsofthefBmanditsratio,wherethecomputationofthe fractionalunitrootdistributionremainstobedone,whereasanapproximationtothetruedistributionisproposedandcomputed.Chapter14introduces thefractionalOrnstein–Uhlenbeckprocess,onwhichthestatisticalinferenceis discussed.Inparticular,themaximumlikelihoodestimatorofthedriftparameterisconsidered,andasymptoticsasthesamplingspanincreasesarediscussed.

Chapter15,thelastchapter,givesacompletesetofsolutionstoproblems posedattheendofmostsections.

Thereareabout140figuresand60tables.Mostoftheseareoflimiting distributionsofnonstandardstatistics.Theyareallproducedbythemethods describedinthiseditionandincludemanydistributions,whichhavenever appearedintheliterature.

Thepresenteditionisdedicatedtomywife,Yoshiko,whodiedin1999.

November2016

KatsutoTanaka Tokyo,Japan

PrefacetotheFirstEdition

Thisbookattemptstodescribenonstandardtheoryforlineartimeseries models,whicharenonstationaryand/ornoninvertible.Nonstandardaspectsof thedeparturefromstationarityorinvertibilityhaveattractedmuchattentionin thefieldoftimeserieseconometricsduringthelast10years.Sincethereseem fewbooksconcernedwiththetheoryforsuchnonstandardaspects,Ihavebeen atlibertytochoosemyway.Throughoutthisbook,attentionisorientedtoward themostinterestingtheoreticalissue,thatis,theasymptoticdistributional aspectofnonstandardstatistics.Thesubtitleofthebookreflectsthis.

Chapter1isapreludetothemaintheme.Byusingsimpleexamples,variousasymptoticdistributionsofnonstandardstatisticsarederivedbyaclassical approach,whichIcalltheeigenvalueapproach.Itturnsoutthat,ifmorecomplicatedproblemsaretobedealtwith,theeigenvalueapproachbreaksdown andtheintroductionofnotionssuchastheBrownianmotion,theItointegral, thefunctionalcentrallimittheorem,andsoonisinevitable.Thesenotions arenowdevelopedverydeeplyinprobabilitytheory.Inthisbook,however, aknowledgeofsuchprobabilitynotionsisrequiredonlyatamoderatelevel, whichIexplaininChapters2and3inaneasilyaccessibleway.

Probabilitytheory,inparticularthefunctionalcentrallimittheorem,enables ustoestablishweakconvergenceofnonstandardstatisticsandtorealize thatlimitingformscanbeexpressedbyfunctionalsoftheBrownianmotion. However,moreimportantfromastatisticalpointofviewishowtocompute limitingdistributionsofthosestatistics.ForthispurposeIdonotsimplyresort tosimulationsbutemploynumericalintegration.Tomakethecomputation possible,wefirstneedtoderivelimitingcharacteristicfunctionsofnonstandardstatistics.Tothisend,twoapproachesarepresented.Chapter4discusses oneapproach,whichIcallthestochasticprocessapproach,whileChapter5 discussestheother,whichIcalltheFredholmapproach.Thetwoapproaches originatefromquitedifferentmathematicaltheories,whichIexplainfully, indicatingtheadvantageanddisadvantageofeachapproachforjudicioususe.

Chapter6discussesandillustratesnumericalintegrationforcomputingdistributionfunctionsviainversionofcharacteristicfunctions.Thischapteris necessarybecauseadirectapplicationofanycomputerpackageforintegration

cannotdoaproperjob.WeovercomethedifficultybyemployingSimpson’s rule,whichcanbeexecutedonadesktopcomputer.Thenecessityforaccurate computationbasedonnumericalintegrationisrecognized,forinstance,when closecomparisonhastobemadebetweenlimitinglocalpowersofcompeting nonstandardtests.

Chapters7through11dealwithstatisticalandeconometricproblemsto whichthenonstandardtheorydiscussedinpreviouschaptersapplies.Chapter 7considerstheestimationproblemsassociatedwithnonstationaryautoregressivemodels,whileChapter8considersthosewithnoninvertiblemovingaveragemodels.Thecorrespondingtestingproblems,calledtheunitroottests,are discussedinChapters9and10,respectively.Chapter11isconcernedwith cointegration,whichisastochasticcollinearityrelationshipamongmultiple nonstationarytimeseries.Theproblemsdiscussedinthesechaptersoriginate intimeserieseconometrics.Idescribeindetailhowtoderiveandcompute limitingnonstandarddistributionsofvariousestimatorsandteststatistics.

Chapter12,thelastchapter,givesacompletesetofsolutionstoproblems posedattheendofmostsectionsofeachchapter.Mostoftheproblemsare concernedwithcorroboratingtheresultsdescribedinthetext,sothatonecan gainabetterunderstandingofdetailsofthediscussions.

Thereareabout90figuresand50tables.Mostoftheseareoflimitingdistributionsofnonstandardstatistics.Theyareallproducedbythemethodsdescribed inthisbookandincludemanydistributions,whichhaveneverappearedinthe literature.Amongthesearelimitingpowersandpowerenvelopesofvarious nonstandardtestsunderasequenceoflocalalternatives.

Thisbookmaybeusedasatextbookforgraduatestudentsmajoringineconometricsortimeseriesanalysis.Ageneralknowledgeofmathematicalstatistics, includingthetheoryofstationaryprocesses,ispresupposed,althoughthenecessarymaterialisofferedinthetextandproblemsofthisbook.SomeknowledgeofaprogramminglanguagelikeFORTRANandcomputerizedalgebralike REDUCEisalsouseful.

ThelateProfessorE.J.Hannangavemevaluablecommentsontheearly versionofmymanuscript.Iwouldliketothankhimforhiskindnessandfor pleasantmemoriesextendingovertheyearssincemystudentdays.Thisbook grewoutofjointworkwithProfessorS.Nabeya,anotherrespectedteacherof mine.Hereadsubstantialpartsofthemanuscriptandcorrectedanumberof errorsinitspreliminarystages,forwhichIammostgrateful.Iamalsograteful toProfessorsC.W.Helstrom,S.Kusuoka,andP.SaikkonenforhelpfuldiscussionsandtoProfessorG.S.Watsonforhelpofvariouskinds.Mostofthe manuscriptwaskeyboarded,manytimesover,byMs.M.Yuasa,andsomeparts weredonebyMs.Y.Fukushima,tobothofwhomIamgreatlyindebted.Finally, Ithankmywife,Yoshiko,whohasalwaysbeenasourceofencouragement.

January1996 KatsutoTanaka Tokyo,Japan

ModelsforNonstationarityandNoninvertibility

Wedealwithlineartimeseriesmodelsonwhichstationarityorinvertibility isnotimposed.Usingsimpleexamplesarisingfromestimationandtesting problems,weindicatenonstandardaspectsofthedeparturefromstationarity orinvertibility.Inparticular,asymptoticdistributionsofvariousstatistics arederivedbytheeigenvalueapproachunderthenormalityassumptionon theunderlyingprocesses.Asapreludetodiscussionsinlaterchapters,we alsopresentequivalentexpressionsforlimitingrandomvariablesbasedon theothertwoapproaches,whichIcallthestochasticprocessapproachandthe Fredholmapproach.

1.1StatisticsfromtheOne-DimensionalRandomWalk

Letusconsiderthefollowingsimplenonstationarymodel:

where ��1 ,��2 ,... areindependentandidenticallydistributedwithcommon mean0andvariance1,whichisabbreviatedas {��j }∼ i.i.d.(0, 1).Themodel (1.1)isusuallyreferredtoasthe randomwalk .Itisalsocalledthe unitroot process intheeconometricsliterature.

Letusdealwiththefollowingtwostatisticsarisingfromthemodel(1.1):

where y = ∑T j=1 yj ∕T .Eachsecondmomentstatistichasanormalizer T 2 ,which isdifferentfromthestationarycase,andisnecessarytodiscussthelimiting distributionas T → ∞.Infact,notingthat yj = ��1 +···+ ��j ,wehave

1ModelsforNonstationarityandNoninvertibility

Itholds[Fuller(1996,p.220)]that

ST 1 = Op (1), ST 2 = Op (1),

where XT = Op (aT ) meansthat,forevery ��> 0,thereexistsapositivenumber T�� suchthat P (|XT | ≥ T�� aT ) forall T .Itisanticipatedthat ST 1 and ST 2 have differentnondegeneratelimitingdistributions.

Wenowattempttoderivethelimitingdistributionsof ST 1 and ST 2 .There arethreeapproachesforthispurpose,whichIcallthe eigenvalueapproach,the stochasticprocessapproach,andthe Fredholmapproach.Thefirstapproachis describedhereindetail,whereasthesecondandthirdareonlybrieflydescribed andthedetailsarediscussedinlaterchapters.

1.1.1EigenvalueApproach

Theeigenvalueapproachrequiresadistributionalassumptionon {��j }.We assumethat ��1 ,��2 ,... areindependentandidenticallynormallydistributed withcommonmean0andvariance1,whichisabbreviatedas {��j }∼ NID(0, 1). Wealsoneedtocomputetheeigenvaluesofthematricesappearingin quadraticforms.Toseethistheobservationvector y =(y1 ,..., yT )′ maybe expressedas

wherethematrix C anditsinverse C 1 aregivenby

Thematrix C maybecalledthe randomwalkgeneratingmatrix andplayan importantroleinsubsequentdiscussions.

Wecannowrewrite ST 1 and ST 2 as

M = IT ee′ ∕T , e =(1,..., 1)′

Letuscomputetheeigenvaluesandeigenvectorsof C ′ C and C ′ MC .The eigenvaluesof C ′ C wereobtainedbyRutherford(1946)(seealsoProblem1.1 inthischapter)bycomputingthoseof (C ′ C ) 1 =

The jthlargesteigenvalue ��j of C ′ C isfoundtobe

Thereexistsanorthogonalmatrix P suchthat P ′ C ′ CP =Λ= diag(��1 ,...,��T ), wherethe k thcolumnof P isaneigenvectorcorrespondingto ��k .Itcanbe shown[DickeyandFuller(1979)]thatthe (j, k )thcomponentof P isgivenby Pjk = 2 √2T + 1 cos (2j 1)(

Ontheotherhand, C ′ MC isevidentlysingularbecausethevector e isthefirst columnof C and M e = 0 sothatthefirstcolumnof C ′ MC isazerovector.In fact,itholdsthat

wherethe (T 1)×(T 1) matrix G∗ isgivenby

Here C∗ and e∗ arethelast (T 1)×(T 1) and (T 1)× 1submatricesof C and e,respectively,whereas e1 =(1, 0,..., 0)′ ∶(T 1)× 1.Theeigenvaluesof

1ModelsforNonstationarityandNoninvertibility

canbeeasilyobtained(Problem1.2).Wealsohave |G 1 ∗ | = T .Thenthe jth largesteigenvalue ��j of G∗ isfoundtobe

1

Thereexistsanorthogonalmatrix Q ofsize T 1suchthat Q′ G∗ Q =Γ= diag(��1 ,...,��T 1 ),wherethe k thcolumnof Q isaneigenvectorcorresponding to ��k .Itcanbeshown[Anderson(1971,p.293)]thatthe (j, k )thcomponentof Q isgivenby

Qjk = √ 2 T sin jk

Wenowhavethefollowingrelations:

Notingthat �� ∼ N(0, IT

and ��

IT 1 ),wecancomputetheexactdistributionsof ST 1 and ST 2 byderivingthecharacteristicfunctions(c.f.s)as

Thenthedensitiesof ST 1 and ST 2 canbecomputednumericallyfollowingthe inversionformula

whereRe(z) istherealpartof z.Thesedensitieswillbedrawnlatertogether withthelimitingdensities.

Becauseofthepropertiesoftheeigenvalues ��j and ��j ,itroughlyholdsthat, as T → ∞,

Infact,itcanbeshown(Problem1.3)that

whichleadsustoderive

where ⇒ signifiesconvergenceindistribution. Thelimitingdistributionscanbecomputedbyderivingthec.f.sof

and S2 . Wehave

wherewehaveusedthefollowingexpansionformulasforcosandsinfunctions:

Figure1.1drawsthedensitiesof ST 1 for T = 10, 20, 50,and ∞.Thesewere computednumericallyfollowingtheinversionformulain(1.11).Thenumerical computationinvolvesthesquarerootofcomplexvariables,andhowtocomputethistogetherwithnumericalintegrationwillbediscussedinChapter5.It isseenfromFigure1.1thatthefinitesampledensitiesconvergerapidlytothe limitingdensity,althoughtheformerhaveaheavierright-handtail.

Figure1.2drawsthedensitiesof ST 2 for T = 10, 20,and ∞.Thesewerecomputedinthesamewayasthoseof ST 1 .NotethatFigure1.2doesnotcontain thedensityfor T = 50becauseitwasfoundtobeveryclosetothatfor T =∞, whileitisnotascloseinFigure1.1.

Probabilitydensitiesof ST ��

Probabilitydensitiesof ST �� .

Figure1.1
Figure1.2

Thenormalizer (T + 1 2 )2 for ST 1 ,insteadof T 2 ,couldmakefinitesample densitiesclosertothelimitingdensity.Morespecifically,wehavethefollowing expansionforthec.f.ofthemodifiedstatistic(Problem1.4).

Itisnoticedthattheexpansioncontainsnotermof O(T 1 ).Thisisnotthecase ifweuse T 2 asanormalizer.Ontheotherhand,wehave(Problem1.5)

Notethatthisexpansiondoesnotcontainthetermof O(T 1 ),whichexplains rapidconvergenceof ST 2 tothelimitingdistribution.

Table1.1reportspercentpointsandmeansfordistributionsof

y2 j ∕ (T + 1∕2)2 for T = 10, 20, 50,and ∞,where“E”standsforexactdistributions while“A”fordistributionsbasedontheasymptoticexpansiongivenin(1.16). Table1.2showsdistributionsof ∑T j=1 (yj y)2 ∕T 2 ,wheretheasymptotic expansion“A”isbasedon(1.17).Itisseenfromthesetablesthatthefinitesampledistributionsarereallyclosetothelimitingdistribution.Especially,percent

Table1.1 Percentpointsfordistributionsof

Probabilityofasmallervalue

1ModelsforNonstationarityandNoninvertibility

Table1.2 Percentpointsfordistributionsof

Probabilityofasmallervalue

pointsfor T = 50areidenticalwiththosefor T =∞ withinthedeviation of3/10,000.Asymptoticexpansionsalsogiveafairlygoodapproximationto finitesampledistributions.Inmostcases,theygiveacorrectvalueuptothe fourthdecimalpoint.

Itisaneasymattertocomputemomentsofthesedistributions.Let �� (j) T

be the jthordercumulantforthedistributionof

)2 basedon theasymptoticexpansionin(1.16).Define �� (j) T 2 similarlyforthedistributionof ∑T j=1 (yj y)2 ∕T 2 basedontheasymptoticexpansionin(1.17).Thenwehave (Problem1.6),upto O(T 2 ),

Cumulantsforthelimitingdistributionsaregiven(Problem1.7)by �� (j) 1 =lim T →∞ �� (j) T 1 = (j 1)! 23j 2 (22j 1) (2j)! Bj , �� (j) 2 =lim T →∞ �� (j) T 2 = (j 1)! 23j 2 (2j)! Bj , where Bj ’saretheBernoullinumbers: B1 =

=

= 1 30 ,andso on.Theskewness �� (3) 1 ∕(�� (2) 1 )3∕2 andkurtosis �� (4) 1 ∕(�� (2) 1 )2 3are2.771and8.657, respectively,while �� (3) 2 ∕(�� (2) 2 )3∕2 = 2.556and �� (4) 2 ∕(�� (2) 2 )2 3 = 7.286.

,

Theeigenvalueapproachhasbeensuccessfulsofar.Thisisbecause eigenvaluesassociatedwithquadraticformscanbeexplicitlycomputed,which israrelypossibleinmorecomplicatedsituations.Theothertwoapproaches, however,donotrequiresuchcondition,whichwewilldiscussnext.

1.1.2StochasticProcessApproach

Wecontinuetodealwiththerandomwalkmodel(1.1)andconsiderthesecond momentstatistics ST 1 and ST 2 givenin(1.2),wherewedonotassumenormality on {��j },butjustassume {��j }∼ i.i.d.(0, 1).

Thestochasticprocessapproach,whichwillbefullydiscussedinChapter3, startswithconstructingacontinuoustimeprocess {YT (t )} definedon t ∈[0, 1].Theprocess {YT (t )} isdefined,for (j 1)∕T ≤ t ≤ j∕T ,by

where YT (0)= 0and YT (1)= yT ∕√T .Theprocess {YT (t )} iscalledthe partialsumprocess,whichiscontinuousandbelongstothespaceofcontinuous functionsdefinedon [0, 1].Thentheprocess YT ={YT (t )} convergesweaklyto the standardBrownianmotion (Bm) W ={W (t )},whichwewriteas YT ⇒ W . Theconvergenceofthismodeentailsthenotionoftheweakconvergenceof stochasticprocessescalledthe functionalcentrallimittheorem (FCLT).Related materialswillbediscussedinChapter2.

Usingthepartialsumprocess {YT (t )},thetwostatistics ST 1 and ST 2 maybe rewrittenas

where RT 1 and RT 2 areremaindertermsthatconvergeto0inprobability.This willbeprovedinChapter2togetherwiththefollowingweakconvergence:

1ModelsforNonstationarityandNoninvertibility

TheaboveresultscanbeobtainedviatheFCLTandthe continuousmapping theorem (CMT).Thislasttheoremisanextensionofthecaseofcontinuous functionstothatofcontinuousfunctionalsandisdiscussedinChapter2.The stochasticprocess

iscalledthe demeanedBm,whichhasmean0ontheinterval [0, 1]. Thefollowingdistributionalequivalenceshouldnowhold.

Wehavealreadyobtained,bytheeigenvalueapproach,thec.f.softhelimiting distributionsof ST 1 and ST 2 .Thiswaspossiblebecausetheeigenvalueswere explicitlyknown.Supposethatwehavenoknowledgeabouttheeigenvalues.In thatcase Girsanov’stheorem,whichtransformsthemeasureonafunctionspace tothatonanotherandwillbediscussedinChapter3,enablesustocompute thec.f.softhedistributionsoftheresultingstatistics.

1.1.3TheFredholmApproach

Thesecondmomentstatistics ST 1 and ST 2 havethreekindsofexpressions describedin(1.5)and(1.6),respectively.Hereweusethelastexpressionsof these,thatis,

whereweonlyassume {��j }∼ i.i.d (0, 1).Thesestatisticscanbeseentobecharacterizedby

wherethereexistsacontinuousandsymmetricfunction K (s, t ) definedon [0, 1]×[0, 1] thatsatisfies

Then,asisdiscussedinChapter4,itholdsthat

where {W (t )} istheBm,whereastheintegralistheRiemann–Stieltjesdouble integralwithrespecttotheBmtobediscussedinChapter2.Thusitcanbe shownthat

Thec.f.softhelimitingrandomvariablesofthepreviousformcanbederived byfindingthe Fredholmdeterminant (FD)ofthefunction K (s, t ),fromwhich thepresentapproachoriginates.DetailswillbediscussedinChapter4. Sofarwehaveobtainedthefollowingdistributionalrelations:

Wenoteinpassingthatthefollowingrelationholds:

whichwillbeshowninChapter3byderivingthec.f.sbythestochasticprocess approach.Thestochasticprocess

W (t )= W (

(1.32) iscalledthe Brownianbridge (Bb),whichhasthepropertythat W (0)= W (1)= 0.

1ModelsforNonstationarityandNoninvertibility

Italsoholdsthat

whichwillbeshowninChapter4byderivingthec.f.sbytheFredholm approach.

1.1.4AnOverviewoftheThreeApproaches

Intheprevioussubsectionswediscussedthethreeapproachestoderiving asymptoticdistributionsofsecondmomentstatisticsarisingfromtherandom walk.Althoughdetailsarepostponeduntillaterchapters,itmaybeofsome helptogiveanoverviewofthethreeapproacheshere,makingcomparisons witheachother.Forthispurposewetakeupthestatistic

Thenthethreeapproachesmaybesummarizedintermsofthefollowing viewpoints,where(A),(B),and(C)refertotheeigenvalue,stochasticprocess, andFredholmapproaches,respectively.

• Expressionsforthestatistic:

• Assumptionsandtheoremsnecessaryforconvergenceindistribution:

(A)Distributionalassumptionsneedtobeimposedon {��j } andknowledge oftheeigenvaluesisrequired.

(B)Thepartialsumprocessneedstobeconstructed,andtheFCLTand CMTareusedtoestablishweakconvergence.

(C)ThekernelfunctionneedstobefoundandtheusualCLTisrequiredto establishweakconvergence.

• Derivationofthec.f.:

(A)Thelimitingexpressionisaninfinite,weightedsumofindependent �� 2 (1) randomvariables.Itsc.f.canbeeasilyderived.

(B)ThelimitingexpressionisasimpleRiemannintegralofthesquaredBm. Itsc.f.canbederivedviaGirsanov’stheorem.

(C)ThelimitingexpressionisaRiemann–StieltjesdoubleintegralofasymmetricandcontinuouskernelwithrespecttotheBm.Itsc.f.canbe derivedbyfindingtheFredholmdeterminantofthekernel.

Problems

Intheproblemsbelowitisassumedthat C istherandomwalkgenerating matrixdefinedin(1.4),whereas M = IT ee′ ∕T and e =(1,..., 1)′ ∶ T × 1. Wealsoassumethat {��j }, {��k }∼ NID(0, 1) and

1.1 Showthattheeigenvaluesof (C ′ C ) 1 are1∕��j (j = 1,..., T ).

1.2 Showthatthenonzeroeigenvaluesof C ′ MC are

1.3 Provethat

1.4 Derivethefollowingexpansion:

1.5 Derivethefollowingexpansion:

1ModelsforNonstationarityandNoninvertibility

1.6 UsingtheasymptoticexpansionsobtainedinProblems1.4and1.5,computecumulantsgivenin(1.18).

1.7 Showthatthedistributionwiththec.f. ��1 (�� )=(cos √2i�� ) 1∕2 hasthe jth ordercumulantgivenby �� (j) 1 = (j 1)!23j 2 (22j 1) (2j)! Bj , where Bj istheBernoullinumber.Showalsothatthedistributionwiththe c.f. ��2 (�� )=(

jthordercumulantgivenby

1.8 UsingthefactthattheinverseLaplacetransformof e c√�� is c exp(−c2 ∕ (4x))∕(2√�� x3 ) for c > 0,showthat

, where {��n }∼ NID(0, 1) and Φ isthedistributionfunctionofN(0, 1).

1.2ATestStatisticfromaNoninvertibleMoving AverageModel

Letusnextconsiderthefirst-ordermovingaverage[MA(1)]model:

,

where ��0 ,��1 ,...,areNID(0,�� 2 ) randomvariables.Theparameter �� isrestricted tobe |�� | ≤ 1becauseoftheidentifiabilitycondition.TheMA(1)model(1.33) issaidtobe noninvertible when |�� | = 1.Variousinferenceproblemsassociated withthenoninvertiblecasewillbediscussedinChapter7.

LetusconsiderheretestingiftheMA(1)modelisnoninvertible,thatis, totest

H0 ∶ �� = 1versus H1 ∶ ��< 1

ForthispurposeweconductascoreorLagrangemultiplier(LM)typetest.The log-likelihood L(��,�� 2 ) for y =(y1 ,..., yT )′ isgivenby L

Letusput Ω=Ω(1).Itisnoticedthatthematrix Ω isexactlythesame,except forsize,as G 1 ∗ givenin(1.8).Forlaterpurposeswealsonotethat

[min(j, k )− jk

,k =1,...,T , where C istherandomwalkgeneratingmatrixdefinedin(1.4),whereas e =(1,..., 1)′ , e1 =(1, 0,..., 0)′ , eT =(0,...,

Thenthemaximumlikelihoodestimators(MLEs)of �� and �� 2 under H0 are ̂�� = 1and ̂�� 2 = y′ Ω 1 y∕T .Itcanbecheckedthat

wherewehaveusedtheformula

Theseyield

wherewehaveusedthefactsthattr(Ω 1 )= T (T + 2)∕6,whichfollowsfrom (1.34),and

TheLMtestconsideredhererejects H0 ifthesecondderivativeofthe log-likelihoodunder H0 islarge,thatis,if

(1.35) takeslargevalues.

Thelimitingdistributionof ST under H0 canbederivedbytheeigenvalue approachasfollows.Put ��

where ��j isthe jthlargesteigenvalueof ٠1 ,whichcanbegiven,from(1.10),by

Itcannowbeseenthatthelimitingdistributionof ST isthesameasthatof ST 2 discussedinthelastsection.Thusitholdsthat,as T → ∞ under H0 , ST = 1 T y′ Ω 2 y y′ Ω 1

Thestochasticprocessapproachdealswith y′ Ω 2 y∕(T 2 �� 2 ) inthefollowing way.Definingarandomwalk z = C �� =(z1 ,..., zT )′ with �� ∼ N(0, IT ),wehave

where RT istheremaindertermof op (1) (Problem2.1).Thus,itfollowsfrom (1.21)that

Another random document with no related content on Scribd:

The Project Gutenberg eBook of The martyrdom of Nurse Cavell

This ebook is for the use of anyone anywhere in the United States and most other parts of the world at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this ebook or online at www.gutenberg.org. If you are not located in the United States, you will have to check the laws of the country where you are located before using this eBook.

Title: The martyrdom of Nurse Cavell

Author: William Thomson Hill

Release date: May 22, 2022 [eBook #68149]

Language: English

Original publication: United Kingdom: Hutchinson & Co, 1915

Credits: Tim Lindell and the Online Distributed Proofreading Team at https://www.pgdp.net (This file was produced from images generously made available by The Internet Archive/American Libraries.)

*** START OF THE PROJECT GUTENBERG EBOOK THE MARTYRDOM OF NURSE CAVELL ***

The Martyrdom of Nurse Cavell.

The Life Story of the Victim of Germany’s Most Barbarous Crime.

WITH ILLUSTRATIONS.

LONDON: HUTCHINSON & CO. PATERNOSTER ROW ᛭ 1915

NURSE CAVELL’S

LAST MESSAGE TO THE WORLD.

“But this I would say, standing as I do in view of God and eternity, I realise that patriotism is not enough. I must have no hatred or bitterness to anyone.”

LIST OF ILLUSTRATIONS.

The latest portrait of Nurse Cavell. (From Photo Copyright Farringdon Photo Company) (see cover)

Nurse Cavell (Photo by courtesy of Illustrated London News) Frontispiece

The Rev. Frederick Cavell, father of Nurse Cavell. (Daily Mirror Photograph) facing page 16 Mrs. Cavell, mother of Nurse Cavell. (Daily Mirror Photograph) ” 17

Nurse Cavell when a child, with her mother and elder sister. (Photo Copyright Farringdon Photo Company) ” 32

The Rectory, Swardeston, where Nurse Cavell was born. (Daily Mirror Photograph) ” 33 Nurse Cavell in her garden. (Daily Mirror Photograph) ” 48

Nurse Cavell, from a photograph taken in Brussels. (Photo Copyright Farringdon Photo Company) ” 49

CHAPTER I. CHILDHOOD.

In the early seventies there were living at the country rectory of Swardeston, near Norwich, a clergyman and his wife and little family. There was a “New” and an “Old” Rectory. Both are still standing, much as they were then, except that the trees are older, and the “New” Rectory has long ago lost any signs of newness. It is one of the ways of Old England to call some of its most ancient things New, as if it could never learn to tolerate change kindly, even after centuries of wont.

There is a Newtimber Place in Sussex whose walls were built before the Armada. There is a New Building in Peterborough Cathedral which was completed before the Reformation. New Shoreham took the place of Old Shoreham before Magna Charta was signed.

The Rector, the Rev. Frederick Cavell, lived with his family at the New Rectory. It is a pleasant sunny house with a large garden. Such parsonages are common in all the unspoiled rural parts of England. A little gate leads to the churchyard close by.

In a great city no man would live willingly close by a cemetery. In such a village as Swardeston the nearness of the graveyard is a consecration. New graves appear among the old ones from time to time. The oldest of these others have faded gently into the grass. Nobody is left to tend them or to remember whose bones they cover. Yet the history of many a family can be traced back for three centuries on the lichen-covered stones.

Some day, when the war is over, another grave may be dug in this quiet spot. If the poor mutilated frame of Edith Cavell is ever permitted to be brought back home, her countrymen will come here to look upon the place where she lies. In this October of 1915 she sleeps in a land ravaged by war, and those who killed her will not stoop even to the tardy pity of giving back her body.

But in those early seventies the village churchyard was not a place of sadness to the Rectory children. They played hide-and-seek among the sloping tombstones. The church and churchyard were, as they still are, the centre of the village life. Gay doings, such as a wedding, took place under the shadows of the elms and yews.

The whole community assembled there on any day of special interest. The churchyard was the Trafalgar Square of Swardeston. For it was not remote from the houses, as many village churchyards are. Norfolk labourers swung their heels on the wall in the long evenings of the days before village institutes and reading rooms were invented.

In these early seventies the village talk still harked back sometimes to the War of the French and Prussians. Its politics dealt with such names as “Dizzy” and Gladstone and Joseph Arch, the agricultural reformer—and, what was more to the point, a Norfolk man. In later years the village church saw the celebrations of Queen Victoria’s two Jubilees and King Edward’s Coronation—“a Norfolk landlord, and a rare good ’un,” as they liked to say in Swardeston.

CHAPTER II

LIFE IN THE RECTORY.

Home life in the Rectory was tinged, as was that of most English homes at the time, with Evangelical strictness. On Sunday all books, needlework, and toys were put away. The day began with the learning of collect or Catechism. As soon as the children were big enough they attended services in the morning and afternoon.

Evening services were not yet introduced in Swardeston. Light was not cheap, and the way across the country fields to church was no adventure for Sabbath clothes on dark winter nights. Thus the closing hours on Sunday were home hours for Rectory and village. Let those who have no memories of such times scoff if they think fit. A memory is better than a jest.

Edith Cavell’s father was Rector of this parish for more than fifty years. He is dead now, but the villagers remember him well. His portrait shows him with a mouth and chin of unusual firmness. His eyes are kindly, but there is little sense of humour about them. It is notably the face of an upright man. Surely capable of sternness, he would be just to the point of inexorableness unless his face belies him. A sense of duty is implicit in every line; and we have the best of reasons for knowing that he transmitted this part of his character to his daughter Edith.

“The clever Miss Cavell” she was called in later years when she worked at a London hospital; but a more dominant characteristic was a rigid insistence upon what she deemed to be right. This was the constant theme of the father’s sermons to his village flock. He would not hesitate to reproach from the pulpit any member of the

congregation, whatever his station, whom he considered guilty of grave fault.

The mother (who is now eighty years old, and lives very quietly at Norwich) brought a gentler influence to bear upon the Rectory life. There is a picture of her with two of her little girls. The mother wears the wide flounces which to-day are among the earliest memories of the “Men of Forty.” Flounces that were a protection and a promise. Something for little hands to cling to when the legs were not yet sure of their way. These flounces made a royal road from earth to the children’s heaven. The grown-up world far out of reach was always within call of a pull at the ample skirts.

Mrs. Cavell was a happy mother, and her children were happy too. So early as the days we are speaking of her eyes had something wistful in them. It was almost as if some inner consciousness had told her then of the distant, poignant future.

So the family grew up in a contented, well-ordered home, with plenty of outdoor games and sunshine, such as country children have. Long afterwards, in the midst of London slums, Edith Cavell would talk of the ripening blackberries far away in the Norfolk lanes, and of the great jam-making times which followed.

CHAPTER III.

WORK IN LONDON.

Like Charlotte Brontë, another vicar’s daughter, Edith Cavell first learned something of the wider world in a Brussels school. It was commoner then than now—meaning by “now” before the war—for English girls to be sent to Belgium to school. Charlotte Brontë’s Brussels life has left us at least one imperishable book. Edith Cavell has left no written memorials of those times; but if we would reconstruct her life we may imagine some such background as that of “Villette”: the strangeness of a foreign city, fascinating by its novelty yet repelling by alien atmosphere.

The lot of a school-girl is not too happy at the best among new companions. When their language and ways are those of a foreign country they can become a source of torture to a sensitive child. Some of these school-girl irritations Edith Cavell had to bear; yet such early annoyances evidently left little mark on her, for she returned many years later to Brussels of her own free will, and conquered the affections of the Belgians a second time.

Edith Cavell’s early womanhood was spent in London—at the London Hospital, the St. Pancras Infirmary, and the Shoreditch Infirmary in Hoxton. Her training was obtained at the London Hospital, the great institution in the Whitechapel Road which is now nursing many wounded soldiers. The women who train in this hospital pass through a hard school. All hospital nurses work hard, but the nurses who come from “The London” think they know more of the strain of their calling than any others.

“The London” proposes to raise a memorial to Nurse Cavell. It is their right and hers that this should be done. For “The London” gave

her the thorough training which enabled her to become the skilful teacher of others, and to instruct the nurses who should succour with equal care the wounded of all nations.

At the end of her arduous training at the London Hospital in 1896, Miss Cavell went to St. Pancras Infirmary as Night Superintendent. She stayed there for a little more than three years. Then she became Assistant Matron at the Shoreditch Infirmary in Hoxton. She left Hoxton in 1906 to start the work in Brussels which ended only with her cruel death.

Including the training years at the London Hospital, Edith Cavell had given twenty-two years to nursing the sick. She was twenty-one years old when she began this work. She was forty-three when she met her death. Thus she had given up the best years of a woman’s life without a break, save for the occasional precious holidays, of which we shall say a word presently.

The work in London was one of unvarying routine in the most dismal surroundings. Nothing but a real devotion to the task could have made the monotony tolerable.

The writer asked one of those who worked with her for part of this time what was the reason that decided Edith Cavell to become a nurse. “She felt it was her vocation,” was the simple answer; “isn’t that enough?” The vocation, in these great London infirmaries, consisted in preserving a cheerful face day in and day out; in ruling, with kindness but also with firmness and an unfaltering tact, old men and women, children from the poorest slums; in being constantly in contact with pain and suffering and in the near presence of death. Those who remember her work in London—and they are very many —speak of her unselfishness and of a shy pride about the details of her labours.

What she did for her patients she liked to be a secret between herself and them. She would follow up the “cases” to their homes. The Matron and her fellow-nurses guessed some of these acts of week-day holiness; but Nurse Cavell never spoke of them. She went about doing good among the neat beds of the wards and in the unlovely surroundings of the neighbouring streets, doubtless thinking

sometimes of the Norfolk village where the sun was shining beyond the fog, yet never letting the patients see that she had any thoughts except for them.

But with this sympathy went a rare strength of mind. Her name “Clever Miss Cavell” was not used in envy. It was a simple recognition of the fact that she had what is called a capable brain. She always knew what to do in a difficult situation. A fellow-nurse in trouble was always advised to consult Miss Cavell.

CHAPTER IV.

UPHILL WORK IN BRUSSELS.

Edith Cavell needed all her strength of character in her first years in Brussels. When she went there nine years ago as Matron of a Surgical and Medical Home, English nursing methods were not appreciated on the Continent as they are now. Nursing was regarded as one of the functions of the Church. Miss Cavell was a Protestant as well as a foreigner. She was felt to be a rival of the nuns and sisters working under religious vows.

The authorities of the Catholic Church looked coldly upon an enterprise which, from their point of view, had an aspect of irreligion and freethinking. But it was not long before the Matron’s efficiency and tact carried the day. A well-known priest trusted himself to the English lady. His tribute to her devotion and skill brought public opinion to her side. In 1909 she established a training home for nurses. The authorities recognised and encouraged her; and shortly before the outbreak of war she was provided with a modern and well-equipped building.

The first warning of the war came when she was spending a holiday at home with her mother at Norwich. During these years in Brussels two holidays a year had been spent in England. They were happy halting places in a rough journey. What made them so pleasant to Edith Cavell was that she could spend them with her mother.

The love of the younger woman for the old was one of the most beautiful aspects of her character. “People may look upon me as a lonely old maid,” she said once to a friend; “but with a mother like mine to look after, and, in addition, my work in the world which I love,

I am such a happy old maid that everyone would feel envious of me if they only knew.”

That was her secret—her love for her mother and her work. It was that which enabled her to look upon the world as a beautiful garden, where there was always something to do for sickly plants. The real flowers, and the care of them which could only be given in English holidays, were almost a passion to her from the earliest Rectory days.

Her success as a nurse, both in Brussels and the slums of London, owed three-parts of its efficacy to her overflowing sympathy “It was her gentle way,” said an old patient, “that did most to make me well again; I felt she was a minister of God working for my good.” And there are wounded British soldiers who have pressed the doctors to send them back quickly to the firing line. “We will go back willingly,” they say, “to avenge this great woman’s death.”

Daily Mirror Photograph.

THE REV. FREDERICK CAVELL, FATHER OF NURSE CAVELL.

Daily Mirror Photograph.

MRS. CAVELL, MOTHER OF NURSE CAVELL.

Every holiday in England was spent with the aged mother, who looked forward to these meetings as much as the daughter. Without warning, the war broke into the last of these holidays in the full summer of 1914. Edith Cavell made her mind up promptly. Her

holiday was not yet over, but she hurried back at once. “My duty is out there,” she said; “I shall be wanted.”

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.