Research supported by an Australian Research Council Federation Fellowship and
Stories about productivity
Australian Research Council Federation Fellow, University of Queensland
Risk & Sustainable Management Group
Schools of Economics and Political Science
University of Queensland
Australian Public Policy Program Working Paper: P06_4
This version: 4 April, 2006
Stories about productivity
Australian Research Council Federation Fellow
School of Economics and School of Political Science and International
University of Queensland
PHONE + 61 7 3346 9646
FAX +61 7 3365 7299
I thank Nancy Wallace for helpful comments and criticism.
This research was supported by an Australian Research Council Federation Fellowship.
Stories about productivity
When asked about their methodological practices, most economists claim
to practise some form of instrumentalist positivism, as advocated by Friedman
(1953) or appeal to Popper’s (1963) notions of falsification, implemented by the
hypothesis-testing procedures of classical inference. More philosophically
sophisticated members of the profession might, following Blaug (1980), refer to
Lakatos’ (1970) methodology of scientific research programs.
In practice, however, particularly in debates over economic policy,
economists rely heavily on a ‘story-telling’ approach. What matters in this
approach is not the formulation of decisive tests of statistical significance, but
the application of economic reasoning to the relevant ‘stylised facts’ to produce a
convincing narrative account. Although most prescriptive methodologists regard
such story-telling as relying on inappropriate appeals to implicit presumptions
and collective authority, this approach has been strongly defended by McCloskey
Beginning in the late 1990s, a narrative developed around the idea of a
‘new’ or ‘miracle’ economy in Australia. Although strong macroeconomic
performance, particularly during the Asian economic crisis, contributed to the
appeal of this narrative, the crucial element of the story was derived from
estimates of multifactor productivity (MFP) developed by the Australian Bureau
of Statistics (various years), first published by the in the late 1990s, and
presented by the Productivity Commission (Parham 1999, 2000) as evidence of
the success of microeconomic reform.
The ABS estimates showed a surge in productivity beginning around
1993-94, and were interpreted as showing the benefits of the micro-economic
reforms undertaken by the Hawke–Keating government. Although the
improvement in estimated MFP growth rate was not sustained beyond 1998-99,
the story of a productivity surge driven by reform continues to be told.
The statistical basis for the ‘new economy’ story has been disputed by
critics. Quiggin (2001) argued that the observed upsurge in estimated rates of
MFP growth could be explained, in large measure, by recovery from recession
and by an unsustainable increase in work intensity.1 Quiggin (2005) claimed
that the slowdown in MFP after 1998-99 supported this view. A more
fundamental criticism was offered by Hancock (2005), who fitted simple linear
and quadratic models to the data and found that the null hypothesis of a
constant rate of MFP growth could not be rejected. In response, Parham (2005a)
argued that an appropriate analysis of data, using smoothing, error-correction
and appropriate timing of cyclical breaks supported claims of a productivity
‘surge’ in the 1990s.
In this paper, it will be argued that, given its relatively short duration
and high year-to-year variability, the MFP data set does not contain enough
information to allow clear statistical discrimination between competing
hypotheses. As a result of this lack of information, combined with the human
predilection for observing patterns, a range of alternative stories, each of which
may be supported by an appropriate interpretation of the data, has been
Three such stories are described here. The first is the ‘New Economy’
story put forward by Parham and others. The second story agrees with the first
regarding the 1990s, but interprets the subsequent decline in productivity
growth as the result of a failure to pursue microeconomic reform with sufficient
vigour. The third story rejects the idea of a productivity miracle in the 1990s and
argues instead that productivity growth rates experienced a sharp decline at the
end of the postwar ‘Golden Age’ around 1970, and that this decline has been
sustained, although with fluctuations around the trend.
1 For reasons that are not clear, Parham (2005a) summarises this paper as follows: ‘A self-
professed sceptic, Quiggin (2001) has come to accept that there was a productivity surge, though
his is at the low end of estimates’. The conclusion of Quiggin (2001) was: ‘The claim that
economic performance in the 1990s was comparable with that of the golden age of the 1960s is
inconsistent with the empirical evidence. On the basis of the available data, it is not even
possible to conclude with confidence that the underlying rate of productivity growth was higher
in the 1990s than in the 1980s.’ This conclusion does not support Parham’s interpretation.
Beginning in the late 1990s, the Australian Bureau of Statistics began
producing estimates of MFP for the market sector, going back to 1963-64. The
crucial requirement for the development of these estimates was the construction
of estimates of the capital stock, thereby permitting the derivation of estimates
of MFP in place of partial labour productivity measures.
The ABS statistics were organised using the concept of productivity
cycles, typically of about six years, which were inferred from the properties of the
annual MFP series. Although productivity cycles typically corresponded fairly
closely to expansion and contraction phases of macroeconomic cycles, no
exogenous information was used in dating cycles.
From a statistical viewpoint, the central question is whether the data set
contains enough information to make reliable claims about average levels and
trends in productivity growth and about the occurrence or absence of a structural
break in those trends in the early 1990s.
It is important to observe that the ability to derive robust inferences
from a data set typically declines each time the data is differenced. Thus, the
data contains more evidence on the level of MFP than on the rate of growth of
MFP, and more evidence on the rate of growth of MFP than on trends in the rate
of growth of MFP. Attempts to detect a structural break in the trend rate of
growth of MFP are therefore likely to be fraught with difficulty.
The data set allows a decisive rejection of the obvious null hypothesis
relating to the rate of growth of MFP, namely that the rate is zero. On the other
hand, as Hancock (2005) shows, using Ordinary Least Squares to estimate a
simple linear trend of the form
MFP(t) = a + bt,
the null hypothesis b!=!0 cannot be rejected at standard levels of significance.
Similar results are obtained using a quadratic functional form
MFP(t) = a + bt + ct2.
Parham (2005a) criticises Hancock’s analysis arguing that the failure to
find statistically significant results reflects short-term noise in the data, and
that it is better to focus on smoothed data. If the ABS analysis of productivity
cycles is accepted, it seems most natural to focus on a data set consisting of such
cycles, as is done in most of the informal discussion of the topic. This data set is
presented in Table 1.
Table 1: Multifactor productivity growth since 1964-65
a: Percentage points, annual average
b Percentage points
1. Source: Australian Bureau of Statistics
2. The long-term average rate from 1964-65 to 2004-05 is 1.2 per cent
3. Excluded observation for 2004-05, -1.7 per cent
The obvious problem for statistical analysis is that there are only eight
data points. This would be enough to detect a trend if the series declined or
increased steadily. There are seven differences in the series, so the probability
that they are all of the same sign (either positive or negative), under the null
hypothesis of stationarity, is 2-n+1 or 1.5 per cent for n=7, which is sufficiently
small to reject the null hypothesis. However, no such monotone trend is present
in the data set.
Any more complex pattern is impossible to confirm or reject with such a
limited data set, using standard classical inference testing. Parham (2005a)
claims that visual inspection of the smoothed data set supports the hypothesis of
1964-65 to 68-69
1968-69 to 1973-74
1973-74 to 1981-82
1981-82 to 1984-85
1984-85 to 1988-89
1988-89 to 1993-94
1993-94 to 1998-99
1998-99 to 2003-04
in MFP b
a productivity surge in the 1990s. However, the ability of the human eye to
detect apparent patterns in random data is notorious. It is not difficult to find
interpretations consistent with prior beliefs, but it seems clear from the course of
the debate so far that a wide range of prior beliefs can be supported by visual
inspection of the data.
Neither the raw annual data, nor the cyclical arrangement preferred by
the ABS justifies rejection of the null hypothesis of random variation about a
constant mean. There are a range of intermediate options, involving either the
use of smoothed versions of the data set or the application of error-correction
models to the raw data.
Parham (2005b) reports that statistical modelling using error correction
methods yielded support for the hypothesis of a structural break in 1990-91.
However, since there are a great many possible models incorporating structural
breaks at different points in the data series, it is hard to assess the power of
statistical tests. Given nearly 40 possible choices for a break point, it would be
not be surprising to find the null hypothesis of a constant trend rejected for at
least one of these points using standard statistical tests.
It seems unlikely that such results, mostly estimated over the period up
to 1998-99, would prove robust if the same model were fitted to the out-of-sample
data that has subsequently been observed. McKenzie (2005) using data for the
period up to 2002-03 finds no evidence that the ‘spike’ in productivity during the
mid-1990s was sustained.
In the absence of any clear statistical resolution of disputes over the
correct interpretation of the productivity data, it may be useful to consider an
analysis of the debate based on a rhetorical or ‘storytelling’ approach to
descriptive methodology. In this approach, rather than considering a description
of research in terms of the formulation and testing of hypotheses, we consider
various alternative stories that can be told about the data, and the factors that
might lead to these stories being accepted or rejected in a given community.
The ‘New Economy’ story
According to the ‘New Economy’ the program of microeconomic reform
that began with the floating of the Australian dollar in 1983 has, after some
initial disruption, produced a new, more flexible and more productive Australian
economy. Thus, the pain of structural adjustment has been more than offset by
the gains from sustained high economic growth.
The ‘New Economy’ story was developed with a focus on estimates of MFP
for the 1990s, and particularly the cycle from 1993-94 to 1998-99. After declining
fairly steadily from the ‘Golden Age’ of the 1960s to the early 1990s, estimated
rates of MFP growth showed a sharp upturn for the cycle beginning in 1993–94,
matching or exceeding those of the ‘Golden Age’. The final estimates for the cycle
from 1993-94 to 1998-99 show a productivity growth rate of 2.0 per cent, the
highest of any period in the ABS data set.
Advocates of the New Economy story, most notably Parham (2005b), have
explained the relatively weak MFP growth estimated for the period since
1998-99 as the product of a variety of temporary factors, including a ‘short-term
blip’ in 2000-01, possibly associated with the introduction of the GST, drought in
2002-03 and 2003-04, the Olympic Games in 2004, and, more recently,
bottlenecks constraining growth in mineral exports.
Given underlying strong MFP growth, however, each of these temporary
shocks should have been followed by above-normal MFP growth, as productivity
returned to its long-term growth path. The idea of a ‘productivity cycle’
incorporates the notion that short-term effects like those discussed above should
wash out over the course of a cycle.
This point is developed further by Dolman et al. (2006) who conclude that
‘in an historical context, the productivity surge of the late
1990s, rather than the most recent productivity cycle,
appears to be the more unusual experience.’
Dolman et al. consider a variety of explanations of the surge in measured
productivity during the 1990s, including microeconomic reform, but find that
none of them are consistent with the slowdown observed after 1999.
The ‘New Economy’ story gains more support from macroeconomic than
from microeconomic outcomes. The current expansion, which has already lasted
fifteen years since the trough of the 1990-91 recession, is one of the longest on
record in Australia, and one of the longest for any OECD country in recent
decades. It was continued growth during the Asian economic crisis of 1997 and
1998 that led to Krugman (1998) describing Australia’s as a ‘miracle economy’.
There is, however, no clear reason to link microeconomic reform to
macroeconomic stability. Supporters of the ‘New Economy’ thesis argue that
market-oriented microeconomic reform has increased the flexibility of the
economy. However, observation of the global business cycle over the past two
centuries gives little support for this view. Even though economic intervention
was very limited in the 19th century, severe recessions and depressions occurred
regularly. Similarly, government intervention, regulation and unionisation were
very limited in the United States in the 1920s, but that did not prevent the
occurrence of the Great Depression.
Despite these objections, the rhetorical appeal of the New Economy story
is obvious. The narrative is one of virtue rewarded, always a popular theme. The
rapid growth in wealth observed over the past few years is attributed, not to
macroeconomic good fortune and a global decline in interest rates, but to the
hard work and sacrifice of the 1980s and early 1990s.
‘The light that failed’
An alternative story, also popular with some advocates of microeconomic
reform agrees with the New Economy story in presenting rapid growth during
the cycle from 1993-94 to 1998-99 as representing the benefits of microconomic
reform. However, the period from 1999-00 onwards is viewed more
pessimistically. The slowdown in productivity growth is regarded as real, and the
result of a slowdown in the pace of microeconomic reform.
The major difficulty for this story is one of timing. While the Howard
government has taken a less consistent approach to reform than its Labor
predecessors, it introduced a number of major reforms in its first few years in
office. Many of these were measures that had long been demanded by advocates
of radical reform but resisted by the Labor government because of political
sensitivities. These included the Workplace Relations Act 1996 (Cwlth), the
partial privatisation of Telstra in 1998 and 1999, waterfront reform in 1998, and,
most notably, the Goods and Services Tax, introduced in 2000.
Moreover, many reforms introduced by the Hawke–Keating government
did not begin to take effect until after 1998. The most notably of these is
National Competition Policy. Most states did not even complete their legislative
reviews or set up their general regulatory bodies until 1998, and the National
Competition Policy process, with associated payments to the states was still not
completed by 2003-04.
If such an array of reforms is not sufficient to maintain even an average
rate of productivity growth, the whole rationale of microeconomic reform is called
into question. Far from generating sustained growth, the ‘light that failed’ story
suggests that the decade or more of microeconomic reform that began with the
floating of the dollar in 1983, produced only five years of above average
productivity growth before requiring a renewed burst of reform merely to sustain
In view of the substantial adjustment costs associated w i t h
microeconomic reform, the ‘light that failed’ story implies that, in many cases,
the net present value of microeconomic reform must be negative. It is generally
conceded, for example, that the short-term consequences of financial
deregulation included a relaxation of lending standards that contributed to the
severity of the 1990-91 recession. If the benefits of this reform were only
temporary, being exhausted by 1998-99, the present value, viewed from an ex
ante perspective, was almost certainly negative.
The most convincing argument for the ‘light that failed’ story is based on
the idea that international competition is becoming steadily more intense,
necessitating ever more radical reform. But this idea is obviously inconsistent
with the claim that microeconomic reform is associated with increasing welfare
and economic productivity. The whole point of economic progress is that the
choice set available to society should expand, not contract.
Again, however, the ‘light that failed’ story has a strong rhetorical appeal.
Surprisingly, perhaps, calls for sacrifice are always popular, and the view that it
is never time to rest on your oars can always count on a hearing. The ‘light that
failed’ story combines this rhetorical appeal with adherence to the claim that
Australia did indeed experience a productivity miracle.
The ‘blip’ story
The central point of the ‘blip’ story (Hancock 2005, McKenzie 2005,
Quiggin 2005) is that the productivity ‘surge’ of the 1990s was a statistical
illusion. The central theme of the story is that the notion of a ‘productivity cycle’
is misleading, since the correct basis for comparison is derived from the business
cycle rather than from internal properties of the series of productivity estimates.
Dividing business cycles into two or more productivity cycles is likely to produce
alternating periods of weak (contraction phases) and strong (expansion phases)
productivity growth. This point is observed by Dolman et al. (2006) who note (p.
42) ‘A period of strong multi-factor productivity growth is not typically followed
by another similar period.’
In addition to the general pattern of alternating expansion and
contraction phases, Quiggin (2001) points to a number of temporary factors that
led to an overestimation of MFP growth for the mid-1990s cycle. The most
important was an increase in work intensity, correlated with an increase in
reported and unreported working hours, and supported by widespread anecdotal
evidence. Reported working hours for full-time workers peaked around 2000, as
did popular discussion of increased work intensity. Thus, it seems likely that
gains in measured productivity from this source during the 1990s were, at least
partially, reversed after 2000.
Consideration of the MFP data supports this view. For the entire period
since 1993-94, including the most recent observation for 2004-05, the average
rate of MFP growth is 1.2 per cent, exactly the same as for the entire data
period. For the current incomplete macroeconomic cycle, beginning at the last
cyclical peak in 1988-89, the rate of MFP growth is below the long term average
Thus, the ‘blip’ story is parsimonious as an explanation, and fits the data
well. On the other hand, a negative finding lacks rhetorical appeal. The
publication bias against such findings is well-known (Scargle 2000).
To the extent that the ‘blip’ story has rhetorical appeal, it does so by
enhancing the contrast between the ‘Golden Age’ from World War II to the early
1970s and the long period of poor economic performance that began with the
breakdown of the Bretton Woods system in the early 1970s. Although the ABS
MFP data only goes back to the early 1960s, and begins with an anomalous
decline in MFP from 1964-65 to 1965-66, it seems likely that MFP growth was
strong throughout the Golden Age, which was also characterised by full
employment and steady reductions in the inequality of income distribution. By
contrast, the period since the early 1970s has been disappointing in all these
An analysis showing higher productivity growth for the period before
1970 compared to the subsequent period is appealing for those who prefer the
policies of the Golden Age, including Keynesian macroeconomic stabilisation, and
an expanding welfare state, to the less interventionist policies associated with
the program of microeconomic reform.
Considered in Popperian or Lakatosian terms, the ‘New Economy’ claim
that Australia experienced a productivity surge in the 1990s, driven by
microeconomic reform in the preceding decade must be regarded as a refuted
hypothesis. In the Popperian approach, the crucial test of a hypothesis is
prediction and potential falsification.
The natural prediction of the New Economy hypothesis, put forward at
the time the hypothesis was formulated was that the further reforms of the
1990s would generate continued strong growth in measured multifactor
productivity. This prediction was refuted by the observed outcome. Subsequent
attempts either to explain away contradictory evidence or to ‘save the
phenomena’ by advancing the modified ‘light that failed’ hypothesis may be seen
as evidence, in the terminology of Lakatos, of a degenerating scientific research
Considered in the rhetorical terms proposed by McCloskey (1983),
however, the ‘New Economy’ story and its variants remain highly successful. The
view that Australia’s long-running economic expansion, and the associated
growth in household wealth is the product of tough decisions in the past, rather
than a combination of asset inflation and adroit macroeconomic management has
This appeal may be found both in the optimistic ‘New Economy’ version
which projects a renewal of strong productivity growth as a result of reforms
already undertaken, and in the more pessimistic ‘light that failed’ story in which
we are in danger of losing the gains of the 1990s. Notably, while giving directly
opposed interpretations of the data for the most recent productivity cycle,
advocates of the two stories derive the same policy conclusion: more reform is
Australian Bureau of Statistics (various years), ‘Australian System of National
Accounts Cat No 5204’, ABS, Canberra.
Australian Bureau of Statistics (various years), ‘Indexes of productivity and
related measures spreadsheet 5204022’, ABS, Canberra.
Blaug, M. (1980) The Methodology of Economics : Or How Economists Explain,
Cambridge University Press, Cambridge.
Dolman, B., Lu, L. and Rahman, J. (2006), ‘Understanding productivity trends’,
Treasury Economic Roundup, Summer, 35–52.
Friedman, M. (1953), ‘The methodology of positive economics’, pp. 3–43 in Essays
in Positive Economics, (Ed, Friedman, M.) University of Chicago Press,
Gruen, D. (2001), ‘Australia’s strong productivity growth: will it be sustained?’,
Reserve Bank of Australia Bulletin, February, 62-69.
Hancock, K. (2005), ‘Productivity growth in Australia 1964-65 to 2003-04’,
Australian Bulletin of Labour, 31(1), 28–32.
Krugman, P. (1998), "I Know What The Hedges Did Last Summer,”Fortune,
http://web.mit.edu/krugman/www/xfiles.html, visited 6/4/06.
Lakatos, I. (1970), ‘Falsification and the methodology of scientific research
programmes’, in Criticism and the Growth of Knowledge, (Eds, Lakatos,
I.M. & A.) Cambridge University Press, Cambridge.
McCloskey, D. (1983), ‘The rhetoric of economics’, Journal of Economic
Literature, 21, 481–517.
McKenzie, M. (2005), ‘Microeconomic reform and productivity in Australia –boom
or blip’, Australian Conference of Economists, Melbourne.
Parham, D. (1999), ‘The new economy: A new look at Australia's productivity
performance’, Productivity Commission Staff Research Paper, Ausinfo,
Parham, D. (2000), ‘A more productive Australian economy’, 7(1), 3–16.
Parham, D. (2002), ‘Productivity growth in Australia: Are we enjoying a
miracle?’, Melbourne Institute and The Australian Conference, Towards
Opportunity and Prosperity, Melbourne.
Parham, D. (2005a), ‘Is Australia’s productivity surge over?’, Agenda, 12(3), 252–
Parham, D. (2005), ‘Australia's 1990s productivity surge: a response to Keith
Hancock's challenge’, Australian Bulletin of Labour, 31(3), 295–303.
Popper, K. (1963) Conjectures and Refutations: The Growth of Scientific
Knowledge, Harper and Row, New York.
Quiggin, J. (2001), ‘The Australian productivity miracle: a sceptical view’,
Agenda, 8(4), 333–48.
Quiggin, J. (2005), The surge we didn't have, Australian Financial Review, 10
Scargle, J. (2000), ‘Publication bias: The “file-drawer” problem in scientific
inference’, Journal of Scientific Exploration, 14(1), 91–106.