Scribd
Upload
238 views5 pages

Data Prep - Homogeneity (Excel)

Homogeneity, or the assumption that a time series sample is drawn from a stable/homogeneous process. In this document, we'll start by defining the homogeneous stochastic process and stating the minimum requirements for time series analysis. Then we demonstrate how to examine the sample data, draw a few observations, and highlight some underlying intuitions behind them. For the example spreadsheet and/or tutorial video, visit us at: http://bitly.com/IkJpyI

Uploaded by

NumXL Pro
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
238 views5 pages

Data Prep - Homogeneity (Excel)

Homogeneity, or the assumption that a time series sample is drawn from a stable/homogeneous process. In this document, we'll start by defining the homogeneous stochastic process and stating the minimum requirements for time series analysis. Then we demonstrate how to examine the sample data, draw a few observations, and highlight some underlying intuitions behind them. For the example spreadsheet and/or tutorial video, visit us at: http://bitly.com/IkJpyI

Uploaded by

NumXL Pro
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

Homogeneity

Inthisissue,thethirdtutorialinourdatapreparationseries,wewilltouchonthethirdmostimportant assumptionintimeseriesanalysis:Homogeneity,ortheassumptionthatatimeseriessampleisdrawn fromastable/homogeneousprocess. Wellstartbydefiningthehomogeneousstochasticprocessandstatingtheminimumstationary requirementsforourtimeseriesanalysis.Thenwedemonstratehowtoexaminethesampledata,draw afewobservations,andhighlightsomeunderlyingintuitionsbehindthem.

Background
Instatistics,homogeneityisusedtodescribethestatisticalpropertiesofaparticulardataset.In essence,itstatesthatstatisticalpropertiesofanypartofanoveralldatasetarethesameasanyother part. Whatdowemeanbystatisticalproperties?Astrictwayoflookingathomogeneitywouldinvolve examiningthechangestothewholeofthemarginaldistribution,buttimeseriesanalysisonlydemands thatweconsiderthelocationstabilityovertime(versustrend)andthestabilityoflocalfluctuationover time.

What does this mean?


Intimeseriesanalysis,weareconcernedwiththestabilityoftheunderlyingstochasticprocessover time.Dowehavestructuralchanges?Ifchangesexistbutgoundetected,wefindourselvesinoneof severaldifficultsituations: 1. Theproposedmodelofferslittleexplanationforthedatavariationovertime 2. Themodelsparametervaluesvarysignificantlywhenwerecalibrateusingeitherasubsetof thesample,orbyincorporatingnewobservations 3. Inextremecases,theselectionofthebestmodeltypeororder(s)canbeinfluencedbythe selectionofsampledata

Why do we care?
Theobjectiveoftimeseriesanalysisandmodelingisusuallytheconstructionofoutofsampleforecasts. Howcanwegeneratetheseforecastsusingamodelwithtimevaryingparameters?Howmuch confidencecanweputinthoseforecasts?Aretheforecastrobust?Letsfindout.

Why does it happen?


Thereareseveralcausesforheterogeneity(oppositeofhomogeneity)inatimeseries: (1) Theunderlyingmodelsstatisticalpropertiesareevolvingovertime.Inthiscase,tryingtofita modelwithfixedparametervalueswouldnotbeoptimal,despiteourbestefforts.Weneedto DataPreparationHomogeneity 1 SpiderFinancialCorp,2012

examineadvancedmodelingtechniquestocapturethedynamicsofthestatisticalpropertiesof theprocess.This,unfortunately,isoutsidethescopeofthispaper. (2) Theunderlyingprocessisnotstationary(e.g.possessestrendovertime). (3) Theunderlyingprocessisheteroskedasticwherevolatilityexhibitsclusteringandmean reversion. (4) Theunderlyingprocesshadundergonefewbutmajorstructuralchangesduetoexogenous events,suchasthepassingandenforcementofnewrelevantlawsoramajordevelopmentin theprocessitself. Example I: OzonelevelindowntownLosAngelescase(refertotheHowdoesitfitissue)

Throughoutthesampletimebetween1955and1972,thereweretwomajordevelopments: (1) Rule#76forgasolinemixandcombustionenginedesign (2) OpeningofafreewaytodiverttrafficfromdowntownLA Obviously,thoseexogenouseventsaffectthenumberofcarsindowntownLA,andconsequently theamountofOzoneemittedinthearea.Onecanarguethattheprocessafterthoseevents(1972) isnotthesameastheprocessin1960. Example II: USConsumerPriceIndexanditsderivative,theinflationrate: TheinflationrateintheUSreflectstheeffectivenessofgovernmentpublicpolicies,sothroughout thesamplehorizonbetween1913and2009,itisnosurprisethatthedatacharacteristicsbeforeand afterWorldWarIIarefundamentallydifferent.Alsoconsiderthatinthe1970s,thesuddenrisein inflationevidentinourdatareflectsafundamentalchange(orfailure)inpublicpolicy.

DataPreparationHomogeneity

SpiderFinancialCorp,2012

Mostimportantly,theinflationrateunderlyingtheprocessafterthe1970sisverydifferentthanin prioryears,foranumberofreasons:(1)fundamentalchangesinpublicpoliciesand(2)amandate fortheFederalReservetofighttheinflationrateandunemploymentin1977.

Insum,onemayarguethatthepost1977processisverydifferentfromthepre1977process. Conclusion Theinvestigatormustbringrichpriorknowledgeandstronghypothesesabouttheunderlyingprocess structureanditsdriverstohisinterpretationofadataset.Theliabilityofpowerfulanalyticalmethodsis

DataPreparationHomogeneity

SpiderFinancialCorp,2012

thepotentialforarichdiversityofalternativesolutionsthatcanhaveverydifferentpropertieswhen extrapolatedfromthesituationfromwhichthedatawasoriginallysampled.

Checking For Homogeneity


Theinitialstagesintheanalysisofatimeseriesmayinvolveplottingvaluesagainsttimetoexaminethe homogeneityoftheseriesinvariousways:namely,stabilityacrosstime(asopposedtoatrend)and stabilityoflocalfluctuationsovertime. Inastatisticalsense,atestforhomogeneityisequivalenttoatestofastatisticaldistribution.Inplain English,wewishtodetectachangeintheunderlyingdistribution.Forthat,wecanexaminethe distributionmoments:mean,variance,skew,andkurtosisforchanges. Fortimeseriesanalysis,wewilllookintothe1sttwomoments:meanandvariance,andexamineany shiftovertime.Herearefewteststoaidus: StandardNormalHomogeneityTest(SNHT): Q:Dowehaveashiftinthemeanorvariance?

H o : r ~ N (0,1) H1 : Thereisashift

Where r arethestandardizedratios(anobservationsvaluecomparedtotheaverage). PettittsTestdetectingashiftinvarianceNonparametrictest(i.e.noassumptionaboutthe distributionofdata). Q:Dowehaveashiftinthevariance?When? ThePettitt'stestisanadaptationoftherankbasedMannWhitneytest,whichallowsyouto identifythetimeatwhichtheshiftoccurs. TestsfordetectingashiftinthemeanNonparametrictest(i.e.noassumptionaboutthe distributionofdata). Q:Dowehaveashiftinthemean?When? Where o

H o : t c H 1 : k c

H o isthenullhypothesis,whichstatesthat xt followsoneormoredistributionsthat
havethesamemean.

DataPreparationHomogeneity

SpiderFinancialCorp,2012

H1 isthealternativehypothesis,whichstatesthatthereexistsatimekfromwhichthe

variableschangemean. BartlesTest(rankedversionofVonNeumannratiotest)forrandomness Q:Isthesampledatarandom?Dowehavepatterns? o o NullHypothesis( H o ):timeseriesishomogeneous. Alternativehypothesis( H1 ):timeseriesisnothomogeneous.

Hold on, doesnt homogeneity sound a look like stationarity?


Stationarityandhomogeneityarecloselyrelated;stationaritylooksintothestabilityofthejoint distribution FX ( xt1 , xt2 ,..., xtN ) ,whilehomogeneityexaminesthestabilityofthewholemarginal distributionovertime. Anonstationarytimeseriesisnonhomogeneous,buttheoppositemaynotalwaysbetrue.

My time series is not homogeneous over time; what can I do?


Ifahomogeneousassumptionfailstohold,weneedtotakeacloserlookandunderstandthetime series: (1) Isthetimeseriesstationary?Ifso,transformthedatatobringittostationarity. (2) Identifyandunderstandthedriversoftheunderlyingprocess: a. Dowehaveexogenousdrivers/factors(e.g.laws,events,etc.)thatcouldaffectthe valuesoftheobservations? b. Hastheunderlyingprocesschangedpermanentlyovertime? c. Doweexpecttheexogenousfactortochangeagaininthefuture? d. Whendidtheprocessmeanorvariancechange? IntheUSCPIexample,thechangemadein1977bycongresstomandatetheFederalReservetoadopt publicpolicytocontrolinflationisamajorturningpoint,andweareinclinedtoconcludethatprocess underwentapermanentchangeasaresultofthatdevelopment.Inthiscase,Iwoulddisregardall observationbeforethattime. IntheOzonelevelindowntownLAexample,theopeningofafreewaydivertingtrafficfromdowntown isastructuralchangeintheunderlyingprocess.Thesamecanbesaidaboutthelawsforgasolinemix andenginedesign.Again,Iwoulddisregarddatabeforethechangestookeffect,andonlyconcern myselfwithobservationsthatoccuraftertheseevents.

DataPreparationHomogeneity

SpiderFinancialCorp,2012

You might also like