Embed Size (px)
Citation preview
THE SHAPE OF THE WORLD: MEASURING GLOBAL DEVELOPMENT
Angus Deaton, Princeton UniversitySir Richard Stone Memorial Lecture, Cambridge, November 2010
SIR RICHARD STONEPrologue
Two closely related themes
Measurement & Modeling Measurement
National Income Accounting work with James Meade Which led to the Nobel Prize Work with UN and others on SNA and on international comparisons Social Accounting Matrices: demographics as well as economics
Modeling Most famous is the linear expenditure system The great book on demand analysis: both m & m Input output analysis: half accounting, half modeling Filling in the SAMs with models
My own work has mostly been about modeling Recently much more on measurement Especially the international work It is difficult, challenging, great (unsung) people
Stone’s skepticism
Richard Stone (1949) “Why do we need to compare the U.S. with, say, India or
China? Everybody knows that one country is very rich and another country very poor, does it matter whether the factor is thirty or fifty or what?”
PWT6.2 says 36 times in 1952
Without an international order, little of policy consequence hinges on these numbers These numbers are not like domestic unemployment or
price index estimates I am going to argue that Stone may have been right Or at least closer than you might think as a regular user of
the Penn World Table, or the World Development Indicators
1. PURCHASING POWER PARITY EXCHANGE RATES
The shape of the world
Who is poor and who is rich? How many poor are there in the world?
How big are the differences? What is the ratio of US to Chinese income?
The global distribution of income? Over countries Over the citizens of the world
For these, and other questions, we use national incomes not in local currencies not in US $ at market exchange rates: non‐traded goods International PPP currency, usually dollars
I shall not talk today about other “shapes”, like health or life evaluation, but about PPPs and what they do
Where do PPPs come from?
The Penn World Table & the World Bank’s World Development Indicators, among others
Ultimately from the International Comparison Program (ICP) ICP collects prices on comparable goods in many countries
To construct multilateral price indexes for each country relative to a base, such as the US
For consumption, investment, GDP, etc Used to deflate nominal local currency amounts to give “real” common
unit international PPP measures PPPs are international price indexes, with all usual price index
issues Multilateral, rather than bilateral (compare A with B) price indexes Some special difficulties
ICP 1993
PWT6 uses price data collected in 1993, updated for inflation rates since then
Important missing countries, including India and China, both imputed not measured
A regional system with each region collecting prices on its own, and calculating its own PPPs with regional numeraire
Weak center with ad hoc links between regions Between regional links are Achilles heel of ICP Involve hard comparisons between countries with different
patterns of demand and relative prices Think of comparing a Bihari laborer who eats only rice with a
Congolese farmer, or Japanese factory worker UN (1997) report concluded that the ICP 1993 had lost
credibility
ICP 2005
Sought to do much better: global office housed by World Bank
146 countries (186 in 2011, next ICP) Including India and China Many African countries never previously included
Regional structure again, each region pricing its own regional list Makes sense, but some regions very diverse
A “ring” of 18 countries, at least 2 in each region Ring countries priced a special ring list of more than 1,100
commodities These prices were then used to link the regions Using regional price indexes: only 4 numbers, e.g. Africa relative
to OECD (“tectonic” indexes)
2. KEY RESULTS OF ICP 2005
Headline result
Per capita GDP of both India and China both much reduced using the new data
China in 2005 from $6,757 to $4,088 India in 2005 from $3,452 to $2,222 Note that the US is numeraire So we could just as well say that the US got richer Essentially, India and China moved further away from the US and
other rich countries Ratio of China to US went from 6.2 times to 10.2 times Their PPPs relative to the US increased, so “real” amounts fell Some indication that the Chinese wanted this outcome
Not only India and China
.51
1.5
22.
5R
atio
of n
ew to
old
PP
P fo
r 200
5
6 7 8 9 10 11Logarithm of per capita GDP in 2005 international $
Congo, DR
Burundi
Sao Tome & Principe
Cape Verde
LesothoGuinea
Ghana
CambodiaTogo
Guinea Bissau
India
Philippines China Namibia
Tonga
Yemen Congo, R Lebanon
GabonKuwait
Fiji
Nigeria
TanzaniaAngola
Bolivia
Ethiopia
Vietnam
Bangladesh
.5.5
2.5
4.5
6.5
8.6
1970 1980 1990 2000 2010year
Post 2005 ICP
Pre 2005 ICP
Gini coefficient for per capita GDP, weighted by population
.45
.5.5
5.6
1960 1970 1980 1990 2000 2010year
WDI 2008, 2005 prices
WDI 2007, 1993 prices
PWT 5.6, 1985 prices
PWT 6.2, 1993 prices
Gini coefficient for per capita GDP, weighted by population
Dollar a day poverty?
Poor world is now poorer relative to the rich world Rich world is now richer relative to the poor world Many more people than before live beneath the new
international dollar a day than lived below the old international dollar a day Because PPPs convert $1 into higher amounts in local
currencies in poor countries, and more people below This would approximately double the world poverty count From about 900 million to about 1.8 billion in 2005
However the World Bank global poverty line is defined from poor country poverty lines Average of poor country lines in international dollars
Poverty in India
In 2005, Indian poverty lines were 538.6 (urban) and 356.3 (rural) rupees per person per month Average is 403.7 per month = 13.3 rupees pp per day Old PPP for 2005 was 11.08 rupees per $ So $1.20 pp per day in 2005 international $ US inflation 1993 to 2005 was 1.35 So $0.89 pp per day in 1993 international $
ICP 2005 increased measured Indian PPP PPP is 15.60 rupees per $ Indian poverty line is $0.85 in 2005 $ Only $0.63 pp per day in 1993 international $
New value, $0.85 in 2005 $ is actually lower than old value, $0.89 in 1993 dollars In spite of US inflation from 1993 to 2005 Upward revision to PPP is larger than US inflation
Poverty from the poor world
In the Indian example, of course, there is no change in domestic Indian poverty
But the $ value of the Indian line falls sharply For global $‐a‐day poverty, the global line is an average of
poor country poverty lines expressed in international dollars
Most of the PPPs have increased, so global line has fallen in 2005 dollars By an amount similar to the fall in India
Little change, or some decrease, in global poverty Essentially ICP did not change the global poverty counts But it sharply reduced the global poverty line
World Bank poverty
The WB, who is the official scorer for the MDG, increased estimate of global poverty by about 500 million Because they increased the global poverty line by changing
the countries in the average Dropping India (low line) in favor of countries with higher lines
Using the original countries to compute the average, global poverty falls a little, but not much change
Whether we should maintain a rich world standard or a poor world standard is a matter of debate Rich world standard seems more like what people perceive
when they think of what a $ a day means Rich world citizens are the audience for such numbers
Does the level matter?
Given that downward trend is much the same There are nearly 200 million Indians living
between $1.00 and $1.25 a day Many fewer Africans
Raising the line makes global poverty relatively more Indian, and relatively less African Numbers shape consciousness of global poverty
India will now no longer meet Millennium Development Goal for poverty reduction Rate of reduction is the same, but base is higher
WHY THE CHANGES? ARE THEY CREDIBLE?
Primer on the 2005 ICP
ICP 2005 collected price data on about 1,000 goods & services in each of 146 countries
Consumption has 110 basic headings, Basic headings are identical in all countries & regions Basic headings are matched to expenditure data from
national accounts
Within basic headings, lists differ by region, and there are no expenditure data to tell us which are more important Mud crabs and squid in Asia, Nile perch, kapenta, and bonga in Africa
More primer
Two stage construction of prices, 1. up to basic headings without weights, 2. from basic headings to overall PPPs with weights Geometric means or generalizations Standard multilateral indexes from BH up to country
Two stage regional procedure: 1. PPPs for countries within a region with no
comparisons across regions, or across countries in different regions:
2. Gluing the regions together using a set of 18 strategically chosen countries (the “ring”) who price a ring list of 1,100 items
Combining the regions
Five regions: OECD‐Eurostat‐CIS, Asia/Pacific, Africa, South America, Western Asia
Across regions, 18 ring countries (Brazil, Chile, Cameroon, Egypt, Estonia, UK, Hong Kong, Jordan, Japan, Kenya, Sri Lanka, Malaysia, Oman, Philippines, Senegal, Slovenia, South Africa, Zambia) Each prices goods & services from the ring list This is where it gets tough: pricing identical goods in
Cameroon and Japan, Senegal and UK
Continent wide price indexes
The ICP accepts a political constraint that the within region PPPs should not change when the global office glues the world together For Eurostat, this is legally mandated
So ICP 2005 collapsed all the ring prices into four price indexes, with OECD region as base, one for each of the four other regions These price indexes give us price indexes for Asia/Pacific, Africa, Western Asia, and South America relative to OECD
Some concerns
Changes or errors in the five super‐PPPs from the ring move whole continents, e.g. Africa or Asia relative to the OECD “Tectonic” super price indexes Potentially important for inequality Or for India and China relative to the US Recall the increase in PPPs for Africa and Asia relative to the US
This all comes from the ring
Concerns in more detail
Why did the ICP increase inequality between nations? Perhaps because of problems with the ring But remember the weak links in 1993
Immediate focus is the more precise matching of quality Possibly gone too far Brooks Brothers shirt example would overstate price in Senegal,
just as “shirt” understated it Concern about goods that are only available in expensive
specialist shops No weights within the basic head to minimize this
A central problem for ICP: Goods need to be comparable Goods should be locally common and representative These two are not compatible in general!
Tentative answers
Some evidence of quality‐matching problem Other cereals: has Kellogg’s cornflakes and Frosted flakes as
items in BH No weighting within basic head This item consistently appears
Yet no quality matching for many services, including medical services
But air travel, cars, and telephone calls in Kenya are genuinely very expensive Prices are OK The problem is the weights! Little local consumption When we compare with UK, the weight is 50% local and 50% UK This is how superlative price indexes work
It’s the theory, stupid!
Confess that we don’t really know what we are doing In the strict version, price indexes and superlative indexes
require identical homothetic tastes If true, these weighting problems would not exist But implausible at this level of disaggregation Without homothetic identical tastes, e.g. with income effects and
taste shifters index is a COLI for a country with intermediate income and tastes
Not very helpful No good theory of quality that is operational here Except in simple cases Perhaps we just can’t make useful price comparisons between
Africa and Europe? We need to know what goods and services are for
Margins of uncertainty
Paasche Laspeyres spread may be the true margin of uncertainty
For US versus Tajikistan, Laspeyres index is 9.6 times the Paasche index
Ratio of US to Tajik GDP is 9.6 times larger in US prices than in Tajik prices
For China and India, numbers are “only” 1.66 and 1.61 30 or 50 or what?
Splitting the difference hardly “solves” the problem Which is what superlative indexes do In the absence of common tastes, superlative indexes are just
averages with attractive properties No solution when consumption patterns don’t overlap Perhaps Dick Stone was right?
UPDATING PPPS
What happens between rounds?
Typically, nothing World Bank discards previous ICP results Takes current benchmarks Growth rates from domestic national accounts
Makes sense if you think revisions are mostly methodological So new ICP is just better than previous one The leading position
In principle, this violates theory underlying PPP adjustment in the first place At least over long enough time periods
‐1.5
‐1‐.5
0.5
Log of ra
tio of P
PP to Excha
nge Ra
re: U
S=1
4 6 8 10 12Logarithm of per capita GDP at market exchange rates
Balassa‐Samuelson effect
Cross‐section v time‐series
If Balassa Samuelson holds in the cross‐section, why would it not hold in the time‐series?
Should not the price level in China rise as it gets richer? But wouldn’t that be picked up in its domestic CPI? So that standard updating of PPPs, using relative CPIs
with US, should be OK?
PPPs, even when they are done every year, as by Eurostat, are not consistent with domestic CPIs Why not? Quality effects, different bundles, different programs
-2-1
.5-1
-.50
.5
5 7 9 11Logarithm of p.c. GDP in 1993 and in 2005 international $
Price leve
l: ratio of P
PP to Excha
nge Ra
te
1993 (yellow)
2005 (blue)
1993
2005
Argentina
Armenia
Bolivia
Congo
Czech Rep
Egypt
Estonia
Fiji
Gabon
Georgia
Guinea
Hong Kong
Ireland
Kenya Latvia
Mongolia
Malaysia
Nepal
Romania
Singapore
Sierra Leone
Vietnam
Yemen-4-2
02
46
-5 0 5 10
Ann
ual %
cha
nge
in p
rice
leve
l of G
DP
Annual % rate of growth of real p.c. GDP, 1993 to 2005
Random revisions?
Consistent with the data from 1993 to 2005 So perhaps OK to use domestic real growth rates
to interpolate Madison does this, so does World Development
Indicators Penn World Table does at least some
interpolation If we were to interpolate between rounds World inequality would not be falling Poverty would not be falling if we used a fixed $1‐a‐
day global line The world would be reshaping very differently
.45
.5.5
5.6
1960 1970 1980 1990 2000 2010year
WDI 2008, 2005 prices
WDI 2007, 1993 prices
PWT 5.6, 1985 prices
PWT 6.2, 1993 prices
Gini coefficient for per capita GDP, weighted by population
Why do we need PPPs anyway?
No domestic relevance within countries Not used by World Bank for concessional aid Some use in IMF & WB voting formulas Global poverty counts and inequality measures Do these have policy relevance? Used by activists and IFIs to argue for more money for aid
We need domestic price indexes because there is a domestic government There is no international government “Cosmopolitan” philosophers argue that the WB or other IFIs
should somehow assume that role Others (Rawls, Nagel, etc.) argue that this is wrong
Why do we need to know?
A DIFFERENT APPROACH
Income data
Any procedure for income poverty needs PPPs and needs a global line Can criticize how it is done But it is genuinely difficult
Where do the income data come from? The Bank uses many hundreds of household surveys
from around the world Not designed for this purpose Different questionnaires, reference periods, timing, and degrees of statistical competence
What about using a single, coherent survey?
40
The Gallup World Poll
Respondents are asked to report their household income 155 countries since 2006: national samples Now half way through 2010 round
Converted to international $ using same PPPs Advantages Identical (core) questionnaires: comparability Timeliness Coverage
Disadvantages A single income question is not best practice OK for wage earners, who get a pay slip Not good for self‐employed people in agriculture Most people in the poor world
Many missing values, rich and poor countries
41
020
4060
8010
0
0 20 40 60 80 100World Poll average
LiberiaBurundi
Zambia
Mali
Burkina Faso
Sierra Leone
Tanzania
Nigeria
Bangladesh
India
DRCUganda
Egypt
Ghana
Ivory Coast
Ethiopia
China Kenya
Philippines
Indonesia
PakistanSouth Africa
RwandaMalawi
Mozambique
Madagascar
Brazil
Wor
ld B
ank
2005
Vietnam
Uzbekistan
42
Discussion of comparison
WB for 2005, WP for average 2006 to 2010 as many years as available
WP simplest procedure No imputations Income in international 2005 $ less than $1.25 per person per day Fraction of non‐missing values less than this
Correlation is 0.83 Correlation with log GDP per head is only ‐0.37 and ‐0.42 Lower for GDP per head, ‐0.15 and ‐0.16 Lower for national poverty count, 0.54 and 0.58
Gives support to both procedures, WB and WP I don’t believe that the WB is the gold standard Both have strengths and weaknesses
43
Monitoring over time
World Poll has a substantial advantage World Bank updates using GDP and predictions in the absence of survey data
Perhaps too early to use the World Poll First year income data is sometimes questionable 2010 not yet complete Soon. . . .
Examples . . .
44
Where the poor are, top 12(millions, ranked by WB 2005)
2006 2007 2008 2009 2010
IndiaChinaNigeriaBangladeshIndonesiaPakistanCongo, DRTanzaniaEthiopiaPhilippinesVietnamNepal
..
..
..1421..............
330194464547....1832..12..
26021140468353..14....83
325128892874303621..
2898
..
..
..3264............11
45
Poverty by world regions(Low and low middle income)
2006 2007 2008 2009 2010
East AsiaCentral AsiaLatin AmericaMid East & N AfricaSub‐Saharan AfricaSouth Asia
Total
206101452431164
876
299201452327490
1,200
3397
1835
350375
1,124
2458
2229361378
1,043
27910144634
296
981
Imputations by country using country/year factor model estimated by region
46
What about hunger?
In principle, easier to deal with No PPPs, no poverty lines, no household incomes Hunger is what people think of in global poverty Also part of MDG1
But what exactly is hunger? Not having enough to eat Under‐nutrition
Physical/medical evidence of poor nutrition Malnutrition Requires measuring people: weight and height Skinny kids and short adults
47
Intake measures
Wonderful if World Poll measured people Not going to happen, at least for a while!
Nor does FAO in its food security numbers What does FAO do? Estimates total food and calorie availability in each country “Food balance sheets” Divides by number of people to get average Uses survey data to estimate dispersion over people: some
get more and some get less This gives a distribution and an estimate of people getting
less than 2,100 calories a day (with some flexibility)
48
Difficulties. . .
Poor correlation with malnutrition But they are different things Malnutrition depends on disease, work load, and things
other than calories Latest years, which get the headlines, are based on
forecasts by USDA Distributional data are very weak, especially in Africa Balance sheets are not very accurate Animal feed hard to estimate Cereals pretty good, other foods much harder Stocks hard to measure FAO Indian data disagree with both surveys and Indian
Department of Agricultural estimates, even in trend
49
Why don’t we just ask?
World Poll did just that “Have there been times in the last 12 months when you did not have enough money to buy food that you and your family needed?”
50
Fractions hungry
2005 2007 2008 2009 2010
East AsiaChinaIndonesia
0.370.29
..0.25
0.160.22
0.170.23
..0.25
South AsiaIndiaPakistanBangladesh
0.350.350.25
0.260.260.24
0.230.280.27
0.290.340.23
..
..0.29
SS AfricaNigeriaEthiopiaSAfricaKenya
0.580.270.450.73
0.560.390.480.56
0.55..
0.560.68
0.60..
0.550.63
..
..
..0.57
Latin AmericaBrazilMexico
0.200.36
0.210.28
0.210.33
0.200.34
..
..51
Millions hungry
2006 2007 2008 2009
East AsiaEurope & Central AsiaLatin AmericaMiddle East & N AfricaSub‐Saharan AfricaSouth Asia
Total
38350
10878
409511
1,539
4045411078
364400
1,410
36550113102410367
1,407
4005511367
411439
1,485
52
Evaluation
WP shows large increase 78 million into 2009, following food price spike and financial crisis Like FAO, but a bit smaller And from a larger base
Gallup is in the lead here Some will argue for the calorie based question But the data are very bad Calories are a bad measure even if well measured Not having enough money to buy food is important in
itself
53
CONCLUSIONS
Comparing countries
We do not know how to make cost‐of‐living or income comparisons between very different countries Increases in global inequality from one ICP to the next are little
understood Possibly related to the Balassa‐Samuelson effect We need some radical rethinking of theory Superlative indexes don’t solve deep conceptual problems Within groups of “similar” countries, much better
Stone was right, both on uncertainty, and on that it doesn’t matter Perhaps these numbers are only of academic interest ICP 2005 cost $45 million, and generated 145 x 130 x 2 public use
numbers=$1,194 per number OECD different, within EEC transfers (supranational
government)
Lessons for academics
Important not to take data for granted Without understanding where they come from Teaching measurement?
Many studies using PWT are not robust Especially those that use data at annual frequency Or are concerned with volatility
Without understanding Balassa‐Samuelson better We are on weak ground in making long‐term growth comparisons
56
