top incomes over the twentieth centurypiketty.pse.ens.fr/files/atkinsonpiketty2007.pdf · on...

TOP INCOMES OVER THE TWENTIETH CENTURY

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page i 2.12.2006 9:19pm

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page ii 2.12.2006 9:19pm

Top Incomes over theTwentieth Century:

A Contrast Between European andEnglish-Speaking Countries

Edited by

A. B. ATKINSON,

NuYeld College, Oxford,

and

T. PIKETTY,

PSE, Paris

1

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page iii 2.12.2006 9:19pm

3Great Clarendon Street, Oxford ox2 6dp

Oxford University Press is a department of the University of Oxford.It furthers the University’s objective of excellence in research, scholarship,

and education by publishing worldwide in

Oxford New York

Auckland Cape Town Dar es Salaam Hong Kong KarachiKuala Lumpur Madrid Melbourne Mexico City Nairobi

New Delhi Shanghai Taipei Toronto

With oYces in

Argentina Austria Brazil Chile Czech Republic France GreeceGuatemala Hungary Italy Japan Poland Portugal SingaporeSouth Korea Switzerland Thailand Turkey Ukraine Vietnam

Oxford is a registered trade mark of Oxford University Pressin the UK and in certain other countries

Published in the United Statesby Oxford University Press Inc., New York

! Oxford University Press, 2007The moral rights of the author have been assertedDatabase right Oxford University Press (maker)

First published 2007

All rights reserved. No part of this publication may be reproduced,stored in a retrieval system, or transmitted, in any form or by any means,

without the prior permission in writing of Oxford University Press,or as expressly permitted by law, or under terms agreed with the appropriate

reprographics rights organization. Enquiries concerning reproductionoutside the scope of the above should be sent to the Rights Department,

Oxford University Press, at the address above

You must not circulate this book in any other binding or coverand you must impose the same condition on any acquirer

British Library Cataloging in Publication Data

Data available

Library of Congress Cataloguing in Publication Data

Data available

Typeset by SPI Publisher Services, Pondicherry, IndiaPrinted in Great Britain

on acid-free paper byBiddles Ltd, King’s Lynn, Norfolk

ISBN 0–19–928688–4 978–0–19–928688–1

1 3 5 7 9 10 8 6 4 2

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page iv 2.12.2006 9:19pm

Contents

List of Figures, Tables, and Boxes vi

Preface xv

1 Top Incomes over the Twentieth Century:A Summary of Main Findings 1T. Piketty

2 Measuring Top Incomes: Methodological Issues 18A. B. Atkinson

3 Income, Wage, and Wealth Inequality inFrance, 1901–98 43T. Piketty

4 The Distribution of Top Incomes in theUnited Kingdom 1908–2000 82A. B. Atkinson

5 Income and Wage Inequality in theUnited States, 1913–2002 141T. Piketty and E. Saez

6 The Evolution of High Incomes inCanada, 1920–2000 226E. Saez and M. Veall

7 The Distribution of Top Incomes in Australia 309A. B. Atkinson and A. Leigh

8 The Distribution of Top Incomes in New Zealand 333A. B. Atkinson and A. Leigh

9 Top Incomes in Germany Throughout theTwentieth Century: 1891–98 365F. Dell

10 Top Incomes in the Netherlands over theTwentieth Century 426W. Salverda and A. B. Atkinson

11 Income and Wealth Concentration in Switzerlandover the Twentieth Century 472F. Dell, T. Piketty, and E. Saez

12 Long Term Trends in Top Income Shares in Ireland 501B. Nolan

13 Towards a UniWed Data Set on Top Incomes 531A. B. Atkinson and T. Piketty

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page v 2.12.2006 9:19pm

List of Figures, Tables, and Boxes

FIGURES

1.1 The fall of top capital incomes in France, 1913–98 10

1.2 The top 1% income shares in France, the UK, and the US, 1913–98 12

1.3 Wealth concentration in Paris and France overall, 1807–1994 16

2.1 Share of top 1% and overall Gini coeYcient in US, 1947–2002 20

2.2 ‘Taxable capacity’ of top 1% in the UK, 1937–2000 21

2.3 Globally rich as % world population, 1910–92 25

2.4 Personal income control totals for the UK, 1908–99 31

2.5 Interpolation into open upper interval, UK 2000 data 33

3.1 The top decile income share in France, 1900–98 48

3.2 The income shares of fractiles P90–95, P95–99, and P99–100 in France,1900–98 49

3.3 The top decile and the top percentile wage shares in France, 1913–98 53

3.4 Factor shares in France, 1913–98 58

3.5 The average estate left by fractiles P90–95 and P99.99–100 in France, 1902–94 60

3.6 EVective average income tax rates in France, 1915–98 62

4.1 Share of total gross income of the top 0.05%, 0.1%, and 0.5% in the UK,1908–2000 92

4.2 Share of total gross income of the top 1%, 5%, and 10% in the UK,1908–2000 95

4.3 EVect on share of top 1% of adjustment for retained earnings,UK 1937–65 101

4.4 Shares within shares, UK 1918–2000 102

4.5 Pareto-Lorenz coeYcients, UK 1908–2000 103

4.6 Share of total personal after tax income of the top 0.05%, 0.1%, and 0.5%,UK 1937–2000 105

4.7 Share of total personal after tax income of the top 1%, 5%, and 10%,UK 1937–2000 106

4.8 Percentage reduction in after tax shares compared with before tax shares,UK 1937–2000 106

4.9 Composition of adjusted total income, UK 1949–2000 108

4.10 Composition of income for diVerent groups, UK 1937–98 109

4.11 Composition of income of top 1%, UK 1937–2000 110

4.12 Contribution to share of top 1%, UK 1949–2000 111

4.13 Shares of top earners and top wealth holders in UK, 1923–2000 112

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page vi 2.12.2006 9:19pm

5.1 The top decile income share, US 1917–2002 147

5.2 The income shares of P90–95, P95–99 and P99–100 in US, 1913–2002 147

5.3 The top 0.01% income share, US 1913–2002 149

5.4 Income composition of top groups within the top decile in US, 1929and 1999 151

5.5 The capital income share in the top 0.5% in US, 1916–99 153

5.6 Capital income in the corporate and personal sector, US 1929–2003 154

5.7 The top 0.1% wealth share in US, 1916–2000 156

5.8 The top decile wage income share, US 1927–2002 159

5.9 Wage income shares for P90–95, P95–99, and P99–100 in US, 1927–2002 159

5.10 Shares of oYcers compensation and wage shares, P99.5–10 andP99–99.9 in US, 1917–60 160

5.11 CEO pay vs. average wage income, US 1970–2003 163

5.12 Top 0.1% income shares in the US, France, and the UK, 1913–98 166

5A.0 Average real income and consumer price index, US 1913–2002 170

5A.1 Average real income of bottom 90% and top 1% in US, 1917–2002 174

5A.2 Top 1% income shares in US: The role of capital gains 1913–2002 175

6.1 Average real income and consumer price index in Canada, 1920–2000 230

6.2 Top income shares in Canada, 1920–2000 233

6.3 The income shares of the top income groups in Canada and US 1920–2000 234

6.4 Capital income in the corporate and the personal sector in Canada,1926–2000 237

6.5 Salary vs. wage earners in manufacturing sector in Canada, 1915–48 239

6.6 Income composition of top groups within the top decile in Canada,1946 and 2000 240

6.7 The share of wage income in upper income groups in Canada, 1946–2000 241

6.8 The top wage income shares in Canada, 1972–2000 243

6.9 The top 1% wage income share of Quebec Francophones vs. all Wlers fromthe rest of Canada, 1982–2000 246

6.10 Top 1% wage income share for individuals and families in Canada,1982–2000 247

6.11 The role of stock options in the surge in top wage income shares inCanada, 1995–2000 248

6.12 Mobility of high incomes in Canada, 1982–2000 251

6.13 Marginal income tax rates in Canada for various percentiles, 1920–2000 253

6.14 Marginal tax rates and income share for the top 0.1% in Canada and US,1960–2000 254

6A.1 Income shares with and without capital gains of top income groupsin Canada, 1972–2000 262

6A.2 Average income tax rates in Canada within top decile, 1920–2000 264

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page vii 2.12.2006 9:19pm

List of Figures, Tables, and Boxes vii

6A.3 Average income tax rates in Canada within top percentile, 1920–2000 264

7.1 Shares of top 1%, 0.5%, and 0.1%, Australia 1921–2002 317

7.2 Comparing Victoria, 1912–23, with Australia, 1921–31 318

7.3 Share of next 4% and second vintile in Australia, 1921–2002 319

7.4 Shares within shares in Australia, 1921–2002 320

7.5 Pareto-Lorenz coeYcients, Australia 1921–2002 321

7.6 Fraction of income from salary and wages, Australia 1954–2002 322

7.7 Contributions to share of top 1%, Australia 1954–2002 322

8.1 Shares of top 1%, 0.5%, and 0.1% in New Zealand, 1921–2002 342

8.2 Shares of next 4% and second vintile in New Zealand, 1921–2002 342

8.3 Comparison with other top income groups in New Zealand, 1921–2002 344

8.4 Comparison with other studies of New Zealand: shares of top 10%and 20%, 1921–2002 348

8.5 Shares within shares in New Zealand, 1921–2002 349

8.6 Pareto-Lorenz coeYcients, New Zealand 1921–2002 349

9.1 Series of Müller and Geisenberger (1972) for Prussia 367

9.2 Share of the top decile, Germany 1891–1998 371

9.3 Share of P90–95 and P95–99, Germany 1891–1998 376

9.4 Share of the top percentile, Germany 1891–1998 376

9.5 Share of P99–99.5, P99.5–99.9, and P99.9–99.9, Germany 1891–1998 378

9.6 Share of the top 0.01%, Germany 1891–1998 378

9.7 Share of the top percentile within the top decile, France, US, andGermany 1891–1998 379

9.8 Share of P99.99–100 in top percentile, Germany 1891–1998 379

9.9 Share of the bottom part of the top decile (P90–99), France, US,and Germany 1891–1998 380

9.10 Share of the top part of the top decile (P99–100), France, US,and Germany 1891–1998 380

9.11 Sources of income in top income groups in Germany, 1928 381





9F.1 German DAX index, 1988–2000 395

9F.2 German DAX index, 1950–2002 395

9F.3 Implicit capital gains in the last bracket, German tax data, 1961–98 396

9G.1 Evolution of the overall Prussian population; evolution of the shareof tax units actually Wling tax returns, 1891–1918 403

9G.2 Overall population, tax units, Weimar Republic, and Third Reich, 1925–38 403

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page viii 2.12.2006 9:19pm

viii List of Figures, Tables, and Boxes

9G.3 Overall population, households, and tax units, Federal Republic ofGermany, 1946–2002 404

9H.1 Net personal income of private households and total taxable incomeFederal Republic of Germany, 1950–98 406

9H.2 Aggregates of the German national accounts after the Second World Warand adjusted net personal income of private households, 1950–2004 406

9H.3 Unemployment in Germany, 1925–38 409

9H.4 Net personal income of private households and total taxable income,Weimar Republic and Third Reich 1925–38 410

9H.5 Average tax unit income over the twentieth century in Germany 414

10.1 Years for which data in the Netherlands, 1914–99 433

10.2 Real gross average tax unit income and consumer prices Netherlands,1914–2000 441

10.3A Gross income shares of top 10%, 5%, and 1%, Netherlands 1914–99 442

10.3B Gross income shares of top 0.5% and 0.1%, Netherlands 1914–99 442

10.3C Gross income shares of next 4% and second vintile group, Netherlands1914–99 443

10.4A Gross income shares within shares, Netherlands 1914–99 444

10.4B Gross income Pareto-Lorenz coeYcients of gross incomes, Netherlands1914–99 445

10.5 Disposable income shares within shares, Netherlands 1959–99 447

10.6 Ratio of disposable income to gross income top shares, Netherlands1959–99 447

10.7 Capital income shares within gross income of top 10%, 1%, and 0.1%,Netherlands 1952–99 450

10.8 Composition of top shares by source of income, Netherlands 1952, 1977,and 1999 450

10.9A Wage income contributions to gross income of top 10%, 1%, and 0.1%,Netherlands 1952–99 452

10.9B Wage income contributions to gross income of top 1% and 0.1%,Netherlands 1952–99 453

10.9C Wage income contributions to gross income of top 0.1%, Netherlands1952–99 453

10.10 EVective tax rates on gross income of top 10%, 1%, and 0.1%, Netherlands1914–99 457

10.11 Relative eVective tax rates on gross income of top 10%, 1%, and 0.1%(average ¼ 1), Netherlands 1914–99 458

10B.1 Tax units (x 1000), Netherlands 1914–99 466

10B.2 Control totals of gross income and known gross income as % of nationalaccounts personal income total, Netherlands 1914–99 467

11.1 Average real income and consumer price index in Switzerland, 1901–2000 488

11.2 Top 10% and top 5% income shares in Switzerland, 1933–96 488

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page ix 2.12.2006 9:19pm

List of Figures, Tables, and Boxes ix

11.3 Top 1%, top 5–1%, and top 10–5% income shares in Switzerland, 1933–96 489

11.4 Top 0.1%, top 0.5–0.1%, and top 1–0.5% income shares in Switzerland,1933–96 490

11.5 Shares within shares in Switzerland, 1933–63 490

11.6 The top 0.1% income share in France, the US, and Switzerland, 1933–97 491

11.7 Top 10–5%, top 5–1%, and top 1% wealth shares in Switzerland, 1913–97 492

11.8 Top 1–0.5%, top 0.5–0.1%, and top 0.1% wealth shares in Switzerland,1913–97 493

11.9 The top 1% wealth share in the US and Switzerland, 1915–2000 493

11.10 The fraction of foreign income earners and non-residents in topincome groups Switzerland, 1957–91 496

12.1 Share of top 0.1% in total income, Ireland 1922–90 511

12.2 Shares of top 1% and top 0.5% in total income, Ireland 1938–2000 512

13.1A Share of top 10% in English speaking countries 540

13.1B Share of top 10% in continental European countries 540

13.2A Share of top 1% in English speaking countries 541

13.2B Share of top 1% in continental European countries 541

13.3A Share of top 0.1% in English speaking countries 542

13.3B Share of top 0.1% in continental European countries 542

13.4A Share of top 1% in income of top 10% in English speaking countries 543

13.4B Share of top 1% in income of top 10% in continental European countries 544

TABLES

1.1 Raw top income tabulations, France 1919 (originally published inBulletin de statistique et de legislation compare, March 1923, tome 93) 4

1.2 Raw income composition tabulations, France 1919 (originally publishedin Bulletin de statistique et de legislation compare, March 1923, tome 93). 6

1.3 The age proWle of wealth at death in Paris, 1817–1994 17

2.1 Example of income tax data: UK super-tax 1911–12 26

3.1 Income growth and income shares in France, 1900–10 and 1990–98 50

3.2 The impact of progressive taxation on capital accumulation 64

3A.1 Top income shares in France, 1900–98 (I) 71

3A.2 Top income shares in France, 1900–98 (II) 73

3A.3 Sources for French income tax data, 1915–98 75

3A.4 Income and population totals for France, 1900–98 77

4.1 Shares in total before tax income, UK 1908–2000 93

4.2 Shares in total after tax income, UK 1937–2000 104

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page x 2.12.2006 9:19pm

x List of Figures, Tables, and Boxes

4A.1 Sources for UK super-tax and surtax data, 1908–72 115

4A.2 Sources of UK SPI data, 1918–2000 117

4B.1 UK control totals for tax units (individuals) and income, 1908–2000 126

4C.1 Derivation of control totals (£ million) for income in theUK 1945/46–2000/01 132

4C.2 Derivation of control totals (£ million) for income in the UK,1908/09–44/45 135

5.1 Thresholds and average incomes in top income groups, US 2000 144

5.2 Shares of each occupation within the top 1% in US, 1916 152

5A.0 Reference totals for tax units and income, US 1913–2002 171

5A.1 Top fractiles income shares (excluding capital gains), US 1913–2002 176

5A.2 Top fractiles (deWned excluding capital gains) income shares(including capital gains), US 1913–2002 179

5A.3 Top fractiles (deWned including capital gains) income shares(including capital gains), US 1913–2002 182

5A.4 Top fractiles income levels (excluding capital gains), US 1913–2002 185

5A.5 Top fractiles (deWned excluding capital gains) income levels(including capital gains), US 1913–2002 188

5A.6 Top fractiles (deWned including capital gains) income levels(including capital gains), US 1913–2002 191

5A.7 Income composition by fractiles of total income, US 1916–99 199

5A.8 Capital gains by fractiles of total income, US 1916–2002 207

5B.1 Aggregate series on wage income, US 1917–2002 211

5B.2 Top wage income shares, US 1927–2002 215

5B.3 Average salary and threshold for each fractile (in 2000 dollars),US 1927–2002 217

5B.4 CEO pay vs. average wage, US 1970–2003 220

6.1 Thresholds and average incomes in top groups within the top decilein Canada in 2000 229

6.2 Marginal tax and US eVects on Canadian top income shares, 1920–2000 256

6A.1 Reference totals for population, income, and inXation in Canada, 1920–2000 259

6B.1 Top income shares in Canada, 1920–2000 266

6B.2 Top income shares including capital gains in Canada, 1972–2000 269

6B.3 Top fractile income levels (excluding capital gains) in Canada, 1920–2000 271

6C.1 Shares of total tax returns in each occupation in Canada, 1920–41 278

6C.2 Shares of each occupation within the top 10% in Canada, 1942 279

6C.3 Income composition by fractiles of total income (excluding capital gains)in Canada, 1946–2000 280

6C.4 Share of capital gains in total income for upper groups in Canada,1972–2000 285

6D.1 Aggregate series on wages in Canada, 1972–2000 288

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page xi 2.12.2006 9:19pm

List of Figures, Tables, and Boxes xi

6D.2 Shares of wage income for upper groups in Canada, 1972–2000 289

6D.3 Average wage income and threshold for each fractile (in 2000 Canadiandollars) in Canada, 1972–2000 292

6D.4 Top wage income shares, Francophones in Quebec vs. all Wlers from restof Canada, 1982–2000 294

6D.5 The role of stock options in top wage income shares in Canada, 1995–2000 296

6E.1 High income mobility in Canada, 1982–2000 298

6F.1 Marginal income tax rates in Canada, 1920–2000 301

6F.2 Average tax rates in upper groups in Canada, 1920–2000 304

7.1 Top income shares, Australia 1921–2002 315

7.2 Top income shares, Victoria, Australia, 1912–23 318

7A.1 Population totals for Australia, 1912–2002 324

7B.1 Personal income totals for Australia, 1912–2002 327

7C.1 Sources of income tax data for Australia, 1921–2002 329

7C.2 Sources of income tax data for Victoria, Australia, 1912–23 330

8.1 Top income shares, New Zealand 1921–2002 340

8.2 Top income percentiles (% mean), New Zealand 1921–2002 345

8A.1 Sources of income tax data for New Zealand, 1921–2002 352

8B.1 New Zealand population totals (thousands), 1921–2002 356

8C.1 New Zealand personal income totals and coverage, 1921–2002 359

8D.1 New Zealand comparison groups for top income shares, 1921–2002 361

9A.1 Income tax publications used, Germany 384

9C.1 Tax units (Tu) in the micro-data set for Germany in the 1990s 385

9C.2 The accuracy of quantile estimation for Germany in the 1990s 386

9F.1 Capital gains and the various aggregates, Germany 1992 392



9G.1 Tax units (Tu) control total for Prussia, 1891–1918 399

9G.2 Tax units (Tu) control total, Germany 1891–1998 401

9H.1 Income control total for Prussia, 1891–1918 411

9H.2 Income control total, 1891–1998 412

9I.1 Nominal thresholds and nominal average income of top income groups,Prussia 1891–1918 415

9I.2 Nominal thresholds and nominal average income of top income groups,Germany 1925–38 416

9I.3 Nominal thresholds and nominal average income of top income groups,Federal Republic of Germany 1950–98 (1) 417


Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page xii 2.12.2006 9:19pm

xii List of Figures, Tables, and Boxes


9I.6 Top income shares, Germany 1891–1998 (1) 420



10.1 Overview of income tax data sources for the Netherlands 432

10.2 Top shares in gross income, Netherlands 1914–99 434

10.3 Top shares in disposable income by range of disposable income,Netherlands 1959–99 446

10.4 Composition of gross income top shares by source of income,Netherlands 1952–99 449

10.5 Composition of aggregate gross income by socio-economic categoryof receiving tax unit, Netherlands 1952, 1977, and 1999 451

10.6 EVective top share tax rates, Netherlands 1914–99 455

10A.1 Sources for data on total gross income and summary statistics,Netherlands 1915–99 460

10A.2 Sources for data on disposable income and summary statistics,Netherlands 1959–99 463

10B.1 Population totals (thousands), Netherlands 1914–99 464

10B.2 Reference income totals (million guilders) and prices, Netherlands 1914–99 468

11.1 Reference totals for population, income, and inXation in Switzerland,1901–2002 480

11.2 Top income shares in Switzerland, 1933–95/96 484

11.3 Top wealth shares in Switzerland, 1913–97 486

11.4 Fraction of non-residents and residents with income abroad in topincome groups in Switzerland, 1949/50–1991/92 495

11.5 Capital income earned through Swiss accounts and tax evasion, 1950–2002 497

12.1 Sur-tax payers classiWed by income ranges, Ireland 1936–37 503

12.2 Personal income classiWed by income ranges, Ireland 1938 and 1943 504

12.3 Income tax payers classiWed by income ranges, Ireland 2000 505

12.4A Control totals for number of tax units, Ireland 1922–2000 507

12.4B Control totals for income, Ireland 1922–2000 509

12.5 Shares of top income groups, Ireland 1922–2000 513

12.6 Top income shares estimated from ‘gross incomes’, Ireland 1989/90–2000 519

12.7 Composition of top incomes, Ireland 1989/90 and 2000 520

12.8 Share of top income groups in top incomes, Ireland 1938–2000 521

12A.1 Source of income data used in deriving ‘total’ income shares,Ireland 1922–2000 523

12A.2 Source of income data used in deriving ‘gross’ income shares,Ireland 1989–2000 526

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page xiii 2.12.2006 9:19pm

List of Figures, Tables, and Boxes xiii

12A.3 Estimated indices for national income in money terms, ‘general prices’,and ‘real income’ 526

13.0 Key features of estimates for ten countries 533

13.1 Shares in total before tax income, France 545

13.2 Shares in total before tax income, UK 547

13.3 Shares in total before tax income, US 549

13.4 Shares in total before tax income, Canada 551

13.5 Shares in total before tax income, Australia 553

13.6 Shares in total before tax income, New Zealand 555

13.7 Shares in total before tax income, Germany 557

13.8 Shares in total before tax income, Netherlands 559

13.9 Shares in total before tax income, Switzerland 561

13.10 Shares in total before tax income, Ireland 563

BOXES

2.1 Pareto distribution 24

10.1 Summary of approach adopted in Netherlands estimates 440

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page xiv 2.12.2006 9:19pm

xiv List of Figures, Tables, and Boxes

Preface

The origins of this volume, and the companion volume to follow, lie in the studyof top incomes in France over the twentieth century published by one of us (TP)in 2001. The study used data from income tax and other sources to show theevolution of income inequality over a much longer continuous period thanhad previously been investigated (see Piketty 2001). This study, summarized inChapter 3, inspired the other editor (ABA) to examine the same topic forthe United Kingdom, and the results are presented in Chapter 4. Piketty andEmmanuel Saez extended the comparison further by making estimates for theUnited States (summarized in Chapter 5). Since then, the fruitfulness of incometax data in providing long run evidence about the top of the distribution has ledto estimates being constructed for a sizeable number of countries (covered here inChapters 6 to 12 and in a forthcoming second volume).

The aim of the project is to assemble in one place the studies of top incomes fora wide range of countries (ten in this volume). A number of the chapters arebased on research that has already been published in journal articles (see theBibliography, Chapters 1 and 2 in this volume), but the present versions containmore extensive accounts of the sources and methods as well as further and, insome cases, more recent results. Present journal editorial practice does nottypically allow space for full documentation of methods, but we believe that itis important that these be recorded and discussed. The preparation of neweconomic data such as those presented here involves a large number of operationsand recourse to a diversity of sources. Along the way, the data constructor hasinevitably had to make assumptions and corrections; it is not simply a matter ofcopying tables. If this process is not documented in full, then the reader is unableto assess the validity of the Wnal series. We have therefore encouraged authors toexplain their methods in detail.

The volume is not intended to be a comparative study. Although a number ofthe chapters refer to evidence for other countries, it will be clear that each countrystudied has its own speciWcities with regard to systems of income taxation, to theways in which data are collected, and to the wider processes of income deter-mination. We cannot assume that the series are fully homogeneous acrosscountries, and the literature on cross-country growth regressions warns us ofthe pitfalls in merging data without regard to the speciWcities of both data andreality. The emphasis is therefore on the historical experience of each of the tencountries. At the same time, as discussed in Chapter 1, the studies presented hererepresent a necessary Wrst stage in any comparative analysis. The series wereconstructed by using the same raw data sources for all countries and applyingthe same methodology to derive the Wnal series. Although fully homogenous,cross-country data sets do not exist, we have done our best to make our database

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page xv 2.12.2006 9:19pm

as homogenous as possible, and to provide users with adequate guidance andtechnical information. We have therefore, in the Wnal chapter (Chapter 13),assembled the key series for the ten countries. In the second volume, we hopeto cover the Nordic countries, countries from Southern Europe, India, China,Brazil, and Indonesia, which will extend considerably the range of experience.

The bibliographic references for the Wrst two chapters are grouped together,but we have kept separate bibliographies for the individual country chapters(even though this means some duplication,) on the grounds that some readersmay only be interested in one country, and wish to see the sources for thatcountry collected together.

A number of the chapters were presented at a conference organized as part ofthe CHANGEQUAL network meeting at NuYeld College, Oxford, in September2003. Atkinson worked on the Wnal preparation of the manuscript while holdinga Chaire Blaise Pascal at ENS-PSE. The editors would like to thank Lin Sorrell andCathy Douglas for their help at NuYeld, and the authors for their contributionsand patience.

A.B. Atkinson and T. Piketty

REFERENCE

Piketty, T. (2001). Les hauts revenus en France au XXe sièle: inégalités et redistributions,1901–1998. Paris: Grasset.

Atkinson & Piketty / Top Incomes over the 20th Century 00-Atkinson-prelims Page Proof page xvi 2.12.2006 9:19pm

xvi Preface

1

Top Incomes Over the Twentieth Century:A Summary of Main Findings1

T. Piketty

1.1 INTRODUCTION

This introductory essay presents some of the key Wndings and perspectivesemerging from the detailed country chapters published in this volume. Allchapters are part of a collective research project on the long-run dynamics ofincome and wealth distribution. The general objective of this project was toconstruct a high quality, long-run, international database on income and wealthdistribution using historical tax statistics. The resulting database now includesannual series covering most of the twentieth century for over 20 (mostly Western)countries. The present volume focuses upon the contrast between continentalEuropean countries and English-speaking countries and includes ten case studies:France, UK, US, Canada, Australia, New Zealand, Germany, the Netherlands,Switzerland, and Ireland. A forthcoming volume will complete the study bycovering Scandinavian and Northern Europe (including Sweden, Finland, andNorway), Southern Europe (including Italy, Spain, Portugal), as well as a numberof Latin American (including Argentina, Brazil) and Asiatic countries (includingIndia, China, and Indonesia).

The primary motivation for this project was a general dissatisfaction withexisting income distribution databases. The international databases on inequalitythat existed were not high quality (they display little homogeneity over time oracross countries),2 they are not long-run (typically they cover only a couple ofisolated years per country, generally restricted to the post-1970 or post-1980period), and they almost never oVer any decomposition of income inequality intoa labour income and a capital income component. This latter feature of existingdata sets is unfortunate, because the economic mechanisms at work can be very

1 The references to this chapter are given at the end of Chapter 2.2 See, e.g., the Atkinson-Brandolini (2001) criticism of the World Bank (Deininger-Squire) sec-

ondary database. The database is ‘secondary’ in the sense that it is based on the collection of inequalitymeasures computed by others using various income data sets and methodologies for diVerentcountries and time periods. In contrast, our inequality measures were computed by ourselves usingthe same primary data sources and methodology for all countries and time periods.

Atkinson & Piketty / Top Incomes over the 20th Century 01-Atkinson-chap01 Page Proof page 1 2.12.2006 8:15pm

diVerent for the distribution of labour income (demand and supply of skills,labour market institutions, etc.) and the distribution of capital income (capitalaccumulation, credit constraints, estate taxation, etc.), so that it is fairly heroic totest for any of these mechanisms using such data. The fact that existing databaseare not long run is also most unfortunate, because structural changes in incomeand wealth distributions are relatively slow and very often span over severaldecades. In order to properly understand such changes, one needs to be able toput them into broader historical perspective.3

Our database also suVers from strong limitations (in particular, our long-runseries are generally conWned to top income and wealth shares and contain littleinformation about bottom segments of the distribution), and fully homogenous,cross-country data sets do not exist. However, our database has the followingadvantages:

. we use the same raw data sources for all countries and apply the samemethodology to derive the Wnal series;

. the series are typically annual and cover a long-run of years;

. the data are mostly broken down by income source.

This means that they oVer a unique opportunity to understand better thedynamics of income and wealth distribution and the two-way interaction be-tween inequality and growth.

We should stress that the main objective of the chapters collected in thisvolume is to describe how the series were constructed, and to oVer Wrst cutanalysis of the long-run dynamics of inequality in each individual country. Suchanalytical narratives and detailed case studies are useful, but in our view theyshould be seen as complements (rather than substitutes) to a more systematicstatistical exploitation of the complete database, which we do not oVer in thisvolume. We very much hope that future researchers will use our database toexplore causal mechanisms in a more systematic way, and in particular that ourdata will contribute to renew the literature on cross-country inequality/growthregressions.4

The rest of this introductory essay is organized as follows. In section 1.2, webrieXy present the basic data and methodology used to construct the database.Section 1.3 presents some of the main descriptive Wndings and conclusions, withparticular emphasis to the Kuznets’ curve debate. Section 1.4 attempts to illus-trate how our database could potentially be used to renew the cross-countrystructural analysis of the interplay between inequality and growth, with betterhopes of success than the previous literature. We then discuss some of theprospects for extending the database using additional published historical taxtabulations and collecting historical individual tax data (Section 1.5).

3 This was Wrst stressed by Kuznets (1955).4 One of the key reasons why the literature on cross-country inequality/growth regressions failed to

deliver robust conclusions (see, e.g., Banerjee and DuXo (2003) for a critical appraisal) is the poorquality of existing databases.

2 T. Piketty AQ1


1.2 . CONSTRUCTING A NEW DATABASE: PRIMARY DATA

AND METHODOLOGY

Household income surveys are a relatively recent venture: they virtually did notexist on a national basis prior to 1950, and in most countries they are not availablein a homogenous, machine-readable format until the 1970s–80s. The only datasource that is consistently available on a long-run basis is tax data. Progressiveincome tax systems were set up in most Western countries at the beginning of thetwentieth century (1913 in the US, 1914 in France, etc.), and in all countries withan income tax system the tax administration started compiling and publishingtabulations based on the exhaustive set of income tax returns.5 These tabulationsgenerally report for a large number of income brackets the corresponding numberof taxpayers, as well as their total income and tax liability. They are usually brokendown by income source: capital income, wage income, business income, etc.

In order to give a sense of what our primary data sources look like, we reproduceon Table 1.1 the raw top income tabulations for France in 1919, as they wereoriginally published by the Finance Ministry. One can see for instance on this tablethat 181 French taxpayers reported tax income above one million francs in 1919(a pretty large income at that time). We also reproduce on Table 1.2 the raw incomecomposition tabulations for France in 1920. One can see that out of the 722 millionFrench francs reported by French taxpayers with individual income above 1 millionfrancs in 1920, 322 million francs took the form of ‘revenus des valeurs et capitauxmobiliers’ (interest and dividend income), 356 million francs took the formof ‘bénéWces industriels et commerciaux’ (business income), and only 2.2 millionfrancs took the form of ‘traitements publics et privés, salaires, etc.’ (wage income).

One can then use standard Pareto extrapolation techniques to compute topfractiles thresholds and average incomes using such data. This methodology isdescribed in a detailed manner in Chapter 2. Here it is suYcient to recall that thePareto law for top incomes is given by the following distribution function:

1! F(y) ¼ (k=y)a (k > 0, a > 1) (1:1)The corresponding density function is given by f (y) ¼ aka=y(1þ a). The keyproperty of Pareto distributions is that the ratio between the average income y *(y ) of individuals (or households or tax units) with income above y and y doesnot depend on the income threshold y :

y*(y) ¼h Z

z > yzf (z)dz

i=h Z

z > yf (z)dz

i

¼h Z

z > ydz=za

i=h Z

z > ydz=z(1 þ a)

i¼ ay=(a ! 1) (1:2)

i:e: y$(y)=y ¼ b, with b ¼ a=(a ! 1)

5 Full details about the administrative publications where the raw tabulations were originallypublished are given in the country chapters.


Top Incomes Over the Twentieth Century 3

Tab

le1.

1R

awto

pin

com

eta

bu

lati

on

s,Fr

ance

1919

(IM

PÔ

TG

ÉN

ÉR

AL

SUR

LE

RE

VE

NU

)A

Q2

MO

NT

AN

TD

ES

DÉ

DU

CT

ION

S

CA

TÉ

GO

RIE

SD

ER

EV

EN

US.

NO

MB

RE

de

co

nt

rib

ua

ble

sin

scri

tsd

ans

les

rôle

s.

MO

NT

AN

Td

esr

even

us

imp

osé

s.

po

ur

sit

ua

tio

nd

efa

mil

le.

po

ur

ch

ar

ges

de

fam

ille

MO

NT

AN

Tb

ru

td

el’i

mp

ôt.

MO

NT

AN

Td

espØ

na

lit

Øset

dro

its

ausu

s.1,

500

fr.

5,00

0fr

.1

23

45

67

8

fr.

fr.

fr.

fr.

fr.

fr.

6,10

0à

10,0

00fr

ancs

130,

787

1,17

0,32

4,80

012

3,91

5,00

07,

110,

000

34,4

06,0

003,

805,

400

170,

500

10,1

00à

20,0

0019

3,67

92,

851,

910,

400

417,

507,

000

25,4

10,0

0019

4,08

2,00

021

,056

,000

759,

100

20,1

00à

30,0

0058

,894

1,47

7,04

5,80

013

7,51

7,00

08,

983,

500

97,7

40,0

0018

,687

,300

755,

300

30,1

00à

50,0

0039

,974

1,52

9,51

2,70

093

,711

,000

6,23

5,50

079

,134

,000

40,0

61,4

001,

025,

200

50,1

00à

100,

000

23,8

821,

592,

572,

500

62,7

33,0

003,

354,

000

46,8

94,0

0094

,486

,600

1,90

7,70

010

0,10

0à

200,

000

9,48

71,

517,

031,

000

21,7

68,0

001,

513,

500

50,5

30,0

0014

2,41

3,80

02,

820,

500

200,

100

à30

0,00

02,

289

556,

396,

900

6,65

1,00

031

5,00

05,

456,

000

99,5

24,9

0096

5,90

030

0,10

0à

500,

000

1,38

852

7,73

4,80

03,

204,

000

138,

000

3,08

0,00

012

6,02

4,70

01,

228,

500

500,

100

à1

mil

lio

n57

638

7,08

2,90

01,

380,

000

46,5

001,

318,

000

130,

956,

900

1,68

0,80

0A

u-d

essu

sd

e1

mil

lio

n18

145

1,96

8,10

042

0,00

013

,500

336,

000

206,

785,

300

883,

400

To

ta

ux

467,

137

11,8

67,5

88,9

0086

8,91

1,00

055

,119

,500

492,

776,

000

883,

801,

200

12,1

77,0

00

(con

td.)


Tab

le1.

1(C

ontd

.)

MA

JOR

AT

ION

du

epa

rle

sc

on

tr

ibu

ab

les

cØl

Øba

ta

ires

.(2

5p

.10

0.)

MA

JOR

AT

ION

du

epa

rle

sm

Øna

ges

san

sin

fan

ts.

(10

p.

100.

)

No

mb

red

eco

ntr

ibu

able

ssu

pp

ort

ant

lam

ajo

rati

on

.

Mo

nta

nt

des

reve

nu

sd

esin

téré

ssés

Pro

du

itd

ela

maj

ora

tio

n.

No

mb

red

eco

ntr

ibu

able

ssu

pp

ort

ant

lam

ajo

rati

on

.

Mo

nta

nt

des

reve

nu

sd

esin

tére

ssés

.

Pro

du

its

de

lam

ajo

rati

on

.

MO

NT

AN

Td

esd

Ødu

ct

ion

sp

ou

rch

arge

sd

efa

mil

le.

PR

OD

UIT

NE

Tt

ot

al

de

l’im

pô

t.9

1011

1213

1415

16

fr.

fr.

fr.

fr.

fr.

fr.

45,1

9034

0,33

4,70

043

0,70

011

,900

111,

048,

300

70,0

0010

5,60

04,

570,

800

21,6

0230

1,51

8,90

087

5,60

029

,401

413,

678,

300

354,

800

727,

800

22,5

18,6

005,

162

130,

728,

500

1,02

6,20

06,

712

168,

608,

500

470,

500

096,

200

20,2

25,1

003,

398

132,

038,

600

1,07

3,90

04,

225

168,

148,

800

431,

000

1,80

1,20

040

,777

,300

2,04

914

3,37

0,60

02,

067,

800

2,40

716

8,39

0,60

01,

034,

800

5,02

8,80

095

,868

,100

746

99,9

47,6

003,

173,

800

904

125,

934,

300

1,55

7,00

010

,161

,500

139,

805,

600

167

39,9

50,6

001,

886,

300

196

46,7

76,5

0089

3,00

04,

704,

100

98,5

66,0

0011

433

,245

,200

2,15

3,30

012

345

,315

,200

1,13

7,00

03,

080,

000

127,

403,

800

3524

,508

,500

2,08

7,50

045

29,9

41,9

001,

086,

100

1,51

8,00

013

4,49

3,50

023

49,2

47,6

005,

993,

800

1733

,763

,300

1,50

6,70

033

6,00

021

4,83

3,20

0

To

ta

ux

78,4

921,

294,

870,

800

20,7

70,9

0055

,930

1,51

1,00

5,70

08,

590,

900

28,6

20,2

0089

6,71

9,80

0

Not

e:T

ab

lea

upr

ésen

tan

t,à

lada

tedu

30av

ril

1922

,la

déco

mpo

siti

on,

par

caté

gori

esde

reve

nu

s,de

sré

sult

ats

des

rôle

sét

abli

sau

titr

ede

l’an

née

1920

(rev

enu

sde

1919

)

Sou

rce:

Ori

gin

ally

pu

bli

shed

inB

ulle

tin

dest

atis

tiqu

eet

dele

gisl

atio

nco

mpa

re,

Mar

ch19

23:

vol.

93.


Tab

le1.

2R

awin

com

eco

mp

osi

tio

nta

bu

lati

on

s,Fr

ance

1919

(IM

PÔ

TG

ÉN

ÉR

AL

SUR

LE

RE

VE

NU

.)

CA

TÉ

GO

RIE

SD

ÉC

OM

PO

SIT

ION

DE

SR

EV

EN

US

GLO

BA

UX

SUIV

AN

TL

ES

DIV

ER

SES

SOU

RC

ES

D’O

ÙIL

SP

RO

VIE

NN

EN

T(a

).

de

rev

enu

s.

MO

NT

AN

Tt

ot

al

des

reve

nu

egl

ob

aus

(A)

RE

VE

NU

Sd

esp

rop

riét

ésb

âtie

s

RE

VE

NU

Sd

esp

rop

riet

ésn

on

bât

ies

RE

VE

NU

Sd

esva

leu

rset

cap

ilau

xm

ob

ilie

rs.

BÉ

NÉ

FC

LE

Sd

el’e

xplo

itat

ion

agri

cols

.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

12

34

56

78

910

mil

lio

ns.

mil

lio

ns.

%m

illi

on

s.%

mil

lio

ns.

%m

illi

on

s.%

6,10

0à

10,0

00fr

...

1,10

065

5.9

313.

114

813

.518

1.6

10,1

00à

20,0

003,

832

205

5.3

100

2.6

497

13.0

822.

120

,100

à30

,000

2,04

412

70.

363

3.1

301

17.7

472.

330

,100

à50

,000

2,13

214

26.

762

2.9

402

21.7

401.

950

,100

à10

0,00

02,

281

143

0.3

592.

658

625

.036

1.6

100,

100

à20

0,00

01,

803

975.

430

1.7

514

28.5

181.

020

0,10

0à

300,

000

751

344.

510

1.3

233

31.0

50.

730

0,10

0à

500,

000

699

203.

78

1.1

227

32.5

60.

950

0,10

0à

1m

illi

on

...

...

530

173.

24

0.7

186

35.1

40.

8A

u-d

essu

sd

e1

mil

lio

n..

.72

212

1.7

50.

732

244

.63

0.4

To

ta

ux

etm

oy

enn

es16

,897

868

5.5

375

2.4

3,53

622

.225

91.

6

(con

td.)


Tab

le1.

2(C

ontd

.)

BÉ

NÉ

FIC

ES

Ind

ust

riel

set

com

mer

ciau

x.

BÉ

NÉ

FIC

ES

de

l’exp

loit

atio

nm

iniè

res.

TR

AIT

EM

EN

TS

pu

bli

cset

pri

vés,

sala

ires

,et

c.

PE

NSI

ON

Sd

ere

nto

svi

rgù

res

BÉ

NÉ

FIC

ES

des

pro

fess

ion

sn

on

com

mer

cial

es.

RE

VE

NU

Sd

esch

arge

set

oY

ces.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

Mo

nta

nt.

Pro

po

rtio

n.

1112

1314

1516

1718

1920

2122

mil

lio

ns.

%m

illi

on

s.%

mil

lio

ns.

%m

illi

on

s.%

mil

lio

ns.

%m

illi

on

s.%

107

9.7

10.

166

560

.435

3.2

242.

23

0.3

820

21.6

30.

11,

873

48.9

721.

914

83.

826

0.7

651

31.8

30.

163

431

.029

1.4

102

5.0

271.

370

535

.02

0.1

502

23.5

221.

010

14.

734

1.6

800

39.0

20.

142

213

.517

0.7

924.

037

1.6

800

44.4

30.

225

714

.28

0.4

503.

120

1.1

353

47.0

10.

187

11.0

30.

418

2.4

71.

031

449

.21

0.1

689.

73

0.3

142.

03

0.5

200

50.2

10.

243

8.1

20.

28

1.5.

’’’’

356

49.3

30.

416

2.2

’’’’

50.

7’’

’’

To

ta

ux

etm

oy

enn

es5,

358

33.7

209.

14,

567

28.7

189

1.2

568

3.6

167

1.0

Not

es:I

MP

ÔT

ÉT

AB

LE

AU

TIT

RE

DE

L’A

NN

ÉE

1921

,—B

ÉN

ÉF

ICE

SB

YR

EV

EN

US

RÉ

AL

ISÉ

SA

UC

OU

RS

DE

L’A

NN

ÉE

1920

.Tab

leau

prés

enta

nt,

pou

rle

sco

ntr

ibu

able

sin

scri

tsda

nts

les

rôle

sém

isdu

1st

jan

vier

1921

au30

avri

l19

22,

ladé

com

posi

tion

dure

ven

ugl

obal

(Rev

enu

sd

écla

rés

scu

lem

ent)

les

diV

éren

tes

sou

rces

de

reve

nu

s.(a

)A

van

tto

ute

déd

uct

ion

auti

tre

des

char

ges

gro

van

tle

reve

nu

glo

bal

.(C

on

trib

uti

on

sd

irec

tes

assi

mil

ées,

per

tes

résu

ltan

td

’un

déf

icit

d’e

xplo

itat

ion

,in

térê

tsd

ede

llos,

etc.

).(b

)A

uco

ne

con

cord

ance

ne

peu

tex

iste

ren

tre

lem

on

tan

td

ere

ven

us

ind

iqu

ésau

pré

sen

tta

ble

auet

lem

on

tan

td

esre

ven

us

qu

else

rvi

de

bas

eau

xim

pô

tscé

du

lair

esp

ou

rl’a

née

1921

.To

us

les

con

trib

uab

les

assu

jett

isau

xim

pô

tscé

du

lair

esn

eso

nt

pas

,en

eVet

,po

ssib

les

de

l’im

pô

tgé

nér

alet

,in

vest

emen

t,ce

rtai

ns

reve

nu

sen

tran

td

ans

laco

mp

osi

tio

nd

ure

ven

ugl

ob

also

um

ià

l’im

pô

tgé

nér

aln

esa

nt

pas

frap

pés

par

l’im

pô

tcé

du

lair

ep

arce

qu

ele

ur

mo

nta

nt

ne

dép

on

sep

asla

som

me

affr

anch

ied

el’i

mp

ôt

dan

sla

céd

ule

corr

esp

on

dan

te.

Sou

rce:

Ori

gin

ally

pu

bli

shed

inB

ulle

tin

dest

atis

tiqu

eet

dele

gisl

atio

nco

mpa

re,

Mar

ch19

23:

vol.

93.


That is, if b¼2, the average income of individuals with income above 1100,000 is1200,000, and the average income of individuals with income above 11 million is12 million. Although this law is only an asymptotic approximation (in practice,estimated b coeYcients vary slightly with y), it works remarkably well for topincomes, as was Wrst noted by Vilfredo Pareto (1896, 1896–97) in the 1890s usingtax tabulations from Swiss cantons. In this volume, we do not address theinteresting issue as to why this law holds, and we solely use is as an interpolationtechnique allowing us to compute top fractile thresholds and average incomesfrom grouped income data. It is important to note that although the b coeYcientis (almost) invariant with y for a given country and a given year, it does varysubstantially over time and across countries.6 A higher b coeYcient means a fatterupper tail of the income distribution, which generally implies higher inequality(for a constant mean). For instance, the b coeYcient declined from about 2.3–2.4to about 1.7–1.8 in France during the twentieth century, as top income sharesdropped. The b coeYcient went through a similar decline in all countries whereinequality dropped, and it started rising again in countries where inequality rosesince the 1970s, e.g. in the United States (where the b coeYcient is now back toabout 2.3–2.4).

Pareto extrapolation techniques are fairly powerful, but they do not allowextrapolation on income ranges for which we have no data. In that respect, onemajor limitation of tax data is that the income of individuals not subject to thetax is excluded from the data. Prior to the Second World War, the proportion ofindividuals subject to progressive income taxation hardly exceeded 10–15% inmost countries, so that one can only compute top decile income series (andabove) over the entire period. In order to construct top fractile income sharesseries from top fractile income data, one needs a total income denominator,which can be computed using aggregate income sources (national accounts andtheir ancestors). Constructing homogenous numerator and denominator seriesrequires special care and raises a number of issues, many of which are addressedin Chapter 2.

1 .3 BASIC DESCRIPTIVE FINDINGS: THE KUZNETS’

CURVE, 50 YEARS LATER

The Wrst economist to use these data sources and methodology in a systematicway was Kuznets (1953).7 He exploited US income tax tabulations covering the

6 Most authors refer to a¼b/(b!1) (rather than b) as the ‘Pareto coeYcient’. Note, however, thatthe b coeYcient has a more intuitive economic meaning. One could for instance refer to b!1 as the‘income advantage of the rich’ (IAR) coeYcient. During the twentieth century the IAR coeYcientdeclined from 130–140% to 70–80% in France, i.e., the income advantage of the rich nearly halved.

7 Earlier authors (e.g., Bowley 1914 and Stamp 1916) used income tax data in a sophisticated way(see Chapter 4), but Kuznets was apparently the Wrst scholar to use control totals to construct topincome shares series.


8 T. Piketty

1913–48 period and computed corresponding top decile and top percentileincome shares series. These were the Wrst long-run income distribution seriesever produced (income distribution had been at the centre of speculative eco-nomic thought at least since the time of Ricardo and Marx, but few data wereavailable). Unsurprisingly, these series had a major impact on economic thinking,especially after Kuznets (1955) proposed his famous ‘Kuznets curve’ theory inorder to account for the 1913–48 decline in income inequality that he witnessedfor the United States. According to this theory (which Kuznets himself viewed ashighly speculative),8 income inequality should follow an inverse-U shape alongthe development process, Wrst rising with industrialization and then declining, asmore and more workers join the high productivity sectors of the economy.

In a sense, all what we are doing in this project is to extend and generalize whatKuznets did in the early 1950s—except that we now have 50 more years of data,and over 20 countries instead of one. In addition, note that Kuznets had access toa fairly limited data processing technology, which probably explains why he didnot use all available data as systematically as possible. In particular, Kuznetsdid not fully use the tabulations broken down by income source, and his topincome shares series are only deWned for total income (for instance, he did notcompute separate series for wage income or capital income).

The fact that we have 50 more years of data, over 20 countries and series brokendown by income source led us to adopt a fairly diVerent perspective than Kuznets asto why income inequality dropped in Western countries during the Wrst half of thetwentieth century. First, as one can see on Figure 1.1, where we plot the basic seriesfor the French case, the decline in top income shares witnessed by Kuznets for theUS also took place in France, but it came to an end right after the Second WorldWar. The secular decline in income inequality took place during a very particularand politically chaotic period, namely during the 1914–45 period (and especiallyduring both World Wars and the early 1930s). This raises serious doubts about agradual, Kuznets type explanation. If the decline in income inequality was due toa continuous reallocation process between from a low productivity to a highproductivity sector (say, from rural to urban sector, as in Kuznets’ originalmodel), then it is hard to understand why the timing of the fall should be soparticular.

Next, and most importantly, one can see from Figure 1.1 that the 1914–45 dropin top income shares is entirely due to the fall of top capital incomes: top wageshares actually did not decline at all. One gets the same picture by using otherinequality measures, e.g., by looking at the top decile share rather than the toppercentile share. In particular, the striking fact that the wage distribution in acountry like France has been extremely stable in the long run during the twentiethcentury appears to be very robust, irrespective of how one measures wageinequality (for instance, the 90–10 ratio—and not only top wage shares—hasalso remained stable in the long run); see Piketty (2003) and Chapter 3. Labour

8 ‘This is perhaps 5% empirical information and 95% speculation, some of it possibly tainted bywishful thinking’ (Kuznets 1955: 26).



reallocation of the kind described by Kuznets did take place (the bottom 30% ofthe French wage distribution was made up almost exclusively of rural workers atthe beginning of the twentieth century, and rural workers have virtually disap-peared by the end of the twentieth century), but this did not lead to a compres-sion of the wage distribution: low wage rural workers have been replaced by lowwage urban workers, and the wage hierarchy remained more or less the same (inspite of the fact that real wages have been multiplied by Wve over the course of thecentury).

The fact that the drop in income inequality is solely due to the fall in top capitalincomes, and that the fall took place mostly during wartime and the GreatDepression, suggests an obvious explanation: for the most part, income inequal-ity dropped because capital owners incurred severe shocks to their capital hold-ings during the 1914–45 period (destruction, inXation, bankruptcies, etc.) Thisinterpretation is conWrmed by available wealth and estate data. Note that theidea that capital owners incurred large shocks during the 1914–45 period andthat this had a big impact on income distribution is certainly not new (Kuznetsalready mentioned this factor). What is new is that there is not much elsegoing on.

The more challenging part that needs to be explained is the non-recovery oftop capital incomes during the post-1945 period (see Figure 1.1). Here theproposed explanation is that the 1914–45 capital shocks had a permanent impactbecause the introduction of high income and estate tax progressivity (there wasvirtually no tax progressivity prior to 1914, and top rates increased enormouslybetween 1914 and 1945) made it impossible for top capital holders to fullyrecover. Simple simulations (Piketty (2003) and Chapter 3) suggest that the

4%

6%

8%

10%

12%

14%

16%

18%

20%

22%

1913

1918

1923

1928

1933

1938

1943

1948

1953

1958

1963

1968

1973

1978

1983

1988

1993

1998

Top 1% income share(income distribution)Top 1% wage share (wage distribution)

Figure 1.1 The fall of top capital incomes in France, 1913–98

Source : Piketty 2001, 2003, Chapter 3 this volume: Table 3A.1; authors’ computations using income tax returns.


10 T. Piketty

long-run impact of tax progressivity on wealth concentration is indeed largeenough to explain the magnitude of the observed changes.9

The French case depicted on Figure 1.1 is interesting, because it appears tobe fairly representative of what happened in other OECD countries.10 In allcountries for which we have data, the secular decline in income inequality tookplace for the most part during the 1914–45 period, and most of the decline seems tobe due to the fall of top capital incomes. The 1914–45 drop was larger in countriesthat were strongly hit by the war (e.g., France and Germany) than in the US, andthere was no drop at all in countries not hit at all (such as Switzerland), which isconsistent with the proposed explanation based on capital shocks. Moreover wealthconcentration seems to have better recovered during the post-war period incountries with less tax progressivity (especially estate tax progressivity) such asGermany, which again seems broadly consistent with the tax explanation.

There are however important diVerences between rich countries. First, incomeinequality did keep declining during the 1950s–60s in a number of countries(such as the UK), albeit at a lower pace than during the 1914–45 period.11 Next,during the post-1970 period, one does observe a major divergence between richcountries. While top income shares have remained fairly stable in France andother continental European countries over the past three decades, they haveincreased enormously in the US, where they are now back to their interwar levels(see Figure 1.2). The UK and other Anglo-Saxon countries tend be somewhere inbetween the European pattern and the US pattern. Note that the rise of US topincome shares is not due to the revival of top capital incomes, but rather to thevery large increases in top wages (especially top executive compensation). As aconsequence, top executives (the ‘working rich’) have replaced top capital owners(the ‘rentiers’) at the top of the US income hierarchy over the course of thetwentieth century. This contrasts with the European pattern, where top capitalincomes are still predominant at the top of the distribution (albeit at lower levelsthan at the beginning of the twentieth century).12 This provides yet anotherexample as to why it is vital to be able to break down income distribution seriesby income source (without such a decomposition, it is virtually impossible tounderstand the forces at play). Note however the new US pattern might notpersist for very long: capital accumulation by the ‘working rich’ is likely to leadthe revival of top capital incomes at the following generation, especially in acontext of large cuts in US income and estate tax progressivity.

Although most countries covered in this volume do follow this general pattern(abrupt decline of top capital incomes during the 1914–45, sudden rise of topwages in Anglo-Saxon countries since the 1970s), a careful reading of the countrychapters collected in this volume will reveal many interesting particularities.

9 See Piketty (2003) and Chapter 3 in this volume.10 See the country chapters collected in this volume.11 This might be partly due to the steeply progressive tax structure applied in those countries

(especially in the UK), but there are other explanations as well.12 See especially the striking contrast between the evolution of income composition patterns by

fractile in the US (Saez 2005: Wg. 4) and Germany (Dell 2005: Wg. 5).



We already mentioned the very special case of Switzerland, where top shares havebeen basically Xat in the long run. Countries like Ireland, Australia, and NewZealand, which were less strongly aVected by the wars than other countries, alsowitnessed a limited inequality decline during the 1914–45 period (albeit lesslimited than in Switzerland, for reasons that probably have to do with diVerencesin trade structures with countries at war). Top income shares in Canada haveincreased dramatically since the 1970s, thereby conWrming the existence of adistinct Anglo-Saxon pattern, as opposed to continental Europe (e.g., France,Germany, and the Netherlands), where top shares hardly changed during the past30 years. The case of Germany reveals another interesting pattern: although topGerman capital incomes were strongly hit by the Second World War, they seem tohave recovered fairly quickly and to be structurally higher than in other Westerncountries, for reasons that might be related to the limited tax progressivity of theGerman Wscal system (more on this below).

1 .4 . NEW FRONTIERS (I) : RETURN TO CROSS-COUNTRY

STRUCTURAL ANALYSIS

So far, most of the eVort in our collective project has been devoted to construct-ing homogenous series and producing consistent analytical narratives as to why

0%

2%

4%

6%

8%

10%

12%

14%

16%

18%

20%

22%

1913

1918

1923

1928

1933

1938

1943

1948

1953

1958

1963

1968

1973

1978

1983

1988

1993

1998

FranceUKUS

Figure 1.2 The top 1% income share in France, the UK, and the US, 1913–2000

Source : France: Piketty Chapter 3 (this volume): Table 3A.1; UK: Atkinson Chapter 4 (this volume): Table 4.1; US:Piketty and Saez Chapter 5 (this volume): Table 5A.1; authors’ computations using income tax returns.


12 T. Piketty

income distribution evolved the way it did in the various countries. Although webelieve one can learn a lot from carefully done case studies, the overall objectiveof the project is to provide a suYciently rich database (with cross-country,temporal, and income source variations) so that one can conduct some rigorouscross-country testing of the various theoretical mechanisms at play. Althoughcross-country analysis will always suVer from severe identiWcation problems, ourhope is that richer data will allow a renewal of the analysis of the interplaybetween inequality and growth.

The Wrst relationship that one might want to test in a systematic way is theimpact of tax progressivity and other factors (such as fertility). Using standardstochastic models of capital accumulation, one can show that long run capitalincome or wealth concentration depends negatively on top income and estate taxrates and fertility:

b ¼ G(t , n, . . . )Where b ¼ E(wjw > w0)=w0 ¼ IAR (Pareto) coeYcient, (1:3)t ¼ top tax rate (Gt < 0),n ¼ fertility (Gn < 0)A high coeYcient b means a fat upper tail of the distribution, i.e., high wealthconcentration. Note that according to theoretical models, tax progressivity andfertility should have an impact on the concentration of wealth and capitalincome, but not on the concentration of labour income. One can then calibratethese theoretical formulae to see whether observed diVerences in tax progres-sivity and fertility across countries can account for observed diVerences inwealth concentration. By going through such a calibration exercise, Dell(2005) concludes that relatively small diVerences in top estate tax rates canhave a large impact on long run wealth concentration. In particular, thediVerence in top estate tax rates between France and Germany appears to belarge enough to account for the much higher concentration of wealth observedin Germany.

The other relationship that one might want to test using our data base is theimpact of inequality on growth. Several theories (e.g., the theory of creditconstraints) predict that inequality might have a negative impact on growth.However the testing of these theories has been plagued by serious data problems.One could think of using our data base to run standard cross-country regressionsexplaining the growth rate of country I at time t as a function of the inequality incountry I at time t. If one tries to run such regressions using our long-run database (say for France), then one would Wnd a statistically signiWcant, negativegrowth impact of inequality. The reason is simply that the pre-1914 period (andto a large extent the interwar period) is associated to high inequality andrelatively low growth, whereas the post-1945 period is associated to low inequal-ity and high growth. Although we believe that such regressions are moreinformative than standard cross-country regressions on inequality and growth(our regressions rely on high quality data and Wrst order changes in inequality), itis fairly obvious that this very crude methodology raises serious identiWcation



problems. There are lots of reasons why post-1945 growth was higher than pre-1914 growth (including a simple catching-up eVect following the 1914–45shocks), and there is no way one can properly identify a causal impact of wealthconcentration per se with such a crude regression. Using all countries in the database might allow production of more convincing results.13 In the meantime, onecan safely conclude that the enormous decline in wealth concentration that tookplace between 1914 and 1945 did not prevent high growth from happening.

1 .4 NEW FRONTIERS (II) : EXTENDING THE

INEQUALITY DATABASE

Although the international long-run inequality data base presented in thiscollective volume covers a large number of years and countries, it is far frombeing complete. First, historical income tax tabulations do exist for many morecountries than the ten countries covered in the present volume, and the com-panion volume will include additional countries in Scandinavia and NorthernEurope, Southern Europe, Latin America, and Asia. More countries are yet to beexplored, both in the OECD and in the developing world. Note that our long-rundata base is bound to be devoted for the most part to OECD countries. Onereason is simply that a number of LDCs introduced a modern income tax onlyrecently, so it is often impossible to construct long-run income distribution seriesfor these countries. There are, however, some exceptions. For instance, a progres-sive income tax was introduced in 1922 in India, which allows the computation ofthe 1922–2000 top income share series for India (Banerjee and Piketty 2005). Inaddition to the countries covered in the companion volume, there probably exista number of other non-OECD countries (especially ex-colonies) where tax dataspanning reasonably long time periods are available. Note that even in LDCswhere the income tax was introduced only recently, income tax returns datashould probably be used more often as a useful supplement to standard incomesurveys.14

Next, the series constructed for the ten countries covered in the present volumeare incomplete, in the sense that an exhaustive use of all published tax tabulationsin these countries would allow the construction of a number of additional series.For all countries, we oVer annual homogenous series on top income shares

13 For a Wrst attempt to use the data base to conduct panel cross-country regressions, see Atkinsonand Leigh (2004) and Leigh (2006).

14 For instance, it is only in 1980 that a modern progressive income tax was introduced in China(following the 1979 reforms), so that it is impossible to construct long-run Chinese inequality series.However, Chinese tax data available for the 1980s–90s oVers a useful supplement to standard surveys,e.g., in order to compare inequality dynamics in China and India during the recent period (see Pikettyand Qian 2004). In particular, one problem with standard surveys is that they severely under-estimatetop incomes (this is true everywhere, but especially so in LDCs), and tax data allows us to addresspuzzling facts such as the Indian ‘growth paradox’ of the 1990s (see Banerjee and Piketty 2005).


14 T. Piketty

covering most of the twentieth century. However, available tax tabulations alsoallow us to calculate eVective income tax rates series for each top income fractile.This is a fairly tedious work (this requires collecting exhaustive information ontax law and taking into account all variations in family structure, childrenallowances, etc.), and such series have been constructed for only a handful ofcountries.15 Available income composition data was used for most countriescovered in this volume, albeit not always on an annual basis.16 In countrieswith a progressive estate tax, there also exists a whole set of historical estate taxtabulations, which could be used to compute top estate shares series (wealthdistribution among decedents), as well as top wealth shares series (wealth distri-bution among the living) using the estate multiplier.17 In the context of thisvolume, we chose to concentrate on income tax tabulations and top incomeshares series, and we did not attempt to use estate tax tabulations in a systematicway.18 Extending the data base in this direction raises technical diYculties butwould be a useful step in order to enrich cross-country structural regressions.

Finally, and most importantly, one of the most exciting avenues for extendinghistorical inequality data sets in the future probably consists of collecting micro-level tax data from individual tax returns available in national archives. As thisvolume attempts to illustrate, published tax tabulations are a useful date sourceand allow us to gain a better understanding of the long-run determinants andconsequences of income inequality. However it is obvious that one could do a lotmore if micro-level data sets were available. In most OECD countries, micro-leveltax returns data sets are available only for the post-1970 or post-1980 period, andthey usually cover a limited number of years and use a fairly low sampling rate.19The only way to construct micro-data sets for earlier periods and with adequatesampling rate is to go back to individual tax returns stored in national archives

15 Note that available data on family structure, number of children, etc., for each income bracketcould also be used to study marriage and fertility behaviour for each top income fractile and to analysethe behavioural impact of changing Wnancial incentives.16 In some countries (e.g., France and the US), separate tabulations by wage brackets were also

published and have been used to compute top wage shares series (and not only top income shares andtop income composition series).17 In countries with a comprehensive tax on the wealth of the living (this is less common than a

comprehensive estate tax), the corresponding data can also be used to compute top wealth sharesseries.18 Estate tax tabulations were used in a systematic way by Atkinson and Harrison (1978) for the UK

(earlier authors did use estate tax data to produce top wealth shares estimates, albeit for shorterperiods; oYcial top wealth shares are now published every year by the UK Inland Revenue) and byLampman (1962) (the resulting top wealth shares series have recently been extended until the presentday for the US by Kopczuk and Saez (2004)). Similar series are also available for France (see Piketty2001, 2003; Chapter 3; and Piketty et al. 2004). The chapter on Switzerland (Chapter 11) also useswealth data, although not in a systematic way.19 One exception is the US, where the Internal Revenue Service (IRS) released annual micro-level

data sets for income tax returns starting in 1960 and with large over-sampling at the top (see Pikettyand Saez 2003 and Chapter 5 in this volume). In most countries, micro-level data sets with large over-sampling at the top (or sometime exhaustive date sets) have been used by tax authorities since the1970s but are diYcult to access for researchers.



(older returns were destroyed in some countries, but properly stored in others)and scan hundreds of thousands of them. Depending on technological evolutionand Wnancial resources made available for such projects, scholars working onhistorical changes in income distribution might throw away tax tabulationsand start working on long run micro-level tax returns data set in ten years, 50years, or more.

In order to illustrate what micro-level data sets could bring to the analysis ofhistorical changes in inequality, we take the example of a recent study on wealthconcentration in Paris and France over the 1807–1994 period. In France, a modern,universal estate tax was introduced in 1791, and individual estate tax returns havebeen stored and can be accessed in the local archives of each département. When theestate tax became progressive in 1902, the tax administration started compiling andpublishing tabulations by estate brackets. No such tabulation was compiledbetween 1791 and 1902, when the estate tax was purely proportional. In order toput twentieth century top wealth shares series in perspective, Piketty et al. (2004)collected large samples of estate tax returns for all decedents with positive wealth inParis every ten years between 1807 and 1887, as well as a similar sample for 1902, inorder to ensure the consistency of the nineteenth century series with the post-1902tabulations based series. As one can see from Figure 1.3, the basic Wnding is thatwealth concentration in Paris and France kept rising right until the First WorldWar. This is important, since this conWrms that there was no pre-existing, Kuznets-type trend in inequality priori to the 1914–45 capital shocks. If anything, theupward trend in wealth concentration appears to accelerate at the end of thenineteenth century and at the beginning of the twentieth century, which again

20%

25%

30%

35%

40%

45%

50%

55%

60%

65%

70%

75%

80%

1807

1817

1827

1837

1847

1857

1867

1877

1887

1902

1913

1929

1938

1947

1956

1994

Top 1% estate share (Paris)Top 1% estate share (France overall)

Figure 1.3 Wealth concentration in Paris and France overall, 1807–1994

Source : Piketty etal. 2004; authors’ computations using estate tax returns.


16 T. Piketty

contradicts the Kuznets view of a stabilization or a reversal of the inequality trendafter the initial wave of industrialization.

Most importantly, the fact that we now have micro-samples of estate taxreturns (with detailed information on age, occupation, types of assets, etc.) alsoallows us to shed some new light regarding the impact of inequality on growth.Per se, the existence of credit constraints does not necessarily imply that highwealth concentration is bad for growth. If most of the wealth is owned by activeentrepreneurs who keep re-investing their assets in proWtable projects, highwealth concentration is not necessarily bad. However if most of the wealth isowned by retired rentiers investing their wealth in low yield assets, then highwealth concentration can entail substantial eYciency costs. Here the strikingWnding is that wealth was getting older and older in France during the nineteenthcentury and until the First World War (see Table 1.3). There is also evidence thattop wealth holders were investing a rising fraction of their wealth in low yieldassets such as public bonds. Although this is not suYcient to prove that inequalityhad a negative growth impact, this shows that the very high levels of wealthconcentration that prevailed in France at the eve of the First World War wereassociated with retired rentiers rather than with active entrepreneurs (withpotential damaging growth eVects). The data set also makes it possible to studythe evolution of the share of aristocratic fortunes, to test hypothesis about thechanging share of women in top wealth fractiles, etc.20 With suYcient resourcesone could also construct panel data sets and follow the same individuals ordynasties over time. If and when such data sets become available for a largenumber of countries, both for income and estate tax returns, the scientiWc studyof income distribution will take a new turn. But in the meantime, we very muchhope that this volume will convince the reader that a systematic use of publishedtax tabulations allows us to make progress in this direction.

20 See Piketty et al. (2004) for a detailed analysis.

Table 1.3 The age proWle of wealth at death in Paris, 1817–1994

20–29-yr-old

30–39-yr-old

40–49-yr-old

50–59-yr-old

60–69-yr-old

70–79-yr-old

80–89-yr-old

90–99-yr-old

1817 26 22 28 100 54 59 59 —1827 44 50 53 100 88 87 60 —1837 133 90 107 100 116 123 110 —1847 87 73 102 100 117 204 132 —1857 84 77 101 100 104 109 145 —1867 67 58 136 100 141 125 154 —1877 66 73 63 100 197 260 430 —1887 45 33 63 100 152 233 295 —1902 29 40 80 100 253 272 401 —1947 31 51 73 100 113 105 105 1091994 — 11 45 100 87 93 95 68

Note : Average estate left by 50–59-yr-old ¼ 100.Source : Piketty et al. 2004; authors’ computations using estate tax returns.



AUTHOR QUERY

AQ1: Running head was not supplied we have follwed previous title of this sameformat Please Check.

AQ2: Table 1.1 and 1.2: No other source to fit these two tables, is it ok? Pleaseadvice.


2

Measuring Top Incomes:Methodological Issues

A. B. Atkinson

2.1 INTRODUCTION

There has been a marked revival of interest in the study of the distribution of topincomes using income tax data. Beginning with the research by Piketty of thelong-run distribution of top incomes in France (Piketty 2001, 2003 and Chapter 3this volume), there has been a succession of studies, as evidenced by the chapterscontained in this volume. In using data from the income tax r

top incomes over the twentieth centurypiketty.pse.ens.fr/files/atkinsonpiketty2007.pdf · on...

Documents