FORECASTING MONTHLY UNEMPLOYMENT BY ECONOMETRIC SMOOTHING TECHNIQUES

Transcription

1 Professor Vergil VOINEAGU, PhD Presiden of Naional Insiue of Saisics, Romania Silvia PISICA, PhD Naional Insiue of Saisics, Romania Nicolea CARAGEA, PhD Naional Insiue of Saisics, Romania FORECASTING MONTHL UNEMPLOMENT B ECONOMETRIC SMOOTHING TECHNIQUES Absrac. In he las year, he need for high-qualiy saisics for labour policy became increasingly significan. Requiremens of European official saisics for shorerm indicaors on labour marke saisics creaed he opporuniy o develop a naional mehodology for deriving monhly unemploymen esimaes direcly from he Labour Force Survey (LFS) resuls. LFS is carried ou by Romanian Naional Insiue of Saisics on quarerly basis, according o he European regulaions and mehodology. The unemploymen is measured in line wih Inernaional Labour Organizaion crieria, ensuring full comparabiliy a European level. In response o his requiremen Romanian Naional Insiue of Saisics has developed such a mehodology ha generaes he forecas of monhly unemploymen rae using daa from he series of Labour Force Surveys. The journey for reaching he presen model was a complex and difficul one. The main par of he sudy is aiming o presen comprehensively he work performed for developing he acual mehod used for regularly producing he monhly unemploymen esimaes. The oupus of he mehod are presened in he second par of he paper showing he usefulness brough by he model. I also emphasizes he seasonaliy of he ime series suggesing an improvemen of he Hol mehod. Main unme expecaions of he mehodology are closing he presen paper. Key words: unemploymen, labour force survey, esimaion, forecasing JEL Classificaion: C, C, J. Inroducion A new approach o forecasing he rend and seasonaliy in ime series daa is presened. I is based on he Labour Force Survey resuls using he relaionship beween he hisorical and forecased daa. Concerning his approach, hree forecasing daa series were produced in order o smooh he values of daa series and also o

2 Vergil Voineagu, Silvia Pisica, Nicolea Caragea exhibi fairly good accuracy. The preliminary resuls show ha for he rend and seasonal daa, he proposed mehods produce beer forecass han he oher mehods esed.. Lieraure Review A ime series is a sequence of observaions which are ordered chronologically. In he collecion of daa aken over ime i is inheren o capure some forms of random variaion. Special mehods for reducing he random variaion effecs are used, of which "smoohing" echniques mos widely. These mehods, when properly applied, reflec more clearly he underlying rends. A he end of 0's, Hol and Winers proposed a mehod for smoohing and forecasing ime series wih locally linear rend (Winers, 0) nowadays referred usually as Hol mehod. They employed he exponenially weighed moving average approach o esimae he level and newly also he slope of he ime series. Several smoohing echniques were also developed and described by Jelinek and Mercer (0), Kaz (), Bell, Cleary, and Wien (0), Ney, Essen, and Kneser (), and Kneser and Ney (). The main mehods of ime series smoohing are presened in Box. Box. The main smoohing echniques The simples smoohing echnique is o ake an average of pas values. Unforunaely, his also compleely obscures any rends, changes, or cycles wihin he daa. More complicaed averages eliminae some bu no all of his obscuring and sill end o lag as forecasers, no responding o changes in rends unil several observaions afer he rend has changed. Moving Averages: Moving averages rank among he mos popular echniques for he pre-processing of ime series. They are used o filer random "whie noise" from he daa, o make he ime series smooher or even o emphasize cerain informaional componens conained in he ime series. In oher words, he moving average echnique only uses he mos recen observaions or a weighed average ha values some observaions more han ohers. Exponenial smoohing represens an aemp o improve upon hese defecs. Exponenial Smoohing: This is a very popular scheme o produce a smoohed Time Series. Whereas in Moving Averages he pas observaions are weighed equally, Exponenial Smoohing assigns exponenially decreasing weighs as he observaion ge older. In oher words, recen observaions are given relaively more weigh in forecasing han he older observaions. The benefi of simple exponenial smoohing is ha i allows for a rend in how he smoohed daa is changing. I does poorly, however, a separaing changes in he rend from he random variaions inheren o he daa. For ha reason, double and riple exponenial smoohing are also used, inroducing addiional consans and more complicaed recursions in order o accoun for rend and cyclical change in he daa.

3 Forecasing Monhly Unemploymen by Economeric Smoohing Techniques Double exponenial smoohing - also known as Hol's Linear Exponenial Smoohing - is a refinemen of he popular simple exponenial smoohing model bu adds anoher componen which akes ino accoun any rend in he daa. Simple exponenial smoohing models work bes wih daa where here are no rend or seasonaliy componens o he daa. When he daa exhibis eiher an increasing or decreasing rend over ime, simple exponenial smoohing forecass end o lag behind observaions. Hol mehod has achieved a broad populariy among forecasers due o is simpliciy and good performance. Several modificaions of his original mehod have appeared in lieraure and are used in pracice (T. Hanzak, 00).. Descripion of he economeric model o forecas he monhly unemploymen The saring poin of he monhly unemploymen smoohing and forecas in he Romania s case is he Hol mehod, using wo differen parameers and allowing o apply he forecasing for series wih rend. In classical model, he sandard Hol s mehod is given by: y ( )( b ) b b where: - is he smoohed level of he series, compued afer y is observed; b - is he smoohed rend a he end of period ; y b 0 - are he smoohing parameers. Hol s wo parameers exponenial smoohing mehod is an exension of simple exponenial smoohing. I adds a growh facor (or rend facor) o he smoohing equaion as a way of adjusing for he rend. The iniializaion process for Hol s linear exponenial smoohing requires wo esimaes: One o ge he second smoohed value for he level of he series; The oher o ge he rend b. Sar values used o iniialize he model are he followings: y b y y. Forecas m periods ino he fuure are given by: F m mb

4 Vergil Voineagu, Silvia Pisica, Nicolea Caragea where: m - represens he periods o be forecas ino he fuure. Using an improved Hol mehod, i was produced he firs esimaor for he unemploymen provided by Labour Force Survey, considered he reference daa. The reference ime series are obained dividing he iniial quarerly sample of LFS in hree sub-samples, one for each monh of he quarer and each monhly sub-sample is grossed-up separaely, applying a reduced calibraion scheme. The esimaor is given by: y ( b ) b b where: - is he smoohed level of he series, compued afer y is observed; b - is he smoohed rend a he end of period ; y b 0 - are he smoohing parameers. In his case, he iniializaion process for Hol s linear exponenial smoohing requires wo esimaes: One o ge he firs smoohed value for he level of he series; The oher o ge he rend b. Sar values used o iniialize he model are he followings: y and b 0. Also assuming: y. The firs equaion correcs he level of ime series wih he las wo rend values b -. The improvemen of he classical mehod helps o deermine he prediced values for he end of he quarer on he basis of he real values regisered a begins of he quarer. Smoohing coefficiens 0, are deermined by minimizing a measure of forecas error such as MSE (Minimum Square Errors): T T ˆ e min T 0 T 0 The forecas error is given by: e. ˆ

5 Forecasing Monhly Unemploymen by Economeric Smoohing Techniques In order o asses he qualiy of he esimaions of he proposed mehod, he resuls were compared wih he benchmarked daa series, given by equaion: where: - represens he benchmarked daa series; Ŷ b Q - represens he quarerly average of he reference daa series (LFS quarerly values provided from he sample); b Q / Ŷ ˆ Q - represens he quarerly average of he Hol s esimaion series. The predicion is given by: ' Ŷ Forecas h periods ino he fuure are given by: ' h Ŷ h Q h where: h - are he periods o be forecas ino he fuure. On he base of his algorihm, he forecas for he firs nex period (h=) is calculaed as following: ' Ŷ. The resuls obained consis in monhly series of final esimaes in non-seasonal adjused form. Beside he monhly series (number of unemployed persons and unemploymen rae) adjused for level and rend (non-seasonally adjused series), seasonally adjused (hus, removing he effec of seasonal variaions) and rend (which represen he series from which, boh, he seasonal and irregular effecs, were removed) series are also calculaed. Seasonal adjusmen was performed using DEMETRA sofware (TRAMO/SEARTS mehod), which also esimaes he calendar effec (Orhodox Easer, leap year and oher naional holidays) and idenifies and correcs he ouliers (occasional ransiory or permanen changes in level). The esimaion of unobserved componens: rend-cycle, seasonal and irregular componen was made wih SEATS program based on ARIMA models. Seasonally adjused series were obained by removing he seasonal componen from he original daa. Trend was obained by removing he irregular componen from he seasonally adjused series. Benchmarked daa series are he original esimaes obained regularly (i.e. quarerly from he enire sample) from he Romanian LFS

6 Vergil Voineagu, Silvia Pisica, Nicolea Caragea. Applicaion of he model The model was applied on he labour marke daa provided by LFS for he period To gross-up he daa for each monh, he following variables are used for calibraion: populaion by sex and four age groups (less han years, -, -, and over), area of residence (urban/rural), regions (NUTS level) and oal number of individual households. Beside he monhly series (number of unemployed persons and unemploymen rae) adjused for level and rend (non-seasonally adjused series), seasonally adjused (hus, removing he effec of seasonal variaions) and rend (which represen he series from which, boh, he seasonal and irregular effecs, were removed) series are also calculaed. Thus, monhly series are buil and disseminaed in hree forms: nonseasonally adjused, seasonally adjused, and rend. An exemplificaion of hese daa series is shown in he figures a and b. Figure. a. Unemploymen rae (- years old).... originale SA_r.. Jan-0 Jan-0 Jan-0 Jan-0 Jan-0 Jan-0 Jan-. Figure. b. Unemploymen rae (- years old).. originale SA_r.. Jan-0 Jan-0 Jan-0 Jan-0 Jan-0 Jan-0 Jan- Source: Auhors calculaion based on LFS daa, Naional Insiue of Saisics, 0.

7 Forecasing Monhly Unemploymen by Economeric Smoohing Techniques Because of he small number of cases of observaion, he reliabiliy of esimaes for he indicaors corresponding o he caegory of young people (age group - years) is exremely low, he series obained showing a high degree of volailiy. Therefore, he monhly esimaes for youh caegory are no disseminaed. The firs release of daa was performed on s of Augus 0. Negaive impac of recession on employmen The unemploymen rae is an imporan indicaor wih boh social and economic dimensions. From an economic perspecive, unemploymen could show unused available labour. Increasing unemploymen may be also refleced in loss of income for individuals and rising pressure on governmen spending on social benefis. The unemploymen rae according o he Inernaional Labor Organizaion (ILO) definiion is he mos widely used labor marke indicaor because of is inernaional comparabiliy. ILO unemployed are he persons aged - years who simulaneously mee he following hree condiions: (i) have no job, (ii) are available o sar working in nex wo weeks, (iii) were acively seeking a job a any ime during he las four weeks. ILO unemploymen rae is he proporion of ILO unemployed in he oal economically acive populaion. The official saisics in Romania calculae and disseminae hree unemploymen raes: (i) regisered unemploymen wih monhly periodiciy, a NUTS (couny) level, defined according o naional legislaion, provided by adminisraive sources Naional Agency for Employmen and Labour Force; (ii) ILO unemploymen (according o Inernaional Labour Office crieria) wih quarerly periodiciy (and as annual average), a NUTS (region) level, defined according o Inernaional Labour Office (ILO) crieria, provided by saisical sources Labour Force Survey (LFS), Naional Insiue of Saisics (INS); (iii) monhly ILO unemploymen - wih monhly periodiciy, a NUTS 0 (counry) level, defined according o ILO crieria, provided by saisical sources LFS and esimaion model. In order o provide inpu for shor-erm employmen policy, he European Communiy Saisical Office (EUROSTAT) requesed Member Saes o develop heir own models o esimae he ILO unemploymen by direc derivaion of he resuls of quarerly Labour Force Survey (LFS), wihou using any oher auxiliary variables obained from oher daa sources. In he nex secion he monhly unemploymen esimaes are analyzed for period January 00-January 0, emphasising he impac of he economic crisis on he labour marke. The Romanian economy knew a coninuous posiive GDP growh during , he unemploymen significanly decreasing, when he labour marke followed one of he economical principals: he unemploymen ends o fall when growh is high. In ha circumsance, he unemploymen rae (calculaed for - age group) decreased from. percens in he firs monh of 00 o. percens in Augus 00, when he value was he lowes during he enire analysed period. The middle of he year 00 is he

8 Vergil Voineagu, Silvia Pisica, Nicolea Caragea saring poin of he global economic and financial crisis ha affeced also mos of he counries across he world. Romania was no excepion. As expeced, mid of 00 was an inflexion momen marking he insallaion of he unemploymen increase. The unemploymen sared o grow especially unil he firs quarer of 0 bu no as much as in oher European counries also affeced by he crisis. Several reasons can explain he phenomenon. On one hand, Romania has a quie srong share of underground economy, and consequenly a high share of employmen in he informal economy. As sudies (Vasile, Pisică, 0) shown one of he effecs of he economic crisis was a shif of employmen from formal ino informal area, he share of persons engaged in informal secor growing in hree years by percenage poins (from. percens in 00 o. percens in 0. On he oher hand, Romania is an emigraion counry. I is noiced ha, a facor influencing he employmen dimension in Romania is he inernaional migraion, especially in he group of working age populaion (- years), he phenomenon increasing since 00 afer joining he European Union. As he oher Member Saes of EU declared o Eurosa he flows of Romanian ciizens arriving in hose counries (and of course leaving Romania) grew from less han 00 housand persons in o more han half million in 00 and over 00 housand persons in 00 (afer ha phenomenon slowing down quie significanly since, because of he financial crisis, he job opporuniies became very limied in he old Member Saes). Monhly unemploymen figures are based on resuls of he model presened in he secion before (figures come from he seasonally adjused series SA, wih unemploymen raes calculaed for boh age groups - and - years old figure.a and.b). Eurosa Saisical Office of he EU Eurosa daabase, hp://epp.eurosa.ec.europa.eu/poral/page/poral/eurosa/home

9 Forecasing Monhly Unemploymen by Economeric Smoohing Techniques Figure a. Monhly unemploymen rae (- years) - % - Non-seasonally adjused (NSA), Seasonally adjused (SA) and Trend NSA SA Trend Source: Auhors calculaion based on LFS daa, Naional Insiue of Saisics, 0 The unemploymen rae has been rising since Augus 00, reaching he highes value in March 0 (. percens for - age group, respecively. percens for - age group)..0 Figure.b. Monhly unemploymen rae (- years) - % - Non-seasonally adjused (NSA), Seasonally adjused (SA) and Trend NSA SA Trend Source: Auhors calculaion based on LFS daa, Naional Insiue of Saisics, 0

10 Vergil Voineagu, Silvia Pisica, Nicolea Caragea Mos par of he unemploymen belongs o he male populaion. The share of men in oal unemployed persons varies from. percens in June 00 o almos wo hirds, i.e.. percens in January % -.0 Figure a. Monhly unemploymen rae by gender (- years) - % - Seasonally adjused (SA) Source: Auhors calculaion based on LFS daa, Naional Insiue of Saisics, 0. M-SA F-SA One pariculariy of he Romanian labour marke is he sense of he unemploymen dispariies by gender. In mos Member Saes of he European Union unemploymen is deeper among women as compared o men. In Romania is he opposie: he unemploymen rae for male exceeds he one for female. The gap beween he wo varies for hose aged - from 0. percenage poins (June 00) o. percenage poins (June 00) and for he group - years old from 0. percenage poins (June 00) o. percenage poins (December 00).

11 Forecasing Monhly Unemploymen by Economeric Smoohing Techniques - % -.0 Figure b. Monhly unemploymen rae by gender (- years) - % - Seasonally adjused (SA) Source: Auhors calculaion based on LFS daa, Naional Insiue of Saisics, 0. For he las couple of years (saring wih he las monhs of 00) he gap beween he wo raes of unemploymen has he endency of decreasing while he share of male in he oal number of unemployed persons is more seadily close o 0 percens (if before 0 he value of he share could vary wihin an inerval of percenage poins, during las wo years he inerval decreased o percenage poins).. Conclusions Based on he one improved Hol s model, he mehod described in his sudy is he mos suiable model for forecasing monhly unemploymen rae in Romania. Each model ype has unique characerisic which fis o a paricular daa series. More forecasing echniques should be explored o ensure finess o longer series of unemploymen rae. There are several advanages ha can be obained when using double exponenial smoohing echnique: exponenial smoohing models could fi wih compuer sysem; in his case, a simple program such as Microsof Excel can be used o generae new forecass; he models of exponenial smoohing reac more quickly o changes in daa paerns han he moving average; The main problem could be encounered when using his mehod is he deerminaion of he size of smoohing coefficiens. The crierion is o choose α such ha he MSE is M-SA F-SA

12 Vergil Voineagu, Silvia Pisica, Nicolea Caragea minimum. Wih he assisance of solver faciliy available in Microsof Excel, he amoun of work is saved when searching for he bes parameer α value. Economeric models provide formulaions and assumpions on variables subjec o forecas, reflecing he orienaion of he behaviour of he sudied phenomenon. Limis due o he forecas model are he followings: exisence of he forecas errors, defined as he difference beween esimaes and real values of he raw series; he model impose he minimizaion of hese errors by he condiion ha mean of squares of forecas errors o be minimal; deviaions beween forecased values and values of he reference series (benchmark) due o heir dependence on he rend of he previous period, and facors ha can ac unpredicably over economic environmen. These deviaions are larger in he case of youh aged - unemploymen daa series, he differences being exacerbaed by he volailiy of he series for his segmen of he populaion. REFERENCES []Hanzak, T. (00), Improved Hol Mehod for Irregular Time Series, WDS'0 Proceedings of Conribued Papers, Par I, ; []Hol, C.C (00), Forecasing Seasonal and Trends by Exponenially Weighed Moving Averages - Journal of Economic and Social Measuremen, ISSN - (Online) hp://iospress.meapress.com/conen/pyurkxkbqq/ ; []Jula, D., Jula, M. (0)), Meode de prognoza, Musang Publishing, Buchares; []Lynwood, A. Johnson, D.(0), Forecasing and Time Series Analysis, McGraw-Hill,Inc, nd ediion; []Vasile, V., Pisica, S.(0), Trends of Employmen in Informal Secor in Romania during Crisis; Economic and Social Effecs ; h ediion of he Inernaional Conference "EUROPEAN INTEGRATION NEW CHALLENGES, Faculy of Economic Sciences Oradea Universiy, Oradea, 0, ISBN , p.-; []Voineagu, V., Caragea, N., Pisica, S.(0), Mehodological Proposal for Compiling he ILO Unemploymen wih Monhly Periodiciy ; Romanian Saisical Review no., p. -0; []Winers, P. R. (0), Forecasing Sales by Exponenially Weighed Moving Averages; Managemen Science, pp. -; []European Commission Eurosa, Direcorae F: Social Saisics and Informaion Sociey, Uni F. Labour marke, working documens hp://circa.europa.eu/members/irc/dsis/ employ/library?l=/&vm=deailed&sb []Eurosa daabase, hp://epp.eurosa.ec.europa.eu/poral/page/poral/eurosa/home []Unemploymen saisics (Doc.: Eurosa/D/LAMAS//0)

13 Forecasing Monhly Unemploymen by Economeric Smoohing Techniques []Harmonised monhly unemploymen raes (Doc.: Eurosa/F/LAMAS/ 0 /0) []Monhly unemploymen (Doc.: Eurosa/F/LAMAS/ /0) []Harmonised monhly unemploymen raes (Doc.: Eurosa/F/LAMAS/ 0 /0) []Revised mehodology for he calculaion of monhly unemploymen raes (Eurosa/F/LAMAS/ a /0) []Feasibiliy sudy on fl ash esimaes of monhly unemploymen raes using EU sampling on weekly LFS samples (Doc.: Eurosa/F/LAMAS/ b /0) []Insiuul Naional de Saisică (INS) România, meadaa daabase hp://colecaredae.insse.ro/meadaa/public.hm?locale=ro