M E ECONOMIC RECORD, VOL 71. NO. 214. SEPTEMBER 1995.

259-270

A Time-Series Perspective on Convergence:

Australia, UK and USA since 1870*
LES OXLEY and DAVID GREASLEY
Department of Economics, DepaHment of Economic History.
University of Edinburgh and University of Edinburgh
University of Western Australia

Tests of the Convergence Hypothesis or the tendencyfor per capita

income levels to narrow over time, have generally utilized cross-
sectional data and have usually found conflicting evidence. In this
study we utilize time-series data on Australia, UK and USAfor the
period 1870-1992, and time-series tests, to consider both 'catching-
up' and 'long-run convergence'. The paper finds evidence in favour
of long-run convergence in per capita income levels for the UK and
Austrulia for the period 1892-1992 and catching-up in the case of
U K and USA, and Australia and USA, giving some s u p p o for~ the
exogenous approaches to economic growth.

I Introduction involve long periods of 'catching-up'. Further

Tests of the convergence hypothesis, or the ten- Williamson (1995) and Barn and Sala i Martin
dency for per capita income levels to narrow over (1995) consider the implications of international
time, have attracted considerable attention. see, factor and commodity flows for the pace of
for example, Barro (19911, Barro and Sala i convergence.
Martin (1992). Baumol (1986), Dowrick and In sharp contrast the recent work on endoge-
Nguyen ( 1989). Durlauf ( I 989). and Quah ( 1993). nous growth models associated with for example
The interest partly derives from the simple eco- Rebelo (1991), has different convergence impli-
nomic underpinnings of the convergence hypoth- cations. Here technological shocks may have per-
esis. In the standard Solow-type neoclassical manent effects and preclude convergence. In
growth model, convergence in per capita output endogenous models physical and educational
stems straightforwardly from a concave produc- investment is assumed, unlike the case of Solow-
tion relationship and the nature of exogenous type models. not to encounter diminishing returns,
technical progress, making long-run growth inde- suggesting leadership might persist. As such dif-
pendent of initial conditions such as the physical ferent assumptions: about the nature of technical
capital stock. Extensions to this standard approach progress, for example exogenous Solow versus
which 'augment' the Solow model with human endogenous Rebelo, the shape of the production
capital include Barro (1991). Barro and Sala i relationship, and the mobility of factors and com-
Martin (1992) and Mankiw, Romer and Weil modities, have strong implications for conver-
(1992). also imply some form of convergent gence and therefore the time-series patterns of
behaviour although it may be protracted and relative per capita output measures.
Most empirical tests of the convergence hypoth-
*We would like to thank panicipants at the 1995 Eco- esis utilize cross-sectional data to investigate the
nomic History Society Conference, at which a version correlation between initial per capita GDP levels
of this paper was presented, and two anonymous refe- and growth rates for groups of countries, see for
rees of this lourno!, for helpful comments. example, Barro (1991) and Mankiw, Romer and

259
'? 1995. The Economic Society of Australia. ISSN 0013-0249.
260 ECONOMIC RECORD SEPTEMBER

Weil (1992). This research generally finds in FIGUREI

favour of convergence for the industrialized coun- Relarive GDP per Capi11I
tries, especially in the years after World War 11. (1985 S adjusted for P.P.P.)
O’Rorke, Williamson, and Hatton, (1994) and
Williamson (1995) also report convergence
between New World primary producers and
industrial Europe in the period 1870-1914.
y1
However, DeLong (1988). and Levine and
Renelt’s (1992) survey based upon the Heston and l .0
1 ’ 8
Summers (1988) wider sample of data for 130
countries for the post-I950 period, repons less
evidence of convergence.
Where time-series tests are utilized. no evidence
favouring convergence has been repned. even .a. - I , . I .

among industrial countries; see, for example. 1170 I190 1910 I930 1910 1970 1990
Bernard and Durlauf (1994. 1995). Time-series
tests investigate differences across countries in the
long-mn behaviour of per capita GDP.where con-
vergence implies that relative shifts must be tran-
sitory. Bernard and Durlauf (1994) consider why
the time-series and cross-sectional appmaches
typically give different results and we will reit-
erate some of these reasons in Section 11. Fur-
thermore. they generally favour the time-series
approach when ‘adjudicating theoretical disputes
in the growth literature’, in the main because of
the fundamental differences in approach to defin-
ing convergence. but also due to potential mis-
specification problems in Cross-Sectional tests.
Additionally the time-series perspective offers
insights into the macroeconomic experience of
individual countries.
This paper deploys time-series unit mot-based
tests to consider the convergence in per capita
GDP between Australia. the UK. and the USA
during the period 187b1992, and pays part~culu
attention to the distinctive macroeconomic history
of Austnlia. Australia’s per capita GDP levels in
the 1870s exceeded those of the UK and the USA.
but declined relatively thereafter. However,
whereas Australian and British per capita income
0.3‘
levels were little different by the 1980s. levels in UWUSA
the USA were considerably higher, see Figure I .
Faster GDP per capita growth in Australia and
the UK since 1945 has so far only modestly
dented the USA’s lead established during the first
half of the twentieth century.
The close and complimentary economic links
between Australia and the UK, shaped by trade. tive income advantage in the period 1870- 1900 as
investment and migration during the century after factor and commodity flows shifted both natural
1870s. suggests a natural test of convergence resource population balances and the terms of
between the two countries. Williamson (lW5) trade. Exceptionally, the USA resists the tendency
notes the tendency for regions of cccent European for New World incomes to converge to European
settlement, including Australia, to lose their rela- levels, becoming the UK’s major industrial com-
 I995 TlME-SERIES TESTS OF CONVERGENCE 26 I

petitor during the 1890s. and maintaining global

ascendancy for much of the twentieth century; see
Nelson and Wright (1992). The convergence where Y, denotes output, A, the level of produc-
hypothesis suggests Australia and the UK should tivity, Kt and H, physical and human capital, L,
have experienced per capita GDP convergence labour and E , a productivity shock. F is crucially
with the USA during the twentieth century. Inves- assumed to exhibit non-increasing returns to scale
tigating pairwise GDP per capita, convergence and standard Inada-type conditions and K , H and
between Australia, the UK. and the USA should L the following dynamics:
shed light on British-New World economic rela-
tions before 1900, and on the convergence
hypothesis more generally, in relation to the
USA's twentieth-century global leadership.
The time-series approach gives rise to two def- Lt = ( 1 + n)'h J
initions of the convergence hypothesis, and these
are considered in the next section. There we where sK and sH denotes savings rates, 6, and
extend the work of Bernard and Durlauf ( 1994) 6, depreciation rates and n the population growth
by giving attention to the possibility of structural rate and 0 represents an initial value. Using such
discontinuities in the convergence process. The a structure, Bernard and Durlauf (1994). consider
subsequent section introduces the historiography, the limiting behaviour of the level of output, nor-
particularly of Australia's macroeconomic record, malized by the levels of productivity and labour,
to motivate the investigation of discontinuities in i.e., A;' L;' Y,and demonstrate how via the con-
convergence experience. The possibility of stmc- cavity of F such a limit will be independent of
tural breaks in the growth record of the individual KO, Ho and h.Countries with the same savings.
countries are also Considered via endogenous tests depreciation, and population growth rates; the
based upon the Zivot and Andrews (1992) same E'S and Fs, will exhibit (steady-state) con-
approach. Finally we report the pairwise time- vergence. Furthermore, given the standard neo-
series test results and consider their implications classical assumptions this steady-state will be
for the convergence hypothesis. unique.1
' A referee has suggested an alternative way of
11 Time-Series Analysis and Convergence gaining insights into the distinction between catching-
The economic underpinnings of the conver- up and convergence within an exogenous growth model
gence hypothesis arise naturally within the stan- by considering the following equation taken from Barro
dard or augmented Solow neoclassical gruwth and SJa-i-Mmin (1992). which describes a linear
model. Here differences in initial endowments are approximation to the out of (stationary), steady-state
behaviour of (log) per capita output within an exoge-
seen to have no long-term effects on growth with nous growth model:
deficient countries able to catch up to the leaders
who suffer from diminishing returns. In contrast. y, = y@)e+ + y' (1 - e-43 (3)
Rebelo-type models imply leadership can be where y, is defined as (log) output per capita, ~(0) (log)
maintained with non-convergence the likely initial output per capita. yo is the stationary steady-state
outcome. As such, not only are tests of conver- (log) output per capita and thc speed of adjustment
gence interesting in their own right, but they which depends on various model parameten. If we con-
emerge as one natural testable implication of alter- sider two countries i and j . and assume a common p,
native models of growth. we can re-write (3) as (3'):
The main feature of the time-series approach to
tests of convergence within a neoclassical. Solow
(1956)-type model, stems from the basic conclu-
sion of that model. In particular. the model pro- Funhermore, if we assume sufficient time elapses such
that catching-up has
been completed then. e-r' I 0 and
duces the property that the long-run behaviour of ( 1') reduces to:
the economy will be independent of initial con-
ditions. To see this clearly, consider the following CY;, - Y, = Cvf -Y,? (3")
version derived from Bernard and Durlauf (1994). where the difference in log per capita output will be
Assume the following standard neoclassical pro- stationary without trend, i.e.. countries i and j will have
duction function: converged. Notice the absence of any association
262 ECONOMIC RECORD SEFTEMBER

When this result is translated into considering capita output differences between countries con-
output diferences, between countries, the time- tains a unit root. since output shocks generating
series tests require that these differences must be relative GDP movement infinitely persist causing
stationary with no statistical association with economic divergence-an implication of the
initial values. This is in sharp conhast to the basic endogenous growth models of Rebelo. Utilizing
cross-sectional tests of convergence which arc slightly different definitions to Bernard and
based upon a negative association between output Durlauf (1994). this can be illustrated via the con-
differences and initial output levels. This particu- cepts of catching-up and long-run convergence.
lar feature represents the crucial potential incom-
patibility between cross-sectional and time-series Defdtioa:
tests. Caching-up: consider two countries i and j,
Contrast such conclusions with Rebelo (199 I)- and denote their per capita real output as yi and
type endogenous models where the production yi. Catching-up implies the absence of a unit root
function exhibits non-constant (increasing) in their difference yj'- yi. p i s concept of con-
returns. In this case output differences between vergence relates to economies our of long-run
economies can become unbounded. Furthermore, equilibrium over a fired interval of time, but
notice the possibility thru convergence may assumes that they are sufficiently similar to make
involve a multiplicity of steady states. a feature tests (and rejections). of the hypothesis non-trivial.
utilized by Durlauf and Johnson (19941, to dem- In this case catching-up relates to the tendency for
onstrate the problems faced by traditional cross- the difference in per capita output to narrow over
section tests of exogenous versus endogenous time. Hence non-srationarity in yj - y.. I
il stochas-
growth models--an issue returned to later below. tic trend, must violate the proposition although the
Convergence, however, is but one implication occurrence of a non-zero time trend, a determin-
of such models and does not in itself represent a istic trend in the process. in itself. would not. A
full test of the competing approaches. In order to sufficient condition for catching-up would be sto-
test for convergence some form of clear definition chastic cointegration between yi and y j .
and some appropriate form of time-series data are Long-run convergence: consider two countries
required where, as we will see. the crucial feature i and j . and denote their per capita real output as
to be exploited are the time-series properties of yi and y j . Long-run convergence implies both the
the data. absence of a unit root i n their difference yi - yj
Following Nelson and Plosser's (1982) pio- and a time trend in the deterministic process, i.e.,
neering research considerable effort has been the absence of both a stochastic and deterministic
devoted to establishing whether or not individual trend. In this case a sufficient condition for con-
output series typically contain a unit root and are vergence would imply both stochastic and deter-
difference stationary. The issue lies at the centre ministic cointegration between yi and y,.
of modern macroeconomics since the existence of Catching-up differs from long-run convergence
a unit mot implies output fluctuations infinitely in that the latter relates to some particular period
persist to shape mean output and employment. T equated with long-run steady-state equilibrium.
Durlauf ( 1989), and Bernard and Durlauf (1995). In this case the existence of a time trend in the
utilize the Dickey-Fuller unit mot testing pmce- non-stationary yi - y j would imply a narrowing
durc as a time-series based test of convergence. of the (per capita output) gap or simply that the
Here convergence implies output innovations in countries though catching-up had not yef con-
one economy should be transmitted internation- verged. Conversely, the absence of a time trend
ally. The absence of transmission implies that per in the stationary series implies that catching-up
has been completed.
Clearly long-run convergence and catching-up
between initial values and the definition of convergence. are related in that both imply stationary yi - yj.
Alternatively. write (3') as: However, long-run convergence relates only to
Q, - y,,) =u + be-"' (3'") (similar) economies in long-run equilibrium and
where b = (v(O), - )to),) - (vl* - y,?. and a = yl* - therefore represents a much stronger version of
' ) 1 Here (3"') captures the notion of catching-up the convergence hypothesis.
h'v etined above where the diffemnce in log per capita As defined above, statistical tests of catching-
output will be stationary. but with significant trend up and long-run convergence hinge, therefore, on
effects. the time-series properties of yi - yj. The natural
 199s TIME-SERIES TESTS OF CONVERGENCE 263

route for such tests involves Dickey-Fuller type D P = t - TB. and DT = t if t > TB and 0
tests based on the bivariate difference in per capita otherwise,
output between pairs of countries, i and j , i.e.. and TB refers to the time of the break.
Models A, B, and C use the specified dummy
variables to respectively allow the incorporation
n
of a crash, trend change and joint crash and trend
changes, when testing the unit root null. a = I.
Under the rejection alternative a < 1. and signif-
where y indicates the logarithm of per capita icance is tested against the Perron ( 1989). critical
output. If the difference between the output series values, which have larger absolute size than the
contains a unit root. a = 1. output per capita in Dickey-Fuller T.
the two economies will diverge. The absence of a Applying Perron's unit mot testing strategy
unit root. a < I. indicates either catching-up, if requires the prior specification of breakpoint
f3 f 0. or long-run convergence if p = 0. years. Christian0 (1992), and Zivot and Andrews
The main reservation surrounding the robust- (1992). favour an alternative approach based upon
ness of unit root tests in general, and therefore recursive searching for endogenous discontinuities
their application to tests of convergence in at every year within the sample. They deploy
particular, concerns the possibility that structural dummies similar to Perron's to incorporate cnsh
discontinuities in the series may lead to erroneous and trend breaks, but dispense with the single year
acceptance of the unit root hypothesis. Perron dummy, D(TB). Critical values reported by Zivot
( 1989). and Rappoport and Reichlin ( 1989). con- and Andrews ( 1992). for testing the significance
sider the importance of incorporating the effects of a with breaks at any year are much greater in
of discontinuities when investigating the statistical absolute value than Perron ( I 989), raising a strin-
propenies of long-run historical series. For gent b h e r to the rejection of a unit root.
example. Perron ( 1989) has challenged Nelson Before considering the results, it is useful to
and Plosser's (1982) findings that US production consider the relationship between previous cross-
series typically contain unit roots since allowing sectional tests and these new time-series
for the effects of the great contraction around approaches. Within the cross-sectional test frame-
1929 appears to overturn the unit root null. Alter- work. convergence is identified if (bivariate) dif-
natively, Zivot and Andrews (1992). contend that ferences in growth rates are negatively correlated
discontinuities have been too readily accepted and with initial income levels-'the cross-section
that macroeconomic time series usually contain a tests consequently examine whether the average
unit root. These analysts consider extensions to change in per capita output of the initially poorer
Dickey-Fuller type regression tests which embed country exceeds that of an initially richer
the unit root null. a = I within the models: country,' Bernard and Durlauf (1994). In conViLFt
the time-series tests of convergence require that
yI =p + 8Du + Pt + dD(TB) these differences in growth rates are stationary
n
with no statistical association with initial wealth.
As such, the time-series version of catching-up,
i- I which permits for time trend effects, captures a
version of the cross-sectional tests of conver-
gence. The time-series perspective is therefore a
stronger notion of convergence which subsumes
the cross-sectional tests as tests only of catching-
up and is one mason for Bernard and Durlauf s
preference for the approach. A second stems from
=p + BDu + pt + yDT + dD(TB) the robustness of the time-series approach when
n
discriminating between for example. endogenous
versus exogenous models of growth. In particular
+ + 2 6; AY,-~+ el
i- 1
(C) using an example from Durlauf and Johnson
(1994). they illustrate how a significant p (as
where y is the variable of interest; defined in fn. I), can emerge when assuming an
D(TB) = I i f f = TB, 0 otherwise; exogenous growth process. but the true process
Du = I if t > TB, 0 otherwise; stems from an endogenous growth model with
264 ECONOMIC RECORD SEPTEMBER

multiple equilibria. No such ambiguity would per capita during the half century after 1890 in
pertain to the time-series notion of convergence their survey of long-run Australian growth, while
because of the required absence of association. Jackson (1977) reckons that the 1890s dividing
line may have been of fundamental importance. I t
Ill Macroeconomic History and Structural appears. see Butlin (1962) and Boehm (1971). that
Discontinuities Australia's export potential was unable to meet
The issue of structural discontinuity figures growing overseas liabilities, and that the 1891
prominently in Ausmlian macroeconomic history. crash was a mechanism for equilibrating the
For the pre- I9 I4 era we have argued that the 185I balance of payments.
gold discoveries and the 1891 depression seg- Twentieth-century Australian macroeconomic
mented Australian growth, see Greasley and history also highlights a number of likely discon-
Oxley ( 1994a). Here we focus on the period since tinuities. Most distinctively interwar Australian
1870 to allow comparison with the U K ' s and the GDP per capita peaked in 1925. Pope (1987) con-
USA's GDP per capita. Both Australian and siders the post- I925 downturn was connected to
British GDP per capita appear to follow a differ- levels of domestic investment, especially building.
ence stationary, (DS). process for the period 187G In part patterns of building investment reflected
1992. whereas the results in Table I suggest the demographic swings, but the UK's return to the
USA's data are trend stationary, (TS). gold standard in 1925 was accompanied by con-
Elsewhere we suggest that structural disconti- straints on capital flows to Austrilia. see Pressnell
nuities associated with World War I, 1914-18. (1978) and Tsokhas (1994). The Australian
and the great contraction. 1929-33. disguise the economy was already on a downward path when
true TS model for the U K , see Greasley and Oxley the USA's crash around 1929 heralded a collapse
(1994b). It seems possible that Australian GDP of world trade. Although Australian GDP per
per capita also follows a segmented trend station- capita rose after 1931. 1925 levels were not
ary process during the period 1870-1997,. see Raj regained until 1940.
( 1992). Nineteenth-century expansion of the Aus-
Other landmarks in Australia's twentieth-
tnlian economy was interrupted by the 1890s century macroeconomic history which may have
depression. The timing and causes of the down- k e n associated with structuril discontinuity
turn have k e n widely debated. but typically ana- accord more closely with the major international
lysts consider the 1890s as a watershed between shocks. The disruption to world trade during
nineteenth-century expansion and a period of World War I was particularly detrimental to
slower growth lasting until 1939. Maddock and Australia. whereas Australian GDP per capita
McLean ( 197 I ) highlight the slight rise in GDP surged over 30 per cent between 1939 and 1944
as increased demand lifted the economy from
TABLEI depression. Maddock ( 1987) emphasizes the long
Testing for Unit ROOISin GDP per Cupita 1870-IWJ boom in the Australian economy to 1970 dated
( without discontinuities? from the onset of World War If. The oil price
shocks around 1973 marked the end of the boom.
and inevitably slowdown in the world economy
~ ~~ ~ ~ ~~~~

adversely affected Australia. However Pagan

Australia - I .577 0.864 1.419 ( 1987) argues the extent of retardation was exac-
UK + -0.85 I 0.564 1.198 erbated by a loss of competitiveness due to real
us -4.363. 0.037 5.883
exchange n t e appreciation both during the 1970s
ADF refers to the Dickey-Fuller statistic with 2 aug- and the early 1980s mining boom.
mentations except + which refers to ADF(6). Within the period 1870-1992 the years most
*denotes significant at the 5% level using the Mac- likely to have witnessed discontinuity in Austral-
Kinnon (1991) critical values. Lh4(SC) relates to a ia's GDP per capita are, 1891, 1914-.18. 1925,
L a p g e Multiplier test for first-order serial correlation 1929. 193945, and 1973. Zivot and Andrews's
and Q6 is a Ljung-Box test based upon 6 lags.
T h e degree of augmentation h u been chosen so as to ( 1992) recursive search procedure offers the sim-
produce serially uncorrelated error terms, significant plest approach to assessing the significance of
lagged difference terms and pmimony. in an attempt to single postulated breaks. Figure 2 illustrates the
maximize test power. Such criteria w e n used to choose ADF(2) statistics obtained by respectively incor-
the level of augmentation in dl cases. porating crash, trend change, and joint crash and
 1995 TIME-SERIES TESTS OF CONVERGENCE 265

FIGURE2 Zivot and Andrews critical values. However a

Zivor and Andrews ( 1992) Apprcwh Perron-type approach with the a priori specifica-
tion of the 1925 discontinuity does overturn the
DS null. with an ADF(2) of -4.54. Moreover
a single endogenous trend break around 1931
appears sufficient to reject a unit root. even
though Zivot and Andrews's critical values raise
a stringent barrier to the rejection of the DS null.
Neither of the world wars or the 1973 oil price
shocks appear to have segmented Australia's GDP
per capita growth record.
It seems likely that the strength of the 1931
break partly stems from measuring accelerating
growth from the depth of the interwar trough,
which casts some doubt on the historical signifi-
cance of the trend discontinuity. Moreover Zivot
and Andrews's procedure only searches for single
year breaks, whereas discontinuities may have
occurred in combinations of years. Furthermore,
ignoring a significant 1925 effect when searching
for further breaks may introduce measurement
error and hence mislocate the true discontinuity.
The results in Table 2 investigate the possibility
of multiple breaks. focusing on the combined
effects of a I89 1 crash, I925 joint crash and trend
break, and a 1931 trend change.
A combination of a 1891 crash and 1925 join1
crash and trend change yields the largest absolute
ADF(2) statistic. Zivot and Andrew's critical
values are not appropriate for multiple break
models. but Inwood and Stengos (1991). report
5 per cent level critical values for similar two- and
three-break models respectively at -4.86 and
-5.13. Although not precisely comparable with
the Inwood and Stengos values, the results here
lend support to the segmented TS model. and
highlight the likelihood of 1891 and 1925 discon-
tinuities in Australia's GDP per capita record.
The case in favour of a 1931 trend change
appears less strong in the multiple break frame-

TABLE2
Unit Root Tests for Austruliun GDP per Cupiru

Year of discontinuities
I89 I C and I89 I C, I925C&T. I89 I C and
1925C&T 1931T 1931T
trend change dummy variables within the unit root ADF(2) -6.450 -6.427 -4.975
testing strategy. LM(SC) 0.001 0.545 0.639-
The year 1891 appears as the strongest crash in Q6 1.637 3.013 1.772
Australian macroeconomic history, and 1925 as
the major joint crash and trend change, although C denotes crash, T trend break. and C&T joint crash
neither ADF(2) statistic overturns the DS null on and trend change
266 ECONOMIC RECORD SEPIZMBER

work. Adding a I93 I break to the 189 I and 1925 TABLE3

discontinuities reduces the ADF(2) statistic. Unit Roor Tests
Clearly Australian trend GDP per capita growth UK-Ausrralian. US-Ausrralian and UK-US,
accelerated between the world wars, which GDP per Cupira
explains why the Zivot and Andrews type results (wirhout discontinuities)
pinpoint the 193 I trough as the single most prom-
inent trend break. However CI priori consideration Countries Sample ADF LM(SC) Q6
of the historical record and multiple break mod-
elling yields strong evidence of a two-break seg- UK-Australia 1870-1992 -3.250 0.792 4.029
mented trend stationary representation with UK-Auswlia # 1892-1992 -5.531* 0.m 7.969
US-Auswlia + 1870-1992 -2.301 0.047 1.823
discontinuities around 1891 and 1925. Figure 3 UK-US + 1870-1992 -3.160 0.994 5.818
illustrates the preferred two-break model incor-
porating a 1891 crash and 1925 joint crash and 'denotes significant at. the 5% level based upon
trend discontinuity. MacKinnon ( 1991). AD6denotes ADF(4) except those
marked + which relate to ADF(2). 8. p value on
IV Time-Series Test Results for Convergence included time trend, (p in eq. 4) = 0.627
This section reports painuise tests for long-tun
(steady-state) convergence, and catching-up. On the likelihood of structunl discontinuities in the
the basis of the results in Table 3 based on equa- Australian (and the British) growth record, for
tion 4. for the 1870-1992 period, neither version example that associated with the crash of 1891,
of the convergence hypothesis receives support, suggests their impact on the convergence process
since a unit root cannot be rejected i n the cross- warrants investigation. Interestingly the British
country differences in GDP per capita. However and Australian economies appear to have been

FIGURE3
+ t + dulH9I
Australian GDP per cupiru: y = c + dt1925 f du1925

1.1475

,61717 -

.m-

- .413u
1981 1932 1963 199

MICNCIL Fitted .............

 I995 TIME-SERIES TESTS OF CONVERGENCE 267

convergent during the century following the dis- and importantly the possible causes of economic
continuity associated with the 1891 Australian growth. In particular, in the UK-Australia case
crash. The absence of both a unit root and a deter- where long-run convergence is inferred, support
ministic trend in the UK-Australia data for the for the Solow growth model emerges. Catching-
period 1892-1992, favours the strong version of up in the UK-USA and USA-Australia case is
the convergence hypothesis. supportive of the augmented Solow model. while
Whether or not the failure of the time-series no case supports one important implication of the
approach to typically identify convergence stems Rebelo model, i.e.. long-tern non-convergence.
more widely from discontinuities in the process However, as stated earlier such implications do
can be assessed by applying Zivot and Andrews's not constitute full tests of the growth models.
search procedure to the comparative series. The
results in Table 4 report the maximum absolute
ADF statistics obtained by endogenous searching V Conc(uding Remarks
over the period 1870-1992 for crash, trend, and Doubts have grown in recent years about the
joint crash and trend changes in a naturally value of the convergence hypothesis. Partly
extended version of equation 4. the scepticism concerns the inability of cross-
All three pairwise results reject the existence of sectional studies to discern convergence other
a unit root in some variant of the model and are than between a narrow range of countries at
supportive of some form of the convergence particular times, but more fundamentally the basic
hypothesis. The UK-Australia results support utility of the cross-sectional approach has been
long-run convergence, with the 1891 crash questioned. Moreover, the recent development of
marking a discontinuity i n the process. time-series convergence tests has led to results
The UK-Australia results do contain a signifi- which hereto have been wholly unfavounble to
cant joint crash and trend change discontinuity in the convergence hypothesis, even for industrial
1891. but the absence of a significant individual economies. The novelty of this paper's statistical
trend break and the closeness of the crash and findings lies in the support found for the conver-
joint crash and trend break ADF statistics point to gence hypothesis via a time-series perspective.
the dominance of the 1891 crash. Alternatively Incorporating discontinuities led to the rejection
both the US-Australia and UK-USA results of a unit root for all the comparative Australian,
contain significant trend discontinuities, and UK, and USA GDP per capita series.
hence favour the weaker catching-up version of In themselves these results offer support for the
the convergence hypothesis. Catch-up towards the catching-up version of the convergence hypothe-
USA's GDP per capita dates from 1950 for the sis. Further the absence of a deterministic trend in
UK and the years of World War I1 for Australia. the UK-Ausualia series points to the attainment
The results therefore show how the omission of of long-run (steady state) between these two econ-
significant discontinuities can lead to incorrect omies. Such results have broader utility when
inferences being drawn regarding convergence related to current debates on exogenous versus

TABLE4
Unit Root Tests-Diferences in GDP per Capita
Zivot and Andrews Approach

Country k Year Crash Trend Crash & Trend

UK-Australia # 4 1870- I992 -5.418" -4.192 -5.440*

[I8911 [ I8991 [I8911
US-Australia 2 1870-1992 -4.085 -4.483" -4.585
[I8911 [ 1943-441 [I9411
UK-US 3 1 870- I992 -4.303 -4.828* -5.342*
(19661 [ 19501 [I9411

k is the degree of augmentation +denotes significant at the 5% level based upon Zivot and Andrtws (1992).
[ 1 denotes the year of the maximum absolute value of the ADF. 1, p value on included time trend (p in eq. A).
= 0.495;p value on time trend, (p in eq. C), = 0.364 and p value on change in trend (y in eq. C), = 0.398.
268 ECONOMIC RECORD SEPTEMBER

endogenous growth. Solow, and Solow-type constraining her ability to meet growing overseas
models augmented by human capital and inter- liabilities. Essentially the 1891 crash appears to
national factor mobility, of exogenous economic be an income adjustment for correcting a balance
growth imply convergence or catching-up. of payments disequilibrium, which served to
However, Rebelo-type models allow the possibil- equalize Australian and UK GDP per capita
ity for perpetual leadership. The results here give levels, with price deflation principally providing
some support for the exogenous over endogenous the equilibrating mechanism. Butlin's (1962) GDP
models although care should be taken not to take deflator for Australia falls 20 per cent relative
such implications too far without full tests of the Feinstein's (1972) UK GDP deflator between
appropriate economic models. 1890-94, and Australian property prices fell even
The results also shed light on the particular faster, Melbourne house rents collapsed by 30 per
macroefonomic history of the three economies. cent between 1890-92. The timing of the adjust-
Most interestingly the UK and Australian econo- ment may have partly. been influenced by extra-
mies appear highly integrated throughout the neous events, the Bariing crisis of 1890 generally
1870-1992 period. with the effects of d e t e d British overseas investment and the UK's
macroeconomic shocks readily transmitting domestic investment cycle peaks in the 1890s. but
between the two economies, although the 1891 the underlying cause of GDP per capita equali-
crash segments comparative Australia-UK eco- zation appears to be the erosion of Australia's
nomic performance. In effect the UK and advantage derived from abundant natural
Australia attain pmllel economic performance in resources as international migration and commod-
the 1870-1891 period 31 a higher level of Austn- ity flows promoted convergence.
lian GDP per capita. reflecting the abundance of The USA was faced with a similar adjustment
Australian natural resources. Levels of Australian problem in the 1890s. but in contnst to Aus-
GDP per capita substantially in excess of the tralia's experience. the USA's GDP per capita
UK's (and the USA's) apparent in the 1870s were surged ahead to global leadership during the
unsustainable. Williamson (1995) articulates the early decades of the twentieth century. The
forces, principally migration and an adverse terms capital inflows needed to fund the USA's current
of trade for New World primary producers as the account deficit were not forthcoming in the early
utilization of new land led to rising agricultural 1890s and Friedman and Schwartz (1967) argue
supply and lower prices. eventually undermining that gold outflows and price deflation were exac-
the Australian advantage. While the increasingly erbated by doubts surrounding the USA's adher-
integrated international market after 1870 steadily ence to the gold standard. Price deflation, as
eroded the Australian wage premium over the U K Vatter (1975) shows, was one element promoting
from 84 per cent in 1870 to 35 per cent in 1890 the USA's industrial competitiveness. Clearly
and 16 per cent in 1913. the statistical results here certain foundations for industrial success had
suggest a stationary GDP per capita gap between been laid earlier, substantial industrial import
the two countries of around 25 per cent between substitution took place behind tariff barriers in
1870-91, followed by long-run parity after the the 1880s and a well-developed education system
1891 crash.? contributed to the development of an American
It seems likely that the Australian GDP per system of manufactures. However Greasley and
capita advantage to 189 I was sustained by capital Oxley (1996) suggest the 1890s did mark a dis-
inflows which temporarily masked a fundamental continuity in the USA's industrial development.
shift in Australia's economic position and main- Further, Lamoreaux ( 1988) argues the 1890s
tained Australian land and property prices despite deflation hit capital intensive firms hard, stimu-
a deteriorating terms of trade. The collapse of the lating the merger wave which shaped the USA's
Melbourne land boom in I888 heralded the crash, corporate structure for much of the twentieth
and Butlin ( I 962) stresses the role of low land, century, and helping facilitate mass production
pastoral and railway investment returns during the for the mass market.
1880s in limiting Australia's export potential and The emergence of a competitive industrial
economy around the 1890s sets the USA apart
:Pope and Withers (1994). doubt the existence of from Australia and other New World economies.
Australia-UK wage convergence before 1914. though While migration and international trade tended to
they accept relative Australia GDP per capiw might fall erode the advantages of other resource abundant
in the 1890s due to h e collapse in farm incomes. economies, the USA surged to global leadership.
 I995 TIME-SERIES TESrS OF CONVERGENCE 269

It was not until after World War I1 that Australia's and Convergence'. American Economic Review 79.
and the UK's GDP per capita show a tendency I010-30.
towards catch-up to the USA's levels, highlighting Durlauf. S. N. (1989). 'Output Persistence, Economic
the baniers to technology transfer during the first Structure, and Choice of Stabilisation Policy', Brook-
ings Pupcrs in Economic Activiiy. vol. 2, 69-1 16.
half of the twentieth century. Moreover the
modest excess of trend growth in the two catch- - and Johnson. P. (1994). 'Multiple Regimes and
Cross Country Growth Behaviour', Journal of
ing-up economies over the USA's rate after 1950. Applicd Econometrics (to appear).
suggests the convergence process will be long Feinstein. C. H. (1972), National Incomc. Expenditure
drawn out. Typically, on the basis of results and Output of the United Kingdom 185s-1965.
derived from cross-sectional studies Australia, the Cambridge University Press. Cambridge.
UK, and the USA have been considered part of Friedman. M., and Schwartz. A. J. (1967). A monetary
the same convergence club. The time-series history of thc Unircd Stares 1867-1960, Princeton
results here offer a different perspective, high- University Press. Princeton.
lighting the closeness of the UK's and Australia's Grrasley, D. and Oxley, L. (1994a). 'Segmenting the
economic links, and the distinct economic expe- Contours: Australian Economic Growth 1898- 19 13'.
working paper, University of Edinburgh.
rience of the USA. More generally the results - and - (1994b). 'Comparing British and United
suggest that factor and commodity flows. as well States' Economic and Industrial Performance 1860-
as technological transfers are important for con- 1993: A Time Series Perspective'. working paper.
vergence, and that conditions for convergence University of Edinburgh.
were more favourable before 1914 and after 1945 - and - (1996). 'Explaining United States industrial
than during the period between the world wars. growth 1860- I99 I : Exogenous versus endogenous
models', Bulletin of Economic Research (to appear).
Heston, R. and Summers. A. (1988). 'A New Set of
International Comparisons of Real Product and Price
REFERENCES Levels: Estimates for 130 Countries. 1950-1985'.
Review of Income and Weulrh 34. 1-25.
Barro. R. J. (1991). 'Economic Growth in a Cross- Inwood. K. and Stcngos, T. (1991). 'Discontinuities in
Section of Countries'. Quurrerly Journal of Econom- Canadian Economic Growth 1870- 1985'. E.rpl,lora-
iL.S 106. 407-45.
rions in Economic History 28. 2 7 4 8 6 .
- and Sala i Manin, X. ( 1992). 'Convergence', Journal Jackson, R. V. (1977). Australian Economic Devrlop-
of Political Economy 100, 223-5 I .
Baumol. W. J. (1986). 'Productivity Growth, Conver- men1 in !he Ninefeenth Centu?, Canbern.
gence, and Welfare: What the Long-Run Data Show', Lamoreaux. N . R. ( 1988). The Greor Mer*qcr Movement
American Economic Review 76, 1072-85. in American Business. Cambridge University Presh.
Bernard, A., and Durlauf, S. ( 1994). 'Interpreting tests Cambridge.
of the convergence hypothesis', Journal of Econo- Levine, R. and Renelt. D. (1992). 'A Sensitivity Anal-
metrics (to appear). ysis of Cross Country Growth Regressions', Atnrri-
- and - (1995), 'Convergence in international can Economic Review 82. 9 4 2 4 3 .
output'. Journal ofApplied Econometrics 10.97-108. MacKinnon. J. (1991). 'Critical values for Cointegration
Boehrn. E. A. (1971). Prosperity and Depression in Tests', in R. F. Engle and C. W. J. Granger, L q y -
Australia. 1887-1897. Clmndon Press, Oxford. Run Economic Relationships, Oxford University
Butlin. N. G. (1962). Australian Domestic Product. PESS.
Investment and Foreign Borrowing, 1861-193U39. Maddison, A. (1991 ), Dynamic Forces in Capitalisr
Cambridge University Press, Cambridge. Development, Oxford University Press.
- (1986). 'Contours of the Australian Economy Maddock, R. (1987). 'The Long Boom'. in R. Maddock
1788-1 860', Australian Economic Histo? Review R. and I. W. McLean. The Australian Economy in rhe
26. 96-125. Long-Run. Cambridge University Press.
Christiano, L. J. (1992). 'Searching for a Break in - and McLean. 1. W. ( 197 1 ), The Australian Economy
GNP, Journal of Business and Economic Staristics in the Long-Run, Cambridge.
10, 237-50. Mankiw. N . G., Romer, D., and Weil, D. N. (1992).
DeLong. B. J. (1988). 'Productivity Growth. Conver- 'A Contribution to the Empirics of Economic
gence, and Welfare', American Economic Review 78, Growth', Quuncrly Journal of Economics 107.
I 138-54. 407-37.
Dickey, D. A. and Fuller, W. A. (1981). 'Likelihood Nelson. C. R. and Plosser, C. I. (1982). 'Trends and
Ratio Tests for Autoregressive Time Series with a Random Walks in Economic Time Series', Journal
Unit Root'. Economerrica 49, 1057-72. of Monetary Economics 10, 139-62.
Dowrick, S. and Nguyen, D. T. (1989). 'OECD Com- Nelson, R. R. and Wright, G. (1992). 'The Rise and Fall
parative Economic Growth 1950-1985: Catch-Up of American Technological Leadership: The Post-
270 ECONOMIC RECORD SEFTEMB ER

War E n in Historical Perspective'. Journal of €co- Raj, B. ( I 992). 'International Evidence on Persistence
nomic Lrerature 30. I93 I -64. in Output in thc Presence of an Episodic Change',
ORorkc. K., Williamson. J. G.. and Hatton. T. J. Journal of Applied Econometrics 7 . 28 1-93.
(1994h 'Mass Migntion. Commodity Market Inte- Rappopon, P. and Reichlin. L. (1989). 'Segmented
gration and Real Wage Convergence', in J. G. Wil- T m d s and Non Stationary Time Series', Economic
liamson. and T. J. Hatton, Migration and rhe Journal (Supplement) 99, 1361-1401.
Intermuionol Lobour Marker. Routledge. London. Rebclo. S. (1991). 'Long-Run Policy Analysis and
Pagan, A. (1987). 'The End of the Long Boom'. in Long-Run Growth'. Journal of Polirical Economy 99.
Maddock and McLean. 5W21.
Pcrron. P. (1989). 'The Great Crash. the Oil Price Tsokhas. T. (1994). 'The Australian Role in Britain's
Shock. and the Unit Root Hypothesis'. Economcrrica Return to the Gold Standard', Economic History
99, 1361-1401. Review 47. 129-46.
Pope. D. (1987). 'Population and Australian Economic V m r ( 1975). The Drive to Industrial Maturity, Green-
Development'. in Maddock and McLem. wood Press. London.
- and Withers. G. (1994). 'Wage Effects of Immign- Williamson, J. G. (1995). 'The Evolution of Global
tion in Late Nineteenth Century Australia'. in Wil- Labor Markets Since 1830: Background Evidence
liamson and Hatton. and Hypotheses', Erplorarions in Economic Histor_v
Pressnell, L. S. (1978). '1925, The Burden of Sterling', 32. 1.11-96.
Economic Histop Review 21, 67-88. Zivor. E. and Andnws. D. (1992), 'Funher Evidence on
Quah, D. ( 1993). 'Empirical Cross-Section Dynamics in thc Great C m h , the Oil Price Shock, and the Unit
Economic Growth'. European Econumic Review 37. Root Hypothesis', Journal of Business ond Economic
426-34. Statistics 10. 25 1-70.