public investment in agricultural r&d and extension: an ......yu sheng, emily m gray, john d...

Yu Sheng, Emily M Gray, John D Mullen* and Alistair Davidson

ABARES research report 11.7

September 2011

Public investment in agricultural R&D and extension: an analysis of the static and dynamic effects on

Australian broadacre productivity

Australian Bureau of Agricultural and Resource Economics and Sciences

Australian Government

www. aba r e s . g ov. a u

S c i e n c e a n d e c o n o m i c s f o r d e c i s i o n - m a k e r s

© Commonwealth of Australia 2011

This work is copyright. The Copyright Act 1968 permits fair dealing for study, research, news reporting, criticism or review. Selected passages, tables or diagrams may be reproduced for such purposes provided acknowledgment of the source is included. Major extracts or the entire document may not be reproduced by any process without the written permission of the Executive Director, Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES).

The Australian Government acting through ABARES has exercised due care and skill in the preparation and compilation of the information and data set out in this publication. Notwithstanding, ABARES, its employees and advisers disclaim all liability, including liability for negligence, for any loss, damage, injury, expense or cost incurred by any person as a result of accessing, using or relying upon any of the information or data set out in this publication to the maximum extent permitted by law.

Sheng, Y, Gray, EM, Mullen, JD and Davidson, A 2011, Public investment in agricultural R&D and extension: an analysis of the static and dynamic effects on Australian broadacre productivity, Australian Bureau of Agricultural and Resource Economics and Sciences, Canberra, September.

ISSN 1447-8358 ISBN 978-1-921448-97-3

Australian Bureau of Agricultural and Resource Economics and Sciences Postal address GPO Box 1563 Canberra ACT 2601 Australia Switchboard +61 2 6272 2010 Facsimile +61 2 6272 2001 Email [email protected] Web abares.gov.au

ABARES project 43218

Acknowledgements This research was funded by the Grains Research and Development Corporation and is part of the Harvesting Productivity Initiative.

The authors thank Cheryl Gibbs, Alistair Davidson, Peter Gooday, Jammie Penm and Terry Sheales (ABARES) for their contributions to the report, as well as Shiji Zhao and Prem Thapa (previously ABARES). The authors also gratefully acknowledge the feedback provided by Zoltan Lukacs (GRDC) and Gordon MacAulay (Principal Economist GrainGrowers and Emeritus Professor of Agricultural Economics at the University of Sydney) as well as participants at the Harvesting Productivity Initiative Working Group meeting in Canberra in 2009.

This report draws heavily on data collected in ABARES surveys of broadacre industries. The success of these surveys depends on the voluntary cooperation of farmers, their accountants and marketing organisations in providing data. The dedication of ABARES survey staff in collecting these data is also gratefully acknowledged. Without this assistance, the analysis presented in this report would not have been possible.

* John Mullen is an adjunct professor at the Institute for Land, Water and Society, Charles Sturt University, Orange.

iii

Foreword

Australian governments and industry stakeholders are committed to increasing productivity growth rates in agricultural industries. Many of the technologies and practices that have driven growth in agricultural productivity in Australia are the result of past public investments in agricultural research and development (R&D), and the extension of those innovations to farmers. But the rate of productivity growth in the broadacre industry has slowed in recent decades, due in part, to a combination of slow growth in public investment in agricultural R&D and drought. In response, some stakeholders have proposed various initiatives directed at achieving nominated target productivity growth rates.

This research provides evidence of the important contribution of public R&D and extension to broadacre total factor productivity in Australia. It finds that, over the past 50 years, knowledge and technology accumulated from past public investments in R&D and extension in Australia and overseas have accounted for almost two-thirds of average annual broadacre productivity growth. Further, given that creation of new technologies and management practices largely depends on the stock of research knowledge accumulated over decades, past R&D investment decisions are likewise expected to largely dictate productivity outcomes for some considerable time.

Funded by the Grains Research and Development Corporation as part of the GRDC–ABARES Harvesting Productivity Initiative, this research offers government and industry stakeholders a range of insights into the effects of R&D policies on farm performance. In particular, it will help those tasked with strategically balancing agricultural R&D and extension portfolios and underscore the importance of making strategic investment decisions that have long-term payoffs.

Phillip Glyde Executive Director ABARES

September 2011

iv

ContentsAcknowledgements ii

Foreword iii

Summary 1

1 Introduction 3

2 Trends in agricultural R&D and extension and broadacre productivity 5

3 Past research evaluating public R&D policies 9

4 Methodology 14

5 Data sources and variable definitions 21

6 Results 25

7 Policy implications and further research 33

Appendixes A Constructing knowledge stocks 37

B Derivation of the total knowledge stock treatment 40

C Alternative specifications (treatment of knowledge stocks) 43

D Derivation of the error correction model 44

E Results 46

F Contribution of public R&D and extension knowledge stocks to annual broadacre TFP growth 57

G Estimation of internal rate of return to public investment in agricultural R&D and extension 58

References 60

Boxes1 Shifting trends in public agricultural R&D policy in Australia 6

2 Construction of R&D and extension knowledge stocks 15

v

Public investment in agricultural R&D and extension ABARES 11.7

Figures1 Real public agricultural R&D and extension investment in Australia: 1952–53

to 2006–07 5

2 Share of extension in total public agricultural RD&E expenditure: 1952–53 to 2006–07 6

3 Broadacre TFP and the terms of trade, Australia, 1952–53 to 2006–07 8

4 Broadacre TFP by farm type,1977–78 to 2007–08 8

5 Real public investment in agricultural R&D (1952–53 to 2006–07) and rural Australia (1917–18 to 1952–53) 22

6 Domestic public R&D knowledge stocks with 35-year lags: 1952–53 to 2006–07 23

7 US public R&D knowledge stocks with 35-year lags: 1952–53 to 2006–07 23

8 Domestic public extension knowledge stock: 1952–53 to 2006–07 23

9 Moisture availability index: 1952–53 to 2006–07 24

10 Percentage of students aged between 4 and 19 enrolled in schools (5-year moving average): 1952–53 to 2006–07 24

11 Relative contributions of public R&D and extension to annual broadacre TFP growth (%) 28

12 Distributed lag functions (35-year lags) 38

13 Distributed lag functions (16-year lags) 39

Tables1 Median and average rates of return to public investments in agricultural research

and extension 9

2 Elasticities of broadacre TFP to public R&D and extension knowledge stocks and other explanatory factors 26

3 Contribution of R&D and extension knowledge stocks to broadacre TFP growth: 1952–53 to 2006–07 27

4 Elasticities of broadacre TFP to public R&D and extension knowledge stocks and other explanatory factors 30

5 Average IRR to domestic public investment in broadacre R&D and extension (%) 31

6 Marginal internal rate of return (MIRR) to domestic public investment in broadacre R&D and extension (%) 32

7 Projected effects on broadacre TFP of changes to public funding of agricultural R&D 34

E1 Unit root tests for selected variables 46

E2 Co–integration tests (TS specification) 47

E3 Correlation Matrix: Domestic public R&D expenditures, foreign public R&D expenditures and domestic public extension expenditures 47

E4 Elasticities of broadacre TFP to public domestic and foreign R&D knowledge stocks and domestic extension knowledge stock 48

E5 Elasticities of broadacre TFP to public domestic R&D knowledge stocks and domestic extension knowledge stock 49


vi

E6 Elasticities of broadacre TFP to a public domestic R&D and extension knowledge stock (Mullen & Cox 1995 specification) 50

E7 Elasticities of broadacre TFP to a total public R&D knowledge stock and domestic extension knowledge stock 51

E8 Elasticities of broadacre TFP to a total public R&D knowledge stock and domestic extension knowledge stock: comparing the linear, log-linear and quadratic models 53

E9 Elasticities of broadacre TFP to a public domestic R&D and extension knowledge stock: comparing the linear, log-linear and quadratic models (Mullen & Cox 1995 specification) 54

E10 Elasticities of broadacre TFP to a total public R&D knowledge stock and domestic extension knowledge stock: ARIMA models for R&D distributed lag models with 35-year lags 55

E11 Elasticities of broadacre TFP to a public domestic R&D and extension knowledge stock: ARIMA models for R&D distributed lag models with 35-year lags (Mullen & Cox 1995 specification) 55

E12 Elasticities of broadacre TFP to a public domestic R&D and extension knowledge stock: Error Correction Model for R&D distributed lag models with 35-year lags (Mullen & Cox 1995 specification) 56

F1 Contributions of the public R&D and extension knowledge stocks to annual broadacre TFP growth: 1952–53 to 2006–07 57

1

Summary

Productivity growth in the broadacre industry has slowed recently, focusing attention on the instruments available to government and industry to enable increased productivity. Over time, public investment in R&D and extension has been a key ‘lever’ Australian governments have used to promote agricultural productivity growth. However, debate continues over the role governments should play in funding agricultural R&D and extension.

The objective of this report is to evaluate the economic effect of publicly funded agricultural R&D and extension by investigating the relationship between public investments in R&D and extension and broadacre total factor productivity (TFP) growth in Australia over the period 1952–53 to 2006–07. Productivity growth in the broadacre industry (essentially, non-irrigated crops, beef cattle and sheep) attributable to public R&D and extension was estimated using an autoregressive integrated moving average (ARIMA) model. Conceptually, broadacre TFP was modelled as a function of the stocks of usable knowledge available to farmers (as well as some other factors), where the stocks of knowledge reflect past public investments in R&D and extension accumulated over many years. The average effects of public investment in R&D and extension on broadacre TFP were also decomposed into their short-run (year-to-year) and long-run marginal effects using an error correction model.

The analysis of the average effect of the R&D and extension knowledge stocks on broadacre TFP between 1952–53 and 2006–07 demonstrated that public investment in R&D and extension has had a significant and positive effect on broadacre TFP. Past public investments in broadacre R&D and extension have generated average rates of return that could be as high as 28 per cent and 47 per cent a year, respectively. While little is known about the opportunity cost of public investment in agricultural R&D and extension, this rate of return is comparable to rates of return estimated for other developed countries (Alston et al. 2010a). Further, the growth in domestic public R&D and extension knowledge stocks arising from this investment has accounted for annual TFP growth in the broadacre industry of 0.33 per cent and 0.27 per cent, respectively (an aggregate of 0.60 percentage points a year).

An important aspect of this study was to control for the effect of foreign public R&D on Australian broadacre productivity growth. Technology and knowledge spillovers from foreign public R&D have enabled substantial productivity growth in the broadacre industry. Growth in foreign public R&D knowledge stocks accounted for an estimated 0.63 per cent TFP growth annually in the broadacre industry. The results suggest that the relative contributions of foreign and domestic research (including domestic extension) to broadacre TFP growth have been roughly equal (0.63 percentage points a year and 0.6 percentage points a year, respectively) and account for the bulk (1.23 percentage points a year) of average annual broadacre TFP growth (1.96 per cent a year).

The analysis of the dynamic relationship indicates that public R&D research strategies that invest over the long-term result in higher returns than research strategies that invest over the short-term. This finding can be used to inform the choice between short-term and long-term public investment in R&D at an aggregate level. Although slowing broadacre productivity


2

growth could prompt policy makers (government or industry) to consider a temporary increase in R&D funding, the results from the dynamic error correction model (ECM) analysis suggest that a short-term response is not efficient and will not yield the highest payoff. Notwithstanding the nature of ongoing debates concerning the extent to which government should fund R&D, the results from this study suggest that even plausible and immediate increases in public R&D are unlikely to improve broadacre productivity growth significantly for many years.

Finally, in comparing alternative strategies to increase productivity growth, it is important to consider the likely trade-offs between investing in R&D and extension. Increased investment in extension in the short run can enhance TFP growth by bringing forward the adoption of currently available technologies and knowledge. Although individual projects should be evaluated on their own merits, at an aggregate level, reallocating existing R&D funding toward extension is unlikely to maximise long-term productivity growth.

1

3

Increasing productivity continues to be an important policy objective of agricultural industries and Australian governments. Productivity growth is important for maintaining farm incomes against a persistent decline in their terms of trade (output prices relative to input prices) and improving the international competitiveness of domestic agricultural industries. Given increasing pressure on the natural resource base, long-term growth in production and the ongoing competitiveness of the sector will depend largely on increases in productivity.

In the long term, agricultural productivity growth largely reflects technical change, and this has been the experience of Australia’s broadacre cropping industry. Between 1977–78 and 2006–07, technical change was the key driver of productivity growth in the broadacre cropping industry (Hughes et al. 2011). With a decline in the rate of technical change since 2000, and slow growth in broadacre total factor productivity (TFP) since the mid 1990s (Sheng et al. 2011), there is renewed focus on the policy instruments available to government and industry to raise productivity growth rates. Recently, key stakeholders with an interest in the grains industry reaffirmed their commitment to achieving total factor productivity growth in the grains industry of greater than 2.5 per cent per a year within a decade (Primary Industries Standing Committee 2011).

Public investment in R&D and extension has been a key ‘lever’ that Australian governments have used to promote agricultural productivity growth. Public R&D has been an important means of developing new technologies and management practices. In turn, facilitating adoption of such innovations through extension has served to improve long-term agricultural productivity growth. However, notwithstanding Australian governments’ commitment to increasing productivity, debate continues over the role governments should play in funding agricultural R&D and extension (Productivity Commission 2011).

As a contribution to the debate, this research uses regression analysis (specifically, an autoregressive integrated moving average [ARIMA] model) to evaluate the economic impacts of public investment in agricultural research by estimating the productivity growth in the broadacre industry (essentially, non-irrigated crops, beef cattle and sheep) attributable to public R&D and extension over the past 50 years. Conceptually, broadacre TFP is specified as a function of public knowledge stocks available to farmers (as well as some other factors), where the stocks of knowledge are determined, in part, by current and past public R&D and extension expenditure (Griliches 1979).

Although public R&D and extension are expected to have a positive effect on broadacre productivity, the relative magnitude of their effects may differ, including over time (that is, in the short and long run). Recognising this, the average effects of public investment in R&D and extension on broadacre productivity are also decomposed into their short-run (year-to-year)

Introduction


4

and long-run marginal effects using an error correction model. While the results of the ARIMA model may be used to measure average returns, the results of the error correction model are useful for comparing the returns to short and long-term aggregate public R&D and extension investments.

Finally, a measure of the profitability of public investments in R&D and extension—the internal rates of return (IRRs)—are calculated. IRRs can be calculated from elasticities estimated using the ARIMA and error correction models, when the model used to convert past investments in R&D and extension into the corresponding knowledge stocks is known. Estimates of the IRR to public investment provide a measure of the return to public expenditure on agricultural R&D and extension (in terms of increased productivity), which may help to improve industry stakeholders’ understanding of the relative importance of R&D and extension.

Structure of this reportThe report is structured as follows. Chapter 2 contains descriptions of trends in public agricultural R&D and extension and broadacre productivity. The literature analysing the relationship between public investment in R&D and extension and agricultural productivity in Australia and overseas is briefly reviewed in chapter 3. The model specification and estimation strategy are outlined in chapter 4. Chapter 5 defines the variables and sources of data used in the analysis. Chapter 6 presents the results and provides estimates of the effects of public R&D and extension on broadacre productivity and the internal rates of return to public investment in agricultural research. Chapter 7 summarises the main findings, outlines policy implications and considers options for further research.

5

2

Observers understand that the bulk of public (and private) agricultural R&D and extension is aimed at improving agricultural productivity but measuring the relationship remains challenging. This chapter provides a background to recent trends in Australian public investment in agricultural R&D and extension and TFP growth in broadacre agriculture, which accounts for around 62 per cent of the gross value of agricultural production in Australia.

Public investment in agricultural R&D and extensionIn Australia, public investment in agricultural R&D has, in real terms, increased over the past 50 years, from $131 million in 1952–53 (2006–07 dollars) to around $778 million in 2006–07. In addition, the share of total agricultural R&D funded by the public sector has typically exceeded that of the private sector—generally more than 90 per cent, although by 2007 this had decreased to around 80 per cent (Mullen 2010).

Historically, Australian and state government research agencies have undertaken the biggest share of agricultural research; in 1995 they undertook 22 per cent and 53 per cent of agricultural R&D, respectively. However, by 2009 these shares had declined to 16 per cent and 50 per cent, respectively, with the share of publicly funded agricultural R&D provided by universities increasing from 14 per cent to 34 per cent (Australian Bureau of Statistics 2010).

While growth in public agricultural R&D expenditure was strong until the late 1970s, it has since slowed, with the average annual growth rate declining from around 7 per cent a year between 1952–53 and 1977–78 to around 0.6 per cent a year from 1977–78 to 2006–07 (figure 1). Moreover, research intensity (defined as the ratio of public R&D expenditure to agricultural gross domestic product [GDP]) peaked at over 5 per cent in 1977–78, before declining to 3 per cent in 2006–07. The policy rationale underlying these shifts in public agricultural research investment is explored in box 1.

Trends in agricultural R&D and extension and broadacre productivity

Real public agricultural R&D and extension investment in Australia: 1952–53 to 2006–07

1

public investment in agriculture R&D

rese

arch

inte

nsity

(%)

financial year ended

public investment in extension research intensity (right axis)

20051995198519751965195520001990198019701960

2007$m

200

400

600

800

1000

1200

0

1

2

3

4

5

6

Notes: Public agricultural R&D and extension includes expenditure by Australian, state and territory governments, and researchinstitutions and universities. Funds from research and developmentcorporations (excluding grower levies) and other external fundersfor agriculture (excluding research in fisheries and forestry) are alsoincluded.Source: Estimated with data from Mullen (2010)


6

continued...

The importance of public extension has also declined, from 24 per cent of total public agricultural R&D and extension in 1952–53 to around 19 per cent in 2006–07 (figure 2). This decline reflects the withdrawal of state and territory governments from providing extension services and has been accompanied by increasing private sector involvement.

The slowing growth of Australia’s public agricultural R&D and extension investment reflects similar slowdowns in other developed countries as well as globally. Between 1981 and 2000, global public investment in agricultural R&D increased at a rate of 2.1 per cent a year, from $15.2 billion in 1981 to around $23 billion in 2000 (in 2000 ‘international dollars’, calculated by converting local currency units to US dollar equivalents using purchasing power parities). However, global public investment in R&D

increased at an average annual rate of 1.2 per cent a year between 1991 and 2000, compared with 2.6 per cent between 1981 and 1990. Further, developed countries’ public investment in R&D fell by 0.58 per cent a year between 1991 and 2000, compared with an increase of 2.3 per cent a year between 1981 and 1990 (Alston and Pardey 2006).

Share of extension in total public agricultural R&D and extensionexpenditure: 1952–53 to 2006–07

2

perc

enta

ge o

f tot

al p

ublic

RD

&E e

xpen

ditu

re

20051995198519751965195520001990198019701960


0

5

10

15

20

25

30

Source: Estimated with data from Mullen (2010)

box 1 Shifting trends in public agricultural R&D policy in Australia

Broadly speaking, two distinct periods characterise public funding of agricultural R&D in Australia: strong growth relative to production coinciding with a marked expansionary phase in agriculture post-World War II; and slower growth from the late 1970s (figure 1).

A possible reason can be advanced to explain the strong growth. In the 1950s, the Commonwealth Government perceived a need to boost agricultural exports to deal with balance of payments problems—Australia had difficulty earning enough income abroad to finance its domestic growth (Campbell and Dumsday 1990). Increasing investment in R&D would be consistent with this objective given the prominence of agriculture in exports revenue—wool alone contributed almost two-thirds of Australia’s total export earnings in 1951.

By the late 1970s, however, growth in agricultural R&D investment in Australia had begun to wane. While the issues appear complex and interrelated, commentators postulated various reasons. From a political economy perspective, public support for agriculture may have fallen, resulting in a decline in spending on agriculture (including R&D) relative to total government spending. Jessup and Dun (1982) support this view and suggest that a perceived decline in the importance of the agricultural sector to the economy may have contributed to the shift in R&D focus. Jarrett (1990) provides further evidence, suggesting that CSIRO’s movement away from its focus on agricultural sector R&D may be part of a general change in the direction of government R&D policy in Australia towards non-agricultural sectors. CSIRO’s expenditure on production R&D in agriculture fell from 40 per cent of its budget in the 1950s to only 20 per cent in 1994 (Mullen et al. 1996).

7


Nevertheless, the Australian system contrasts strongly with those in other OECD countries. On average, more than half the total investment in agricultural R&D in OECD countries in 2000 came from the private sector, while the share in Australia was less than 20 per cent. Similarly, research intensity in Australia generally exceeded average research intensity in other developed countries (2.6 per cent of agricultural GDP; Mullen 2010). Thus, the effect on agricultural production of reduced investments in public R&D and extension could be stronger in Australia than in other OECD countries.

Trends in Australia’s broadacre total factor productivityA key objective of agricultural R&D and extension is to improve farm performance, particularly in relation to farm productivity. Productivity, as measured by TFP, reflects the efficiency with which producers combine all market inputs to produce market outputs.

Broadacre TFP in Australia trended up at an average annual rate of almost 2 per cent a year between 1952–53 and 2006–07 (figure 3). Over this period Australian broadacre farms increased production by 2.7 per cent a year on average, more than offsetting a 0.7 per cent increase in total input use (comprising land, labour, capital and intermediate inputs). The increase in total inputs was mainly attributable to increasing use of intermediate inputs (such as fertiliser, chemicals and contracted services) partly as substitutes for land, labour and capital, but also through a shift in activity mix from livestock to cropping.

Broadacre productivity growth has helped ease the effects on farmers’ incomes of a persistent decline in their terms of trade (figure 3). Productivity growth has also helped maintain Australia’s competitiveness in global commodity markets and build resilience within the farm sector to overcome the declining availability and increasing variability of some natural resources.

box 1 Shifting trends in public agricultural R&D policy in Australia ...continued

On the other hand, Alston et al. (1999) suggest that public sector support for agricultural R&D may have fallen for reasons specific to agricultural R&D, such as increased private sector roles. Over the period 1981 to 1993, agriculture’s share of Australia’s total public R&D budget experienced a marked decline (from 20 per cent to 15 per cent; Pardey et al. 1999). Jarrett (1990) links the changing allocation of scientific resources to attempts by policy-makers in the public sector to shift the responsibility for funding agricultural research into private hands. Jarrett (1990) further suggests the belief that the capital-intensive agricultural sector would not provide employment opportunities for the Australian economy may have contributed to the change in government attitudes toward agricultural R&D.

However, notwithstanding Australian governments’ recently-renewed aim of increasing agricultural productivity, and the centrality of R&D and extension to productivity growth, debate continues over the role governments should play in funding agricultural research and the returns to such public expenditure (Productivity Commission 2011).


8

However, recent evidence of a slowdown in broadacre productivity growth since the mid-1990s, particularly in the cropping industry, has concerned some stakeholders (figure 4). Broadacre TFP growth averaged around 2.2 per cent a year before 1993–84 (a turning point year determined by Sheng et al. 2011), but dropped to 0.4 per cent a year thereafter.

Evidence shows that slow growth in public investment in R&D since the late 1970s may have contributed to the slowdown in broadacre productivity growth (Sheng et al. 2011). Although a range of factors, including droughts, is likely to have contributed to the slowdown, that stagnating public investment in R&D should also be identified as a contributing factor is not surprising, given the predominant underlying objective of such investment. However, it is not clear to what extent public investment in R&D and extension has influenced broadacre productivity.

Broadacre TFP and the terms of trade, Australia, 1952–53 to 2006–07

3

20051995198519751965195520001990198019701960

terms of tradebroadacre TFP

index

50

100

150

200

250

300

350

400

Source: The terms of trade is the ratio of an index of prices received by farmers to an index of prices paid byfarmers for all agriculture (ABARE 2009). TFP is the ratioof a quantity index of aggregate output to a quantityindex of aggregate input (see Gray et al. 2011)

Broadacre TFP by farm type,1977–78 to 2007–084

mixed crop–livestockall broadacre

cropping specialists

index

50

100

150

200

250

200520001995199019851980

Source: Nossal & Sheng (2010)

9

3

There is an extensive body of literature evaluating the effect of public investments in agricultural R&D and extension on farm performance. This chapter reviews estimates of the returns to public investments in agricultural R&D and extension and key technicalities that can significantly affect empirical results, namely: the specification of R&D and extension variables (specifically, the assumed length and shape of the R&D lag profile) and the importance of technology spillovers for domestic productivity growth. The chapter concludes by drawing out the implications for this study.

Returns to public investments in agricultural R&D and extensionEconomic evaluations from overseas and Australia indicate the returns to public investments in agricultural R&D are high, with little evidence to suggest they are declining over time. Although the nature of the underlying causal factors is not well understood, a review of the empirical evidence supports this conclusion. For example, Echeverria (1990) reviewed over 200 projects and aggregate level analyses of agricultural R&D globally, and found that most estimated rates of return were greater than 20 per cent. Evenson (2001) confirmed this finding (as did Fuglie & Heisey (2007) for public agricultural research in the United States) and also found that the majority of estimated rates of return exceeded 20 per cent, although estimates were widely dispersed. Alston et al. (2000a) and Evenson (2001) report that overall, the median rate of return to research has exceeded 40 per cent (table 1).

Researchers have also reported high returns to public extension. Alston et al. (2000b) found an overall median rate of return to extension of 63 per cent (table 1), which was higher than the median return to research (48 per cent). This may be due to the more immediate effect of extension on productivity (Huffman and Evenson 2006).

In Australia, high internal rates of return (IRR) to public investments in R&D and extension have also been achieved. In their study of Australian broadacre agriculture, Mullen and Cox (1995) estimated returns to have averaged between 15 and 40 per cent between 1952–53 and 1987–88. Mullen (2007) also found that similar rates of return had persisted over the period 1952–53 to 2002–03.

Past research evaluating public R&D policies

1 Median and average rates of return to public investments in agricultural research and extension

Rate of return (%)

Study focus Median Average

Research only 48 100Extension only 63 85Research and extension 37 48All studies 44 81

Source: Data from Alston et al. (2000b)


10

Some commentators have questioned the credibility of these findings. For instance, several have pointed to the possibility of selection bias which could over-represent successful research programs that report high returns to public investment (Evenson 2001; Productivity Commission 2011). In addition, estimates are typically sensitive to model specification choices (discussed in the next section), including the specification of R&D and extension variables, the limited availability of time series R&D investment data, and the use of datasets covering different time periods (Makki et al. 1999a; Alston et al. 2010a; Productivity Commission 2011).

Nevertheless, the pool of evidence overwhelmingly points to relatively high returns to public investment in agricultural R&D and extension, at least well beyond a risk-free rate of return. Evaluations of aggregate level R&D and extension programs, which are unlikely to suffer from selection bias by reason of including both successful and unsuccessful programs, also report consistent and good returns to public investment. In this respect, Evenson (2001:613) argues that high returns are not inconsistent with actual productivity growth experiences.

Specification of R&D and extension variablesIntuitively, agricultural productivity in a given year does not depend on the current level of R&D and extension expenditure, but on the stock of usable knowledge derived, in part, from past investments (Alston and Pardey 2001). The standard method in econometric studies is to model productivity as a function of a knowledge stock constructed from distributed lags of past research investments in R&D and extension, to proxy the (unobserved) stock of knowledge available to farmers. Choices about the assumed length and distribution shape of the R&D lag profile have important implications for the estimation results, and errors in estimates of, or assumptions about, the lag profile can result in overstated or understated estimates of the returns to R&D (Alston et al. 2000a).

Length of R&D lagThe length of the assumed R&D lag profile can significantly affect estimated rates of return. There are often long lags before farmers begin accessing the outputs of R&D investments (and to a lesser extent, extension). As a result, underestimating the maximum number of years over which a given investment can affect productivity may overstate returns (Alston and Pardey 2001). For example, in their meta-analysis of studies that estimated the returns to agricultural research, Alston et al. (2000a) found those that included lags of 15 years or more for full realisation of the benefits from research investments had significantly lower rates of return.

Further, recent research has revealed that R&D lags are much longer than many researchers had realised (Alston and co-authors 2008; 2010a; 2010b). Early studies commonly specified short lags (less than 20 years), the assumed lag length being not much longer than available time series R&D investment datasets (Alston et al. 2009). However, Alston et al. (2009) note recent studies have tended to use longer lags, in line with expectations about the likely duration of the effects of R&D. For example, some researchers believe the lagged effects of Australian public investments in agricultural R&D could exceed 30 years (Mullen et al. 1996; Mullen and Strappazzon 1996; Cox et al. 1997).

11


Moreover, ignoring the possibility of a long-run relationship between productivity and R&D knowledge stocks may result in overstated rates of return. Makki et al. (1999a; 1999b) suggest that estimates of the long-run dynamic relationship between R&D knowledge stocks and productivity are improved by error correction modelling, as this technique brings together short and long-run information on how R&D knowledge stocks and other factors affect productivity growth.

Shape of lag profileEstimates of the returns to public investments in R&D and extension also depend on the time weights (or lag profile) used to aggregate past R&D and extension expenditures. The shape of the lag profile represents the dynamics of knowledge creation, use and depreciation, as the impact of past research on current productivity rises to a maximum before fading into unimportance (Griliches 1979; Alston et al. 2009). Since the return to an investment in subsequent years depends on these weights, lag profiles that assign larger weights to more recent years generate higher estimated returns.

Approaches to estimating the lag profile include estimating ‘free-form’ lags and the use of distributed lag models. A few studies have estimated time weights as ‘free-form’ estimates of individual lag coefficients, by including a number of lagged R&D and/or extension variables in the regression. However, this approach is not recommended as it requires too many coefficients to be estimated given available time series data (Huffman and Evenson 2006). Moreover, due to high multicollinearity between lagged research variables, coefficients tend to alternate between positive and negative values (Evenson 2001; Alston et al. 2009).

Instead, researchers recommend that beliefs about the shape of the lag profile may be imposed on R&D and extension data by assuming a specific functional form for the distributed lag model (for example, Griliches 1998). Although this approach can impose a strong structure on the lag profile, it also imposes smooth marginal effects of consecutive R&D and extension expenditures on productivity and reduces the demands on the data to identify coefficients (Evenson 2001; Huffman and Evenson 2006).

The literature commonly uses a range of functional forms and distributions for the lag profile including:

• the trapezoid distribution (Huffman and Evenson 1993; Mullen and Cox 1995)• the geometric distribution (reflecting the perpetual inventory method) (Shanks and

Zheng 2006)• the gamma distribution (Alston et al. 2010a).

The most appropriate distribution for the lag profile is not apparent, although the perpetual inventory method is considered inconsistent with the expectation that agricultural R&D investments have little effect in the early years because of long lags in adoption (Alston et al. 2010a). While Alston et al. (2010a) concluded that US agricultural R&D lags were best represented by a gamma distribution, it is difficult to discriminate econometrically between the trapezoid, gamma and other broadly similar distributions, absent consideration of wider aspects of the model specification (Alston et al. 2008).


12

Role of technology spilloversIn addition to indigenously generated knowledge, Australian agricultural productivity is likely to be influenced by spillovers of technology from other countries. R&D conducted overseas can be a source of spillover productivity gains, whether as ideas gained from the research of others or through foreign technology adapted to suit local conditions. The extent to which technology and knowledge ‘spill-ins’ influence agricultural productivity growth in Australia is not well understood, although Mullen (2010) speculates that it could account for productivity growth of up to 0.8 percentage points a year in broadacre agriculture, which has averaged around 2 per cent a year since 1953. A possible consequence of not taking into account spillovers from R&D in other states, countries and the private sector is that rates of return to domestic public R&D may be overstated (see Alston and Pardey 2001; Alston et al. 2009).

Foreign R&D may be especially important for small, open economies such as Australia’s. Openness to trade and investment can increase the transfer of knowledge and technology between countries and, in effect, facilitate access to the outputs of foreign R&D (Gutierrez and Gutierrez 2003; Hall and Scobie 2006). Overseas studies that have considered foreign spillovers have found that foreign R&D is as important—if not more so—as domestic R&D (Alston 2002; Alston et al. 2010a). For example, in a study of economic growth in G7 and 15 smaller countries, Coe and Helpman (1995) found that a country’s productivity growth also depended on the R&D knowledge stocks of trading partners, and that this effect was strongest for smaller and more open economies. Some researchers have suggested that existence of foreign knowledge may be a necessary condition for achieving productivity growth in such economies (Hall and Scobie 2006).

In general, spillover productivity gains from external R&D are greater when the technology or knowledge is sourced from regions (or countries) that have similar agroecological conditions, as less investment in adaptive research is needed (Sunding and Zilberman 2001; Gutierrez and Gutierrez 2003). Some studies have allowed for spillovers from states in the same geo-climatic region, on the basis that states in close proximity have similar agroclimatic and geoclimatic conditions that make them respond similarly to new technologies (Huffman and Evenson 2006). However, Alston and Pardey (2001) have argued that this is defining spillovers according to geopolitical boundaries and geographic proximity, rather than by agroecological similarity. Instead, Alston (2002) and Alston et al. (2010a) recommended a measure of the similarity of agricultural technology between states, defined by their agricultural output mix.

13


Implications for this studyThe modelling strategy followed in this analysis was informed by key findings from the literature on the returns to public investment in agricultural R&D and extension. Following Alston et al. (2010a), this analysis adopts a comprehensive estimation strategy to investigate the returns to public investment in Australian broadacre R&D and extension. This includes evaluating alternative model specification choices, including:

• the length (16 or 35 years) and shape (the gamma, trapezoid and geometric distributions) of the research lag profile

• the functional form of the model relating broadacre total factor productivity and R&D and extension knowledge stocks.

An important advance on earlier studies of the returns to Australian agricultural research is to account for broadacre productivity gains arising from technology spill-ins from overseas research and to distinguish between the relative contributions of foreign and domestic R&D and domestic extension to broadacre total factor productivity growth.

The efficiency of alternative estimation techniques is compared, taking into account the time series properties of the available data. Specifically, the efficiency of the ordinary least squares (OLS) and the autoregressive integrated moving average (ARIMA) model estimators are compared. Finally, an error correction model is estimated to describe the short-run dynamics and the long-run equilibrium relation between Australian broadacre total factor productivity and public investment in R&D and extension.

14

4

At any point in time, agricultural productivity does not depend on R&D and extension expenditure in that year, but rather on the stock of usable knowledge accumulated from past investments made over many years. Regression results (as discussed in chapter 3) are sensitive to the method used to construct knowledge stocks, as well as variable and model specification choices. In these matters, economic theory does not suggest an obvious estimation strategy, although past empirical studies can provide some guidance (for example, Mullen and Cox 1995; Alston et al. 2010a).

Base modelFollowing the literature (for example, Griliches 1979; Mullen and Cox 1995), Australian broadacre total factor productivity (TFP) is assumed to be a function of the current stock of knowledge available to farmers, as well as other control variables not explicitly reflected in the TFP index. The stock of knowledge is determined, in part, by past R&D and extension expenditure. The process for estimating knowledge stock variables is described in box 2.

An unconstrained base model representing the relationship between TFP and the R&D and extension knowledge stocks and other variables is given by:

!"#! = ! !"!,!"!,!"#!,!"!;!! + !! (1)

where !"#! = ! !"!,!"!,!"#!,!"!;!! + !! is the broadacre TFP index in time t and !"#! = ! !"!,!"!,!"#!,!"!;!! + !! , !"#! = ! !"!,!"!,!"#!,!"!;!! + !! , !"#! = ! !"!,!"!,!"#!,!"!;!! + !! and !"#! = ! !"!,!"!,!"#!,!"!;!! + !! are knowledge stocks in time t, pertaining to past expenditures on domestic public R&D, domestic private R&D, domestic extension and foreign public and private R&D, respectively. A specific functional form is denoted by f(.) and !"#! = ! !"!,!"!,!"#!,!"!;!! + !! is an error term.

The model includes a vector of other variables cited in previous studies (Zt ) namely, soil moisture availability, the terms of trade, and farmers’ level of education. The rationale for including these is as follows:

• Soil moisture availability can substantially depress TFP estimates in drought years because the broadacre industries (grain, beef and sheep production) are predominately dryland (non-irrigated) enterprises. Soil moisture availability is therefore an important factor contributing to volatility in the broadacre TFP index.

• Human capital formation is a driver of agricultural productivity growth which, beyond experience, is partly reflected by the education level of farmers. Were labour inputs differentiated according to education and weighted by prices indicative of labour quality, improvements in human capital would be effectively embodied in the labour input.

Methodology

15


Methodologybox 2 Construction of R&D and extension knowledge stocks

Lags between investments in R&D and productivity reflect the dynamics of knowledge creation, use and depreciation. At the project level, the distribution of benefits over time will differ. For example, although there may be long lags before basic research into crop or livestock genetics generates commercial technologies, the benefits may also persist (such as the development of dwarf varieties in the grains industry). In contrast, some new crop varieties may be short-lived as new pest and disease pressures emerge.

At the aggregate level, the relationship between annual R&D expenditure and the net knowledge stock formation (which is equal to the gross knowledge stock formation minus the depreciation of the knowledge stock) for each year after the investment can be approximated by a single distribution representing the average benefits for all types of agricultural R&D. Two key dimensions characterise the distribution of benefits: the shape of the lag profile and the maximum time lag (figure A1 in appendix A). Aggregating all net knowledge stock formations due to annual R&D investment in the past generates the knowledge stocks. The general format of such models is given in appendix A (equation A-1).

The choice of models used to construct the knowledge stock variables was informed by the findings of previous international and Australian studies (for example, Mullen and Cox 1995; Alston et al. 2000b; Alston et al. 2010a) and the authors’ econometric experimentation with the vector autoregression model to identify the shape of the preferred lag profile. The authors selected a small group of models that had sound econometric properties, based on a series of econometric tests, including the Ramsey RESET test and the root mean square error (RMSE) test.

Following Mullen and Cox (1995), for simplicity this study only considered R&D lags of up to 16 and 35 years in constructing the R&D knowledge stocks. To describe the shape of the lag profile, three functional forms were used: the gamma, trapezoid and geometric distributions. These distributions have been widely used in the literature: the gamma distribution in US studies (for example, Alston et al. 2010a), and the trapezoid distribution in Australian and US studies (for example, Mullen and Cox 1995; Huffman and Evenson 2006). The geometric distribution was included because it reflects the perpetual inventory method (PIM) approach, which has been used to construct knowledge stocks for the Australian market sector (for example, Shanks and Zheng 2006). However, the geometric distribution should be viewed with caution as the PIM approach is inconsistent with the expectation that agricultural R&D investment will have little effect in its early years because of long lags in adoption (Alston et al. 2010a).

In total, R&D knowledge stocks constructed using 10 different distribution functions were compared: a gamma distribution with the peak impact occurring after seven years, a trapezoid and geometric (PIM) distribution, and two gamma distributions that mimic the trapezoid (Gamma_T) and geometric (Gamma_P) distributions, for both 16-year and 35-year lags (appendix A). Two other assumptions defined the knowledge stock variables. First, the flow of services from the domestic and foreign knowledge stocks and the domestic extension knowledge stock were assumed to be constant over time. Second, the efficiency with which R&D and extension expenditures contribute to their respective knowledge stocks was assumed to be constant over time.

The domestic extension knowledge stock was derived as a weighted sum of extension expenditures over three years, with weights of 0.5, 0.25 and 0.25 (Huffman and Evenson 2006). This is because extension was expected to have a much quicker, but still lagged, effect on productivity, in contrast to the relatively long R&D lag profiles.


16

However, ABARES survey data differentiates labour only according to whether it is hired labour, or owner–operator and family members (with services provided by shearers for wool production specifically separated). Therefore, the effect of improved human capital on broadacre productivity will not be captured by the TFP index, but will be reflected in the TFP model estimates.

• Changes in the terms of trade may, in the short run, induce farmers when profit-maximising to choose combinations of inputs and outputs that reduce their overall productivity (Productivity Commission 2008; O’Donnell 2010). For example, farmers may expand cropping into relatively marginal land in response to increases in expected output prices.

Other factors that could influence agricultural productivity are not included in the vector of control variables. For example, the agriculture sector has experienced spillover productivity gains from government investment in transportation and communication infrastructure (Parham 2004). Changes in the structure of the farm sector are likely to be sources of productivity growth as well. To the extent these variables are not correlated with the knowledge stock variables, excluding them is unlikely to result in biased estimators for the total public R&D and extension knowledge stocks.

As a first step to estimate the relationship between public investment in R&D and extension and agricultural productivity, it is necessary to select an appropriate functional form for equation (1). For simplicity, previous studies have generally specified f(.) in equation (1) as a linear or log-linear function. The major difference between the two specifications is that the coefficients in the log linear function should be interpreted as elasticities, whereas the coefficients in the linear specification are multipliers of broadacre TFP with respect to the knowledge stocks. Mullen (2007) also found that including a quadratic knowledge stock term, which allows the elasticity of broadacre TFP to the R&D and extension knowledge stock to change over time, added to the explanatory power of the model.

In this study, linear, log-linear and quadratic specifications of equation (1) were all considered initially. However, as the log-linear specification was later preferred to the linear and quadratic specifications due to a series of statistical tests, the model (and results) are discussed for the log-linear specification. More detailed analysis of the choice among different model specifications of equation (1) is given in appendix E.

Therefore, equation (1) is restated in log form, where the superscripts k and j denote the lag length and distribution (or shape) of the research lag profile of the R&D knowledge stocks (box 1):

ln !"#! = ! + !!ln !"!!" + !!ln !"!

!" + !!ln !"#! + !!ln !"!!" + !ln !! + !! (2)

Due to multicollinearity owing to the high correlation between the R&D knowledge stock variables, it is not possible to estimate equation (2) directly. Past approaches have usually excluded the foreign R&D knowledge stock variable and combine the domestic public R&D and extension knowledge stocks. However, excluding knowledge stock variables that are correlated with other explanatory variables in equation (2) may result in endogeneity and bias

.

17


the estimators for domestic public R&D and extension knowledge stocks (b1 and b3 ). In this report, two constraints were imposed on equation (2) to deal with these identification issues.

First, following Mullen and Cox (1995), private R&D knowledge stocks ln !"#! = ! + !!ln !"!!" + !!ln !"!

!" + !!ln !"#! + !!ln !"!!" + !ln !! + !! were excluded

from equation (2) due to data inadequacy. Not including private R&D (domestic and foreign) may result in biased estimates of the coefficients of public knowledge stock variables if private and public knowledge stocks are correlated. For example, Alston and Pardey (2001) argued that, should private R&D be positively correlated with public R&D, its omission would bias upward the estimates of the coefficient on public R&D. However, sufficiently long time series data on private R&D expenditure in Australian agriculture are not available, which makes it difficult to include private R&D knowledge stocks ln !"#! = ! + !!ln !"!

!" + !!ln !"!!" + !!ln !"#! + !!ln !"!

!" + !ln !! + !! in the regression.

Although omitting private R&D knowledge stocks could potentially bias the results, the effects are likely to be minor for Australian agriculture. To the extent that farmers pay for private sector research (such as advances in the design of farm machinery or the efficacy of a new herbicide), such payments will be captured as inputs in the TFP index and the attendant benefits reflected in outputs. In other words, the productivity gains from an increase in output would be at least partially offset by the measured increase in higher quality inputs. Some of these effects are accounted for in the broadacre TFP index, as ABARES (partially) adjusts inputs and outputs for quality (Zhao et al. 2010). In contrast, the productivity benefits generated by publicly funded R&D are likely to be reflected in the estimates of TFP growth insofar as they are largely derived from adopting disembodied, unpriced management practices and technologies.

Further, in Australia, the private share of agricultural R&D has been small relative to public investment, exceeding 10 per cent only in recent years. Given the long lags between research investments and their contributions to the stock of knowledge, it is likely that the effect of domestic private R&D on broadacre TFP has been small. Nevertheless, excluding foreign private R&D remains a possible source of bias of unknown importance and an area for future research.

Second, rather than estimate the individual effects of domestic and foreign public knowledge stocks (as in equation (2)), it was necessary to construct a total public research knowledge stock variable to avoid their high correlation (table E-3, appendix E). Foreign public R&D is expected to contribute directly or indirectly to productivity growth in Australia through cross-country technology spillovers, including:

• direct adoption of advanced farming techniques and management skills developed overseas

• learning by doing through international cooperation• import of foreign machinery and materials• receipt of foreign direct investment.

Since not controlling for the effect of foreign public knowledge stocks may result in omitted variable bias (most likely overestimating the domestic contribution if domestic and foreign public knowledge stocks are positively correlated), a total public R&D knowledge stock variable was preferred.


18

Two assumptions guided construction of the total public R&D knowledge stock variable: a) domestic and foreign public R&D knowledge stocks were assumed to have the same lag profiles and b) the foreign public R&D knowledge stock was assumed to have a smaller effect on broadacre TFP than the domestic public R&D knowledge stock. In the latter case, factors such as differences in agricultural production techniques and possible trade and non-trade barriers to agricultural knowledge transfers may reduce the efficiency with which technology is transferred between countries.

The approach to selecting the weight on foreign public R&D knowledge stocks (π) was similar to that used by Alston et al. (2010a), which was based on the degree of similarity in production patterns in the United States and Australia. A measure of Australia’s openness to trade, which is widely believed to be positively associated with foreign technology spill-ins, was also taken into account. The weight on foreign public R&D knowledge stocks (π) was set to 0.1 (see appendix B for the assumptions informing this decision). This yielded the total public R&D knowledge stock variable, ln !"!

!" = ln !"!!" + ! ln !"!

!" , such that

ln !"!!" = ln !"!

!" + ! ln !"!!"

A range of alternative specifications of the relationship between broadacre TFP and the R&D and extension knowledge stocks was also considered, based on the specifications estimated by Mullen and Cox (1995). These specifications are outlined appendix C.

Time series analysis: An ARIMA modelGiven these specification choices, the model for examining the relationship between public R&D and extension knowledge stocks and broadacre TFP is:

ln !"#! = ! + !!ln !"!!" + !!ln !"#! + !!ln !"#! + !!ln !"#$ + !!ln !"!! + !!

Comparing equations (2) and (3), ln(Zt ) has been replaced by three specific variables, including:

• a measure of soil moisture availability (essentially a water stress index) ln !"#! = ! + !!ln !"!!" + !!ln !"#! + !!ln !"#! + !!ln !"#$ + !!ln !"!! + !!

• farmers’ education attainment as a proxy for the unobserved human capital of broadacre farmers !ln !"#! = ! + !!!!ln !"!

!" + !!!!ln !"#! + !!!!ln !"#! + !!!!ln !"#$! • the farmers’ terms of trade for Australian agriculture ln !"#! = ! + !!ln !"!

!" + !!ln !"#! + !!ln !"#! + !!ln !"#$ + !!ln !"!! + !! .

These variables are included because they may affect productivity, but are not reflected specifically in the TFP index (Mullen and Cox 1995). Not accounting for their effects may lead to biased estimation.

In previous studies, the ordinary least squares (OLS) technique was typically used to estimate equation (3). However, OLS may estimate a ‘spurious’ relationship between broadacre TFP and its determinants. If the time series data are non-stationary (positively correlated with time), coefficients estimated using OLS may be statistically significant even when there is no true

(3)

19


relationship between the dependent and explanatory variables (Granger and Newbold 1974). OLS estimates may only be unbiased when there is a co-integrating relationship between TFP and its determinants (that is, there is a linear combination of the variables such that the error term is stationary, even if some variables are not).

To avoid estimating a spurious relationship between TFP and the explanatory variables, the Dickey-Fuller Test and the Phillips-Perron Test were used to test all variables in equation (3) for stationarity. As these tests provided strong evidence that ln(TFPt ) and the (logged) knowledge stock variables were non-stationary, the Johansen Co-integration Test was used to determine whether a co-integrating relationship also existed.

Although OLS estimates are expected to be unbiased if broadacre TFP and the explanatory variables are co-integrated, they are not necessarily efficient (Greene 2007). Therefore, assuming the residuals (εt ) follow a normal distribution (based on the co-integration test results), a time series regression technique—the autoregressive integrated moving average (ARIMA) model—was used to examine the relationship between broadacre TFP and the R&D and extension knowledge stocks.

Long-term vs short-term effects: An error correction modelThe ARIMA model can be used to estimate the average effect of the R&D and extension knowledge stocks on broadacre TFP for a given period of time (as defined by the length of available time series data). However, the ARIMA model cannot describe the long-run properties and the short-run dynamics of the relationship between R&D and extension and broadacre TFP.

In the long run, the growth rate of broadacre TFP depends on the stable (or equilibrium) growth rate of R&D and extension knowledge stocks. In turn, this depends, in part, on the accumulated outputs of past research and extension. In contrast, in the short run, broadacre TFP is also influenced by adjustments in the levels of current knowledge stocks, after accounting for volatility largely due to seasonal conditions and price movements on world markets.

In the event that TFP and the R&D and extension knowledge stocks are co-integrated, an error correction model (ECM) is derived from the ARIMA model to bring together the short and long-run information on how the R&D and extension knowledge stocks (and the control variables) affect the dynamics of broadacre TFP growth. An advantage of the ECM is that it can be used to decompose the average impacts (specific to the time frame being analysed) of the R&D and extension knowledge stocks on TFP into their short and long-run components.

Assuming a co-integrating relationship exists between broadacre TFP and its determinants (subject to the Johansen Co-integration Test), residuals obtained from equation (3) would be identically distributed and independent of time (such that E(εt )=E(εt-1 )=0). Thus, since E(εt-1 )=0, lagged values of ln(!"#! = ! !"!,!"!,!"#!,!"!;!! + !! ) and its determinants can be added to equation (3) in equilibrium, making it possible to accommodate the dynamic response of broadacre TFP to its determinants. An ECM can be derived from equation (3), and re-arranged to give equation (4):


20

(4)

!ln !"#! = ! + !!!!ln !"!!" + !!!!ln !"#! + !!!!ln !"#! + !!!!ln !"#$!

+ !!!!ln !"!! − !!!!! + !!

where dln(∙), denotes the first logarithmic difference of broadacre TFP and the explanatory variables (dln(∙)=ln(∙)t – ln(∙)t-1 ), and the derivation of the error correction model from equation (3) is given in appendix D.

In equation (4), the change in broadacre TFP !ln !"#! = ! + !!!!ln !"!!" + !!!!ln !"#! + !!!!ln !"#! + !!!!ln !"#$!

depends on changes in the R&D and

extension knowledge stocks (!ln !"#! = ! + !!!!ln !"!!" + !!!!ln !"#! + !!!!ln !"#! + !!!!ln !"#$!

and !ln !"#! = ! + !!!!ln !"!

!" + !!!!ln !"#! + !!!!ln !"#! + !!!!ln !"#$!

) and the control variables, exogenous shocks (μt ), and the gap between and its determinants in the previous period, εt-1 , where:

(5)

The long-run information on broadacre TFP and its determinants is contained in the term κεt-1 in equation (4). κ is the adjustment coefficient, which links the short-run and long-run changes in broadacre TFP, following incremental changes in the knowledge stocks, control variables or other exogenous shocks to the relationship.

Assuming no change in the R&D and extension knowledge stocks and the control variables in period t for simplicity (dln(∙)=0), the adjustment coefficient is expected to be negative (κ > 0). This implies that if ln(TFPt ) is above its long-run equilibrium value (εt-1 > 0), the adjustment in broadacre TFP to restore equilibrium will be given by!ln !"#! = ! + !!!!ln !"!

!" + !!!!ln !"#! + !!!!ln !"#! + !!!!ln !"#$!

< 0. The adjustment coefficient, κ, also determines how quickly broadacre TFP returns to its equilibrium value following external shocks.

From equations (4) and (5), the long-run elasticities of broadacre TFP with respect to the R&D and extension knowledge stocks and the control variables are β1, β2 and γ1, γ2, γ3. These elasticities indicate how, in the long-run, broadacre TFP adjusts to permanent changes in the R&D and extension knowledge stocks and the control variables. The short-run elasticities are β1‘, β2‘ and γ1‘,γ2‘,γ3‘. These elasticities indicate the immediate effect on broadacre TFP of a temporary change in knowledge stocks and control variables.

The ECM provides a tool for analysing the dynamics of broadacre TFP growth and its determinants by distinguishing between the long-run and short-run elasticities of broadacre TFP to the R&D and extension knowledge stocks. The R&D and extension knowledge stocks are expected to have a positive effect on broadacre TFP. However, the magnitude of their marginal effects and their pattern over time may differ between the short and long run.

!!!! = ln !"#!!! − !!ln !"!!!!" − !!ln !"#!!! − !!ln !"#!!!

− !!ln !"#$!!! − !!ln !"!!!!

21

5

Broadacre total factor productivity

The measure of productivity used in the analyses is the ABARES broadacre total factor productivity (TFP) index (figure 3). ABARES broadacre TFP estimates are defined as a ratio of a Fisher quantity index of total output for the broadacre industry to a Fisher quantity index of total input. An exposition of the concepts, theories and empirical methods underlying the ABARES TFP estimates for the broadacre (and dairy) industries can be found in Gray et al. (2011) and Zhao et al. (2010).

All related data used for estimating the broadacre input and output indexes were collected through the ABARES broadacre farm surveys. ABARES has collected survey data since 1952–53. However, changes in 1977–78 to the data collected mean that comparable TFP estimates for the broadacre industry based on a consistent methodology are usually calculated only for the period 1977–78 to the present.

To calculate a longer but still consistent TFP series than usual, for the period between 1952–53 and 2006–07, outputs were directly aggregated into four time-consistent indexes of crops, livestock, wool and other outputs at the industry level (rather than first aggregating outputs into 19 specific output categories). Inputs were directly aggregated into four time-consistent indexes of land, capital, labour and materials and services at the industry level (rather than into 27 specific input categories). The four output indexes and four input indexes were then aggregated to give a total output index and a total input index, using the corresponding output and input price indexes as weights.

Public R&D and extension expendituresIn this study, domestic public R&D investment in agriculture was defined as Australian, state and territory governments’ expenditure on public R&D on plants and animals (excluding fish and forestry R&D) and expenditure on extension. Data were obtained from two sources:

• data for 1994–95 to 2006–07 were sourced from the Australian Bureau of Statistics biannual Australian Research and Innovation Survey (ABS 2008)

• data for 1952–53 to 1993–94 were drawn from Mullen et al. (1996), who sourced data from the Commonwealth Department of Science and the published financial statements of the state departments of agriculture and their counterparts.

Data sources and variable definitions


22

The absence of a long time series on R&D investments prior to 1952–53 also presented a challenge. Given previously identified research lags up to 35 years, public R&D expenditure data from 1917–18 were needed to make full use of the TFP dataset from 1952–53. To address this problem, public R&D expenditure data for the period 1917–18 to 1952–53 were generated by backwards extrapolation using a regression of log real R&D data against a time trend for the period between 1952–53 and 1971–72.

Data on extension were derived from state departments’ total expenditure records. As a breakdown of total expenditure on R&D and extension is generally unavailable, extension expenditure was estimated using past department surveys of time spent by staff on functions such as research, extension and regulation. The share of extension in total research expenditure for the period 1952–53 to 1993–94 ranged from 27 per cent (in 1964–65) to 39 per cent (in 1957–58), with no apparent trend (Mullen and Cox 1995). Data on investment in extension prior to 1952–53 were backcast with the proportion of public investment in extension assumed to be one-third of the state departments’ total investment in R&D and extension. The approach taken in constructing the extension dataset in this manner may affect the results, but the nature and extent are not clear.

Backcasting the R&D and extension investments prior to 1952–53 allows the full broadacre TFP series from 1952–53 to 2006–07 to be analysed. In addition, it reduces the possibility of an upward bias to the estimated rates of return which could arise from specifying insufficiently long research lags (Alston and Pardey 2001). Further, backcasting does not introduce additional information beyond that present in the R&D trend growth after 1952–53 and so is unlikely to unduly affect interpretation of possible causes of changes in TFP growth rates. Moreover, the backcast data are broadly consistent with total public investment in rural Australia, which provides an upper bound for public broadacre R&D investment up to the early 1950s (figure 5).

R&D and extension investment in the broadacre industry (as distinct from the agriculture sector as a whole) was derived by multiplying broadacre agriculture’s share of the total value of production in

agriculture to total public investment in agricultural R&D and extension. The GDP deflator was used to calculate real public R&D and extension expenditure.

Due to data constraints, US public expenditure on R&D related to agricultural production was used as a proxy for foreign public R&D. This treatment is justified because the US has a pivotal role in global agricultural R&D. This is not only in terms of the size of its investment, but also that it largely invests in areas that, from an Australian perspective, may be readily adapted or applied to Australian farming systems.

Real public investment in agricultural R&D (1952–53 to 2006–07) and rural Australia (1917–18 to 1952–53)

5

public investment in rural Australiapublic investment in agricultural R&D

2007$m

200

400

600

800

1000

20071987196719471927


23


Data for the period 1970 to 2007 were obtained from the Economic Research Service of the US Department of Agriculture. The pre-1970 data were aggregated state-level data from Huffman and Evenson (2006). To ensure consistency with Australian public R&D data, calendar year data were converted into financial year data by taking the average of expenditure in consecutive years.

Since the 1980s, developing countries’ share of global public agricultural R&D investment has increased dramatically, to 55.7 per cent of total public investment in 2000, predominantly in the newly industrialised China, India, Brazil, Thailand and South Africa (Pardey et al. 2006). However, it is reasonable to assume that developing country R&D is not a significant source of spillover technology, as Australia mainly imports agricultural technologies from developed countries. Consequently, excluding developing country spillovers is unlikely to bias the analysis.

Australian and US public research expenditures were aggregated to form the corresponding knowledge stocks, as described in chapter 3 (appendixes A and B contain detailed discussion). These are presented for R&D distributed lag models with 35-year lags in figures 6 and 7). Domestic extension knowledge stocks are presented in figure 8.

Domestic public R&D knowledge stocks with 35-year lags: 1952–53 to 2006–07

6

gamma PIM geometric

gamma traptrapeziod

gamma PIM

2007A$m

200

400

600

800

1000

1200

20052000

19951990

19851980

19751970

19651960

1955


US public R&D knowledge stockswith 35-year lags: 1952–53 to 2006–07

7

gamma

PIM geometricgamma traptrapeziod

gamma PIM

2007US$m

1000

2000

3000

4000

5000

20052000

19951990

19851980

19751970

19651960

1955


Domestic public extension knowledge stock: 1952–53 to 2006–07

8

2007A$m

20052000

19951990

19851980

19751970

19651960

1955

50

100

150

200



24

Control variablesSoil moisture availability (WSIt in equation (3)) was approximated by a water stress index for broadacre agriculture (figure 9). The water stress index—more precisely, the annual wheat water stress index (Potgieter et al. 2002)—is a measure of the relative water stress of the crop accumulated throughout the growing season. The index is a measure of grain growth, which is derived from daily rainfall and average weekly radiation data, maximum and minimum temperatures, location-specific soil data and crop-specific water requirements. The index reflects the cumulative stress endured by the crop throughout the growing season relative to its initial value of 100. The index was constructed from three data sources. The annual wheat water stress index at the national level between 1952–53 and 1987–88 was obtained directly from the Agricultural Production Systems Research Unit (APSRU). Due to data constraints, the national level water stress index was approximated by a weighted average of the annual wheat water stress index at the farm level for the period 1988–89 to 2003–04, and a weighted average of farm-level total rainfall for the period 2003–04 to 2006–07.

The proportion of school-age students in the total population enrolled in schools was used as a proxy for broadacre farmers’ education attainment (EDUCt in equation (3)) using ABS data (see Mullen and Cox 1995) (figure 10). Enrolment was defined as ‘school attendance’. The education index used here is a crude proxy for the real variable of interest, the human capital

stock of broadacre farmers. Since farmers’ education attainment is likely to differ from that of the total population, future research into the relationship between agricultural productivity growth and investment in R&D and extension would possibly benefit from development of a more appropriate measure (such as education levels of the rural population) for the human capital stock of farmers.

The farmers’ terms of trade (ln !"#! = ! + !!ln !"!!" + !!ln !"#! + !!ln !"#! + !!ln !"#$ + !!ln !"!! + !! in equation (3)) is defined as the ratio of the average price

farmers received for their outputs to the average price paid for farm inputs (figure 3). It covers all agriculture, not just broadacre (ABARE 2009).

Moisture availability index: 1952–53 to 2006–079

index

20052000

19951990

19851980

19751970

19651960

1955

50

100

150

200


Percentage of students aged between 4 and 19 enrolled in schools (5-year moving average): 1952–53 to 2006–07


10

%

20052000

19951990

19851980

19751970

19651960

1955

70

75

80

85

25

6

This chapter presents the results of the time series regression analysis, and provides estimates of the effect of public R&D and extension knowledge stocks on agricultural productivity and the internal rate of return to public investment in agricultural research in Australia.

Model selection testsTo ensure efficient estimates of the relationship between broadacre TFP and the R&D and extension knowledge stocks, a range of models that differed with regard to research lag profile, the treatment of knowledge stock variables, functional form, and estimation methods were compared according to a number of criteria.

As a first step, OLS estimates provide statistical criteria for selecting between alternative models (following Alston et al. 2010a). The Ramsey RESET test (Ramsey 1969) (which uses non-linear combinations of the fitted values of the dependent variable to test the null hypothesis that the model does not suffer from omitted variables bias) was used to determine whether combining domestic and foreign R&D to construct a total public research knowledge stock leads to biased estimators. Models with alternative functional forms and R&D lag profiles were compared on the basis of the smallest root mean square error (RMSE) statistics (the square root of the variance of the residuals of the regression).

On the basis of these statistical tests, it was possible to identify the preferred lag length, functional form and estimation strategy (see appendix E for more detail):

• combining domestic and foreign R&D to construct a total public research knowledge stock variable was preferred to the past practice of omitting foreign R&D (as in Mullen and Cox 1995)

• a 35-year lag period for capturing the effects of past R&D expenditures was preferred to a 16-year period (the models with 16-year lags did not pass the RMSE specification test and are not discussed further)

• the gamma distribution with peak impact occurring after seven years was preferred over alternative distributions (Trapezoid, Gamma_T, PIM and Gamma _P)

• the log-linear function form for equation (3) was preferred to linear and quadratic functional forms.

Although the OLS estimator is widely used in the literature, it may be biased if the broadacre TFP series and explanatory variables data series are non-stationary. To test for the presence of unit roots, a series of unit-root tests were carried out to examine each variable. On the basis of the Phillips-Perron and Dickey-Fuller tests, broadacre TFP, the total public R&D knowledge stock, and the terms of trade series are determined to be non-stationary and highly correlated with time (tables E1 and E2, appendix E).

Results


26

Because a co-integrating relationship exists between broadacre TFP and its determinants, an ECM can be used to describe the short-run dynamics as well as the long-run equilibrium relation between broadacre TFP and the public R&D and extension knowledge stocks.

In the following sections, the estimation results for the ARIMA model and the ECM are discussed for the preferred model specification identified above. The results for models with alternative distributions (Gamma_T, Gamma _P, Trapezoid and PIM) are also provided as robustness checks. The results for alternative model specifications are provided in appendix E (including models with alternative functional forms and treatment of knowledge stock variables based on the specifications used by Mullen and Cox (1995)).

Static effects of R&D and extension knowledge stocks on broadacre TFP The estimated elasticity of broadacre TFP with respect to public R&D knowledge stocks was positive and significant for all distribution profiles (table 2). In the preferred gamma specification, the coefficient on public R&D knowledge stocks was 0.23, implying that a 1 per cent increase in the public R&D knowledge stock is likely to lead to a 0.23 per cent increase in broadacre productivity, on average, all other things being equal.

2 Elasticities of broadacre TFP to public R&D and extension knowledge stocks and other explanatory factors

Gamma Gamma_T Trapezoid Gamma_P PIM

ln(TSt ) 0.234*** 0.225*** 0.202*** 0.242*** 0.201*** (0.051) (0.053) (0.046) (0.057) (0.048)ln(EXTt ) 0.100*** 0.104*** 0.143*** 0.072** 0.146*** (0.035) (0.034) (0.033) (0.036) (0.034)ln(WSIt ) 0.275*** 0.270*** 0.276*** 0.264*** 0.275*** (0.054) (0.055) (0.055) (0.055) (0.056)ln(EDUCt ) 0.562 0.662* 0.019 0.853** 0.293 (0.368) (0.379) (0.398) (0.386) (0.385)ln(TOTt ) −0.266*** −0.240** −0.262*** −0.255*** −0.261*** (0.081) (0.096) (0.084) (0.095) (0.088)Constant −0.325 −0.853 1.843 −1.372 0.614 (1.693) (1.817) (1.570) (1.906) (1.646)/sigma 0.063*** 0.065*** 0.063*** 0.065*** 0.064*** (0.007) (0.007) (0.007) (0.007) (0.007)Number of observations 55 55 55 55 55

Notes: ARIMA model of equation (3), broadacre TFP as a function of a total public R&D knowledge stock, a domestic extension knowledge stock, and the control variables. Statistical significance at the 1 per cent, 5 per cent and 10 per cent levels are represented by ***, ** and *. The values in parentheses are standard errors. ‘Gamma’ refers to the preferred model specification in which a gamma distribution was used to construct knowledge stocks, with a peak impact occurring seven years after investment.

27


Similarly, the results suggest that increases in public extension knowledge stocks have had a significant and positive effect on productivity, with an elasticity of around 0.1 per cent. The effect of the extension knowledge stock on TFP was, on average, around half that of the public R&D knowledge stock, where R&D and extension knowledge stocks both increased at the same rate.

Of the three control variables, soil moisture availability and the farmers’ terms of trade had significant effects on broadacre TFP. The estimated elasticities of TFP with respect to soil moisture availability was 0.28, indicating that a 1 per cent increase in soil moisture over the growing season would be expected to increase productivity by, on average, 0.28 per cent, all other things being constant.

In contrast, the farmers’ terms of trade had a negative effect on broadacre TFP. The elasticity of TFP with respect to the terms of trade was −0.27 in the preferred gamma distribution, indicating that a 1 per cent improvement in the farmers’ terms of trade would, on average, lead to a 0.27 per cent fall in broadacre productivity, all other things being constant. As indicated, a possible explanation is that improvements in the terms of trade may induce farmers, when profit-maximising, to choose combinations of inputs and outputs that, in the short term, reduce their overall productivity.

The elasticity of TFP with respect to the level of education attainment was positive but insignificant. To some extent this is unexpected since human capital can facilitate technology adoption and improve farmers’ ability to organise and maintain complex production processes. As suggested, the national education attainment index used in the analysis may not be a good proxy for the human capital stock of broadacre farmers.

The relative contributions of R&D and extension knowledge stocks to broadacre TFP growth over the past five decades can be estimated using the elasticities of broadacre TFP to its determinants (the steps followed are outlined in appendix F). For example, the contribution of total public R&D can be calculated by multiplying the broadacre TFP elasticity (from table 2) by the annual growth rate of the corresponding total public R&D knowledge stocks. This provides an estimate of the annual percentage broadacre TFP growth attributable to growth in the total public R&D and extension knowledge stocks between 1952–53 and 2006–07.

3 Contribution of R&D and extension knowledge stocks to broadacre TFP growth: 1952–53 to 2006–07

Average annual Growth rate (%) Elasticity a TFP growth (%)

Total R&D knowledge stocks (TSt ) 4.13 0.234 0.97Domestic R&D knowledge stocks (DSt ) 7.71 – 0.33Foreign R&D knowledge stocks (FSt ) 2.29 – 0.63Extension (EXTt ) 2.65 0.100 0.27Other factors b – – 0.73Broadacre TFP Growth (TFPt ) – – 1.96

a From table 2. b Includes the control variables and other unobserved factors not in equation (3). Notes: Contribution of R&D and extension knowledge stocks to broadacre TFP growth calculated for the preferred gamma R&D distributed lag model with 35-year lags and a peak impact occurring seven years after investment, estimated using the ARIMA model.


28

Between 1952–53 and 2006–07, growth in public R&D and extension knowledge stocks accounted for more than half the annual TFP growth in the broadacre industry. Broadacre TFP growth averaged around 1.96 per cent a year between 1952–53 and 2006–07. Over this period, growth in public R&D knowledge stocks accounted for approximately half the broadacre TFP growth (around 0.96 percentage points a year). This comprised 0.33 percentage points a year from the accumulation of domestic public R&D knowledge stocks and 0.63 percentage points a year from the accumulation of foreign public R&D knowledge stocks. Growth in public extension knowledge stocks contributed around 0.27 percentage points a year to TFP growth. This suggests that, between 1952–53 and 2006–07, the relative contribution to broadacre TFP growth of domestic and foreign R&D and domestic extension was in the ratio of 1:2:1 (figure 11).

Dynamic effects of R&D and extension knowledge stocks on broadacre TFPThe ECM simultaneously provides estimates of the long-run relationship between broadacre TFP and its determinants and the short-run response of broadacre TFP to incremental changes in the explanatory variables. Therefore, the model can be used to decompose the ARIMA estimates (the average elasticity of broadacre TFP to its determinants) into their long-run and short-run components, also known as marginal elasticities. The results of the ECM are reported in table 4. The results of the ECM estimated for broadacre TFP modelled as a function of a single domestic R&D and extension knowledge stock and the control variables (Mullen and Cox 1995), are given in table E-12, appendix E.

The ECM results confirm the significant and positive long-run equilibrium relationship between broadacre TFP and the R&D and extension knowledge stocks. In the long run, a (permanent) 1 per cent increase in the total public R&D knowledge stock increases TFP in the broadacre industry by 0.25 per cent (for the preferred gamma specification), all other things being equal. The extension knowledge stock has a small (but significant) marginal effect on broadacre TFP in equilibrium, with a (permanent) 1 per cent increase in the extension knowledge stock increasing broadacre TFP by 0.09 per cent in the long run.

To the extent that public R&D is a source of the new technologies and best practice methods that have enabled farmers to increase productivity in the long term, these results are

domestic R&D0.33 percentage points a year (17%)

extension0.27 percentage

points a year(14%)

foreign R&D0.63 percentage points a year(32%)

Relative contributions of public R&D and extension to annual broadacre TFP growth (%)11

other factors0.73 percentage

points a year(37%)

total publicR&D

0.96 percentage points a year

(49%)

29


consistent with recent findings that technical change has been the key driver of long-term productivity growth in Australian broadacre cropping farms (Hughes et al. 2011).

These results also suggest that public investments in R&D that increase, as opposed to just maintain, the stock of knowledge available to farmers can significantly promote long-term productivity growth. Given the long lags between research investments and their contributions to the stock of knowledge, slow growth in public research in recent years may cause a significant drop in broadacre TFP growth in the future, to the extent that there is no growth or a reduction in the public R&D knowledge stock.

In contrast, an increase in the extension knowledge stock will have a more significant and positive marginal effect on broadacre TFP growth in the short run than an increase in the R&D knowledge stock. Estimates of the short-run elasticities of broadacre TFP to the total public R&D knowledge stock are not robust to the distribution used to construct the R&D knowledge stock and statistically insignificant at the 5 per cent level. However, a 1 per cent increase in the domestic extension knowledge stock (relative to its value in the previous period) increases broadacre TFP by 0.44 per cent in the short run, and the estimated marginal elasticities of broadacre TFP to extension knowledge stocks were robust across distributions, ranging from 0.44 to 0.57, and significant at the 5 per cent level.

The ECM results suggest that a temporary increase in the extension knowledge stock (through a one-off increase in investment) can significantly increase broadacre TFP, although this effect will be short lived. Public extension can increase the diffusion of technology and best practice within the broadacre industry, increasing technical efficiency in the industry. However, the marginal effect of public extension on broadacre TFP may decline over time as new technologies and best practice methods are made available to farmers (requiring new extension).

Compared with the effect of a temporary increase in public extension, a temporary increase in the R&D knowledge stock is unlikely to bring about a substantial increase in broadacre TFP growth. In practice, lags in adoption and implementation of new knowledge and technologies may defer application of new technology. This means the productivity effects of public investment that increases the R&D knowledge stock can only be realised in the long run rather than the short run.

As expected, soil moisture availability and the terms of trade significantly contribute to short-run growth in broadacre TFP. A (temporary) 1 per cent increase in soil moisture (a measure of the relative water stress of the crop compared with the previous growing season) increases TFP in that year by 0.28 per cent. A (temporary) 1 per cent improvement in the terms of trade leads to a 0.21 per cent fall in broadacre TFP. Again, this is consistent with the expectation that favourable terms of trade may induce farmers to choose combinations of inputs and outputs that reduce their overall productivity (but increase profitability) in the short run (Productivity Commission 2008; O’Donnell 2010).

A persistent decline in the terms of trade was evident over the period 1952–53 to 2006–07, although the rate of decline has flattened over the past two decades (figure 3). In the short run, a 1 per cent decline in the terms of trade leads to a 0.26 per cent increase in broadacre productivity. In the long run, the decline in the farmers’ terms of trade may provide an incentive to substitute more efficient technologies to improve productivity, to maintain competitiveness and, in turn, viability.


30

4 El

astic

ities

of b

road

acre

TFP

to p

ublic

R&

D a

nd e

xten

sion

kno

wle

dge

stoc

ks a

nd o

ther

exp

lana

tory

fact

ors

G

amm

a

Gam

ma_

T

Trap

ezoi

d

Gam

ma_

P

PIM

LR

SR

LR

SR

LR

SR

LR

SR

LR

SR

ln(T

S t )

0.25

0***

1.

114*

0.

231*

**

−0.

044

0.25

0***

0.

019

0.21

6***

0.

736

0.21

5***

1.

916*

(0

.053

) (0

.675

) (0

.054

) (0

.724

) (0

.059

) (0

.673

) (0

.048

) (0

.714

) (0

.050

) (0

.990

)ln

(EXT

t )

0.08

7**

0.43

7**

0.09

5***

0.

476*

* 0.

062*

0.

479*

* 0.

133*

**

0.47

0**

0.13

7***

0.

569*

*

(0.0

34)

(0.1

89)

(0.0

34)

(0.1

95)

(0.0

36)

(0.2

08)

(0.0

32)

(0.2

21)

(0.0

32)

(0.2

27)

ln(W

SIt )

0.

259*

**

0.27

7***

0.

257*

**

0.27

8***

0.

251*

**

0.27

5***

0.

259*

**

0.27

9***

0.

258*

**

0.27

5***

(0

.055

) (0

.027

) (0

.057

) (0

.028

) (0

.057

) (0

.028

) (0

.057

) (0

.027

) (0

.058

) (0

.027

)ln

(ED

UC

t )

0.63

5*

−0.

425

0.72

8*

−0.

728

0.92

8**

−0.

632

0.05

6 −

1.00

8 0.

345

−0.

821

(0

.364

) (0

.866

) (0

.383

) (0

.971

) (0

.390

) (0

.970

) (0

.388

) (1

.020

) (0

.378

) (1

.057

)ln

(TO

T t )

−0.

259*

**

−0.

206*

* −

0.24

2**

−0.

204*

* −

0.25

6***

−

0.20

7**

−0.

255*

**

−0.

211*

* −

0.25

1***

−

0.20

2**

(0

.081

) (0

.081

) (0

.095

) (0

.088

) (0

.094

) (0

.085

) (0

.084

) (0

.084

) (0

.088

) (0

.081

)RE

SD

– −

0.99

1***

–

−0.

944*

**

– −

0.94

1***

–

−0.

957*

**

– −

0.95

8***

–

(0.1

37)

– (0

.137

) –

(0.1

39)

– (0

.138

) –

(0.1

41)

Con

stan

t −

0.58

8 −

0.04

0*

−1.

016

0.00

2 −

1.57

9 −

0.00

1 1.

729

−0.

032

−1.

533

0.00

1

(1.6

83)

(0.0

22)

(1.8

27)

(0.0

30)

(1.9

17)

(0.0

26)

(1.5

36)

(0.0

40)

(1.8

51)

(0.0

35)

/sig

ma

0.06

2***

0.

059*

**

0.06

4***

0.

061*

**

0.06

4***

0.

061*

**

0.06

2***

0.

061*

**

0.06

2***

0.

061*

**

(0.0

08)

(0.0

07)

(0.0

07)

(0.0

07)

(0.0

07)

(0.0

06)

(0.0

08)

(0.0

07)

(0.0

08)

(0.0

07)

No.

of o

bse

rvat

ions

54

54

54

54

54

54

54

54

54

54

Not

es: E

rror

cor

rect

ion

mod

el (e

quat

ion

4), b

road

acre

TFP

as

a fu

nctio

n of

a to

tal p

ublic

R&

D k

now

led

ge

stoc

k, a

dom

estic

ext

ensi

on k

now

led

ge

stoc

k, a

nd th

e co

ntro

l var

iab

les.

Sta

tistic

al

sign

ifica

nce

at th

e 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

. ‘G

amm

a’ re

fers

to th

e p

refe

rred

mod

el

spec

ifica

tion

in w

hich

a g

amm

a di

strib

utio

n w

as u

sed

to c

onst

ruct

kno

wle

dg

e st

ocks

, with

a p

eak

imp

act o

ccur

ring

seve

n ye

ars

afte

r inv

estm

ent.

31


Return to public investment in agricultural R&D and extension: a cost–benefit frameworkThe above analysis provides evidence of the significant productivity effect of public investment in agricultural R&D and extension on broadacre agriculture. However, it is of further interest from a policy perspective to weigh these payoffs and investment costs within a cost–benefit framework.

Internal rates of return (IRRs) to public investment can be calculated from estimated elasticities when the model used to convert past investments in R&D and extension into the corresponding knowledge stocks is known (the steps followed in calculating the IRRs are outlined in appendix G). Estimates of the IRR to public investment provide a measure of the return to public expenditure on agricultural R&D and extension (in terms of increased productivity), which can be used to evaluate past and inform future investments.

Using the elasticities estimated in the ARIMA model, over the period 1952–53 to 2006–07, the average IRR to public investment in broadacre R&D was 28.4 per cent a year in the preferred model, but ranged from 15.4 to 38.2 per cent in the other specifications (table 5). To some extent the variation in IRRs across the distributions are due to the different weights assigned to the lagged years, since the estimated elasticities are quite similar in magnitude. Generally, distributions that assigned greater weights to more recent years generated higher IRRs (figure A-1, appendix A).

Public extension generated significantly higher average IRRs than those for public R&D. Over the period 1952–53 to 2006–07, the IRR estimated from the preferred gamma specification for public extension was 47.5 per cent, ranging from 32.6 per cent to 57.1 per cent. These are relatively consistent with the median rates of return in the international literature reported in Alston et al. (2000b) (table 1). This suggests that more public investment in extension is more profitable than public investment in R&D. However, this may not be the case since the higher IRR to public extension may be due to the more immediate effect of extension on productivity (Huffman and Evenson 2006).

Internal rates of return calculated with estimated elasticities from the ARIMA model are independent of how public funding of R&D and extension is invested over time (excluding the impact of the pre-determined distributed-lag model used to construct the knowledge stocks). However, short-term (one-off) investment strategies and long-term (smooth) investment strategies may generate different benefits, depending on whether they change the knowledge stocks permanently or temporarily.

5 Average IRR to domestic public investment in broadacre R&D and extension (%)


Public R&D 28.4 14.0 15.4 38.2 51.9Extension 47.5 35.0 32.6 57.1 79.5

Note: IRRs calculated for the R&D distributed lag models with 35-year lags, as estimated by the ARIMA model.


32

The elasticities estimated from the ECM model (table 6) can guide resource allocation, at an aggregate level, between public R&D and extension with short and long-term investment horizons. It is likely that short and long-term investment strategies will affect formation of research and extension knowledge stocks over time, and therefore productivity, differently. Thus, a return to alternative investment strategies can aid decisions on the mix of public R&D and extension expenditure and how to allocate such expenditure over time.

For this purpose, the marginal IRRs calculated with the short-run elasticities can be interpreted as a measure of the impact of a $1000 increase in public expenditure, spent as a single investment, which results in a temporary increase in the R&D or extension knowledge stock. IRRs imputed from the long-run elasticities can be interpreted as a measure of the impact of a $1000 increase in public expenditure, spread smoothly over a period of time, which results in a permanent increase in the R&D or extension knowledge stock.

Public R&D research strategies that invest over the long-term, with a view to permanently increasing the R&D knowledge stock, result in higher returns than research strategies that invest over the short-term. Estimates of the short-run marginal IRRs were not robust to the distribution used to construct the R&D knowledge stock and statistically insignificant at the 5 per cent level. On the other hand, investments that permanently increase the public R&D knowledge stock generate a marginal IRR of 32.7 per cent in the long run (ranging from 14.3 per cent to 54.5 per cent and robust to the distribution used to construct the R&D knowledge stock).

In contrast, short-term investments in public extension yield larger returns than investments over the long-term. Estimates of the short-run marginal IRRs to extension ranged from 170 per cent to 183 per cent. The long-run marginal IRRs to extension were also significant and positive, ranging from 18 to 56 per cent. The high rates of return to extension may be due to the short lags assumed in constructing the extension knowledge stocks.

6 Marginal internal rate of return (MIRR) to domestic public investment in broadacre R&D and extension (%)

Overall Gamma Gamma_T Trapezoid Gamma_P PIM significancePublic R&D Long-run elasticity 0.25 0.231 0.25 0.216 0.215 YesLong-run MIRR (%) 29.7 14.3 15.8 40.5 54.5 YesShort-run elasticity 1.114 −0.044 0.019 0.736 1.916 NoShort-run MIRR (%) 68.5 –1.1 0.0 105.0 105.0 No

Extension Long-run elasticity 0.087 0.095 0.062 0.133 0.137 YesLong-run MIRR (%) 32.7 37.1 18.0 56.0 51.5 YesShort-run elasticity 0.437 0.476 0.479 0.47 0.569 YesShort-run MIRR (%) 170.0 182.0 183.0 180.0 180.0 Yes

Notes: IRRs calculated for the R&D distributed lag models with 35-year lags as estimated by the error correction model. IRRs calculated using the elasticity of broadacre TFP to a domestic public R&D and extension knowledge stock (Mullen and Cox 1995) are given in appendix G.

33

7

Policy implicationsExamination of the relationship between public R&D and extension and broadacre (total factor) productivity in this report offers several insights for public strategies for investing in agricultural research. The strong empirical relationship indicates that public investment in R&D and extension is an effective lever for government to promote agricultural productivity growth and, in turn, changes in R&D policies will significantly affect productivity growth rates in the long run.

ABARES notes that an outcome sought by the agriculture subgroup at the Australia 2020 Summit was for Australian agriculture to achieve an annual total factor productivity improvement of 3.5 per cent by 2020 (Commonwealth of Australia 2008). More recently, key stakeholders with an interest in the grains industry (including the GRDC, various state agriculture departments, the CSIRO and the Australian Council of Deans of Agriculture) collaborated to develop the Grains Industry National Research, Development and Extension Strategy and reaffirm their commitment to achieving TFP growth in the grains industry of greater than 2.5 per cent per a year (within a decade) while sustaining the resource base and improving market position as a core outcome of the Grains Industry National Research, Development and Extension Strategic Plan (Primary Industries Standing Committee 2011).

Achieving substantial productivity growth quickly to meet aspirational targets would likely involve a trade-off in realising higher productivity growth in the long run. Policy makers need to find the optimal balance in allocating scarce R&D and extension budgets, at an aggregate level, between R&D that generates payoffs over the long run and extension that can generate higher returns over the short run; and R&D with a stable, long-run focus versus finite funding directed at a short to medium term payoff.

Increased investment in public (and private) extension could help realise such targets by bringing forward adoption of existing knowledge. Of course, reallocating the current quantum of R&D and extension expenditure would have short-term payoffs, but a longer-term downside insofar as less knowledge will ultimately be created over a given time. However, if stakeholders are committed to achieving aspirational targets, budgetary (and efficiency) increases would presumably need to be considered in the range of policy responses. In considering opportunities for enhancing public extension initiatives, it is important for decision makers to consider the scope for emphasising extension initiatives directed at accelerating foreign public knowledge and technology spill-ins within the strategic policy mix—rather than limiting the concept of extension simply to indigenously generated knowledge. Such spill-ins have benefitted the broadacre industry to a considerable extent—around one-third (0.63 per cent a year) of average TFP growth (1.96 per cent a year) over the past 50 years or so.

Policy implications and further research


34

Although slowing broadacre productivity growth could prompt policy makers (government or industry) to consider a temporary increase in R&D funding, the results from the dynamic ECM analysis suggest that a short-term response is not efficient and will not yield the highest payoff. Among a range of factors, funding uncertainty and attendant short-term employment contracts are likely to constrain research efficiency, including through reduced job satisfaction among scientists (Productivity Commission 2007). They are also less likely to attract venturous proposals which may generate serendipitous, high-payoff outcomes. Such basic research is often undertaken on the understanding that there may well be no apparent impact from a short to medium term viewpoint (Productivity Commission 2007, p.321).

In addition, notwithstanding the nature of ongoing debates concerning the extent to which government should fund R&D and extension, even plausible and immediate increases in public R&D are unlikely to improve broadacre productivity growth significantly for many years. Lags associated with development and commercialisation of new technologies and management practices delay realising expected productivity benefits. For example, under a range of alternative funding scenarios directed at increasing public R&D, the productivity growth rates projected suggest little improvement until at least the late 2020s, all other things being equal (table 7).

7 Projected effects on broadacre TFP of changes to public funding of agricultural R&D relative to broadacre TFP growth rate or public agricultural R&D funding level in 2006–07

2006–072006–07 to

2019–202006–07 to

2049–502006–07 to

2049–50

Hypothetical public R&D funding scenario

Annual TFP growth (%)



Additional avg. public R&D investment

($m/year)

No change 1.96 2.02 2.03 85

Permanently increased by 10% in 2007–08 1.96 2.02 2.09 170

Permanently increased by 20% in 2007–08 1.96 2.02 2.16 520

Increased annually by 2% in real terms starting 2007–08 1.96 2.02 2.17 1470

Increased annually by 4% in real terms starting 2007–08 1.96 2.02 2.32 590

Maintained at 3.1% of GVAP to 2014–15 then no change to 2049–50 1.96 2.02 2.39 1120

Maintained at 3.1% of GVAP to 2014–15 then increased annual by 2% in real terms until 2049–50 1.96 2.02 2.43 85

Notes: Scenarios are hypothetical and have been designed to illustrate a range of possible effects. All values in real terms (2006–07 dollars). Simulations do not take into account a wide range of issues, including any potentially binding constraints on government expenditure to various industries under the Primary Industries and Energy Research and Development Act 1989.

35


Notwithstanding the significant contribution public R&D can make to productivity growth, many other factors influence productivity growth, including macroeconomic influences; institutional settings; and drivers of innovation such as sharpening price incentives, improving operating flexibility and building capabilities (Productivity Commission 2008). These factors, in conjunction with public R&D investments, will also influence the extent to which industry may realise target growth rates.

Related scenario modelling for the United States also highlights how past public R&D investment decisions strongly shape future productivity outcomes. Heisey et al. (2011) examined the potential effects of increasing public R&D investment on agricultural TFP in the United States for the period 2010–50 by considering three scenarios relative to the 2005–09 average:

1 public R&D spending is constant in nominal terms (that is, it declines at 3.73 per cent a year in real terms)

2 public research spending is constant in real terms

3 public R&D spending increases by 1 per cent a year in real terms.

In all three scenarios, the annual TFP growth in United States agriculture would decrease from present levels (around 1.3 per cent a year) and then level off after 2030, mainly due to the lagged effects of past declines in public R&D investment since the 1980s.

Beyond the period where past investment decisions ‘wash through’, the US results demonstrate that different scenarios may be expected to impose markedly different effects on agricultural productivity growth in the next half century. Under scenario 1, TFP growth declines to under 0.75 per cent a year, while under scenarios 2 and 3, TFP growth stabilises at about 1.4 per cent a year and 1.6 per cent a year respectively, slightly above the initial level (1.3 per cent a year for the past two decades).


36

Further researchDespite the efforts made by this research, significant opportunities remain to extend the analysis. In the discussion of the model specification and data, a number of limitations were raised that further research might usefully address. The limitations arise in part from data constraints, including the unavailability of domestic and foreign private research expenditures and the lack of a more appropriate proxy for the human capital of broadacre farmers. Education is, with other types of human capital, an important determinant of farmers’ ability to effectively adopt and integrate productivity-increasing innovations into existing farming systems (Nossal and Lim 2011). Given a better proxy, further research could be expected to identify that the human capital of farmers has made a significant contribution to broadacre productivity growth. Future extensions to this analysis would also benefit from a consistent dataset for soil moisture from which to estimate the water stress index, and a formal derivation of the value of the weight on spill-ins from foreign public R&D knowledge stocks.

Further research might also usefully explore the interaction between domestic and foreign R&D knowledge stocks and how the broadacre industry can increase spillover benefits from foreign R&D. This study identified that foreign and domestic public R&D have made a substantial contribution to Australian broadacre productivity growth, but does not provide any insights into the mechanisms, direct and indirect, through which foreign public R&D enables domestic productivity growth. Research into industry’s capacity to identify, adapt and exploit technologies developed outside Australia would further aid efficient design of domestic R&D and extension investment strategies.

Finally, this study has focused on quantifying the private returns to public investment in R&D and extension. However, a range of social benefits from publicly funded research may arise through application of agricultural public R&D outputs beyond the broadacre sector and/or incidental effects on environmental quality or human health and safety. To the extent that public investments in agricultural R&D and extension benefit society more broadly (that is, beyond broadacre farmers), accounting for such social benefits would translate into higher internal rates of return to public investments in agricultural R&D and extension than those presented here.

37

Aappendix

37

In general terms, a model linking the stock of knowledge to past R&D expenditures can be written in the form of equation (A-1) (Alston et al. 2010a):

!"!! = !! !!! ,!!!!! ,… ,!!!!!!! (A-1)

where !"!! denotes the knowledge stocks corresponding to various activities i, ! = !",!",!"#,!" where DS, PS, EXT and FS are past expenditures on domestic public R&D, domestic private R&D, domestic extension and foreign public and private R&D, respectively. The investment at time t is denoted by !"!! = !! !!! ,!!!!! ,… ,!!!!!!

! and the maximum time lag for each knowledge stock variable is !!! . The distribution functions for alternative R&D lag profiles are denoted by !! ∙ .

The distribution function for lagged weights specific to research activity i, !! ∙ , is equal to

,

where !!! is the non-negative weight on an investment made in time t (reflecting that an investment will contribute to the knowledge stock for !!! years), where the weights on a given investment are normalised such that

.

Equation (A-1) implies that each knowledge stock variable will be affected by two factors: the maximum time lag (!!! ) and the distribution function for time lags characterised by the weights (!!! ). It is not possible to estimate the parameters associated with these two factors in an unconstrained model. This is because it would be necessary to estimate a large number of research effects for each year within the lag period, which cannot be estimated independently since research expenditures over time are likely to be highly correlated (Alston et al. 2010b). Therefore, the distribution function and the maximum time lag need to be specified before constructing the knowledge stock variables.

In this report, a range of models was used to reflect prior knowledge and assumptions about the length and shape of the lag distribution (among other factors) and a small group of models was selected that had better econometric properties. More specifically, 10 distribution functions defining R&D knowledge stocks as a function of past research expenditures were considered: three gamma distributions, a trapezoid distribution and a geometric distribution

Constructing knowledge stocks

.


38

for 16-year and 35-year lags. The maximum length of time lags is specified as 16 years and 35 years following Mullen and Cox (1995), who argued that research lags in Australian agriculture should be close to these two settings.

The shape parameters for the first gamma distribution were obtained from a vector auto-regression model (with data at the industry level), which showed the peak impact of research occurred seven years after investment. A peak impact seven years after investment was earlier than expected (for example, Alston et al. 2010a), who concluded that US agricultural research lags were best represented by a lag structure that allowed for a peak impact 24 years after investment). For this reason, the second gamma distribution (Gamma_T) with a later peak impact occurring 13 years after investment (similar to the trapezoid distribution) was also considered. The third gamma distribution (Gamma_P) mimicked the geometric distribution. The models used in this report are defined as follows.

R&D distributed lag models with 35-year lags (figure 12)Gamma: gamma distribution with peak impact occurring after 7 years (with shape parameters λ = 0.6 and β = 0.7)

• Trapezoid: trapezoid distribution with peak impact in years 9 to 15, used previously by Mullen and Cox (1995)

• Gamma Trap: gamma distribution with peak impact occurring after 13 years to mimic the trapezoid distribution (with shape parameters λ = 0.7 and β = 0.8)

• PIM (perpetual inventory method): geometric distribution with a peak impact in the first year, after which the impact depreciates at a rate of 15 per cent a year

• Gamma PIM: gamma distribution mimicking PIM (with shape parameters λ = 0.8 and β = 0.3).

Distributed lag functions (35-year lags)12

gamma

PIM geometricgamma traptrapeziod

gamma PIM

weights

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

35302520151051

39


Distributed lag functions (16-year lags)13

gamma

PIM geometricgamma triangulartriangular

gamma PIM

weights

0.025

0.050

0.075

0.100

0.125

0.150

0.175

0.200

161284 141062

R&D distributed lag models with 16-year lags (figure 13)• Gamma: gamma distribution with peak impact

occurring after 5 years (with shape parameters λ = 0.5 and β = 0.7)

• Triangular: inverted-V research profile with peak impact occurring 8 years after investment, used previously by Mullen and Cox (1995)

• Gamma Triangular: gamma distribution with peak impact occurring after 8 years to mimic the triangular distribution (with shape parameters λ = 0.6 and β = 0.8)

• PIM (perpetual inventory method) : geometric distribution with a peak impact in the first year, after which the impact depreciates at a rate of 25 per cent a year

• Gamma PIM: gamma distribution mimicking PIM (with shape parameters λ = 0.8 and β = 0.1).

40

appe

ndix

40

B

This report uses the total public R&D knowledge stock (ln !"!!" = ln !"!

!" + ! ln !"!!" ) (defined as the weighted sum

of domestic and foreign public R&D knowledge stocks) in the regression to avoid the omitted variable problem associated with excluding the foreign public R&D knowledge stock and the multicollinearity problem associated with directly including foreign and domestic public R&D knowledge stocks separately. A detailed derivation can be used to justify the treatment as below.

Define ln !"!!" = ln !"!

!" + ! ln !"!!" and ln !"!

!" = ln !"!!" − ! ln !"!

!" as two orthogonal linear combinations of the domestic and foreign R&D knowledge stocks, respectively: namely, the total public R&D knowledge stock and the ratio of the domestic public R&D knowledge stock to the foreign public R&D knowledge stock:

ln !"!!" = ln !"!

!" + ! ln !"!!" (B-1)

ln !"!!" = ln !"!

!" − ! ln !"!!" (B-2)

Equations (B–1) and (B-2) can be solved for ln !"!!" and ln !"!

!" , such that:

ln !"!!" =

12 ln !"!

!" + ln !"!!" (B-3)

ln !"!!" =

12! ln !"!

!" − ln !"!!"

(B-4)

Substituting equations (B-3) and (B-4) into the unconstrained base model (2) yields equation (B-5)

ln !"#! = ! + !!12 ln !"!

!" + ln !"!!" + !!ln !"#!

+!!12! ln !"!

!" − ln !"!!" + !ln !! + !! (B-5)

which can be rearranged as (B-6)

ln !"#! = ! +!!2 +

!!2! ln !"!

!" + !!ln !"#! +!!2 −

!!2! ln !"!

!" + !ln !! + !!

Assuming yields ! = !!,!!,!! =!!2 +

!!2! , !!,

!!2 −

!!2!

ln !"#! = ! + !!ln !"!

!" + !!ln !"#! + !!ln !"!!" + !ln !! + !! (B-7)

Derivation of the total knowledge stock treatment

41


Equation (B-7) is exactly same as equation (2), with ln !"!!" = ln !"!

!" + ! ln !"!!" and ln !"!

!" = ln !"!!" − ! ln !"!

!" being less correlated from a statistical perspective since they are orthogonal to each other. Thus, a regression based on equation (B-7) can be used to estimate the impact of R&D knowledge stock on broadacre TFP without losing any information.

The t-statistic test was used to exclude the term ln !!!!" = ln !"!

!" − ! ln !"!!" ,

followed by the Ramsey RESET specification tests for the potential omitted variable problem due to the new specification. The results showed that the use of the weighting factor to aggregate domestic and foreign knowledge stocks generated no statistically significant bias in the unconstrained model specification.

Finally, since the coefficient on ln!"!!" is not significant at the 10 per cent level (which implies

that TFP is insensitive to the relative shares of domestic and foreign knowledge stocks since they might be indifferent in nature), ln!"!

!" should be eliminated from equation (B-7) which results in equation (3) in chapter 3.

ln !"#! = ! + !!ln !"!!" + !!ln !"#! + !ln !! + !! (3)

Equation (3) is preferred to two alternative specifications (widely used in previous literature): equation (2) (the unconstrained model in chapter 3) and equation (B-8) (excluding foreign R&D knowledge stocks),

(2) ln !"#! = !! + !!ln !"!!" + !!ln !"#! + !!ln !"!

!" + !ln !! + !!

ln !"#! = µμ! + µμ!ln !"!!" + µμ!ln !"#! + !ln !! + !! (B-8)

This is because equation (3) includes the impact of foreign public R&D knowledge stocks without introducing additional correlation between domestic and foreign R&D knowledge stocks. A Ramsey RESET test has also been carried out to examine the fitness of equation (3) to the real data and the results suggest that it performs better than equations (2) and (B-8).

Choice of the weight on the foreign public R&D knowledge stockThe total public R&D knowledge stock variable, ln !"!

!" = ln !"!!" + ! ln !"!

!" , was defined as the weighted average of domestic and foreign public R&D knowledge stocks. This was guided by two assumptions: (i) the lag profiles associated with domestic and foreign public R&D are the same; and (ii) foreign public R&D has a smaller effect on broadacre TFP than the domestic public R&D knowledge stock by a factor π. The total public R&D knowledge stock was defined as:

ln !"!!" = ln !"!

!" + ! ln !"!!" (B-9)

The weight π on ‘spill-ins’ of foreign public R&D knowledge stocks was derived based on the agro-ecological proximity approach proposed by Alston et al. (2010a), who used it as a measure of the similarity in output mix to weight interstate spillovers from US state research


42

agencies. In this report, a US public research knowledge stock is used as a proxy for spill-ins to Australia from overseas research.

First, an agro-ecological similarity index was estimated for the United States and Australia based on all major agricultural commodities produced, classified into five categories including meat livestock, other livestock products, grains, oilseeds, and other agricultural products. The agro-ecological similarity index (ω) is given by:

! =!!"!!"!

!!!

!!"!!!!!

! ! !!"!!!!!

! ! (B-10)

where fim and fjm are the value of production of output , expressed as a share of the total value of agricultural output in country i (that is, Australia) and country j (that is, the United States), where there is a total of M different commodity categories for Australia and the United States, and M=5. Data on fim and fjm for Australia were obtained from the ABARES Australian Agricultural and Grazing Industries Survey (AAGIS). Data on fim and fjm for the US were obtained from ERS-USDA.

For the period 1952–53 to 2006–07, the average value of the agro-ecological similarity index (ω) is 0.96, which suggests that agricultural production in Australia and the US is very similar for major crops and livestock products.

Second, following Shanks and Zheng (2006), the agro-ecological similarity index should be further deflated by share of imports in GDP for Australia to obtain the weight on spill-ins of foreign public R&D knowledge stocks (π in equation A-2). This is to take into account the trade and non-trade barriers to agricultural knowledge transfer across countries. According to World Development Indicators statistics (http://data.worldbank.org/data-catalog/world-development-indicators), the share of imports in GDP ranged from 8 per cent to 23 per cent between 1952–53 and 2006–07.

Based on these considerations, the weight on foreign public R&D knowledge stocks (π) was defined as the agro-ecological similarity index multiplied by the share of imports in GDP. For the period 1952–53 to 2006–07, the average value of π was 0.1. It is to be noted that the choice of the value of π may affect the fitness of the model, but the sensitivity of results to the assumption that π =0.1 was not tested.

43

appe

ndix

43

C

In the first alternative specification (equation C-1), broadacre TFP was modelled as a function of domestic and foreign R&D knowledge stocks and a domestic extension knowledge stock, and the control variables (soil moisture availability, the terms of trade and farmers’ level of education):

ln !"#! = !! + !!ln !"!!" + !!ln !"#! + !!ln !"!

!" + !ln !! + !! (C-1)

The second alternative specification (equation C-2) was based on the model estimated by Mullen and Cox (1995), who excluded the foreign public R&D knowledge stock because of the lack of suitable data. In this specification, broadacre TFP was modelled as a function of domestic R&D and extension knowledge stocks and the control variables:

ln !"#! = µμ! + µμ!ln !"!!" + µμ!ln !"#! + !ln !! + !! (C-2)

However, due to a high degree of collinearity between R&D and extension, Mullen et al. (see Mullen and Cox 1995; Mullen 2007; Binenbaum et al. 2008) were unable to estimate the effects of public R&D and extension knowledge stocks on broadacre TFP separately. Instead, they aggregated the R&D and extension knowledge stocks, weighting the extension knowledge stock by a factor of 0.1.

The third alternative specification (equation C-3) followed Mullen and Cox (1995) by modelling broadacre TFP as a function of a single domestic R&D and extension knowledge stock and the control variables.

ln !"#! = !! + !! ln !"!!" + 0.1ln !"#! + !ln !! + !! (C-3)

Recall that in the specification described in chapter 4, a total public R&D knowledge stock variable was defined to deal with the high correlation between foreign and domestic public R&D knowledge stocks. Broadacre TFP was modelled as a function of the total public R&D and extension knowledge stocks and the control variables:

ln !"#! = ! + !!ln !"!!" + !!ln !"#! + !ln !! + !! (C-4)

Alternative specifications (treatment of knowledge stocks)

.

44

appe

ndix

44

Equation (3) in chapter 4 describes the equilibrium relationship between public R&D and extension knowledge stocks and broadacre TFP, which can be re-arranged as an error correction model (ECM) to allow for a range of dynamic effects of R&D and extension knowledge stocks and the control variables on broadacre TFP. More specifically, starting with the equilibrium relationship in equation (3) implies:

!!!! = ! + !!ln !"!!!!" + !!ln !"#!!! + !ln !!!! − ln !"#!!! (D-1)

If there is a co-integrating relationship between broadacre TFP and its determinants, residuals obtained from equation (3) would be identically distributed independent of time (or ! !! = ! !!!! = 0 ). Adding equation (D-1) to equation (3), a general dynamic relationship between broadacre TFP and its determinants can be specified by including one-period lags of broadacre TFP and its determinants: (D-2)

When broadacre TFP and its determinants are in their long-run equilibrium relationship, ln !"#! = ln !"#!!! , ln !"!

!" = ln !"!!!!" and so forth, with !! = 0 . Setting

ln ∙ ! = ln ∙ !!! and !! = 0 yields: (D-3)

1− ! ln !!"! = !! + !!! + !!!! ln !"!!" + !!! + !!!! ln !"#! + !! + !!! ln !!

Rearranging equation (D-3) gives: (D-4)

ln !"#! =!!

1− ! +!!! + !!!!

1− ! ln !"!!" +

!!! + !!!!

1− ! ln !"#! +!! + !!!

1− ! ln !!

Assuming equation (D-4) corresponds to equation (3) implies

! =!!

1− ! !! =!!! + !!!!

1− ! ! =!! + !!!

1− ! !! =!!! + !!!!

1− !

Letting κ = (1-Ɵ) gives α‘ = κα;β1‘‘ = (κβ1- β1‘);β2‘‘ = (κβ2- β2‘); γ‘‘ = (κγ-γ‘)

Substituting κ, α‘, β1‘‘, β2‘‘, γ‘‘ and into equation (3) yields:

Derivation of the error correction modelD

45


(D-5)

Rearranging equation (D-5) yields

(D-6)

And defining d1n(.) = 1n(.)t-1n(.)t-1 equation (D–6) is the structural form of an error correction model:

(D-7)

The ECM says a change in broadacre TFP (d1n(TFPt )) can be explained by the changes in its determinants, d1n(TFPt

kj ), d1n(EXTt ), and d1n(Zt ), in the current period (or the short-term

effect) and the residuals from the long-term relationship (that is, the lagged residual εt-1 , where εt-1 is the difference between 1n(TFPt-1 ) and its equilibrium value in period t-1).

In practice, the ECM is estimated to decompose the average impact of public R&D and extension knowledge stocks into the long-term and short-term components, captured by βi and β‘i in equation (D-7).

46

appe

ndix

46

Testing for unit roots and co-integrationThe time series variables were tested for unit roots, with the results for selected variables presented in table E-1. The results of the Phillips-Perron and Dickey-Fuller tests indicate that broadacre TFP, the R&D and extension knowledge stock variables, the education index and the terms of trade have unit roots. The two-step Engel-Granger test was then performed, with results indicating that a co-integrating relationship exists between broadacre TFP and its determinants (table E2).

As a result, the OLS estimators, controlling for heteroscedasticity, are unbiased and valid to compare model specifications. However, OLS estimates may still be inconsistent.

Results

E1 Unit root tests for selected variables

Phillips-Perron Test Dickey-Fuller Test

MacKinnon MacKinnon Z(ρ) Z(t) P-Value Z(t) P-Value

ln(TFPt ) −3.28 −1.66 0.45 −1.88 0.34ln(WSIt ) −42.00 −6.13 0.00 −6.76 0.00ln(EDUCt ) −6.43 −2.47 0.12 −2.42 0.14ln(TOTt ) −2.16 −1.47 0.55 −1.50 0.54ln(EXTt ) −3.17 −4.29 0.00 −6.16 0.00

R&D distributed lag models (35 years) ln(TSt ) : Gamma 0.04 0.11 0.97 0.39 0.98ln(TSt ) : Gamma_T −0.56 −2.07 0.26 −3.82 0.00ln(TSt ) : Trap −0.57 −1.99 0.29 −3.65 0.00ln(TSt ) : Gamma_P −0.04 −0.18 0.94 −0.20 0.94ln(TSt ) : PIM −0.16 −0.80 0.82 −1.30 0.63

R&D distributed lag models (16 years) ln(TSt ) : Gamma −1.29 −2.90 0.05 −4.94 0.00ln(TSt ) : Gamma_T −1.47 −3.35 0.01 −6.13 0.00ln(TSt ) : Trap −1.34 −2.91 0.04 −5.23 0.00ln(TSt ) : Gamma_P −1.08 −2.30 0.17 −3.97 0.00ln(TSt ) : PIM −1.45 −4.36 0.00 −7.66 0.00

Notes: The null hypothesis is that there is unit root. The critical values for Z(ρ) are −18.9, −13.3 and −10.7 at the 1 per cent, 5 per cent and 10 per cent levels of significance, respectively. The critical values for Z(t) are −3.58, −2.93 and −2.60 at the 1 per cent, 5 per cent and 10 per cent levels of significance, respectively.

E

47


Testing for multicollinearity and omitted variable bias: knowledge stocksAs expected, domestic and foreign public R&D expenditures and domestic extension expenditures were highly correlated (table E3). As a result, a model including the domestic and foreign public R&D knowledge stocks and the extension knowledge stock appeared to suffer from multicollinearity, meaning that the coefficients on the knowledge stock variables could not be precisely estimated. In particular, over a range of the distribution profiles, the domestic public R&D knowledge stock was insignificant and the estimated effect of the extension knowledge stock on broadacre TFP was negative (table E4).

However, models that excluded foreign public R&D knowledge stocks to reduce multicollinearity had larger RESET statistics, suggesting those models potentially suffered from omitted variable bias (tables E5 and E6). Combining domestic and foreign public R&D knowledge stocks (to avoid causing further multicollinearity problems) resulted in lower RESET statistics and the failure to reject the null hypothesis of no omitted variables at a 10 per cent level of statistical significance (table E7).

E2 Co-integration tests (TS specification)

Gamma Gamma_T Trapezoid Gamma_P PIMR&D distributed lag models with 35 year lagsDickey-Fuller Test −4.8 −6.1 −4.7 −4.9 −5.0 (0.000) (0.000) (0.000) (0.000) (0.000)White Noise Test Not Reject Not Reject Not Reject Not Reject Not RejectJohansen Co-integration Test 723.0 668.9 830.2 613.4 680.3 (0.000) (0.000) (0.000) (0.000) (0.000)

R&D distributed lag models with 16 year lagsDickey-Fuller Test −4.8 −2.6 −4.7 −2.4 −4.8 (0.000) (0.075) (0.000) (0.078) (0.000)White Noise Test Not Reject Not Reject Not Reject Not Reject Not RejectJohansen Co-integration Test 686.1 631.8 711.5 584.4 641.0 (0.000) (0.000) (0.000) (0.000) (0.000)

Notes: The null hypothesis is no co-integration. The numbers in parenthesis are P-values.

E3 Correlation Matrix: domestic public R&D expenditures, foreign public R&D expenditures and domestic public extension expenditures

Domestic R&D Foreign R&D Domestic Extension

Domestic R&D 1.00 – –Foreign R&D 0.96 1.00 –Domestic Extension 0.83 0.91 1.00


48

E4

Elas

ticiti

es o

f bro

adac

re T

FP to

pub

lic d

omes

tic a

nd fo

reig

n R&

D k

now

ledg

e st

ocks

and

dom

estic

ext

ensi

on

know

ledg

e st

ock

R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs

R&D

dis

trib

uted

lag

mod

els

with

16-

year

lags

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

ln(D

S t )

0.14

* 0.

18

0.45

**

0.14

* 0.

39

0.09

0.

20

0.28

**

0.10

0.

03

(0.0

7)

(0.1

9)

(0.2

2)

(0.0

8)

(0.2

9)

(0.1

1)

(0.1

7)

(0.1

2)

(0.1

0)

(0.1

3)ln

(EXT

t )

0.07

−

0.01

−

0.15

**

0.06

−

0.01

−

0.15

* −

0.26

**

−0.

25**

−

0.14

**

−0.

17**

(0

.08)

(0

.06)

(0

.07)

(0

.07)

(0

.10)

(0

.08)

(0

.13)

(0

.11)

(0

.07)

(0

.08)

ln(F

S t )

0.05

0.

06

−0.

43**

0.

05

–0.6

2 0.

44**

* 0.

40**

0.

220*

* 0.

42**

* 0.

69**

(0

.24)

(0

.21)

(0

.22)

(0

.23)

(0

.77)

(0

.17)

(0

.17)

(0

.10)

(0

.16)

(0

.29)

ln(W

SIt )

0.

28**

* 0.

27**

* 0.

29**

* 0.

28**

* 0.

28**

* 0.

29**

* 0.

30**

* 0.

27**

* 0.

29**

* 0.

30**

*

(0.0

6)

(0.0

6)

(0.0

5)

(0.0

6)

(0.0

6)

(0.0

5)

(0.0

5)

(0.0

5)

(0.0

5)

(0.0

5)ln

(ED

UC

t )

0.33

0.

61

0.11

0.

35

0.68

1.

10**

* 0.

96**

1.

17**

1.

09**

* 0.

81**

(0

.63)

(0

.52)

(0

.45)

(0

.62)

(0

.48)

(0

.42)

(0

.46)

(0

.51)

(0

.42)

(0

.38)

ln(T

OT t )

−

0.23

**

−0.

23**

−

0.23

**

−0.

23**

−

0.21

**

−0.

30**

* −

0.30

***

−0.

29**

* −

0.29

***

−0.

26**

(0

.10)

(0

.11)

(0

.10)

(0

.10)

(0

.10)

(0

.11)

(0

.09)

(0

.10)

(0

.11)

(0

.11)

Con

stan

t 0.

80

–0.1

9 3.

98

0.69

2.

16

−2.

35

−1.

61

−2.

20

−2.

34

−2.

32

(2.8

4)

(2.3

7)

(2.5

2)

(2.7

8)

(2.3

7)

(2.0

7)

(2.1

9)

(2.4

4)

(2.0

6)

(2.1

0)N

umb

er o

f ob

serv

atio

ns

55

55

55

55

55

55

55

55

55

55R-

squa

red

0.

96

0.96

0.

96

0.96

0.

96

0.96

0.

96

0.96

0.

96

0.96

RESE

T Te

st (H

0: no

omitt

ed v

aria

ble)

F-

stat

isti

cs

1.45

1.

73

1.38

1.

54

1.60

1.

62

1.36

1.

48

1.63

1.

16P-

valu

e 0.

18

0.10

0.

21

0.14

0.

12

0.12

0.

19

0.13

0.

12

0.35

RMSE

0.

067

0.06

8 0.

066

0.06

7 0.

068

0.06

9 0.

069

0.07

0 0.

069

0.06

8

Not

es: O

LS m

odel

of t

he fi

rst a

ltern

ativ

e sp

ecifi

catio

n, b

road

acre

TFP

as

a fu

nctio

n of

dom

estic

and

fore

ign

R&D

kno

wle

dg

e st

ocks

and

a d

omes

tic e

xten

sion

kno

wle

dg

e st

ock,

and

the

cont

rol

varia

ble

s. S

tatis

tical

sig

nific

ance

at t

he 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

.

49


E5

Elas

ticiti

es o

f bro

adac

re T

FP to

pub

lic d

omes

tic R

&D

kno

wle

dge

stoc

ks a

nd d

omes

tic e

xten

sion

kno

wle

dge

stoc

k

R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs

R&D

dis

trib

uted

lag

mod

els

with

16-

year

lags

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

ln(D

S t )

0.15

***

0.20

***

0.23

***

0.16

***

0.15

***

0.27

***

0.45

***

0.41

***

0.26

***

0.30

***

(0

.03)

(0

.05)

(0

.05)

(0

.04)

(0

.04)

(0

.08)

(0

.11)

(0

.11)

(0

.07)

(0

.08)

ln(E

XTt )

0.

09**

0.

005

–0.0

3 0.

08**

0.

08**

−

0.14

* −

0.35

***

−0.

31**

* −

0.12

* −

0.16

**

(0.0

4)

(0.0

5)

(0.0

5)

(0.0

4)

(0.0

4)

(0.0

8)

(0.1

2)

(0.1

1)

(0.0

7)

(0.0

8)ln

(WSI

t )

0.28

***

0.27

***

0.27

***

0.28

***

0.28

***

0.28

***

0.29

***

0.29

***

0.28

***

0.28

***

(0

.06)

(0

.06)

(0

.06)

(0

.06)

(0

.06)

(0

.05)

(0

.05)

(0

.05)

(0

.05)

(0

.05)

ln(E

DU

Ct )

0.

25

0.52

0.

51

0.27

0.

27

1.14

**

0.80

* 0.

89*

1.12

**

0.98

**

(0.4

0)

(0.3

8)

(0.3

8)

(0.3

9)

(0.3

9)

(0.4

7)

(0.4

7)

(0.4

7)

(0.4

6)

(0.4

4)ln

(TO

T t )

−0.

23**

−

0.23

**

−0.

22**

−

0.23

**

−0.

23**

−

0.32

***

−0.

35**

* −

0.34

***

−0.

31**

* −

0.29

***

(0

.10)

(0

.10)

(0

.10)

(0

.10)

(0

.1)

(0.1

1)

(0.0

9)

(0.1

0)

(0.1

1)

(0.1

1)C

onst

ant

1.22

0.

30

0.28

1.

13

1.17

−

1.29

0.

41

−0.

03

−1.

31

−0.

83

(1.6

6)

(1.7

2)

(1.7

1)

(1.6

6)

(1.6

5)

(2.2

4)

(2.0

6)

(2.0

8)

(2.1

9)

(2.0

5)N

umb

er o

f ob

serv

atio

ns

55

55

55

55

55

55

55

55

55

55R-

squa

red

0.

96

0.96

0.

96

0.96

0.

96

0.95

0.

95

0.95

0.

95

0.96

RESE

T Te

st (H

0: no

omitt

ed v

aria

ble)

F-

stat

isti

cs

2.30

4.

28

4.41

2.

21

1.95

4.

12

3.94

4.

00

4.33

4.

45P-

valu

e 0.

09

0.01

0.

01

0.10

0.

13

0.01

0.

01

0.01

0.

01

0.01

RMSE

0.

066

0.06

7 0.

067

0.06

6 0.

066

0.07

1 0.

071

0.07

1 0.

070

0.07

1

Not

es: O

LS m

odel

of t

he s

econ

d al

tern

ativ

e sp

ecifi

catio

n, b

road

acre

TFP

as

a fu

nctio

n of

dom

estic

pub

lic R

&D

kno

wle

dg

e st

ocks

and

a d

omes

tic e

xten

sion

kno

wle

dg

e st

ock,

and

the

cont

rol

varia

ble

s. S

tatis

tical

sig

nific

ance

at t

he 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

.


50

E6

Elas

ticiti

es o

f bro

adac

re T

FP to

a p

ublic

dom

estic

R&

D a

nd e

xten

sion

kno

wle

dge

stoc

k (M

ulle

n &

Cox

199

5 sp

ecifi

catio

n)

R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs

R&D

dis

trib

uted

lag

mod

els

with

16-

year

lags

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

ln(D

S t )+

0.1l

n(EX

T t )

0.17

2***

0.

208*

**

0.21

4***

0.

181*

**

0.18

0***

0.

157*

**

0.14

4***

0.

146*

**

0.16

3***

0.

166*

**

(0.0

35)

(0.0

36)

(0.0

37)

(0.0

35)

(0.0

36)

(0.0

35)

(0.0

36)

(0.0

36)

(0.0

35)

(0.0

35)

ln(W

SIt )

0.

266*

**

0.27

1***

0.

275*

**

0.26

8***

0.

269*

**

0.28

4***

0.

287*

**

0.28

6***

0.

283*

**

0.28

2***

(0

.063

) (0

.055

) (0

.054

) (0

.062

) (0

.062

) (0

.050

) (0

.050

) (0

.050

) (0

.050

) (0

.050

)ln

(ED

UC

t )

0.12

9 0.

513

0.54

7 0.

156

0.15

6 1.

064*

* 0.

910*

0.

936*

1.

060*

* 0.

973*

*

(0.4

18)

(0.4

12)

(0.4

23)

(0.4

13)

(0.4

12)

(0.5

00)

(0.5

42)

(0.5

35)

(0.4

90)

(0.4

95)

ln(T

OT t )

−

0.29

0***

−

0.23

0**

−0.

235*

* −

0.27

1***

−

0.27

8***

−

0.41

1***

−

0.49

5***

−

0.48

1***

−

0.39

0***

−

0.39

9***

(0

.100

) (0

.095

) (0

.094

) (0

.098

) (0

.098

) (0

.081

) (0

.070

) (0

.072

) (0

.083

) (0

.080

)C

onst

ant

2.81

2 0.

271

0.03

7 2.

467

2.52

5 –0

.718

0.

518

0.30

7 –0

.859

–0

.487

(1

.958

) (1

.963

) (2

.003

) (1

.922

) (1

.925

) (2

.332

) (2

.455

) (2

.425

) (2

.294

) (2

.283

)N

umb

er o

f ob

serv

atio

ns

55

55

55

55

55

55

55

55

55

55R-

squa

red

0.

957

0.95

9 0.

959

0.95

8 0.

957

0.95

2 0.

949

0.95

0 0.

952

0.95

3

RESE

T Te

st (H

0: no

omitt

ed v

aria

ble)

F-

stat

isti

cs

2.20

1.

76

1.74

2.

09

2.18

2.

12

2.06

2.

12

2.08

2.

25P-

valu

e 0.

032

0.09

1 0.

095

0.04

2 0.

034

0.03

9 0.

045

0.03

9 0.

043

0.02

9RM

SE

0.06

9 0.

067

0.06

6 0.

069

0.06

8 0.

072

0.07

4 0.

074

0.07

2 0.

072

Not

es: O

LS m

odel

of t

he th

ird a

ltern

ativ

e sp

ecifi

catio

n (M

ulle

n an

d C

ox 1

995)

, bro

adac

re T

FP a

s a

func

tion

of a

dom

estic

pub

lic R

&D

and

ext

ensi

on k

now

led

ge

stoc

k, a

nd th

e co

ntro

l var

iab

les.

St

atis

tical

sig

nific

ance

at t

he 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

.

51


E7

Elas

ticiti

es o

f bro

adac

re T

FP to

a to

tal p

ublic

R&

D k

now

ledg

e st

ock

and

dom

estic

ext

ensi

on k

now

ledg

e st

ock

R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs

R&D

dis

trib

uted

lag

mod

els

with

16-

year

lags

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

ln(T

S t )

0.23

4***

0.

225*

**

0.24

2***

0.

202*

**

0.20

1***

0.

361*

**

0.46

2***

0.

425*

**

0.22

1***

0.

395*

**

(0.0

53)

(0.0

55)

(0.0

60)

(0.0

49)

(0.0

50)

(0.0

91)

(0.1

15)

(0.1

04)

(0.0

74)

(0.0

99)

ln(E

XTt )

0.

100*

**

0.10

4***

0.

072*

0.

143*

**

0.14

6***

−

0.11

2*

−0.

215*

* −

0.18

2**

0.01

6 −

0.12

0*

(0.0

37)

(0.0

36)

(0.0

38)

(0.0

35)

(0.0

35)

(0.0

65)

(0.0

87)

(0.0

78)

(0.0

49)

(0.0

67)

ln(W

SIt )

0.

275*

**

0.27

0***

0.

264*

**

0.27

6***

0.

275*

**

0.28

1***

0.

270*

**

0.27

0***

0.

262*

**

0.27

8***

(0

.056

) (0

.057

) (0

.058

) (0

.058

) (0

.059

) (0

.049

) (0

.053

) (0

.052

) (0

.057

) (0

.051

)ln

(ED

UC

t )

0.56

2 0.

662*

0.

853*

* 0.

019

0.29

3 1.

338*

**

1.39

3***

1.

453*

**

1.29

7***

1.

129*

**

(0.3

88)

(0.3

98)

(0.4

05)

(0.4

18)

(0.4

05)

(0.4

53)

(0.4

60)

(0.4

68)

(0.4

96)

(0.4

25)

ln(T

OT t )

−

0.26

6***

−

0.24

0**

−0.

255*

* −

0.26

2***

−

0.26

1***

–0

.285

***

−0.

303*

**

−0.

302*

**

−0.

352*

**

−0.

262*

*

(0.0

85)

(0.1

01)

(0.1

00)

(0.0

89)

(0.0

93)

(0.1

02)

(0.0

95)

(0.0

94)

(0.1

04)

(0.1

06)

Con

stan

t −

0.32

5 −

0.85

3 −

1.37

2 1.

843

0.61

4 -2

.372

−

2.19

6 −

2.51

9 −

2.02

4 −

1.77

4

(1.7

81)

(1.9

08)

(2.0

02)

(1.6

49)

(1.7

30)

(2.1

72)

(2.1

47)

(2.1

96)

(2.4

51)

(2.0

45)

Num

ber

of o

bse

rvat

ions

55

55

55

55

55

55

55

55

55

55

R-sq

uare

d

0.95

9 0.

957

0.95

6 0.

959

0.95

8 0.

955

0.95

5 0.

955

0.95

1 0.

956

RESE

T Te

st (H

0: no

omitt

ed v

aria

ble)

F-

stat

isti

cs

1.62

1.

39

1.49

1.

61

1.65

1.

37

1.77

1.

71

1.82

1.

31P-

valu

e 0.

12

0.21

0.

16

0.12

0.

11

0.22

0.

08

0.10

0.

07

0.25

Not

es: O

LS m

odel

of e

quat

ion

(3),

bro

adac

re T

FP a

s a

func

tion

of a

tota

l pub

lic R

&D

kno

wle

dg

e st

ock

(ln(T

St )=

ln(D

St )+

0.1

ln(F

St )

) and

a d

omes

tic e

xten

sion

kno

wle

dg

e st

ock,

and

the

cont

rol

varia

ble

s. S

tatis

tical

sig

nific

ance

at t

he 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

.


52

Comparing functional formsFocusing on the models that included the total knowledge stock variable, the log-linear functional form was preferred based on a comparison of the RMSE values and estimated coefficients of the linear, log-linear and quadratic model specifications (table E8). The root mean square error (RMSE) values were significantly higher in the linear models than for the quadratic and log-linear models, suggesting the linear model is a poor fit to the data. The RMSE values for the log-linear and quadratic models were similar and low. However, whereas the coefficients in the log-linear models were significant at a 1 per cent level of significance for most distributions, the coefficients in the quadratic models were not significant at a 10 per cent level of significance.

The linear, log-linear and quadratic model specifications for the Mullen and Cox (1995) model are also compared in table E8. In this (third alternative) specification, broadacre TFP is modelled as a function of a domestic public R&D and extension knowledge stock.

In addition, the gamma distribution with peak impact occurring after seven years was preferred over the alternative distributions (Trapezoid, Gamma_T, PIM and Gamma _P). For the log-linear total knowledge stock specification, the gamma and PIM distributions had the lowest RMSE values. However, as discussed previously, the PIM approach is inconsistent with the expectation that agricultural R&D investment will have little effect in its early years (table E8).

Comparing estimatorsAlthough OLS is unbiased due to the existence of a co-integrating relationship between broadacre TFP and its determinants, an ARIMA model was still the preferred regression method since its estimators are more efficient. Based on the log-linear function form, the efficiency of the two estimators was compared in tables E7 and E10. Although the OLS and ARIMA estimators generate the same coefficient values, the ARIMA estimates have smaller standard errors, indicating that the ARIMA estimator is more efficient. This was also the case when the OLS and ARIMA estimators were compared for the log-linear specification of the Mullen and Cox (1995) model (tables E6 and E11).

53


E8

Elas

ticiti

es o

f bro

adac

re T

FP to

a to

tal p

ublic

R&

D k

now

ledg

e st

ock

and

dom

estic

ext

ensi

on k

now

ledg

e st

ock:

co

mpa

ring

the

linea

r, lo

g-lin

ear a

nd q

uadr

atic

mod

els

R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs

R&D

dis

trib

uted

lag

mod

els

with

16-

year

lags

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

Line

ar M

odel

TSt

0.01

0***

0.

009*

**

0.01

0***

0.

009*

**

0.00

0***

0.

011*

**

0.01

5***

0.

014*

**

0.01

0***

0.

012*

**

(0.0

02)

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

00)

(0.0

02)

(0.0

02)

(0.0

02)

(0.0

02)

(0.0

02)

EXT t

0.00

0 0.

000

0.00

0 0.

000*

0.

000

−0.

000*

* −

0.00

1***

−

0.00

0***

−

0.00

0 −

0.00

0***

(0

.000

) (0

.000

) (0

.000

) (0

.000

) (0

.000

) (0

.000

) (0

.000

) (0

.000

) (0

.000

) (0

.000

)RM

SE

15.0

87

15.0

40

15.1

02

15.3

02

15.4

36

16.8

72

16.4

84

16.5

30

16.6

39

16.3

84

Log-

linea

r Mod

elln

(TS t )

0.

234*

**

0.22

5***

0.

242*

**

0.20

2***

0.

385*

**

0.36

1***

0.

462*

**

0.42

5***

0.

221*

**

0.39

5***

(0

.053

) (0

.055

) (0

.060

) (0

.049

) (0

.093

) (0

.091

) (0

.115

) (0

.104

) (0

.074

) (0

.099

)ln

(EXT

t )

0.10

0***

0.

104*

**

0.07

2*

0.14

3***

0.

134*

**

−0.

112*

−

0.21

5**

−0.

182*

* 0.

016

−0.

120*

(0

.037

) (0

.036

) (0

.038

) (0

.035

) (0

.035

) (0

.065

) (0

.087

) (0

.078

) (0

.049

) (0

.067

)RM

SE

0.06

7 0.

068

0.06

9 0.

067

0.06

8 0.

070

0.07

0 0.

070

0.07

3 0.

069

Qua

drat

ic M

odel

ln(T

S t )

1.92

7 0.

525

1.36

7 1.

689*

6.

741

1.93

1*

2.59

7 2.

188

0.13

0 2.

228

(1

.244

) (1

.458

) (2

.246

) (0

.950

) (4

.769

) (1

.004

) (1

.698

) (1

.414

) (2

.954

) (1

.366

)ln

(TS t )

^2

−0.

095

−0.

015

−0.

060

−0.

086

−0.

256

−0.

088

−0.

113

−0.

093

0.01

0 −

0.10

1

(0.0

72)

(0.0

83)

(0.1

27)

(0.0

55)

(0.1

93)

(0.0

59)

(0.0

98)

(0.0

82)

(0.1

64)

(0.0

78)

ln(E

XTt )

2.

082

1.81

4 2.

692

0.64

2 0.

321

−0.

923

0.48

7 1.

016

2.80

5 −

1.14

9

(1.4

44)

(1.8

62)

(1.7

90)

(1.5

45)

(1.6

52)

(2.1

10)

(2.3

40)

(2.2

17)

(2.4

07)

(2.2

91)

ln(E

XTt )

^2

−0.

094

−0.

077

−0.

121

−0.

028

−0.

013

0.03

0 −

0.04

1 −

0.06

1 −

0.12

5 0.

039

(0

.065

) (0

.081

) (0

.078

) (0

.067

) (0

.072

) (0

.093

) (0

.104

) (0

.099

) (0

.108

) (0

.100

)RM

SE

0.06

8 0.

069

0.06

9 0.

066

0.06

9 0.

070

0.07

0 0.

071

0.07

4 0.

070

Not

es: O

LS m

odel

of e

quat

ion

(3),

bro

adac

re T

FP a

s a

func

tion

of a

tota

l pub

lic R

&D

kno

wle

dg

e st

ock

(ln(

TSt )

=ln

(DS t )

+0.

1 ln

(FS t )

) and

a d

omes

tic e

xten

sion

kno

wle

dg

e st

ock.

Sta

tistic

al

sign

ifica

nce

at th

e 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. R

esul

ts fo

r the

con

trol

var

iab

les

not r

epor

ted

. The

val

ues

in p

aren

thes

es a

re s

tand

ard

erro

rs.


54

E9

Elas

ticiti

es o

f bro

adac

re T

FP to

a p

ublic

dom

estic

R&

D a

nd e

xten

sion

kno

wle

dge

stoc

k: c

ompa

ring

the

linea

r, lo

g-lin

ear a

nd q

uadr

atic

mod

els

(Mul

len

& C

ox 1

995

spec

ifica

tion)

R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs

R&D

dis

trib

uted

lag

mod

els

with

16-

year

lags

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

Line

ar M

odel

DS t+

0.1E

XTt

0.00

0***

0.

000*

**

0.00

0***

0.

000*

**

0.00

0***

0.

000*

**

0.00

0***

0.

000*

**

0.00

0***

0.

000*

**

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

RMSE

15

.381

14

.922

14

.924

15

.410

16

.730

17

.278

18

.389

18

.120

16

.934

16

.730

Log-

linea

r Mod

elln

(DS t )

+0.

1ln(

EXT t )

0.

172*

**

0.20

8***

0.

214*

**

0.18

1***

0.

166*

**

0.15

7***

0.

144*

**

0.14

6***

0.

163*

**

0.16

6***

(0

.035

) (0

.036

) (0

.037

) (0

.035

) (0

.035

) (0

.035

) (0

.036

) (0

.036

) (0

.035

) (0

.035

)RM

SE

0.06

9 0.

067

0.06

6 0.

068

0.07

2 0.

072

0.07

4 0.

074

0.07

2 0.

072

Qua

drat

ic M

odel

ln(D

S t )+0

.1ln

(EXT

t )

0.65

8***

0.

276

0.15

8 0.

627*

**

−0.

502

−0.

419

−1.

141

−0.

994

−0.

342

−0.

502

(0

.199

) (0

.269

) (0

.295

) (0

.206

) (0

.505

) (0

.516

) (0

.796

) (0

.736

) (0

.472

) (0

.505

)(l

n(D

S t )+0

.1ln

(EXT

t ))^

2 −

0.02

0**

−0.

003

0.00

2 −

0.01

8**

0.02

8 0.

024

0.05

2 0.

047

0.02

1 0.

028

(0

.008

) (0

.011

) (0

.013

) (0

.009

) (0

.021

) (0

.021

) (0

.032

) (0

.030

) (0

.020

) (0

.021

)RM

SE

0.07

0 0.

071

0.06

7 0.

069

0.07

3 0.

073

0.07

5 0.

075

0.07

3 0.

074

Not

es: O

LS m

odel

of t

he th

ird a

ltern

ativ

e sp

ecifi

catio

n (M

ulle

n an

d C

ox 1

995)

, bro

adac

re T

FP a

s a

func

tion

of a

dom

estic

pub

lic R

&D

and

ext

ensi

on k

now

led

ge

stoc

k, a

nd th

e co

ntro

l var

iab

les.

St

atis

tical

sig

nific

ance

at t

he 1

per

cen

t, 5

per

cen

t and

10

per

cen

t lev

els

is re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

.

55


E10 Elasticities of broadacre TFP to a total public R&D knowledge stock and domestic extension knowledge stock: ARIMA models for R&D distributed lag models with 35-year lags


ln(TSt ) 0.234*** 0.225*** 0.202*** 0.242*** 0.201*** (0.051) (0.053) (0.046) (0.057) (0.048)ln(EXTt ) 0.100*** 0.104*** 0.143*** 0.072** 0.146*** (0.035) (0.034) (0.033) (0.036) (0.034)ln(WSIt ) 0.275*** 0.270*** 0.276*** 0.264*** 0.275*** (0.054) (0.055) (0.055) (0.055) (0.056)ln(EDUCt ) 0.562 0.662* 0.019 0.853** 0.293 (0.368) (0.379) (0.398) (0.386) (0.385)ln(TOTt ) −0.266*** −0.240** −0.262*** −0.255*** −0.261*** (0.081) (0.096) (0.084) (0.095) (0.088)Constant –0.325 –0.853 1.843 –1.372 0.614 (1.693) (1.817) (1.570) (1.906) (1.646)Sigma-statistic 0.063*** 0.065*** 0.063*** 0.065*** 0.064*** (0.007) (0.007) (0.007) (0.007) (0.007)Number of observations 55 55 55 55 55

Notes: ARIMA model of equation (3), broadacre TFP as a function of a total public R&D knowledge stock (ln(TSt )=ln(DSt )+0.1 ln(FSt )) and a domestic extension knowledge stock, and the control variables. Statistical significance at the 1 per cent, 5 per cent and 10 per cent levels is represented by ***, ** and *. The values in parentheses are standard errors.

E11 Elasticities of broadacre TFP to a public domestic R&D and extension knowledge stock: ARIMA models for R&D distributed lag models with 35-year lags (Mullen & Cox 1995 specification)


ln(DSt )+0.1ln(EXTt ) 0.172*** 0.208*** 0.214*** 0.181*** 0.166*** (0.034) (0.035) (0.035) (0.034) (0.034)ln(WSIt ) 0.266*** 0.271*** 0.275*** 0.268*** 0.282*** (0.060) (0.053) (0.052) (0.059) (0.048)ln(EDUCt ) 0.129 0.513 0.547 0.156 0.973** (0.402) (0.396) (0.407) (0.397) (0.476)ln(TOTt ) −0.290*** −0.230** −0.235*** −0.271*** −0.399*** (0.096) (0.092) (0.091) (0.094) (0.077)Constant 2.812 0.271 0.037 2.467 –0.487 (1.881) (1.887) (1.925) (1.850) (2.196)Sigma-statistic 0.065*** 0.063*** 0.063*** 0.065*** 0.068*** (0.008) (0.007) (0.006) (0.008) (0.006)Number of observations 55 55 55 55 55

Notes: ARIMA model of the third alternative specification (Mullen and Cox 1995), broadacre TFP as a function of a domestic public R&D and extension knowledge stock, and the control variables. Statistical significance at the 1 per cent, 5 per cent and 10 per cent levels is represented by ***, ** and *. The values in parentheses are standard errors.


56

E12

Elas

ticiti

es o

f bro

adac

re T

FP to

a p

ublic

dom

estic

R&

D a

nd e

xten

sion

kno

wle

dge

stoc

k: E

rror

Cor

rect

ion

Mod

el fo

r R&

D d

istr

ibut

ed la

g m

odel

s w

ith 3

5-ye

ar la

gs (M

ulle

n &

Cox

199

5 sp

ecifi

catio

n)

G

amm

a G

amm

a_T

Trap

ezoi

d G

amm

a_P

PIM

LR

SR

LR

SR

LR

LR

SR

LR

SR

LR

ln(D

S t )+

0.1l

n(EX

T t )

0.17

8***

0.

957

0.20

6***

0.

204

0.21

2***

0.

190

0.18

8***

1.

038*

0.

188*

**

1.41

6**

(0

.034

) (0

.609

) (0

.034

) (0

.366

) (0

.035

) (0

.361

) (0

.034

) (0

.615

) (0

.035

) (0

.647

)ln

(WSI

t )

0.25

0***

0.

279*

**

0.26

1***

0.

281*

**

0.26

5***

0.

283*

**

0.25

2***

0.

278*

**

0.25

2***

0.

277*

**

(0.0

61)

(0.0

29)

(0.0

54)

(0.0

29)

(0.0

53)

(0.0

29)

(0.0

60)

(0.0

28)

(0.0

60)

(0.0

28)

ln(E

DU

Ct )

0.

211

–0.9

95

0.58

7 –0

.542

0.

620

–0.4

36

0.24

0 –0

.940

0.

238

–0.9

35

(0.3

86)

(0.9

35)

(0.4

00)

(0.8

59)

(0.4

11)

(0.8

26)

(0.3

84)

(0.9

15)

(0.3

83)

(0.8

96)

ln(T

OT t )

−

0.27

5***

−

0.24

3***

−

0.23

8***

−

0.23

1***

−

0.24

3***

−

0.22

9***

−

0.25

6***

−

0.23

5***

−

0.25

9***

−

0.23

2***

(0

.096

) (0

.093

) (0

.091

) (0

.085

) (0

.089

) (0

.083

) (0

.094

) (0

.089

) (0

.094

) (0

.086

)RE

SD

– −

0.89

5***

–

−0.

896*

**

– −

0.89

5***

–

−0.

911*

**

– −

0.92

7***

–

(0.1

37)

– (0

.138

) –

(0.1

38)

– (0

.136

) –

(0.1

36)

Con

stan

t 2.

377

−0.

053

0.06

4 0.

000

−0.

159

0.00

1 2.

025

−0.

057

2.05

5 −

0.08

3*

(1.8

29)

(0.0

44)

(1.9

14)

(0.0

25)

(1.9

50)

(0.0

24)

(1.8

08)

(0.0

44)

(1.8

02)

(0.0

46)

Sigm

a–st

atis

tic

0.06

3***

0.

062*

**

0.06

3***

0.

062*

**

0.06

3***

0.

062*

**

0.06

3***

0.

062*

**

0.06

3***

0.

061*

**

(0.0

09)

(0.0

07)

(0.0

07)

(0.0

07)

(0.0

07)

(0.0

06)

(0.0

09)

(0.0

07)

(0.0

09)

(0.0

07)

Num

ber

of o

bse

rvat

ions

54

54

54

54

54

54

54

54

54

54

Not

es: B

road

acre

TFP

as

a fu

nctio

n of

a d

omes

tic p

ublic

R&

D a

nd e

xten

sion

kno

wle

dg

e st

ock,

and

the

cont

rol v

aria

ble

s. ‘G

amm

a’ re

fers

to th

e p

refe

rred

mod

el s

pec

ifica

tion

in w

hich

a g

amm

a di

strib

utio

n w

as u

sed

to c

onst

ruct

kno

wle

dg

e st

ocks

, with

a p

eak

imp

act o

ccur

ring

seve

n ye

ars

afte

r inv

estm

ent.

Stat

istic

al s

igni

fican

ce a

t the

1 p

er c

ent,

5 p

er c

ent a

nd 1

0 p

er c

ent l

evel

s is

re

pre

sent

ed b

y **

*, **

and

*. T

he v

alue

s in

par

enth

eses

are

sta

ndar

d er

rors

.

57

appe

ndix

57

The elasticities in table 2 can be used to calculate the relative contributions of total public R&D and extension knowledge stocks to annual broadacre TFP growth between 1952–53 and 2006–07.

From equation (3), taking the first-order total derivative of both the dependent and explanatory variables with respect to time yields:

(F-1)

From equation (F-1) it can be seen that the annual growth of broadacre TFP (TFPt ) is equal to the sum of the annual growth rates of the explanatory variables multiplied by their corresponding elasticity (βi ) (table 2). The average annual growth rates and estimated elasticities of the total public R&D and extension knowledge stocks, and their relative contributions to annual broadacre TFP growth between 1952–53 and 2006–07 are reported in table F1.

The estimate of the contribution of the total public R&D knowledge stock to broadacre TFP growth was also decomposed into the relative contributions of domestic and foreign R&D.

First, an ARIMA model of equation (C-1) in appendix C was estimated

(C-1) ln !"#! = !! + !!ln !"!!" + !!ln !"#! + !!ln !"!

!" + !ln !! + !!

Although this model suffers from multicollinearity, estimates of the elasticities of broadacre TFP to the domestic and foreign public knowledge stocks (δ1 and δ3 ), multiplied by their growth rates, can be used to decompose the contribution of the total public R&D knowledge stock to broadacre TFP growth into the shares attributable to domestic and foreign public R&D.

Contribution of public R&D and extension knowledge stocks to annual broadacre TFP growth

F1 Contributions of the public R&D and extension knowledge stocks to annual broadacre TFP growth: 1952–53 to 2006–07

Gamma distribution

Contribution Growth rate (%) Elasticity to TFP growth

Total public R&D knowledge stocks (TSt ) 4.13 0.234 0.97Domestic R&D knowledge stocks (DSt ) 7.71 – 0.33Foreign R&D knowledge stocks (FSt ) 2.29 – 0.63Extension knowledge stocks (EXTt ) 2.65 0.100 0.27Other factors – – 0.73TFP Growth 1.96 – 1.96

F

58

appe

ndix

58

To measure economic benefits obtained from public investment in R&D and extension, it is necessary to convert broadacre TFP growth due to public R&D and extension investments into a gross value of output, which can be compared with the associated investment costs. In this report, a cost–benefit framework has been used to estimate the internal rate of return (IRR) to public investment in R&D and extension.

The procedure for estimating the IRR for an increase in investment in R&D and extension (say, $1000) at time t (Rt

i in equation (A-1), i = {DS, EXT}) consists of three steps.

The contribution of the public R&D and extension knowledge stocks to the gross value of output (TVKi

t ) is calculated. This is done by multiplying the elasticities of TFP with respect to the R&D and extension knowledge stocks (βi

) by the total contribution of TFP to output value for the period after the investment has been made (T-t ) with t ≤ Li

R (which is equal to TFP growth divided by output growth, multiplied by the average value of output):

(G-1)

where TFP and OUTPUT are the growth rates of the TFP index and the output quantity index, AGVPT-t is the average gross value of output in the period after the investment is made, and T represents the time period under consideration.

Note that equation (G-1) is derived from the definition of TFP. TFP is defined as the ratio of output to input, such that TFP = OUTPUT/INPUT , which can be rearranged to give TFP + INPUT = OUTPUT . Therefore, the value of growth in output can be decomposed into the component attributable to growth in TFP and the component attributable to growth in input, such that

Since the intention is to estimate the return to an increase in public investment in R&D (or extension) (Ri), rather than the return to an increase in the knowledge stock, the second step is to convert the contribution of knowledge stocks to output value into the return to public R&D or extension investment. For a given R&D lag profile (gi(.) in equation A-1), the return to the investment in each year after the investment is made is estimated by multiplying the contribution of knowledge stocks to output value by the normalised series of weights

Estimation of internal rate of return to public investment in agricultural R&D and extension

G

.

59


The discounted return at time t + j to an investment made at time t (VMPit+j ) is given by:

(G-2)

where ri is the internal rate of return (IRR), the interest rate at which the flow of discounted benefits exactly offsets a one unit change in Ri.

Third, a change in public R&D investment in year t will add to the knowledge stock for LiR (in

this study either 16 or 35) years. The total discounted return, TVMPit , in year t should be equal

to the change in R&D investment in year t (Rit ), which can be used to estimate the IRR (ri ).

!"#$!! = !"#!!!!!!

!!!= !!! (G-3)

The IRR is a measure of the average benefits from an increase in Ri over a period of time, which can be used ex post, as a measure of returns achieved, and ex ante, to aid resource allocation. In this report, the IRR to public R&D and extension investments in both the short run and the long run were also estimated using the elasticities estimated from the error correction model.

Finally, similar to Mullen and Cox (1995), the term VMPit+j and Rt are assumed to be changing

over time. For simplicity, the practice in past studies of setting VMPit and Ri at their geometric

means throughout the whole period is followed.

60

ReferencesABARE 2009, Australian Commodity Statistics, Canberra.

Alston, JM 2002, ‘Spillovers’, Australian Journal of Agricultural and Resource Economics, vol. 46, no. 3, pp. 315-46.

Alston, JM, Andersen, M, James, JS and Pardey, PG 2010a, Persistence Pays: U.S. Agricultural Productivity Growth and the Benefits from Public R&D Spending, Springer, New York.

—— 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.

Alston, JM, Chan-Kang, C, Marra, MC, Pardey, PG and Wyatt, TJ 2000a, A Meta-Analysis of the Returns to Agricultural R&D: Ex Pede Herculem?, International Food Policy Research Institute, Washington D.C.

Alston, JM, Harris, MS, Mullen, JD and Pardey, PG 1999, ‘Agricultural R&D policy in Australia’, in JM Alston, PG Pardey & VH Smith (eds), Paying for Agricultural Productivity, IFPRI, Washington, D.C., pp. 118-71.

Alston, JM, Marra, MC, Pardey, PG and Wyatt, TJ 2000b, ‘Research returns redux: A meta-analysis of the returns to agricultural R&D’, Australian Journal of Agricultural and Resource Economics, vol. 44, no. 2, pp. 185-215.

Alston, JM and Pardey, PG 2001, ‘Attribution and other problems in assessing the returns to agricultural R&D’, Agricultural Economics, vol. 25, pp. 141-52.

—— 2006, ‘Developing-country perspectives on agricultural R&D: New pressures for self-reliance?’ in PG Pardey, JM Alston & RR Piggott (eds), Agricultural R&D in the Developing World: Too Little, Too Late?, International Food Policy Research Institute, Washington, DC, pp. 11-28.

Alston, JM, Pardey, PG, James, JS and Andersen, M 2009, ‘The economics of agricultural R&D’, Annual Review of Resource Economics, vol. 1, pp. 537-66.

Alston, JM, Pardey, PG and Ruttan, VW 2008, Research lags revisited: Concepts and evidence form U.S. agriculture, University of Minnesota, Department of Applied Economics Staff Papers, Staff Paper P08-14.

Australian Bureau of Statistics 2008, 2008 National Innovation Survey, Cat. No. 8158.0, Canberra.

—— 2010, Research and Experimental Development, All-Sector Summary, Australia, 2008-09, Cat. No. 8112.0, Canberra.

Binenbaum, E, Mullen, JD and Wang, T 2008, ‘Has the return on Australian public investment in agricultural research declined?’ paper presented to The Australian Agricultural and Resource Economics Society 2008 Conference, Canberra, Australia, 5-8 February, 2008.

Campbell, KO and Dumsday, R 1990, ‘Land policy’, in DB Williams (ed.), Agriculture in the Australian Economy, Sydney University Press, Sydney, pp. 162-71.

Coe, DT and Helpman, E 1995, ‘International R&D spillovers’, European Economic Review, vol. 39, pp. 859-87.

61


Commonwealth of Australia 2008, Australia 2020 Summit Final Report, Canberra.

Cox, TL, Mullen, JD and Hu, WS 1997, ‘Nonparametric measures of the impact of public research expenditures on Australian broadacre agriculture’, Australian Journal of Agricultural and Resource Economics, vol. 41, pp. 333-60.

Echeverria, RG 1990, ‘Assessing the impact of agricultural research’, in RG Escheverria (ed.), Methods for Diagnosing Research System Constraints and Assessing the Impact of Agricultural Research, International Service for National Agricultural Research, The Hague, Netherlands.

Evenson, RE 2001, ‘Economic impacts of agricultural research and extension’, in BL Gardner & GC Rausser (eds), Handbook of Agricultural Economics, Elsevier, Amsterdam, vol. 1A, pp. 573-628.

Fuglie, KO and Heisey, PW 2007, Economic Returns to Public Agricultural Research, Economic Brief No. 10, U.S. Department of Agriculture, Economic Research Service, Washington D.C.

Granger, CWJ and Newbold, P 1974, ‘Spurious regressions in econometrics’, Journal of Econometrics, vol. 2, no. 2, pp. 111-20.

Gray, EM, Jackson, T and Zhao, S 2011, Agricultural productivity: concepts, measurement and factors driving it, RIRDC Publication No. 10/161, Canberra.

Greene, WH 2007, Econometric Analysis, 6th edn, Prentice Hall, Upper Saddle River, New Jersey.

Griliches, Z 1979, ‘Issues in assessing the contribution of research and development to productivity growth’, The Bell Journal of Economics, vol. 10, no. 1, pp. 92-116.

—— 1998, ‘The Search for R&D Spillovers’, in Z Griliches (ed.), R&D Productivity: The Econometric Evidence, University of Chicago Press, Chicago, pp. 251-68.

Gutierrez, L and Gutierrez, MM 2003, ‘International R&D spillovers and productivity growth in the agricultural sector. A panel cointegration approach’, European Review of Agricultural Economics, vol. 30, no. 3, pp. 281-303.

Hall, J and Scobie, GM 2006, ‘The role of R&D in productivity growth: The case of agriculture in New Zealand: 1927 to 2001’, paper presented to New Zealand Treasury Working Paper 06/01.

Heisey, PW, Wang, SL and Fuglie, KO 2011, Public Agricultural Research Spending and Future U.S. Agricultural Productivity Growth: Scenarios for 2010-2050, EB-17, U.S. Department of Agriculture, Economic Research Service, July 2011.

Huffman, WE and Evenson, RE 1993, Science for Agriculture, Iowa State University Press, Ames, IA.

—— 2006, ‘Do formula or competitive grant funds have greater impacts on state agricultural productivity?’ American Journal of Agricultural Economics, vol. 88, no. 4, pp. 783-98.

Hughes, N, Lawson, K, Davidson, A, Jackson, T and Sheng, Y 2011, Productivity pathways: climate adjusted production frontiers for the Australian broadacre cropping industry, ABARES research report 11.5, Canberra.

Jarret, FG 1990, ‘Rural research organisation and policies’, in DB Williams (ed.), Agriculture in the Australian Economy, 3rd edn, Sydney University Press, Sydney, pp. 82-96.

Jessop, JE and Dun, RB 1982, ‘Organization and administration’, in DB Williams (ed.), Agriculture in the Australian Economy, 2nd edn, Sydney University Press, Sydney, pp. 106-24.


62

Makki, SS, Thraen, CS and Tweeten, LG 1999a, ‘Returns to American agricultural research: Results from a cointegration model’, Journal of Policy Modeling, vol. 21, no. 2, pp. 185-211.

Makki, SS, Tweeten, LG and Thraen, CS 1999b, ‘Investing in research and education versus commodity programs: Implications for agricultural productivity’, Journal of Productivity Analysis, vol. 12, no. 1, pp. 77-94.

Mullen, JD 2007, ‘Productivity growth and the returns from public investment in R&D in Australian broadacre agriculture’, Australian Journal of Agricultural and Resource Economics, vol. 51, no. 4, pp. 359–84.

—— 2010, ‘Trends in investment in agricultural R&D in Australia and its potential contribution to productivity’, Australasian Agribusiness Review, vol. 18, pp. 18-29.

Mullen, JD and Cox, TL 1995, ‘The returns from research in Australian broadacre agriculture’, Australian Journal of Agricultural Economics, vol. 39, no. 2, pp. 105-28.

Mullen, JD, Lee, K and Wrigley, S 1996, Agricultural production research expenditure in Australia: 1953-1994, NSW Agriculture, Orange.

Mullen, JD and Strappazzon, L 1996, ‘The relationship between research investment and productivity growth: Australian broadacre agriculture: 1953-94’, paper presented to Global Agricultural Science Policy for the Twenty-First Century, Melbourne, Australia, 26th August 1996.

Nossal, K and Lim, K 2011, Innovation and productivity in the Australian grains industry, ABARES research report 11.06, Canberra, July.

Nossal, K and Sheng, Y 2010, ‘Productivity growth: Trends, drivers and opportunities for broadacre and dairy industries’, Australian Commodities, vol. 10.1 March Quarter, pp. 216-30.

O’Donnell, C 2010, ‘Measuring and decomposing agricultural productivity and profitability change’, Australian Journal of Agricultural and Resource Economics, vol. 54, no. 4, pp. 527-60.

Pardey, PG, Roseboom, J and Craig, BJ 1999, ‘Agricultural R&D investments and impacts’, in JM Alston, PG Pardey & VH Smith (eds), Paying for Agricultural Productivity, IFPRI, Washington, D.C., pp. 31-68.

Parham, D 2004, ‘Sources of Australia’s productivity revival’, The Economic Record, vol. 80, no. 249, pp. 239-57.

Potgieter, AB, Hammer, GL and Butler, D 2002, ‘Spatial and temporal patterns in Australian wheat yield and their relationship with ENSO’, Australian Journal of Agricultural Research, vol. 53, no. 1, pp. 77-89.

Primary Industries Standing Committee 2011, Grains Industry: National Research, Development and Extension Strategy, Primary Industries Standing Committee—R&D Sub-Committee, http://www.daff.gov.au/agriculture-food/innovation/national-primary-industries/grains_industry (accessed 17 June 2011).

Productivity Commission 2007, Public Support for Science and Innovation, Research Report, Productivity Commission, Canberra.

—— 2008, Annual Report 2007-08, Productivity Commission, Canberra.

—— 2011, Rural Research and Development Corporations, Report No.52, Final Inquiry Report, Canberra.

63


Ramsey, JB 1969, ‘Tests for specification errors in classical linear least squares regression analysis’, Journal of the Royal Statistical Society B, vol. 32, pp. 350-71.

Shanks, S and Zheng, S 2006, Econometric modelling of R&D and Australia’s productivity, Productivity Commission Staff Working Paper, Canberra.

Sheng, Y, Mullen, JD and Zhao, S 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, Canberra, May.

Sunding, D and Zilberman, D 2001, ‘The agricultural innovation process: Research and technology adoption in a changing agricultural sector’, in BL Gardner & GC Rausser (eds), Handbook of Agricultural Economics, Elsevier, Amsterdam, vol. 1A, pp. 207-61.

Zhao, S, Sheng, Y and Gray, EM 2010, ‘Productivity of the Australian broadacre and dairy industries’, paper presented to Conference on the Causes and Consequences of Global Agricultural Productivity Growth, Washington D.C., 11-12 May 2010, 11-12 May, 2010.

ReseaRch funding ABARES relies on financial support from external organ isations to complete its research program. As at the date of this publication, the following organisations had provided financial support for Bureau research in 2009–10, 2010–11 and 2011–12. We gratefully acknowledge this assistance.

AusAID

Australia Indonesia Governance Research Partnership (ANU)

Australian Centre of Excellence for Risk Analysis

Australian Competition & Consumer Commission

Australian Fisheries Management Authority

Australian Government Department of Innovation, Industry, Science and Research

Australian Government Department of Climate Change and Energy Efficiency

Australian Government Department of Resources, Energy and Tourism

Australian Government Department of Sustainability, Environment, Water, Population and Communities

Australian Government Department of the Treasury

Australian National University

Bureau of Resource and Energy Economics

Cooperative Research Centre for National Plant Biosecurity

CSIRO

Dairy Australia

Department of Primary Industries, Parks, Water and Environment Tasmania

Ensis (joint venture between the CSIRO (Aust) and Scion (NZ))

Environment ACT

Environment Hydrology Associates

Fisheries Research and Development Corporation

Food Standards Australia New Zealand

Forest & Wood Products Australia

Goulburn-Murray Water

Grains Research and Development Corporation

Grape and Wine Research and Development Corporation

Horticulture Australia Limited

Industry & Investment NSW

Inovact Consulting Pty Ltd

Korea Rural Economics Institute

Meat & Livestock Australia

Murray–Darling Basin Authority

National Water Commission

New Zealand Institute of Veterinary, Animal and Biomedical Sciences

Office of Northern Australia

Peel Harvey Catchment Council

Plant Health Australia

Queensland Competition Authority

Queensland Department of Employment, Economic Development and Innovation

Rural Industries Research and Development Corporation

Sinclair Knight Mertz

South Australian Research & Development Institute

Southern Cross University

University of Melbourne

Western Australian Agriculture Authority

public investment in agricultural r&d and extension: an ......yu sheng, emily m gray, john d...

Documents