Comparing agricultural total factor productivity between Australia, Canada, and the United States, 1961-2006.
Sheng, Yu ; Ball, Eldon ; Nossal, Katerina 等
Global agricultural output has more than tripled over the past half
century, driven by new technologies and increased input use. This output
growth has helped to satisfy increasing demand for food and fibre as
population and income per capita have increased, and has thereby
stabilized global food prices. Agricultural productivity growth has
contributed significantly to these gains (Fuglie and Wang, 2012).
One of the most important drivers of agricultural productivity
growth is technological progress. This progress has followed two
distinct paths in developed countries, depending on the initial
endowment of resources. Those possessing relatively abundant capital and
land, for example Australia, Canada and the United States, have been
lead adopters of capital-intensive technologies such as reduced-till
cropping, yield mapping and mechanised mustering, and thus have achieved
high levels of output per worker. In contrast, land-scarce, labour-rich
developed economies such as Japan, South Korea and Taiwan have adopted
labour-intensive technologies such as green-housing and vertical farming
technologies, and thus have achieved high yields per unit of land
(Fuglie, Wang and Ball, 2012).
However, recent evidence suggests that agricultural productivity
growth is either stagnant or slowing in many countries (Alston, Beddow
and Pardey, 2010; World Bank, 2007; Sheng, Mullen and Zhao, 2011). This
is particularly the case in countries such as Australia and Canada that
have historically relied on the adoption of capital-intensive
technologies to drive productivity growth, where the inelastic supply of
natural resources (i.e. land) decreases the marginal benefits obtained
from adopting the embodied technology. In turn, this creates concern
about the sustainability of capital-intensive technological progress as
a source of ongoing productivity growth, compared with labour-intensive
technological progress. These concerns are heightened by decreasing
marginal returns to capital and tightening agricultural land supply
throughout the world.
To gather more empirical evidence on this issue, this article
calculates and compares total factor productivity (TFP) levels and
growth rates between Australia, Canada and the United States for the
1961-2006 period. Comparing output, input and productivity across
countries requires data on relative output and input prices, based on
purchasing power parity (PPP) estimates. We obtain these relative prices
by combining output and input prices with their quantities in each
country. In the estimation process, the Tornqvist index with the
Caves-Christensen-Diewert formula for transitivity is employed.
Productivity levels in each country are defined as the ratio of real
output to real input, which in turn are constructed as their values
divided by corresponding prices. Finally, we use a dynamic panel
regression analysis to link these TFP measures to potential determinants
in the three countries.
This article builds on previous research (for example, Ball et al.,
2001 and 2010) by constructing a consistent production account with
which to compile price and quantity data for agricultural outputs and
inputs in Australia, Canada, and the United States. In addition, the
accounting identity (whereby total output value equals total input
value) is used to derive unobserved returns to labour, enforcing the
assumption of constant returns to scale. Finally, a quality adjustment
has been applied to land and certain intermediate inputs to eliminate
the undesirable impact of embodied technological progress when
estimating TFP.
The article is organized into five sections. Section 1 provides a
review of methods and data used in cross-country comparisons of
agricultural productivity. Section 2 develops the data base for each
country and describes the method used to develop comparable productivity
estimates. Section 3 documents data sources for Australia, Canada, and
the United States. Section 4 presents the results and compares
agricultural productivity and its drivers between Australia, Canada, and
the United States. Section 5 concludes.
Cross-Country TFP Comparison in Agriculture: A Literature Review
While many studies have used index number methods to estimate
agricultural TFP in individual countries (Fuglie, Wang and Ball, 2012),
international comparisons remain challenging. Obtaining data remains the
most problematic issue, with some economists warning of
'insurmountable data constraints' in producing detailed
commodity datasets for the agriculture industry in different countries
(Craig, Pardey and Roseboom, 1997). Where established datasets are
available, differences in the treatment of variables limits the
comparability of input and output data (Capalbo, Ball and Denny, 1990).
Given these limitations, most cross-country comparisons have drawn
on data from the United Nations Food and Agriculture Organization (FAO).
Although it lacks price information and does not cover all inputs, the
FAO dataset covers many countries over a long time period. For example,
Craig, Pardey and Roseboom (1994 and 1997) estimated agricultural land
and labour productivity for 98 countries between 1961 and 1990 and found
that input mix, input quality and public infrastructure were significant
factors explaining agricultural productivity growth differences between
countries. While such partial productivity measures are likely to
overstate overall efficiency improvements (because they do not account
for changes in the use of capital and intermediate inputs), they
nonetheless provide some indication of factor-saving technical change
(Fuglie, 2010).
Coelli and Rao (2005) used FAO data to compare agricultural TFP for
93 countries between 1980 and 2000 using a Malmquist index and data
envelopment analysis (DEA). The Malmquist index method allows inputs and
outputs to be aggregated through a distance function, without the need
for price data. The results show that agricultural TFP growth was strong
across all countries before 2000, with some evidence of catch-up between
low and high performing countries. Ludena et al. (2007) also used the
Malmquist index method to estimate TFP growth for subsectors of the
agriculture industry (crops and ruminant and non-ruminant livestock) for
116 countries between 1961 and 2006. The study found that TFP growth in
developing and developed countries was converging for crop and
non-ruminant livestock production activities, and diverging in the
ruminant livestock sector.
While the Malmquist index method has some advantages (for example,
no price information is needed for TFP estimation), it also has
disadvantages. In particular, it is sensitive to the set of countries
compared, and the number of variables in the model (Lusigi and Thirtle,
1997). Without a large cross-section of countries, TFP estimates are
likely to suffer from measurement errors. Also, estimates from Malmquist
index numbers often seem implausible (Coelli and Rao, 2005; Headey,
Alauddin and Rao, 2010), possibly because of the unrealistic implicit
shadow prices derived for aggregation (Coelli and Rao, 2005).
For these reasons, wherever reliable price data are available,
'superlative' index methods are preferred. Superlative index
number methods are widely adopted by national statistical agencies and
are recommended by the OECD (2001) for productivity statistics.
Fuglie (2010) used a Tornqvist index to estimate and compare
agricultural TFP growth for 171 countries. While FAO data were used, and
were augmented using a fixed set of average global prices from Rao,
Maddison and Lee (2002) for revenue shares, and using input elasticities
from country-level case studies for cost shares. Fuglie (2010) found
that global agricultural TFP growth had accelerated in recent decades,
particularly among developing countries such as China and Brazil. This
contrasts with recent estimates of yield and labour productivity which
indicate a global slowdown (Alston, Beddow and Pardey, 2010).
After considering various approaches for performing inter-region
comparisons of agricultural prices, quantities and productivity, Ball et
al. (1997) identified two suitable options: the Fisher index with an
Elteto-Koves-Szulc formula (Elteto and Koves 1964; Szulc 1964) and the
Tornqvist index with the Caves-Christensen-Diewert formula (Caves,
Christensen and Diewert, 1982). Ball et al. (2001, 2010) conducted
empirical studies to examine these approaches. To address the data
challenges facing international comparisons of agricultural
productivity, Ball et al. (2001, 2010) developed an internationally
consistent production account system for collecting agricultural input
and output data from individual countries.
Ball et al. (2001) compared agricultural TFP between the United
States and nine European Union countries. Using 1990 as the base year,
Ball et al. (2001) derived bilateral Fisher price indexes adjusted by
purchasing power parity and then by the Elteto-Koves-Szulc (EKS) formula
(for transitivity). Indirect quantity indexes of outputs (inputs) were
then estimated as total output (input) value divided by the
corresponding price index. The results showed that agricultural
productivity converged between the United States and the nine European
Union countries between 1973 and 1993. Accordingly, most of the observed
disparity in output levels between these countries is caused by
differences in input use.
Ball et al. (2010) further developed the method for comparing TFP
across countries by applying Tornqvist price indexes and the
Caves-Christensen-Diewert formula (to impose transitivity across
countries). These studies compared competitiveness between the United
States and eleven European Union countries over the period 1973 to 2002.
Ball et al. (2010) found that the apparent catch-up of the European
Union countries was reversed after the mid-1990s, and significantly
weakened the competitiveness of European Union agriculture relative to
that of the United States.
Using the method advanced by Ball et al. (2010), this article uses
production account data for agriculture in Australia, Canada, and the
United States to compare agricultural TFP between countries and identify
its potential drivers.
Measuring Output, Input and TFP in Agriculture
In this section, we briefly discuss the index number method used
for multilateral comparison of agricultural productivity levels. When
using this method, the TFP index is defined as the ratio of the index of
real output to real input, which in turn are obtained from the nominal
values of output and input by the corresponding price indexes. The
construction of the output and input price indexes takes into account
prices and quantities of the individual components in each country, and
is adjusted for cross-country comparability. To identify its underlying
drivers, measures of TFP are then regressed against factors such as
climate conditions, public R&D knowledge stock, and infrastructure.
Index Method for TFP Estimates
Theoretically, TFP is measured as real output, [Y.sub.t], divided
by real input, [X.sub.t], and its growth is measured as the difference
between output and input growth rates (estimated using logarithmic
differentials to time t).
(1 [TFP.sub.t] = [Y.sub.t]/[X.sub.t]
(2) dln ([TFP.sub.t])/dt = dln ([Y.sub.t])/dt - dln[X.sub.t])/dt
where [X.sub.t] includes land, capital, labour and intermediate
inputs.
Both direct and indirect methods can be used to derive real output
and input. In practice, an indirect approach is usually preferred,
whereby real output and input quantities are measured as the gross value
of outputs or inputs divided by a corresponding price index, since value
data for most outputs and inputs are more readily available than
quantity data. Assuming perfect competition and a linearly homogenous
production function, direct and indirect quantity estimates are
equivalent when using a superlative index that satisfies the factor
reversal test (Diewert, 1992). In this sense, the estimation of real
output, real input and productivity is converted into the estimation of
output and input relative prices.
For each country, output and input price indexes can be obtained by
using a Tornqvist index to approximate a linearly homogeneous translog
function, such that
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
(4) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [R.sub.i] is the revenue share of the ith output and
[S.sub.j] is the cost share of the jth input. [P.sub.i] and [W.sub.j]
are the prices of the ith output and jth intput, respectively.
We use the Tornqvist index for two reasons. First, although the
Tornqvist index only satisfies the weak factor reversal test (which
states that the product of price and quantity indexes should yield the
expenditure), it nonetheless provides a reasonable second-order
approximation. (2) Second, Ball et al. (1997) also showed that the
Tornqvist index retains a high degree of characteristicity when combined
with the Caves-Christensen-Diewert (CCD) formula for transitivity
(Drechsler, 1971). This means that a price index estimated when using
this method is not dependent on the basket of goods in one particular
country that is used in the comparison.
Purchasing Power Parity Adjustment
To enable cross-country level comparisons, output and input price
indexes measured in domestic currencies must be converted to a common
'international' currency. The common currency estimates of
relative input and output price levels produced by market exchange rates
do not necessarily represent the purchasing power parity estimates.
Instead, relative price indexes for agricultural output and input were
constructed to capture each country's purchasing power parity
(PPP). For example, the PPP of wheat in Australia was defined as the
amount of Australian dollars required to purchase the same quantity of
wheat as one 2005 U.S. dollar.
In this article, we used the CCD formula (Caves, Christensen and
Diewert, 1982), derived from the geometric average of bilateral
Tornqvist indexes, to compare output and input prices between countries
in a given base year (2005). Compared with the Fisher index adjusted by
the EKS formula, this method has the advantage that a complete matrix of
bilateral Tornqvist indexes is not required, but instead a man-made
country average can be used as a numeraire.
Specifically, the difference between logarithms of the price of
output for any two countries can be expressed as weighted averages of
the differences between logarithms of the component prices and the
geometric average of component prices for the three countries.
Therefore, relative to the United States in the base year, the output
price for other countries in the same year can be written as:
(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
and [R.sup.d.sub.i] is the value share of the components in the
output aggregates. d = AU, CA, US denotes Australia, Canada, and the
United States, and C is the number of countries in the comparison.
Similarly, we can also write the input price for other countries
relative to the United States as:
(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
[bar.ln[W.sub.j]] = 1/c x [summation over (d)][W.sup.d.sub.j] and
[s.sup.d.sub.j] is the cost share of the components jn the input
aggregates.
The price indexes in Equations (5) and (6) represent the PPP
between the currencies of the two countries expressed in terms of
agricultural output and input respectively. Finally, the Tornqvist index
was used to chain-link the 2005 cross-country comparable prices to
construct a time series in each country.
Data Sources
In this section, we briefly discuss the two types of data used in
this paper. Data used for productivity measures were sourced from
Australia, Canada, and the United States. A cross-country consistent
production account was developed for agriculture and the same definition
and method was used to derive each variable, although data were
collected from different sources (a complete list of variables is
provided in Appendix A). Most variables were collected for the period
from 1961 to 2006, except for capital investment and asset prices, for
which a longer time series was used. (3) Data used to construct
independent variables for the productivity regression were obtained from
internationally consistent databases such as the World Bank's World
Development Indicator database and the Global Historical Climatology
Network.
Australia
Agricultural output quantity and value data were sourced primarily
from the Australian Bureau of Agricultural Research Economics and
Sciences' (ABARES) Agricultural Commodity Statistics. For some
smaller commodity items price data were not available, and so an ABARES
index of farm prices received was used instead.
Capital investment data were taken from the Australian Bureau of
Statistics (ABS) National Accounts Database from 1960, and backcast to
1860 using data from Butlin (1977) and Powell (1974). Since no data are
available for the deflator for transportation vehicles between 1920 and
1960, it is assumed to be the same as that for plant and machinery.
Data from the ABS Agricultural Census was used to estimate the land
area used for agricultural production. Land prices were estimated using
ABARES' Australian Agricultural and Grazing Industry Survey data
after 1978 and backcast to 1960 using a GDP deflator. For the base year
(2005), more detailed data on land area and prices across 226
statistical local areas were collected for a hedonic regression
analysis. Data on intermediate inputs (including total expenditure and
price indexes) were sourced from ABARES' Agricultural Commodity
Statistics.
The labour input quantity was estimated as the total number of
hours worked each year, calculated by multiplying the number of workers
by the average number of hours worked in a week, and the number of weeks
worked each year. The average number of hours worked was obtained from
the ABS Population Census and it is assumed there are 52 weeks of work
each year.
Canada
Output quantity data were not directly available for Canada, but
were estimated from total income from sales to processors, consumers,
exporters and farm households (including within-sector use, waste,
dockage, loss in handling and changes in closing stocks). Output price
data were available from Statistics Canada CANSIM tables. Some
non-separable forestry outputs were included in the aggregate output
estimates.
A capital investment data series was compiled for the period 1926
to 2006. Data were not available for some early of this period, and so
imputations were applied at the beginning of the investment series.
Investment deflators (i.e. a price index) were constructed for the
period 1926 to 1935 using import price data taken from CANSIM tables.
For other years, disaggregated deflators for each asset grouping were
taken directly from the national account statistics.
Land area data were sourced from the Canadian Agricultural Census,
while land price data were obtained from the Canadian Agricultural
Value-Added Account. All data series started from 1981, and were
backcast using a fixed proportion of agricultural land in the total
land, which was derived from the Census.
Data on intermediate input quantities and values were taken from
the Statistics Canada publication Supply Disposition Balance Sheets, and
other industry statistics. Individual price indexes were obtained from
Statistics Canada or were imputed using a combination of prices.
Finally, for inputs where data were unavailable, values were estimated
to be 1 to 3 per cent of total costs and were added into the production
account of agriculture.
The hired labour input was estimated using data from the Canadian
Labour Force Survey and the Population Census of Canada. Estimates of
the self-employed labour input (defined as the number of hours worked)
were based on data from the Canadian Agricultural Census. The number of
days worked were then converted into number of hours worked assuming 10
hours a day worked for 1961 to 1991, and using actual hours worked
(obtained from the Canadian Labour Force Survey) for 1991 onwards. The
input of unpaid family members was estimated as a proportion of the
self-employed labour input.
United States
Agricultural output values were constructed by aggregating
state-level data on farm cash receipts compiled by the United States
Department of Agriculture Economic Research Service (USDA ERS). Price
data were sourced from the USDA for most outputs and intermediate
inputs.
Capital investment data were sourced from the Bureau of Economic
Analysis, and deflators for transport vehicles were obtained from the
Bureau of Labor Statistics. For non-dwelling buildings and structures,
the implicit price deflator from the US National Accounts was used.
County-level land area data were collected from the US Census of
Agriculture with interpolation between census years using spline
functions and prices were obtained from the annual USDA survey on
agricultural land values.
Intermediate input data were sourced from the USDA state farm
income database. Price data were sourced from the National Accounts, the
US Monthly Energy Review and the USDA agricultural prices database.
Labour input data for hired and self-employed workers were sourced
from the US Census of Population and the US Current Population Survey.
Variable Definition for Potential Productivity Drivers
Estimating equation (7) requires data for a series of variables
that reflect potential productivity drivers. These variables include the
stock of research and development (R&D) knowledge, the
capital-labour ratio, rural infrastructure levels, the urbanisation
rate, temperature, and rainfall. These variables are defined below.
The knowledge stock of R&D in agriculture is considered to be a
better indicator of disembodied technological progress than R&D
investment. This is because there are often long lags before farmers
begin accessing the outputs of R&D investment. We define the R&D
knowledge stock as the weighted average of past public investments in
agricultural R&D following the methods used by Alston et al. (2010a)
and Sheng et al. (2011). Specifically, weights are obtained from an
assumed R&D lag profile that reflects the dynamics of knowledge
creation, use and depreciation. For all the three countries, we assumed
that the R&D lag profile takes the form of a gamma distribution with
the length of average service life of 35 years. Data on public R&D
investment in agriculture are obtained from ABARES for Australia, USDA
ERS for the United States and Statistics Canada for Canada.
To capture differences between countries in the extent of capital
deepening and therefore in associated embodied technology, we use the
capital-labour ratio as a control variable. (4) This variable is defined
as the aggregate capital input divided by the labour input. For all
three countries, the capital input is consistently defined and derived
from the stock of three depreciable assets, namely non-dwelling
buildings and structures, transportation vehicles and other plant and
machinery. The labour input is defined as the total number of hours
worked by both hired workers and unpaid proprietors, adjusted by their
age, education and experience.
Road transport plays an important role in agricultural production
in all three countries, and so we used the average per-capita length of
roads in rural areas to approximate the level of rural infrastructure
available to farmers. Specifically, this variable was defined as the
total length of roads in the rural areas of each country divided by the
rural population. In addition, we also used the urbanisation ratio,
defined as the proportion of the urban population in the total
population, as a control variable to represent changing economic
development levels in each country. Data used to construct those
variables were obtained from the World Bank World Development Indicator
database, which provides cross-country consistent measures of these
variables. (5)
[GRAPHIC 1 OMITTED]
[GRAPHIC 2 OMITTED]
Finally, all three countries have significant shares of
non-irrigated cropping and grazing in their agriculture sectors, hence
changes in rainfall and temperature will affect agricultural
productivity from year to year. To consider these effects, we used total
precipitation and average temperature for the crop growing season in
each country. Reflecting the difference in seasons between the Northern
and Southern hemispheres, the growing season for Australia is defined as
September to April while for Canada and the United States it is defined
as March to October. Data used to construct these two variables were
obtained from the Global Historical Climatology Network.
Agricultural TFP Estimates
Using the index method and production account data, we estimate and
compare agricultural TFP between Australia, Canada, and the United
States. A dynamic panel regression technique is then used to link the
productivity differences between countries to some of its potential
drivers.
Productivity Comparison Between Australia, Canada, and the United
States
Australian agricultural TFP was generally below the level achieved
by the United States and Canada from 1961 to 2006 (although in 2001
Australia's TFP level briefly exceeded the level achieved by
Canada) (Chart 1), but its growth was relatively strong. Between 1961
and 2006, the annual growth rate of agricultural TFP in Australia was
1.6 per cent a year on average, higher than in Canada (1.2 per cent a
year), (6) and only slightly lower than in the United States (1.8 per
cent a year). Australia's relatively strong TFP growth allowed it
to improve its TFP level relative to Canada and to maintain its TFP
level at around 70 per cent of the United States (Chart 2).
While Canada and the United States had similar levels of
agricultural TFP during the 1960s, they have since diverged. The average
level of agricultural productivity in Canada fell to 70 per cent of that
in the United States in 2001, before rebounding to about 80 by the
mid-2000s (Chart 2). A further analysis of productivity growth in the
most recent decade of the dataset showed that Canada experienced a
downturn in agricultural productivity associated with drought in the
early 2000s, but this slowdown was not sustained. In contrast, Australia
experienced a slowdown in agricultural productivity growth from 1998
onwards. This slowdown widened the TFP gap between Australian
agriculture and its North American competitors, particularly between
2002 and 2006. This finding is consistent with Sheng, Mullen and Zhao
(2011) who identified a turning point in broadacre agricultural
productivity in Australia after the mid-1990s, associated with poor
seasonal conditions and a decline in the intensity of public R&D
investment.
Sensitivity Check
To examine the sensitivity of our agricultural TFP estimates for
the three countries, we compare them with those obtained from Fuglie and
Rada (2013) for the period 1961-2006 and those from Coelli and Rao
(2005) for the period 198 02000. Differences in estimated TFP growth
rates between studies reflect differences in both methodology and data,
and provides some useful insight into cross-country comparisons of
agricultural productivity.
Table 2 presents the comparison results. Generally, the magnitude
of TFP growth rates estimated in this article are similar to those
obtained in earlier studies during both periods, but our estimates are
closer to those of Fuglie (2010). Since both Coelli and Rao (2005) and
Fuglie (2010) used the Malmquist index with FAO data, this result
reflects (to some extent) the relatively weak performance of the
Malmquist index for generating cross-country consistent estimates of
agricultural productivity compared with the Fisher or Tornqvist indexes.
Given the Malmquist index only uses the quantity information for outputs
and inputs, this highlights the importance of collecting price
information when performing cross-country comparisons of agricultural
productivity.
In addition, there were significant differences between countries
in the variation in estimated TFP growth across studies. In particular,
the difference in TFP growth rates across studies for Canada was around
three times larger than that of Australia and the United States (Table
2).
This difference might reflect the use of different data sources in
earlier studies, which reinforces the importance of constructing a
production account that is comparable across countries when performing
international comparisons of agricultural productivity.
Agricultural TFP Drivers Estimation Strategy for TFP Driver
Analysis
To investigate cross-country differences in agricultural
productivity, we use a dynamic panel data regression to analyse the
relationship between TFP measures and some potential drivers. Following
Alston et al. (2010a), Ball et al. (2001, 2010) and OECD (2012), we
focus on three such drivers, namely technological progress, capital
deepening and the availability of infrastructure, while controlling for
the impacts of climate conditions and the market environment. The
empirical specification is, for simplicity, assumed to take the
log-linear form so that we can interpret coefficients in terms of
percentage changes in TFP:
(7) ln[TFP.sub.it] - [[beta].sub.0] + [summation over
(j)]ln[X.sup.j.sub.it] + [summation over
(k)][[gamma].sub.k]ln[Z.sup.k.sub.it] + [[epsilon].sub.it]
where ln[TFP.sub.it] is the logarithm of agricultural TFP of
country i at year t. ln[X.sup.j.sub../.] denotes the logarithm of
potential drivers of productivity across countries, ln[X.sup.j.sub..,.]
denotes the control variables and [[epsilon].sub.it] are the residuals.
[[beta].sub.j] and [[gamma].sub.k] are the coefficients associated with
productivity drivers and control variables respectively, which capture
their marginal effects on agricultural productivity. The null hypothesis
is that by is insignificant, suggesting that there is no causal
relationship between the potential drivers and cross-country
productivity growth and vice versa.
Implementing equation (7) may encounter potential endogeneity
problems because of possible correlation between independent variables
and the residual. To avoid this problem, we use a flexible combination
of lagged independent variables and the differential of rainfall and
temperature as instruments, (7) and adopt the generalized method of
moments (GMM) to perform the estimation. A difference GMM is used as it
eliminates country-specific fixed effects. In addition, we have also use
the Arellano-Bond test to examine autocorrelation of the error terms (8)
and the Sargan/Hansen test to examine the identification issues. (9)
Three scenarios are specified. In scenario one (Model 1), we
consider a baseline model which only incorporates the variables that
represent climate conditions into the regression. Scenario two (Models
2-4) individually adds disembodied technological progress, the
capital-labour ratio and infrastructure-related variables into the
baseline model to explore their roles in explaining cross-country
productivity differences. Scenario three (Model 5) adds all these
factors together into the baseline model and investigates their combined
effects.
A Dynamic Panel Data Regression Analysis
Many factors have been used to explain differences in agricultural
productivity between countries. These include climate conditions,
technological progress and innovation, public infrastructure and
domestic policies (Alston, Beddow and Pardey, 2010; Ball et al., 2010;
OECD 2012). In this research, a dynamic panel regression analysis was
used to estimate the relative contribution of these factors to
agricultural productivity growth based on annual data for the 1961-2006
period for Australia, Canada, and the United States. Reflecting concerns
on the limited sample size, standard errors were corrected (by using the
finite-sample regression procedure) to avoid the potential for downward
bias (Windmeijer, 2005) and country-specific cluster effects were
accounted for. The results obtained from different scenarios, and the
corresponding Arrellano-Bond test and the Sargan/Hansen tests are
reported in Table 3. Three main findings are discussed below.
First, R&D spending has played an important role in promoting
agricultural productivity growth across the three countries. After
controlling for climate conditions and services from public
infrastructure, the coefficients associated with R&D knowledge
stocks (which generate the service flow of disembodied technologies) are
0.34 and 0.35, which are positive and significant at the 1 per cent
level. The results are consistent throughout different scenarios (Models
2 and 5), implying that a one per cent increase in the R&D knowledge
stock tends to raise the agricultural TFP level by more than 0.3 per
cent. Similar results were also obtained in Alston et al. (2000), Pardey
et al. (2006) and Alston (2010), which showed productivity improvements
in agriculture were strongly associated with lagged R&D investments.
This suggests that further increasing agricultural R&D investment
remains an effective way for policy makers to achieve long-term
productivity growth in agriculture.
Second, simply increasing the capital-labour ratio without adopting
new technology and changing farm practices does not necessarily increase
agricultural TFP in the three countries. In our regressions, the
coefficients attached to the capital-labour ratio are not significant at
the 10 per cent level and become negative when other factors are
accounted for (Column 5 in Table 3). This implies that increasing the
capital-labour ratio through making more investment does not necessarily
contribute to improved TFP levels. (10) For example, to implement
reduced-till technology in the cropping industry, farmers in Australia,
Canada, and the United States invested heavily in more powerful
tractors/machinery and larger pieces of land during the 1980s and 1990s.
By the 2000s, however, this technology was widely adopted, and further
investment in powerful tractors/machinery encountered decreasing
marginal rates of return, particularly on farms with relatively modest
size (Sheng et al., 2015).
Although this finding appears inconsistent with that of Ball et al.
(2001), who found that an increase in the capital-labour ratio
significantly contributed to reducing the difference in agricultural
productivity between European countries and the United States, it is
nonetheless a reasonable result. On one hand, all three countries in
this analysis have been widely adopting capital-intensive technologies
such as minimum 10 or no tillage and yield mapping for some time, and
accordingly, further investment in physical capital may lead to
decreasing marginal returns to capital. As such, labour and land
productivity can still increase with more investment, but total factor
productivity will decline. On the other hand, a positive correlation
between a change in the capital-labour ratio and agricultural TFP in
Ball et al. (2001) could result from the interaction between the
capital-labour ratio and other productivity drivers (i.e. R&D
knowledge stock), which have not been well considered in this study.
Empirically, this follows from the observation that the marginal effects
of the capital-labour ratio on productivity become negative when other
variables are controlled for (Models 3 and 5 in Table 2).
Third, climate conditions and public policies targeted to improve
the supply of public infrastructure services may also contribute to
explaining cross-country differences in TFP growth. In all scenarios,
the coefficients estimated for growing-season rainfall and temperature
are positive and significant at the 1-5 per cent level (Models 1 and 5
in Table 3). This suggests that agricultural TFP is sensitive to average
growing-season rainfall and temperature in the three countries, which
implies that differences in climate condition could be causes of
differences in TFP growth. In addition, the results also show that the
availability of public infrastructure and the urbanisation ratio (a
proxy for the level of economic development) appear to have positive
impacts on productivity. For example, the coefficients associated with
the rural infrastructure index and the urbanisation ratio are all
positive and significant at the 1-10 per cent level (Models 4 and 5 in
Table 3).
Robustness Check
To establish whether or not the findings obtained from the analysis
of productivity drivers are sensitive to the choice of methods and the
independent variables included, we carried out two robustness tests.
First, with respect to the choice of regression methods, it has
been argued that the dynamic panel data regression technique is less
efficient than the standard panel data regression technique when a long
time-series of data is available (Zilak, 1997; Wooldridge, 2010). To
check the sensitivity of our finding to the regression method used, we
re-estimated equation (7) using a panel data regression with fixed
effects (for Scenario 1 in Table 4). The results obtained from this
regression are similar to those obtained from the dynamic panel data
regression.
Second, with respect to the choice of independent variables, it
could be argued that the three countries produce different mixes of
output, and produce some country-specific products, which may bias the
regression results. For example, Canada and the United States produce a
large amount of maize and soybeans in the crop sector and beef and
cattle in the livestock sector while Australia produces more canola in
the crop sector and more sheep and wool in the livestock sector. Without
accounting for this disparity in output mix between countries, the
estimated contribution of various productivity drivers would be biased
since TFP could vary between farms. To examine the sensitivity of our
findings to this possibility, we added an index of output similarity to
the regression. (11) The results are similar to those obtained from the
basic model, although model fit is higher.
Conclusion
This article has estimated and compared agricultural total factor
productivity in Australia, Canada and the United States between 1961 and
2006. To do this, a consistent production account for the agriculture
sector in all three countries was developed, and a multilateral index
was applied to construct comparable price and quantity estimates for
output and input in each country.
Our results show that these countries have experienced different
TFP levels and growth patterns over the past four decades, despite
primarily using capital-intensive technologies and possessing similarly
well-developed production systems. In particular, Australian agriculture
has experienced rapid productivity growth over four decades, which has
improved Australia's productivity level relative to Canada and
maintained it relative to the United States. In recent years however,
Australia's productivity growth rate has slowed relative to that of
Canada and the United States.
Further empirical analysis shows that agricultural productivity
differences are likely to be related to each country's capacity for
investing in R&D and the availability of infrastructure. Differences
in climate conditions are also found to be important causes of
differences in agricultural TFP between Australia, Canada, and the
United States. These findings provide useful insights into the
importance of public policies in promoting public R&D investment in
agriculture and providing infrastructure to the farm sector to sustain
productivity growth.
Although our estimates measure agricultural output, input and TFP
in the three countries, a shortcoming is that the time series ends at
2006 because of data constraints. Additional studies could better inform
policy-makers by updating the series, and by applying additional
regression analysis to examine the sources of the various productivity
experiences of these countries.
References
Alston, J.M., M.C. Marra, P.G. Pardey, and TJ. Wyatt (2000) "A
Meta-Analysis of Rates of Return to Agricultural R&D: Ex Pede
Herculem?" IFPRI Research Report No. 113.
Alston, J. M. (2010) "The Benefits from Agricultural Research
and Development, Innovation, and Productivity Growth," OECD Food,
Agriculture and Fisheries Working Papers No. 31, OECD Publishing.
Alston, J. M., J. M. Beddow and P.G. Pardey (2010) 'Global
Patterns of Crop Yields and Other Partial Productivity Measures and
Prices', in J. Alston, B.A. Babcock and P. Pardey (eds.), The
Shifting Patterns of Agricultural Production and Productivity Worldwide,
Midwest Agribusiness Trade Research and Information Center, Iowa State
University, Ames, Iowa.
Alston, J.M., M.A. Andersen, J.S. James and P.G. Pardey (2010a)
'Shifting Patterns of Agricultural Production and Productivity in
the United States', in J. Alston, B.A. Babcock and P. Pardey
(eds.), The Shifting Patterns of Agricultural Production and
Productivity Worldwide, Midwest Agribusiness Trade Research and
Information Center, Iowa State University, Ames, Iowa.
Alston, J.M., M. Andersen, J.S. James and P.G. Pardey (2010b)
"Revisiting the Returns to U.S. Public Agricultural Research: New
Measures, Models, Results and Interpretation," University of
Minnesota, Department of Applied Economics Staff Papers, Staff Paper
P10-8.
Andersen, M.A., J.M. Alston and P.G. Pardey (2011) "Capital
Services in US Agriculture: Concepts, Comparisons, and the Treatment of
Interest Rates," American Journal of Agricultural Economics, Vol.
93, pp. 718-38.
Ball, E., J.C. Bureau, J.P. Butault and R. Nehring (2001)
"Levels of Farm Sector Productivity: An International
Comparison," Journal of Productivity Analysis, Vol. 15, pp. 5-29.
Ball, V.E., A. Barkaoui, J.C. Bureau and J.P. Butault (1997)
"Aggregation Methods for Intercountry Comparisons of Prices and
Real Values in Agriculture: a Review and Synthesis," European
Review of Agricultural Economics, Vol. 24, pp. 183-206.
Ball, V.E., J.P. Butault, C. Mesonada and R. Mora (2010)
"Productivity and International Competitiveness of Agriculture in
the European Union and the United States," Agricultural Economics,
Vol. 41, pp. 611-27.
Box, G.E.P. and D.R. Cox (1964) "An Analysis of
Transformations," Journal of the Royal Statistical Society, Series
B, Vol. 26, pp. 211-52.
Butlin, M.W. (1977) "A Preliminary Annual Database, 1900-01 to
1973-74," Research Bank of Australia, Research Discussion Paper,
May.
Capalbo, S.M., V.E. Ball and M.G.S. Denny (1990)
"International Comparisons of Agricultural Productivity:
Development and Usefulness," American Journal of Agricultural
Economics, Vol. 72, pp. 1292-97.
Caves, D., L.R. Christensen and E. Diewert (1982)
"Multilateral Comparisons of Output, Input, and Productivity using
Superlative Index Numbers," Economic Journal, Vol. 92, pp. 73-86.
Coelli, T. & D.S.P. Rao (2005) "Total Factor Productivity
Growth in Agriculture: A Malmquist Index Analysis of 93 Countries,
1980-2000," Agricultural Economics, Vol. 32, pp. 115-34.
Craig, B.J., P.G. Pardey and J. Roseboom (1994) "International
Agricultural Productivity Patterns," University of Minnesota, CEFAP
Working Paper No. WP94-1.
Craig, B.J., P.G. Pardey and J. Roseboom (1997) "International
Productivity Patterns: Accounting for Input Quality, Infrastructure, and
Research," American Journal of Agricultural Economics, Vol. 79, pp.
1064-77.
de Avillez, Ricardo (2011) "A Half Century of Productivity
Growth and Structural Change in Canadian Agriculture: An Overview,"
International Productivity Monitor, Number 22, Fall, pp. 82-99.
Diewert, W.E. (1978) "Superlative Index Numbers and
Consistency in Aggregation," Econometrica, Vol. 46, pp. 883-900.
Diewert, W.E. (1992) "Fisher Ideal Output, Input and
Productivity Indexes Revisited," Journal of Productivity Analysis,
Vol. 3, pp. 211-48.
Drechsler, L. (1973) "Weighting of Index Numbers in
Multilateral International Comparisons," Review of Income and
Wealth, Vol. 19, pp. 17-34.
Elteto, O. and P. Koves (1964) "On a Problem of Index Number
Computation Relating to International Comparison," Szatistikai
Szemle, Vol. 42, pp. 507-18.
Eswaran, H., F Beinroth and P. Reich (2003) "A Global
Assessment of Land Quality," in K. Weibe (ed.), Land Quality,
Agricultural Productivity and Food Security (Edward Elgar: Cheltenham,
UK.)
Fuglie, K. and S. Wang (2012) "New Evidence Points to Robust
but Uneven Productivity Growth in Global Agriculture," Amber Waves,
September, Vol. 10, pp. 1-6.
Fuglie, K.O. (2010) "Total Factor Productivity in the Global
Agricultural Economy: Evidence from FAO Data," in J. Alston, B.A.
Babcock and P. Pardey (eds.), The Shifting Patterns of Agricultural
Production and Productivity Worldwide, Midwest Agribusiness Trade
Research and Information Center, Iowa State University, Ames, Iowa, pp.
63-95.
Fuglie, K.O., S. Wang and E. Ball (2012) "Productivity Growth
in Agriculture: An International Perspective," (CAB International:
Cambridge, MA.)
Headey, D., M. Alauddin and D.S.P. Rao (2010) "Explaining
Agricultural Productivity Growth: An International Perspective,"
Agricultural Economics, Vol. 41, pp. 1-14.
Huffman, W.E. and R.E. Evenson (2006) "Do Formula or
Competitive Grant Funds Have Greater Impacts on State Agricultural
Productivity?" American Journal of Agricultural Economics, Vol. 88,
pp. 783-98.
Jorgenson, D.W. and P. Schreyer (2012) "Industry-Level
Productivity Measurement and the 2008 System of National Accounts,"
Review of Income and Wealth, Vol. 59, pp. 185-211.
Ludena, C.E., T.W. Hertel, P.V. Preckel, K. Foster and A. Nin
(2007) "Productivity Growth and Convergence in Crop, Ruminant and
Non-Ruminant Production: Measurement and Forecasts," Agricultural
Economics, Vol. 37, pp. 1-17.
Lusigi, A. and C. Thirtle (1997) "Total Factor Productivity
and the Effects of R&D in African Agriculture," Journal of
International Development, Vol. 9, pp. 529-38.
OECD (2001) "Measuring Productivity: Measurement of Aggregate
and Industry-Level Productivity Growth," OECD Manual, OECD, Paris.
OECD (2012) "Cross Country Analysis of Farm Performance,"
Trade and Agriculture Directorate, Committee for Agriculture, OECD,
Paris.
Pardey, P.G., N. Beintema, S. Dehmer and S. Wood (2006)
"Agricultural Research: A Growing Global Divide?" Agricultural
Science and Technology Indicators Initiative, International Food Policy
Research Institute, Washington, DC.
Penson, J. B., D. W. Hughes and G. L. Nelson (1987)
"Measurement of Capacity Depreciation Based on Engineering
Data," American Journal of Agricultural Economics, Vol. 35, pp.
321-29.
Powell, R. A. (1974) "Technological Change in Australian
Agriculture, 1920-21 to 1969-70," Thesis, University of New
England, Armidale.
Rao, D. S. P, A. Maddison and B. L. Lee (2002) "International
Comparison of Farm Sector Performance: Methodological Options and
Empirical Findings for Asia-Pacific Economies, 1900-1994," in A.
Maddison, W.F. Shepherd, and R. Prasada (eds.), The Asian Economies in
the Twentieth Century (Edward Elgar, Northampton, MA), pp. 27-52.
Romain, R., J.B. Penson and R. Lambert (1987) "Capacity
Depreciation, Implicit Rental Prices, and Investment Demand for Farm
Tractors in Canada," Canadian Journal of Agricultural Economics,
Vol. 35, pp. 373-78.
Sanchez, P.A., C.A. Palm and S.W. Buol (2003) "Fertility
Capability Soil Classification: A Tool to Help Assess Soil Quality in
the Tropics," Geoderma, Vol. 114, issues 3-4, pp. 157-85.
Sheng, Y., E. Gray, J.D. Mullen and A. Davidson (2011) "Public
Investment in Agricultural R&D and Extension: An Analysis of the
Static and Dynamic Effects on Australian Broadacre Productivity,"
ABARES Research Report 11.7, Australian Bureau of Agricultural and
Resource Economics and Sciences, Canberra.
Sheng, Y., J.D. Mullen and S. Zhao (2011) "A Turning Point in
Agricultural Productivity: Consideration of the Causes," ABARES
Research Report 11.4 for the Grains Research and Research and
Development Corporation, Australian Bureau of Agricultural and Resource
Economics and Sciences, Canberra.
Sheng, Y., S. Zhao, K. Nossal and D. Zhang (2015)
"Productivity and Farm Size in Australian Agriculture:
Reinvestigating the Returns to Scale," Australian Journal of
Agriculture and Resource Economics, Vol. 59. pp. 16-38.
Shi, U. J., T.T. Phipps and D. Colyer (1997) "Agricultural
Land Values Under Urbanizing Influences," Land Economics, Vol. 73,
pp. 90-100.
Szulc, B. (1964) "Indices for Multiregional Comparisons,"
Przeglad Statystyczny, Vol. 3, pp. 239-54.
World Bank (2007) World Bank Development Report 2008, World Bank,
Washington, DC.
World Bank (2015) World Development Indicator Database, World Bank,
Washington, DC, available at the website of http://data.worldbank.org/
data-catalog/world-development-indicators.
Zilak, J. P. (1997) "Efficient Estimation with Panel Data When
Instruments Are Predetermined: An Empirical Comparison of
Moment-Condition Estimators", Journal of Business and Economic
Statistics, Vol. 15, pp. 419-431.
Wooldridge, J. (2010) Econometric Analysis of Cross Section and
Panel Data, 2nd Edition, (MIT Press: Cambridge, MA).
Appendix A
Cross-Country Consistent Production Account for Agriculture:
Methodology and Variable Definition
The agricultural production account is defined and collected
consistently between countries, and data are obtained from a range of
sources in each country. This appendix provides a definition for each
output and input (summarized in Table A1). All data were collected on a
calendar year basis. For Australia, this meant converting financial year
data by taking a simple average of two consecutive financial years.
Outputs
Output variables were collected under three categories: crops,
livestock and other outputs. Crop outputs included grains and oil seeds,
vegetables, and fruits and nuts. Livestock outputs included slaughter
livestock (red meat), poultry and eggs, and other animal products (dairy
and wool). Other outputs included 'non-separable secondary
activities' such as income from machinery hire and contract
services.
Primary agricultural outputs included deliveries to final demand as
well as intermediate demand and on-farm use. Primary output is
approximated as total sales plus non-market transactions (that is,
cross-industry transfers through long-term contracts and on-farm use
such as animal feed). Where production statistics are not directly
available, primary output is approximated from changes in inventory for
each commodity.
Outputs from non-separable secondary activities are defined as
goods and services whose costs cannot be observed separately from those
of primary agricultural outputs. Two types of secondary activities are
included: on-farm production activities such as the processing,
packaging and marketing of agricultural products, and service provision
such as machinery hire and land lease.
Government taxes are included in agricultural outputs, since the
value of inputs is inclusive of indirect taxes. We recognise that
differences in government subsidies or taxes between countries may
create differences in the measured value of total output.
Equation (3) is used to aggregate output prices using their
corresponding revenue shares. The implicit aggregate output quantity
index is then defined as the total value of agricultural output divided
by the aggregate price index.
Inputs
Input variables were collected under four categories: capital,
land, labour, and intermediate inputs. Capital and land inputs are
estimated as service flows.
Capital
Following Ball et al. (2001 and 2010), three types of capital input
are distinguished: non-dwelling buildings and structures, plant and
machinery, and transportation vehicles. While relevant, the inventory of
crops, livestock, and other biological resources, such as vineyards and
orchards, are not included because of inadequate value data. However,
these capital inputs are likely to represent a relatively small
proportion of total capital.
Measurement of the capital input begins with using investment data
to calculate the stock of three types of capital goods. At each point in
time t, the stock of capital K(t) is the sum of all past investments, It
_ t, weighted by the relative efficiencies of capital goods of each age
t, St.
(A1) [K.sub.t] = [[infinity].summation over ([tau]=0)]
[S.sub.[tau]][I.sub.t - [tau]]
When using equation (A1) to estimate the capital stock, the
efficiencies of capital goods must be defined explicitly. Similar to
Ball et al. (2010), this is done by using two parameters, namely service
life of the asset, L, and a decay parameter, [beta], to specify the
functional form, S(.) such that:
(A2) S([tau]) = (L - [tau])/(L - [beta][tau]), if 0 [less than or
equal to] [tau] [less than or equal to] L
S([tau]) = 0 if [tau] > L
Each type of capital asset has an assumed distribution of actual
service life which provides a mean service life [bar.L]. In this
analysis, the asset lives for non-dwelling buildings and structures,
plant and machinery, and transport and other vehicles are assumed to be
40 years, 20 years and 15 years, respectively, with an assumed standard
normal distribution truncated at points two standard deviations before
and after the mean service life.
The decay parameter [beta] can take values between 0 and 1, with
[beta] = 1 implying that the capital asset does not depreciate over its
service life. Although there is little empirical evidence to define
appropriate values of [beta], it is reasonable to assume that the
efficiency of a capital asset declines smoothly over most of its service
life. Following Ball et al. (2001), decay parameters are set to be 0.75
for non-dwelling buildings and structures and 0.50 for all other capital
assets, reflecting an assumption that efficiency declines more quickly
in the later years of service (Penson, Hughes and Nelson, 1987; Romain,
Penson and Lambert, 1987).
The aggregate efficiency function was constructed as a weighted sum
of individual efficiency functions where the weights are the frequency
of occurrence.
Rental Rate
Assuming the sector invests when the present value of the net
revenue generated by an additional unit of capital exceeds the purchase
price of the asset, the farm sector will invest in capital stock
formation until the output price P satisfies:
(A3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where c is the implicit rental price of capital, r is the real rate
of return and [W.sub.K] is the price of an additional unit of capital
(or investment).
The rental price c consists of two components: the opportunity cost
associated with investment, [rW.sub.K], and the present value of the
cost of all future replacements required to maintain the productive
capacity of the capital stock,
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].
Let F denote the present value of the rate of capital depreciation
on one unit of capital according to the mortality distribution m:
(A4) F = [[infinity].summation over (t=1)] [(1 + r).sup.-t]
where m([tau]) = -[S([tau]) - S([tau] - 1)] for [tau] = 1, ... L.
It can be shown that
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
such that
(A5) c = [rW.sub.K]/1 - F
Following Ball et al. (2010), the real rate of return, r, is
approximated with an ex-ante rate, estimated as the nominal rate of
return minus inflation. The nominal rate of return is obtained using the
exogenous approach, and is derived from returns on government bonds with
a range of different maturities. The choice of interest rate is widely
debated. Andersen, Alston and Pardey (2011) argued that use of a fixed
interest rate generates more plausible estimates of capital services in
the United States than annual market rates, while Jorgenson and Schreyer
(2012) proposed using the residual of output value after deducting input
costs to derive an endogenous real interest rate. To test the
sensitivity of measured capital services to different real interest
rates, both ex-ante and ex-post rates were estimated. The ex-ante rate
was chosen for this study as it was less volatile than the ex-post rate.
Land
The value of land service flows is given by the product of the land
stock and rental price. The stock of land was estimated from the total
land areas operated. The rental price of land was obtained using
Equation (9) with the assumption of zero depreciation, i.e. c =
[rW.sub.L]. As explained below, the land price, [W.sub.L], was derived
from a hedonic function.
In particular, agricultural land prices can be affected by many
factors unrelated to agricultural production, such as urbanisation
pressures, distance to major cities and government land release
policies. Also, spatial differences in land quality may prevent direct
comparison of prices between regions and over time. To address these
problems, comparable land price indexes for each country were
constructed using the hedonic regression method.
In this article, the hedonic price of land is a generalised linear
function of its characteristics relevant to agricultural production and
some control variables. The function uses the Box-Cox (1964)
transformation to represent the dependent variable and each continuous
independent variable:
(A6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where the price of land, [W.sub.L]([[lambda].sub.0]), is the
Box-Cox transformation of real observations, when [W.sub.L] > 0, that
is:
(A7) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Similarly, [X.sub.n] ([[lambda].sub.n]), a vector of land
characteristics associated with agricultural production, is the Box-Cox
transformation of the continuous quality variable Xn where:
(A8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
and D is a vector of country dummies used to control for external
factors. For simplicity, D is approximated by a group of region and time
dummy variables and not subject to transformation; [lambda], [alpha] and
[gamma] are unknown parameter vectors to be determined in the regression
and e is a stochastic disturbance term. This expression can assume
linear, logarithmic and intermediate nonlinear functional forms
depending on the transformational parameter.
To apply the hedonic regression model, regional land prices and
land characteristics were observed for each country in 2005. Land
characteristic data for 2005 were sourced from the USDA World Soil
Resources Office and selected following Eswaran, Beinroth and Reich
(2003) and Sanchez, Palm and Buol (2003). GIS mapping was used to
overlay country and regional boundaries with land characteristics data
relating to particular soil categories, including soil acidity,
salinity, and moisture stress. Across the three countries more than 18
common variables were used to capture environmental attributes.
Two additional attributes affecting the price of agricultural land
should also be considered: irrigation and population accessibility.
Irrigation (the percentage of cropland irrigated) was included as a
separate indicator of production capacity in water-stressed areas, as
well as an interaction term between irrigation and soil acidity. A
population accessibility index could be used to control for the impact
of urbanisation and economic development on land prices; however, it was
not included in this analysis due to data constraints. Such indexes have
been constructed in previous studies by using a gravity model of urban
development, and provided a measure of accessibility to population
concentrations (Shi, Phipps and Colyer, 1997).
Intermediate Inputs
Intermediate inputs comprise all materials and services consumed,
excluding fixed capital, land and labour inputs. They include 10
categories, namely: fuel, electricity, fertilisers and chemicals, fodder
and seed, livestock purchases, water purchases, marketing services,
repairs and maintenance, plant and machinery hire, and 'other
materials and services'.
Fuel (including lubricants) and electricity are estimated from the
total quantity consumed and an index of prices paid by farmers for
petrol, diesel, liquefied natural gas and electricity. A fuel price
index was calculated using quantities of petrol, diesel and liquefied
natural gas consumed as weights. The total quantity of fuel consumed was
obtained by dividing total expenditure by this fuel price. The price of
electricity was estimated separately and used to deflate electricity
expenditure to obtain the quantity consumed.
Other intermediate inputs were estimated as implicit quantities.
Price indexes were sourced domestically, except for pesticides and
chemicals where quality-adjusted price data from 2005 were sourced from
the World Bank World Development Indicator database and FAO statistics.
The quality-adjusted data for 2005 were used with domestic time-series
prices to impute a trend.
Consistent with the treatment of output, intermediate inputs were
valued at farm-gate prices, including direct taxes and excluding
indirect taxes and subsidies.
Labour
Labour is defined as total hours worked by hired, self-employed and
unpaid family workers. Because data were only available on the number of
people employed in the agriculture industry, total hours worked were
imputed by multiplying the number of workers by the average hours worked
per week and the number of weeks. For consistency, we use 52 weeks a
year for this imputation.
Wages were not used to estimate the value share of labour inputs.
This is because hourly wages are unlikely to capture total compensation
to farm workers, since additional employee benefits (such as lodging and
superannuation contributions) are not included in wage statistics. In
addition, compensation to self-employed workers is not directly
observable.
Instead, the real cost of total labour use was derived using the
accounting assumption that the value of total output equals the value of
total input. This enabled real wages to be estimated as real labour
compensation (or total output value minus capital, land, and
intermediate input costs) divided by the total number of hours worked.
Finally, hired, self-employed, and unpaid family workers were
distinguished, and differences in education levels and work experience
were used to adjust prices for labour quality in all three countries.
Table A1
Summary List of Variables
Outputs
Crops
Grains and oil Fruits and
seeds nuts Vegetables
Barley Almonds Asparagus
Canola Apples (fresh,
Caster Apricots processing)
Cottonseed Avocadoes Snap beans
Flaxseed Bananas Bean (dry,
Hay and silage Cherries processing)
Maize (sweet) Broccoli
Oats Cherries (tart) Cauliflower
Peanut Cranberry Cabbage
Rice Dates Capsicum
Rye Figs Celery
Safflower Grapefruit Cucumber
Sorghum Grapes (fresh,
Soybean Hazelnuts processing)
Sunflower Lemons and Corn (fresh,
Triticale limes processing)
Wheat Macadamias Honeydew
Mandarins Lettuce
Other crops Mangoes Lentils
Nectarines Onions
Cotton lint Olives Peas (dry,
Tobacco Oranges green)
Horticulture Peaches Potatoes
Floriculture Pears Rock melon
Greenhouse Pecans Spinach
nursery Plums (fresh,
Sugar beet Prunes processing)
Sugar cane Strawberries Sweet
Mushrooms Tangelos potatoes
Tangerines Tomatoes
Other crops Walnuts (fresh,
not inlcuded Other fruit processing)
elsewhere and nuts Watermelon
Other
vegetables
Outputs
On-farm
Livestock activities
Cattle and Marketing
calves Packaging
Ducks Processing
Chickens and
broilers Services
Eggs
Hogs Contract
Milk, butter, services
cheese Machinery hire
Sheep and Land lease
lambs Other services
Turkey
Wool
Inputs
Land Capital Labour Materials
Land services Buildings and Operator Chemicals
structures labour Electricity
(non- Hired labour Fertiliser
dwelling) Unpaid Fodder and
workers seed
Plant Fuel and
machinery lubricant
Transportation Livestock
and other purchases
vehicles Water
purchases
Other
materials
Services
Marketing
Plant and
machinery
hire
Repairs and
maintenance
Veterinary
services
Other services
Yu Sheng
Australian Department of Agriculture
Eldon Ball
U.S. Department of Agriculture
Katerina Nossal
International Trade Centre (1)
(1) Yu Sheng is a senior economist of Agricultural and Resource
Economics and Sciences (ABARES) at the Australian Department of
Agriculture and Water Resources; Eldon V. Ball is a Senior Economist
with the Economic Research Service, the US Department of Agriculture,
and Instituto de Economia, Universidad Carlos III de Madrid; Katarina
Nossal is an independent consultant and a former economist of
Agricultural and Resource Economics and Sciences (ABARES) at the
Australian Department of Agriculture and Water Resources. The authors
thank two anonymous referees for comments and all remaining errors
belong to the authors. Emails: yu.sheng@agriculture.gov.au;
eball@ers.usda.gov; katarina.nossal@gmail.com
(2) When the factor reversal test is satisfied, the direct and
indirect methods will lead to the same results. Under certain
assumptions, Diewert (1978) showed that the failure to satisfy the
factor reversal test is not a major problem when the Fisher or Tornqvist
indexes are used to estimate the price index.
(3) We use the perpetual inventory method to estimate the capital
stock and capital input for depreciable assets. Depending on the service
life of each capital asset, this method requires a long time-series of
data on investment and the purchasing price of each capital asset before
the starting period. For example, the average service life of
non-residential buildings and structures is 40 years. Given that the
real service life of most assets is distributed within two standard
deviations of the average service life, to construct the capital stock
of nonresidential buildings and structures in I960, we needed investment
data for at least 80 years prior to 1960.
(4) Theoretically, embodied technological change (in capital)
should have little effects on TFP if the associated adoption costs equal
to the benefits (or being well considered from the input or cost
perspective). However, in practice, the situation could be more complex
depending on the interactive relationship between the embodied and
disembodied technological progress (Kohli 2015). Specifically, if the
embodied technology progress brings more benefits than the adoption
costs (or it is positively correlated to the disembodied technology
progress), TFP would be higher; and vice versa.
(5) More details about the database are available at
http://data.worldbank.org/data-catalog/world-developmentindicators
(World Bank, 2015).
(6) This TFP growth rate is consistent with the TFP estimate of 1.0
per cent annually in Canadian crop and animal production over the
1961-2006 period based on Statistics Canada's official TFP
estimates available in CANSIM Table 383-0022.
(7) Our argument for using the differential of exogenous climate
change conditions as a valid instrument is based on the observation of
significant amount of efforts having been put into dealing with
adaptation to climate change at the national levels for all the three
countries. Specifically, lagged and differential of those variables
generally will not directly affect agricultural productivity of the
current period when land quality is well controlled (when TFP are
estimated). But, they are correlated to rainfall and temperature in the
current period, given that the climate is gradually evolving over time
(or time-contigent). In addition, when observing rainfall and
temperature in the previous period, farmers are willing to adapt to the
situation through changing R&D knowledge stock, infrastructure,
capital-labor ratio to adapt to the new environment.
(8) The Arrellano-Bond test is designed to examine whether there is
autocorrelation in the idiosyncratic disturbance term, and helps to
identify which periods of lagged variables could be used to obtain
instruments. The null hypothesis is that once fixed effects are removed,
the disturbance term is not autocorrelated.
(9) The Sargan/Hansen tests are designed to examine whether the
model is over-identified, and tests for the joint validity of the moment
conditions in the GMM framework. The null hypothesis of these tests are
that the chosen moments (or instruments) are jointly valid.
(10) Of course, increased capital intensity boosts labour
productivity. For example, de Avillez (2011) showed that capital
deepening accounted for just under one half of the 3.77 per cent
increase in output (value added) per hour in Canadian agriculture
between 1961 and 2007.
(11) An output similarity index relative to the US at the base year
(i.e. 2005) was estimated for Australia, Canada, and the United States
based on all agricultural outputs. The output similarity index ([omega])
is given by [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] where
[f.sub.im] and [f.sub.jm] are the value of production of output m,
expressed as a share of the total value of agricultural output in
country i (that is, Australia, Canada or the United States) at time t
and country j (that is, the United States at the based year 2005) where
there is a total of M different commodity categories for Australia (or
Canada) and the United States, and M = 16. Data on fm for Australia,
Canada and the United States, and data on [f.sub.jm] for the United
States at the base year are obtained from the output value estimates at
current prices. For a more detailed technical discussion, see Alston,
Beddow and
Table 1
Comparison of Relative Agricultural TFP Levels in
Australia, Canada and the United States, 1961-2006
US = 100 in 2005
United
Year Australia Canada States
1961 0.329 0.389 0.452
1962 0.338 0.457 0.455
1963 0.356 0.486 0.471
1964 0.354 0.454 0.482
1965 0.337 0.481 0.499
1966 0.353 0.519 0.500
1967 0.351 0.457 0.525
1968 0.353 0.477 0.542
1969 0.383 0.489 0.549
1970 0.377 0.477 0.546
1971 0.388 0.512 0.588
1972 0.386 0.484 0.584
1973 0.393 0.494 0.604
1974 0.411 0.464 0.566
1975 0.422 0.512 0.617
1976 0.438 0.517 0.607
1977 0.449 0.523 0.646
1978 0.459 0.530 0.616
1979 0.465 0.489 0.634
1980 0.454 0.507 0.607
1981 0.467 0.550 0.680
1982 0.428 0.557 0.695
1983 0.446 0.535 0.601
United
Year Australia Canada States
1984 0.504 0.527 0.710
1985 0.482 0.554 0.754
1986 0.482 0.596 0.740
1987 0.483 0.576 0.750
1988 0.483 0.540 0.713
1989 0.487 0.590 0.780
1990 0.532 0.637 0.814
1991 0.540 0.635 0.821
1992 0.529 0.629 0.895
1993 0.571 0.568 0.843
1994 0.536 0.650 0.902
1995 0.541 0.663 0.825
1996 0.620 0.710 0.901
1997 0.652 0.682 0.915
1998 0.658 0.709 0.897
1999 0.679 0.742 0.895
2000 0.674 0.695 0.939
2001 0.683 0.662 0.943
2002 0.624 0.670 0.937
2003 0.591 0.723 0.965
2004 0.643 0.785 1.024
2005 0.677 0.797 1.000
2006 0.652 0.770 1.003
Source: Authors' calculations.
Table 2
Agricultural TFP Growth Rates
(per cent)
United
Australia Canada States
Time period for
comparison: 1961-2006
Fuglie (2010) 1.53 1.57 1.67
Our estimates 1.64 1.24 1.80
Relative difference -0.07 0.27 -0.07
to our estimates
Time period for
comparison: 1980-2000
Coelli and Rao (2005) 2.60 3.30 2.60
Our estimates 2.14 1.73 1.99
Relative difference 0.21 0.91 0.31
to our estimates
Note: TFP growth rates are estimated using the
regession method.
Table 3
Dynamic Panel Regression on Agricultural TFP Levels,
Difference GMM Estimation Results
Scenario 1
Model 1 Model 2
Dependent
variable: lnTFP
ln_rainfall 0.056 ** (0.027) 0.054 ** (0.027)
(growing seasons)
ln_average_temp 0.068 *** (0.013) 0.064 *** (0.015)
ln_R&D knolwedge stock -- 0.337 *** (0.082)
ln_capital_labour_ratio -- --
ln_infrastructure_index -- --
urbanisation ratio -- --
Number of observations 135 135
F-statistics 417 1,148
Arrellano-Bond 0.191 0.164
test for AR(1)
Arrellano-Bond 0.331 0.235
test for AR(2)
Sargan test of overid. 1.000 0.424
restrictions
Hansen test of overid. 1.000 1.000
restrictions
Scenario 2
Model 3 Model 4
Dependent
variable: lnTFP
ln_rainfall 0.058 ** (0.028) 0.056 ** (0.004)
(growing seasons)
ln_average_temp 0.071 *** (0.010) 0.069 *** (0.007)
ln_R&D knolwedge stock -- --
ln_capital_labour_ratio 0.061 (0.065) --
ln_infrastructure_index -- 0.179 *** (0.013)
urbanisation ratio -- 0.059 * (0.031)
Number of observations 135 135
F-statistics 262 217
Arrellano-Bond 0.196 0.111
test for AR(1)
Arrellano-Bond 0.308 0.461
test for AR(2)
Sargan test of overid. 0.426 0.388
restrictions
Hansen test of overid. 1.000 1.000
restrictions
Scenario 3
Model 5
Dependent
variable: lnTFP
ln_rainfall 0.051 *** (0.005)
(growing seasons)
ln_average_temp 0.068 *** (0.009)
ln_R&D knolwedge stock 0.349 *** (0.112)
ln_capital_labour_ratio --0.001 (0.039)
ln_infrastructure_index 0.196 *** (0.013)
urbanisation ratio 0.031 ** (0.014)
Number of observations 135
F-statistics 321
Arrellano-Bond 0.119
test for AR(1)
Arrellano-Bond 0.406
test for AR(2)
Sargan test of overid. 0.341
restrictions
Hansen test of overid. 1.000
restrictions
Note: the numbers in parenthesis below coefficients are
"robust" standard errors taking into consideration of
heterosk-edasticity H(1), and *** p<0.01, ** p<0.05, *
p<0.1. Statistics for Arrellanno-Bond test for
autocorrelation and Sargan-Hansen test for over
identification are p-values.
Table 4
Robustness Checks on Estimation Method and Variable Choices
Scenario 1 Scenario 2
Panel Fixed Panel Fixed Difference
Effect Effect GMM
Dependent variable: InTFP
ln_rainfall 0.098 *** 0.135 *** 0.138 ***
(growing seasons) (0.008) (0.008) (0.008)
ln_average_temp 0.062 *** 0.063 *** 0.062 ***
(0.005) (0.004) (0.005)
R&Dknowledgestock 0.316 *** 0.213 *** 0.231 ***
(0.057) (0.039) (0.038)
ln_capital_ -0.018 -0.050 -0.053
labour_ratio (0.043) (0.046) (0.046)
ln_infrastructure_ 0.304 *** 0.350 *** 0.354 ***
index (0.059) (0.050) (0.050)
urbanisation_ratio 0.043 *** -0.037 *** 0.037 ***
(0.010) (0.010) (0.009)
similarity_index -- 0.838 *** 0.885 ***
(0.154) (0.155)
Constant 1.586 *** 0.458 *** --
(0.450) (0.028)
Number of 138 138 135
observations
Adjusted R-squared 0.883 0.901 --
Arrellano-Bond -- -- 0.150
test for AR(1)
Arrelanno-Bond -- -- 0.101
test for AR(2)
Sargan test of -- -- 0.350
overid.
restrictions
Hansen test for -- -- 1.000
overid.
restrictions
Note: the numbers in parenthesis below coefficients are
"robust" standard errors that account for heterochasticity
H(1), and *** p < 0.01, ** p < 0.05, * p < 0.1. Statistics
for Arrellanno-Bond test for autocorrelation and Sargan-
Hansen tests for over identification are p-values.
