The seven steps needed for good-quality backcasting1
! Christine Pinel* The new version of the French classification of activities, NAF Rev. 2, has become the benchmark classification for structural business statistics. To accompany the change, INSEE has developed a method for backcasting NAF Rev. 1 to NAF Rev. 2. The first step consists in processing “branches,” i.e., the primary activities of enterprises, identified in the Annual Enterprise Survey (Enquête Annuelle d’Entreprise: EAE). Next, we recalculate the enterprise’s principal economic activity (Activity Principale Exercée: APE). The method relies on anad hoc“intermediate” classification. The branches are converted into the final classification in four steps—two deterministic, two probabilistic. Three further steps are needed to obtain APEs in the final classification. The backcast EAE data follow the statistical-processing sequence for structural business statistics all the way to end users, including the National Accounting system.
When a classification changes, we cannot simply publish results under the old version until the chosen crossover date, then work under the new version. There are two reasons why this is not possible: – First, survey management procedures may refer to the previous survey period, or even the two previous periods—but all the data must be coded in the same classification. – Second, the survey results will be used by entities that will not all switch to the new classification at the same date.
We must therefore develop a method for backcasting the series. For our present purposes, we need to translate the activities of enterprises under nAF Rev. 1 into nAF Rev. 2—and vice versa to perform an extrapolation. The Annual Enterprise Survey: a central role in backcasting
Box: Method for determining an enterprise’s principal economic activity (APE) APe codes are determined in Annual enterprise Surveys (eAes) by means of an “up and down” algorithm. We begin with the breakdown of sales by industry at the most detailed level of the classification (three activities for the enterprise in the example below) and we add the sales in industries with the same parent. At each step, we move up to a more aggregated level in the classification. once we reach the summit—the point that contains the enterprise’s total sales—we move back down the tree; at each level, we select the activity that posts the largest sales figure. The activity selected at the most detailed level becomes the AP e code for the enterprise. Industries and AP es are, of course, coded under the nAF system. note: According to its official definition, the AP e code must be determined from the value-added breakdown. As the distribution of value added by industry is generally not known, we use “proxy” variables instead—namely, sales or, in some cases, number of employees.
The nAF Rev. 2 implementation timetable set January 2008 as Two core variables are coded in relating to an individual industry. AP e the date for the switchover in the nAF and therefore directly affected serves to classify the enterprise in French business register, SIR ene. by the classification change: (1) the an activity sector. The results for a The system of structural business breakdown of an enterprise’s sales given sector therefore also include statistics had been preparing for the (turnover) by industry, i.e., the change since 2005. In distribution of its total sales across 2cl0a0s7s,i fisctaattiisotni cians implemented tools its different basic activities, and (2) its noitazinomraH s, onsiviDi tht*A el worked at theB sunise suSvryetie omewrf initc ,gsirhenitniP and adjusted applications to prepare principal economic activity (Activitywhich forms part of InSee’s Business Statistics the crossover from nAF Rev. 1 Principale exercée: APe). The firstsieh dsall yuplbpt étapes “Les seiDtcerri onagiator1.e. to nAF Rev. 2 and to obtain the variable is used in national accountingnécessaires pour une rétropolation de qualité,” materials needed for the backcasting to determine industry accounts, whichnCoovu.r-riDeercd.e2s0s0ta8t,i sptipq.u7e1sn .ose,)esircn h(Fre ,57-, 25 1e.fr/h tt:p//ww.wnies and extrapolation work ahead. measure all economic activity strictlyfr/ffc/docscff_1sc/.l52.fdp
courrier des statistiques, english series no. 15, 2009
the output of “minority” activities more complex classification, we can generated by enterprises whose AP e achieve greater consistency of the corresponds to the sector studied. sales figures provided. With the classification change, the challenge is to obtain robust results inA solid “carapace” the final classification, both for the sales breakdown and for the APe codes. SIRene crossed over in January That is the goal of all backcasting and 2008. The 2008/2007 eAe— extrapolation exercises. conducted in 2008 to gather data on The sales breakdown by industry 2007—had to be administered under is collected only in the AnnualFood industry: Danone plant in Bieruh, Polandirr foe th7002002/ 6wen ehtfissalc ion.icatordi Acc ,htgnylseit euq enterprise Survey (enquête Annuelle onna e d’entre is covers 140000p re: eAe). The eAerket-esAkee (dd aetnat eornp r2is0e0s6 tcoo llreecstpeod nidn 2u0si0n7g) sect,or entenrpornis-feisn anecviearly yemaarIts saw uesr vurwevyoe uyml dam ngaaangiane gdne oud niundnfeodrr emnr aAntFiAo RFn eRov.en v1t. ht2o, tahe intermediate classification. In the . ri e event, the response rate and quality fsocoodp ei nidnucslutrdy,e s semrvainceufsa, ctwuhonlge, satlhe/ttrhaen sifoirtimal aticolans soiffi caetaiocnh: iinndtou stwryh icienh were as high as in previous years. nd The first backcasting scheduled in rceotnasilt rutcrtiaodne., At tprraenssepnot,r itta itsi omna, naaged iitn dcuonsntvre(iretes)d ,o fa tnhdei finwalh actl apsrsoifpicorattiioonns i?s the project timetable consisted in b fo ifferent statistical offices. translating the 2006/2005 eAe data Wyhen utr hed enterprise is flagged by Aosn tbhaics kicnafosrtimnga tia onnnd , ewxet ranpeeolda tiao dni rreelcyt collected under nAF Rev. 1 into in an eAe the nAF Rev. 2. Failing this operation, the AanP e “ocbosdeer viast iroenc”alculated from, the foobr sea rvsautrivoeny unwdaevre .b oWthe cslahsalsli ftichautsi obnes computer application—which requires breakdown by industry (Box). These able to determine the conversion ke s data on two previous years—would recalculated APes serve to da not have been able to function for the SIRene register. A backcuapstinteg for switching from the old classificatioyn the 2008/2007 eAe. This backcasting r ex took place between December 2007 (aoctit raclpaolation) ioen xewriclil ser esihn uffalen to the new one.and January 2008. The tool used is bothv itiyndustrsys ifbirceaatkdowns and APe For this purpose, we cannot simply an SAS “macro-instruction” named on the cARAPAce, as in “CalculApproprié scuordveesy.s ,I t asw iltl hneey caerses atrhiley orenllyy source toohnbees eortuvhene drt ewru on tdbherere atkohdledo wnneclswa .s sseiWfpieac raahttiaoevlnye,, pour la Rétropolation desActivités allowing such an operation therefore built an “intermediate”Principales and desActivités ventilées . parChiffre d’affaires et/ ouEffectifs ” An intermediateivcty it iniptsnnoofr oreorscm cl auic aeltne aly o o onds to(itaceitfiilssalcsiethSou lweoafbhl larca e“ny,: classificatio for observingeach classification (Diagram). each of principal activities and backcasting nactivities broken down by sales and/ industries or number of employees”).item of the intermediate classification corresponds to a non-empty The backcasting method chosen for intersection of the cartesian product Before explaining the backcasting structural business statistics rests on of nAF Rev. 1 and nAF Rev. 2. And method for the 2006/2005 eAe, let two key principles. First, we decided to as we ask enterprises to supply a us take a brief look at the chaining work at the most detailed level, i.e., to single breakdown rather than two, of backcasting operations in the backcast each individual (enterprise), even if the activities are coded in a French system of structural business rather than directly convert the aggregated series without going back to the individual information. The goalDiagram is to reconstitute the sales breakdownThe intermediate classification bAeyta ctihhn ed uuensintt rdy i no af tnthdhe e afipsnrsaiolg cnec slasa,sn s iwfiAec Pantei eoetnod. tacinoiveR FAN2Intev. AF R. 1Nficealidtareem ss only reaggregate the individual data to obtain the backcast series. note that we shall also need to directly backcast the APes of enterprises whose sales breakdown by industry is not known.
The second point concerns the method for using the eAe to gather information on the classification change. If we merely switched from a
The seven necessary steps for good-quality backcasting statistics. After the annual survey We look at each enterprise and begin is instantaneous and the entire sales (eAe), the processing chain moves by translating the breakdown of its figure for the industry is recorded through (1) the Unified System of sales by industry (activity). under the new nAF Rev. 2 code. Business Statistics (Système Unifié de Statistiques d’entreprises: SUSe), The goal is to restate the sales For all industries whose which merges eAe data and tax data breakdown by industry under the correspondence with the new on business earnings, then (2) the new classification, so as to determine classification is not univocal, the Intermediate enterprise System the APe under nAF Rev. 2. The procedure is more complicated. (Système Intermédiaire d’enterprises: quality of the APe coding will be Let us imagine an activity Z under SIe), which provides full coverage higher if we work from a backcast nAF Rev. 1 that is split into two of all non-financial enterprises and breakdown and not exclusively from activities in nAF Rev. 2: X and Y. ensures accounting-data consistency. the information provided by the APe Some enterprises perform only the The final stages involve the national under nAF Rev. 1. Accordingly, for part of activity Z corresponding to X, accounts and eurostat, the main end- enterprises that did not supply a others the part of Z corresponding users of structural business data. sales breakdown in the 2006/2005 to Y, and still others engage in all of tehAe e2, 0w02e l2o0o0k4 fwora vtehse. Iifn fito remxiasttiso, n wine Z and therefore in both X and Y (the In the wake of the 2006/2005 eAe, - breakdown between X and Y can we backcast the three previous eAe r waves under nAF Rev. 2 in early va y). 2008. We then supplied the backcast To backcast such industries in the eAe data for the 2002-2005 waves 2006/ we under the new classification and the 2005 eAe, need to refer to the breakdown of the 2007/2006 eAe 2007/2006 eAe data based on the under the intermediate classification. intermediate classification to the SUSe This gives, among all po ible administrator, who incorporated them combinations between nAF Rev.s 1s and into the SUSe databases. An additional backcasting of SUSe bases was bnyA Fe aRcehv . e2n,t etrhporissee . acTtauballely 1r epsohrotewds needed to translate into nAF Rev. 2 us the form in which th i formation the principal economic activity (APe) is presented. We takeet hen example codes of enterprises not covered byConstruction industry: “Le Phare,”of enterprises engaging in (among the eAe. In the third quarter of 2008,a multi-purpose auditorium under construction the SIe retrieved SUSe data under thein Chambéry, Franceother activities) 452c, “construction new classification to incorporate and of civil-engineering structures,” which complete them so as to obtain bases is divided into six different activities in entirely under nAF Rev. 2. These apply the previous years’ breakdown nAF Rev. 2. enterprise A is involved results are now available to national to 2005 sales. in three subdivisions of 452 c. Its accountants. sales in industry 452c are distributed The first stage in backcasting among its activities under nAF Rev. 2 Since end-2008, InSee has also breakdowns is simple: it consists in the following proportions: 20%, been conducting extrapolations from in processing the activities under 50% and 30%. nAF Rev. 2 to nAF Rev. 1. The reason nAF Rev. 1 that correspond to a single for the operation is that the 2007 activity under nAF Rev. 2. For all these The second stage in backcasting edition of eAe, and therefore of SUSe “univocal” industries, the translation breakdowns consists in comparing and SIe, was prepared under the new classification, but the data complyingTable 1: Breakdown of 2007/2006 EAE under intermediate classification with the eurostat regulation on structural business statistics for 2007 must be supplied under nAF Rev. 1.rirpnAseteennIetmrde FeR.v1 ssificatiate cla RAFnnioet2D. evas deliaaerb selnTotkdowsaleal . 1Rsve.v2 -eRon crsven ioctfaro We shall also need to extrapolate all subsequent annual waves up to 20% 4,922 26,021 452c01 4213AA 452c the national-accounting switchover 50% 4222Z 12,556 26,021A 452c 452c02 (currently scheduled for 2011). A 452c 452c06 4299Z 8,543 26,021 30% A multitude of possible correspondences between 4221Z 881 977 90%B 452c 452c03 NAF Rev. 1 and NAF Rev. 2 B 452c 452c06 4299Z 96 977 10% Having described the setting and the basic concepts, we can discuss thec 452c 297 100% 297 4221Z 452c03 backcasting of the 2006/2005 eAe.
courrier des statistiques, english series no. 15, 2009