J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Wherehave(almost)all the wealthy gone? Spatialdecomposition of trendsin ,1820-1939

Jérôme BOURDIEU*, MartaMENÉNDEZ**, GillesPOSTEL-VINAY***, Akiko SUWA-EISENMANN*

*ParisSchool of Economics(INRA),Paris **UniversitéParis-Dauphine,PlaceduMaréchaldeLattredeTassigny,F-75775Pariscedex16, France;ParisSchool of Economics(INRA) e-mail:[email protected] *** ParisSchool of Economics(INRA)etEHESS, Paris

Summary –Thispaperexaminesthe evolution of wealthdistribution in Franceduring the urbanization process of the nineteenthcentury,based on acomprehensivedatasetof individualinheritances.Itpresents aspatial decomposition between ruraland urbanareas,distinguishing Parisfrom othercities.Weuseanon-parametric approachbased on wealthdensity functions.Changesin the level of wealthexplained most of the spatialevolution of wealthduring 1820-1939;atthe turnofthe century however,the effectof urbanization on wealthdistribution increased gradually.

Keywords: wealthdistribution,decomposition,semi-parametricand nonparametricmethods

Oùvontlesrichesses?Décomposition spatiale despatrimoinesen France,1820-1939

Résumé –Cetarticle analysel’évolution de ladistribution de larichesseenFrancedurantle processus d’urbanisation entre1820 et1939,àpartird’une basededonnéesindividuellesde successions.Nous exploronsladimension géographiquedel’évolution de larépartition despatrimoinesen distinguantParis, lesautresvillesetlesterritoiresruraux.Nous utilisonsdestechniquesde décompositionsnon paramétriquesetde micro-simulationspour mesurerl’importancedelacroissanceéconomiqueetde l’urbanisation dansl’accumulation dupatrimoine. Leschangements dansle niveaude richesseexpliquent une grande part de l’évolution de larépartition de larichessependanttoutelapériode entre1830 et1939, tandisquel’effetde l’urbanisation sur ladistribution dupatrimoine augmenteàlafin du19e siècle.

Mots-clés: répartition dupatrimoine,décomposition,méthodessemi-paramétriquesetnon para- métriques

Descripteurs JEL :C14,J11,H20,O18

5 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

1. Introduction The period between the earlynineteenthcentury and the Second World Warwasone of considerable structuralchangesin France. Onceruraland mainlyagricultural,it progressivelyturned intoanindustrialand urbanized country.While less thanone fifthofthe population lived in townsaround 1800,morethanone half did in the 1930s.Thispaperstudieschangesin privatewealthdistribution thatparalleled these structuralchanges,withafocus on geographicaldifferences,between ruraland urban areas,and among the latter,between Parisand othercities. The evolution of income during the process of developmentfrom an agriculturaleconomytoanindustrialone hasbeen modeled in the dualframeworkof A.W.Lewis(1954). Inthatframework,Kuznets (1955) conjectured thatariseinthe proportion of individualsemployed atahigherwage in the modernsectorwould lead first torising and then tofalling overall inequality (the so-called Kuznets ‘inverted-U’ curve). Keyassumptionsrelated on the proportion of individualsemployed in each sector,the degree of inequality within sectors,and the differencebetween the mean incomesacross sectors. Extending Kuznet’sconjecturetowealthdistribution isnotstraightforward. Indeed,asafirst approximation,the evolution of wealthshould besimilartothatof income,perhapswithatime lag(sincewealthcould beconsidered asastockresulting from accumulated flows of income). However,wealthcanalsobeindependentfrom income forvarious reasons.First,alarge part of wealthisinherited and depends thereforenotonlyon one’sownlife-cycle earnings.Next,specificshocksmightaffect wealthmorethanincome,suchasinflation leading todifferentiated valuation of assets orwardestructions.Moreover,wealthismoreconcentrated thanincome. The top of the distribution ownsmost of the totalamountof wealth. Thus,usualindicators such asmeansaresensitivetowhathappensatextreme levels,whichmightbenot representativeofotherparts of the wealthdistribution. Thiscaveatextendstoscalar inequality indicesbased on means.Atthe otherend,people mightearnaliving without detaining anyassets.Actually,according tothe datasetused in thispaper,only around one half of the population owned wealth. Moreprecisely,the shareof individualsleaving aninheritancedecreased overtime:ashigh as68% of all decedents in 1820-47,itdropped to54% in 1919-39(Bourdieu etal.,2003). Thesezerovalues must betaken intoaccount,the moresobecausetheirevolution overtime could be related tothe same structuralchangesthataffected the amounts of wealth. Despitethesedifficulties,asmall numberof casestudieshavenevertheless examined the long-runevolution of the wealthdistribution during the process of development.Inthe ,afterarelativelyegalitariancolonialera, wealth inequality increased and then decreased in three steps,during the Civil War(for Southerninequality),during WWI and,most of all,after1929(Williamson and Lindert,1980). ForBritain,Atkinson and Harrison (1978) havealsofound falling wealthinequality afterWWI.Using Swedishdataforthe period 1873-2005,Roine and Waldenström (2007)showed thatwealthconcentration wasalreadyhigh in the agrarianstate;anincreaseinwealthconcentration occurred during industrialization

6 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 but itwassmall and limited tothe top percentile. Thesestudieshoweverleaveaside the problem of the zerowealth. Inthe caseofFrance,Daumard(1973)studied estatesin fivecities(Paris,Lille, Lyon,Bordeaux and Toulouse) atthree points in time overthe period 1815-1914. She showed that,overall,the amounts of wealthincreased and the fivecitiesstudied by Daumarddid notevolveinthe same way.Lille seemed tohavebenefited most from industrialization in termsof capitalaccumulation,and Toulousethe least.InParis,the wealthiest of all Frenchlived side byside withthoseinextreme .Another recentpapershedslighton the long-runevolution of the top percentile of wealth holders in France(Piketty etal.,2006). Wealthconcentration increased during the nineteenthcentury upuntil WWI, driven bylarge industrialand financialestates: the shareoftotalwealthheld bythe richest percentile of the population grewfrom 43.4% in 1807 to54.9% in 1914. The riseininequality did notslowdownafter 1870,especiallyin Paris.The decline afterWWI wasabruptand mostlydriven by exogenous shocks. Thispaperstudiesthe spatialdistribution of wealthinthe nineteenthcentury and first half of the twentiethcentury.Wedepart from previous studiesin three ways. First,weconsidernotonlywealthowners but alsothosewho do notleaveany . Second,wecontrol forthe influenceofextreme valueson means.We choosethe ratherdrasticoption of getting rid of the top percentile. Indeed,the trajectory of the top percentile doesnotnecessarilyreflectwhathappened tothe great majority of the Frenchpopulation. The weightof the top percentile in the total amountof wealthcrushesand masksevents atmoremodest but perhapsmore representativewealthlevels:thoseofordinary people. Third,wegobeyond summary indicessuchasmeansorscalarindicesof inequality.Weuseanon parametric decomposition approachinordertoestimatethe role of urbanization and the rural- urbanwealthgapinexplaining changesin the wealthdistribution. Our method follows the line of Jenkinsand VanKerm(2005) adapted forwealth asin Jenkinsand Jäntti(2005). Contrary tothe latter,wedonotencounterthe problem of negativewealth,asthe definition used in thispaperisgross assets,in estatesregistered forfiscalpurpose. Changesin the density function of wealthare decomposed in twoelements,one thataccounts forchangesin the amounts of wealth and anotherelementthataccounts forchangesin the spatialstructureofthe population. The latterelementactuallyrepresents bothurbanization process and the factthatthe shareofwealthowners did notevolveinthe same wayin ruraland urbanareas.Changesin the amountof wealth,in turn,arealsodecomposed between changesin the meanwealths(the rural/urbanwealthgap) and changesin the formof the distribution. The sizeofeachelementismeasured byasummary indexand is compared,between them and overtime. Our results suggest thatchangesin wealthdistribution in Franceweredominated during the nineteenthcentury bythe change in the numberof assetowners (decreasing overtime) and the rural/urbanwealthgap. The period between the end of nineteenth century and WWII sawagrowing importanceofurbanization and changesin the shape of the distribution. Wealthownership increased in Paris,but notenough to

7 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 overcome the generaldrop observed in otherareas.Putting togetherthe evolution of wealthowners and thoseleaving no inheritance,our results suggest apolarization of Frenchsociety in the long run. The structureofthe paperisasfollows.The next section presents the datasetused. Section 3describesour non-parametricdecomposition of changesin the wealth distribution,whichallows forabroad-brushidentification of the relevantsourcesof distributionaltrends.Section 4analysesthe spatialdecomposition of wealth distribution in Francefrom 1820 to1939and discussesthe contribution of the differentfactors in the observed changesin the . The last section concludes.

2.Dataand definitions The datasetcomesfrom the TRA-wealthsurvey,based on Frenchestatetaxreturns (Bourdieu etal.,2003). Since1791,the Frenchtaxadministration hascollected the totalamounts of wealthofall decedents fornearlyall typesof property and atanylevel of wealth. When originallysetup,arandom selection wasmade bysampling individualswhosefamilyname started withthe letters “TRA”. Thisuniquesurveyis knowntogiveareliable pictureofthe nationaldistribution of wealthwithin the generalpopulation,although itisill-suited forthe studyof the super-rich(see Bourdieu etal .,2004;and Piketty etal .,2006). The sample includes41,489 adult individuals,over20years old when theydied,in the period between 1820 and 1939. The variable used asanindicatorof wealthisgross assets atdeath,deflated bya nationalcost of living index(all results areexpressed in 2007 Euros) 1 . Inthe following,the century and ahalf undersurveyhasbeen divided into5sub- periods,omitting WWI.The first period,1820-1847,ischaracterized byslow economicgrowthduring the Restoration,beforeadeep crisison the eveofthe Second Republic.The second period,1848-1869,covers twodecadesof marked industrialization during the Second Republic(1848-1851) and the Second Empire. Thisperiod of prosperity wasinterrupted bywar,the loss of tworichregionsof Eastern Franceand the Commune in Paris; along deceleration followed in the thirdperiod, 1870-1895. Growthresumed in the fourthperiod (1895-1914). The last period (1919- 1939) ischaracterized bythe upsand downsof the inter-waryears.Aquarterof the individualsin the datasetbelong tothe deceleration of 1870-1894,withthe rest being equallydivided overtime. The originalsourcesof the TRA surveyarethe “ tablesdessuccessionsetabsences ” (hereafterTSA),and the “ registresde mutationspardécès ”(hereafterRMD). Despitethe remarkable continuity of thesesourcesfrom the beginning of the nineteenthcentury, the degree of detail hasevolved overtime. First,theyprogressivelytook intoaccount newtypesof assets astheyappeared. Second,and moreimportantly,the waywealthis

1 The priceindexisbased on acost of living indexcomputed byD.R.Weirupto1913(Weir, 1991),the corresponding indexin Piketty (2001) and INSEE series.WearegratefultoD.R.Weir who generouslyprovided us withthisseries.

8 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 reported evolved. Forabout half of the period underconsideration,wecantake advantage of the workofthe estatetaxadministration. Forevery person who died,the administration opened anentry in afirst setof volumes(the TSA)and then completed it,eitherwhen estatetaxeshadbeen paid orwhen the administration wassatisfied that the individualhadleftno wealthbehind. Theseentriesinclude name,occupation, residence,maritalstatus,age,and,forindividualswithwealth,information about heirs and the dateatwhichthe declaration wasfilled out.Upuntil 1870 the TSA also included asummary of the individual’sestate,broken downintomoveable wealthand realestate. Henceour datauptothe 1860s reliesheavilyon the TSA.Atlaterdates, wealthinformation wasno longerrecorded in the TSA.Forthesecases,westart from the TSA and,foreachentry withapositivebut unknownamountof wealth,weturn toasecond setof registers (the RMD)whereacompletedescription of the estates appears,and then append thiswealthinformation tothe first setof entriesgleaned from the TSA.Asthismatching process isextremelylaborintensive,ithasnotbeen carried out forthe whole sample. Consequently,wehaveless wealthdatain the second half of our period,after1870. The lackofreadilyavailable information on the preciseamountof wealthafter 1870 forpart of our sample could beasourceofbias.Thereisno straightforward solution tothisproblem. Tocounterbalancethiseffect,wehavechosen toreweigh the sample byapplying raising factors after1870 togroupsof individualsforwhom we knowthe amountof wealthdeclared and who sharethe same characteristicsasthosefor whom welackwealthinformation. Thesegroupsaredefined bytime period,region, urban-rurallocation,age atdeathand gender 2 .Forinstance,women living in Paris, who deceased in 1871atages45 to64and whoseinheritanceareknown,aredeemed toberepresentativeofthe sub-groupofwomen aged 45 to64,who deceased during 1870-94,in Parisand whosewealthisunknown,but knowntobedifferentfrom zero. Wealthdatapresentspecificchallenges,compared tomoreregularincome data (Jenkinsand Jäntti,2005). First,wealthisconcentrated in the handsof onlyafew, whichmeansthatits distribution exhibits alarge spike atthe valueofzero. Second, even among the luckyfew,wealthishighlyunequallydistributed. Thismeansthat wealthdistributionsareusuallyright-skewed withextremelylong and sparseright- hand tails.Third,wealthdistributionssufferfrom the problem of influential observationswithextreme values.Inotherwords,usualmoments,suchasthe mean, aredriven byasmall numberof estatesatthe very top of the distribution,and arethus impreciselyestimated. The best solution tothisproblem would havebeen toinclude exhaustivelythe super-richinour sample,whichwecannotdo. The typicalsolution is totrim the dataatthe very top and performsensitivity analysistoevaluatethe impact of alternativetrimmings.Thisisthe strategywefollowin thispaper.Wetrim the data using anexogenous 99th percentile thresholdsestimated in Piketty etal .(2006),and

2 Region isdefined by6categories:NorthWest,NorthEast and Ile de France,Paris,SouthEast, SouthWest and Other(thisincludescolonies,“ territoiresd’outre-mer ”and foreign countries). Age at deathhasalsobeen grouped intothree broadcategories:from 20 to44years old,from 45 to64, and 65orover(youngerindividualsareexcluded from the sample). Results obtained without weights areavailable from the authors upon request.

9 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 wethen carry out sensitivity tests based on the 99th and 95th percentile thresholds found in the originalTRA sample (table A1inthe appendixprovidesbasic information about originaland finalsamplesforcomparison). Leaving out the top percentile isbyno meansanegligible assumption,becauseitentailsthatafarlarger shareofwealthisneglected and previous workfocusing on top fortunesduring the same period in Franceshowed thatthesewerecloselyrelated togrowthand industrialization (Piketty etal.,2006). Toevaluatethe consequencesof trimming,table 1providesahypothetical exhaustivesample made upofour sample complemented bythe super-richpicked up in Piketty etal .(2004,2006). Trimming halvesthe meanwealthvalues,asthe mean estatevalueofthe top one percentof the distribution issohigh. Unsurprisingly,this differenceisgreatest in Paris,wherepresumablythe majority of the top one percent arelocated. The gapdeepensoverthe nineteenthcentury and decreasesafterWWI. Table 1underlinesthatthe world of the super-richdiffers strikinglyfrom the world of ordinary people. Thisissonotonlyin termsof amounts of wealthbut alsoregarding the timing and the ratesof wealthaccumulation. While the Frenchsuper-richenjoyed theirhighest increaseinmeanwealth(by60%) from 1848-69to1870-94,thiswasthe caseonlylaterforthe rest of the Frenchpopulation (and withamoremodest riseof 31%). From nowon,between the twoworldsof the super-richand ordinary people,we will focus onlyon the latter(who still represent99% of the totalpopulation). Our story isone of wealthholders in France,notof wealthperse. Consistentwithprevious studiesforFrance(Daumard,1973; Piketty etal .,2006), weintend tocontrast Parisand othercities.Ofcourse,while itmightbepossible to “locate” preciselythe sourceofincome,itismoredifficultto“locate” wealth. Real estateisof coursebydefinition,preciselylocated. But,itmayhappen thatarural estateisowned byaNoble living elsewhere,in aprovincialtownorin Paris.The diffusion of financialassets throughout the nineteenthcentury makesthislocalization even morecomplex.The definition weuseinthe paperistorelateindividualstothe areatype 3 wheretheydied:ruralareas,Parisorothercities(called hereafter“provincial cities” 4 ).

3 Onthe rural-urbandevide in ahistoricalperspective,see Pirenne (1927),Duby etal. (1983),De Vries(1984),Hohenberg(2004). 4 Herewedefine asrural(urban) a“commune” (the smallest administrativeunit)withfewer (greater)than5000 inhabitants.Inordertotake intoaccountchangesin sizeovertime,weuse three population censuses,in 1841,1876 and 1926,and relateeachdecedentin our sample tothe surveyof the yearclosest toheryearof death. Asaresult,around 9percentof all “communes” shifted from ruraltourbanduring the sample period. Asisoften the case,Parisisanoutlierand all the moresointhatthe city limits changed overtime. Forsimplicity,wechooseamiddle of the roadsolution:forthe whole period surveyed,wesupposethatParisisdefined byits current boundaries.Asaresult,earlynineteenthcentury suburbs(incorporated in the city in 1859) are supposed tobe“Paris”whereasmoredistantsuburbsareconsidered asprovincialcities.

10 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Table 1. Meanwealthlevels,totalpopulation and trimmed population 1820-18471848-18691870-1894 1895-19131919-1939 France (1) Meanwealth,P99-P100*594 688 850333 1367 163 1716274646176 (2)Meanwealth,trimmed 6719 924711 77915444 9019 (3)Meanwealth,total12599 17658 25333 32 45215 391 (4) :(2)/(3)0.530.520.460.48 0.59 Paris (1) Meanwealth,P99-P100*23402393808107 5959 094 7992802 3286677 (2)Meanwealth,trimmed 70617894 11 603 11 184 8019 (3)Meanwealth,total30 39245 8967107891000 40805 (4) :(2)/(3)0.23 0.170.160.120.20 *P99-P100 isthe top percentile of the distribution. Note :in 2007 Euros Sources :(1) TablesA2and A3in Piketty etal. (2004).(2)TRA surveytrimmed atP99 withPiketty etal. (2004) thresholds.(3)TRA surveytrimmed forP0-99 and Piketty etal. (2006)forP99-100.

3.Accounting forchangesin wealth:The decomposition methodology Toovercome the limitation of scalarmeasures,researchers haveexplored inequality decompositionsin aless parametricframework(see DiNardo etal.,1996; Jenkinsand VanKerm,2005). Inthissection,wefollowthisline of researchand appealtoa density function decomposition approach. Using the additivedecomposability property,wecanrepresentthe totalwealth density function gt ()y atanydate t asthe combination of apointmass atvaluezero capturing the proportion of decedents without wealthplus acertain density function forthe assetowners:

 A t   A t  gt ()y = 1−  ⋅0 +   ⋅ ft ()y (1)  N t   N t  whereiANttsthe shareofassetowners in the totalpopulation in period t .The wealthdensity function forassetowners in period t ,,ft ()y canagain berepresented as aweighted sumofthe densitiesfor k mutuallyexclusivesubgroupsof population (in our case, k =1,2,3,corresponding toruralareas,provincialtownsand Paris). Thus, wecanwrite: A 3 A k 3 t t k k k gt ()y =⋅∑∑⋅=ft ()yvt ft ()y (2) N t k ==1 A t k 1 k A k whereiv t sthe sample fraction of wealthowners who died in region k attime t ()t k overthe totalnumberof decedents ( N t )and f t isthe wealthdensity function of wealth owners in region k in period t .

11 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Changesin the distribution of wealthcanbewritten asthe sumoftwo components.The first represents the part thatcanbeimputed toachange in ‘levels’, herethe amounts of wealththatassetowners possess within eachregion k ;thisis represented bythe regionalwealthdensity functions.Wecall thisthe wealthamount changesamong assetowners noted hereafter C D (y) .The second elementrepresents the part thatcanbeimputed to‘quantities’,or changesin the structureofthe population, C P (y) ;the latterrepresents urbanization,inclusiveofthe evolution of the regionalshareofassetowners:

3 3 kk k k ΔΔg()yv=∑∑t + 1 f()yv+Δ ft ()y= CD ()y ++ CP ()y (3) k ==1 k 1

Borrowing from the literatureonincome distribution and development,which offers anumberof differentmethodsfordecomposing distributionalchangesintoa growthcomponentand aninequality component(see,forexample,Kakwani,1993; Datt and Ravallion,1992),weadditionallyconstructcounterfactualdistributionsbased on additivetransformationsand decomposethe change in the wealthdensity function of eachregion k, Δ fk ()y intotwocomponents:the change duetoa ceterisparibus shift in regionalmeanswealthamong assetowners ( meanseffect ),and the change duetoa ceterisparibus shiftin the regionaldistribution around aconstantmean(pure distribution effect ):

Δ fk ()yy=ϕ k (;yk )(− fk y ) +−fk ()yyϕ k (;y k ) TUoooot oo oot + 1 oooooot o V TUoooooot ++11oVooooooot t (4) meanseffect puredistrii butioneffect The first componentcapturesthe distribution-neutralenrichmentthatasset owners in eachregion experienced overthe period. The second componentpicksup distributionalchangesaround aconstantmeanwhichoccurred in eachregion atany pointalong the wealthline (hence,going beyond ascalarmeasureofinequality). The aim istodetermine,within eachregion,whichofthe twocomponents hasbeen more importantin explaining observed changesin the wealthdistribution.

The change in population structure(C P (y) )canalsobedecomposed intotwo terms 5 .Weexpress the shareofeachregion v k asthe productof the proportion of individualsin region k ( N k /N )bythe proportion of assetowners in region k ( s k =A k / N k ). The change in the lattercomponent A k /N k identifiesregionalmovements in the proportion of individualsleaving anestate. The formerelement N k /N measures the pureurbanization effect(duetochangesin location,independentof changesin the numberof assetowners byregion).

5 Our wealthdecomposition draws on the one developed byJenkinsand vanKerm(2005). The essentialdifferencebetween the twoisthattheyadd anadditionalcomponenttothe decomposition of changesin wealthamounts,while weinclude anewdecomposition of changesin the population structure.

12 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Wenowsubstituteinequation (3)forthe structuraland wealthchangesobserved in eachregion bytheirrespectivesub-components.Wearethus able towritethe changesin the distribution of wealthinFranceasthe sumoffour explanatory effects 6 : 3 k k k k Δ g()y=CDP()y+C()yv=∑ t ++11 ϕ t (yy;t )(− ft y )  k = 1 33 k k k k +−∑ vt ++11 ft ()yyϕ t (;y t + 1 )  k = 1 3 N k A k k + ∑ Δ ()⋅ ()⋅ ft ()y N N k t + 1 k = 1 (5) 3 A k N k k + ∑ Δ ()⋅ ()⋅ ft ()y N k N tt k = 1

=CMI()y++C()yCUS()y+ C()y wherethe first twocomponents capturechangesin the levelsof wealth: C M (y) the regionalmeanseffect forFranceand C I (y) the intra-regionalpuredistribution effect forFrance. Asnoted above,the last twocomponents capturerespectivelythe pureurbanization process of the Frenchpopulation (holding everything else constant), C U (y) ,and the regionalchange in the shareofassetowners in France, C S (y) . Notethatthe regionalmeanseffect forFrance C M (y) takesintoaccountnotonlythe common globalenrichmentenjoyed byall regions(the French meanwealtheffect ), but alsothe factthatactualwealthgrowthratesvaried from one region toanother(the changesin the urban/ruralwealthgaps).

4. The spatialdecomposition of wealthdistribution in France from 1820-48 to1919-39 4.1. The wealthdensity function Figure1plots kernel density estimatesof the wealthdistribution forFranceand by location in 1820-1847,1870-1894 and 1919-1939(the horizontalaxisison a logarithmicscale,though intermediatevaluesin 2007 Eurosarealsogiven toease interpretation). Atabulation of the databywealthrangesisalsopresented in table A2 of the appendix. The kernel function in Francemovesprogressivelytothe rightbetween 1820- 1847and 1919-1939,especiallyforthe bottom part of the wealthdistribution,aswell

6 Thistype of nonparametricdecomposition isessentiallyanaccounting exercisecarried out by generating appropriatecounterfactualdistributions,and isnotinsensitivetothe yeartaken asthe referencepointin the weighting procedure(eitherbaseyearorfinalyearvalues). Wepresent the methodologywiththe choiceofaggregation weights thatcorrespond tothe results shownin the paper,thatis,finalyearvaluesfor v and initialyearvaluesfor f k ,but the robustness of our main conclusionstodifferentweighting choiceshasbeen tested and isdiscussed below.Results forthe differentcombinationsof weights areavailable from the authors upon request.

13 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 b y lo ca t ion F r a n c e nd in i str b ut ion W e a l t hd . F ig ur e1

14 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 assome (but notall) of the upperquantiles.Itisclearthat,becausethe majority of our sample isrural(80%in1820-48,and still 57%in1919-40)the curveforall of France resemblesthatforthe ruralareas,especiallyin the initialperiod. Nevertheless, the curvesforParisand the provincialcitiesarehigherin the finalperiod,showing the growing role of citiesin wealthdistribution. Figure1shows thattherewasnotonlya broadenrichmentof assetowners in France(despitethe factthatwecompareour initialperiod withthe situation afterWWI and its devastating effects),but,more interestingly,the actual shapes of the distribution functionschanged,particularlyin provincialtownsand in Paris.Here,wesee ahollowing out of the lowand high ends of the assetdistribution towardsintermediatelevels,uptothe pointthat,in the caseof Paris,weactuallypass from abi-modaltoaunimodaldistribution. InPiketty etal. (2006)the super-rich(the top percentile) experienced fallsin wealthsharesafterWWI. Infigure1wesee thatsuchmovements atthe higherend of the distribution werealso shared atmoreordinary levelsof wealth(on our trimmed sample),especiallyin Paris, and (although less so) in provincialtowns.Notealsothatthe “missing part”ofthe story,absentfrom figure1,isthe numberof decedents without assets in France,which rosefrom about 31% in 1820-47toabout 46%in1919-39(see table 2). Putting these individualsbackintothe picture,the kernel curveforFranceshows apolarization between anincreasing mass atzerowealthand aprocess of mass concentration at around 6400 Eurosin ruralareasand at13000 Eurosin urbanzones.

4.2.The elements of decomposition Beforedescribing the ceterisparibus contribution of eachtermofour decomposition to wealthdistribution changes,wewill describethe actualevolution of eachtermof interest overtime. Letusstart bythe historicalprocess of urbanization of the French population (see table 2). Between 1820 and 1847,80%ofthe Frenchpopulation died in ruralareas.Between 1919 and 1939,57%ofthe population wasstill rural. Thus, therewasaslow,but constantmigration toprovincialtownsand tothe capital. Paris’ shareinthe totalpopulation reached 8percentatthe end of the nineteenthcentury and stayed atthislevel thereafter. Next,letuslook atthe second factorof the change in the structureofpopulation, thatis,the shareofassetowners byregion. Assaid above,assetowners aredefined as decedents who leftapositiveestate,no matterhowsmall the actualquantity.Table 2 suggests three importantfindings.First,thereisastrong discrepancybetween geographicalareas.In1820-1847,most ruraldecedents held some wealth(74% on average). Onthe contrary,the majority of urbandecedents leftnothing behind (while about half of the population leftnothing behind in provincialtowns,thisproportion climbed uptoastriking figureof70%inParis). Thisranking wasmaintained overthe century.The greatershareofassetowners in the countryside isduetorealestateand datesbacktothe “ Ancien Régime”. Itwasenhanced bothbyRevolutionary redistribution 7 and bythe ruraldevelopmentthroughout the nineteenthcentury of agriculturaland non-agriculturalactivities,suchashandicrafts,cottage industriesand

7 See Bodinierand Teyssier(2000)

15 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Table 2. Trendsin population shares

1820-18471848-18691870-1894 1895-19131919-1939 France Totalpopulation 7577 7878106877419 7923 Shareofassetowners 68.2%66.5% 60.6%56.3%53.8% Ruralareas Population share80.0%75.5% 67.5% 64.1% 57.6% Shareofassetowners 73.8% 74.9% 71.8% 68.1% 64.4% Paris Population share4.5% 6.0%8.2%8.2%8.2% Shareofassetowners 30.4% 24.5% 22.6%22.5% 29.4% Provincialtowns Population share15.5% 18.5% 24.3%27.7%34.2% Shareofassetowners 50.1% 45.7%42.4% 39.0%41.8% Source :Trimmed TRA survey the publicsector.Inthe countryside,itwaseasy tobuy apieceofland,becausethere wasno downwardlimiton the sizeofparcel holdings.Onthe contrary,in large cities like Paris,itwasonlypossible toownwhole buildings(Rosenthal,2002). Thus,the fixed cost of real-estateownership wasvery high in cities.Moreover,itwasprofitable forfarmers toownpart of theirland,while in most otheractivitiesworkers could not owntheirmeansof production. Last,in the countryside,owning even atinyparcel would entitle the ownertoparticipateinlocalactivitiesorobtain access tolocalpoor relief and creditin badtimes. Second,the shareofdecedents who leftanestatefell sharplyfrom 68% to54% between 1820 and 1939. Thisconfirmsastylized fact,whichwasestablished,and tested forrobustness and significance,in aprevious paper(Bourdieu etal .,2003), wherethisdecline in the numberof wealthowners wasshowntobehomogeneous with respecttogender,age groupand professionalstatus. Third,the fall in the proportion of population owning assets wasgeneralthrough- out France,withone exception. Between 1820-47and 1919-39the shareofasset owners in ruralareasand in provincialtownsdropped by9.4 and 8.4 percentage points respectively.InParishowever,the fall wassmaller,atonly1.4 percentage points.In fact,the percentage of assetowners in Parisissimilarin the first and last time periods, showing aU-shaped patternovertime:itfell by8percentage points between 1820- 1847and 1870-1913,beforeincreasing by7percentage points afterWWI. Itisof interest tonotethatrural-urbanmigration cannotexplain thissecular decreaseinassetowners,sincethe drop isobserved bothinruralareasand cities. Moreover,if the probability of asset-holding washigherin ruralareasthanincities, whydid people (and notnecessarilythe moredevoid of) migratetocities?Two arguments canbebroughtup. First,therewasageneralchange in the role of wealth- holding in :withthe riseofwage earners,salaried employees,and the public sector,notonlyin citiesbut alsointhe countryside,itbecame less necessary tohold

16 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 assets.Anadditionalexplanation appealstowealthlevels:if city dwellers end upfar richerthanruralinhabitants,even withalowerprobability,rural-urbanmigration would still bejustified in thisframework. Inthe following,wewill showthatthiswas indeed the case,especiallyuntil 1895-1913,forthe wealthiest. Letusnowturntothe changesin the amounts of wealth(table 3). Wealthlevelsof assetowners in all three areasand atall points of the wealth distribution followasimilarpattern:theyincreaseuntil WWI and then drop in 1919- 1939. The fall in wealthafterWWI wasgreatest forthe richest quartiles:the top 25% wealthiest lost 40%onaverage,while the bottom 25% lost only13.4%,and thoseat the median26%. Considering regionsseparately,meanwealthlevelsof assetowners werelowest in the countryside,followed byprovincialtownsand Paris.Inthe latter,meanwealthwas overtwiceaslarge asthatin ruralareas.The greatest gappertained in 1870-1894,a period of greathardship in the countryside 8 .During theseyears,the ratio of wealth

Table 3. Trendsin wealthlevelsof assetowners:Meanand quartiles

1820-18471848-18691870-1894 1895-19131919-1939 France Meanlevel 9854 1391619 433 27 418 16767 1st quartile 846132516342637 2283 Medianlevel 3013432158808351 6170 3rdquartile 8967 12647165402793416936 Bysubpopulation groups: Ruralareas Meanlevel 8954 1206514960 24037 14 452 1st quartile 8821338154627652134 Medianlevel 293441825515 78166013 3rdquartile 845311 659 14 42523 889 15 184 Paris 00000 Meanlevel 23 218 32 18751 39349 59627231 1st quartile 777 787334021083584 Medianlevel 302531521679714 78610074 3rdquartile 22 38333359 49 88360967 29487 Provincialtowns Meanlevel 14 345 23 1103470837 27420 995 1st quartile 72814871789 2160 2507 Medianlevel 3423 6018 6984 9388 6976 3rdquartile 1295322526 2908642173 18 396 Note :in 2007 Euros Source :Trimmed TRA survey

8 See Grantham(1996)

17 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Tableau4. Wealthinequality levelsamong assetowners

1820-18471848-18691870-1894 1895-19131919-1939 Theil indices France 1.0651.0641.130 0.9800.854 (0.020)(0.021) (0.023)(0.019) (0.014) Ruralareas1.0290.998 1.000 0.9520.818 (0.025) (0.027)(0.031) (0.023)(0.020) Paris1.0801.176 0.9100.8800.754 (0.088) (0.087)(0.066)(0.090)(0.048) Provincialtowns1.0571.0501.136 0.9750.873 (0.044) (0.046)(0.035) (0.037)(0.023) Note :Standarderrorestimatesbased on 100 resampling bootstrapsareprovided in parenthesis Source :Trimmed TRA survey between Parisand the ruralareasreached 3.5,withananalogous figurebetween provincialtownsand ruralareasof 2.4. Thiswealthgapthen decreased in 1895-1913. AfterWWI, the ratioswithrespecttoruralareasdropped to1.9 forParisand 1.5 for provincialtowns.The medianwealthlevels(of the sub-sample of assetowners)add moredetail tothe picture. First,the medianwealthgapwasstill in favorof provincial townswithrespecttothe countryside,byafactorwhichvariesbetween 1.2and 1.5. However,the position of Parisin the medianwealthladderreversed itself. Until 1870, medianwealthinParislagged behind ruralwealth. In1870-1894,thisgapreversed: medianwealthinParisjumped tothree timesthatof its ruralcounterpart,then decreased toaratio of 1.6by1919-1939. Forquantilesof wealthbelowthe median, thereisno constantranking between geographicalareas.Asaresult,the dominanceof citiesoverthe countryside reflected in the meanlevelsismostlyduetothe top quartilesof the wealthdistribution. The fourthelementof our decomposition,corresponding tothe observed changes in the distribution of wealthamong assetowners,hasalreadybeen described in the previous subsection,bylooking atthe actualshapesof the density functions.Letus nevertheless provide additionalevidencebylooking atstandardinequality indices. Table 4presents Theil indicesforFranceand disaggregated byarea, forwhich bootstrapped standarderrors havebeen computed. Inequality among assetowners did notsignificantlyincreaseoverthe nineteenthcentury until 1870-94,and then significantlydeclined,particularlyin the interwarperiod. Inequality started declining first in 1870-94 in Paris,and onlylaterin provincialtownsand ruralareasatthe turn of the nineteenthcentury.Intermsof levels,inequality isgreatest in Paristhroughout the whole period,followed byprovincialtownsand lowest in ruralareas. 4.3.The contribution of the differentfactors Wehavedescribed the evolution of the differentfactors of interest.Wewanttobeable torank the relativeimportanceofeachfactorin the observed changesin the distribution of wealth. Figure2represents and quantifiesthe ceterisparibus

18 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Figure2. Counterfactualdistribution changesin France,1820-47to1919-39

Note :The curvesrepresented correspond tothe choiceofweights of equation(5). The horizontalaxisison a logarithmicscale,withintermediatevaluesin 2007 Euros. contributionsof the four differentexplanatory factors in our decomposition of changes in the distribution of Frenchwealthbetween 1820-47and 1919-39. The closerthe curveistozeroatanywealthlevel,the less importantthatfactorisin accounting for totalwealthchanges.The upperleftgraph of figure2represents the regionalmeans effect( C M (y) )and shows howthe increaseinmeanwealthlevelsin every geographical location hasshifted density mass tothe right.The intra-regionalpuredistribution effect( C I (y) )atthe upperrightgraph of the figureshows howthe distribution of wealthexhibits falling density atbothhigh and lowwealthlevels,illustrating the process of wealthconcentration among wealthowners 9 .The twobottom graphsshow the contributionsof population changestowealthdensity changes.Onthe left,the decreaseinthe regionalshareofassetowners ( C S (y) )inFrance,everything elsebeing equal,will byconstruction increasethe mass pointatzerowealthand proportionately

9 Wecould speakofincreasing polarization between wealthowners and wealthdeprived. Esteban etal .(1994) describeapolarization process asaresultof greaterintra-grouphomogeneity (what theydefine asthe “identification function”) and greaterinter-groupheterogeneity (their“alienation function”). Our regionalmeanseffectwould capturethe increasing gapbetween assetowners and decedents withzerowealth(sinceatthe same time the shareofassetowners isdecreasing). Wealth density concentration captured bythe pureintra-regionaldistribution effectwould relatetoa process of ‘identification’among the wealthowners.

19 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

decreasethe wealthdensity mass everywhereelse. The pureurbanization effect( C U (y) ) iscaptured in the bottom right-hand figure. The ceterisparibus wealthchangesdueto the decreaseinthe ruralpopulation (wealthmass losses)arenotcompletely compensated bythe increaseinthe urbanpopulation (implying mass gainsatslightly higherwealthlevels). Togauge the relativeimportanceofthesedifferentdecomposition components in observed changesin wealthdistribution,weneed ameasureoftheirsize. Here,the size of eachcomponentisestimated bythe surfacebetween the horizontalaxisand the curvedepicting the contribution of eachcomponentin figure2.Technically,thissize ismeasured bythe integral,in absoluteterms,of the kernel density change corresponding toeachcomponent.Thisvaluegivesanorderof magnitude thatis comparable across components and overtime. Table 5indicatesthe dominantfactors explaining the change in the wealth distribution between 1820-1847and 1919-1939. Tosee if the ranking across components haschanged overtime,weseparatethe nineteenthcentury changes(from 1820-47to1870-94) from morerecentchanges(from 1870-94 to1919-39). Results shownarerobust todifferentweighting choices( i.e. across differentspecificationsof equation (5)). Confidenceintervals(atthe 95% level) arealsocomputed viabootstrap resampling and reestimation of the decomposition of wealthchangesfor100 replications. Suchbootstrapassessesthe robustness of the results withrespecttothe choiceofthe sample points used todrawthe kernel functions.When the dominanttermiswritten between brackets itdenotesinsignificanceofits relativeimportance,according toour bootstrapped confidenceintervals.Table 6providesthe detail of integralmeasuresand the valuesof the confidenceintervals,forthe weighting procedureused in equation (5).

Table 5. Summary of dominantfactors across differentcomponents of our decomposition of wealthchangesusing aggregatekernel density estimates

Changesin distribution, Inter-regionalmeans Pureurbanization C D (y) , vs changes, C M (y) , vs changes, C U (y) , vs Changesin population, Puredistribution Changesin %asset C P (y) changes, C I (y) owners, C S (y) France Changesfrom 1820-47to1919-39 C D (y)C M (y)C S (y) –Changesfrom 1820-47 to1870-94 C D (y)C M (y)(C S (y)) –Changesfrom 1870-94 to1919-39 (C D (y)) C I (y)(C S (y)) Note :The relativeimportanceofeachfactorismeasured bythe numericalintegration of absolutevaluekernel density changesattributable toeachcomponent(see Table 6),and the dominantfactoristhe one represented in eachcase. TheserankingsforFrancearepreserved no matterthe weighting procedureused in equation(5). Termsin brackets denoteinsignificantdifferencesbetweenfactors according to95% bootstrapped confidenceintervals obtained through resampling. Source :Trimmed TRA survey

20 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

Table 6. Relativeimportanceofdifferentcomponents toaggregatekernel density changes Changesfrom Changesfrom Changesfrom 1820-47to1919-38 1820-47to1870-93 1870-94 to1919-38 Intra-regionalchangesin the amounts of wealth 0.203 0.181 0.092 Confidenceinterval0.176 0.230 0.163 0.200 0.0830.110 Changesin population structure 0.1460.076 0.073 Confidenceinterval0.132 0.1610.063 0.089 0.062 0.088 Intra-regionalpuredistribution changes 0.099 0.0430.089 Confidenceinterval0.081 0.1160.036 0.0650.081 0.107 Intra-regionalmeanschanges 0.166 0.1610.026 Confidenceinterval0.1500.188 0.144 0.1790.0130.043 Intra-regionalchangesin sharesof assetowners 0.089 0.032 0.046 Confidenceinterval0.0750.106 0.018 0.044 0.0340.061 Pureurbanization effect 0.066 0.048 0.029 Confidenceinterval0.0610.0710.0430.054 0.0250.035 Note :The relativeimportanceofeachfactorismeasured bythe numericalintegration of absolutevaluekernel density changesattributable toeachcomponent.Valuesshowncorrespond tothe choiceofweights of equation(5). Confidence intervals(at95%) werecomputed viabootstrapresampling thatassessesthe robustness of the results withrespectto the samplepoints used todrawthe distribution curve. Source :Trimmed TRA survey

Overall,the evolution of wealthinFrancefrom 1820-47to1919-39shownin figure1 isessentiallydetermined bydistributionalchanges, C D (y) ,and more particularlybyregionalmeanschangesin distribution ( C M (y) ,thatis,growthand the rural/urbanwealthgap). However,the relativeimportanceofthiscomponentisnot constantovertime. During the nineteenthcentury,thiswasclearlythe most influentialfactor(withanintegralvalueof0.16,three timeslargerthanthe next most importantfactor),but from 1870-94 tothe interwarperiod,relativepositionschanged and the dominantexplanatory factorbecame C I (y) ,the intra-regionalpuredistribution changes(withanintegralvaluenowfour timeslargerthanthe former). Inotherwords, after1870,the purechangesin the wealthdistribution functionsin eachregion (the shape of the curvesitself) wasmoreimportantthanthe meandifferenceacross regions oreven globalenrichment.

Changesin the population structure, C P (y) ,though notnegligible,played a relativelyless importantrole throughout the period,especiallybeforethe turnofthe nineteenthcentury (withintegralvaluesof about 0.07,while distributionalchanges

21 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 wereupto0.18). From 1870-94 tothe interwarphase,its relativeweightin explaining wealthchangesincreased,the differencebetween the distribution and population changesbecoming insignificantatthe 95% confidencelevel to bootstrapped resampling. Inthislast period,changesin bothlocation and in the regionalshareofassetowners became nearlyasimportantasthe evolution of the amounts of wealth. 5. Conclusion Thispapertracesbackthe evolution of wealthdistribution in Franceinthe nineteenth century and the first half of the twentiethcentury and measuresthe role of urbanization. According toKuznets,income inequality first increasesthen decreases along the developmentpath. The process isdetermined,in adualframework,bythe transition from agriculturetoindustry.Three parameters summarizethistransition in Kuznets’framework:the sectoralshareofpopulation,the meanincome gapand within-sectorinequality.Our results conformwithothers in the literaturein underlining thatotherparameters need tobeincluded in the analysisof the relationship between inequality and growth. Thispapershows that,in termsof wealthatdeath,anothersignificantparameter should beconsidered:the shareofthe population leaving no estateatall. Thisshareof the population leaving no estatematters in explaining the overall distribution. It increased by14 percentage points between 1820 and 1939,throughout all the population,independentof age group,genderorsectorof activity.However,in Paris, thereweretwoseparatetrends:anincreaseof8percentage points of the shareofthose leaving no inheritancebetween 1820-1847and 1870-1913wasfollowed byadecrease of the same magnitude afterWWI. Contrary tothe Kuznetsianassumption of aconstantrural-urbangap(atleast for income),the wealthgapwasneitherconstantovertime,norhomogeneous across differentlevelsof wealth. Forthe top of the distribution –oralmost the top,tobe precise–,the ratio of relativeurban/ruralwealthfollowed aninverted U-shape,with the maximumbeing reached in 1870-1894. Onthe contrary,forthosewithmedian wealthorbelow,the rural/urbanwealthgapfluctuated overtime,without any particularpattern. Itisimportanttonotethatasthe meanlevel of wealthisinfluenced bythe extreme valuesof the upperend of the distribution,suchspecificevolution at lowerlevelsof wealthisneglected when one looksonlyataggregateinequality indices and standardinequality decompositions,suchasthoseobtained using Theil indicesand based on meanwealth. Thus,anon-parametricanalysiswasperformed,thatprovidesa detailed evaluation of the whole distribution. Wealthdensity changesaredecomposed intochangesin the localization of the population and changesin the amounts of wealth;the latterwasthe dominantfactorin explaining overall wealthevolution over the period of study.However,the role of urbanization hasincreased overtime,by whichwemeannotonlypopulation migrationsbut alsothe change in the regional sharesof assetowners.The patternofdistribution changed too,atleast in urbanareas. Inprovincialtowns,therewasaconcentration in the intermediatepart of the distribution,withahollowing out of the density curveinthe lowerand upperendsof the distribution. InParis,the shape of the density curveevolved from abi-modaltoa

22 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25 uni-modaldistribution. Foroverall France,the density function of wealthdistribution moved tothe right(wealthowners gotricherovertime) witharising mass atzero (thereweremoreand moreindividualsleaving no inheritanceatall). Putting both elements together,wecanconclude toapolarization of Frenchsociety between those who hadnone and thosewho werebetteroff. Finally,the paperhighlights severaladditionalmeasurementissues.First,itis difficulttorelatepreciselyanindividualand his/herassets toone given location. Indeed,realestateisof coursepreciselylocated. But,itmayhappen thataruralestate isowned byancity dweller.The diffusion of financialassets throughout the nineteenth century makesthislocalization even morecomplex.Itisnotuncommon tosee aperson living in aremoteruralareacorresponding withbanksornotariesin the city and investing in financialassets.Second,in thispaper,aperson isdefined asruralifshe/he died in aruralarea.Thisdefinition mightoverlook life-time migration. Forinstance,a person borninthe countryside who wenttothe city,succeeded thereand returned backtohis/herbirthplaceforretirementwill belabeled rural. Itisalsopossible that rural-urbanmigrants wereselected from the poorest of the ruralpopulation,oron the contrary,from the richest (thosethatcould affordthe cost of migration). Thispoint extents toabroaderissue,thatof wealthmobility.Take forinstance,familieswho stood atthe top of the wealthdistribution beforeWWI.Inwhatpart of the distribution could theirancestors befound in the earlynineteenthcentury?These extensionsdeservefurtherresearch.

References Atkinson A.B.,Harrison A.J.(1978) The Distribution of PersonalWealthinBritain, Cambridge,Cambridge University Press,xiv+330 p. BodinierB.,TeyssierE.(2000) L’événementle plus importantde laRévolution :laventedes biensnationaux (1789-1867)enFranceetdanslesterritoiresannexés ,[PréfacedeJ.M. Moriceau,participation de FrançoisAntoine], Collection Mémoiresetdocuments d’histoiredelaRévolution française,Paris,Ed. duCTHS, Sociétédesétudes robespierristes,503 p. BourdieuJ.,Postel-VinayG.etSuwa-Eisenmann A.(2004) Défenseetillustration de l’enquêtedes3000 familles.L’exemple de son voletpatrimonial, Annalesde démographie historique 107(1),19-52. BourdieuJ.,Postel-VinayG.etSuwa-Eisenmann A.(2003)Pourquoi larichessenes’est- elle pasdiffusée aveclacroissance?Ledegrézérodel’inégalitéetson évolution en France,1800-1940, HistoireetMesure XVIII(1/2),147-198. Datt G.,Ravallion M.(1992)Growthand redistribution components of changesin poverty measures:Adecomposition withapplicationstoBrazil and Indiain the 1980s, JournalofDevelopmentEconomics 38(2),275-295. DaumardA.(1973) LesfortunesfrançaisesauXIXe siècle,Paris,Ed. Mouton. DeVriesJ.(1984) EuropeanUrbanization,1500-1800,Cambridge,HarvardUniversity Press,xvii+398 p.

23 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

DiNardo J.N.,Fortin N.and Lemieux T.(1996)Labormarketinstitutionsand the distribution of wages,1973-1992:Asemiparametricapproach, Econometrica 64 (September),1001-1044. DubyG.(éd.) (1983) HistoiredelaFranceurbaine. La ville àl’âge industriel,Paris,LeSeuil, 671p. EstebanJ.M.,RayD.(1994) Onthe measurementof Polarization, Econometrica 62 (July), 819-851. GranthamG.(1996)The Frenchagriculturalcapitalstock,1789-1914, Researchin EconomicHistory 16,39-89. HohenbergP.M.(2004) The historicalgeographyof Europeancities:Aninterpretative essay, in:Handbook of Regionaland UrbanEconomics, Henderson J.V.,Nijkamp P., MillsE.S.,CheshireP.C.and ThisseJ.-F.(eds),Elsevier,volume 4,3021-3052. JenkinsS.P.,Jäntii M.(2005) Methodsforsummarizing and comparing wealth distributions,ISER Working Paper2005-5,40p. JenkinsS.P.,VanKermP.(2005) Accounting forincome distribution trends:Adensity function decomposition approach, JournalofEconomicInequality 3(April),43-61. Kakwani N.(1993)Poverty and economicgrowth,withapplication toCôted’Ivoire, Reviewof Income and Wealth 39(June),121-139. Kuznets S.(1955) Economicgrowthand economicinequality, AmericanEconomic Review 45 (March),1-28. LewisA.W.(1954) Economicdevelopmentwithunlimited suppliesof labour, ManchesterSchool 22 (March),139-91. Piketty T.(2001) Leshauts revenus en Franceau20 e siècle -Inégalitésetredistributions1901- 1998, Paris,Grasset,807 p. Piketty T.,Postel-VinayG.and RosenthalJ.L.(2006)Wealthconcentration in a developing :Parisand France,1807-1994, AmericanEconomicReview 96 (March),236-256. Piketty T.,Postel-VinayG.and RosenthalJ.L.(2004) Wealthconcentration in a developing economy:Parisand France,1807-1994,CenterforEconomicPolicy Research,Discussion Paper4631. Pirenne H.(1927) LesvillesduMoyen Âge,essaid’histoireéconomiqueetsociale,Bruxelles,Ed. Lamertin,203 p. Roine J.,Waldenström D.(2007)Wealthconcentration overthe pathofdevelopment: 1873-2005,Working PaperSeriesin Economicsand Finance677, Stockholm School of Economics. RosenthalJ.L.(2002)The sizeofthe Ante:Inequality,financialmarkets and growthin Paris1780-1907,mimeo. WeirD.R.(1991) Lescriseséconomiquesetlesoriginesde laRévolution française, AnnalesESC 4(Juillet-Août),917-47. Williamson J.G.,Lindert P.(1980)Long-termtrendsin Americanwealthinequality, in: Modeling the Distribution and IntergenerationalTransmission of Wealth ,SmithJ.D. (ed.),Chicago,University of Chicago Press,9-94.

24 J.Bourdieuetal. -Reviewof Agriculturaland EnvironmentalStudies,87(2008-2),5-25

APPENDIX

Table A1. Comparing datasets 1820-18471848-18691870-1894 1895-19131919-1939 OriginalTRA sample Total76567927 1079375318026 Assetowners (1),wealthknown4618 4836 3054 918 1121 Assetowners,wealthunknown628448 3530 3373 3244 Noasset24102643420932403661 Reweighted TRA sample (2) Total76567927 10789 7515 8026 Assetowners 52465284 658042754365 Noasset24102643420932403661 Reweighted and trimmed (at99thpercentile) sample (3) Total757878751068774177924 Assetowners 51685232 64784177 4263 Noasset24102643420932403661 Largest wealthvalue1073 0471626 8972374914 2539507 1126 396 Notes : (1) Assetowners aredefinedasdecedents leavingapositivebequest. (2) Anydifferenceinthe numberof asset owners beforeand afterreweighting isduetorounding. (3) The thresholdsused fortrimming arethosereported by Piketty etal. (2004). Source :TRA survey

Table A2. Sharesof population bywealthinterval(%) 1820-18471848-18691870-1894 1895-19131919-1939 Noassets 31.81 33.55 39.3843.67 46.21 Less than1000 Euros18.4213.3810.91 7.41 7.24 Between 1000 and 10000 34.0133.02 27.5023.0126.40 Between 10000 and 50000 13.03 15.98 16.9617.70 15.36 Between 50000 and 100 000 1.70 2.532.91 4.253.28 Between 100 000 and 500 000 1.03 1.55 2.343.961.50 Source :TRA survey

25