BackgroundDefectsofPresentMethodsInchina,themostrecentvaluesarecomparedwiththecorrespondingvaluesayearagotoobtainatimeseriesfreeofseasonalfluctuations.However,anapproachbasedonyear-on-yearcomparisonshasfollowingdefects:1.Influencedbytheseasonalfactorsofthebaseperiod.2.Turningpointswouldalwaysbedelayed.3.Seasonalfactorswouldnotbeeliminatedthoroughly.4.AccurateseasonalmodecannotbemeasuredMeasuringMethods1.TRAMO/SEATS(model-based)[SeeBjörnFISCHER(1995)]2.X-12-ARIMA(empirical-based)[SeeD.F.Findleyetc(1998)]EmpiricalResearchMainPricesIndexesCPI-ResidentsConsumerPriceIndexRCPI-RetailCommodityPriceindexesPPIR-PurchasingPriceIndexofRawMaterials,FuelsandPowerPPI-ProducerPriceIndexofIndustryProductsOriginalData:ChinaEconomicInformationNetworkDatatransformationLengthoftimeseries&Baseyear“Increasingthelengthofthetimeserieswillgenerallyincreasetheaccuracyofestimation,butnotnecessarilyindirectproportiontotheamountofdata.Oneoftheproblemswithparticularlylongtimeseriesisthatfrequentlythepatternchangeswithtimeandusingalongertimespanmayinfactgivelessaccurateresultsthanusingonlytheminimumnumberofdatanecessary.”——seeDerekBlades(2001)ExaminethepresenceofseasonalitybyX-12-ARIMASelectingRules:1.LargerFs-valueintestforthepresenceofseasonality2.Easytocomparehighly-relatedindexes3.Lesspresenceofmovingseasonality1996asbaseyearforCPI&RCPI(132months)1998asbaseyearforPPI&PPIR(108months)ComparisonbetweentwomethodsTestforseasonalityCorrelationtestGraphComparison(1)GraphComparison(2)ComparisonofseasonalmodesamongmainpricesindexesRangeofseasonalfluctuation(SeeGraph)CorrelationcoefficientoffinalseasonalfactorsFinalSeasonalFactor/ComponentPerMonthDiagnosticIdempotencycriterionIfisthefilteroutputattimetafterusingtheseasonaladjustmentfilterforthefirsttimeandisthefilteroutputattimetafterusingthefilterforasecondtime,Misnumberoftimeseriesdata,ameasureisthemeanabsolutepercentagedifference(MAPD)SlidingspansTherearefourestimatesSi,j(1),…,Si,j(4)fortheseasonalfactorofXi,jint