information to users - mcgill universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer...

165
INFORMATION TO USERS This manuscript has been reproduced trom the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be tram any type of computer printer. The quallty of thl. Npracluctlon 1. .pendent upon the quallty of the copy submitted. Broken or indistinct prim, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction. ln the unlikely event that the author did not send UMI a complete manuscript and there are missin; pages, these will be noted. AllO, if unauthorized copyright material had ta be removed, a note will indicate the deletion. Oversize materials ( •.g., maps. drawings, charts) are reproduced by sectioning the original, beginning st the upper Ieft-hand corner and continuing trom 18ft ta right in equal sections with small overtapa. Photographs induded in the original manuscript have been reproduoed xerographically in thi. copy. Higher quality e- x 9- black and white photographie prints are available for any photographs or illustrations appearing in this capy for an additional charge. Contad UMI directly to arder. Bell & Howell InformatiOn and Leaming 300 North Z8eb Road, Ann Arbor, MI 48108-1346 USA 800-521-0800

Upload: phunghanh

Post on 01-Feb-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

INFORMATION TO USERS

This manuscript has been reproduced trom the microfilm master. UMI films

the text directly from the original or copy submitted. Thus, some thesis and

dissertation copies are in typewriter face, while others may be tram any type of

computer printer.

The quallty of thl. Npracluctlon 1. .pendent upon the quallty of the

copy submitted. Broken or indistinct prim, colored or poor quality illustrations

and photographs, print bleedthrough, substandard margins, and improper

alignment can adversely affect reproduction.

ln the unlikely event that the author did not send UMI a complete manuscript

and there are missin; pages, these will be noted. AllO, if unauthorized

copyright material had ta be removed, a note will indicate the deletion.

Oversize materials (•.g., maps. drawings, charts) are reproduced by

sectioning the original, beginning st the upper Ieft-hand corner and continuing

trom 18ft ta right in equal sections with small overtapa.

Photographs induded in the original manuscript have been reproduœd

xerographically in thi. copy. Higher quality e- x 9- black and white

photographie prints are available for any photographs or illustrations appearing

in this capy for an additional charge. Contad UMI directly to arder.

Bell & Howell InformatiOn and Leaming300 North Z8eb Road, Ann Arbor, MI 48108-1346 USA

800-521-0800

UMI~

Page 2: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure
Page 3: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Anthony Q.Q. Aboagye

. Finaneial Flows, Macroeconomie Poliey andThe Agricultural sector in Sub-Saharan Africa

Faeulty of Management

Ph.D

Page 4: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

National Ubraryof Canada

Acquisitions andBibliographie Services

39S wellington SI,eet0Itawa ON K1A 0N4Canada

BibliothèQue nationaledu Canada

Acquisitions etservices bibliographiques

395. rue WellingtonOttawa ON K1A 0N4C8nadII

The author bas granted a non­exclusive licence allowing theNational Library ofCanada toreproduce, loan, distribute or seOcopies of this thesis in microform,paper or electronic formats.

The autbor retains ownership of thecopyright in this thesis. Neither thethesis nor substantial extracts from itmay he printed or otherwisereproduced without the author'spemusslon.

L'auteur a accordé une licence nonexclusive permettant à laBibliothèque nationale du Canada dereproduire, prêter, distribuer ouvendre des copies de cette thèse sousla fonne de microfiche/film, dereproduction sur papier ou sur formatélectronique.

L'auteur conserve la propriété dudroit d'auteur qui protège cette thèse.Ni la thèse ni des extraits substantielsde celle-ci ne doivent être imprimésou autrement reproduits sans sonautorisation.

0-612-50100-0

Canadrl

Page 5: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

ACKNOWLEDGEMENTS

1am extremely grateful to all the members of my thesis committee for their willingness tobe on my committee even though 1had not worked closely with sorne of them previously.The committee comprises F. Grimard, K. Gunjal, (co-supervisor), J. Jorgensen, J-O.Simonato and G. A. Whitmore (co-supervisor). Members ofmy committee went beyondthe cali ofduty in providing helpful suggestions and advice, in being generous with theirtime and for their words of encouragement and challenging me to do even better. 1amvery thankful for that. 1am particularly grateful to Professor Whitmore for providing ashoulùer to lean on dwing the later half ofmy studies here.

[ also wish to express my appreciation to the University of Ghana, the Govenunent ofGhana, MeGill University and Prof. Whitmore for providing me financial and other forrosof support at various stages ofmy studies.

[ have as weIl henefited from my association with several people, both staffand studentshere in McGiIl and elsewhere including B. Campbell, M. Dagenais and J-C. Duan. andwish to thank themall.Ialso wish to thank my external examiner, Fidel Ezeala-Harrisonfor his useful comments.

Final1y, 1wish to dedicate this thesis to my parents, Yao and Victoria Aboagye, Foanor.Lynette, and the rest of the family for unwavering support and ail the lime we spent apart.

Somelhing has broken down in Africa - throwing billions ofdol/ars will nof changeanything. Africa must learn ta became wha/e again ... We are not just wulking digestivetracts. Africa ;s more than a client jôr international soup ki/chens. We must build ourself.l'om the ground up ... infrastructure, logistics and superstructure, including the bra;n.

Kizerbo (1989)

Page 6: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

ii

ABSTRACT

This thesis focuses on the effects ofdevelopment assistance (ODA), private foreigncommercial capital (PFX), domestic savings (SAV), the openness of the economy andproducer priees on agrieultural output, and on export and domestic shares ofagrieulturaloutput in sub-Saharan Afiiea (SSA). This study uses panel data spanning 27 eountriesand the period 1970 ta 1993.

The production function is a Cobb-Douglas type. Statie export and domestic shareequations are derived from a specification of the agricultural gross domestie productfunclion. Transformed auto-regressive distributed-Iag versions of the slatie share modelsare used to investigate long-run dynamics, persistence and implementation lags in theshare response mode!.

Agrieultural output is affected as follows. OOA, PFX and SAV have small positive ornegative impact depending on agrieultural region or economic poliey environment. Theimpact ofopenness of the economy is negative in ail agricultural regions, however, thereis evidence of positive effect ofopenness within improved policy environment. None ofthese effects are statistically significant.

Export share is affeeted as follows. OOA, PFX and SAV have small positive impact insorne agrieultural regions and poliey environments, both in the short-run and in the long­run. PFX is not signifieant anywhere. ODA is significant only when eountries aregrouped by poliey environment in the short-run. SAV is significant in the short-run onlyin sorne regions, and significant in the long-run only in others. Openness has positiveimpact in the short-run. This is significant in many regions. hs long-run impact is mostlypositive but not significant anywhere. The impact ofprodueer price is mostly positive butnot significant.

Efforts to encowage economic activities in rural communities such as improvements indomestic terms of trade in favor of agriculture, together with the provision ofinfrastructure are likely to stimuJate output. Strategies to diversify and processagricultural exports in the face of falling agricultural commodity prices should bepursued.

Page 7: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

iii

RÉsUMÉ

Cette thèse s'intéresse aux effets de l'aide au développement (A1D), à ceux du capitalprivé, étranger et commercial (CPE), à ceux de l'épargne des ménages (EM), à ceux del'ouverture de l'économie et à ceux des "prix de producteurs" sur le produit agricole ainsique sur les parts nationales et à l'exportation du produit agricole en Afriquesubsaharienne. Cette étude utilise des données en panel allant de 1970 à 1993 et portantsur vingt sept pays.

Une fonction de production du type Cobb..Douglas est utilisée. Les équations statiquesd~s parts nationales et à l'exportation sont dérivées à panir d'une spécification de tafonction du produit agricole national brut. Des versions transformées autorégressives àretards distribués des "modèles statistiques pour les parts" sont utilisées. Cela permetd'étudier les dynamiques à long terme, les effets de persistance ainsi que l'implementationdes retards dans les réponses des "modèles pour les parts".

Le produit agricole est affectée comme suit. L'(A1D), le (CPE) et 1'(EM) ont un faibleeffet, positif ou négatif. Cela dépend de la région agricole ou de l'économiqueenvironnementale. L'impact de l'ouverture de l'économie est negatif dans les régionsagricoles. Pourtant, un effet positif de l'ouverture de l'économie en moins del'amélioration de l'économique environnementale a identifié. Aucun de ces effets n'eststatistiquement significatif.

La part à l'exportation est affectée comme suit. L'(AID), le (CPE) et 1'(EM) ont un faibleeffet positif dans quelques-uns régions agricoles et dans quelques-uns économiquesenvironnementales, à la fois à court et à long terme. Le (CPE) n'est pas significatif.L'(A1D) est significatifquand les pays sont régroupés par l'économiqueenvironnementale. L'effet de l'épargne des ménages est significatifde court terme danscertaines regions et de long tenne dans d'autres. L'ouverture de l'économie a un effetpositif à court terme. Cet effet est significatif dans plusieurs régions. Son effet à longterme est surtout positif sans être partout significatif. L'effet des "prix de producteurs" estpresque toujours positif mais non significatif.

Les efforts à encourager les activités économiques dans les communautés rural sont denature à favoriser la croissance du produit agricole. Il s'agirait notamment d'améliorer lesechanges commerciaux intérieurs en faveur de l'agriculture ainsi qu'un meilleurapprovisionnement en infrastructure. Par ailleurs et face à la baisse des cours mondiauxdes produits agricoles, des stratégies visaut à diversifier et à transformer les exportationde produits agricoles doiveut être poursuivies.

Page 8: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

TABLE OF CONTENTS

Chapter 1: INTRODUCTION

Chapter 2: LITERATURE REVIEW 12Production Function Studies 12Short..Run supply Response Studies 13Long-Run Supply Response Studies 15Other Related Studies 15

Chapter 3: ECONOMIC THEORY AND METHODOLOGY 17Panel Data 17Economie Growth Theory 18Rational for Decomposing Capital 23Econometrie Specification .. Production function 25Econometrie Specification .. Analysis ofagricultural shareresponse in the short..run 28Econometrie Specification .. Dynamics ofagricultural shareresponse 34Econometrie and Other Related Issues 44

CHAPTER 4: DATA 48Data Description and Sources 48

CHAPTER 5: EMPIRICAL RESULTS .. PRODUCTION FUNCnON 54Regression Results 55Human Capital Variable 56Diagnostic Tests of Regression Adequacy 58Agro-Climatic Regions 59Policy Groups 72Swrumary 82

CHAPTER 6: EMPIRICAL RESULTS - ANALYSIS OF AGRICULTURALSHARE RESPONSE IN THE SHORT-RUN 85Regression Results 86Diagnostic Tests of Regression Adequacy 86Agro-Climatic Regions 87Policy Groups 98Summary 106

iv

Page 9: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

CHAPTER 7: EMPIRICAL RESULTS - DYNAMICS OF AGRICULTURALSHARE RESPONSE 110Estimation 110Results and Discussion 112Diagnostic Tests of Regression Adequacy 112Parameter Estimates 116Summary ofDynamic Share Response 130Comparison of Long.Run and Short·Run Estimates 133

CHAPTER 8: CONCLUSIONS AND POLICY IMPLICATIONS 137Policy Implications 137Increasing Agricultural Output 138Increasing Export Earnings 142Limitations of this Study 145

REFERENCES 147

APPENDIX 155

v

Page 10: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 1

Introduction

This thesis examines the impact of foreign financial fiows, domestic savings, labor

as well as the role of the macroeconomic environment, external terms of trade and

variability of the weather on the agricultural sector of sub-Saharan Africa (SSA).

Importance of Agriculture in SSA. In SSA, agriculture is practised mostly in

the nrraI areas, where it is the mainstay of most of the rllral people. Agriculture has

been described as the "lifebIood" and the "engine of growth" of sub-Saharan Africa.

In a nutshell, this refers to its importance as a source food, income, foreign exchange

earnings, employment, etc. for many Africans. The importance of agriclùture in SSA

can be seen from the fact that agriclÙtural share of overall gross domestic product

(GDP) in 1993 was about one-third (World Bank, 1995). Further, The Food and

Agricultl1ral Organization of the United Nations (1996) estimates that about two­

thircls of the labor force in SSA was engaged in agric\.Ùture in 1995. Foreign exchange

earnings froID 8gricultural exports account for very high proportions of total foreign

1

Page 11: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

2

exchange earnings of many COlmtrïes. The following table illustrates the proportions

for four countries.

Table 1.1: Agriclùtural exports as a percentage of total merehandise exports

Year 1980 1985 1990 1993

Burkina Faso 50 37 -13 32

Cote d'Ivoire 67 78 53 50

Ghana 68 64 46 32

Central Mrican Republic 42 50 30 24

l\Iali 94 90 75 75

Kenya 55 73 68 51

l\Ialawi 89 93 94 89

Rwanda 49 61 92 88

Tanzania 70 69 59 89

Source: Computed from World Bank (1996a).

Unfortunately, this sector has not received sufficient attention. Helleiner (1992),

among others, has argued that ''probably the single most important poliey mistake

in the 1960s and 1970s was the neglect of agriclÙture," in SSA. Furthermore, the

World Bank (1989) reports that in spite of the contribution of agriculture to these

economies, the proportion of government expenditure devoted to agriclùture during

the early 1980s was less than 10 percent on average. By and large, investment in

research and development, extension services, or infrastructure has been inadequate.

Page 12: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

3

Indeed, agriclùture in SSA is characterized by the following.

Agricultural Systems and Rural Communities: Agriculture in SSA is dom­

inated hy peasant farmers using ooly basic fanning implements. There is heavy

relianee on nature (for rain and sail fertility). Further, only minimal use is made of

inputs snch as high yiclding sccds and fcrtilizcr. Storage facilities are poor, leaJiug

ta waste during harvest seasons and shortages during leau seasons. AIso, the terms

of trade between agriclùtllral output and the rest of the economy are largely biased

against agriclùture due mostly to lùgh direct and indirect taxes on agrÏclùtnral out­

put. In addition, agriclùtural output faces clifficlÙt access ta markets due ta poor

infrastructure snch as roOOs. Institutional support in the form of access to extension

services or financiaI credit are aIso very weak. 1

As a consequence, agriclùtural output has suffered. and with it the welfare of t.he

people. most of whom live in the nlfaI areas where agriclùture has been and still is the

mainstay. AEriea Recovery (1997) reported that up to 40 percent of the people of SSA

are technically lmder-nourished. This is due in part to the fact that food production

has Dot kept up with poptùation growth. It adds that over the period 1965 to 1993

agricultural output grew at an average rate of 1.8 percent while population grew at

about 2.9 percent. Rodrik (1997) adds that, per capita real gross domestic prodnct

(GDP) in most of SSA is currently below the 1970 levelon average. It is people who

live in nlral areas who hear the brunt of this reduetion.

Rural non-agrictùtural activities in SSA are aIso at a very low level due to the

1 Base<! on authorls own knowledge, but see aIso UNCTAD (1997).

Page 13: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

4

direct link between them and agriclÙture. Haggblade, Hazell and Brown (1987) have

documented that the link between agricultural and non-agricultural activities in SSA

is lower than it is in Asia or Latin America. This weak linkage is attributed to low nse

of inputs, low output and low growth in agriculture. Typically, the linkage between

agriclùture and nrraI non-agricultlrral activities takes three forros. Increased use of

inputs in agrictùturc crcatcs a buch."\vard linkage bctween agrictùture and the suppliers

of these inputs, Second, increased output (due at least in part to more inputs) trans­

lates into higher spending on agriclÙtural and nOll-agriclùtural activities by farmers

(constunption linkages). Consumption linkages are closely tied to the third fonn of

linkages, the forward linkages. These are the links between agrictùtural output and

non-agriclùtural activities that process agrictùtnral ontput (for local consumption),

Consequences of strong linkages will include a halt to rural-tO-urban migration, as

non-land owners and others find non-agrictùtlrral employment. Also, most of these

rural non-agricultural activities are labor intensive, they thus have the potential to

rednce twemployment. Being labor intensive, nrral industries generally require little

investrnent in capital or foreign exchange, thus, being potentially easier and cheaper

to establish. Processing of agricultural output also help improve the terms of trade

between the nrral areas and the urban centers. The importance of the potentiallink­

ages that agriclùture can provide may he obtained frOID the following statement in

Africa Recovery (1998) in reference to SSA economies.

"One stnlctural concem is that higher growth achieved in sorne cOlwtries

in recent years is to sorne extent driven byenergy and mining sectors.

Page 14: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•5

These are relatively isolated from the rest of the economy and thus have

little direct impact on living standards of the majority of the poptùation."

Indeed, the World Bank has recognized the importance of rural economies in de-

veloping cOlmtries and has developed a nrraI sector strategy.2 It is based on the

realization that L'Rlrral welfare cannot improve without a thriving nrraI economy of

which a prosperons agriCtÙtlrre is a necessary condition." Indeed, Haggblade et al.

(1987) present sorne evidence showing that expenditure elasticity for locally produeed

non-farm goods is high. The World Bank's strategy calls for L'public investment to

complement the private sector." This author agrees with the Bank's analysis regard-

ing the importance of agriculture in nrraI economies. Further, this thesis takes the

view that the weak linkages that are cllrrently suggested to exist in SSA provide the

potential for grawth. The extent of these linkages shall be econametrically established

in this study by focusing on the role of capital in agriclùture in SSA.

l'vIany govermnents that did not pay much attention ta agricultlue in the past

reasoned that industrializatian was the way forward. They went about their indus-

trialization efforts by transferring resoluces away from agriclùture. In the process.

they practically killed the LLgeese that laid the golden eggs". The faet of the matter is

that, in addition to its raIe in providing employment, incarne and foreign exchange,

attention ta agriclùture will advance industrialization efforts by providing the raw

materials for industry. Farmers and their dependents will be ready markets for in-

dustrial output, while savings made out of agriclùtural incorne will he available for

2 14Revitalizing the World Bank's Approach to Rural Development." Talle presented at the McGillEconomie Policy ~Ianagement Seminar Series by A. l'tIcCalla, (1998).

Page 15: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

6

investment in industry. (See for example, Ezeala-Harrison, 1996).

Financial Flows and Macroeconomie Policy. A major objective of this study

is to investigate the impact of the factors of production on agricultural output to

understand how agricultural output may he increased. In particular, tms study fo­

cuses on the role of capital and macroeconomic policy. For SSA: inv~tmflnt. rapit.al

may come from domestic savings, or foreign sources. Foreign finandal flows to SSA

take two forros, official development assistance (ODA) and private foreign commercial

flows (PFX).

The term official development assistance generally refers to resources that are

provided on concessional terms. That is, terms less stringent than will be obtained

in capital markets. The Organization for Economk Co-operatioll and Development

(OECD, 1994) prescrihes three criteria (at least one of which mnst be satisfied) for

a flow of resonrces to qualify as ünA. These are, (i) resollrces are provided by offi­

cial agencies, induding state and local governments or their execlltive agencies. (ii)

resources are provided with the promotion of economic development and welfare as

the main objective, or, (iii) resources are provided on concessional terms, and as well

convey a grant element of at least twenty-five per cent.

In recent years, the importance of development assistance in providing funds to

SSA has increased, as domestic savings and private foreign flows have fallen. OECn

(1994) reports that in 1992, 31.1%, of all ODA funds went to SSA, a slight decline from

the 1990 level. Unfortunately, while the proportion of total ODA to SSA has remained

high, total ODA to developing countries has been declining in real terms since 1990,

Page 16: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

7

(World Bank, 1997). For example, total aDA in 1995 declined by 10 percent in real

terms over the level in 1994 (World Bank, 1997). Another important aspect of aDA

is described by the same source as "... the composition of ODA has shifted, with a

significant portion being used to fund emergency relief and peacekeeping activities

and less going toward long-term development needs."

PFX arc dcfincd to include foreign direct investnlellt, ~Onllllel'dalloalls fl'uru for­

eign banks and other sources, as weIl as portfolio eqlùty investments in financial and

eqlùty nlarkets. The amount of PFX going to the region has fallen sllbstantially

from the peak of an annnai average of 8.9% of all total private flows to developing

c01mtries dllring 1977 ta 1982, to an annual average of 1.6% between 1990 and 1995,

(Bhattacharya, ~Iontiel and Sharma, 1997).

Over the period 1970 to 1993, domestic savings for a sampIe of 27 countries (llsed in

this stndy) averaged jllSt Imder Il % of GDP. Wide variations exist among the cOlln­

tries of SSA. Table 1.2 presents a SllInmary of actual üDA, PFX and SAY amolmts

in millions of constant 1987 United States dollars (US$).

Table 1.2. Average amotmts of ODA, PFX and SAY in 1987 US$ in SSA.

Period aDA PFX SAY

1975-79 5,616 1,860 38,283

1980-84 6,801 1,734 20,032

1985-89 8,184 1,048 17,783

1990-93 10,147 1,520 14,208

Source: Computed from data on hand. See chapter 4.

Page 17: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

8

As can be seen from the table, domestic savings are indeed the largest source of

investment nmds, but these amounts have been falling steadily. aDA funds have in­

creased substantial1y to a sizeable proportion. But as noted, sinee 1990 these amounts

have been falling tao. PFX amolmts are small eomparatively.

[t is argued in this thesis that the three eomponents of capital will have different

cffccts on agrictùturc. A dollar of ODA flaw is providcd on humanitarian and ethical

grollnds and is development oriented. On the other hand, the decision for a dollar

of PFX flow is made on a risk and retum hasis, and is clearly trade oriented (World

Bank, 1997). Thus, ODA flows are unlikely to flow into the same projects as PFX.

Domestic savings are likely to he invested in areas of the eeonomy that are different

from those into whieh foreign flows are invested. ~Iore about this in chapter three.

Another issue investigated in this thesis is the relationship between the agrkultural

seetor and macroeconomie poliey. ~Iacroeconomie poliey issues include, tight control

of fiscal budgets, tight monetary poliey and liberalization of international trade and

foreign exchange markets. These measures are snpposed to lead ta more efficient nse

of resourees, henee inereased economic growth. This thesis investigates the impact

of poliey at two levels. First, by including a poliey variable that changes with time.

Second, by considering groups of cOlmtries classified by changes in macroeconomic

policy.

A further objective of this thesis is to investigate the response of the domestic and

export shares of agriclùture to producer priee changes, factor inputs, economic poliey

and external shocks. This is done for severa! reasoDS. For one thing, the value of

agricwtural exports as a share of total agricwtural output has been falling. In 1993,

Page 18: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

9

the average export share was 25 percent, having been as high as 50 per cent in the

late 1970s.

Falling export share of agrictùturaI output has been happening in the face of

anecdotal evidence that foreign finandaI flows, recent economic recovery programs

and government polides favor export agriclùture relatively more than agriclùture for

domcstic constunption. The apparent bias in favor of exports takes the form of higher

producer priees for exports, availability and distribution of inputs and infrastructure

in areas that grow export crops (or rear animaIs for export). As well, research and

development efforts favor agriclùturaI exports more. Indeed, falling export share of

agriclùture is happening in the face of indications of food security problems in sorne

countries.

Generally, crops grown for export are called "cash crops", enlphasizing the fact

that these are the crops that have been grown traditionally for incorne. Support

services sucb as marketing boards that existed in the past catered principal1y to export

produce. By providing ready markets for these produce, the boards created huge

incentives to farmers of snch prodnce who no longer had to contend with marketing

their produce in the face of poor raad netwarks, etc.

Objectives of This Study. In sum, the objectives of this thesis are the following.

In the first place, to investigate the impact of domestic savings, development assis­

tance, private foreign commercial capital, macroeconomic policy and extemal shocks

on aggregate agricultural output using panel data for SSA countries spanning the pe­

riad 1970 to 1993. Then the response of the export and domestic shares of aggregate

Page 19: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

10

agricultural output to these factors as weil as producer priees are investigated, first in

a static model, and then in a dynamic model for short-nm and long-nm implications.

The analysis is performed when data from (i) countries from aIl parts of the continent

are pooled together, (ii) COtmtries are grouped together according ta agra-climatic re­

gions, and (Hi) in the case of agrictùtural output and static share investigations, when

~olmtries are grouped by improvements in economic policy.

Contribution of This Study. This study makes contribution to the literatllre

in severa! respects. By disaggregating investments into three eomponents: official

development assistance, private foreign cornmerdal flows and domestic savings, the

impact of each fonn of capital on total agriclùtllral output., and on export a.nd da­

mestic shares of total agriclùtural output in the short-run and in the long-nm can be

investigated.

This study also contributes to lmderstanding the impact of macroecononlÎc policy

and extemal shocks on agriclùtural output as weIl as on export and domestk shares

of agriclùtural output in SSA.

Another contribution of tms study is adoption of an approach that allows equaI

study of bath agriclùture for exports and agriclùture for domestic cOIlSlunption.

Studying agriculture for domestic consumption has posed sorne challenge in the past,

(see Jaeger (1992) for example). In studying farmers' response ta agriclùtural ex­

ports and agriculture for domestic consumption, focus is not on prices only, but also

on factors of production and economic policy.

Yet a further contribution of this study is to go beyond individual cotmtry case

Page 20: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Il

studies (sucb as Bautista and Valdes (1993) and Krueger, Schifr and Valdes, 1991)

to provide econometric evidence that holds across regions and cOlmtries over time by

pooling cross-colmtry and time series data. More on this in chapter three.

Organization of the Study. This thesis is organized as follows. Chapter two

reviews pertinent literature on agrictùtural production nUlctions, short-run and long­

nm agrictùtural supply response, and other related studies. Chapter three develops

the econornic and econometric basis for the study, while chapter four discusses the

data used in the study and their sources. In chapters five, six and seven, empirical

estimates of the parameters of the meta-production flUlction, the static and dynamic

share respanse models respectively, are presented and discussed. Chapter eight con­

clncles by painting out policy implications of this stlldy as weB as its liuùtations.

Page 21: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 2

Literature Review

The introduction to this thesis has discussed the importance of agriculture as well

as agriclùtural systems in SSA. This chapter reviews selected previons studies that

have pertinent bearing on the CUITent one. It starts with past studies on production

flIDctions, and then discusses studies on agriclùtural supply responses.

2.1 Production Functions Studies

Lan and Yotopowos (1989) formalized the concept of a common agricll1tural produc­

tion flIDction for severa! COlIDtries. This is called a meta-production nmction. They

explained that the lmderlying assumption is that these countries have access to the

same production techn01ogy. Different cOlIDtries may however operate on clifferent

parts of this production flIDction dne to the nature of each C01IDtry'S endowment of

natura! resources, relative prices of inputs, macroeconomic environment, etc. They

used panel data for 43 cOlIDtries ta estimate Cobb-Douglas and transcendentalloga­

rithmic production functions. Their sample included both developed and developing

cOlmtries.

Norton, Ortz and Pardey (1992) investigated the impact of aDA on agricultural

12

Page 22: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

13

output in 98 developing countries of the world. They postulated. and estimated a

polynomiallag structure for annual ODA flows. They fOlmd the coefficient of ODA

to he positive in sorne specifications of their production flUlctions, and negative in

others (depending on the variables in the model).

Kherallah, Begrun, Peterson and Ruppel (1994) also investigated the impact of

ODA on agricultural growth in a simlùtancous cquations model for 56 developing

COIUltries. They also considered the impact of PFX and SAY. However, while t.hey

specified a polynomial lag structure for GDA, they only considered cnrrent annnal

flows of PFX and SAY. They fOlUld the impact of ODA and SAVon growth of

agriclùtural output to he positive, but that of PFX to be negative.

Gllnjal and Gichenje (1997) investigated the short-nUl and long-niD impacts of

ODA on agrictùtural out.put in SSA. They controlled for the effect. of economic policy

environment by llsing a dummy variable to identify those eOllntries that were lUlder­

going economic reforms and those that were not. They fOlmd the impact of OOA

in the aggregate SSA sample ta be significant and positive. Depending on t.he Sllb­

sample of cotmtries used, they found the impact of ODA to be positive or negative.

They did Dot, however, recognize other forms of capital snch as PFX or SAY in their

models.

2.2 Short-Run Supply Response Studies

Binswanger, Yang, Bowers and l\'Iundlak (1987) investigated aggregate short-run sup­

ply response of agriculture to priee and various public inputs using panel data for 58

countries. Their study eovered the period 1969 to 1978. They found that the main

Page 23: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

14

determinants of output are variables representing infrastructure and that the efIect

of priee was relatively weak.

Jaeger (1992) studied the short-run response of total agrieultural exports and

individual export crops to real producer priees, real effective exchange rates, weather

and disaster variables. He used panel data for 21 cOtmtries of SSA covering the period

1970 to 1987. He aIso examined the possibility that export agrieulture may crowd

out food production. He fOlmd that in the short-nm, elasticities of tree crop exports

are only moderately responsive ta price incentives. Annnal crops exports are more

elastic. He also fOl.md that "growth in export agrietùture does Ilot appear to come

at the expense of food production". He added that there was evidence to suggest

that poor policies have had a major role in the decline of African agriculture, but

cOlnmented that lack of price data for food crops prevented him from studying the

impact of poliey on food production the way he did for agricultural exports.

Elmi (1994) estimated short-nm and long-run agrictùtural export and food crop

output to real producer priees, weather, fertilizer and a lagged dependent variable.

His data consisted of 20 SSA cOlmtries, and covered the period 1974 ta 1989. He round

that "aggregate agrictùtural export and food sllpply responses ta priees in tropical

Africa are both positive and significant but inelastic".

SchifI and ~Iontenegro (1997) surveyed the literature on sllpply response and

provide sorne evidenee in support of the complementarity between producer priees and

public goods (infrastructure, supporting services, legal and institutional framework)

using data for 18 countries of the world of whieh three are in sub-Saharan Africa.

Page 24: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

15

2.3 Long-Run Supply Response Studies

The literatllre on long-run response is limited. Bond (1983) estimated aggregate and

individual crop long-nm and short-run supply response to producer prices and lagged

dependent variables for individual SSA cOlmtries. The average value of her long-nID

estimates is sometimes quoted as the long-run value for SSA.

As noted above, Elmi (1994) too has estimated long-nm supply response for SSA.

He found that long-nm response to producer priee increases exceeds short-nm re-

sponse.

2.4 Other Related Studies

Other relate<! studies include Leie (1990). who undertook a \Vorld Bank sponsored

individual country case studies on the impact of aOA on agriclùtl1ral development

in 6 sl1b-Saharan African cOlmtries. She fOlmd the impact to he "surprisingly sInall" .

Knleger, Schiff and Valdes (1991) edited a World Bank sponsored comparative

study (individl1al cOlmtry case studies) that assesses the impact of direct and indirect

intervention in agriclùtural priees in 18 cOlmtries. Titree of these are SSA countries.

They found the effect of direct intervention (direct taxation) on all selected products

(inc1uding 'exportables') to he negative. This effect was however dominated by the

effect of indirect intervention (industrial protection poliey and over-valuation of real

exchange rate).

Similarly, Bautista and Valdes (1993) also edited a volume that investigated the

effect of trade and macroeconomic policies on agriclùture in 18 eountries (not identical

Page 25: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

16

to the Knleger et al. sample}. Two ofthese cOlUltries are in sub-Saharan Africa. They

also conclude that restrictions on trade, foreign exchange rates, direct and indirect

taxes constitute bias against agric\ùture.

One realizes from this review that, none of these studies has explicitly investigated

the impacts of the factors of production. This study attempts to fill this gap in the

literaturc. In particular, physical capitul is disaggrcgatcd iuto aOA, PFX and SAV.

In addition. tlùs study investigates the effect of the macroeconomic environment at

two levels, (i) by using a variable that uses the openness of the economy as a ffieasllfe

of the macroeconoic environment, a variable that has received empirical support in

recent years (Sachs and \Varner, 1995 and Edwards~ 1998), as weil as pooling cOlmtry

data using on classifications based on improvements in economie policy. Further, the

methodology adopted here enables equal investigation ofhoth agriclùtllrc for domestic

consumption, as weIl as agrÎClùture for export.

Page 26: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 3

Economie Theory andMethodology

This chapter discusses the methodology and economic theory underlying this stlldy.

It discusses panel data, economic growth theory. disaggregation of capital. and the

three models that are investigated in this study.

3.1 Panel Data

This study pools crosS-ColUltry and time series data for severa! SSA COlmtrïes. AlI

cOlmtries with total poplùation of at least one million and for which pertinent data

are available are included in the study. The study covers the period 1970 to 1993.

The meta-production nmction is estimated llsing data for 27 countries. The studies

on export and domestic shares of agriculture eonsist of data for 23 eOlmtries. The

eonstraint was produeer priee data.

The advantages of using panel data are many and are discussed in Baltagi (1995),

Hsiao (1986) and Lau and Yotopoulos (1989) among others. They include increased

number of observations, inereased ranges of variation of the variables in the model,

thereby allowing for more precise estimates, and reduced multicollinearity among

17

Page 27: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

18

explanatory variables. In addition, use of panel data makes it possible to differentiate

between economies of scale and technical change, as weIl as the ability to study

dynamic effects (which cannot he done using cross-sectional data ooly). Thus, use

of panel data in this thesis will enable one to make valid inferences about the state

of affairs in the region beyond what can be done using only individual cOlmtry case

study dat.a.

However, when one pools data across many cOlmtries, one must recognize t.he

potential of differences in definition, measurements and even qualities of input.s across

COIUltries. There is also the question of differences in economic environment across

COlUltries. In a.ddition, one should be careflù not to carry the assumption of a commOll

production technology too far. The latter issue may be addressed by statistical tests

however. This is discussed later in sub-section 3.7.3.

3.2 Economie Growth Theory

The models employed in this study use ideas from both neoclassical and endogenous

growth theories.

3.2.1 Neoclassical and Endogenous Growth theories

Starting from the theoretical production function stated in terms of stocks of physical

capital and labor, the neoclassical growth theory of Solow (1956), for example, is

derived by considering steady state values of capital per effective unit of labor. Neo­

classical growth theory emphasizes the role of factor accUIDlùation in the production

process. However, with land fixed for example, increases in variable factors such as

Page 28: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

19

Labor eventually Lead to decline in marginal productivity of Labor. In Solow's version of

the neo-classical growth model, per capita incorne increases with the rate of saving in

the economy, but decreases with the rate of poplùation growth. This modeL envisages

a steady state Level of incorne per capita growth which depends on an exogenousLy

given rate of technical progress.

Endogenous growth modeLs build upon the ideas of neoclassical growth theory by

allowing intangible inputs such as knowledge acquisition, human capital (e.g. skills

acquired by Labor) as weIl as factors that enhance the efficiency of inputs ta affect

the production process. These models are thus able to explain non-decreasing retarns

to reprodllcible factor inputs (Romer. 1986, 1990). Within this context, there is no

steady state LeveL of incorne. Incomes can continue growing withollt bOlmds. Poor

economic growth in SSA has been blamed especially on inadeqllate investment and

more recently on bad macroeconomic policy environment which has hampered the

efficiency with which inputs are nsed.

There exists another strand of the literature (Boserup, 1981, and Kelly and

Schmidt, 1994, for example) which argues that poplùation increase may induce changes

or innovdtion in production technology, in social organization, in economies of scale or

agglomeration of economies (density of economic activity) , leading ta greater prodac­

tivity. This view is certainly not universally held. rvIany authors (e.g., Binswanger and

Deininger, 1997) have argued that not all societies experiencing poplÙation growth

have shown increased productivity.

Page 29: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

20

3.2.2 More on Endogenous Growth Theory

This sub-section highlights selected issues from endogenous growth theory.

Ruman Capital

Endogenous growth theorists have argued strongly for inclusion of human capital in

production functions cithcr bccause it is a direct input into research (RoIller, 198G,

1990), or because of its positive externalities (Lucas, 1988). While the intuition for

this inclusion is easy to fol1ow, appropriate measures of human capital are not easy

to find. School em'ollment rates, literacy rates, years of schooling! even wage rates of

the labor force have been used as proxies in many studies. On the whole! estimat.es

of the coefficient of these variables have yielded mixed results in empirical stndies.

In agricultural production functions, use of these proxies has sometimes yielded

Imacceptable restùts. Lan and YotOPOlÙOS (1989) report l' ... the estinlated coeffi­

cients for general education still appear unreasonably large" and subseqnently drop

this variable from their mode!. These authors also qllote similar finding by Kawagoe!

Hayami and Ruttan (1985). Binswanger et al. (1987) aIso obtained similar (lmac­

ceptable) filldings. In a widely cited study, Levine and Renelt (1992) fOlmd that

common measures of hllman capital do not have robust influence on GDP growth.

Rather, the statistical relationship is sllbject to the COtUltries included in the sample,

the time period tmder consideration, as well as other variables included in the mode!.

~Iore recently, Levin and Raut (1998) have provided evidence that suggests that it is

the manufactllrîng export sector, not primary commodity exports, that benefits froID

education. Indeed, sorne formulations (such as Loayza, 1994) of endogenolls growth

Page 30: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

21

models treat human capital as an efficiency factor (see the following section) rather

than as a factor input.

This study is formtùated in terms of stocks of factor inputs and uses the average

years of schooling of the labor force as a proxy for the stock of human capital.

EfBciency Factors and Policy Issues

As mentioned above, endogenous growth models posit a linkage between public poli­

cies and long-nID growth. In partictùar, outward oriented trade policies are said to

promote competition and thereby raise the efficiency with which factor inputs are

used. l'rade ë:ÙSO enhances externalities resulting from access to improved technology

(Grossman and Helpman, 1991), KhéUl (1987), Lucas (1988) and Ramer (1986, 1990).

Enlpirical findings in favor of the impact of trade have been provided by Sachs and

Warner (1995) and Edwards (1998), among others. Other efficiency factors that have

been fOlmd ta affect growth worldwide (and aIso within SSA) indude social capital

(institutions. associations, etc.) and public services (infrastrncture, etc.). See for

example Collier and Glmrung (1997) and Haque and Aziz (1997). In particlùar, the

raIe of infrastructure in promoting agriewtural output has been emphasized by Antle

(1983) and Binswanger et al. (1987).

Bad poliey environment in SSA has been blamed partly for SSA's poor economie

performance, (e.g., World Bank, 1989). Prodded by the Intemationall\tlonetary Fl.lnd

(I~[F) and the World Bank, most sub-Saharan African eountries have embarked on

sorne fonn of poliey reforme These are generally referred to as Structural Adjustment

Programs (SAP). The object of these reforms is threefold: (i) to eliminate or reduce

Page 31: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

22

fiscal deficits, (ü) to rein in monetary poliey 50 as to reduee inflation, and (ili) to

remove domestic cnrreney over-valnation by domestic currency liberalizing foreign

exchange markets. The idea is that these policies will make SSA economies more

efficient and competitive with the rest of world. The region will then benefit from the

advantages associated with international trade. These policies are having a profound

impact on SSA economies. Given the extensive theol'etit:al al.lù eillpirical support,

this study uses a trade variable as a proxy for the macroeconomic environment.

External Shocks

For a typical SSA co\mtry, changes in its external terms of trade affect her foreign

currency receipts from agric\ùtural €",,<ports. This in turn affects the resources (priee

and non-priee) that these governments devote to agriculture (e.g., proportion of in­

ternational prices paid to producers, provision of infrastnlctnre, etc.). For most of

SSA, agric\ùtural exports are a major sonrce of their foreign exchange earnings (over

which they have direct control), yet these cOlmtries have virtually no control over

international commodity priees. From their empirical study of growth rates over

time, Easterly, Kremer, Pritchett and Summers (1993) conclude that shocks, espe­

cially terms of trade, are important in determining long-nm growth. Naturally for

agric\ùture, weather is also an important exogenous shock.

This study uses these ideas from neo-classical and endogenous growth models to

investigate the impact that the stocks of the variables (whose growth rates enter

growth theory models) have on the level of aggregate agricultural incarne as weIl as

on the export and domestic shares of agrictùtural output in SSA.

Page 32: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

23

3.3 Rationale for Decomposing Capital

It is the position of this thesis that the three eomponents of capital have different

impact on agriclùture. A dollar of aOA flow is provided on humanitarian and ethical

grounds and is development oriented. On the other hand, the decision far P FX flow

is made on a risk and return basis, and is clearly trade oriented (World Bank, 1997).

In faet, a major rationale for OOA is imperfections in international capital markets

(Le., faihITe of foreign private capital to flaw to certain regions). It wO\ùd appear that

countries that carmot attract PFX are the ones most in need of OOA. Further. while

sections of the anA literature argue that such flmds are fungible and may actuéùly

help dissuade cornlpt and rent-seeking gavernments from pursuing rational economic

policies, PFX, especially foreign direct investments, are Imlikely to be diverted or

miSllSed.

The rationale for disaggregating domestic savings (SAV) from foreign financial

flows is essentially ta emphasize the ideas of the "two-gap planning mode!" (Chenery

and Strout, 1966), and the Harrod-Oomar model, (Domar. 1946, and Harrad, 1948).

These suggest that, foreign financial flows to a developing COlmtry play two roles,

(i) providing investment flmds, and (ii) making possible investments in areas t.hat

require foreign capital, that would otherwise not have been possible with domestic

savings only.

In addition, in the case of private domestic savings (by individuals or corpora­

tians), who are assumed ta he individual utility maximizers, these savings are in rel­

atively small amounts, and decision making regarding where to invest snch amounts

Page 33: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

24

is potentially different than decisions involving much larger aOA and PFX amalmts.

Government investments using domestic savings (taxes) may he influenced. by political

and other considerations that may Dot influence aOA or PFX.

The essence of the argument is that, different sources of capital target different

types of investments. In the empiricalliterature, Papanek (1973) first disaggregated

forcign financiai fio\vs iuto forcign aid, forcign privatc invcstment and other flows in

a study of the impact of investment ou economic growth.

3.3.1 Stocks of Capital Versus Polynomial Lags

Since investments take time to come ta fnlition, analysis of the effects of these flows

must recognize this time lag. A time lag of 6 years has been used in a nnrnber of

analyses that have investigated the effects of GOA. These stlldies indllde, Ivlosley,

Hudson, and Horrel (1987), Norton, Ortiz and Pardey (1992), Kherallah. Beghin.

Peterson and Rllppel (1994), Gunjal and Gichenje (1997). The empirieal basis for

this is the World Bank (1984) report which sllggests that. on average, "Vorld Bank

flmded projects require a total of 7 years for cumulative cash flows to equal the original

investments. Assllming that the six-year lag is appropriate, one wOlùd then have ta

estimate the parameters of the polynomial as well as the degree of the polynoDÙal.

The appropriate equivalent stnlctllres for PFX and SAV will then have ta he estimated

as well.

What is more, this lag structure approach is somewhat ad hoc. It amounts ta

"payback period" appraach to capital budgeting, where only the length of time re­

quired for the sum of dîSColmted cash flows frOID a project to equal the original amount

Page 34: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

25

invested is being counted. One would prefer an approach that is equivalent to the

"net present value method" , which considers all cash flow accrtùng to an investment

irrespective of when they occur. To this end, flow of investments, and of poplùation

and human capital used in growth models are accuInlùated ioto stocks. F\rrther, the

use of stocks rather than current flow variables has the potential to help reduce the

problem of simlùtaneity between inpnts and out.pnt.s.

Above then is the economic basis Imderlying the analysis performed in this study.

The discussion now shifts to economic specifications: first the meta-production flmc-

tion, then the static and dynamic share response models.

3.4 Econometrie Specification: Production Func­tion

This section specifies and comments on the production function modelled here. The

meta-production function has been described by Lau and Yotopoulos (1989) as .4 a

common nnderlying production function that cao be used to represent the input-

output relationship of a given indllstry, e.g. agriclùture, in all countries, ..." The

first objective of this study is to estimate such a flillction for SSA. The Imderlying

asslunption is that these countries have access to the same production technology.

The model is a two-way error component "Cobb-Dollglas type" production tlmction

within the context of utility maximization with competition, where output depends

on factor inputs, efliciency factors and external shocks.

Page 35: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

26

3.4.1 The Model

Following the discussion in the preceding sections, the production function to be

estimated is posited as follows. The transcendentallogarithmic flmction was originally

proposed. Tests that the cross-terms in that specification were zero were accepted,

leading to the following specification.

where,

AGDP is total agriclùtural output per hectare of agriclùtural land;

LAER is agriclùturallabor force per hectare of agriculturalland;

KODA is the stock of official development capital per hectare of agriclùturalland;

I(P FX is the stock of private foreign commercial capital per hectare of agriclù-

tl1ral land;

K S AV is the stock of domestic savings per hectare of agricultural land;

HeAP is the measure of the stock of human capital.

In this specification, i stands for a cOlmtry within the panel of data (i = 1. 2, ... , 27),

and t stands for time, (t = 1970,1971, ..., 1993). The stocks of physical capital are

lagged one year to allow time for investment flow to come ioto the production process. 1

Binswanger, Yang, Bowers and ~hmdlak (1987) used a lag of two years.

The production nmction looks like the typical Cobb-Douglas production nmction.

However, no assumptions are made with respect to the magnitudes or sum of the

elasticities 13t, (32, f33' (34, and 135, That is, decreasing returns to scale has not been

1 Also, in constructing the stock of physical capital, endogeneity between output and investment iscontrolled for by assuming that all investments were made at the beginning of the year.

Page 36: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

27

imposed. The variable Ait represents the efficiency and external shock factors im­

pacting agrictÙtllral production (diSCllSSed above). Linearizing by taking logarithms

yields, the full statistical model

In(AGDP)it - /3tln(LABR)it + ,B2ln(KODA)~t-l + 1331n(KP FX)it-l

+,84 In(KSAV)it-1 + f3sln(HCAP)it + 136(OPEN)it

/37 (EX01) it + 138 (EX02) it + € it • (3.1 )

where,

OPEN is a measure of the openness of the economy;

EX01 measures the terms of trade:

E ..~02 measures variations in weather and other related shocks that impact pr~

dllction. The error tenn, éd, is decomposed into an lIDobservable COlilltry-specific

effect, Qi , (fixed over time), and an tmobservable time-specific effect, Àt • (fixed across

cOlmtries), and a random term. eit. That is,

(3.2)

The country and time specifie components of the error term may he fixed or

random. In this conte.xt, a fixed efIects model is posited since the sample of COlmtrïes

is not random. Rather, these are the SSA cOlmtries for which complete data for this

study are available, and the sample practically exhausts the countries of interest to

us. Inference made in this study may therefore he considered to apply to this group

of countries. In addition, statistical tests for random effect specification (l\tlundlak,

Page 37: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

28

1978) yield evidence against sncb specification.

Country and time effects account for differenees between countries and time that

affect inputs and outputs, and hence help control the potential for simultaneous equa-

tians bias. Country dummy variables capture differences in soil fertility, differences in

government policies such as infrastnlcture, the impact of marketing boards, pricing,2

external markets for exports, clùtllral/institutional differences, clifferences in techni-

cal efficiency in production across countries, etc. They are also assllmed ta capture

inter-conntry differences in definitions, measllrements and qualities of outputs and

inputs. Time dummy variables captnre effects that are specifie to time, for example,

changes in total factor produetivity over time for aU cOlmtries.

While the two-way error component model allows one to capture differences be-

tween cOlmtries and over time, this approach uses np many degrees of freedom. Having

established that the fixed-effect specification is Dot reject.ed, t.he time trend is llsed

ta capture alltonomous growth in productivity over time, (Hicks neutral technical

change3 ), and the time dummy variables were dropped.

Static domestic and export share equations are discussed next.

3.5 Econometrie Specification: Analysis of Agri­cultural Share Response in the Short-Run

This section explains the econometric approach adopted in investigating the supply

response of farmers in the export and domestic sub-sectors of sub-Saharan African

agriculture to priees, factor inputs, macroeconomic poliey, and exogenous shoeks in

2 Produeer priees are investigated in the next two papers.3 It will aIso capture the effects of omitted trending variables.

Page 38: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

29

the short-run.4 Static supply responses are widely estimated in the literature.

The gross domestic product (GDP) funetion (also ealled profit ftmction) approach

is used to obtain domestic and export shares. Export share is defined as the value of

total agriclùtural exports divided by total agrietÙtural output. Domestic share is one

minus the export share.

Use of the profit funetion approach enables one ta analyze the domestic sub-sector

just as weIl as the export sub-sector. Thus, one cireumvents sorne methodological

and data problerns alluded to by Jaeger (1992) among others. This approach also

makes it possible for one to investigate the response of these sub-sectors not orùy ta

price, but also ta changes in factor inputs, external shocks, and the effect of policy

environment. Duality between profit and cost functions allows specification of a

flexible profit funetion without worrying about the specifie form of the production

ftmction.

3.5.1 Profit Function

Consider lltility maximizing economie agents making decisions with respect to factor

inputs, output priees and other variables that affect their profits. Given perfect com-

petition in input and output markets, the decision they face is stated as maximizing

the value of output quantity y, subject to available production technology, T, fac-

tor endowment vector, x, (capital, land, labor and human capital), and a vector of

prevailing positive output priees, p. Priees of agriclùtural exports are exogenously de-

termined by governments or in the world market, while domestic agrielÙtural priees

.. Often, this study shall refer ta the agricultural sector as being made up of two sub-sectors: theexport sub-sector and the domestic sub-sector.

Page 39: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

30

may be determined by the demand and supply situation in the home country or

sometimes by governments.

By definition, the profit nmetion ?r, in terms of output priees and input vector

may he written as the solution to a maximization problem5

rr(p, x) = max{p'y: (x, y) f T,p» O},y

(3.3)

Restrictions are then imposed to ensure that the flllction is weil behaved, (i.e.,

existence of a solution and a ma.ximtun, giving rise ta the concept of a restricted

profit function). The restrieted profit funetion is homogenous of degree one, convex,

increasing in output priees and non-increasing in input prices.6 Under the assllmp-

tion of constant returns to seale, it is aIso homogenalls of degree OOP, concave and

increasing in qllantities of fixed inputs. The asslunption of constant retllrns ta scale

is used often. In this context, it is justified on the grolmds that. agrieultllral land,

the only factor input that is nat repraducible is far now, at least. not a canstraint in

SSA.

~IcFadden (1971) has shown the existence of a one-to-one correspondence between

the set of concave production flmctions and the set of profit fllnctions. This observa-

tion allows one not ta worry about the specifie fonn of the production hmction.

3.5.2 Specializing the Model

The agrictÙtural profit nmction (GDP function) is represented here as a transcenden-

tallogarithmic nmction. ft is a second order Taylor's approximation to an lmknown

5 The GDP function has been studied by ~IacFadden (1971)t Lau and Yotopoulos (1971}t Samuelson(195~)t Diewert (1974) among others.6 See for example, Chambers (1988)t Diewert (1974)t and Varian (1992).

Page 40: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

31

nmctional form (Kmenta, 1967, Christensen, Jorgenson and Lau, 1973). This special-

ization of the GDP function enables one to obtain value shares of each sub-sector of

agrictùture (export and domestie) in terms of factor inputs, producer priees, external

priees, a policy variable, and the weather, by difierentiating with respect to producer

priee in each sub-sector. The two sub-sectors span total agrieultural production.

The translog function also allows the elasticity of substitution between inputs to

be flexible, and does not impose input-output separability. It has been used by Kohli

(1978) and Lawrence (1989) to study substitution possibilities between Canadian im-

ports, exports and domestic inputs or outputs. ~Iore recently ~Iartin and Warr (1993.

1994) llsed it to study the decline of agriclùtlrre vis-a-vis Inanufacturing and service

sectors of Indonesia and Thailand. As indicated above. the agriclùtnral production

fllDction equation (3.1) is obtained as a restricted translog flmction. Here~ we restate

it as a translog flIDction with additional priee terms. These extra terniS l'anse little

trouble since the impact only the precision of estimates. Even then, this is not the

model that is eventually estimated.

The GDP flmction is then written as

lnAGDP

+

2 1 2 2 6

= 80 + L 8i lnPi + ? L L T'ij ln pdn Pj + L Bk ln Fki=l .. i j k=l

1 6 6 2 6

22: 2: 6km ln Fk ln Fm + 2: L'lik ln pdn Fk,k=l m=l i=l k=l

(3.4)

where, AGDP is the agriclùtlrral GDP; the Pi are the producer priees prevailing in

the domestic (i = 1) and export (i = 2) sub-sectors; and the Fk are the six factor

inputs, namely, the stock of anA (KI), the stock of PFX (K2), the stock of SAY

Page 41: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

32

(K3), agriculturallabor force, (LABOR), agrieulturalland (LAND), and the stock of

hlunan, (HCAP).

Export and domestic share eqllations are derived as follows. By Yotmg's theorem

(symmetry of the mixed partial second derivatives), "Yij = "Yji, Dkm, =8mk , in the GDP

ftmction equation (3.4). In addition, homogeneity of degree 1 in priees requires that

E;=l (Ji = 1; E;=l rij = 0; 2:;=1 'Yij = 0; E:=t L;=1 rij = 0; L1 T/ik = o.

The share of each sub-sector in agrÏctùtural output is then obtained by logarithmic

differentiation with respect to the respeetive priees as

and

SI = (JI + 1'11 ln Pl + 1'12 10 P2 + 1111 ln KI + 1/l21n K2 + '113 ln K3

+1/14 ln LABOR + TJ15 ln LAI'!D + TJ16 ln HeAP,

S2 - (J2 + 1'21 ln Pl + "Y22 ln P2 +'121 1nKl + TJ22 1n K2 +'1231nK3

+1]24ln LABOR + '1]25 10 LAND + 1]261n HCAP,

(3.5)

(3.6)

where, SI is the share of the domestic sub-sector in agriclùtural GDP and 82 is the

share of the export sub-sector. Factors of produetion are assluned to be mobile be­

tween the two sub-sectors with the rental price of each given by its marginal praduct.

Share equations are then augmented with a measure of the macroeconamic pal­

iey variable, and external shocks variables. Allowanee is made for the influence of

country-specifie and time-specific effects as done for the production function above.

Priees are then expressed relative to each other, since the share equations being

profit functions must be homoge~ousof degree one in priees. Given the assumption

Page 42: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

33

of constant returns to seale in factor inputs, factor inputs are deflated by the LAND

variable. This foeuses the discussion on yield (per hectare). The normalization aIso

helps in eontrolling for heteroskedasticity among the eOllntries (of different sizes).

Thus, a representative share equation is DOW written as

s = a~ + a~ ln Pl + 0; ln KODA + a~ ln K PFX + Q~ ln KSAVP2

+o~ InLABR + a~ ln HCAP +~. (3.7)

where, [<ODA, KPF..:'(, KSAV,and LABR, are as defined for the production flmc-

tion. The share equations to be estimated are then written as

+04 1n(KSAV)it-1 + Qsln(LABR)it + Q6 1n(HCAP)it

ADON/it - ,8t ln(PRICE)it + {32 ln(!(ODA)it-1 + f33 ln(KPF"Y)it-1

+{34 ln(KSAV)it-1 + 13s ln(LABR)it + ,86 In(HCAP)it

for the export and domestic shares AEXP and ADOAt/ respectively. PRICE is the

ratio of the index of real produeer priee of agrieultural exports to the index of the

real producer priee of agriculture for domestic constunption. The variables 0 PEN,

EXOl, and EX02 may he viewed as components of o~ in equation (3.7). They have

Page 43: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

34

the same meaning as in the production function above. AIso, ri and 6i are the country-

specific effects, and (t and Vt are the time-specific effects as in the production fUDction.

The two shares must add up ta one (adding-llp condition). Hence the relationship

between the slope coefficients in eqnations (3.8) and (3.9) are,

Here too, having established that the fixed-etfect specification is not rejected, the

time dummy variables are dropped and a time trend is used to capture autonomO\lS

growth in prodnctivity over time (as weIl as effects of omitted trending variables).

The trend coefficients in the two equations are also the negatives of each other, while

the cOtmtry dtunmy variables equal one minus the other.

3.6 Econometrie Specification: Dynamics of Agri­cultural 8hare Response

This section models the dynamics of aggregate export and domestic shares of snb-

Saharan African (SSA) agriclùt.ure. In spite of interest in static share response in its

own right, one notes that short-run response obtained from static models does not

take into account the fact that the empirical relationslùps on the ground may not be in

long-run equilibrhun. Indeed, agriclùtural snpply may not respond immediately and

ft.ùly to changes in the explanatory variables. This may be due to habits, persistence,

implementation lags, misinterpreted real priee changes, and other factors.

This section modifies the static specifications of the previons section to allow for

deviations from steady state. An autoregressive distributed-lag model is adopted

Page 44: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

35

(Hendry, Pagan and Sargan, 1984). The specification is general enough to allow

estimation of both long-run equilibrium and short-nm dynamic parameters, both

being of interest in this study.

To illustrate. Tree crops naturally take time to come into production after being

planted. Farmers and others engaged in one sub-sector of agrÏCtùture will require tirne

to leam new skills shatùd they wish to move from one sub-sector ta another. But they

will probably take even longer to put aside previously acquired habits (Le., ways of

making a living). Land dedicated to one crop may require substantial time lag ta be

ready for another crap. A piece of land previously nsed as a cocoa, rubber or coffee

farm for example, takes long ta be rid of the fibrous roots (and taproots tao), as

well as changes in acidity of the soil caused by these trees. Further, physkal capital

nlay not he readily withdrawn from one application to another since retooling may he

called for. Expenditnre of additional S1.uns of money to acqtùre new capital eqllipment

may require severa! years of planning and reallocation of profits from one sub-seetar

to the other. Yet, it is known that farmers being utility maximizers respond t.a priee

changes, technical change and other factors that enhance their welfare. Long-nln

aggregate supply responses will include the effect of reallocation of resOllrces from

one snb-sector of agriclùtl1re ta another, or from non-agrictùtllral activities ta the

agricultural sector (or vice-versa), changes in government expenditures, etc. The

approach adopted in this paper precisely addresses these concerns by building upon

the findings of the static share eqllations above.

Jaeger (1992), Binswanger (1989), and Binswanger, Yang, Bowers and l\JItmdlak

(1987) among others, who have themselves estimated agriclÙtural supply elasticities

Page 45: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•36

from static equations have argued that supply elasticities obtained from static equa­

tions, though common, are best viewed as short-nm responses. Short-nm responses

may however be of interest in their own right: for short term impact evalnation of

poliey and priee changes for example. Or where it is believed that the data generation

proeess has changed rendering long-nm estimates diflielùt or impossible to compute.

Starting \Vith the static equations. a general autoregressive distribllted-Iag modpl

(ADL) of the dynamics of each share equation involving static parameters (from wruch

long-nm parameters may be inferred) is derived. ~[aking inferences indirectly from

the ADL specification has been shawn to be problematic. For one thing, additional

computations are ealled for. For another, since these computations involve ratios

of regression coefficients, problerns arise in regard to estimating standard errors of

the long-nln coefficients from finite sample estimates. These have been discllssed

by Kmenta (1984) and Bewley and Fiebeg (1993) and Kesavan. Hassan. Jensen and

Johnson (1993).

In this study, the ADL is transformed 50 as ta be able to estimate directly long­

nm parameters and their standard errors. The transformation also allows one to

investigate the extent of persistenee and of implementation lags. Thus, no additional

computations shall be reqnired. The transformation used was proposed by Bewley

(1979) and has been diseussed by Bewley and Fiebeg (1993), Banerjee, Dolado. Gal­

braith and Hendry (1993), and Wickens and Breusch (1988). I<esavan et al. (1993)

for example, used such an approach ta study the demand for meat (beef, chicken

and pork) in the United States. ~Iartin and Warr (1993, 1994) also llsed a similar

specification to study the dynamics of the decline in the agriclùtural sector vis-a-vis

Page 46: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

37

manufacturing and services sectors in Indonesia and Thailand. Anderson and Blun-

deU ( 1983) used a sirnilar approach that combined both long-nm relationships and

short-nm behavior ta test consmners' expenditure patterns in Canada.

3.6.1 Distributed Lag Specification

Consider a static two-way error component model

P

Yit = L {3kx ikt + ê it ,

k=1

(3.10)

where Xik is the data pertaining ta variable k, (k = 1, .,., p) in COlmtry i, at time t, and

êit has both cotmtry-specific and time-specific cOlnponents. Naw allow lagged depen-

dent and explanatory variables ta enter the model ta account for habits. persistence.

adjustment costs, etc.. that cause agents nat to adjust their activities immediately.

This resnlts in an ADL which may be written as

m p n

Yit = L CtjYit-j + L L ,BkjXikt-j + êlt,j=1 k=l j=O

(3.11 )

where the Qj are the coefficients of the lagged dependent variables. the Bkj are the

coefficients of the jth lag ( j = 0, 1, ... , n) of the kth explanatory variable of eqnatian

(3.10), k = l, ... ,p.

Ta make inference about long-run parameters withant tlSing a transfonnation snch

as Bewley's, one wOlùd first have to estimate an the Ctj and Pkj and then proceed

to compute the equilibrium parameters of the relatiansmp between the dependent

variable and the explanatory variables. The long-nm impact 8k of the kth explanatory

variable is then derived as,n

8k = À L{3kj,j=O

(3.12)

Page 47: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

38

where

For long-rtm equilibrium relationship to exist, it is req\ùred that

(3.13)

The Bewley transformation is achieved by the following algebraic manip\ùation.

One, for each exogenous variable Xk, k = l, ...~p~ subtract 13kjXkt-J from each ,BkjXkt-j

(j = 1,2, ... , n) term on the right of the ADL, equation (3.11), and add the same to

the PkOXkt term (Le., levels of the variable) also on the right. Second, for each QjYit-j

on the right, subtract GjYit from both sides of the equation. Then re-arrange and

collect terms. Thus the Bewley transfonnation is obtained as

Yi' = -,\ fOj(Yil - Yil-j) - t [,\ t,3k] (X'k' - X'k.-i )])=1 k=l )=1

+~ [,\ ~8k]] X,k' + ,'lé", (3.14)

where, À r:.;=o 13kj is the long-nUl effect on y due ta a change in variable Ik. This

transformed model is linear in parameters which can thus be estimated directly (co-

efficients and their standard errors). Note that the first two terms on the right capture

short-nm dynamics (Le., persistence and implementation lags).

The autoregressive model, the partial adjustment model and the static model are

all nested in equation (3.14). See Kesavan et al. (1993) or Hendry et al. (1984) for

discussions on appropriate restrictions that will yield each of these reduced models.

If required, these restrictions can he tested to determine which of the restricted forms

is more appropriate (depending on the situation one is modelling).

Page 48: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•39

Next, the dynamic domestic and expart share equations are derived.

3.6.2 Dynamics of Agricultural Share Equations

Consider the static export share equation analyzed in the preceding section with time

dummy variables replaced by a time trend variable,

AEXPit - Ài + P1ln(KODAitJ + 132 ln(l{ P FXid + .L13 ln(K.sA \lid

+t14 ln(LABRïd + I3s0PENit + (36 EXOl it

(37E~~02it+ /3sln(PRIC Eit ) + ,8gT REND + eit, (3.15)

where 'i refers to country (i - 1, ... , 23). in the panel data, and t is time (t =

1970,1971, ... ,1993).

The ..\i are the cOlmtry-specific dummy variables. t.he 8j , j - l. 2..... 9 are the

coefficients of the varions variables and eit is a random errar term.

Up ta five annuallags of the dependent and explanatory variables are hypothesized

ta impact enrrent share. The rationale for five years rests on the grolmds that tree

crops may take up to five years ta come on stream, land previously used for one crop

may t.ake that long to be ready for another. Farmers may take this long to adapt

their habits, plan and realloeate resources from one sub-sector to another.

A general-to-specific modelling approach that does not reqlùre one to specify a

priori that the sequence of lag coefficients progresses in any particlùar manner is used

to achieve a parsimonious specification of the ADL. Tests for significance of lags were

conducted by considering difIerences among successive lags (that is, each additional

lag must provide new information), to obtain a parsimonious specification for the

Page 49: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

40

ADL. The ADL is analyzed to ensure that the residuals are stationary as weil as

for the existence of long-mu relationship between the dependent variable and the

explanatory variables.

Once an ADL is obtained, the steps outlined above are followed to achieve the

corresponding Bewley transformation. Four Bewley transformations were obtained.

One for the overall data (23 countries frem 1970 te 1993 callcd 'Overall'), one for the

Western agro-climatic region (West), one for the Eastem-Southern agro-dimatie re­

gion (East) and one for the Sudano-Sahel agro-climatic region (Sudano). This section

does not model the stnlctural adjustments period (1981 ta 1993), because the use of

lagged values leaves too few observations for meaningflù estimation. Policy implica­

tions will be made with respect to the macropolicy variable. OPEN. Stability of the

data generating process was analyzed for the statie model, as weil due consideration

was given ta the presence of lagged dependent variables on the right.

By way of illustration, the ADL for the Overall specification is

+1PlOln(KODAitl + 'rlhsln(KODAit - S )

+l/J20 ln(KP F X it ) + 'W21 ln(KP FXit -d + 'W22ln(KP F )(it-2)

+1/130 ln(K S AVit) + 1/;34 ln(K S AVit-4)

+1/140 In(LAB14t) + 'l/J43In(LAB~t-3) + 'w4s ln(LAB14t-s)

+WsoOPENit + t/J6oEX01it + 1/;6sEX0 1t-S

+1/J70E ~"(02it + 1/J71EX02it- 1

Page 50: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

+t/Jsoln(PRICEit ) + 'l/J84,ln(PRICEit - 4 )

+1/JgT REND + éit.

41

(3.16)

The following are the Bewley transformations obtained for the export shares:

(i) Overall

AE...Y:Pit = Gi + (3t(AEXPit - AE...Y:Pit-d + f33(AE ...Y:Pit - AE...Y: Pit-3)

+1'10 In(KODAit ) + 1'20 m(KP F...Y,t ) + 1'30 In(KSAV':d

+1'40 In(LAB~t) + 1s00PENit + 160EX'Ol it

+1'70E)(02it + 1'so In(PRICEit )

+"'(ls(ln(KODAit ) -ln(KODAit - s))

+1':u(ln(KPF)(ie) -m(KPF...Y:1t-d)

+1'22(ln(KPFXit ) -ln(KPFX1t- 2 ))

+1'34(ln(KSAVid -ln(KSAVit-4))

+1'43(ln(LAB~t)-ln(LAB~t-3))

+1'45(ln(LAB~t)-ln(LAB~t-s))

+1'6s(EXOlt - EXOl t - s )

+1'71(EX02it - E.~02it-l)

+Î84(ln(PRICEit ) -ln(PRICEit - 4 ))

+i9TREND+éit, (3.17)

Page 51: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

(ii) East

AEXFit - Qi + {31(AEX~t - AEXFit-d + /32(AEX~t - AEXPit- 2 )

+{3s(AEXPit - AEXPit- S )

+'10 ln(KODAit ) + 1'20 In(KPFXit) + 1'30 ln KSAVit)

+'40 In(LAB~t) + '"YsoOPENit + '"Y6oEXOlit

+,iOEX02it + 1'soln(PRICEid

+,14(ln(KODAit ) - ln(!(ODAit - 4))

+,ls(ln(KODAid - In(KODAit_s))

+'"Y22(ln(KPFXid -ln(KPF~'(it-2))

+'"Y31(ln(KSAVit) -ln(KSA\lat-d)

+'"Y3s(ln(KSAVid -ln(!(SAVit-s))

+'43(ln(LAB~t)-ln(LAB~t-3))

+'"Y45(ln(LAB~d-ln(LAB~t_5))

+'"Y6s(EXOlt - EXOlt- s)

+'"Yn (EX02 it - EX02it- 1)

+784(ln(PRICEit ) -ln(PRICEit- 4))

+79TREND + êit,

(iü) West,

42

(3.18)

Page 52: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

AEXPit - Qi + (3t(AEXPït - AEX~t-l)

+"'(10 In(KOD~t) + 120 In(KPFX it ) + "'(30 ln(KSAlItt)

+"'(40 In(LAB~t) + 1500PENit + 1'60E..-YOl it

+Î70E ..-Y02it + 180 ln(PRIC Eit )

+Î15(ln(KOD~d -ln(KODAit - s))

+r21(ln(KPFXid -ln(KPF.Xtt-d)

+"'(34(ln(KSAVjt) -ln(KSAVit-4))

+143(ln(LAB~d-ln(LAB~t-3))

+r45(ln(LAB~d- In(LABRit_ s ))

+184(ln(PRICEit ) -ln(PRICEit-,d}

+ÎgTREND + ê1tl

(iv) Sudano,

AEXPit - Qi + f3t(AEXPit - AEXPit-d

+1'10 In(KODAit ) + 1'20 In(KPF Xit) + 130 ln(KSAlItt)

+"'(40 In(LABRït) + 1500PENit + 160EX01it

+170EX0 2it + rsoln(PRICEit )

+112(ln(KOD~t) -1n(KODAit_2 ))

43

(3.19)

Page 53: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

44

+1'2s(ln(KPF Xid - ln(KPF X it- s))

+1'34 (ln(K S AVit) - ln(K S AVit-4))

+1'4s(ln(LABR;t} - ln(LAB~t-5))

+1'84 (ln(PRIC Eit ) - ln(PRICEit - 4 ))

+1'gTREND + eit·

Note that the specifications vary somewhat from one region to another. This must

be due to regional differences in persistence and implementation lags.

ln the literature, a typical panel has a short time series (about 5 observations),

thus the effect of presample observations (Le., observations prior to the start of the

study) on the stlldy Imder consideration have been discllssed in sorne sections of the

literature, e.g., Pakes and Griliches (1984). Given a time series of as many as 23

obsenrations per COlmtry in this study, the impact of tmohserved presample data will

have minimal impact.

3.7 Econometrie and Other Related Issues

This section discusses relevant econometric issues.

3.7.1 Single Equation Specification

It is weIl known that most macroeconomic variables are simtùtaneously determined.

Thus, simtùtaneous equations specification wotùd probably be most appropriate in

general. However, use of single equation specification in economic studies is far more

common. In the current context, one rationalizes the single equation specification on

the following grounds. Inclusion of country-specifie and time-specific dummy variables

Page 54: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

45

in the production function mitigates the effects of unobserved components that may

enter both output and inpnts. Second, measuring inputs as stock variables rather

than actual CUIrent flow variables goes sorne way towards avoiding the problem of

simultaneous determination of output and inputs. (The random distllrbance term is

of course assllmed ta he lillcorrelated with ather right-hand side qnantities).

3.7.2 Heteroskedasticity, Autocorrelation and Errors in Vari­ables

Paoling data for different countries immediately raises the question of whether residual

variances vary from one COlultry to another. In general they do. Use of inclividual

country dummy variables goes sorne way in addressing this problem. In addition.

standardizing ontput and factor inputs by dividing these qllantities by the agricultural

land area also helps reduce the problem of residual heteroskedastidty due to large

differences in cOlmtry sizes. Norton et al. (1992) and Binswanger et al. (1987) for

example used tms approach.

The third step in addressing the heteroskedasticity problem wotùd be to estimate

the variance-covariance matrLx by a robust estimator. If autocorrelation is also fOlwd

to exist, then a modification to the robust estimator that addresses autocorrelation

shaH be used.

One heeds the advice of Dagenais (1994) and lVlankiw (1995) among others and

hesitate to transform the model to remove seriai correlation because of concern that

such transformation may introduce more bias in coefficient estimates than otherwise.

This concem stems froID the very real chance that measurement errors are present

Page 55: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

46

in the data. Economists who work with SSA data generally believe that the data

qllality sholùd he improved further7•

3.7.3 Stability of the Data Generating Process and Pooling

To pool time series data for severa! countries, it is important to check whether there

exists statistical justification. Testing of restrictions (Le., pooling) may proceed along

many tines. This study tested for poolability in the context of parameter stability

for bath the production flmction and the supply response models as follows. First,

the cOlmtries were divided into three agro-climatic regions (Eastem-Sollthern region,

Sudano-Sahel region and Western region).8 Then model pararneters were estimated

for each region over the entire period (1970 - 1993). Tests for structural breaks in the

slope parameters between the following time periods were then performed for each

region:

(i) 1970 - 1980 and 1981 - 1993, (ii) 1970 - 1981 and 1982 - 1993, (Hi) 1970 - 1982

and 1983 - 1993, (iv) 1970 - 1983 and 1984 - 1993, (v) 1970 - 1984 and 1985 - 1993,

and (vi) 1970 - 1985 and 1986 - 1993.

Next, tests were performed for difIerences in slope parameters hetween pairs of

agriclùtlU"al regions over the suh-periods 1970 - 1980 and 1981 - 1993 and over the

entire periode (Many (e.g., the World Bank) consider the 1980s to have been a par-

tictùarly difficlùt time for many African countries. It is also dnring this decade that

structtU"al adjustment programs took hold. A Wald test that recognized heteroskedas-

7 The problem of errors in variables is pervasive however. Duncan and Hill (1985), Altonji and Siow(1987) for example, have reported these problems even with the popular Panel Study of IncorneDynaimcs Data (Psm).8 Countries belonging to each region are indicated in subsequent chapters.

Page 56: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

47

ticity and first order seriai correlation was used).

Then pooling data for all countries, tests for structural break over the following

periods were aIso performed: (i) 1970 - 1980 and 1981 - 1993, (H) 1970 - 1981 and

1982 - 1993, (iii) 1970 - 1982 and 1983 - 1993, (iv) 1970 - 1983 and 1984 - 1993, (v)

1970 - 1984 and 1985 - 1993, and (vi) 1970 - 1985 and 1986 - 1993.

Finally, a group of ninctccn S5A cOtUltrics classificd into thrcc subgroups were

considered. The first SUbgrOllp is classified as having lmdergone LARGE positive

changes in macroeconomic poliey (fiscal, monetary and exchange rate policies) be­

tween 1981 - 1986 on one hand, and 1987 - 1992 on the other. The second as having

lmdergone S~IALL positive changes in macroeconomic policy over the same period.

The third as having lmdergone POOR (negative) changes in macroeconomic poLicy

aiso over the same period. This classification was performed by Bouton. Jones and

Kiguei (1994). Production nmctions and static share equations were estimated for

each sllbgroup as classified. Tests were then perfornled for differences in parameters

between pairs of subgroups. Reslùts are reported in chapters 5 and 6.

Page 57: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•Chapter 4

Data

This section discusses data sources and data set constrnction. The data are drawn

from various sources, mostly publications of the World Bank and the Food and Agri-

clùtllral Organization of the United Nations. The data cover the period 1970 to

1993.

4.1 Data Description and Sources

AGDP, Agricultural Gross Do'mestic Product ai factor cast:

The source of this data is World Bank (1995) and earlier issues. Agriclliturai

output includes agriclùtural and livestock production, fishery and forestry output.

Data are published in current values of domestic enrreney and aiso 1987 values of

domestic curreney. Agricultllral gross domestic produet in constant 1987 domestic

curreney values are eonverted to 1987 United States dollars (US$), using the World

Bank's conversion factor for each country. (The World Bank refers to it as Atlas

method. The method is a blend of both official exchange rates and paraliei market

exchange rates). 1

1 1987 values are used because many publications and data sources do 50, and aIso by this, dollarvalues of agricultural output show increases that generally agree with quantity outputs.

48

Page 58: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

49

LAND, Agricultural land:

This is obtained from FAO Production Yearbook (1996) and earlier issues. Agri­

clùtnrai land is computed as the sum of land used for (i) Arable and Permanent

Crops, (ii) Permanent Pasture, and (Hi) Forest and Woodland

KOD.4. physical stock of ODA:

Net annual Hows from 1960 ta 1995 in CIment United States dollars were supplied

by William Easterly of the World Bank. He obtained the data from üECD sources.

Stock computation: Flow data were first converted to constant 1987 US$ llsing US

consumer priee index (Deaton and l\tIiller, 1995). Next, stock data were const.nlcted

by the perpetuai inventory method: Kt = Kt- t (1 - 8) + /t, where Kt is the stock at

time t. and It is the time t flow. The initial stock of aOA flow is then eomputed as

Ka = /0/ (8+9 ). where 9 is the estimated average growth rate of real aDA investment.

and /0 is the flow at time O. {) is the depreciation rate of capital (discIlssed below). /0

and 9 are bath obtained by regressing the log of real aOA flow during the period 1960

ta 1969 against time and taking anti-Iogs. (Harberger, (1978). Nehru and Dhareshwar,

1993).

Rate of decay, {) : Brown, (1972) computed physical capital stade for Ghana using

survey and investment data obtained from the local statistical authorities. The de­

predation rate implied in his computations is 6 ~ 7%. Nehnl and Dhareshwar (1993)

used 8 = 4% in camputing the physical capital stock for a group of 92 develaping

countries. l\Jlankiw, Romer and Weil (1992) used (8 + a) = 5%; a is the exogenous

rate of growth of technical progress. For both ODA and PFX, it is noted that sorne of

Page 59: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

50

the infiows are Dot physical capital but consumables sucll as fertilizer and teclmical

assistance in the fonn of expartriate salaries. These are consumed within a relatively

short tirne. In this constnlctioD 8 was set to 10%.

For KODA, KPFX, SAY and HCAP (below) total stocks in each ecoDomy are used.

Sectorial data are Dot available. Thus one implicitly assumes that the components

relevant ta agriculture are a constant share of each total, or if not, that dcviations do

not bias restùts in any significant manner. See for example Binswanger et al., 1987).

K P F X, stock of p7'ivate commercial foreign flows:

Net capital armual flow data was also obtained from William Easterly of the

World Bank, and supplemented by data from World Bank (1996b) and earlier issues.

Easterly obtained his data from 11tIF sources. By definition, net private foreign

capital ('omprises foreign direct. investment, debt from foreign banks and other private

creditors, as well as portfolio equity investments. Stocks of PFX were constnlcted

as done for aDA from 1970 ta 1993. Whenever net flow for a particular year was

negative, previous stock was only depreciated, Le.. (~urrent negative flows were not

added to depreciated previous stock. The impact of these negative amollnts is not

10st however, since net negative flows of PFX come out of SAY and/or üDA.

K S AV, stock of domestic savings:

AnnuaI data on gross domestic savings were obtained from World Bank (1995)

and earlier issues. For post 1973 period, tms is available, as percent of GDP, (and

in domestic currency amounts). GDP for this period is published in US$ in World

Bank (1995). Stocks were constructed as done for üDA. Again, whenever savings

were negative, they were set ta zero.

Page 60: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•51

OPENJ openness of an economy:

l'vIeasured as foreign trade share of GDP, i.e., OPEN = (E=por~br;portS). Exports

are exports of goods and non-factor services, (free on board values). Imports are

imports of goods and non-factor services, (cost insurance and freight values). GDP

is obtained from World Bank (1995).

IICAP, hU/nan Capüal :;tvc;k:

The stoek of hUIDan capital is the average nnmber of years of schooling in the total

poplùation over age 15. These are obtained from Barro and Lee (1996) and Nehru,

Swanson and Dubey (1995). The Barro and Lee data set contains quinqllennial

stock data for 23 SSA countries from 1960 ta 1990. The Nehnl data set has annual

observations for 21 SSA cOlmtries from 1960 ta 1987. Starting with the Barro and Lee

data, schooling years for the intervening years were estimated and then extrapolated

for 1991 ta 1993. For the few COllntrîes not common to bath data sets but for wruch

Nehru et al. (1995) have data, these annua! data (1970 - 1987) values were llsed and

then extrapolated for the years 1988 ta 1993.2

LABR, the stock of agriculturallabor force in each economy:

This was abtained from World Bank (1995) and earlier issues and FAO Production

Yearbook (1996) and earlier issues.

E)(OI, Terms of trade:

This is defined as index of export priees divided by index of import priees. Ob-

tained from World Bank (1995) and earlier issues.

2 AIso, the process was reversed, i.e., started with Nehru et al. and complemented with Barro andLee. Both methods produced similar statistical estimates. This variable is discussed further in thenext chapter.

Page 61: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

52

EXD2, Measure of weather variability:

This was constructed as the deviation of cereal yield from trend. Jaeger (1992)

has data from 1970 to 1987. This was extended to 1993. Indices of cereal production

per cotmtry are given in FAO Yearbook, varions issues. It is argued that this variable

also has the potential to measure the impact of unforeseen disnlption to cereal yield

snch é.lS the cffccts of political and social nnrest.

GDP, Gross Domestic Product:

Obtained from World Bank (1995) and earlier issues.

PRIeE, Ratio of the -index of real producer prices for exports to the index of real

producer prices of agriculture for domestie consumption:

Jaeger (1992) supplies Average Real Producer Priee for klajor Expo1t Commodities

(Index, 1980 = 100), and Real Producer Price for Major Food Crops (Index, 1980 =

100) both for the period 1970 to 1987. These are computed as Laspeyres~ indices with

1980 as base and extended to 1992 using representative export and food crop prodllcer

priees in local ctlITencies. Producer priees were obtained from \Vorld Bank (1996a)

and earlier issues. 1980 production quantities were obtained from FAO Produetion

Yearhook (1984). Consumer priee indices were obtained from World Bank (1995).

In addition ta extending the Jaeger price indices to 1992, computations were also

done for cases where the Jaeger data has missing values. Complete price data were

available for 23 countries.

AEXP, share of agricultural exports in total agricultuml output:

Computed as the ratio of the value of agricmtural exports over total agrictùtural

gross domestic product (AGDP).

Page 62: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

53

Agricultural exports are obtained froID FAü Trade Yearbook (1996) and earlier

issues. (AGDP is described above).

ADOJ\;/, domestic consumption of total agricultural gross domestic product:

This equals one minus AEXP.

Page 63: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 5

Empirical Results - ProductionFunction

This chapter presents and discllsses estimation resnlts for the meta-production fllnc-

tion. First, the simple correlation coefficients between pairs of pooled variables in

the production function are presented in Table 5.1. It is observed that factor inputs

are most higlùy correlated with agricultllral ontput. Among explanatory variables.

pairwise correlation is highest between the stock of domestic savings and private for-

eign capital, and between the stock of development assistance capital and the labor

force. OPEN is negatively correlated with AGDP, so is the correlation between the

terms of trade (EXOl) and private foreign capital stock (KPFX) and between terms

of trade and development assistance. Variations in the weather appear not to be

correlated with any variable. High simple correlation coefficients do not necessarily

imply multicollinearity problems, neither do Low coefficients indicate that there are

no multicollinearity problems. Investigation of multicollinearity is reported below.

54

Page 64: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

55

Table 5.1. rvlatrix of correlation coefficients (aIl countries)

AGDP KODA KPFX KSAV OPEN EXOI EX02

AGDP 1

KODA 0.691 1

KPFX 0.511 0.511 1

KSAV 0.618 0.428 0.767 1

OPEN -0.190 0.066 0.283 0.127 1

EXOI 0.011 -0.145 -0.11 0.047 0.105 1

EX02 0.006 0.005 0.019 0.018 0.013 -0.009 1

LABR 0.893 0.722 0.397 0.544 -0.262 0.029 -0.015

Noting that simple correlations may not be very illuminating in a multiple regres­

sion context, attention now shifts ta regression for further insight.

5.1 Regression Results

Prelinùnary tests of ordinary least squares residllals of the meta-prodllction flmction

equation (3.1) indicated that the residuals were heteroskedastic and autocorrelated.

For example, for the panel consisting of aIl cOlmtries, the likelihood ratio test yielded a

chi-square statistic of 1,124 with 26 degrees of freedom for the nlùl of no heteroskedas­

ticity. The Durbin-Watson statistic was 0.67.

The model was then re-estimated by a form of the generalized least squares (GLS)

method that corrects for heterosked.asticity and autocorrelation in estimating a con-

Page 65: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

56

sistent variance-covariance matrix. The model was estimated on a persona! computer

numing Winrats - 32 version 4.3. The variable OPEN is lagged one year, out of

concern for possible simultaneity between OPEI'! and ACDP sinee the numerator of

OPEN involves agriclùtural exports. These exports are included in AGD P, the de­

pendent variable. A consistent estimate of the variance-covarianee matrix is obtained

in the presence of heteroskedasticity and autocorrclation of rcsiduals by specifying the

L'ROBUSTERROR8" and L'LACS" options in ~Vinrats. The ROBUSTERRORS

option is important in situations where sorne forms of the GLS may be 'Linconsistent

bet~ause the regressors (or instnunents) are correlated with past residllals". (Doan.

1992).1 Snch cases have been discussed by Brown éUld ~[aital (1981) and Hayashi

and Sims (1983) among others. COlmtry-efIects and coefficients of other variables of

the meta-production ftmction equation (3.1) are estimated in one step.

Estimates are reported for the panel of all cOllntries (called LOverall'). as weIl as

for each of Western (West), Eastern-Southern (East). and Sndano-Sahel (Sndano)

agrictùtural regians. Anather set of estimates is reported far eotmtries grollped by

macroeconomic policy environment.

5.1.1 Human Capital Variable

The original specification. equation (3.1), included HCAP. Upon estimation however.

the coefficient of this variable was relatively large and significant, while the LABR

variable was insignificant in sorne cases. This was troubling given what is common

knowledge (as weIl as what this author knows about the situation on the grotmd in

l Rats Users ~Ianual Version 4.

Page 66: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

57

SSA). Forward stepwise regression approach was then used ta build the model froID

ground up, starting with LABR and adding one variable at a time. Inclusion of HCAP

in the model at any stage caused many variables to either lose their significance or

even reverse the sign of their coefficients.

Clearly, the assumption that the average years of schooling among farmers equals

the national average is tmtenable, (again based on the author's knowledge of SSA).

Indeed, the majority of African farmers have little or no formai schooling. A typical

Mrican worker who has attended school is uIÙikely to be a full time farmer. Certainly,

these farmers have acquired human capital but years of schooling does not capture this

capital. Also, at a conceptuallevel one finds it difficlùt to justify the position that a

cOlmtry with twice as many years of schooling per average worker as another, has twice

as much farming human capital as this other country (given equal pop\ùation). HCAP

was then dropped from all subsequent estimations. Thus, parameters estimated here

should be viewed as being contingent on human capital. As discllSSed in chapter

three, empirical reslùts in respect of proxies for hllman capital have been troublesome

in sorne cases.

Unfortlmately, in a multiple regression such as this, one cannot easily determine

the direction of bias, if any, on estimated coefficients when HCAP is omitted. Simple

correlation coefficients will not be enough. See Greene (1997) for further discussion.

Further, Hsiao (1986) has ar~led that use of country-specifie effects helps reduce or

even avoid omitted variable bias. The effect of the omitted variable will be absorbed

by the country-specifie dummy variable.

Page 67: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

58

5.1.2 Diagnostic Tests of Regression Adequacy

1'[any standard diagnostic tests were perforrned to assure validity of estimation results.2

They include the following.

Stationarity test

Autocorrelation functions (ACF) of all variables in the regression were computed

and plotted. For all variables, the ACF plots died down quickly. Indeed, less than

1 percent of the elements of any of the ACfs were significantly difIerent from zero.

This leads ta the conclusion that each pooled variable was stationary.

Nfulticollinearity

On an individual country basis, the labor and physkal capital variables for sorne

countries are fairly higWy correlated. But upon pooling, the correlations die clown

substantially. See table 3 for pairwise correlation coefficients of variables of pooled

data. In addition, computed variance inflation factors for the pooled variables in

the model are well below 10. One concludes therefore that mlùticollinearity is not a

problem.

Outliers

Three tests ta isolate olltlîers and inflllential observations were performed. They

are, tests for (i) observations that are olltlying with respect to the explanatory vari­

ables (leverage values), (ii) observations that are outlying with respect to the depen­

dent variable (studentized deleted residuals) and (Hi) observations that have signifi­

cant impact on parameter estimates (Cook's distance).

Leverage values were computed for individual observations. None of the computed

2 See for example Neter, Wasserman and Kutner (1996).

Page 68: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

59

values exceeded 0.12 (which is 2k/n, where k is the nlunber of coefficients estimated

and n is the nlUllber of observations). Bence, none was considered to be outlying

with respect ta the explanatory variables. Tests for outlying cases with respect to the

dependent variable were performed by computing studentized deleted residuals. This

time just under 2 percent of the observations fell within 2.5 percent of either tail area

of thp r'orresponding t distribution. Howevcr, no furthcr test, Cook's distance, sug-

gested that none of these cornes close to having had significant influence on regression

coefficients.

5.1.3 Agro-Climatic Regions

Prior to estimating the model, statistical tests for poolability of the data aver time

and across regions were performed. There was statistical support for pooling the

data as discussed below. Regression reslùts are presented in t.wo parts. First, for

the Overall and the three agro-climatic regions covering the entire period of study

in table 5.2. then for countries grouped by macroeconomic policy improvements in

table 5.3. A test for the eqllality of the capital coefficients in the West. was rejected.

Disaggregated capital coefficients are thus reported in all cases.3

3 Given the large number of parameters being estimated, the risk of type 1 error is controlled at 0.01in ail of the discussions below.

Page 69: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Table 5.2: Estimates of meta-production function parameters1 for

agricultural regions, (1970 - 1993). See legend on next page.

Variahlps2 Ovprall3 West4 East5 Sudano6

ln(KODA) 0.054 -0.006 0.026 0.077

(1.584) (-0.064) (0.483) (1.007)

ln(KPF)() 0.021 0.221** -0.039 0.090*

(0.086) (5.226) (-1.349) (2.299)

ln(KSAv') 0.030 -0.114 0.067 -0.008

(1.060) (-1.662) (1.726) (-0.168)

ln(LAER) 0.278 0.908* 0.782** -0.249

(1.827) (1.951) (4.411) (-1.036)

OPEN -0.033 -0.186 -0.358* -0.065

(-0.394) (-1.721) (-2.102) (-0.478)

EXOI 0.017 0.043 0.066 -0.028

(0.371) (0.946) (1.321) (-0.180)

EX02 0.108* 0.075 0.0104 0.095

(2.263) (0.783) (1.711) (1.069)

TREND 0.003 0.008 -0.004 -0.006

(0.782) (1.195) (-0.582) (-0.790)

D. F. 560 139 224 181

60

Page 70: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1 t-statistics in parenthesis below coefficient estimates.

* and ** indicate statistical significance at 0.05 and 0.01

levels respectively. D. F. is degrees of freedom.

2 Variables are defined in chapter 3.

30verall: West, East and Sudano together.

.tWest: Benin, Cameroon, Cote d'Ivoire, Ghana. Nigeria, Sierra Leone,

and Togo.

5East: Bostwana, Burundi, Ethiopia, Kenya. l\Iadagasca, rvlalawi,

~Iauritius, Rwanda, Tanzania, Zamhia and Zimbabwe.

6Sudano: Burkina Faso, Central Africa Republic, Chad, Cambia,

rvlali, lVlallritania, Niger, Senegal and Slldan.

61

Page 71: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

62

Table 5.2 (continued): Estimates of fixed country-specifie

effects for agricultural regions. Continues next page.

Countries Overall1 West2 East3 Sudano4

Benin 4.215** 5.528**

Cameroon 4.278** 6.510**

Cote d" Ivoire 4.879** 6.752**

Ghana 4.924** 6.600**

Nigeria 4.607** 5.888**

Sierra Leone 3.545** 5.000**

Togo 4.379** 5.753**

Bostwana 1.952* 5.142**

Bllflmdi 5.180** 5.370**

Ethiopia 3.367** 4.271 **

Kenya 3.981** 5.578**

~Iadagascar 3.117** 4.627**

rvlalawi 3.602** 4.585**

rvlauritius 6.500** 7.560**

Rwanda 5.344** 5.432**

Tanzania 3.121** 4.374**

Zambia 2.105** 4.289**

Zimbabwe 3.454** 4.951**

Page 72: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Table 5.2 (continued): Fixed country-specifie effects for"

agricultural regions.

Countries Overall1 West2 East3 Sudano4

Burkina Faso 3.949** 3.131**

Central Afriea Rep. 2.993** 1.074

Chad 2.311 ** 0.289

Gambia 4.355** 3.972**

1vlali 3.616** 2.232**

1vlauritania 2.544** 0.180

Niger 4.173** 3.295**

Senegal 4.300 2.450**

Sudan 4.097** 2.450**

* and ** indicate significance at 0.05 and 0.01 respeetively.

1,2,3,'1 Overall, West, East and Sudano are defined above.

63

Page 73: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

64

Parameter Estimates

Wald tests for stnlctural change were performed as discussed in chapter three. There

was not mnch evidence against parameter stability. For example, Wald tests for

changes in slope pararnet.ers hetween 19iO - 1980 and 1981 - 1993 yielded f'hi-sqnared

statistics of 12.52 (p-value 0.09), 7.49 (p-value 0.3S), and 20.12 (p-value 0.01) for the

West, East and Sudano regions respectively, for the nlùl hypothesis of no change in

the slope parameters.

Similarly, tests for differences in the slope parameters during the period 1970 ­

1980 between the West and the East is 19.42 (p-vall1e 0.01); between the West. and

Sndano it is 17.07 (p-value 0.02); and between the East and Slldano 11.47 (p-value

0.12). Thirdly, when data for all three regions were pooled and the test for stnlctural

change between 1970 - 1980 and 1981 - 1993 was performed, the chi-squared statistic

for the nnll of no stnlctural change was 12.89 (p-valne O.OS), lending support to

the nlùl. This would imply that stnlctural adjl1stment programs which have mostly

taken hold since the beginning of the 19S0s, have not significantly impacted regional

agric\ùtural production.

There was sorne evidence for change in model parameters between 1970 - 1985 and

1986 - 1993 when all cOlmtries were pooled. But given the rednced size of the sample

for the second period, rednced emphasis was placed on this result and parameters

were estimated for the entire period 1970 - 1993.

Indeed, in discussing data pooling, Baltagi (1995) argues that sometimes the ques-

Page 74: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

65

tion is "to choose on 'pragmatic grolmds' between two sets of estimators ... and hence

achieve in a sense, one of the main motivations behind pooling". Continuing, he

quotes Toro-Vizcarrondo and Wallace (1968, p.56D) who write, "... if one is willing

to accept sorne bias in trade for a reduction in variance, then even if the restriction

[Le., pooling} is not trlle one might still prefer the restricted [Le., pooled] estimator" .

Thc finding of insnfficicnt c\idcncc in favar of parameter change suggests that

the structure of agriclùtural production has not changed much in any of the three

regions over this period. This goes to buttress the argument that agriclùtnre has

been receiving little attention. Farming methods and the other difficnlt situations

that farmers have faced in the past apparently persist. Indeed. the coefficient of the

trend variable, the measure of total factor prodnctivity growth. is not signifieant in

any region (presented in table 5.2). The findings with respect to the policy gronps

indicate a somewhat different picture and are reported Iater in table 5.3.

Overseas Development Assistance, KODA: KODA has positive impact on

agriclùtural output in the Overall, East and Sudano specifications. However. the

magnitude of its coefficient is only a small fraction of that of the labor variable in

each region. The p-values for the t-statistics of its coefficients are 0.11, 0.95. 0.63.

0.31 in the Overall, West (where it is negative), East and Sudano respectively. This

leads ta the conclusion that this variable is not significant in determining agrictùtural

output in any region. The finding for the Overall data is probably stronger than is

implied in LeIe (1990) and Burnside and Dollar (1997), but weaker than Glmjal and

Gichenje (1997). LeIe concludes from six SSA country case studies that donors (Le.,

ODA) have "surprisingly small impact on agricultural development ...." Burnside

Page 75: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

66

and Dollar (1997) conclude from their study of aOA and economic growth in 56

developing countries (which included 21 SSA cOlmtries) that ODA has no positive

effect on growth in the presence of poor policies. While Gunjal and Gichenje (1997)

report a coefficient of 0.045 Overall, which is close in magnitude to the finding of this

study, but with 0.01 significance level. Thus one may conclude that the impact of

rlpvplopment. ~",~istan('f' on agrirnltnral output. is small and within reasonable range

of previons estimates.

Foreign private capital flows, KPFX: The impact of this variable on agriclùtllre

varies across the regions. It is small and positive but clearly not significant in the

Overall. It is however, positive and significant in the West (p-value < 0.01) and

practically so in the Sudano (p-value 0.02). In the West, its magnitude (0.22) is large

(compared to other physical capital variables) but only about one-qu~U'ter the Bize of

the labor variable. In the Sudano, it is the factor input with the largest coefficient.

KPFX is negative but not significant (p-value 0.18) in the East. The significant impact

of private foreign commercial flows in the West may be due ta the more prominent

role of agrictùtural exports in sorne of these economies. This region grows permanent

tree crops with well established history of exports earnings which are required ta

service PFX. (Agricultural output grew at 2.5% annually in this region over the

period). Cocoa and coffee in Cote d'Ivoire, Ghana, Cameroon, Togo, and Nigeria4 ,

virtually guarantee export earnings. Also, one may be picking up the tendency for

sorne PFX to go directly into agriclùture in Cote d'Ivoire and Cameroon. Agriclùture

4 Large oil exports complicate the issue for Nigeria.

Page 76: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

67

in Ghana too, may have henefited indirectly from PFX flows into the COlmtry during

the earlier and later years of tlùs study. (There is no substantial foreign investment

in agriculture in Ghana). While there is a positive relationship between KPFX and

agrÏClùtural output in the Sudano, it is unlikely that PFX was invested directly in

agriclùture. Rather agriclùture may have benefited fron1 PFX investments in mining

in t.his rpginn. Th~ negat.ive impact of this variable in the East suggcsts allocation

or re-allocation of PFX out of agriclùture (or agrictùture related activities) due ta

declining international primary agrictùtural cornmodity priees. It wotùd appear that

by pooling all regions, one dilutes the positive effect of K PFJ'y seen in the West (and

Sudano).

Domestic savings, KSAV: K S AV is not significant in the Overall or in any re­

gion. The p-values of the t-statistics of t.he coefficients are 0.29. 0.10, 0.08, 0.87 in the

Overall, West, East and Slldano respectively. The coefficient is positive in the Overall

cmd in the East ooly. It is negative in the West and Sudano. As a factor input, one

would have expected its coefficient ta be positive in ail cases. The positive coeffi­

cient in the East may be explained by the faet that, in the East. producer priees of

agriclùtural exports reflect international priees more than in any other region. (Note

the apparent contradiction with PFX. However, this is consistent with the argument

that domestic savings and private foreign capital are likely to have different impact

on agriclûture). This would impact positively on priees of domestic agriculture. since

otherwise a11 farmers will grow exports. By reflecting international priees, the effect

is ta recluce the bias of domestic terms of trade against the agricultural sectar in

Page 77: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

68

this region. Thus local investors in the East find investments in agriclùture relatively

more profitable than in the West for example. In most countries in the West, the

damestic terms of trade is biased. substantially against agriculture. Thus domestic

investors in the West find agriclÙtural iDvestments lmprofitable and they allocate or

re-allocate resources ta other sectors of the economy at the detriment of agriclùture.

Indccd. Knleger ct al. (19!H) report that direct and indirect taxes on agricultural

exports averaged 51.6 percent in Ghana, Cote d'Ivoire and Zambia over the period

1960 - 84.5 The bias against agriclùture has not improved much in recent years

in spite of liberalization policies in a nllmber of cotmtries. Ghana for example has

recently been described as I4A nation of traders". (Centre for Policy Analysis. 1997),

referring to the buoyant retail trade in imported merchandise. Another example is

Nigeria, where the availability of ail contriblltes to the neglect of agriclùtl1re.

K S AV is very small and not significant in the Sudano regression. Here tao.

peasant agriclùture is not profitable being in the Sahel (near desert) region. On the

whole therefore, it wo\ùd appear that domestic savings tao have rniIÙmal or negative

impact on total agriclùtural output in SSA.

Labor, LABR: The labor variable is positive and almost significant in the Overall

(coefficient 0.28, p-value 0.06). In the West tao, the coefficient is positive and large

(0.91) and almost significant (p-value 0.05). The figures for the East are, coefficient

0.78, p-value less than 0.01. In all these cases, labor is the factor of production

5 Cote d'Ivoire and Ghana are in the West. Zamhia is in the East. However, agriculture is onlya small part of the Zamhian economy, currently about 15% of GDP. The equivalent percentage forGhana and Cote d'Ivoire is about 35.

Page 78: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

69

with the largest coefficient except in the Sudano where it is negative. Positive and

significant labor coefficients are in accord with the observed situation on the grolmd,

and are also predicted by economic theory. The non-significance of the coefficient of

labor in the Sudano may he due to peculiarities of this region (Sahara desert belt

of Africa little snited for agriclùture, sparse poplùation, civil strife, etc.). It is also

possible that poor data qnality (for this region) is contributing ta this finding. 6

Openness, OPEN: The variable OPEN is negative in ail regians. It is not signifi-

cant in the Overall, West or Sudano, but is practically significant in the East ( p-vallle

0.03). Given that enrrent theory Sllggests that opeIUless is conducive to g;rowth, one

may wonder what cOlùd be happening here. It is noted that the benefits ta apenness

arise when competition induees increased efficiency in the nse of factor inputs. ete.

By losing export market share, in the face of increased openness, one snnnises that

the efficiency with which factor inputs are use<! must be lower in SSA than it is in

competing regions. (Ng and Yeats, 1997, disCllSS this and present evidence showing

that between 1962 - 1964 and 1991 - 1993, SSA lost a reasonable portion of its share

of üECD agriclùtural imports. By losing export market share in the face of increased

openness, one surmises that it is the relative inefficiency in the use of factor inputs,

in agriclùture in SSA that is driving the negative coefficient of OPEN. What is more,

to an endogenous growth theorist, the negative coefficient of this variable may be

capturing the fact that these economies have either stagnated or retrogressed.

Here is another view of what is happening in SSA. Recall the definition of OPEN

6 Dropping Sudan because of long ranging civil war from the Sudano sample did not change mattersvery much.

Page 79: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

70

as trade (imports and exports) share of GD? Thus since gross GD? has in general

either increased. or remained steady, a higher value of OPEN means higher value

for imports and exports. However, most countries in SSA continue ta suffer negative

trade balances, most of the imported items being consnmables for retail trade. For

many, increases in total exports have been due to higher minerai exports. Bence as

npenn~s has reslllted in more imports, agric1.ùtural output has suffered.7 This ties

in with allocation of resonrces away from agrictùtllre as discussed above.

To obtain another view of this variable, its interactions with both KODA and

K P FX were considered. The new variables were called K 0 _0PEN and K P_0PE IV

respectively. One finds that the coefficient of KO_OPEN is positive and significant

in the Overall regression. ft is positive and almost significant in the West, but c1early

insignificant in both the East and Sudano. Realizing that 0 D.4 is a rnL'Ced bag of

grants (possibly to relieve hardship) and concessional loans (these days as a response

ta improvements in policy)! it is not clear how to argue on economic grotmds one way

or the other for a positive (or negative) value of the coefficient of this variable. The

argument is more straight Corward when K P _0PEN is considered.

KP_OPEN is negative but not significant in the Overall. 1t is negative and

significant in the West. The variable is not significant in the East, while significant

and positive in the Sudano. Thus, on the whole, one might say that private foreign

capital elasticity of agriclûtural output falls or does not increase the more open the

econoroy. This would not have been expected.8 Things may yet change however.

1 Food imports are small but on the încrease.8 One notes however that in the specification with the interaction terms, private foreign capitale1asticity of agricultural output defined as (:t: :~~k) is the SUIn of two tenns. For the West, (theonly place with significant values) the actual value of this elasticity over the period however, is

Page 80: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

71

These restùts are essentially capturing what has happened in the pasto With the

adoption of new policies, the impact of private foreign capital on agriculture might

change if this sector becomes as competitive as the other sectors, (mining for exampie).

For another insight, this variable is further investigated below when cotmtries are

grouped by macroeconomic policy improvement.

External Shocks, EX01: The external terms of trade variable. EXOI. is weak

(p-values more than 0.18 in all cases). The data show that SSA's external terms of

trade have been falling. The stnlcture of African economies however has been snch

that farmers in most co\mtries are not iInpaeted directly by changes in the terms of

trade, sinee it is domestic governments who set producer priees of agricultllral exports

especially. AlI t.he same, these governments must no donbt consider this variable in

setting domestic producer priees. Deaton and ~Iiller (1995) have suggested that

governments try to avoid having to lower domestic priees in response to falling terms

of trade. The argument is that, governments expecting the terms of trade to falI

wotùd no donbt set today's produeer priees in anticipation of lower t.erms of trade in

the future. This is not the full picture however. Governments may not have lowered

domestic nominal priees in response to falling terms of trade, but falling terms of trade

mean lower revenues from external trade with whieh to make public investments (sueh

as provision of infrastnlctnre), mostly based on imported inputs. Even in cotmtries

where domestic priees track international prices more c1osely, falling external terms

of trade for primary agrieultural exports make the agricultural sector less attractive.

+0.08, showing a positive impact on average.

Page 81: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•72

Weather Variability, EX02: The EX02 variable is almost significant in the

Overall (p-value 0.02) and fairly strong (p-value 0.08) in the East. It is weak in the

West and Sudano. That this variable is positive and fairly strong emphasizes the

dependence on nature in SSA agriclùture. By and large, irrigation (or use of high

yielding seeds) is Ilot at ail significant in the overall picture. Reducing the current

state of complete cxposnrc to the clements is onc fcature of 5ub-Saharall Afrkall

agriclùture that caIls for long-term plans.

Trend and Country Dummy Variables: The coeffident of the time trend is

not significant in any regression. This wotùd sllggest that Hicks nelltral technkal

progress has not been significant in SSA agrictùture.9 Thtls~ increases in AGDP are

coming from increases in inputs only (mostly labor). Trus, theory suggests, restùts in

diminishing returns. COtUltry dummy variables are all significant in the Overall, East

and West. Confirming the hypothesis of significant tUlobserved cotmtry effects. Three

out of nine COtUltry dummy variables in the SUdéUlO region are not. Non-significance

of sorne cOlmtry dtunmy variables (or similarity of estimates), may suggest that these

cotmtries have enough cornmon characteristics to assign them a single dummy, but

trus is not purslled.

5.1.4 Policy Groups

The countries of LARGE, SlVIALL and POOR poliey groups are listed below. To

compare how things stood during the period of interest 1981 - 1993 against the pre-

9 One recognizes the argument of embodied technical progress. Howevert in this same context, onefinds in the next two chapters that Hicks neutral technical chaIlge is significant in export agricultureand domestic agriculture share equations.

Page 82: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

73

ceding period 1970 - 1980, Wald test fol' stability of slope coefficients over the two

periods were performed. For the LARGE group, the clù-squared statistic for the ntùl

of no change in slope coefficients was 38.14 with a p-vallle less than 0.001. Thus, the

nu11 is rejected. Two variables are identified as having slope coefficients that are sig­

nificantly different between the two periods. KODA is sigrùficantly larger during the

second period than the first, while KPFX is significantly smaller during the second

period than the earlier period.

For the SrvIALL group, the chi-squared statistic is 11.43 with a p-value of 0.13,

hence one concllldes there has been no change in the slope parameters over the two

periods. For the POOR group, the chi-squared statistic was 31.40 with p-vallle far

less than 0.001. Hence here tao, one conc1udes that slope parameters have changed.

This time too KODA is identified as being significantly larger during 1981 - 1993

than dllring 1970 - 1980, while KSAV is almost significantly lower during 1981 ­

1993 than 1970 - 1980. For LARGE and POOR, the findings contradict that for

agricultural regions where one could not. reject the nnU of no change in parameters.

This is hardly surprising however. In grouping by agriclùtural regions, one is taking

advantage of a common characteristic of the countries, their c1imate which wo\ùd not

have changed over an eleven year period (1970 - 1980). In grouping into LARGE,

S~IALL and POOR, during the period 1981 - 1993 another characteristic is used.

ehanges in maeroeconornic poliey. However one has no information about any cornmon

eharacteristic that they countries may have shared during 1970 - 1980. Certainly,

eaeh grouping eomprised cOlmtries from different agric\ùtural zones. The rest of this

discussion focuses on the period 1981 - 1993, the period of economic reforms.

Page 83: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

74

Next, one tested for poolability across policy groups over the period 1981 - 1993.

The test suggests that one cotùd ooly pool countries belonging to the LARGE and

SlVIALL groups. Significantly higher positive slope coefficients of KODA, KSAV and

EX01 in the POOR group are identified as the reason for this finding. (These are dis­

cussed below). This finding suggests convergence of sorts of those groups of cOlmtries

t.hat. are responding to poliC'y rflforms. Based on these findings, data for LARGE and

Sl\tIALL cOlmtries are pooled. POOR is presented on its own. A statistical test for

the equality of the coefficients of ln(KODA) and ln(KPF)() for the POOR group

rejects equality.

Page 84: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Table 5.3: Parameter estimates for meta-production function 1

for Pol-icy G'roups, 1981 - 1993.

Variables2• POOR3 4SrvIALL/5 LARGE

ln(/<ODA) 0.514** 0.120

(3.945) (1.342)

ln(KPFX) 0.040 -0.0351

(0.576) (-0.543)

ln(KSAV) 0.111 -0.191*

(1.508) (-1.961)

ln(LABR) 0.623 1.041

(1.316) (1.639)

OPEN -0.182 0.146

(-1.310) (1.186)

EXOI 0.298** -0.447**

(4.318) (-4.145)

EX02 0.085 0.222**

(0.680) (4.058)

TREND -0.012 -0.030

(-1.343) (-1.854)

D. F. 58 135

See legend on next page.

75

Page 85: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

76

l t-statistics in parenthesis below coefficient estimates.

* and ** indicate statistical significance at 0.05 and 0.01

levels respectively. D. F. is degrees of freedom.

2 Variables are defined in chapter 3.

3 POOR: Benin, Cameroon, Cote d'Ivoire, Rwanda, Sierra Leone and Zambia.

4SMALL: Central Africa Republic, Kenya, rvlalawi, l\Iali, Niger, Nigeria,

Senegal and Togo.

5LARGE: Gambia, Ghana, l\tladagasca, Tanzania, and Zimbabwe.

Page 86: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

77

1 Table 5.3 (continued): Fixed country-specifie effeets for Policy groups.

Country POOR S~IALL/LARGE

Gambia 6.901**

Ghana 6.704**

~Iadagascar 8.158**

Tanzarua 7.926**

Zimbabwe 6.457**

Central Afriea Republic 6.319**

Kenya 6.904**

!vralawi 6.954**

~Iali 8.264**

Niger 7.094**

Nigeria 6.541**

Senegal 7.565**

Togo 7.332**

Benin 2.283

Carneroon 3.090

Cote d '1voire 3.209

Rwanda 1.438

Sierra leone 1.592

Zamhia 1.310

* and ** indicate significance at 0.05 and 0.01 respectively.

Page 87: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

78

Regression Parameters

Parameter estimates for the combined LARGE and S~IALL and the POOR groups

are reported in table 5.3 and discussed next.

Group: Combined LARGE & SMALL. These group of COlUltries are dassified

as having shawn positive improvenlent.~ in thf'ir ffiRf'rOPConomir poliC"'y f>nvironment

between the years 1981 - 1986 and 1987 - 1992. The coefficient of the stock of aOA

is positive, small, (about one-tenth the labor coefficient), but not signifieant (p-vallle

0.16). The coefficient of the stock of PFX is negative, small and not signifieant (p­

value 0.59). The stock of domestic savings has a negative coefficient that is one-fifth

the magnitude of the labor coefficient. It is practically significant (p-value 0.05).

Thus. the three forms of capital have different impact on agrîcultllral outpnt. The

negative coefficient of domestic savings is partictùarly t.ronblesoIne. That increases

in domestic savings are having detrirnental effect.s on agrictùture speaks t.o (i) no

investments of domestic savings in agrictùture, émd/or (ii) investments of domestic

savings in other sectors and in the process having negative impact on agrictùture.

The second cornes about for example, when the flow of investments to other sectors

(retail trade, say) restùt in farmers having reduced access to credit (which is already

very low).

Positive policy environment in this context means the economy has become more

liberalized in many respects. One may see this as rednction in indirect taxation

(Le., reduction in domestic currency over-valuation) of agriclùtural output. However,

direct taxes (government price fixing) for the major items - coffee, cocoa, etc. - still

Page 88: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

79

existed over this period. Unfortunately, the supposed reduction in indirect taxation

wOlùd not benefit farmers that much as long as governments continue to set prices

low. To this, one adds that devaluation was generally inflationary and as a farmer,

being on fi.xed prices is no way to keep up with inflation.

Substantial amolmts of development assistance (concessionalloans) may have been

rPrpivpd hy t.hpse governments during this period. This nnalysis shows that thcsc did

nat impact the agricultural sectar significantly. The story with respect ta PFX fiows is

that, the period under consideration coincided with private foreign capital flight (both

foreign direct investment and bank loans) out of SSA. Over the period, agriclùtural

output did grow at a modest 1.5%, hence the negative relation with (generally) KPFX.

The labor coefficient (LAER) is once again the largest factor input c.ontributing

positively to agrÎcnltllral outpnt. It is however not significallt (p-value 0.10). lndeed.

as discussed in the introduction, agriclùture as practised in SSA has traditionally

had a high labor component. But more activities in other sectors of the economy

engendered by liberalization, have been attracting YOlmger and more energetic people

away from farming.

OPEN is positive but not significant (p-value 0.23). Openness has meant. reduced

direct taxation on agriclùture as a whole, but more sa for domestic agriclùtllre, (gov­

ernments not fixing priees for this sub-sector) and may thus he encouraging output.

However, by the coefficient not being significant, the impact of openness has not been

as strong as is implied by theory.

In the specification involving the interactions between openness and foreign capital

flows, it turns out that KP_OPEN is positive (0.12) but not significant (~value

Page 89: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

80

0.20), while KO_OPEN is negative (-0.05) and not significant (p-value 0.45). This

suggests that the elasticity of agriclùtural output with respect to private foreign

capital increases with openness in good policy environment. Opening up economies

ta the rest of the world is a major pIank of the policy reform package by which these

cOlmtries have been judged as having done weIl. This is what growth theory wOlùd

prcdict. It has bccn argucd that it is not obvious how ta explaill t.he KO_OPE1\/

variable, given the composition of ODA.

Back to table 5. The coefficient of the ternIS of trade (E..Y01) is negative, large

(-0.44) and significant (p-value < 0.01). The finding here must be picking up the

positive trend in agriclùtural output (annual growth rate of about 1.5%), against

falling terms of trade of primary agriclùturaI exports over the period. (This may aIso

explain why domestic savings will be invested in those sectors of the economy with

more favorable international (in addition to domestic) terms of trade). The weather

variable (E..Y02) impacts agriclùtural ontput significantly. The inlpact is positive.

This finding speaks ta the extent of reliance on nature. The trend variable is negative

but it is not significant. This suggests deterioration in total factor prodllctivity in

agrÏclùture. This may not be surprising when viewed in the context of insignificant or

negative physical capital coefficients. AlI COlmtry dummy variables for the combined

LARGE and Sl\IALL regression are significant.

Group: PODR. Peculiar things happen here. While labor is pùsitive and not sig­

nificant, this is the only group in which KDDA is positive, large (0.51) and significant

(p-value < 0.01). KPFX is positive, small (0.04) but not significant (p-value 0.56).

Page 90: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

81

KSAV is alse positive (0.11) but net significant (p-value 0.13). The cOlmtries in this

group include Benin, Cameroon, Cote d'Ivoire, Rwanda, Sierra Leone and Zambia.

Thus, growth in agricultural output in these cOlmtries increases with OOA, KPFX

and SAV. Their policy environment may be judged to he bad, but they seem to have

made the crucial connection between agriclùtural output and capital. Indeed, the

annual growth rate in agrictùtural output per hectare for this group is 2.1% (over

the period 1981 to 1993), larger than for the cOlmtries judged to have positive policy.

Of the three, KPFX is smallest and weakest probably for the same reason that this

variable is not significant in most other specifications. The general tendency is for

private foreign flows to go into trade oriented sectors. The unusual positive impact

of KODA on agriclùtural output is not easy ta explain. We can aoly guess that it is

being targeted at agriculture specifically.

The labor coefficient is positive, large (0.62) but not significant. The variable

OPEN is negative but not significant. This follows from being classified as poor pol­

ieyenvironment. Here, the external terms of trade are significant and positive llnlike

our finding for the other group. Poor policy environrnent implies poor performance

in respect of fiscal, monetary and exchange rate policies. Three of the six cOlilltries,

(and practically a fOluth, Rwanda), are members of the CFA currency zone, where

the CFA was considered to have been over-valued for most of the period tmder consid­

eration (UNCTAD, 1997). Given this environment, the exc.hange rate for conversion

of agricultural output to United States dollars will influence the outcome more than

in other regions or groups. One thinks this is probably what EX01 is reflecting. lO

10Also, included in this group is Cote d'Ivoire which has tried in the past ta hold off cocoa exports

Page 91: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

82

The weather variable has no explanatory power this time, suggesting other variables

have stronger explanatory power. The trend coefficient once again suggests nega-

tive productivity growth. The coefficient is however not significant. This suggests

higher total factor productivity than in the LARGE and SlVIALL group of countries.

The finding is however consistent with the positive impact of the capital variables.

Conntry effects are significant, but the test that they are jointly zero is rcjectcd..

Next, the preceding reslùts are summarized.

5.1.5 Summary

This chapter has reported and discllssed estimation of parameters for the meta-

production nmctions of agriclùtural output in 27 sub-Saharan African eOllntries.

Estimates were obtained for (i) aIl cOlmtries pooled together, (ii) cOlmtries pooled

together by agro-climatic region, and (iii) cOllntries pooled together on the basis of

improvement in macroeconomic policy environment. The production ftmction is cast

within the framework of neoclassical and endogenous growth models. Physical capital

is disaggregated into the stock of official development assistance, the stock of private

foreign commercial fiows, and the stock of domestic savings in order to focus on the

impact of each component. Labor, openness of the economy, external terms of trade

and weather variahility are the other regressors. The proxy for human capital, the

average years of schooling within the labor force was judged to be lUlsatisfactory in

captluing human capital of farmers and was therefore excluded.

While statistical tests suggest that the slope parameters are unchanged over the

during periods that they consider the world priee of the commodity to be too Iow. In such a situation,measured output will tum to vary positively with terms of trade.

Page 92: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

83

two periods 1970 - 80 and 1981 - 93, (periods without and with structural reforms

respectively), for each agricultural region, the evidence is diffenmt for policy groups.

For these (policy groups), significant differences exist between the slope coefficients

of the LARGE group over the two periods. The same holds for the POOR group. For

the 1981- 93 period, however, significant differences in slope coefficients exist between

C'ornhinpo LARGE and SIvIALL on one hand, and POOR on the other. Differences

in the slope coefficients of the capital variables and openness between the two are

identified as being statistically significant.

One finds that the impaet of development assistance is positive everywhere except

in the West. This variable has significant impact on agricllitlual output only in t.he

POOR group. In ail the regions as weil as in the combined LARGE and S~IALL

group. the magnit.udes are only a small fraction of the impact of the labor variable.

The magnitude is mllch doser to the labor coefficient in the POOR group. Private

foreign capital has positive impact on output in aIl regressions except those for the

East and combined LARGE and SrvIALL. The impact is signifieant ol1ly in the West

and Sudano. The magnitudes are about the same as for development assistance.

Domestic savings are positive in the Overall, East, and POOR groups, but not signif­

icant anywhere, though almost so in the combined LARGE and S~IALL where the

coefficient is negative. Here too, the magnitude of the impact of this variable is about

the same as for the other capital variables.

Labor is the variable with the highest impact in all cases. Its impact is positive

and significant in all regions but the Sudano-Sahel region where it is negative. It is

positive but not significant in the policy groups. On the whole, the openness measure

Page 93: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

84

bas negative or insignificant impact on agricultural output. With respect to the agro­

climatic groups, the terms of trade have no significant impact on agricultural output.

Its impact is significant for the poliey groups: large and positive for POOR, and

large and negative for combined LARGE and SMALL. On the whole, variations in

the weather have practically significant explanatory power for changes in agrictùtural

output. Technical change is virtually non-existent within the agricultural regiuus

(bath in magnitude and significance). It is negative in both policy groups, (and

alrIlost. significant in the combined LARGE and Sl\IALL).

The next two chapters present and discuss estimation restùts for the static and

dynamic export and domestic shares of agriclùtural output.

Page 94: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 6

Empirical Results - Analysis ofAgricultural Share Response in theShort-Run

This chapter discusses estimation of the static export share eqnation (3.8) with the

time-specific dlunmy variables replaced by a Ume trend. The response of the ex-

port (and domestic) shares of agriclùtllre ta prodllcer priee changes, factor inputs.

economic policy and external shoeks is investigated in order to nnderstand why agri-

cultural exports as a share of total agricultural output has been falling over the years.

This has been so in spite of anecdotal evidence that Foreign financial flows, recent eco-

nantie recovery programs and government polides favor export agrictùtllre relatively

more than agriclùture for domestic consnmption. Short-nln responses are cornmon in

the literature and are of interest in short term impact evaluation of policy and priee

changes. As weil, short term impact of changes in the capital and other variables are

of interest here.

The complete panel consists of 23 out of the 27 co\wtries previously used. in the

production function investigation. Producer priee data ww) not available for 4 of

those countries. Pairwise correlation coefficients between the variables are similar to

85

Page 95: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

86

those presented in table 3. In addition, export share (AEXP) is mildly negatively

correlated with AGDP, KODA, EX02 and LABR. It is positively correlated with

OPEN, KSAV, PRIeE, KPFX, and EXOI. Again, for a more complete understanding

of the relationship between the dependent and independent variables in the model

one turns to regression analysis.

6.1 Regression Results

As done previously, iIÙtial estimation and testing of the model by ordinary least

squares yielded heteroskedastic and alltocorrelated residuals. A generalized least

squares method was then llsed to estimate cOlmtry-effects and coefficients of other

variables of the export share eqllation in one step in ~Vinrats - 32 versian 4.3. 1

6.1.1 Diagnostic Tests of Regression Adequacy

Following the steps taken in the previons chapter for the meta-production fllnction,

standard diagnostic tests were performed to assure validity of estimation reslùts. Au-

tocorrelation ftmction tests snggested that the variables in the model were stationary

in levels, while variance inflation factors compllted for ail explanatory variables were

all weil below 10.

Outliers

Once again, tests for, outlying cases with respect to the regressors (leverage val-

lles of observations) and dependent variables (studentized deleted residl1als) were

performed. No significant leverage values were determined. Just about one percent

1 Again, for a consistent estimate of the varianc~covariancematrix in the presence of heteroskedas­ticity and autocorrelation ofresiduals t the uROBUSTERRORS" and "LAGS" options in Winratsare specified as previously described.

Page 96: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•87

of the studentized deleted residuals feU within 2.5% of either tail of the corresponding

t distribution. Finally, computation of the Cook's distance statistics did not suggest

that any observation had significant influence on parameter estimates.

6.1.2 Agro-Climatic Regions

Regression results are presented in two parts. First, for the Overall and the three

agro..climatic regions covering the entire period of study, then for countries grouped

by macroeconomic policy environment. One year lagged OPEN and PRIeE were

used as instrumental variables for CUITent values of the two variables.

Parameter Stability

For agriclùtural regions, Wald tests for stability of slope coefficients were performed as

before for stability over time within each region. Sample results are the following chi­

sqllared statistics: 18.09 (p-value 0.02), 20.57 (p-value 0.008), 22.10 (p-value 0.006)

and 16.44 (p-value 0.03) for the East, WP.St., Sndano and Overall respectively. for

stability of slope coefficients between 1970 - 80 and 1981 - 93. The chi-squared statistic

for poolability of data between pairs of regions are 17.11 (p-value 0.02) for poolability

between West and Sudano, 13.74 (p-value 0.09) for poolability between West and

East, and 15.55 (p-vall1e 0.05) for poolability between East and Sudano. Thollgh the

values of sorne of these tests throw sorne doubt on the validity of poolability of the

data on statistical grolwds, one goes ahead and pools the data on pragmatic grolmds

as previously discussed. As before, parameter estimates are reported for the Overall,

West, East, and Slldano agriclÙtural regions over the entire period (1970 to 1993) in

table 6.1. The findings with respect to the three poliey groups are different. These

Page 97: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

88

are reported later in table 6.2.2

Parameter Estimates

Estimated parameters for this section are presented in table 6.1 for the export share

equations. For the Overall data, estimation with 23 cOlmtry dtunmy variables yields

t statistics for 17 cOlmtry dllmmy variables were not significantly different from zero.

The values of five of these dllmmy variables were clearly close ta each other. Upon

testing far eqnality of these five, one cOlùd not reject the nlùl hypothesis that they

were equal. The test that they were jointly zero was rejected however. These five

were then pooled to form one dununy variable, and the equation re-estimated. The

impact of this was to change the coefficient and standard errar estimates of sorne

variables somewhat. One reasons that the variances of the conntry dl1mmy variables

were not efficiently estimated initially, possibly due to collinearity between couutry

dummies and sorne variables. By combining the dummy variables. one opts for a

more efficient variance. The priee of this however, is possible bias in coefficient

estimates. As discussed earlier, the rationale for doing so has been discussed by

Baltagi (1995) and others. In the Sudano estimation, two pairs of COlmtries were

assigned a common dummy. While for the West, three sets of cauntries (made np af

four and twa cOlmtries) were assigned one dummy per set. In the East, ail countries

retained their individual dummy variables as all were statistically significant and fairly

different from each other. Equality of the coefficients of In(KPFX) and ln(KSAV)

was rejected. for Sudana.

2 Given the large number of parameters being estimated, the risk of type 1 error is again controlledat 0.01 in all of the following discussion.

Page 98: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

89

1 Table 6.1. Parameter estimates for export share regression1 -

Agricultural regions (1970 - 1993).

Variable2 OveralP West4 East5 Sudano6

ln(KODA) 0.026 0.003 0.075 0.041

(1.354) (0.097) (1.888) (1.417)

ln(KPF.X) 0.019 -0.010 0.047 0.024

(1.377) (-0.340) (1.612) (1.777)

ln(KSAV) 0.013 0.010 0.114** -0.042**

(0.681) (0.362) (3.198) (-2.506)

ln(LAER) 0.035* -0.153** -0.065 0.019

(1.959) (-5.737) (-0.458) (0.288)

OPEN 0.380** 0.367** 0.624 0.722**

(5.006) (3.104) (1.575) (3.648)

EXOI 0.076** 0.062* 0.065* 0.103*

(3.420) (2.180) (2.009) (1.986)

EX02 -0.045 -0.072 -0.046 -0.063

(-1.260) (-0.0976) (-.878) (-1.794)

ln(PRICE) 0.019 0.062* 0.099 -0.020

(1.023) (2.131) (1.476) (-0.364)

TREND -0.012** -0.101** -0.017** -0.009**

(-6.250) (-4.705) (-3.904) (-2.528)

D. F. 478 142 180 140

See legend on next page.

Page 99: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1 t-statistics in parenthesis below coefficient estimates.

* and ** indicate statistical significance at 0.05 and 0.01

leve1s respectively. D. F. is degrees of freedom.

2 Variables are defined in chapter 3.

30verall: West, East and Sudano together.

-lWest: Benin, Cameroon, Cote d'Ivoire, Ghana. Nigeria, Sierra Leone.

and Togo.

5East: Burundi, Ethiopia, Kenya, rvladagasca, ~Ialawi,

Rwanda. Tanzania, Zambia and Zimbabwe.

6Sudano: Burkina Faso, Central Africa Republic. Gambia.

rvlali, Niger, Senegal and Sudan.

90

Page 100: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

91

Table 6.1 (continued): Estimates of fixed country-specifie

effects for agricultural regions. Continues next page.

COlwtries Overall1 West2 East3 Sudand1

Beninw1 -0.167 -0.226

Cameroonw1 0.110 -0.226

Cote dl! Ivoire O.302lt' 0.127

Ghana@,tu2 0.080 -0.034

NigeriaWl -0.211 -0.226

Sierra Leonew2 0.042 -0.034

Togotu 1 -0.258 -0.226

Bltrllndi -0.070 -0.816

EthiopiaC» 0.080 -0.632*

Kenya 0.218 -0.946*

rvladagascar 0.434** -0.305

ivIalawi 0.427** -0.591

Rwanda -0.250* -1.162*

Tanzania 0.241 -0.651

Zamhia -0.070 -1.262*

Zimbabwe 0.514** -0.568

Page 101: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Table 6.1 (continued): Fixed cOlmtry-specific effects for

agricultural regions.

COlmtries Overall1 West2 East3 Sudano't

Burkina Faso 0.002 0.009

Central Afriea Rep. s2 0.214* -0.130

Gambias1 -0.192 -0.150

~[ali''' 0.080 0.060

Nigers1 -0.197 -0.150

Senegarti 0.080 0.141

Sudan,,*,s2 0.080 -0.130

* and ** indicate significance at 0.05 and 0.01 respectively.

1,2,3,4 Overall, West, East and Sudano are defined above.

s1Cmmtries assigned common dummy variable in the Sudano.

s2Colultries assigned common dummy variable in the Slldano.

wlColmtries assigned common dnmmy variable in the West.

w2Countries assigned common dlunmy variable in the West.

gColmtries assigned common dummy variable in the Overall.

92

Page 102: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

93

Stocks of capital KODA, KPFX, and KSAV : The coefficients of the stock

of the development assistance variable in the export share equations are positive in

al! regions. The magnitudes are all small however, (less than a tenth of the coefficient

of OPEN, the variable with the largest coefficient). The coefficient is largest in the

East (0.075) where it is strongest (p-value D.OG). KPFX is positive ill ail regiouti

but the West, and more or less of the same magnitude as KODA. It aIso has its

largest magnitude in the East (0.047). It is nat significant in any region. Thus, the

suggestion that both forms of foreign finanda! flows are likely ta impact export share

of agriclùture positively has sorne support in the Overall, East and Sudano. Recall

that in the production function, KPFX had a very significant and positive coefficient

in the West. By not being significant in the export share equation, one conc1udes

that in relative terms private foreign capital does not promote one sub-sector over

the other in this region.

The stock of domestic savings is significant in both the East and Sudano, (p­

values < 0.01 in both cases). The coefficient is positive in the East, but negative in

the Sudano. This variable is positive but not significant in the Overall and West.

The magnitude of the coefficients of this variable in the various regions are about the

same as they are for KODA and KPFX. The coefficient is largest in the East with a

value of 0.11. Thus one concludes that domestic savings have a tendency to impact

export share positively in the West, Overall and in the East. Significantly 50 in the

latter case. On the other hand, domestic savings have a significantly negative impact

on exports in the Sudano. For this region, the unusual impact of the labor variable

Page 103: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

94

in the production nmction has aIready been noted.. This negative impact of domestic

savings on export share is another. One hastens ta add that sorne countries included

in this latter region are landlocked. Exports must go by air, or travel by land and

then rely on shipping facilities of other cOtmtries. Either scenario will raise the cast of

exports, rendering investments in this sub-sector of agrictùture less attractive. This

e.xplanation is in addition ta the f'let th'lt many of thcsc countries are in the Sahel

zone as previously noted. (Note that negative coefficients in an export share equation

implies positive coefficient in the domestic share equation).

One notes that all forms of capital have greatest positive impact in the East. The

explanation must be that, by allowing producer priees to reflect international priees

more, t.he bias in domestic terms of trade against agrictùture in SSA is smallest in

this region. Hence on a relative basis, more investments are made in agrkulture in

the East than in othee regions.

Labor force, LABR: The labor coefficient is small but positive in the Overall

(0.04) and in the Sudano (0.02) export share equations, (p-values 0.05 and 0.77 re­

spectively). The impact of the variable is larger but negative in the East and West

(coefficients -0.062 and -0.15 respectively). It is significant only in the West. Thus,

one wotÙd conclude that increases in the labor variable lead to practically significant

increases in expart share of agriculture Overall. However, in the West, labor force

increases actually lead to a reduction in export share to the advantage of domestic

share. For the other two regions, the impact of labor force increase does not appear

to favor the share of any particlùar sub-sector significantly. The explanation for the

Page 104: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

95

finding for the West (for example Ghana) may be found in the increasing attrac­

tiveness of the domestic sub-sector due to deteriorating food secllrity situation, little

government involvement in setting prices in the domestic sub-sector, as weil as low

(sometimes falling) real producer prices for exports. (Governments fix nominal prices

orny).

Openness, OPEN: This variable is positive in an regions. It is aIso significant

everywhere but the East where it has a p-value of 0.11. The positivity of this variable

is probably due to the significant role agriclùtural earnings play in SSA trade earnings.

That is, trade share of GDP tends to rise and faH with agriclùtllral exports. Hence

the advice from sorne quarters that agriclùtural export earnings must be given serions

attention. As export shares move in the sarne direction as OPEN, domestk shares

must be moving in the opposite direction, hence openness has strong and negative

impact on domestic share in all regions. Its impact is weaker in the East. OPEN,

the policy variable which also proxies the extent of liberalization of the economy

including reduction in direct and indirect taxes, may not be significant in the East

because lower gains will he realized from further 'opening up' in this region than has

already happened because of the argument that this region more than others already

enjoys sorne of the gains from liberalization, (lower direct taxation of agrictùtural

exports).

Terms of trade, EXOl: That the external terms of trade variable is positive and

practically significant in all regions in the export equation is not surprising. Export

Page 105: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•96

share values are obtained at international priees. And agric1.ùtural exports constitute

a significant portion of the exports of most eountries. Thus, international prices for

agricultnral commodities influence the determination of the terms of trade of most of

these countries. Hence, the export share tend to rise and faIl with the terms of trade.

\Veather VariabHity, E)(02: The coefficient of this variable is not significant in

the any specification of the export share equation. This is what one would expect.

On the whole, one w01.ùd expect variability in weather ta affect bath domestic and

export sllb-sectors almost equally.

Ratio of export-to-domestie produeer priees, PRIeE: The role of price in

sllpply response has been studied extensively. Here, it is positive in all regions but

the Sudano. It is practieally significant (p-value of 0.03) in the West only. Thus

increases in the PRIeE variable lead to higher export share of agriculture except in

the Slldano. This is to be expected sinee an increase in this variable will result from

either an increase in the real producer priees of exports (numerator), or a decrease in

the real domestic agrieultural producer priee (denominator). In the Slldano, increases

in the variable PRIeE have a tendency to lead to a falI in the share of exports.

This appears unusual, but negative elasticities have been noted in the literatnre in

the eontext of subsistence agriclùture. In snch econornies, higher prodncer priee for

agricultural exports means the need to sell fewer quantities to meet onets needs, (e.g.,

Ezea1a-Harrisan, 1996). This is one more unusual finding with respect ta the Sudano

region.

Page 106: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

97

Jaeger (1992), Elmi (1994) and Bond (1983) have reported priee elasticities for

agricultural exports, food crops and aggregate agrictùture. Conceptually, their figures

are not directly comparable ta the ones obtained here since they were obtained from

different models. Be that as it may, the implication of the findings of this study

is that, the role of producer priee increases by themselves though positive in most

countries, is limited (Dinswanger, 1989, allJ Clùùbuel', 1988).

Time TREND: In all Ca':ies, the coefficient of the trend variable is negative,

small and significant in the export share equations, hence positive and significant in

the domestic share specifications. Interpreting this coefficient as productivity growth,

one wo\ùd conclude that in aggregate, productivity changes are having significantly

negative impact on export shares.

Page 107: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

98

6.1.3 Policy Groups

Separate tests were performed for stability of slope parameters with respect ta the

three policy groups, LARGE, S~IALL and POOR as was described in chapters three

and five. First, for stability over time for each poliey group betwccn the pcriods 1!J70

- 1980 (period prior ta being classified) and 1981 - 1993 for example, and then for

poolability across regions. The chi-squared statistic for constancy of slope coefficients

for the LARGE group was 25.24 with a p-value of 0.001. Thus, one cannot accept

the hypothesis of no change over the two periods. The slope coefficients of KODA

and KPFX are identified as being sigIÙficantly different over the two periods. The

test shows that the slope coefficient for KODA was significantly smaller in the 1981 ­

93 period than in the period 1970 - 80, while that for KPFX was significantly larger

during the 1981- 93 period than during 1970 - 80. This wOllld suggest that KODA did

not promote export share as much as it did in the earlier period (probably benefiting

other sectors of the economy in the later period), while KPFX on the other hand,

promoted export share more.

For the SMALL group, the corresponding chi-squared statistic was 18.69 with p­

value 0.02. This gives borderline indication of no change in slope coefficients. (This

time KPFX is identified as the only slope coefficient that is almost significantly dif­

ferent between the two periods). It is lower during 1981 - 93 than 1970 - 80. For the

POOR group, the chi-square statistic was 10.80. Thus, one is unable to reject the

null hypothesis of no change in slope parameters.

Page 108: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

99

Poolability of data across all three groups, or, for any pair during the period 1981­

93 is strongly rejected. The pairwise chi-squared statistics are 36.27, 42.94, and 25.26

for LARGE and S~lALL, LARGE and POOR and SlVIALL and POOR. AlI p-vaiues

are less than 0.001. Equality of the coefficients of ln(KODA) and ln(KSAV) was

rejected for bath LARGE and PODR.

Page 109: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

100

1 Table 6.2. Parameter estimates for export share regression1

for Policy groups, 1981 - 1993.

Variable2 LARGE3 Sl\JIALL4 POORs

ln(KODA) 0.129** 0.057** -0.303*

(2.839) (3.530) (-2.259)

ln(KPFX) -0.032 0.033 -0.079

(-0.464) (0.613) (-0.997)

ln(KSAV) -0.023 0.040 0.051

(-0.258) (1.412) (1.135)

ln(LAER) -0.398 -0.068 1.692**

(-1.135) (-0.232) (2.618)

OPEN 0.105 0.153 -0.329

(1.547) (1.674) (-1.338)

EX01 0.257** 0.084 0.158**

(3.697) (1.529) (2.509)

EX02 -0.091* -0.114** -0.078

(-1.908) (-2.574) (-1.384)

ln(PRIeE) 0.061 0.138** -0.016

(1.819) (3.936) (-0.469)

TREND -0.016 -0.006 -0.027**

(-1.608) (-1.139) (-2.497)

D. F. 46 80 57

See Iegend on next page.

Page 110: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

101

1 t-statistics in parenthesis below coefficient estimates.

* and ** indicate statistical significance at 0.05 and 0.01

levels respectively. D. F. is degrees of freedom.

2 Variables are defined in chapter 3.

3LARGE: Gambia, Ghana, ~Iadagasca, Tanzania, and Zimbabwe.

4SMALL: Central Africa Republic, Kenya, rvlalawi, rvlali, Niger, Nigeria,

Senegal and Togo.

5 POOR: Benin, Cameroon, Cote d'Ivoire, Rwanda, Sierra Leone and Zarnbia.

Page 111: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

102

1 Table 6.2 (continned): Fixed cOlUltry-specific effects.

Export share equations - Policy groups.

Country LARGE SlVIALL POOR

Gambia -0.740

Ghana -0.931

lVladagascar -1.056

Tanzania -1.139

Zimbabwe -0.583

Central Africa Republic -0.392

Kenya -0.466

~Ialawi -0.017

lVlali -0.339

Niger -0.627

Nigeriasl -0.697

Senegal -0.385

Togos! -0.697

Benin 2.756

Cameroon 3.865

Cote d'Ivoire 3.924*

Rwanda 0.779

Sierra leone 2.817

Zamhia 4.621

* indicates significance at 0.05 level.

Page 112: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

103

Parameter Estimates

Estimates of parameters for the policy groups are presented in table 6.2.

Capital, KODA, KPFX, KSAV:

In both the LARGE and S~IALL subgroups, the KODA coefficient is positive and

significant. The magnit.lUip of t.hp ('oeffident in LARGE (0.13) is twice n.s big as that

for SivIALL (0.06). The KODA variable is negative (-0.30) and significant for the

POOR group. The stock of private foreign capital is not significant for any group.

Its coefficients are negative in the LARGE and POOR groups, but positive in the

SlVIALL group. AlI coefficients are smaller in magnitude théUl corresponding ones for

KODA. Domestic savings have virtually no expléUlatory power for the LARGE group,

but have substantially more explanatory power and positive in Sl\·IALL and POOR,

though not significant for either group.

[t wo\ùd appear that linking flow of concessionalloans (a high proportion of OOA

in recent times) to improved policy environment is having heneficial influence on

the export share (over domestic share) of those conntries that. are able to attraet

such loans. In the POOR countries, aDA appears to be benefiting domestic share

of agriclùture instead. If the objective of aDA is to help improve foreign exchange

earnings, in the LARGE and S~IALL groups of cOlmtries, but on the other hand

ta help improve the domestic sub-sector in the POOR group, then aDA is serving

its purposes. These apparently different objectives of ODA are not contradictory

however. For, sorne countries may require aOA to improve the level of agriculture for

domestic consumption, while for others whose agriculture for domestic consumption

Page 113: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

104

is adequate, OOA may be geared towards increasing export earnings.

On the other band, it wOlùd appear that private foreign capital does not signifi­

cantly impact either the export or domestic sub-seetor in any policy group. Domestic

savings too do not significantly impact the share of one sub-sector over the other in

any group either.

Labor, LAER:

The labor coefficient is positive, large and significant only in the POOR export

share equation. This is the only region where labor input positively impacts export

share over domestic share. In the LARGE group, the LABR coefficient is big but

negative, while in S~IALL, it is negative but not big. Both are not significant. Seing

negative implies that they are benefiting the domestk share more. Liberalization in

the face of high direct taxes on agricultllral exports must be driving lltili ty maximizing

farmers to the lower taxed domestic sub-sector.

Openness, OPEN:

Openness is positive and fairly big in the export share equations of LARGE and

SIvIALL countries. This is not surprising sinee policy eompliance includes liberaliza­

tion of the agricultural sector (as weIl as others). This is in addition to the faet that

agriclùtural exports are still a good portion of each country's exports. The variable

is not significant however. This may be due to the faet that there are limits to the

benefits of openness if other steps such as investments, training of people, (Rodrik,

1997) are not taken in addition. The coefficient is negative in the POOR group.

This follows from the classification of these countries as being in a negative policy

environment. That the coefficient is not significantly negative here too follows from

Page 114: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

105

the fact that agriCtÙtural exports are substantial in spite of being classified as poor

poliey environment.

Terms of trade, EXOI :

The extemal terrns of trade variable is signifieant and positive for all groups in the

export share equation. This is not surprising sinee export share values are obtained

at internatiùuéÙ prit2es, and a.') explaiuoo pl'eViOlltily, agl'iClùtlUa! exports coustitute

a significant portion of the exports of most countries and hence they influence the

determination of the terms of trade of most of these countries. Hence, the export

shares tend ta rise and faU with the terms of trade.

Weather variability, EX02 :

The coefficient of this variable in the export share equation is small and negative

aeross ail three groups. The magnitudes are almost equaI aeross the groups. The

coefficient is significant only in the S~IALL group. One is Imable to explain this

later observation except ta note that, there are fewer observations in these policy

regressions, thus pectùiar events may stand out more here.

PRIeE:

This variable is positive in both LARGE and Sl\tIALL gronps. It. is significant

in Sl\tIALL, and has a p"value of 0.06 in LARGE. The coefficient. of this variable

is negative and not significant in POOR. Reduced government intervention in priee

determination is a major plank of poliey reform. Reforming countries are the ones

responding to calls ta liberalize their economies, including giving farmers higher pro­

ducer priees. Thus, it appears that as governments respond ta the calI 50 do farmers

(Knleger et al., 1991, and Bautista and Valdes, 1993). The same argument explains

Page 115: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

106

why in poor poliey environment, (which translates ta lower level of liberalization,

hence higher government control over priees), producer priees have no significant im­

pact on export share. There is a slight tendency for priee increases to favor domestic

share in the POOR countries.

TREND:

The trend coefficient is negative across all groups. In addition, it is significant

in the POOR group. Interpreting tms coefficient as productivity growth. one wOlùd

conclude that aggregate productivity is falling in the export sub-sector across the

groups. The drop is however larger and stronger in the POOR policy group. The

other side of the argument is that productivity appears to be improving more in the

domestic sub-sector in the POOR group, than in t.he other two groups.

6.1.4 Summary

This chapter reports the reslùts of estimating the static response of export shaxe of

agriclùture in 23 sub-8aharan African cOlmtries over the period 1970 to 1993. Share

equations are deduced using the profit nmction (GDP function) approach. Physical

capital is disaggregated into the stock of official development assistance, the stock of

private foreign commercial flows and the stock of domestic savings as before. The

analysis also focuses on the openness of each economy as a measure of economic

poliey, labor, terms of trade, and the role of the weather. The data are pooled

together in several ways: (i) aU countries in one panel, (ii) COlmtries within one agr~

elimatic region, (three regions), and (iü) cotmtries grouped by improvements in their

macroeconomic poliey.

Page 116: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•107

Statistical tests suggest that slope parameters are unchanged between the two

periods 1970 - 80 and 1981 - 93 far agricultural regions. The evidence is somewhat

different for poliey groups. For these, a statistical test suggests a change in slope

coefficients for the LARGE group over the two periods. The two foreign capital

variables stand out as having changed significantly during the second period for the

LARGE group. The iJllpact of ODA wa.s lower in the second period, while that of PFX

was larger. Slope coefficients were lUlchanged for the SJ\.IALL and POOR groups.

The reslùts show that the impact of the stock of development assistance is positive,

small but not significant either Overall or in any of the regional groups. On the other

hand, the stock of development assist.ance impacts export share of agriculture in the

policy groups significantly. The impact of tms variable on the share of agricultllral

exports is highest (positive and significant.) in the LARGE group, followed by the

S~IALL group. Its impact is negative, fairly large and significant in the POOR

group.

Private foreign capital fiows too impact the export share of agricultnre positively

in all the regions but the West, where the impact in negative, small and insignificant.

The impact is small in the other regions tao, and of about the same magnitude as

development assistance. Within the policy groups, the impact of tms variable is not

significant any where. It is negative in bath LARGE and POOR, but positive in

Sl\JIALL.

Domestic savings have positive impact in aIl the regjans but the Sudano (where

they are negative). They are significant only in the East and Sudana. The magnitude

of the impact of this variable is about the same as the other two capital variables.

Page 117: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

108

Within the policy groups, the impact of this variable is not significant, and its magni­

tude is about the saIne as that of private foreign capital. It is negative in the LARGE

group, but positive in Sl\tIALL and POOR.

Labor is positive Overall and in the Sudano. It is negative in the West and East.

It is significant in the West, and almost 50 in the Overall. Among the policy groups,

the impact of labor is negative in the LARGE and S:tvIALL groups, but positive in

the POOR group, the ooly group in which it is significant.

Openness has positive impact on export share everywhere exeept in the POOR

poliey group. However, it is significant only in the Overall, West and Sudano. The ex­

ternal terms of trade are aIso positive and signific~mt or aImost significant everywhere.

The weather variable is negative everywhere, but not significant anywhere.

Domestic priees are positive everywhere except in the Sudano and in the POOR

eountries. They are significant in the SrvIALL cotmtries, and almost significant in the

West and LARGE. Finally, there is ample evidence that changes in productivity are

having negative impact on export share in aIl regions and policy groups.

The findings suggest that once again. the stnlcture of the response of export and

domestic shares have not changed significantly aver the period across the agrictùtural

regions. However, when looked at fram economic policy point of view sorne changes

have oecurred. Development assistance for the cotmtries grouped lmder LARGE

poliey used to have a larger impact in the earlier period (1970 - 80) on export share

than during the second period (1981 - 93). On the other hand, private foreign capital

used to have a sma1ler impact on export share during the earlier period than it did

during the second period.

Page 118: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

109

The findings further suggest that, the raIe of the capital variables is not significant

in determining export shares though the tendency is for capital ta impact export

shares positively. Labor is no longer dorninating other factor inputs in determining

export share the way it did in determining of agricultural output. The impact of

producer price increases though in general positive is not significant in determining

sllb-~ector shares. There is weak evidellce of suLsistellce agriculture in the Sudano.

Opermess and terms of trade are in general significant in determining export shares.

The next chapter investigates share response in adynamie framework.

Page 119: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 7

Estimation and Empirical Results ­Dynamics of Agricultural ShareResponse

This chapter discusses the techniques used to estimate the parameters of the dynamic

share response mode1, and then presents estimation reslùts. In spite of interest in

static share response in the literature. one notes that short-rnn response obtained

from static models do not take into aceount the faet that the empirical relationships

on the grolmd may not be in long-run eqnilibrium. Indeed. agricultnral supply may

not respond immediately ta changes in the explanatory variables. This may be dne

ta habits, persistenee, implementation 1ags, misinterpreted real priee changes. and

ather factors. This ehapter modifies the statie specifieatians of the previons chapter

ta alIow for delayed effects of changes in variables.

7.1 Estimation

The export and domestic share equations eonstitute a set of seemingly unrelated re-

gressions, hence, to avoid singularity of the variance-covariance matrix, one equation

is deleted. In ail cases, only the export share (AEXP) specification is estimated. The

110

Page 120: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

111

slope coefficients in the corresponding domestic share (ADûM) equation are obtained

by reversing the signs of the corresponding coefficients in the export share equations.

The cOtmtry dummy variables of the ADû~I equations equal one minus the corre­

sponding values in the export share equation. Standard errors remain unchanged.

Observe that the transformation from the ADL to the Bewley specification leaves

intact cOlmtry-specific dummy variables. IndeeJ, this tl'ausforIllatioll preserves the

error process. Equations (3.17), (3.18), (3.19), and (3.20) are estimated by instnl­

mental variable (IV) regressions. This is necessary for two reasons: (i) for consistent

estimation due to the presence of contemporaneous terms in the export share vari­

able on the right-hand side, and (ii) to obtain identical estimates as wOlùd have been

obtained by estimating the ADL and then computing long-nm coefficients. This is

shawn in Bewley (1979, 1986), and Wickens and Breusch (1988). The IVs that will

insure these are the regressors of the ADL.

Specifical1y, they are: (i) each regressor that appears in levels in the Bewley

transformation, Xkt, serves as its own IV, except OPEN and PR/CE. These two

variables are possibly simtùtaneollsly determined and one year lags are used as IVs

for these. Thus, the IV for KODAt is [<ODAt , LABRt is the IV for LABRt etc.,

whereas the IV for OPENt is OPENt- 1 and for PR/CEt it is PR/CEt- l • (ii) For

each regressor in the Bewley transformation that is a lagged deviation from levels,

the lagged value is used as the IV. That is, Xkt-j is the IV for (Xkt - Xkt-j). for aIl j

and k. In particular, AEXPt- 1 is IV for (AEXPt - AEXPt-d, ln(KODAt-d is the

IV for [ln(KODA t ) -ln(KODAt_d}, and ln(LABRt _ a) is the IV for [ln(LABRt} ­

m(LABRt- a)}, etc. Further, in a panel data context, Arellano(1989) recommends

Page 121: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

112

these IVs over other candidate IVs.

Each specification is estimated by pooled generalized least squares method using a

robust estimator for the variance-covariance matrix to obtain estimates of all dummy

variables and other variables in the models in one step as in the previous cases.

Estimation is done in Winrats-32 version 4.3 numing on a persona! computer. In all

cases, one cOlùd not reject the nlùl hypothesis of no autocorrelation in the residuals

as the following illustrates. Thus coefficient estimates obtained here are consistent

in spite of the presence of lagged dependent variables on the right-hand side. The

regression of the residual et on its first lag, et-l, (et = pet-l + Ut, where Ut is a

random error term) yields the following estimates for p : 0.02, p-value 0.78; 0.08,

p-value 0.28; -0.01, p-value 0.88; -0.01. p-value 0.97; for the Overall. West, East. and

Sudano respectively. The ntùl hypothesis is that p = 0. 1

7.2 Results and Discussion

First, an analysis of the adeqnacy of the regressions are discnssed, then the parameter

estimates are presented and discussed.

1.2.1 Diagnostic Tests of Regression Adequacy

Estimation reslùts are presented in table 7.1 for the export share equation. The nota-

tian used for deviations of lagged dependent and independent variables from CUITent

level in the tables are given below. On the whole, many cotmtry dummy variables

1 Keane and Runkle (1992) proposed a method for estimating dynamic panel-data modeIs withseriai correlation. This W8S considered but not pursued here, partiy because seriai correlation is nota problem, but aIso because their method is principally aimed at random efFects modeIs (though itmay serve fixed effects models just as weIl).

Page 122: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

113

have t statistics that are not significantly different from zero (or from one another).

For sorne of those not significantly different from zero, statistical tests could not reject

the null hypothesis that these cotmtry-specific dummy variables are equal. Tests that

these cOtmtry effects are jointly zero are however rejected. Four of these countries

were then assigned a common dummy variable and the model re-estimated as done

pl'eviou::ily. By a.':i::iiguiug cOUUllOH JlUillllY variable;, ulle upts for a more efficient

variance at the risk of biasing coefficients as discussed in chapter 6.

Residuals and other Diagnostics

Ordinary least squares residual errors were tested and found ta be heteroskedas­

tic. The chi-squared statistics for the likelihood ratio tests of the ntùl of no het­

eroskedasticity are 1,023.12 with 22 degrees of freedom for the Overall, 123.45 with

6 degrees of freedom for the West, 247.11 with 8 degrees of freedorn for the East,

and 242.01 with 6 degrees of freeclom for Slldano). Ta correct for heteroskedastidty

the ROBUSTERRORS option in Winrats is specified in compllting the variance­

covariance matrLx. Tests for stationarity of the residuals of bath the ADL and the Be­

wley transformation were aIso performed using their alltocorrelation ftmctions (ACFs)

as discussed earller. AlI ACFs were fOlmd to be stationary.

lv/odel stability

The question of the stability of the data generating process of the static share

response models has been addressed in chapter six where support for stability of the

process was found. However, there is an additional complication introduced here

by the presence of lagged dependent variables on the right-hand side. The issue is

that of stationarity of regressors (which now include lagged dependent variables).

Page 123: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

114

First, the existence of long-run relationship between the dependent variable and the

explanatory variables is determined by checking that the condition of inequality (3.13)

in chapter three is satisfied. This is indeed found to be the case in all specifications.

For the ADL, it was confirmed that the absolute value of the sum of the coefficients

of the lagged dependent variables on the right is less than one. For illustration, the

coefficients of the ADL for the Overa1l are given in the appeuŒx. For this ADL, the

SUffi of the coefficients on the lagged AEXP terms is 0.64, which is less than 1.

Second, stationarity of regressors in the presence of lagged dependent variables is

assured if the roots of the polynomial equations formed using the coefficients of the

lagged dependent variables in the ADL as coefficients of the polynomial term alllie

olltside the unit circle, (Kesavan et al. (1993), Bewley (1986, p.61) and Greene, 1997).

This is aIso fotmd to be the case. (For the Overall case, \Ising coefficients of the ADL

given in the appendix, the equation of interest is 1-0.547z - 0.088z3 = O. Its roots lie

outside the unit circle. Other checks for stability that were performed include ordinary

least squares as an altemate estimation method. As expected standard errors of this

approach were smaller than the GLS errors. Also, the data were pertnrbed slightly

by using lags of variables instead of levels. lags as instrumental variables for variables

in levels, and moving averages of variables. AIl yielded reslùts that are not vastly

diHerent from each other.

The following notation is adopted in reporting parameter estimates.

K02 =In(KODAit )-ln(KODA;t-2);

K04 =ln(KODA;t)-ln(KODA;t-4);

K05 =ln(KODA;t) -ln(KODA;t_s);

X015 == EXOl t -EXOl t _ s;

X021 =EX02it-EX02it_l;

Dl =AEXPt - AEXPt - 1 ;

Page 124: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

KP1 =ln(KPFXit ) -ln(KPFXit- 1);

KP2 == ln(KPF...Yit ) -ln(KPFXit- 2);

KSI == In(!(SAVit) -ln(KSAVit-4);

KS4 == In(KSAVid -ln(KSAVit-4);

KS5 == In(KSAVit) -ln(KSAVit-s);

LBI =: In(LABR,t) - ln(LAB~t-d;

LB3 == In(LAB~t) - Ln(LAB~t_3);

LB5 = In(LABRit ) -ln(LABltït-s);

P4 =In(PRICEit ) -ln(PRICEit_td;

115

D2 =AEXPt - AEXPt-2;

D3 == AEXPt - AEXPt - 3 ;

D5 = AEXPt - AEXPt - 5 ;

Page 125: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

116

7.2.2 Parameter Estimates

This sub-section discusses parameter estimates reported in table 8.2

Lagged dependent variables.

In the specifications for the Overall, East and West, coefficients of the deviations

of the lagged dependent variables from current levels are all significant. indkat.ing

strong evidence of persistence (or slow adjustments). In the Overall regrcssion, two

deviations of the dependent variables from CIment levels are significant. In the West,

whereas all deviations (of lags) of the dependent variable (1 ta 5) on their own are

significant, theyall become insignificant in the presence of the first deviation. ResIùts

are reported for this latter case only. In the East, the first, second and fifth deviations

remain significant. In the Sudano however, the t statistic of this coefficient is not

significant (p-value 0.23), suggesting no persistence. This is another unusual finding

for this region. A possible explanation may he the practice of shifting clùtivation

and migration (more prevalent here than in other parts of SSA). This likely helps to

moderate persistence.

2 In aIl of the following discussion, parameters are considered significant when their marginal signif­icance level is 1% or less.

Page 126: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

117

1 Table 7.1: Dynamic share response estimation.

Coefficients of deviations of lagged export share (1970 - 1993). 1

Variable2 Overall6 West3 East4 Sudano5

Dl -1.450** -1.534** -0.805** -0.384

(-5.038) (-3.063) -3.248) (-1.200)

D2 -0.383**

(-2.443)

03 -0.236*

(-2.110)

D5 0.293**

(2.993)

O. F. 376 106 132 104

1t statistics in parentheses below coefficient estimates.

* and ** indicate significance at 0.05 and 0.01 levels respectively.

2 D. F. is degrees of freedom. AlI variables are defined above.

3West~ Benin, Cameroon, Cote d'Ivoire, Ghana, Nigeria, Sierra Leone and Togo.

4East: Ethiopia, Kenya,IvIadagasca, lVlalawi, Bunmdi, Rwanda,

Tanzania, Zambia and Zimbabwe.

5Sudano: Central Africa Republic, Gambia, Ivlali, Niger, Senegal, Sudan

and Burkina Faso.

60verall: West, East and Sudano together.

Page 127: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

118

• Table 7.1 (continued): Dynamic share response estimation.

Estimates of long-run coefficients of explanatory variables l

(export share equations, 1970 - 1993). See legend on next page.

Variable2 Overa1l3 West4 East5 Sudan06

ln(KODA) -0.002 0.024 0.045 -0.111**

(-0.058) (0.360) (1.196) (-2.903)

ln(KPF~~) 0.022 0.063 0.060 0.012

(0.956) (0.641) (1.418) (0.635)

ln(KSA1/) 0.045** 0.032 0.078 0.094**

(2.602) (0.954) (1.569) (3.212)

ln(LAER) -0.116** -0.251 ** -0.049 0.265*

(-4.0189) (-4.912) (-0.548) (2.302)

OPEN 0.248 -0.012 0.242 0.120

(1.687) (-0.041) (0.638) (0.673)

E~~Ol -0.043 0.055 -0.058 0.444**

(-0.853) (0.933) (-0.776) (4.215)

E)(02 -0.057 -0.014 0.108 -0.056

(-0.656) (-0.107) (1.352) (-1.013)

ln(PRICE) 0.055 0.101 0.274* -0.077

(0.911 ) (1.545) (1.983) (-1.426)

TREND -0.014** -0.013** -0.023** 0.034

(-3.978) (-2.313) (-4.147) (1.157)

O.F. 376 106 132 104

Page 128: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1

119

1 t-statistics in parentheses below coefficient estimates. * and ** indicate statis­

tical significance at 0.05 and 0.01 levels respectively. 2 D. F. is degrees of freedom.

Other variables are defined in chapter 3. 3,4,5,6 West, East, Sudano and Overall are

defined above.

Page 129: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

120

Table 7.1 (continued): Dynamic share response estimation.

Coefficients of deviations of lagged variables 1

(export share equations, 1970 - 1993). See legend on next page

Variable2 Overall3 West4 East5 Sudano6

K02 -0.093

(-1.098)

K04 -0.265*

(-2.184)

K05 -0.048 -0.202* 0.220*

(-1.213) (-1.998) (1.986)

KPI -0.087 0.330

(-1.465) (1.236)

KP2 0.093* -0.045 0.011

(2.113) (-0.330) (0.351)

KSI 0.289*

(2.081)

KS4 -0.066* 0.006 -0.077**

(-2.266) (0.061) (-2.744)

KS5 -0.205**

(-3.576)

D. F. 376 106 132 104

Page 130: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1 t-statistics in parentheses below coefficient estimates. * and ** indicate

statistical significance at 0.05 and 0.01 levels respectively. 2Variables are

defined above. D. F. is degrees of freedom. 3,4,5,6 West, East, Sudano and

Overall are defined above.

121

Page 131: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

122

Table 7.1 (continued): Dynamic share response estimation.

Estimates of coefficients of deviations of lagged variables

from Cllrrent level1 (export share equations, 1970 - 1993).

Variable2 Overa1l3 West4 East5 Sudano6

LB1 -0.281

(-0.816)

LB3 0.427 1.781

(1.525) (0.943)

LB5 -0.348* -2.494 -0.095 0.010

(-1.895) (-1.721 ) (-0.528) (0.099)

X015 0.176** 0.156**

(4.853) (3.498)

X021 -0.097 -0.286**

(-1.310) (-3.587)

P4 -0.017 0.044 -0.073 0.034

(-0.5563) (0.787) (-1.286) (1.157)

D. F. 376 106 132 104

1 t-statistics in parentheses below coefficient estimates.

* and ** indicate statistical significance at 0.05 and 0.01 levels respectively.

2 Variables are defined above. D. F. is degrees of freedom.

3,4,5,6 West, East, Sudano and Overall defined above.

Page 132: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

123

Table 7.1 (continued): Dynamic share response estimation.

Fixed cOlmtry-specific effects in export share equations

(1970 - 1993). Continues next page.

Countries Overall West East Sudano

Benin@,wl -0.180* -0.376

Cameroon<c, w2 -0.180* -0.647**

Cot d'Ivoirew3 0.120 -0.216

Ghanaw1 -0.043 -0.376

Nigeriaw2 -0.379** -0.647**

Sierra LeonewJ -0.029 -0.216

Togowl -0.164 -0.376

Burundi 0.208* -0.080

EthiopiaE1 0.080 -0.007

Kenya -0.008 -0.238

rvladagasca 0.206* 0.314

~lalawiE2 0.527** 0.176

Rwanda 0.115 -0.389

TanzaniaE1 0.048 -0.007

Zamnbia -0.428** -0.474

ZimbabweE2 0.390** 0.176

Page 133: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Table 7.1 (eontinue<!): Dynamie share response estimation.

Fixed country-specifie effects in export share equations

(1970 - 1993). Continues next page.

Countries Overall West East Sudano

Burkina Faso -0.039 0.237

Central Afriea Rep.@ -0.180* 0941

Gambias1 -0.025 0.025

Niger~·sl -0.180* 0.025

Senegal 0.044 0.453

SudanQ -0.180* 0.566

* and ** indicate signifieanee at 0.05 and 0.01 respectively.

slColmtries assigned common dummy variable in the Slldano.

52Colmtries assigned eommon dummy variable in the Sndano.

u:1Countries assigned eommon d\unmy variable in the \Vest.

w2Countries assigned common dnmmy variable in the \Vest.

w3Conntries assigned common d\unmy variable in the West.

EIColmtries assigned eommon dummy variable in the East.

E2Countries assigned eommon dummy variable in the East.

'f}Countries assigned common dummy variable in the Overall.

124

Page 134: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

125

Development assistance, KODA: The coefficient of ln(KODA) suggests that

the stock of official development assistance does not have significant long-nm (LR)

effect on AEXP in either the Overall, East or West. In fact, its long-nln impact in the

Overall is almost nil (magnitude -0.002 with a large p-value). It does have significantly

negative LR impact in the Sudano (implying positive impact on ADOlVI). In all cases,

the magnitude of the coefficients are relatively small (being largest in the Sudano).

The deviation of the fifth lag from current level is negative but not significant in

the Overall, negative and almost significant in the West, and positive and almost

significant in the East. Discussion of these reslùts follow after presenting reslùts of

KPFX and KSAV.

Private foreign ftows, KPFX: The coefficient of ln(KPF ~y) suggests that the

stock of private foreign commercial flows does not have significant LR impact on

AEXP in the Overall or any region. The coefficients are all positive (though still

smali) and bigger than corresponding ones for KODA except in the Sudano. The

deviation of the second lag from ctment level has positive and almost significant

impact in the Overall specification, while that of the first lag is negative but not

significant in the Overall. The deviation of the first lag is positive but not significant

in the West.

Domestic savings, KSAV: The coefficient of In(KSAV) suggests that the stock

of domestic capital has significant positive long-run (LR) effect on AEXP in the

Page 135: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•126

Overall and Sudano specifications. The LR impact of this variable is positive but

not significant in the East (p-value 0.12) or in the West (p-value 0.34). Further,

deviations of the fourth lag of this variable from current level (which measures delayed

impact) is significant or almost significant and negative Overall and in the Sudano.

The deviations of the first and fifth lags of this variable from the current level are

signilkaut (or è:ÙJuost so) in the Ea::it, the first beillg positive, the second negative.

Thus, the impact of the three components of physical capital vary ac.cording to

region. The stock of development assistance appears not to have large or significant

LR impact on the export share of agriculture in any region, or Overall. It does

have positive LR impact. on domestic share in the Slldano however. The stock of

private foreign commercial flows too does not appear ta have large or significant LR

effect on AEXP, (hence not on ADO~[ either). Domestic savings have significant

positive LR effect on export share Overal1. Further, a11 capital stock variables aIso

have substantial delayed effect.s. KPFX has the shortest delayed effect.s of up ta two

years only, while it is up to five years for bath KODA and KSAV.

Labor, LABR: The coefficient of ln(LABR) suggests that the LR effect of the

labor variable on export share is negative and significant in the Overall and in the

West. It is positive and significant in the Slldano region. Labor has small, negative

but non-significant LR impact on the export share in the East. It has no significant

delayed eflects in this region either. In the Overall specification, the third and fifth lag

deviations from current levels have substantial (magnitude and marginal significance)

impact on export share. In the specification for the West, ooly the fifth lagged devia-

Page 136: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

127

tion of this variable has substantial negative delayed impact on AEXP. No deviation

of LABR has significant impact on AEXP in Sudano.

The negative long-run relationship in ail but the Sudano may be explained by the

fact that the deteriorating food security situation as weIl as government control over

producer priees of exports make it increasingly attractive for farmers to turn to food

production aver time. Thus, the long-nUl effects of changes in LABR are positive

for ADO~,t That, in the Overall, lagged deviations are substantial is not surprising.

Pensant fanners (easily the backbone of Afriean agrÎClùture) hardly have other skills

with which to make a living. They are thus likely to continue to earn their living

doing what they know how to do best, unless externuating cïrCllIIlstances force them

ta do otherwise. In fact, unlike conventional wisdam, which suggests ease of labor

movement relative ta other factors of production, in this context, labor may weIl he

the most immobile factor input.

Priee, PRIeE: The coefficient of ln(PRICE) suggests that the LR impact of

the ratio of the index of the real producer priee for the export sub-sector ta that of

the domestic sub-sector on AEXP is positive, small and not significant in the Overall

specification (coefficient 0.06, p-value 0.36). It is positive, larger but not significant in

the West, (coefficient 0.10, p-value 0.12). ft is positive, and largest in the East where

it is almost significant (coefficient 0.27, p-value 0.04), and negative, somewhat small

and not significant in Sudano (coefficient -0.08, p-value 0.15). The higher estimate for

the East is probably due ta the higher tendency for countries in this region ta allow

domestic producer prices to track international prices more closely. The impact of

Page 137: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

128

this would likely be that price increases more likely maintain (if not improve) the level

of domestic terms of trade between agriculture and the rest of the ecouomy. Thus

sucb priee changes are likely to be seen by farmers to be worth their while, and so

they respond by producing relatively more exports. The deviations of lagged prices

are not significant in any region or overall.

Estimates obtained here may not be directly comparable to those of .Jaeger (1992),

Elmi (1994) or Bond(1983) because all three models are different than that estimated

here. In particlùar, Elmi and Bond in tlSing the Nerlove technique focus on sup­

ply response to priee, the model here considers other factors that determine output.

Further, Binswanger (1989) and Schiff and ~[ontenegro (1997) have expressed reserva­

tions about the Bond estimates. Indeed, Chhibber (1988) has found that byaccOlmt­

ing for other variables (i.e., other than priee) t.hat. affect sllpply response in models

that emphasize priee only, he obtains lower estimates of priee elast.icities. However,

the general conclusion that aggregate response to prodllcer priees are low is confirmed

here.

Openness (OPEN) and Terms of Trade, EXOl: The coefficient of OPEN

suggests that the LR effect of the extemal trade share of gross domestic product on

AEXP is positive but not significant in the East, Sudano, or Overall. The impact is

relatively large in the Overall and in the East. OPEN is very small and negative in

the West. That the long-run impact of OPEN is positive (except for the West) but

not significant anywhere may be due to the fact that openness (liberalization) which

translates to higher producer prices for farmers will by itself not raise the export

Page 138: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

129

share of agriculture significantly in the long-run. In fact, its impact in the West is

almost non-existent. NIore is called for (this is discussed later). Indeed, export share

has been dropping Overall, and in all regions.

In regard to the terms of trade, the coefficient of EXOI suggests that the LR

effect of this variable on AEXP is positive in the West (where it is small, coefficient

0.06) and Slldano (where it is rather large, coefficient 0.44). It is significant ooly in

the Sudano region. The value of the coefficients are negative in the Overall and in

the East. The deviation of the fifth lag from current level however, has positive and

significant impact on AEXP in the Overall and the East only.

The importance of the terms of trade shock to growth has been enlphasized by

Easterly, Kremer. Pritchett and Summers (1993). That the long-nin EXOI V"dIiable is

not significant in the East, West and Overall may be due to falling share of agricllltural

exports in total exports over time. When agricultural exports are a larger proportion

of total exports, international priees of agrictùtural exports influence a country's

overall terms of trade (the variable being tlsed here) more.

Weather, EX02: This variable does not appear ta have significant LR effect on

either AEXP or ADOrvl as suggested by the coefficient of the EX02 variable. The

deviations of the first lagged values from CUITent level have significant negative impact

on AEXP in the East ooly. No doubt, the weather must impact both sub-sectors

of agriculture. Aboagye (1998a) finds that this variable has sigrùficant impact on

agrictùtural output. One rationalizes that the apparent non-significance of the LR

impact of this variable on export share is because of the faet that the variable very

Page 139: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

130

likely affects both sub-sectors almost equally. It is Dot immediately obvious why the

variable is significant in the East.

TREND: This variable is expected to measure technical change (as weIl as other

omitted trending variables). It is negative and significant in the Overall, West, and

East. It is not significant in the Sudano. One interpretation is that, total factor

praductivity in the export sub-sector is falling, but less than snggested in the static

specification (chapter six), once long-nm effects of changes in factor inputs and other

explanatory variables have worked their way iota the system. If one accepts this

however, one must then also accept that significant productivity gains are taking

place in the domestic sub-sector. The first is easy to accept sinee even in the face

of illcreasing stocks of inputs, SSA's export share has been falling. The second may

also be true, with the explanation that any food security concerns may not be due to

lack of productivity growth on per capita basis.

7.2.3 Summary - Dynamic Share Response

This chapter has estimated a dynamic specification of the export share of agrictùture

in sub-Saharan Africa in arder to investigate the long-ntn relationslùp as well as per­

sistence between the shares of export and domestic sub-sectors of agrictùtural output

and the stocks of development assistance, private foreign commercial capital, dames­

tic savings, producer prices, openness of the economy, tenns of trade and variations

in the weather. Panel data far 23 su~Saharan African countries covering the periad

1970 to 1993 were used.

Estimated coefficients of deviations of lagged dependent variables from CIment

Page 140: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

131

values are significantly different from zero in all regions but the Sudano. These give

indications of slow adjustments in all regions but the Sudano. Estimation restÙts

further suggest that in the long-run, the stock of development assistance appears to

he small and not to have more impact on the share of one suh-sector over the other in

the Overall specification. In the Western or Eastem-Southem agricllltural regions the

long-nm impact of this variable appears ta favar the export 5hare, though ils Ï.lllpact

is not sigIÙficant. In the Sudano-Sahel region however, it appears that the stock

of development assistance has significant negative long-nm impact on export share.

The stock of private foreign commercial flows has small, positive long-nm impact

on the export share in all regions. This impact is however not significant anywhere.

Domestic savings significantly affect export share positively in the Overall and in the

Sudano-Sahel region in the long-nm. In the other two regions, the long-nm ilnpact

of tohis variable is aIso positive but not significant. Of the capital variables, it is the

stock of domestic savings that has the largest positive impact in all the regions. The

stock of private foreign capital is next.

These variables also have sllbstantial delayed effects. Private foreign capital has

the shortest delayed effect of IIp to two years oruy. Delayed effects are up to five years

for bath development assistance and domestic savings. These findings are consistent

with (i) Bhattacharya, ~[ontiel and Shanna (1997), who have documented evidence

that suggests that foreign commercial investments yield high annllal returns on in­

vested capital, (Le., quick payback of invested capital), (ii) a study by the World Bank

(1984) which suggests that payback of invested capital on the bank's funded projects

(development assistance) are realized more slowly (than indicated by Bhattacharya

Page 141: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

132

et al.), and (iü) investment of domestic capital generally taking a longer term view.

The reslùts suggest that, Overall and in the Western region, the long-nm effects of

increases in the labor force impact export share significantly negatively. In the East

tao, the impact is negative but not significant. The impact of this variable in the

Sudano on export share is positive. Further, this variable exercises significant delayed

effects. Oth~r thau iu the East, the magnitude of the impact of this variable on export

share is larger than any of the capital variables. The Labor variable also exercises

significant delayed effects (up ta five years). This wOllld be the consequence of farro

labor not moving easily between agricultllral sub-sectors, or between agriclùtl1re and

other sectors of the economy.

For producer priees, one finds that the long-nm impact of priee increases favor

the export share of agriclùture everywhere except in the Sndano-Sahel region. This

impact is not significant anywhere. In the East, it is almost significant. This is the

region in wmch it has the largest magnitude. The priee variable has no significant

lags.

The long-nm effect of the openness of these economies on the export share is

positive in all regions but the West. It is however, not significant anywhere. In

regard to the extemal terms of trade, the long-nm effect of this variable on export

share is positive in the Western and Sudano-Sahel regions. It is only significant in

the Iater. The impact is negative and Dot significant in the Eastem-Southern region,

or Overall. Changes in the terms of trade affect the shares of the suh-sectors for up

ta five years.

The long-run impact of the weather affects both sub-sectors almost equally. The

Page 142: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

133

measure of technical change suggests that total factor productivity in the export sub­

sector has been falling in all the regions except in the Sudano-Sahel region. In this

region, there is indication of rising productivity in the export sub-sector.

Thus, this exercise has provided evidence of slow adjustments in response ta

changes in factor inputs and other factors that affect profitability of the agriclùtural

sector in SSA. One wOlùd look ta these slow adjustments as the basis for apparent

differences that mayexist between the short-nm and long-nm impact of the variables

in the models. TItis is done next.

1.2.4 Comparison of Long-Run and Short-Run Estimates

This section compares the short-nm (static) and long-nm (dynanùc) effects of the

stocks of development assistance, private foreign capital, domestic savings, the open­

ness of each economy, the terms of trade and variability of the weather on the export

and domestic shan~ of agriclùtural output. Comparison is for regional findings ooly.

sinee no policy gronp estimates are available for the dynamic mode!.

Long-nin price coefficients exceed short-nm coefficients in the Overall, as weIl as in

the East and West. This is consistent with theory, where the arglunent is that over the

short-nUl, total resources allocated ta agriclùture are generally fixed. Hence changes

in aggregate response will be small, compared to the long-nm when reallocation of

resources within the two sub-sectors of agriculture, and to and from agriclùture wOlùd

have taken place, if price signais were perceived to be permanent. Even in the Sudano

where the price coefficients are negative, the magnitude of the coefficient is larger in

the long-run than in the short-mn. This would still be consistent with the subsistence

Page 143: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

134

agriculture hypothesis advanced. (Here too, in the long-nm more resources would be

allocated away from the export sub-sector, in the case of a positive priee signal).

Unfortlmately, the relationships between the long-run and static coefficients iu

respect of the other variables have not been weil established in the literature. This

study finds that the coefficients of the development assistance variable are not signif­

icant in the export share equation either in the short-nul 01' iu the long-nul, iu the

East, West or Overall. Thus, the impact of development assistance does not henefit

one sub-sector relatively more than the other. [n the production function tao, one

finds that tms variable does not have significant impact on total agricllitural output.

Similarly, tms study finds that private foreign commercial flows are not significant

in the export share equation either in the short-nm or in the long-nm in all the

regions. This variable however, has the tendency to favor the export share both in

the long-nID and in the short-nln in all the regions (except possibly in the West).

The short-nm and long-nm estimates with respect ta domestic savings are dif­

ferent. In the short-nm domestic savings do Ilot sigIÙficantly favor any sllb-sector

overall, but they favor (significantly) export share relatively more in the long-nm.

Short-nln estimates measure only one period impact (in this case one period later),

whereas the long-run estimates measltre as weIl the impact of past changes in this

variable. Ta llnderstand what lS happening better, consider the ADL from which

the Bewley transformation was obtained. (Recall that the specific instnunents used

in the estimation ensme that long-run estimates obtained here would be identical

ta those that would have been obtained if one had deduced them from the ADL).

From the ADL, the long-nm estimates would be obtained as a function of the sum of

Page 144: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

135

coefficients of lagged. domestic savings (refer to equation 3.12). Thus, one concludes

that the efIects of the sum of lagged domestic savings are significant and positively

impact the export share relatively more than they impact domestic share. These

lagged effects are not captured in the static specification. The explanation for the

different behavior of this variable in the short-nm and in the long-nm in the vari­

ous regions follows along similar lines. Indccd, this cÀ~lanation for the behaviol' uf

domestic savings is valid for aIl variables in the mode!.

The impact of the labor force in the Overall is different between the short-run and

long-nm specifications. {ts impact is positive in the static equation, but negative in

the long-nm equation. The explanation is similar to that for domestic savings. In

this case however, increases in the labor force are associated with falling export share

in the Overall in the long-nm. Again one looks to the ADL for interpretation. The

long-nm coefficients are a flmction of the SlUll of lagged coefficients. Unfortunately,

even t.hough the labor force has been increasing, export share has been falling. Thus

the long-nm impact of increases in the labor variable on export share is negative.

(The impact of labor in the EWit, West and Sudano have the same signs bath in the

short-nm and in the long-nm. The magnitudes differ however, but the explanation

for these differences are along the same Hnes as provided far the Overall).

The apenness variable is positive in all regions in the short-nm. It is positive in

the long-nm too for the Overall, East and Sudano. The only discrepancy is in the

West, where it is negative (and very small ) in the long-run but not at all significant.

The extemal terms of trade variable is positive and essentially significant in ail speci­

fications in the short-run, negative and non-significant in the Overall and East in the

Page 145: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

136

long-run, but positive in both West and Sudano and significant only in the later. It

wOlùd appear that given the situation faced by fanners the impact of openness would

not be long lasting. Considering that the impact of openness is to reduce direct and

indirect taxation of agriculture, the benefits of this to farmers is higher producer

priee. But farmers face other constraints beyond priee. Infrastnlcture is a major one.

Thus without addl'essing these otIler l.:OllstraïllLs, the efIecL of opeWless on the export

share will not be strong in the long-omo The tenDS of trade too is not significant

in the long-nm because with falling agricultural commodity priees, the proportion of

other items in this index increases, thus Limiting the role of this variable in explaining

export share (of agriclùtllre) variation in the long-nm.

The weather variable is not significant in either specification. The coefficient of

the trend variable is negative and significant both in the short-run and in the long-run

in all but Sudano. In the Sndano, while significant in t.he short-nm it is not in the

long-nm suggesting sorne improvement in prodllctivity in the export sub-sector in the

log-nm.

Page 146: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

Chapter 8

Conclusions and PolicyImplications

This chapter presents the policy implications arising from this study. Attention is

also drawn ta possible short comings of the findings of dûs study.

8.1 Policy Implications

The policy recommendations of this study have two objectives: ensllring higher agri-

clùtural output, thereby generating more economic activity in the rural areas, hence

leading to higher overall p.conomic growth, and secondly, earning more from SSA

agriclùtural exports to help indllstrialization efforts by helping pay for capital equip-

ment imports, etc. Taking the position that the existing situation in the agriclùtural

sector is unsatisfactory, the recommendations aim to achieve these two objectives by

avoiding the drawbacks of the impact of economic reforms on agriclùture, (as being

currently irnplemented in sorne SSA countries). These drawbacks became apparent

in the investigation of LARGE, SlVlALL and POOR policy groups. (This thesis ac-

cepts that over the next several years the worldwide momentum for economic poliey

reforms is unstoppable). The recommendations faIl into two categories: those that

137

Page 147: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

•138

can he implemented immediately and will have quick impact, and those that wotùd

he implemented over time and will have long-term impact. They vary byagric,ùtural

region.

8.1.1 Increasing Agricultural Output

The findings of this study hclp isolate the areas that need attentioll. The fiuwug

of no struct.ural change in the production functions over time across agric,ùtural re­

gions must be addressed sa as ta improve the situation. The structure of agriclùtllral

production must change, in order that agricultural output may increase significantly.

Agric,ùtural output may be increased by either increasing quantities of inputs, or by

ensuring higher productivity, or both. This study recommends a conlbined approach.

Of particular interest to this stlldy are the l'ales of capital and economic poliey. In

the first place, the impact of the three components of capital is unacceptably low

(sometimes even negative). Secondly, the finding with respect to the policy vari­

able suggests that economic policy in most areas is negatively related to agric,ùtlual

output. AlI these show up in the negative total factor productivity growth. The rec­

ommendations of this stndy faIl into two categories: those that can be implemented

immediatelyand are expected to have quick impact, and those that wo,ùd he impIe­

mented over time and will have long-term impact. They vary by agric,ùtural region

and poliey group.

The basic recommendations are summarized as reduction in the bias in domestic

terms of trade against agriculture, and the need for capital investments in infras­

trnctural projects. Improved domestic terms of trade are urgently required in those

Page 148: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

139

places in which the domestic savings have negative coefficients. This is the case in

the countries of LARGE and S~IALL group and those of the West and Sudano agri­

cultural regions. Economie reform policies as being currently implemented result in

domestic terms of trade being biased against agriclùture because of relatively higher

direct taxation of agricwture (producer priee fixing hy governments, especially for

exports). Indeed, in Ghana and UganJa the leaJiug refofllliug cotwtries in SSA,

Africa Recovery (1997) reports that average annual growth rates of earnings from

agrictùtural exports were -4.3% between 1990 and 1995 for Ghana, a deterioration

over the 1985 - 89 average of 3.2%, and -11.3% between 1990 and 1993 for Uganda,

also a deterioration over the 1985 - 90 average of -7.4%. One thinks this reflects at

least in part, the effect of bia.sed damestic tenns af trade against agriclùture.

The recomendation is far reduction in direct ta.xation of agric\ùture. The conse­

quence of this can he felt within a short-period of time as this sectar beeomes more

competitive and farmers respond immediately by taking better care of existing farms

by being able to afford seed and fertilizer inputs as weIl as fanning implements such

as machetes, cutlasses and hoes and improved seeds. Farm implements are probably

best fashioned damestically in factories or hy local blacksmiths and lathe machine

operatars who know domestie fanning techniques very well. (In most cases, steel

for this may have to he imported). Simîlarly, higher quality seeds that have already

been engineered and/or tested loeally and found to be suitable wOlùd naw became

more affordable. Studies on rnrallinkages are seant, but two studies cited in Hagg­

blade et al. (1987) suggest high expenditure elasticity far loeally produced non-farro

gaods. For example, a 10% increase in rural household incorne 100 to 13% increase

Page 149: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

140

in expenditure on non-agricultural goods and services in Sierra Leone and Nigeria.If

this policy is perceived to be credible and long lasting, it will aIso have long term

impact. To balance the expected shortfall in government revenues due to rednction

in taxes on agriculture, governments will have to impose higher taxes on other sectors

of their economies. One has reservations about the use of the value-added tax being

prOu10ted by the \Vorlù Dallk, ou the gl'UlUlds that the rllechanisms by which taxes

collected will make their way iuto government coffers are plagued with loopholes.

The negative impact of development assistance in the cOlUltries of the West caUs

for a review of those areas of the €Conomy iuto which 0 DA is targeted. A modification

or redirection of development assistance wOlùd be reqllired. It nlay not be possible

to take action on this front immediately however.

Private foreign capital has negative impact on agriclùtllral ontput in the conntries

of the LARGE/SrvIALL group as weIl as those of the Eastern region. To reverse this

may require taking actions t.hat will halt the fiight of foreign capital ont of these

countries, as weIl as hait the re-allocation of private foreign capital (within eotmtry)

away from agriclùture. This will happen when cOlmtries adopt credible palicies that

lower bath economic and political risks that foreign capital faces. This too may not

be achievable in a short-time. As part of a long-t.erm plan, governments sholùd create

incentives that will attract private foreign capital into processing agriclùtllral output

for exporte It is the position of these authors that private foreign capital is luilikely

to be attracted to primary agriculture in a significant way.

In all cases, the long-term objective must be to raise the impacts of aIl capital vari­

ables on agricultural output. Improvements in basic infrastructure especially roads

Page 150: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

141

would be a major step forward. These have been shown in the literature (e.g. Bin-

swanger et al., 1987, and Antle 1983) to have significant impact on agrictùtural output.

Development assistance as weIl as government portion of domestic saving shotùd be

directed to tms end. However, constraints on the provision of public goods resulting

from balaneed budgets, required of reforming eOlmtries (good poliey), severely curb

governments' ability to provide infrastructure. Glven that the state of infrastructure

in Illost SSA econornies is woefully inadequate, this constraint is a serions develop-

mental bottleneck. So is the debt burden. Reforming conntries have been granted

snbstantial loans. Debt servicing conditions attached ta these leave little ftmds for

governments ta make investments. For example, Africa Recovery (1996) reports that

in 1994, the ratio of the amotmt of debt service to total exports for Ghana was 24.6%

and 44.2% for Uganda. l

Investment. of domestic savings in research and development as well as extension

services has aIso not received much attention. Thirtle, Atkins, Bottomley, Gonese,

Govereh and Khatri (1993) have shawn that expenditllres on research and extension

services significantly explain total factor productivity in Zimbabwe for both the com-

mercial and commtmal sectors. In Ghana, government spending on agriclùtnre which

was only 3.9% of recurrent expenditure in 1990, declined to 1.4% in 1996. This paper

takes the view that agriculture must be aceorded a higher priority.

Given the long lead times required for public investments to come on stream, and

the generallag in private capital response to public investments, it is necessary that

1 Many international bankers consider 25% to be dangerously high. Uganda has since (1998) bene­lited from the "Highly Indebted Poor Countries" program. This has lowered her debt burden.

Page 151: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

142

attention be paid ta the provision of public goods early in the economic refonn process

so that valllable time is not lost. It would appear from this study that valuable time is

heing lost. One way to complement domestic government resources in providing public

goods wotùd be a return ta the original vision of development assistance: creating

an enabling environment for private sector development. This is a calI for reduction

in the proportion of development assistance flmds alloèateJ for elUel'geucy rdlef and

peacekeeping. (Based on OECD data, UNCTAD (1997) reports that in 1995,24 % of

bilateral aOA commitments were in the form of food aid and emergency assistance).

This calI wOlùd be realistic if the need for such efforts were rednced by addressing

the canses of civil strife and lUUest, (or if tlmds for snch activities came from other

allocations). It may he argued that depressed economic performance is an important

cause of civil strife and lUlrest. Using development assistance for emergency relief and

peacekeeping only amolUlts ta putting out one tire aftel' another withont addressing

the root canse, wlùch is economic to a good extent.

8.1.2 Increasing Agricultural Export Earnings

Increasing agriclùtural export earnings is important for sllb-Saharan African conn­

tries. Ta this end, reduction in indirect taxation through liberalization (openness)

of the economies (devaluation is one component) is being pursued. However, excess

direct taxation of agriclùture (more than in other sectors) remains. From this study,

one finds that the impact of the three capital variables on agrictùtural exports is

small, as was the case for agricultural output. In the short-term, the magnitudes of

the coefficients are least in the West, followed by Sudano. They are highest in the

Page 152: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

143

East where direct taxes are lower. The deduction from this finding is that eOlmtries

in the West and Slldano have more to do by way of improving domestic tenus of trade

than those in the East. This should he an immediate term poliey. The long-term

impact of all capital variables on export share are positive (except development assis­

tance in Sudano) and small, suggesting that in ail regions, efforts aimed. at increasing

the impact of these variables must he pursued as a long-tenu stl'ategy.

The recommendation for increasing export earnings are two fold, just as those for

increasing agriclùtural output. In the short-term, the focus sholùd be to rehabilitate

(increased inputs) existing farms, while the long-term plan wOlùd be for cOlmtries ta

move away from exporting few primarily raw agriclùtural prodllce to exporting di­

versified and processed agrÎClùtural produce. The arguments posed above for the raIe

of domestic savings, development assistance and private foreign capital in increasing

agriclùtural output will naturally contribute to increasing the volume of agriclùtnral

exports. But in addition, if efforts to process agriclùtural produce prior ta exparting

them are successflù, agriclùtural export earnings wOlùd further increase.

One strong reason for cautioning lmdue emphasis on promoting agriclùture for

external trade across SSA is that on the world stage, SSA countries have compar­

ative advantage in producing sorne agriclùtural primary products. The situation in

SSA is that many COlmtries have common agricmtural exports. For example, co­

coa is common to Cameroon, Cote d'Ivoire, Ghana, Nigeria, Sierra Leone and Togo;

coffee is common to Bunmdi, Cameroon, Cote d'Ivoire, Ethiopia, Kenya, Rwanda,

Tanzania, Uganda and Zaïre; groundnuts are common to Gambia, Mali and Senegal.

Together (even individually in sorne cases) they are important suppliers to the world,

Page 153: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

144

(Gersovitz and Paxson, 1990). Increasing individual country output will surely lower

international priees further. This will restùt in what is sometimes ealled "immeseriz-

ing growth", (that is, growth in output that leads to poverty due to lower prices).2

This situation has been deseribed by sorne authors as static comparative advantage,

which is not in the long-nm interest of SSA. (We live in a dynamic world). This twist

may not have ueell cOllsiJereJ by Roùdk (1997) when he caUs for efforts that ~~ill

sharply incl'ease traditional and non-traditional exports." He is not alone.

The idea of encouraging so called non-traditional exports to diversify the export

base is laudable. As the phrase suggests, these wOlùd be new export products. How-

ever, with common climate, groups of cOlmtries are likely to diversify into similar

activities, most probably agriclùture related. (Global competition in the face of weak

industrial base are other reasons). Diversification of exports ta include many prod-

ncts common to many COlmtries will however, be an improvement over the eurrent

situation. This thesis suggests a broadening of the concept of non-traditional exports

to include vallle-added. items. That is, processing of traditional primary agriclùtural

products. To encourage snch investments, govemments will have to provide incentives

that are at least as attractive as those they currently provide ta the rnining sectar.

In this regard, more and more cocoa beans could be processed into cocoa butter, for

example. In Ghana, a major cocoa producing cotmtry, most of the talk on reforming

the export sub-sector centers on diversifying exports (non-traditional exports), rather

than processing traditional exports.

2 International commodity agreements and cartels are notoriously unstable and are not realisticalternatives.

Page 154: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

145

In discussing trade reform, Rodrik (1997) calls for managing the distribution of

the consequences of trade reform so that winners can emerge early. This study rather

suggests that managing these issues sa that consequences are broad-based (benefit a

large proportion of the population) should be the target. A far reaching way to assure

this wOlùd be for benefits of trade reform to the accnle to the agricultural sector, the

s~tor that illlpacts the lives of the Iuajurit.y of the people.

8.2 Limitations of this Study

Four aspects of this study come to mind as areas that cOlùd use sorne improvernent.

The first is the question of possible bias of parameter est.imates due to the faet that

data are pooled across cOlmtries of sub-Saharan Afriea, a large continent. As noted in

the text, the basis of this is the assllmption that all countries have access to the same

production technology. The point here is the realization that it is possible that. aIl

COlIDtries do not have aceess ta the same production technology. This study addresses

this concem in two ways: (i) by performing statistical tests for poolability, and (ii)

by providing regional estimates. Statistical tests provide support for pooling, while

regional estimates are less likely to be biased since countries in a region are more

likely to have more in common, inc1uding production teclmology.

Another area of sorne concern is the quality of data. Gersovitz and Paxson (1990)

express the point c1early when they wrote,

"... it is useful to keep in mind the diffiC1Ùties with respect to data that

confront any inquiry into economic conditions and prospects in Africa.

These countries ... lack sufficient resources to devote to the collection of

Page 155: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

146

statistics. As a result, information on their economies is limited in scope

and quality and is not avaiJable in a timely fashion ...."

This study address this concem by using data from credible sources, mostly from

the World Bank and The Food and Agriclùtural organization. It is recommended

that strong emphasis be placed on data collection and management issues as part of

structural adjustment programs.

A related issue is the question of conversion of amotmts in domestic currencies

to United States dollars (US$). This study has followed the World Bank's approach

by tlsing the Atlas method to convert domestie currencies to U8$. This method

incorporates so called "official" and 44parallel" market rates. An alternative approach

that is common in the literature is the Purchasing Power Parity approach. This

latter approach addresses the question by trying to establish the qnantities of clifferent

currencies that are required ta buy equivalent quantities of goods and services in their

respective cOllntries. 80th ainl to achieve the same objective. Tms study has tlsed

the Atlas approach because its conversion factors are available for a11 cotmtries of

interest. As well, use of the Atlas approach shotùd be more consistent with the rest

of the data, since most of it is coming from the same source (World Bank).

Finally, the question of an appropriate measure of the human capital of farmers

also needs to be addressed for a more complete understanding of the state of affairs.

Page 156: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

147

REFERENCES

Africa Recovery, vol. 10, No. 2, October 1996, United Nations Office of Com­munications and Public Information.

Airiea Recovery, vol. Il, No. 2, October 1997, United Nations Office of Com­munications and Public Information.

Afriea Recovery, vol. Il, No. 3, February 1998, United Nations Office of Com­munications and Public Information.

Altonji. J. G.. and A. Siow~ 1987~ "Testing the R.flsponsp of Consnmpt.înn t·n

Incorne Changes with (noisy) Panel Data," Quarterly Journal of Economies,102, 293 - 328.

Anderson, G., and R. Bhmdell, 1983, " Testing Restrictions in a FlexibleDynarnic Demand System: An Application to Consumers' Expenditure inCanada," Review of Economie Stud'ies, L, 397 - 410.

Antle, J. rvL, 1983, "Infrastnlctllre and Aggregate Agricultural Productivity:International Evidence," Economie Development and Cultural Change, 31, 609- 620.

Arellano, rvL,1989, ~~A Note on the Anderson-Hsiao Estinlator for Panel Dat.a."Economies Letters 59, 87 - 97.

Bagachwa, rvL S. D., and F. Stewart, "Rural Industries and Linkages in Sub­Saharan Africa: A survey," in Stewart, LaU and Wangwe (eds) AlternativeDevelopment Strategies in Sub-Saharan Africa. St.. rvIartin's Press.

Baltagi, B. H., 1995, Econometrie Analysis of Panel Data, John Wiley & Sons.

Banerjee, A.t J. Dolado, J. W. Galbraith, D. F. Hendry, 1993, Co-'integration,Error-Correction, and the Econometrie A.nalysis of Non-Stationary Data, Ox­ford University Press.

Bardsen, G., 1989, "The Estimation of Long-run Coefficients from Error­Correction tvlodels," Oxford Bulletin of Economies and Statistics, 51, 345 ­350.

Barro, R. and J. W. Lee, 1996, " International tvleasures of Schooling Years andSchooling Quality," American Economie Review Papers and Proceedings, 82(2),218 - 223.

Bautista, R. ~1. and A. Valdes, 1993, "The Relevance of Trade and Macroec~

nomic Policies for AgrictÙture," in R. rvr. Bautista and A. Valdes (00), The BiasAgainst Agriculture, Institute of Contemporary Studies.

Bewley, R., 1979, 'The Direct Estimation of Equilibrium Response in a Linearhtlodel," Economies Letters, 3, 357 - 361.

Page 157: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

148

Bewley, R., 1986, Allocation Models, Harper & Row, Inc.

Bewley, R. and D. G. Fiebeg, 1993, "Why are Long-Run Parameter Estimatesso Disparate," The Review of Economies and Statistics, 345 - 349.

Bhattacharya A., P. J..~Iontiel, and S. Sharma, 1997, " Private Capital Flowsto Sub-Saharan Africa: An Overview of Trends and Determinants," Financeand Development, 3 - 6.

Binswanger, H., 1989, "The Poliey Response of Agriclùture," Proceedings of theWorld Bank Annual Conference on Development Economies.

Binswanger, H. and K. Deininger. 1997, "Explaining Agricultural and AgrarianPolicies in Developing Countries," Journal of Economie Literature, x....XXV. 1958- 2005.

Binswanger, H., ~I. Yang, A. Bowers, and Y. ~hmdlak, 1987, "On the Determi­nants of Cross-cOlmtry Aggregate Agrieultnral Supply," J Durnal of Economet­ries, III - 131.

Bond, rvr. E., 1983, "Agriclùtural Responses to Priees in Sllb-Saharan AfricanCOlmtrîes," [MF Staff Papers 30, no. 4: 703-26.

Boone, P., 1996, " Politics and Effectiveness of Aid." European Economie Re­'mew, 40, 289 - 329.

Bosenlp, E., 1981, Population and Teehnological Change: A Study of Long- TermTrends, University of Chicago Press.

Bonton, L., C. Jones and ~I. Kiguel, 1994, "rvlacroeconomic Reforrn and Growthin Mriea," Policy Research Working Paper 1394, ~[acroeconomic Growth Divi­sion, The world Bank.

Brown, T.~I., 1972, "l'vlacroeconomic Data on Ghana (Part 1)," Economie Bul­letin of Ghana, Vol 2, No.1. 25-53.

Brown, B., and S. l\tlaital, 1981, "What Do Economists Know? An EmpiricalStudy of Experts' Expectations," Econometrica, vol 49, 491 - 504.

Burnside, C., and D. Dollar, 1997, "Aid, Policies and Growth," Policy ResearchWorking Paper 1777, fvlacroeconomic Growth Division, The World Bank.

Centre for Poliey Analysis, 1997, Quarterly Report on Ghana, No. 3, Accra.

Chambers, R. G., 1988, Applied Produ.ction Analysis: A Dual Approach, Cam­bridge University Press.

Chenery, H. B., and A. M. Strout, 1966, "Foreign Assistance and EconomieDevelopment," The American Economie Review, vol LVI, No. 4, Part I.

Page 158: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

149

Chhibber, A., 1988, "Raising Agricultural Output: Price and Non-priee Fac­tors," Finance and Development, 44 - 47.

Christensen, L., D. Jorgenson, and L. Lau, 1973, "Transcendental LogarithmicProduction Frontiers," Review of Economies and Statistics, 55, 28 - 45.

Collier, P., and J. W. Gunning, 1997, Explaining African Economic GrowthPerformance. Paper PresentOO at Tenth Anniversary of CSAE Oxford.

Dagenais, Iv!., 1994, L'Parameter Estimation in Regression 1\IIodei with Errars inthe Variables and Autocorrelated Disturbances," Journal of Econometrics, 64,145 - 163.

Deaton, A. S., and R. I. 1\IIiller, 1995, International Commodity Priees, Macroe­conomie Performance and Politics in Sub-Saharan Africa, International FinanceSection, Department of Economics, Princeton University.

Diewert, W. E., 1974, " Applications of Duality Theory," in IvI. D. Intriligatorand D. A. Kendrick (00), Prontiers of Quantitative Economies II.

Doa:n. T. A., 1992, RATS User's Man'ual, Version 4.

Domar, E., 1946, "Capital Expansion, Rate of Growth and Employment,"Econometriea, 14, 137 - 147.

Duncan, G. J., and L4 An Investigation of the Extent and Consequences of rvIea­surement Error in Labor Economic survey data," Journal of LaboT Economies,3,508 - 532.

Easterly, W., 1\11. Kremer, L. Pritchett, and L. H. Summers, 1993, H Good Polieyor Cood LllCk? C01Ultry Growth Performance and Temporary Shocks," Journalof Nlonetary Economies, 32.

Edwards, S., 1998, "Openness, Productivity and Growth: What Do We ReallyKnow?," The Economic Journal 108, 383 -398.

Elmi, O. S., 1994, Agricultural Priees and Supply Response, 1\II.Sc. Thesis, De­partment of Agriclùtural Economies, rvIcGill University.

Ezeala-Harrison, F., 1996, Economie Development: Theory and Policy Applica­tions, Praeger.

Food and Agriclùtural Organization, 1996, FAO Production Yearbook, Rome.

Food and Agricllitural Organization, 1996, F:40 Trade Yearbook, Rome.

Gersovitz, l\tI., and C. Paxson, 1990, The Economies of Africa and the Prieesof Their Exports, International Finance Section, Department of Economics,Princeton University.

Page 159: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

150

Ghura, D., and !vI. T. Hadjimichael, 1996, " Growth in Sub-Saharan Afriea,"[MF Staff Papers, vol 4, no. 3, 605 - 634.

Gichenje, H., The Impact of Official Development Assistance to African Agricul­ture, M. Sc Thesis, Department of Agriclùtural Economies, McGill University.

Greene, W. H., 1993, Econometrie Analysis, Second Edition, !vlacmillan Pub­lishing company.

Greene, W. H. 1997, Econometrie Analysis, Third Edition, tvlacmillan Publish­ing company.

Grossman, G. Iv!., E. Helpman, 1991, "Trade, KnowlOOge Spillovers, andGrowth," European Economie Review, 35, 517 - 526.

Gunjal K., and H. Gichenje, 1997, Economie Impact of International Develop­ment Assistance on African Agriculture. Paper Presented at the Joint Meetingsof Canadian Agrictùtural Economies and Farm Ivlanagement Society and t.heAmerican Agriclùtural Economies Society at Toronto.

Haggblade, S., P. Hazell, and J. Brown, 1987, " FarmjNon-farm Linkages inRural Sub-Saharan Mrica: Empirical Evidence and Poliey Implications," Dis­cussion Paper, Agrictùture and Rural Development Department, The WorldBank, Washington, D. C.

Haque, U. N., and J. Aziz, 1997, "The Quality of Governance: ~Second Gener­ation' Civil Service Refonn in Africa," HvlF Working Paper.

Harberger, 1978, " Perspectives on Capital and Technology in Less DevelopedCountries." in IvL J. Artis and A. R. Nobay (00), Conte'mporary EconomieAnalysis, London.

Harrod, 1948, Towards a Dynamic Economies, ~Iacmillan.

Hayashi, F., and C. A. Siros, 1983, "Nearly Efficient Estimation of Time SeriesIvIodels with Precietermined, But Not Exogenolls, Instnunents," Econometrica,vol 51, 783-798.

Helleiner, G. K., 1992," Structural Adjustments and Long-Term Developmentin Sub-Saharan Africa," in Stewart, Lall and Wangwe (OOs) Alternative Devel­opment Strategies in sub-Saharan Afriea, St. l'vlartin's Press.

Hendry, Pagan and Sargan, 1984, "Dynamic Specification," in Z. Griliches and1\1. Intriligator (eds), Handbook of Econometries, 2, North-Holland.

Hoch, 1., 1962, "Estimation of Production Function Parameters CombiningTime-series and Cross-Section Data," Econometrica, vol 30, no. 1, 34 - 53.

Holtz-Eakin, D., W. Newey, and H. S. Rosen, 1988, " Estimating Vector Au­toregressions with Panel Data," Econometrica, vol 56, no. 6, 1371 - 1395.

Page 160: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

151

Hsiao, C., 1986, Analysis of Panel Data, Cambridge University Press.

Jaeger, W., 1992, "Effects of Economie Policy on Africau AgrictùtuIe," WorLdBank Discussion Paper, Africa Technical Department, series No. 147.

Johnston, J., 1984, Econometrie Methods, third edition, l\tIcgraw-Hill.

Kawagoe, T., Y. Hayami, and V. W. Ruttan, 1985, " The Intercountry Agri­cultural Production Function and Productivity Differences Among COtmtries,"Journal of Development Econornics 19, 113 - 132.

Kcanc, ~'L P, and D. E. , Rllnklc, 1D92, "On the Estimation of Panel-Datal\tlodels with Seriai Correlation When Instnunents Are Not Strictly Exogenous,"Journal of Business & Economie Statisties, vol 10, no. 1, 1 - 9.

Kesavan, T., Z. A. Hassan, H. H. Jensen, and S. R Johnson, 1993, "Dynamicsand Long-Rlm Structure in U.S. lVIeat Demand," Canadian Jo'urnal of Agricul­tural Economies 41, 139 - 153.

Kelley, A., and R. Schmidt, 1994, "Poplùation and Incorne Change: RecentEvidence," World Bank Discussion Paper 249, \Vorld Bank. Washington. D.C.

Khan, l\tL S., 1987, " lVlacroeconomic Adjustment in Developing COlmtries: APolicy Perspective," World Bank Research Observer, 2, 23 - 42.

Kherallah, lVI. W., J. C. Beghin, E. W. Peterson and F. J. Ruppel, 1994, "Im­pacts of Official Development Assistance on Agriclùtllral Growth, Savings andAgriclùtural Imports." AgrieuLtural Economies Il, 99 - 110.

Killick, T., 1985, l'Economic Environment and Agricultllral Development - TheImportance of ~IacroeconornicPolicy," Food Poliey, 29 -40.

Kizerbo, J., 1989, "Discussant's Contribution," in L. Enunerrij, One World orSeveral'l, 257 - 261.

Kmenta, J., 1967, "On Estimation of the CES Production F\mction," Interna­tional Eeonornie Review, 8, 180 - 189.

Kmenta, J., 1986, Elements of Econometries, rvlacmillan.

Kohli, U. R., 1978, "A Gross National Prodllct Flmction and Derived Demandfor Imports and Sllpply of Exports," Canadian Journal of Economies, XI, no.2, 170 - 182.

Krueger, A., ~1. Schiff, and A. Valdes, 1991, The Political Economy of Agrieul­turai Pricing Poliey, Johns Hopkins, Baltimore.

Lau, L. J., and P. A. Yotopoulos, 1971, "A Test of Relative Efficiency andApplication to Indian Agriculture," Ameriean Economie Review, 69, 94 - 109.

Page 161: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

152

Lau, L. J., and P. A. Yotopoulos, 1989, "The Meta-Produetion Function Ap­proach to Technological Change in World Agriculture," Journal of DevelopmentEcono'mies 31, 241 - 269.

Lawrence, D., 1989, "An Aggregator ~'lodel of Canadian Export Supply andImport Demand Responsiveness," Canadian Journal of Economies, 22, 503 ­521.

LeIe, U., 1990, Agrieultural Growth and Assistance to Africa: Lessons of aQuarter Century, les Press.

Lcvin, A., and L. K. Raut, 1!J97, "Complementarities betw€ell Expol'ts àHU

Ruman Capital in Economie Growth: Evidence from the Semî-IndustrializedCOlmtries," Economie Development and Cultural Change, 155 - 174.

Levine, R., and D. Renelt, 1992, " A Sensitivity Analysis of Cross-CountryGrowth Regressions," American Economie Review 32 (9), 1777 - 1795.

Lucas, R. E., 1988, "On the l\lechanies of Economie Development," Journal offlI/onetary Economies, 22, 3 - 42.

Levy, V., 1988, "Aid and Growth in Snb-Saharan Africa: The recent Experi­ence," European Economie Review 32, 1777 - 1795.

Loayza, N. V., 1994, "A Test of International Convergence Hypothesis UsingPanel Data," Policy Research Working Paper 1333, l\Iaeroeconornic GrowthDivision, The World Bank.

l\IacFadden, D. L., 1971, ~, Cost, Revenue and Profit F\metions," in D. L. l\Iac­Fadden (00), The Econometrie Approaeh to Production TheonJ. Amsterdam.

l\IcCalla, A., 1998, "Revitalizing the World Bank's Approach to Rural Devel­opment." Talk presentOO at the l\rcGill Econonùc Policy l\Ianagement SeminarSeries.

l\Iankiw, G. N., D. Romer, D. N. Weil, 1992, "A Contribution to the Empiricsof Economie Growth," Quarterly Journal of Economies, vol 107, 407 - 37.

l\Iankiw, G. N, 1995, The Growth of Nations, Brookings Papers on EconomieAetivity l, 275 - 326.

~Iartin, W. J., and P. G. Warr, 1993, "Explaining the Relative Decline of Agri­clùture: A supply-Side Analysis for Indonesia," The World Bank EconomieReview, vol. 7, No. 3, 383 -401.

lVlartin, W. J., and P. G. Warr, 1994, "Determinants of Agrieulture's RelativeDecline: Thailand," Agricultural Economies 11,219 - 235.

~Iosley, P., J. Hudson, and S. Horrel, 1987, " Aïd, Public Sector and the l\Iarketin Less Developecl Countries," Economie Journal 97, 616 - 641.

Page 162: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

153

Mundlak, Y., 1978, "On the Pooling of Time Series and Cross-Sectional Data,"Econometrica 46,69 - 86.

Nehru, V., and A. Dhareshwar, 1993, "A New Database on Physical CapitalStock: Sources, Methodology, and Results," Revista Analisis de Eeonomico, vol8, no. 1, 37 - 59.

Nehru, V., E. Swanson, and A. Dubey, 1995 "A New Database on HumanCapital Stock: Sources, ~Iethodology, and Results," Journal of DevelopmentEconomies, 379 - 401.

Neter, J., \V. \Vasserman, and hi!. H. Kutner, 1996, Appl~ed L1.near Stat'lstzcaLNJodels, Fourth Edition, Irwin.

Newey, W., and K, West, 1987, " A Simple Positive 8emi-Definite, Heteroscedas­ticity and Autocorrelation Consistent Covariance fvlatrix," Econometrica, 55,703 - 708.

Ng, F., Md A. Yeats, 1997, " Open Economies Work Better! Did Africa'sProtectionist Polides Cause Its rvlarginalization in World Trade?" World De­velopment, Vol 26, no. 6. 889 - 904.

Norton, G. W., J. Ortiz and P. G. Pardey, 1992, L'The Impact of Foreign As­sistance on Agrictùtural Growth," Economie Development and Cultural Change40, 775 - 786.

ûECD, 1994, Development Co-operation: 1993 Report. Paris.

Pakes, A. and Z. Griliches, 1984, "Estimating Distribnted Lags in Short Panelswith an Application ta the Specification of Depreciation Patterns and CapitalStock Constnlcts," Review of Economie Studies 51, 243 - 262.

Papanek, G. F., 1973, "Aid, Foreign Private Investment, Savings and Growthin Less Developed COlmtries," JO'urnal of Political Economy, 120 - 130.

Pritchett, L., 1994, "Poplùation, Factor Accumtùation and Productivity,"Working Paper, ~Iacroeconomic Growth Division, The World Bank.

Pritchett, L., '~Where's all the Education Gone," Working Paper, ~[acroeco­

nomic Growth Division, The World Bank.

Rodrik, D., 1997, T'rade Policy and Economie Performance in Sub-SaharanAfrica. Paper Prepared for the Swedish ~Iinistry of Foreign Affairs.

Romer, P. rvI., 1986, " Increasing Returns and Long-nm Growth," Journal ofPolitical Economy, 94, 1002 • 37.

Romer, P. rvI., 1990, " Endogenous Technical Change," Journal of PoliticalEconomy, 98, 871 - 103.

Page 163: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

154

Ruttan, V. W., 1989, "Why Foreign Assistance," Economie Development andCultural Change 37, 411 - 24.

Sachs, J., and A. Warner, 1995, Brookings Papers on Economic Activity l, 1 ­118.

Samuelson, P. A., 1953..4, "Priees of Factors and Goods in General equilibrium,"Review of Economic Studies, 21, 1 - 20.

Schiff. 1\11., and C. ~Iontenegro, 1997, "Aggregate Agricultural Supply Responsein Developing COlmtries," Economic Development and Cultural Change, vol 45,No. 2, 393 - 410.

Solow, R., 1956, " A Contribution to the Theory of Economie Growth," Quar­terLy Journal of Economics, LXXI, 312 - 320.

Thirtle, C., J. Atkins, P. Bottomley, N. Gonese, J. Govereh and Y. Khatri,1993, " Agricultural Prodnctivity in Zimbabwe," The Economic Journal. 103,474 - 480.

Toro-Vizcarrondo, C., and T. D. Wallace, 1968, " A Test of the lVlean SquareCriterion for Restrictions in Linear Regression," Journal of the American Sta­tistical Association 63, 558 - 572.

UNCTAD, 1997, " The Least Developed COlmtries 1997 Report," New York.

Varian, V.,1992, Microeconomie Analysis. Third Edition, W. W. Norton andCompany.

White, H.~ 1980, "A Heteroscedasticity-Consistent Covariance ~Iatrix Estima­tor and a Direct Test for Heteroscedasticity," Econometrica, 48, 817 - 838

World Bank, 1984, Tenth Annual Review of Project Performance Audit Results.Washington, o. C.

World Bank, 1989, Africa: From Crisis ta Sustainable Development. Washing­ton D. C.

World Bank, 1994, Adjustment in Africa: Refonns, Results and the Raad Ahead,Washington, O. C.

World Bank, 1995, World Tables, Baltimore: Johns Hopkins University for theWorld Bank.

World Bank, 1996a, African Development indicators, Baltimore: Johns HopkinsUniversity for the World Bank.

World Bank, 1996b, World Debt Tables, Washington, D.C.

World Bank, 1997, Global Development Finance, Washington, D. C.

Page 164: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

APPENDIXA

Coefficients of the Overall export share ADL1• Legend on next page.

Variable2 Coefficient Variable2 Coefficient

AEXP{1} 0.547** ln(LABR) -0.013

(12.458) (-0.328)

AE.YP{3} 0.088** ln(LABR){3 } -0.161**

(2.271) (-2.745)

ln(KODA) -0.016 ln(LABR){5} 0.130**

(-1.200) (2.475)

ln(KODA){5} 0.019 OPEN{1} 0.054

(1.414) (1.706)

ln(KPFX) 0.017 EXOI 0.050*

(1.031 ) (2.495)

ln(KPFX){l} 0.026 EX02 -0.058*

(1.171) (-2.564)

ln(KPFX){2} -0.033* ln(PRICE){l} 0.009

(-1.957) (0.702)

ln(KSAV) -0.008 ln(PRICE) {4} 0.004

(-0.768) (0.328)

ln(KSAV){4} 0.025** TREND -0.005**

(2.383) (-4.056)

155

Page 165: INFORMATION TO USERS - McGill Universitydigitool.library.mcgill.ca/thesisfile35672.pdf · shoulùer to lean on dwing the later halfofmy studies here. ... approvisionnement eninfrastructure

1 t-statistics in parentheses below coefficient estimates.

* and ** indicate statistical significance at 0.05 and O.011evels respectively.

2 Variables are defined in chapter 3. Nwnbers in {.} represent lag nwnber.

156