Income inequality in preindustrial societies: What can we learn from it? Branko Milanovic Mostly based on joint papers written with Peter Lindert and Jeffrey Williamson

Summer 2012

Questions ● Is inequality caused by the Industrial Revolution? Or, has inequality been pretty much the same before and after? ● Is inequality in poor pre-industrial economies today about the same as in ancient and pre-industrial economies? ● Was inequality augmented by colonization? ● Have some parts of the world always had different levels of inequality than others?

Two contadictory images of inequality in pre-industrial societies • Everybody about equally poor: if so, measures of inequality cannot show high inequality (like in a despotic regime: everybody equally powerless but one person) • Mass poverty combined with extreme opulence at the top: the top-to-bottom ratio will be very high

Constraints on the Elite in Ancient Pre-Industrial Societies ● Fact: Ancient pre-industrial societies had average income levels usually twice, but at most 5-6 times, the subsistence level. (Rich countries today have 100 times the subsistence.) ● Fact: Low average income, combined with the requirement that few fall below subsistence, meant that the elite’s surplus (and thus inequality) could not be very large. ● Query: What happened when average income and the potential surplus rose? Did the poor subsistence workers get any the added surplus or did the elite grab it all?

A New Measure: the Inequality Possibility Frontier • Divide society into 2 groups: people with subsistence income and elite (fraction ε of total population) that shares the surplus equally among themselves. • There is no overlap between the two classes, and no inequality within each. • Then, the Gini simplifies to: G

1



 yj  yi ) pi p j

• Per capita income of the elite is: yh 

N  sN (1   ) 1  [  s(1   )] N 

where N=total population, μ=overall mean income, s=subsistence. • Per capita income of people is s; and respective population shares are ε and (1-ε). • Substituting all of this into Gini gives G* 

1 



  s(1   )  s   1   (  s) 

If, for simplicity, we express μ as so many (α) subsistence minimums, the Gini becomes

1   1 G*  s(  1)  (1   ) s  limG *  0



 1 

IMPORTANT: The expression gives the maximum Gini compatible with mean income of αs; ε fraction of the elite, and no inequality among either elite or people.

When ε tends to 0 (one Mobutu), G* = (α-1)/α. With α=1, G*=0; with α=2, G*=0.5; if α=100 (like in the US today), G*=0.99. Introduction of inequality among the elite does not affect the maximum Gini.

Another way of deriving the maximum Gini (suggested by Paul Segal)

1

Slope=y/m = s/m =1/α

B ~Ns/Nm=1/α

A 0 Area A = (1/2)x(1/α) Area B = (1/2)-(1/2α)

1

Gini = B/(A+B) = 1- (1/α) = (α-1)/α

Other interpretations • This is the maximum inequality which may exist at a given income level when the entire surplus income is appropriated by (at the extreme) one individual. • The size of the overall income (the pie) limits the level of measured inequality (measured by the synthetic measures like the Gini where all incomes matter). • It is a new and realistic generalization of the Gini index since it requires that the society be sustainable. The ―usual‖ Gini is a special case when s=0.

Link with top income shares literature

New Measurement of Inequality • The ratio between the actual Gini and the maximum Gini (a point on the IPF) is the inequality extraction ratio. • The inequality extraction ratio shows what percentage of maximum feasible inequality an elite is able (or wishes) to extract = ratio A/B (next slide).

The locus of maximum inequalities is the ―inequality possibility frontier‖; the ratio A/B is the inequality extraction ratio 100

Maximum feasible inequality (G*)

80

60

40

B

A 20

Average income as multiple of subsistence minimum (alpha)

Note: Vertical axis shows maximum possible Gini attainable with a given α.

5

4. 8

4. 6

4. 4

4. 2

4

3. 8

3. 6

3. 4

3. 2

3

2. 8

2. 6

2. 4

2. 2

2

1. 8

1. 6

1. 4

1. 2

1

0

IPF can be expressed in terms of other inequality measures • Formula for maximum Theil (0) or standard deviation of logs is ln α. If α=2, then maximum Theil (0)= 69.3. • Maximum Theil (1) or entropy measure is not bounded from above. • The movement of the inequality extraction ratio similar whether we use Gini or Theil but its value tends to be lower with Theil (0) than with Gini.

How are we going to study ―ancient inequalities‖ • There are no household survey data, but.. • There are social tables akin to King’s 1688 table. • We shall use mostly the social tables that have already been produced or the data that can allow us to produce such tables (in some cases from professional censuses). Plus Ottoman censuses of settlements (2 cases) • Inequality (Gini) calculated from such tables assumes that (i) all members of a group have the same income, and (ii) groups are nonoverlapping (i.e, all members of an upper group have higher incomes than all members of a lower group). This is our lower-bound Gini1.

• We relax assumptions (i) by calculating maximum feasible inequality within the income ranges of the groups (thus allowing for an estimate of within-group inequality). But we have to keep (ii) although we know that there are members of (say) nobility who may have lower income than some merchants. This is our upper-bound Gini2. • The ratio between Gini2/G* estimates inequality extraction ratio for a given country.

What countries do we include? • Wherever we could find a social (class) table with estimated mean class income and population shares. • We set time limits: for the developed world, 1810; for the rest, 1929 (with India 1938 as an exception). • Difficult decision to decide what is a country: an officially distinct territory with autonomous or foreign government (the latter is a colony). • We do not include cities (Jerez, Paris, Amsterdam for which data exist).

• This leaves us with 30 data points, ranging from Rome 14 to India 1938. • Four data points from England (1230, 1688, 1759, 1801) and three from Holland though (1561, 1732, 1808) • Number of social classes mostly in double digits except in Nueva España and China (3 classes only), Moghul India (4) and England 1290 (7). Median number of classes = 20, but Tuscany (1427) almost 10,000 households, Levant (1596) 1415 settlements. • Does number of classes matter? Sensitivity analysis suggests Not (see below). • Estimated per capita incomes in 1990 $PPP almost all from Maddison; if not, use the ratio between the estimated mean LC income and estimated subsistence (α) and price the latter at $PPP 300 (Byzantium paper) • In the sample, α ranges from 1.6 to 6.7 (based on a subsistence minimum of $PPP 300).

An example of a social table: Peru 1876 Social Group

Population (in 000)

Per capita income (soles per annum)

Population %

Female spinners

168

59

12.7

Low paid female occupations

167

97

12.8

Farmers (both sexes)

513

117

39.2

Male laborers

276

146

21.1

Poorer artisans in provinces

71

269

5.4

Other earners

84

312

6.5

Poorer artisans in Lima

6

832

0.4

Govt salaried people

10

970

0.7

―Potentates‖

14

3670

1.04

1309

180

100

Total

Source: Shane Hunt’s estimates as revised by Albert Berry (1990, Table 4, p. 47).

Data Sources, Estimated Demographic Indicators and GDI Per Capita…(Contd.) Country/territory

Source of data

Year

Number of social classes

Population (in 000)

Estimated GDI per capita

Roman Empire

Social tables

14

11

55000

633

Byzantium

Social tables

1000

8

15000

533

England

Social tables

1290

7

4300

639

Tuscany

Household survey

1427

9,780

38

978

South Serbia (w/o foreign)

Census of settlements

1455

615

80

443

Holland

Tax census dwelling rents

1561

10

983

1129

Levant(w/o foreign)

Census of settlements

1596

1,415

237

974

England and Wales

Social tables

1688

31

5700

1418

Holland

Tax census dwelling rents

1732

10

2023

2035

Moghul India

Social tables

1750

4

182000

530

Old Castille

Income census

1752

33

1980

745

England and Wales

Social tables

1759

56

6463

1759

Country/territory

Source of data

Year

Number of social classes

Population 000)

(in

Estimated GDI per capita

France

Social tables

1788

8

27970

1135

Nueva España

Social tables

1790

3

4500

755

England and Wales

Social tables

1801-3

44

9277

2006

Bihar (India)

Monthly census of expenditures

1807

10

3362

533

Netherlands

Dwelling rents

1808

20

2100

1800

Kingdom of Naples

Tax census

1811

12

5000

637

Chile

Professional census

1861

32

1702

1295

Brazil

Professional census

1872

813

10167

721

Peru

Social tables

1876

9

2469

653

China

Social tables

1880

3

377500

540

Java

Social tables

1880

32

20300

661

Maghreb

Social tables

1880

8

5002

694

Japan

Tax records

1886

21

38622

916

Kenya

Social tables

1914

13

3816

456

Java

Social tables

1924

12

34984

909

Kenya

Social tables

1927

13

3922

558

Siam

Social tables

1929

21

11607

793

British India

Social tables

1938

8

346000

617

Notes: GDI per capita is expressed in 1990 Geary-Khamis PPP dollars (equivalent to those used by Maddison 2003 and 2004).

18th, 19th and first half of the 20th century included countries

13 countries before the French revolution, 17 countries after… No social tables for the United States (but Williamson and Lindert have produced them now), Russia (but Mironov and Lindert have just done it), Africa (except Kenya and Maghreb)

… but more may be coming North and South USA 1776/1800 (Lindert and Williamson working on it) Czarist Russia (Mironov) Poland Mehmet Ali’s Egypt More Ottoman defters Madagascar Audiencia de Quito

Kingdom of Naples around 1810

Map of Levant 1596-97 (yellow areas included)

Inequality Measures Country/territory/ year

Gini1

Gini2

Maximum feasible Gini with s=300

Actual Gini as % of the maximum

Roman Empire 14

36.4

39.4

52.6

75

Byzantium 1000

41.0

41.1

43.7

94

England and Wales 1290

35.3

36.7

53.0

69

46.1

69.3

67

20.9

32.2

65

Holland 1561

56.0

73.4

76

Levant (w/o foreign) 1596

39.8

69.1

67

Tuscany 1427 South Serbia (w/o foreign) 1455

19.1

England and Wales 1688

44.9

45.0

78.8

57

Holland 1732

61.0

61.1

85.2

72

Moghul India 1750

38.5

48.9

43.4

113

Old Castille 1752

52.3

52.5

59.7

88

England and Wales 1759

45.9

45.9

82.9

55

France 1788

54.6

55.9

73.5

76

Inequality Measures Country/territory/ year

Gini1

Nueva España 1790

Gini2

Maximum feasible Gini with s=300

Extraction ratio: Actual Gini as % of the maximum

63.5

62.0

105

England/Wales 1801-3

51.2

51.5

85.0

61

Bihar (India) 1807

32.8

33.5

43.7

77

Netherlands 1808

56.3

57.0

83.3

68

Naples 1811

28.1

28.4

52.9

54

Chile 1861

63.6

63.7

76.8

83

Brazil 1872

38.7

43.3

58.3

74

Peru 1876

41.3

42.2

54.0

78

China 1880

23.9

24.5

44.4

55

Java 1880

38.9

39.7

54.6

78

Maghreb 1880

57.0

57.1

56.7

101

39.5

67.2

59

Japan 1886 Kenya 1914

33.0

33.1

34.2

97

Java 1924

31.8

32.1

66.9

48

Siam 1927

48.4

48.5

62.1

78

Kenya 1927

41.6

46.2

46.2

100

British India 1938

48.0

49.7

51.3

97

Estimated Gini Coefficients and the Inequality Possibility Frontier 90

IPF 80

70 Chile 1861

Nueva España 1790 60 India 1947

Gini index

Kenya 1927

Maghreb 1880

40

Peru 1876 Brazil 1872

France 1788 England 1801

Kenya 1914

Bihar 1807

England 1759

Florence 1427 England 1688

Japan 1886 England 1290

Rome 14 30

Netherlands. 1808

Siam 1929

Java 1880 Byzant 1000

Holland 1561

Old Castille 1752

India 1750

50

Holland 1732

Levant 1596

Java 1924

Naples 1811 China 1880 Serbia 1455

20

10

0 0

300

600

900

1200

1500

1800

2100

2400

GDI per capita (in 1990 $PPP)

Note: The IPF is constructed on the assumption that s=$PPP300. Estimated Ginis are Ginis2 unless only Gini1 is available

• At α