Download globalisation and convergence

GLOBALISATION AND CONVERGENCE
DeLONG AND DOWRICK
PAPER FOR NBER PRE-CONFERENCE MEETING
TO BE HELD ON NOVEMBER 16TH, 2000
This is a very preliminary draft – some of it still in note form.
POST WAR GLOBALISATION & CONVERGENCE
1: Introduction
Over the past half century the expansion of world trade and capital flows has [MATCHED /
EXCEEDED ?] the globalisation that occurred in the fifty years up to the First World War –
although migration of people has probably not matched the mass migrations of the late nineteenth
century.
Growth theoreticians have been contrasting the convergence predictions of the neo-classical growth
models of Swan (1956) and Solow (1956) with predictions of potential non-convergence from the
newer models of endogenous technological progress of Romer (1990) and Aghion and Howitt
(1998). In most models, globalisation is predicted to speed up convergence through capital flowing
to the capital poor countries where marginal returns are high or through ideas flowing to the
technologically less advanced.
Modern econometric studies have been dominated by controversies and confusion over the
definition and measurement of convergence. Convergence is typically treated as conditional on
some array of pre-determined factors, and recent estimates of the rate range from two to thirty-five
percent per year.
All of this concentration on conditional convergence has tended to obscure the observation1 that
1 Qualification: this statement depends on the sample of countries, measure of dispersion, PPP method, and weighting of
observations. I am referring to the unweighted variance of log GDP per capita from PWT5.6a, extrapolated to 1998, for a
sample of 106 countries. A recent report by the Norwegian Institute of International Affairs (NUPI, October 2000) reports
that the population weighted Gini coefficient for 136 countries has been declining steadily since 1975. We find that the
population weighted variance declines between 1980 and 1998.
1
country income levels have been diverging rather than converging over the past fifty years.
The first puzzle to be explained is why 50 years of globalisation has been accompanied by
divergence rather than convergence.
A mechanical answer is that some of the ‘conditioning’ variables in the standard convergence
regressions are distributed in such a way as to promote divergence – eg faster population growth
and lower rates of investment in poorer countries. In autarchic models, the development poverty
trap is readily explained: a population living close to subsistence is unable to mobilise the surplus
required for substantial domestic investment; they will typically face an exchange rate that is
undervalued relative to purchasing power parity, hence high prices for imported capital goods; they
may well be caught in a prisoner’s dilemma whereby each family substitutes quantity of children
for quality of human capital investment (schooling) in attempting to maximise family welfare,
running foul of diminishing returns to labour in the aggregate.
But globalisation is typically presumed to overcome such poverty traps – especially through the
flow of capital towards capital-poor economies, through trade-induced factor price equalisation and
through knowledge spillovers. Why might globalisation have failed to produced convergence over
the past fifty years? A number of hypotheses have been suggested:

Lucas (1990) suggests that human capital complementarity may block the capital channel –
though that argument requires in turn some explanation, such as moral hazard, for failure of
international capital markets to invest in human capital.

Ventura (1997) shows that convergence is not necessarily induced by trade in a neoclassical two-sector model with a high degreee of substitution between capital and labour
[FURTHER EXPLANATION …].

Knowledge spillovers may be insufficient (due to Abramovitz’s lack of social capability) to
overcome the divergent tendencies of autarchic endogenous technology growth.
Sachs and Warner (1995) argue that globalisation does indeed induce convergence – but only to
those countries that allow relatively free movement of goods and capital.
2
2: Evidence on growth and convergence 1960-98
Descriptive statistics are given in Table 1 for real GDP per capita (RGDP) for 109 countries in
1960, 1980 and 19982, using the Penn World Tables 5.6a.3 We also report real GDP per member of
the workforce (RGDPW) and real GDP per capita as adjusted by Summers and Heston (1991) for
changes in the terms of trade (RGDPTT). This terms of trade adjustment gives a better measure of
changes in average welfare than the fixed price measure. These latter two measures are available
only up until 1992.
Dispersion is measured by the variance of the logarithm. Other measures of dispersion are often
used, particularly in welfare analysis; but the log variance is particularly useful in that it can be
directly related to the regression analysis of growth rates. Table 1 reports changes in population
weighted variances for the whole sample, a measure appropriate to analysis of inequality across
individuals. For the most part, however, we take a positivist approach with a view to hypothesis
testing and treat each country’s performance over a defined period as a single, unweighted
observational unit.
All three measures of real GDP show increasing dispersion over the whole sample. We have
divided the sample of 109 countries into three income groups, depending on whether 1960 RGDP
was above or below I$1,500 or I$5,000 (measured in constant international prices with the I$
normalised to the purchasing power of the US$ in 1985). Divergence has occurred within each
group, except for the richest 19 countries between 1960 and 1980. But the principal cause of
overall divergence has been the failure of the poorest economies to match the growth rates of the
more developed. Between 1960 and 1980, the middle income countries grew fastest, at 3.2 percent
per year, followed by the rich countries at 2.7 percent and the poorest countries at 2.1 percent. Over
the subsequent two decades growth rates slowed for all groups, averaging a meagre 0.6 percent per
year for the 55 poorest economies.
It is this falling behind of the poorest countries, in a period of increasing globalisation, that we
investigate. In particular we examine the ways in which these countries have, or have not, engaged
in the global economy and use regression analysis to identify the relative importance of these
characteristics.
2
To reduce the influence of asynchronous business cycles, the data labelled 1960 are actually five-year averages for the
period 1960-64; similarly 1978-82 averages for 1980, 1988-92 for 1990 and 1994-98 for 1998.
3
We have at the last minute identified some problems with the PWT population and real GDP series between 1970 and
1971 which will require correction and subsequent adjustment to all the empirical results presented in this draft.
3
2.1: NOTES ON COUNTRY CHARACTERISTICS SUMMARISED IN TABLE 2
POOR COUNTRIES ARE CHARACTERISED BY SLOW GROWTH, LOW INVESTMENT,
HIGH POPULATION GROWTH, FALLING RATIOS OF WORKFORCE TO POPULATION,
HIGH PRICES FOR INVESTMENT GOODS, POOR PERFORMANCE ON VARIOUS
MEASURES OF OPENESS AND TRADE. THE SLOWEST GROWERS AMONGST THIS
GROUP ARE EVEN WORSE OFF IN TERMS OF THESE CHARACTERISTICS THAN THE
OTHER COUNTRIES WHICH HAD SIMILAR LEVEL OF RGDP IN 1960.
2.2: EXPLORATORY REGRESSIONS TO UNTANGLE THE RELATIONSHIPS
BETWEEN THE VARIABLES MENTIONED ABOVE AND TO ESTIMATE PARAMETER
MAGNITUDES
-
RELATIONSHIP BETWEEN SIGMA AND BETA CONVERGENCE AND THE
CORRELATION BETWEEN ‘CONDITIONING’ VARIABLES AND INCOME LEVELS;
-
INTERPRETATION OF THE COEFFICIENT ON INITIAL INCOME LEVELS – CAN BE
RATE OF CONVERGENCE TO NEO-CLASSICAL STEADY STATE AND / OR RATE OF
TECHNOLOGY TRANSFER
-
INTERPRETATION OF ‘CONDITIONING’ VARIABLES EITHER AS DETERMINANTS
OF NEO-CLASSICAL STEADY STATE OR AS DETERMINANTS OF LONG-RUN
TECHNOLOGICAL GROWTH
INTERACTION BETWEEN INITIAL INCOME TERM AND MEASURES OF
‘GLOBALISATION’
The regressions reported in Table 3 replicate some of the analysis carried out by Sachs and Warner
(Brookings Paperes, 1995). We use the S&W distinction between open and closed economies for
the period 1960-80 to construct a dummy variable equal to 1 for countries they deemed to be open
4
for the period 1970-89. Since we are also examining growth over the period 1980-98, we extend
the S&W classification to our later period, reclassifying countries as open if S&W report that have
been open for a significant number of years since 1980.4 This enables us to check whether the
Sachs Warner results carry over to the 1990s – in particular for the 24 poor and middle-income
countries that have only recently opened their economies.
Regression 1 in Table 3 confirms the Sachs-Warner result that open economies grew substantially
faster than closed economies over the period 1960-80. Our estimate of a 2.0 percentage point
growth premium is only slightly lower than the S&W estimates for 1970-89. By any standards, it is
a huge premium – implying that 20 years of openness lifts per capita GDP by a cumulative fifty
percent. When we interact openness with initial income, regression 2 indicates that the growth
premium for openness tends to be higher for poorer countries – averaging 3.4 percentage points
compared with 1.0 points for rich countries.
Controlling for openness, these regression show no evidence of conditional convergence, indeed the
beta coefficients are positive implying that there were additional factors slowing the growth of the
poorest relative to the richest countries. Regression 3 confirms that the usual suspects were
involved – investment rates and demographic differences accounted for 1.3 percentage points of
slower growth for the poor countries. Taking account of factor accumulation, and of the differential
effects of openness, there is now some weak evidence of conditional convergence. This should be
interpreted as conditional convergence in multi-factor productivity, proceeding at a slow rate of
only 0.4 percent per year, possibly resulting from international technology transfer. Since the
regression is controlling for trade effects, any such technology spillovers are not operating through
trade.
We perform similar analysis to explain growth between 1980 and 1998 – see Regressions 4 –6 in
Panel B of Table 3. Conditional convergence is still weak. Openness appears to deliver a smaller
growth premium than that of the previous twenty years, although 1.3 percentage points is still a very
substantial addition to annual growth rates. The interactive term, introduced in regression 8,
suggests that the poorer countries benefit less from openness than do rich countries – the opposite of
For example, S&W classify Israel as closed but “open since 1985”, so we reclassify Israel for our second period. The
other countries becoming open are: Benin, Botswana, Chile, Colombia, Costa Rica, El Salvador, Gambia, Ghana,
Guatemala, Guinea, Guinea-Bissau, Guyana, Mali, Mexico, Morocco, New Zealand, Paraguay, Phillipines, Sri Lanka,
Tunisia, Turkey, Uganda and Uruguay.
4
5
the pattern observed for the earlier period.5
Slower capital deepening in the poorer countries now contributes 2.2 points of slower growth.
Technology catch-up is statistically significant and approaching one percent per year over this most
recent period.
The next three rows of Table 3 report regressions on a larger group of countries (109 compared to
96) for which we have derived an alternative measure of openness. This measure, based on the
unsurprising observation that countries with small populations tend to engage in more international
trade than do more populous nations, consists of the residuals from a very simple regression which
explains half of the observed variation in trade shares over the pooled sample:
Log (exports + imports / GDP) = 6.23 - 0.25 log (population) ; n=218, R2 = 0.498.
This measure behaves similarly to the S&W measure in that it is positively correlated with income
level and it increases over time for all three income groups. It has the advantage of being a
continuous variable, it does not require subjective classification rules and it covers more countries.
Use of this measure of openness confirms that whilst there is a significant growth premium
resulting from openness for the period 1980-98, the benefits of openness accrue mostly to the richer
nations. The coefficient on the trade-income interaction term is again positive and significant, see
regression 8, implying that the premium for the average poor country is 1.4 percentage points, but
for the average country in the rich group it is 1.9 points.
Strong correlation between the level of openness and the interactive term (r=0.992) make their joint
inclusion in the regression problematic. Accordingly we revert to the sample of 96 countries for
which the extended Sachs-Warner index is available so that we can use the SW index for one term
and the trade share variable for the other term. The strongly significant coefficients on the tradeincome interaction term in regressions 11 and 12 confirm that the benefits of openness are skewed
towards the richer countries, whether or not we control for factor accumulation.
THIS REMARKABLE TURN-AROUND IN THE DISTRIBUTION OF BENEFITS FROM
OPENNESS REQUIRES MORE DETAILED INVESTIGATION.
5
When the openness variable and the interactive variable are included together, the latter is not statistically significant.
6
3: Non-trade Dimensions of Globalisation and Convergence
?
?
?
?
?
?
?
?
?
?
REFERENCES
7
TABLE 2: AVERAGE CHARACTERISTICS OF INCOME GROUPS1
RICH
MID
POOR
SLOWEST
GROWERS
1960
7117
2433
855
800
1980
11475
4579
1385
978
1990
13416
5365
1555
878
1998
14788
6398
1808
885
1960-80
0.027
0.032
0.021
0.007
1980-98
0.015
0.013
0.006
-0.007
PROPORTION AFRICAN
0.000
0.143
0.618
0.806
PROPORTION OECD
0.895
0.171
0.000
0.000
REAL GDP PER CAPITA
(I$)
RGDP GROWTH RATE
(annual average)
PROPORTION OPEN
1960-80
0.833
0.393
0.120
0.032
(Sachs-Warner)
1980-98
0.889
0.750
0.320
0.226
(IMPORTS+EXPORTS) /
GDP
1960-80
0.621
0.702
0.517
0.479
1980-98
0.712
0.851
0.623
0.540
ADJUSTED TRADE
SHARE
1960-80
0.016
-0.105
-0.221
-0.283
1980-98
0.209
0.171
0.100
-0.017
REAL INVESTMENT /GDP 1960-80
SHARE
1980-98
0.262
0.203
0.113
0.085
0.232
0.178
0.115
0.073
RELATIVE PRICE OF
INVESTMENT GOODS
1960-80
1.022
1.358
2.379
2.899
1980-98
0.938
1.391
2.483
2.992
POPULATION GROWTH
RATE (annual average)
1960-80
0.010
0.020
0.025
0.026
1980-98
0.007
0.016
0.024
0.027
GROWTH OF
WORKFORCE /
POPULATION
1960-80
0.005
0.002
-0.003
-0.005
1980-98
0.005
0.005
0.002
0.000
1. ‘Rich’ group is composed of 19 countries with RGDP 1960-64 averaging above I$5,000; the ‘poor’ group
comprises 55 countries with RGDP 1960-64 averaging below I$1,500. The ‘slow growers’ are the 35
slowest-growing countries within the ‘poor’ group.
8
TABLE 3: EXPLORATORY REGRESSIONS RELATING TRADE AND CONVERGENCE
The dependent variable is the annual growth rate of real GDP per capita, measured at 1995 local prices,
between 1960 and 1980, or between 1980 and 1998, from World Bank World Tables. White’s
heteroscedasticity-adjusted t-statistics are reported in brackets. Coefficients that are significantly different
from zero at the 95% confidence level are shown in bold.
A: VARIABLES AVERAGED OVER 1960-80
1
2
Openness
measure
Sachs &
Warner
dummy
variable
3
No. of
obs.
96
96
96
ln
RGDP0
- 0.0014
(0.7)
0.0026
(1.0)
- 0.004
(-1.5)
Open
0.020
(4.9)
0.108
(3.3)
0.077
(2.1)
Open x
logRGDP0
Investment2
Pop
growth
W/Pop
growth
adj.
R2
0.208
- 0.011
(-2.8)
-0.008
(-1.8)
0.247
0.078
(2.9)
0.09
(0.4)
0.68
(2.1)
0.347
W/Pop
growth
adj.
R2
B: VARIABLES AVERAGED OVER 1980-98
4
5
Openness
measure
Sachs &
Warner
dummy
variable
6
7
8
11
12
96
96
96
Population
adjusted
trade share
9
10
No. of
obs.
109
109
109
Mixed
measures:
(T = pop
adjusted
trade share)
96
96
96
ln
RGDP0
Open
0.004
(2.1)
0.003
(1.4)
- 0.008
(-2.5)
0.005
(3.0)
0.005
(2.9)
- 0.004
(-1.5)
0.013
(2.8)
0.002
(0.9)
0.003
(1.3)
- 0.007
(-2.4)
0.019T
(3.9)
0.010sw
(2.3)
0.005sw
(1.1)
Open x
logRGDP0
Investment2
Pop
growth
0.174
0.002
(3.0)
0.005
(1.2)
0.020
(4.4)
0.176
0.143
(3.8)
- 0.70
(-2.8)
0.67
(1.8)
0.409
0.226
0.002
(4.8)
0.016
(3.6)
0.225
0.084
(2.3)
0.001SW
(2.4)
0.002T
(4.3)
0.002T
(3.1)
- 0.65
(-2.8)
0.46
(1.3))
0.386
0.289
0.286
0.097
(2.5)
-0.82
(-3.4)
0.50
(1.4)
0.451
1.RGDP0 is real GDP per capita at the beginning of the period (Penn World Tables 5.6a). Investment is the
average share of investment in GDP over the whole period, measured at constant international prices
(PWT). W/Pop growth is the growth of the ratio of workforce to population (PWT).
2. Regressions using investment are estimated using 2SLS with beginning of period investment and
investment price as instruments.
3. The population adjusted trade share is the residual from regressing log(trade share in GDP) on
log(population) for the full pooled sample with 218 observations.
9
TABLE 1: BREAKDOWN OF SIGMA DIVERGENCE 1960-98: COUNTRIES RANKED BY RGDP 1960
REAL GDP per capita (1985$)
1960
1980
1998
REAL GDP per worker (1985$)
1960
1980
1990
REAL GDPpc terms of
trade adjusted (1985$)
1960
1980
1990
WHOLE SAMPLE: 1960-80
2454
4170
5544
MEAN
0.025
0.010
Annual Growth Rate
0.815
1.004
1.347
VAR(log)
Change in VAR(log)
+ 0.189
+ 0.343
Change in Var – population
weighted
+ 0.097
- 0.158
RICH SAMPLE: y60>$5,000 : N=19
7117
11475
14788
MEAN
0.027
0.015
Annual Growth Rate
0.042
0.023
0.064
VAR(log)
Change
- 0.019
+ 0.041
MIDDLE SAMPLE: $1,500 < Y60 < $5,000 : N=35
2434
4579
6398
MEAN
0.032
0.013
Annual Growth Rate
0.008
0.035
0.466
VAR(log)
Change
+ 0.027
+ 0.431
POOR SAMPLE: Y60 < $1,500 : N=55
855
1385
1808
MEAN
0.021
0.006
Annual Growth Rate
VAR(log)
Change
0.187
0.329
0.622
+ 0.142
+ 0.293
4079
0.027
0.955
6629
6971
0.005
1.047
1.156
+ 0.092 + 0.109
- 0.002
17168
0.022
0.049
25588
0.010
0.016
- 0.033
6478
0.031
0.009
11393
0.001
0.048
+ 0.039
1850
2945
0.026
0.323
0.003
0.495
+ 0.172
1631
0.027
0.815
- 0.035
28182
0.032
+ 0.016
11551
0.270
+ 0.222
+ 0.084
6906
0.028
0.050
11365
0.014
0.025
- 0.025
2314
0.033
0.007
4201
0.005
0.034
+ 0.027
3027
793
0.575
0.022
0.198
+ 0.080
2628
2831
0.007
1.015
1.209
+ 0.200 + 0.194
- 0.081
13127
0.066
+ 0.041
4410
0.419
+ 0.385
1176
1219
0.004
0.347
0.451
+ 0.149
+ 0.104
Source: Penn World Tables 5.6a, RGDPCH extrapolated to 1998 using World Bank World Tables measures of constant domestic price growth rates