OPTIMAL INVESTMENT DECISIONS OF COFFEE … · COFFEE FARMERS IN VIETNAM ... Nguyen Le Hoa, Truong...

OPTIMAL INVESTMENT DECISIONS OF

COFFEE FARMERS IN VIETNAM

Tran Cong Thang

A thesis submitted for the degree of

Doctor of Philosophy at

THE UNIVERSITY OF WESTERN AUSTRALIA

(School of Agricultural and Resource Economics)

October 2011

Acknowledgements

i

Acknowledgements

I would like to give special thanks to my principal supervisors Prof. Michael Burton and

Dr. Donna Brennan for their role in supervising my work on this thesis. They are

generous with their wisdom and encouragement. They are not only my supervisors but

also my very good friends in Australia. Sincere thanks are due to my supervisor Prof.

Ben White, for his helpful comments and suggestions.

The supportive co-operation of staff and colleagues from the School of Agricultural and

Resource Economics and AusAID officers at the University of Western Australia

gratefully acknowledged - particularly Jan Taylor, Deborah Swindells, Theresa Goh,

Sally Marsh, Vilaphonh, Sharon Harvey, Rhonda Haskell, Cathy Tang, Christine Kerin,

Deborah Pyatt and Alicia Zabah.

My sincere thanks go to Dr. Dang Kim Son for his approval and encouragement.

Sincere thanks are due to my colleagues in the Center for Agricultural Policy for their

useful support: Nguyen Ngoc Que, Nguyen Do Anh Tuan, Tran Thi Quynh Chi,

Nguyen Le Hoa, Truong Thi Thu Trang, Pham Huong Giang, Nguyen Nghia Lan and

Phan Van Dan.

Support from my friends in Perth during my program is gratefully acknowledged;

particularly from Ngoc Linh, Bich Ngoc, Lan Huong, Trung Khanh, Manh Hieu, Thanh

Nhan, Tu-Huong, Ha-Trong, Lam-Huong, Van Liem and Doc Lap.

The generous financial support from the Australian Centre for International Agricultural

Research (ACIAR) I gratefully acknowledge.

Finally, I owe thanks to my parents and my wife for their great support, encouragement

and love.

Perth, August 2010

Abstract

ii

Abstract

For perennial crops like coffee, identifying the prices at which farmers should cut or

replant is a key investment decision. Optimal cutting and replanting can help farmers

use their resource to maximise income. The response of individual households may also

significantly affect the supply response at the aggregate level. In addition, knowledge of

how supply responds to different variables can help planners forecast supply.

While coffee is a major crop in Vietnam, the number of studies on the aggregate supply

response, or on optimal production decisions at farm level, is limited. In this study, two

types of models are developed to analyze the supply response of coffee in Vietnam and

identify optimal investment and production decisions of coffee farmers. The first model

uses the fixed form optimization method to solve a stochastic optimal control model of

the household’s cutting and replanting problem. The model identifies optimal ‘trigger’

coffee prices for cutting and replanting. The second predictive model estimates an

aggregate coffee supply function. This model gives the determinants of coffee area

variation in Vietnam.

The results from the stochastic optimal control model explore the optimal trigger prices

in different scenarios. The first result shows that coffee farmers can have a low ‘trigger’

price for cutting and a high trigger price for replanting. Between these two values is a

range of price values for which there is a ‘hysteresis effect’ where neither cutting

standing trees nor replanting occurs. Second, if this model extends to allow age

dependent trigger prices, as opposed to fixed trigger prices, the income is significantly

increased. These results reckon that farmers should not cut the trees before the 11th

year

of planting

A third set of simulations analyse the importance of credit on the availability of working

capital. Poor or cash constrained households are more likely to remove coffee trees

compared to the unconstrained households. Thus, the cutting frequency of the poor

households is higher than that of the non-poor for all ages of coffee trees. The cash

constraint leads to a higher cutting percentage of coffee trees for poor households,

especially young trees. Furthermore, the poor farmers wait for significantly higher

trigger prices before replanting. Due to their cash constraint, farmers cannot follow the

Abstract

iii

same decisions as their richer neighbors, and this reduces their average income. The

credit available to poor household has a significant impact on the poor farmer’s income

and behavior. However, the importance of credit depends on the age of coffee trees: it is

more important for farmers with young trees that have not started to produce as harvest.

Fourth, if the model is generalised to allow for the short-run application of inputs such

as fertilizer, this has a significant impact on income. Furthermore, farmers are much less

likely to cut if they can adjust input use efficiently in response to coffee price changes.

Fifth, the cutting/replanting decision of farmers is influenced by the profit of alternative

crops. If the profit of the alternative crop falls, farmers are less likely to cut and more

likely to replant.

Previous studies of the coffee supply response for Vietnam assumed that the supply

function of coffee was symmetric with respect to changes in prices. The possibility of

an irreversible response was neglected. The results of this study show that the response

of coffee supply in Vietnam to price is asymmetric: it reacts more to a price increase

than to a price decrease. The asymmetric response of coffee area at the aggregate level

to price changes is consistent with the optimal decisions of farmers because they

optimize their decision by different ‘trigger’ prices for cutting and replanting. The area

response is similar across regions.

Studies of the response of coffee at both aggregate and farm level are useful for

households and planners. Despite some limitations, the modeling approaches used in

this thesis can be applied to supply response and farmers’ decision for other perennial

crops.

Table of Contents

iv

Table of Contents

Acknowledgements ............................................................................................................ i

Abstract ............................................................................................................................. ii

Table of Contents ............................................................................................................. iv

List of Tables.................................................................................................................. viii

List of Figures ................................................................................................................... x

Table of Abbreviations ................................................................................................... xiii

CHAPTER 1. INTRODUCTION ..................................................................................... 1

1.1. Rationale of Study .................................................................................................. 1

1.2. Objectives ............................................................................................................... 3

1.3. Methodology .......................................................................................................... 4

1.4. Data ........................................................................................................................ 6

1.5. Thesis Structure ...................................................................................................... 7

CHAPTER 2. THE COFFEE SECTOR IN VIETNAM ................................................... 9

2.1. Introduction ............................................................................................................ 9

2.2. Agricultural Sector in Vietnam .............................................................................. 9

2.3. Coffee Production ................................................................................................ 13

2.4. Coffee Export ....................................................................................................... 18

2.5. Coffee Households ............................................................................................... 23

2.5.1. Farm Size and Distribution ........................................................................... 23

2.5.2. Starting Year of Coffee Production............................................................... 25

2.5.3. Income Sources ............................................................................................. 27

2.5.4. Profitability of Coffee Production ................................................................. 28

2.5.5. Source of Water............................................................................................. 30

2.6. Conclusion ........................................................................................................... 32

CHAPTER 3. STOCHASTIC OPTIMAL INVESTMENT DECISION FOR

PERENNIAL CROPS: A LITERATURE REVIEW ...................................................... 34

3.1. Introduction .......................................................................................................... 34

3.2. Theoretical Models for Optimal Investment Decision ......................................... 35

3.3. Faustmann Model with Risk ................................................................................ 40

3.4. Stochastic Optimal Control Methods ................................................................... 41

Table of Contents

v

3.4.1. Dynamic Programming (DP) ........................................................................ 41

3.4.2. Real Option Approach .................................................................................. 45

3.4.3. Other Techniques of Solving the Complex Dynamic Stochastic Models ..... 53

3.5. Conclusion ........................................................................................................... 55

CHAPTER 4. OPTIMAL REPLANTING AND CUTTING RULES FOR COFFEE

FARMERS IN VIETNAM: FIXED YIELD MODEL ................................................... 57

4.1. Introduction .......................................................................................................... 57

4.2. Coffee Farm System in Dak Lak .......................................................................... 58

4.3. Model Structure .................................................................................................... 61

4.3.1. Objective Function ........................................................................................ 61

4.3.2. Profit Function .............................................................................................. 63

4.3.3. Decision Rule ................................................................................................ 63

4.3.4. Yield Function ............................................................................................... 67

4.3.5. Production Cost ............................................................................................. 68

4.3.6. Price Simulation ............................................................................................ 69

4.3.6.1. Lagged Price Model ..................................................................................... 70

4.3.6.2. Price Cycle Model ........................................................................................ 72

4.3.7. Procedure for Estimation .............................................................................. 75

4.4. Results of the FY Model ...................................................................................... 78

4.4.1. Optimal Rule with Lagged Price Model ...................................................... 78

4.4.2. Impact of Substitute Crop on Coffee Farmer’s Decision .............................. 84

4.4.3. Optimal Rules with Price Cycle Simulation Model ...................................... 85

4.5. Conclusion ........................................................................................................... 89

CHAPTER 5. OPTIMAL COFFEE PLANTING DECISIONS UNDER A CASH

CONSTRAINT ............................................................................................................... 91

5.1. Introduction .......................................................................................................... 91

5.2. Impact of Cash Constraints on Farmer’s Decision .............................................. 91

5.3. Poverty Trends in Vietnam .................................................................................. 93

5.3.1. Saving and Income Level in Vietnam ........................................................... 98

5.3.2. Relationship between Income and Expenditure of Poor Farmers ............... 102

5.4. Structure of the FY-CC Model ........................................................................... 105

5.4.1. Objective Function ...................................................................................... 107

5.4.2. Profit, Yield and Production Cost Function ................................................ 108

5.4.3. Expenditure, Saving and Loan .................................................................... 108

5.4.4. Decision Rule .............................................................................................. 110

Table of Contents

vi

5.5. Results of the FY-CC Model ............................................................................. 111

5.5.1. Impact of Cash Constraints on Income ....................................................... 112

5.5.2. Effect of Loans and Savings ....................................................................... 113

5.5.3. Optimal Rule for Poor Coffee Farmers ....................................................... 117

5.6. Conclusion ......................................................................................................... 122

CHAPTER 6. SHORT-RUN RESPONSE AND OPTIMAL RULES FOR COFFEE

FARMERS IN VIETNAM ........................................................................................... 124

6.1. Introduction ........................................................................................................ 124

6.2. Review of Literature on Yield Response Functions .......................................... 124

6.3. Coffee Yield Function in Vietnam ..................................................................... 127

6.3.1. Yield Coffee Function Estimation .............................................................. 127

6.3.2. Optimal Cost Specification by Output Price ............................................... 130

6.3.3. Supply Price Elasticity ................................................................................ 131

6.4. Variable Yield Model (VY model) .................................................................... 132

6.4.1. Model Structure ........................................................................................... 132

6.4.2. Adjustment of Yield function...................................................................... 133

6.4.3. Optimal Rule of the VY model ................................................................... 136

6.5. The Variable Yield – Cash Constraint Model (VY-CC model) ......................... 140

6.5.1. Model Structure ........................................................................................... 140

6.5.2. Optimal Rule of the VY-CC model ............................................................ 141

6.6. Conclusion ......................................................................................................... 145

CHAPTER 7. SUMMARY OF THE OPTIMAL MODELS ........................................ 147

7.1. Introduction ........................................................................................................ 147

7.2. Model Development ........................................................................................... 147

7.2.1. Objectives of Models .................................................................................. 147

7.2.2. Rules and Constraints .................................................................................. 148

7.3. Changes in Coffee Farmer’s Decision ............................................................... 150

7.4. Conclusion ......................................................................................................... 154

CHAPTER 8. COFFEE SUPPLY RESPONSE IN VIETNAM ................................... 155

8.1. Introduction ........................................................................................................ 155

8.2. Literature Review on Supply Response Analysis using the Econometric

Approach ................................................................................................................... 156

8.2.1. Nerlovian Approach .................................................................................... 157

8.2.2. Extended Nerlovian Approach .................................................................... 160

8.2.3. Wicken - Greenfield Approach ................................................................... 161

Table of Contents

vii

8.2.4. Price Asymmetric Response ....................................................................... 163

8.2.5. Previous Studies on Supply Response of Coffee in Vietnam ..................... 167

8.3. Empirical Model of Coffee Supply Response in Vietnam ................................. 169

8.3.1. Data ............................................................................................................. 169

8.3.2. Model Results ............................................................................................. 169

8.4. Conclusion ......................................................................................................... 177

CHAPTER 9. CONCLUSIONS ................................................................................... 178

9.1. Background ........................................................................................................ 178

9.2. Key results .......................................................................................................... 181

9.2.1. Response at the Farm Level ........................................................................ 181

9.2.1.1. Cutting and Replanting Decision ............................................................... 181

9.2.1.2. Impact of Cash Constraints on Farmer’s Behaviors .................................. 182

9.2.1.3. Change of Farmer’s Decision with Short-run Response ............................ 183

9.2.1.4. Impact of the Profitability of the Substitute Crop ...................................... 184

9.2.2. Coffee Supply Response at Aggregate Level ............................................. 184

9.3. Policy implications ............................................................................................. 185

9.4. Limitations ......................................................................................................... 186

9.5. Further Studies ................................................................................................... 188

References ..................................................................................................................... 190

Appendix A ................................................................................................................... 199

Appendix B ................................................................................................................... 206

Appendix C ................................................................................................................... 210

Appendix D ................................................................................................................... 211

List of Tables

viii

List of Tables

Table 1.1: Sample of Coffee Farm Survey 2007 in Dak Lak............................................ 6

Table 2.1: Key Economic Indicators of Vietnam............................................................ 10

Table 2.2: GDP in agriculture, forestry and fisheries 2005-2008 (current price, %) ..... 10

Table 2.3: Area and output of crops in Vietnam, 1990-2008.......................................... 11

Table 2.4: Agricultural commodity exports in Vietnam, 2007-2008 (mill. $) ................ 12

Table 2.5: Value and share in import-export of agricultural commodity ....................... 12

Table 2.6: Changes in coffee production in different periods (%) ................................. 14

Table 2.7: Coffee production by region in Vietnam, 2008 ............................................. 17

Table 2.8: Main markets for Vietnamese coffee in 2005 and 2008 ................................ 20

Table 2.9: SWOT analysis of coffee ............................................................................... 22

Table 2.10: Number of perennial crop households and size in Vietnam ........................ 23

Table 2.11: Distribution of coffee household by groups................................................. 26

Table 2.12: Average crop area of coffee households by district (m2) ............................ 27

Table 2.13: Earning sources of coffee households in 2006 ($) ....................................... 28

Table 2.14: Coffee farm performance in Daklak province, Vietnam 2006 ($/ha) ........ 29

Table 2.15: Main source of water (%) ............................................................................ 30

Table 2.16: Is yield limited by water (%) ....................................................................... 31

Table 4.1: Percentage of household with other activities excluding cropping (%) ........ 60

Table 4.2: Percentage of households reducing coffee area ............................................ 60

Table 4.3: Percentage of farmer switched to other crops ................................................ 60

Table 4.4: Coffee production cost by age of tree (US$/ha) ............................................ 69

Table 4.5: Distributions of actual international price and price data set simulated from

two models ...................................................................................................................... 74

Table 4.6: Summarized results of different cutting rules of FY model .......................... 83

Table 5.1: Perceived causes of poverty in Dak Lak Province ........................................ 95

Table 5.2: Coffee farming in Central Highlands ............................................................. 96

Table 5.3: Poverty incidence of coffee farmers in Central Highlands, Vietnam ............ 97

Table 5.4: Household income and expenditure in rural area in 2006 ($/year) ................ 99

Table 5.5: Household income and saving in rural by region in 2006($)....................... 100

Table 5.6: The saving flows of household by types ...................................................... 101

List of Tables

ix

Table 5.7: Number of poor households by region in VHLSS2006 ............................... 103

Table 5.8: Regression between per capita income and expenditure of poor HHs ........ 104

Table 5.9: General data of poor coffee household ........................................................ 105

Table 5.10: Loan amount and duration ......................................................................... 109

Table 5.11: Main loan purpose (% respondent) ............................................................ 110

Table 5.12: Percentage of loan by different sources by districts .................................. 110

Table 6.1: Sample distribution of coffee households in Agrocensus_2006 .................. 127

Table 6.2: Regression Results of Coffee Yield Function .............................................. 128

Table 6.3: Comparison of optimal rule between FY and VY model ............................ 137

Table 6.4: Average cost and yield from the VY model and the FY model................... 139

Table 6.5: The results of FY model with average cost and yield from VY model ...... 139

Table 7.1: Main objective of models ............................................................................ 148

Table 7.2: Decision rules and constraints ..................................................................... 149

Table 7.3: Main results of simulation models ............................................................... 150

Table 8.1: Data series and source .................................................................................. 169

Table 8.2: Estimated results from different models ...................................................... 172

Table 8.3: Results for testing the difference of coefficients among provinces, Modified

Wolfram model with window=6 ................................................................................... 174

Table 8.4: Elasticities of coffee acreage to price .......................................................... 175

List of Figures

x

List of Figures

Figure 2.1: Migration to Dak Lak province, 1976 to 2000 ............................................. 15

Figure 2.2: Coffee area development in Vietnam, 1975-2008 ........................................ 16

Figure 2.3: Coffee area structure by age group in Vietnam, 2007 .................................. 18

Figure 2.4: Quantity and value of coffee export in Vietnam, 1991-2008 ....................... 19

Figure 2.5: Vietnam coffee price export, 1988-2008 (FOB-$ per tonne) ....................... 19

Figure 2.6: Coffee export value of Vietnam by destination (%) ..................................... 20

Figure 2.7: Exporting cost for coffee in some countries (cent/lb) .................................. 21

Figure 2.8: DRC of coffee and other export commodities in Vietnam ........................... 22

Figure 2.9: Coffee Household Structure by farm-size (%) ............................................. 24

Figure 2.10: Distribution of coffee households across Vietnam ..................................... 24

Figure 2.11: Coffee household structure by number of plots (%)................................... 25

Figure 2.12: Percentage of new coffee farmers and farm-gate price .............................. 26

Figure 2.13: Distribution of coffee production cost, 2006 ($ per ha) ............................. 30

Figure 3.1: The determination of PH and PL .................................................................. 51

Figure 4.1: Development of coffee area in Dak Lak, 1986-2008 ................................... 59

Figure 4.2. Example of optimal cutting and replanting rule ........................................... 67

Figure 4.3: Coffee yield by age of tree ........................................................................... 68

Figure 4.4: Fitted and actual value of logarithm of price ($/kg) ..................................... 71

Figure 4.5: Examples of price trajectories predicted from Lagged price model ............. 72

Figure 4.6: Price cycle of coffee in the world market (UScent/lb) ................................. 72

Figure 4.7: Example of price trajectories predicted from Price Cycle model ................. 74

Figure 4.8: Model Structure Map ................................................................................... 77

Figure 4.9: Optimal cutting and replanting rule from the FY model .............................. 79

Figure 4.10: Proportion of actual cut in FY model with Lagged price simulation ......... 80

Figure 4.11: Comparison of optimal RP of FY model and farmer’s estimates .............. 81

Figure 4.12: Optimal quadratic CP and best constant CP from the FY model .............. 82

Figure 4.13: The maximum ENPV per ha among different CP rules ............................. 82

Figure 4.14: ENPV with different starting ages for quadratic CP and no cutting rule ... 83

Figure 4.15: Changes in optimal rule when maize profit varies ..................................... 84

Figure 4.16: Changes in the maximum ENPV when maize profit varies ....................... 85

List of Figures

xi

Figure 4.17: Optimal rule of the FY model with Price cycle model ............................... 86

Figure 4.18: Optimal rules of Price cycle and Lagged price simulations ....................... 86

Figure 4.19: Distribution of farm-gate price data set simulated from two models ......... 87

Figure 4.20: Simulated percentage cut at each age of trees from two data sets .............. 88

Figure 4.21: Different optimal rules of FY model with Price cycle simulation ............. 89

Figure 4.22: Actual cut of the FY with Price cycle model and different CP forms........ 89

Figure 5.1: Poverty trend in Vietnam 1993-2006 ........................................................... 94

Figure 5.2: Poverty trend in Vietnam by regions 1993-2004.......................................... 95

Figure 5.3: Coffee area of poor farmers in Central Highlands by provinces ................. 97

Figure 5.4: Structure of family size of poor and non-poor coffee household ................. 98

Figure 5.5: Per capita income and expenditure by area in 2006 ($/year) ..................... 100

Figure 5.6: Per capita income and expenditure of the poor in rural areas by regions,

2006 ($) ......................................................................................................................... 101

Figure 5.7: Fitted per capita income and expenditure by family size ........................... 104

Figure 5.8: Plot of per capita income and expenditure of the poor ............................... 105

Figure 5.9: ENPV from FY-CC if imposing optimal rule of FY and optimal ENPV from

FY ($/poor farm) ........................................................................................................... 112

Figure 5.10: ENPV from FY-CC at different initial savings at annual loan of 625$ .. 113

Figure 5.11: ENPV of farm income at different annual loans and savings................... 114

Figure 5.12: ENPV of FY-CC with different starting age of trees and loans ............... 116

Figure 5.13: Optimal Rules of the FY-CC model ......................................................... 118

Figure 5.14: Optimal rule of FY model and FY-CC model .......................................... 118

Figure 5.15: Actual cutting percentages by age of trees ............................................... 119

Figure 5.16: Impact of cutting decision by CP rule and by cash constraint in the FY-CC

model ............................................................................................................................. 120

Figure 5.17: Optimal rule of the FY-CC model with different initial savings .............. 121

Figure 5.18: Optimal rules of FY model and FY-CC with initial saving of $1500 ...... 121

Figure 6.1: Coffee yield – age relationship at average cost .......................................... 129

Figure 6.2: Cost –yield relation for 11 year old coffee trees ........................................ 130

Figure 6.3: Simulation of cost and yield relationship (age of tree =11 year old, medium

yield level district)......................................................................................................... 131

Figure 6.4: Simulation of price and coffee yield........................................................... 132

Figure 6.5: Coffee yield by age of tree in FY model .................................................... 133

Figure 6.6: Yield in the FY model and Adjusted Yield in the VY model at cost of

$930/ha .......................................................................................................................... 135

List of Figures

xii

Figure 6.7: Yield variation at different production costs .............................................. 135

Figure 6.8: Optimal cutting and replanting rules in the VY model .............................. 136

Figure 6.9: Optimal rules for FY model and VY model ............................................... 137

Figure 6.10: Percentage of cases in which farmers cut coffee at optimal rule.............. 138

Figure 6.11: Optimal rule of the VY-CC model ........................................................... 142

Figure 6.12: A comparison of optimal rule between FY-CC and VY-CC model......... 142

Figure 6.13: Percentage of cases in which farmers cut coffee at optimal rules ............ 143

Figure 6.14: Percentage of actual cut of the VY-CC model by cutting rule and by cash

constraint ....................................................................................................................... 143

Figure 6.15: Comparison of ENPV from different models at poor farm-size ............. 144

Figure 7.1: Maximum ENPV achieved from models ................................................... 151

Figure 7.2: Optimal cutting and replanting rules in different models ........................... 152

Figure 7.3: Actual cutting percentage at optimal rules in FY and FY-CC model ........ 152

Figure 7.4: Cutting percentage at optimal rule in FY and VY model ........................... 153

Figure 7.5: Optimal replanting prices for different models .......................................... 154

Figure 8.1: Hypothetical response overtime of Wolffram model and Modified Wolffram

model ............................................................................................................................. 165

Figure 8.2: Fitted and actual area from Modified Wolffram model (ha) ...................... 176

Table of Abbreviations

xiii


ACIAR Australian Centre for International Agricultural Research

ADB Asian Development Bank

Agrocensus Agricultural Census Survey

Agroinfo Information Center for Agriculture and Rural Development

CAP Centre for Agricultural Policy

CH Central Highlands

CPI Consumer Price Index

CP Cutting Price

DARDD Department of Agriculture and Rural Development of Dak Lak

DP Dynamic Programming

DRC Domestic Resource Cost index

EMA Equivalent Mature Area

ENPV Expected Net Present Value

FY Fixed Yield Optimal

FY-CC Fixed Yield-Cash Constraint Optimal

GDP Gross Domestic Product

GSO General Statistical Office of Vietnam

ICARD Information Center for Agriculture and Rural Development

ICO International Coffee Organization

IPSARD Institute of Policy and Strategy for Agriculture and Rural

Development

LP Linear Programming

MARD Ministry of Agriculture and Rural Development

Mill. million

MOLISA Ministry of Labour and Invalid Social Affair

MRD Mekong River Delta

NCC North Central Coast

NES Northern East South

NPV Net Present Value


xiv

RP Replanting Price

RRD Red river delta

SCC South Central Coast

VHLSS Vietnam Household Living Standard Survey

VND Vietnam Dong

VY Variable Yield Optimal Model

VY-CC Variable Yield – Cash Constraint Optimal

YMA Yield of Mature Area

$, USD United State Dollar

Chapter 1. Introduction

1

CHAPTER 1. INTRODUCTION

1.1. Rationale of Study

Vietnam has an agriculture-based economy in which the sector accounts for about 22

percent (in 2008) of Gross Domestic Product (GDP). The coffee crop, with its rapid

expansion in the 1990s, became the second largest export agricultural commodity after

rice. In 2007, the total coffee exports of Vietnam were over one million tonnes, with a

value of $1.8 billion. At present Vietnam is the second largest coffee exporter in the

world. In addition, the coffee sector plays an important role in labour absorption in rural

areas. In the peak harvest season, the coffee sector employs about 800,000 workers (The

World Bank, 2002).

Since the early 1990s, the coffee area in Vietnam has increased rapidly, from only

100,000 ha in 1990 to about 600,000 ha in 2000 (GSO, 2001)1. Coffee trees are planted

in 30 provinces in Vietnam, of which Dac Lak, Dak Nong, Gia Lai, Kon Tum and Lam

Dong in the Central Highlands are the main producing areas with 90 percent of national

output.

Natural conditions in the Central Highlands of Vietnam are favorable for coffee

cultivation. In addition, labour in rural areas in Vietnam is abundant and relatively

cheap. Thus, the production costs for coffee in Vietnam are normally lower than that in

other countries such as Brazil, Colombia and Indonesia (PI-IPSARD, 2007). Other

studies by Oxfam (2002) and CAP (2006)2 also show that Vietnam has a comparative

advantage in coffee production.

Coffee production in Vietnam is dominated by small farm households. The average

coffee area of coffee farms in Vietnam is only 8799 m2 and in total 477,000 coffee

farms, over 60 percent had less than one ha of land (GSO, 2007).

1 GSO denotes General Statistical Office of Vietnam

2 CAP signifies Centre for Agricultural Policy, Hanoi.


2

Many coffee farmers are relatively poor and are from ethnic minorities. Over 30 percent

of coffee farmers in the Central Highlands belong to minority ethnic groups3 and about

25 percent of coffee households are poor people4. Thus, coffee farm households are

typically economically vulnerable, and so the coffee sector contributes significantly to

development and household income in the rural economy.

As a major coffee exporter, the coffee price for Vietnamese producers is determined on

the world market. This means that Vietnamese producers are vulnerable to price

variability. This clearly shows by the low prices in the early 2000s. The farm gate

coffee price in Central Highlands reduced from $2 per kg in 1995 to less than $0.25 in

2001 (Thang, 2008). According to the Financial Times, coffee farmers in Vietnam's

Central Highlands lost $172 million from their 2000/01 coffee crop because of low

world prices5. In Dak Lak province, where household incomes had grown by 9 per cent

annually from 1996 to 1999, this growth rate fell by reductions of around 10 per cent

when the price of coffee fell. By 2002, about 45 per cent of coffee growing households

lacked adequate food, 66 per cent had bank debts and nearly 45 per cent had members

of the family who had turned to off-farm wage labour to earn money (Oxfam 2002).

When the price of coffee dropped, many smallholders fell into debt, they could not

afford to pay the loans and input costs. Due to losses, some coffee farmers cut down

trees and switched to other crops. In Dak Lak province alone, during 2002 and 2003

over 25,000 ha of coffee trees were cut down and replaced with other crops6. The rapid

decline in price reduced the coffee area in Vietnam from 600,000 ha in 2001 to about

491,000 ha in 2005. The cutting decision raises several issues.

First, replacing coffee by other crops is a complex and difficult decision for farmers. If

they switch too early, and the coffee price increases sharply then they bear the cost of

re-establishing trees. High replanting costs and other switching costs are a significant

proportion of costs for coffee households, especially poor households.

Second, it is likely that the cutting decision will depend heavily on the age of the coffee

trees. It may be appropriate to cut older trees first as the coffee price reduces. Thus, the

3 Vietnam has 54 people’s groups in which Kinh is major group with occupation of 87% total population.

Others are ethnic people. This number is calculated by author based on Agricultural Census in Vietnam in

2006. 4 According to the definition of the Ministry of Labour and War Invalid and Social Affairs: The people

are defined as poor are ones whose per capita income of less than $12.5 per month for rural areas and

$16.25 per month for urban areas. The poverty issues will be mentioned more detail in Chapter 5. 5 Available at http://www.allbusiness.com, accessed on 14/09/2009

6 DARDD: Department of Agriculture and Rural Development of Dak Lak.


3

relationship between the age of coffee trees and optimal cutting prices are an important

parameters.

Third, the identification of price at which farmers should replant coffee is also important

because it can help farmers (or their advisors) optimize their decision and significantly

increase their income. Farmers may consistently replant before or after the optimal

replanting price.

Fourth, it is possible that many poor coffee households cut earlier to meet cash

requirements for household expenditure, and inputs for subsistence crops. The

replacement of coffee by annual crops such as maize, and rice may help them meet their

cash requirements in the short-term. However, this decision may be more costly in the

long-term than keeping coffee and waiting for a price increase. The availability and cost

of credit is critical to this decision. If credit is cheaply available then farmers could

delay the cutting decision by borrowing to cover living expenditure and input costs.

Fifth, understanding the response of input application to coffee price may help farmers

to reduce short-run production costs to improve their income. The relationship between

the short-run response of inputs and the cutting decision for those under cash constraints

is important to understand.

Finally, coffee farmers may optimize their decision by different ‘trigger’ prices for

cutting and replanting. This asymmetric response of coffee households could reflect in

an asymmetric response of the aggregate coffee area to price changes. Thus, analyzing

the pattern of coffee supply in Vietnam may show an asymmetric response to price.

An understanding of the coffee supply response in Vietnam provides useful insights for

participants in the coffee sector and policymakers. Identification of optimal cutting and

replanting rules will be beneficial to coffee producers.

1.2. Objectives

Although agriculture and rural development in Vietnam has received great attention

from researchers from both Vietnamese institutions and international organizations,

there have been limited attempts to study coffee supply response in Vietnam either in

aggregate or at farm level. The only published research on Vietnamese coffee farmer

cutting and replanting decisions is by Luong and Loren (2006). There have been two


4

studies that present estimates of the coffee supply function (Tien, 2006, EDE-IPSARD,

2007). Both considered a symmetric price response of coffee area in Vietnam and

estimated the supply function based on aggregate time series data. The lack of empirical

research on coffee in general and on farmer’s decisions in particular is a principal

motivation for this study.

The detailed objectives of the study are:

1. Determine the optimal cutting and replanting rule for coffee farmers in Vietnam, in

particular:

identify the price at which farmers should cut and replant coffee to maximise their

income;

investigate how the cutting rule changes with age of the coffee trees;

examine how optimal rules change if profit from a substitute crop varies

2. Assess the loss of farm income due to cash constraints, in particular:

examine the relationship between expenditure and saving of poor farmers;

investigate how saving and loan availability affects the income and optimal rules of

farmers

3. Analyze how much farmers can improve their income if they can adjust crop input

levels:

first, examine the relationship between the yield of coffee and variable costs using

cross-sectional data

second, investigate changes of farmers’ planting/cutting behaviors and income if they

have an optimal short run response

4. Estimate the supply response function for coffee in Vietnam based on time series

data.

1.3. Methodology


5

Determining the optimal replanting and cutting rules for coffee farmers is a stochastic

control problem. In this study, the Dynamic Model with Fixed Form Optimization is

used to solve this problem. The Dynamic Model with Fixed Form Optimization is

structured as a system of equations (such as revenue, cost and profit) and decision rules

for cutting and replanting. The core objective of the models is to find the optimal cutting

rules and replanting price to maximise income per unit of land under price uncertainty7.

A number of optimal models are developed:

The Fixed Yield Optimal Model (FY model) investigates the optimal cutting and

replanting price for coffee farmers. The FY model maximises the net present value from

a unit of land (here one ha). In this model, the coffee yield function only depends on the

age of trees. In addition, representative farmers in the FY model are not restricted by

cash constraints when making replanting decisions.

Fixed Yield-Cash Constraint Optimal Model (FY-CC model) expands the FY model by

adding cash constraints into the FY model. In the FY-CC model, representative farmers

take cutting/replanting decisions in a household context, in which expenditure, other

income and loan factors constrain choices. Farmers spend their income on living

expenses and inputs including hired labour. Thus, the decision rules for the FY-CC

model are more complicated than in the FY model. Farmers cannot continue producing

coffee if their total budget (income and loans) cannot cover household expenditure and

production costs. Similarly, they cannot resume coffee production if their budget is less

than the replanting cost.

Further development of the FY model and the FY-CC model that captures the short-run

response of farmers is presented through Variable Yield Optimal Model (VY model) and

Variable Yield – Cash Constraint Optimal Model (VY-CC). In these models, coffee

yield varies according to the use of variable inputs, and the level of variable inputs is

determined by the price of output. The VY model and VY-CC model answer the main

question of how much farmers can improve their income if they can follow an optimal

adjustment of input application in the short-run.

The “Positive” approach estimates the aggregate coffee supply function using historical

time series data. The data used includes coffee area, output, world price, export price,

7 In the model, it is assumed that in cases where farmers cut their coffee trees they will switch to maize.

Thus, the income from farm land is the sum of coffee and maize income.


6

domestic price, consumer price index (CPI). The data covers a period from 1986 to

2007.

1.4. Data

The data used in this study stems from three main sources. The most important source is

the Coffee Farm Survey in early 2007 in Dak Lak province by the author. The total

sample of the Coffee Farm Survey is 150 households in three districts of Dak Lak: Cu

Mgar, Krong Pak and Eakar. All farmers in the Coffee Farm Survey are private

smallholders.

Table 1.1: Sample of Coffee Farm Survey 2007 in Dak Lak

Districts

Surveyed

Communes

Number of interviewed

households

Cu Mgar Cuor dang 25

Cu se 25

Krong Pak Hoa Tien 25

Ea Kuang 25

Eakar Cu ri 25

Ea pal 25

Total 150

Source: Thang (2008)

The survey collected general information about the households (farm size, land area,

education, sources of income), coffee production (coffee area, yield, output, number of

plots, sale price, input use, input price, irrigation), credit issues and response of coffee

farmers to price uncertainty. The full questionnaire is presented in Appendix C.

The Agricultural Census Survey in 2006 (Agrocensus_2006) from General Statistical

Office consists of several secondary surveys of which Coffee Efficiency Survey and

General Household Survey are used to estimate yield coffee function and analyze coffee

household characteristics of Vietnam.

Vietnam Household Living Standard Surveys (VHLSS) is the source of additional data..

The General Statistic Office (GSO) implemented the first VHLSS in 1992. From 1992

to 2002, the VHLSS was done every 5 years. After 2002, the VHLSS has been

implemented every 2 years. VHLSS covers many aspects of household income and

expenditure. In this study, the VHLSS are used to investigate saving and expenditure of

poor coffee households and to investigate the relationship between income and

expenditure.


7

1.5. Thesis Structure

Following this introductory chapter, Chapter 2 reviews the recent development of the

coffee sector in Vietnam. The main coffee household characteristics discussed in the

chapter are from the Coffee Farm Survey of 2007. An analysis from Chapter 2 provides

a review of existing issues of the coffee sector in Vietnam.

Chapter 3 reviews the literature on optimal planting and clearing decisions. This chapter

discusses different methods of solving the stochastic optimization problems, such as

dynamic programming, real option theory, and approximate dynamic programming.

Chapters 4, 5 and 6 report the alternative models of optimal cutting/replacement rules.

Chapter 4 develops the Fixed Yield Optimal Model (FY model) to determine the cutting

and replanting price to maximise the expected NPV per hectare given uncertain coffee

prices. Chapter 4 also investigates changes in farmer’s decision when the income of the

substitute changes.

In Chapter 5, the Fixed Yield-Cash Constraint Optimal Model (FY-CC model)

investigates modifications to the optimal rules to account for a situation where coffee

farmers are poor and they do not have enough money to cover annual costs or invest in

new trees. A brief analysis of poor coffee farmers’ income- expenditure relationship is

included in this chapter.

Chapter 6 explores the optimal cutting and replanting decisions of coffee farmers with a

variable coffee yield function. An estimated yield function identifies the relationship

between yield, production cost and age of trees. Thus, optimal yield becomes a function

of the coffee price. By including the coffee yield function into the FY model and FY-

CC model two new models are obtained - the Variable Yield Optimal model (VY

model) and the Variable Yield – Cash Constraint Optimal Model (VY-CC model). This

chapter contains a discussion of the income and the farmer’s decision, including the

short-run response.

Chapter 7 presents a synthesis of results from Chapters 4, 5 and 6.

Chapter 8 estimates the supply response function of coffee in Vietnam based on

aggregate series data from 1985 to 2007. This chapter reports both symmetric and


8

asymmetric response functions, with the latter providing an improved explanation of

coffee areas in Vietnam.

Chapter 9 concludes the thesis and summarizes the main results – discussing the

limitations of the study and suggesting ideas for further studies on these issues.

Chapter 2 The coffee Sector in Vietnam

9

CHAPTER 2. THE COFFEE SECTOR IN VIETNAM

2.1. Introduction

This chapter provides an overview of the agricultural sector in Vietnam. It also reviews

recent developments in the coffee sector and highlights the importance of coffee in the

rural economy. In addition, this chapter describes the characteristics of farm households

producing coffee in Vietnam. The chapter is organised as follows. Section 2.2 presents a

brief overview on the agricultural sector in Vietnam. Section 2.3 and Section 2.4 discuss

coffee production and export trends, and Section 2.5 provides an analysis of the main

characteristics of coffee households. Some conclusions are presented in Section 2.6

2.2. Agricultural Sector in Vietnam

Vietnam is a developing country with average GDP in 2008 of $1020 per capita per

annum. The total population of Vietnam in 2008 was 86.2 million people of which

nearly three-fourths live in rural areas. Agriculture plays a central role in Vietnam’s

economy. The GDP from agriculture, forestry and fisheries accounted for 22 percent of

the national economy in 2008 (GSO, 2009). Vietnam has achieved strong growth in

agricultural production and trade over the past twenty years. This is commonly

attributed to infrastructure investment in irrigation and perennial crops before the 1988

“Doi Moi”(Innovation) policy changes that encouraged: market-oriented production and

input use; the allocation of individual land use rights; sound macroeconomic policies;

and improved credit access for farmers. The annual growth rate of the agricultural sector

has been maintained at a record high level of 3.7 per cent per annum for the five years

from 2003 to 2008 (MARD, 2008).

The rapid growth of the economy after “Doi Moi” policy has benefited most of the

Vietnamese population. However, the country remains one of the worlds’ poorest and a

relatively high 16 per cent of the population was below the poverty line in 20068.

8 The poverty incidence in Vietnam will be presented in more detail in Chapter 5.


10

Agriculture contributes about 22 percent of GDP but employs about 70 percent of the

workforce. This reflects the low labour productivity in the agricultural sector compared

to the rest of the Vietnamese economy.

Table 2.1: Key Economic Indicators of Vietnam

Indicators Vietnam

Population in 2008 (000 persons) 86210

Rural population, 2008 (%) 73

GDP per capita (US$) 1020

Agricultural GDP per capita, 2008 (US$) 304

Share of extended-agriculture in GDP, 2008 (%) 22

Share of extended-agriculture in labour force, 2008 (%) 70

Annual agricultural GDP growth, 2004-2008 (%) 3.7

Annual nonagricultural GDP growth, 2004-2008 (%) 8.7

Poverty rate in 2006* (%) 16

Share of rural poor in total poor in 2006* (%) 90

Source: GSO, 2008 *GSO, 2007

Note: Extended agriculture consists of agriculture, forestry and fisheries

Agriculture is the engine of rural development, accounting for 77 percent of GDP for

the extended agricultural sector (agriculture, forestry and fisheries). In recent years,

fisheries have become more important and accounted for nearly 20 percent of GDP in

agriculture, forestry and fisheries (see Table 2.2).

Table 2.2: GDP in agriculture, forestry and fisheries 2005-2008 (current price, %)

2005 2006 2007 2008

Agriculture 75.6 75.3 75.0 77.2

Forestry 5.7 5.4 5.2 4.9

Fisheries 18.7 19.3 19.8 17.9

Total 100 100 100 100

Source: GSO per com.

Growth in production has been across most food and industrial crops, with only jute and

cotton showing a reduction in output. Since the early 1990s, perennial crops have shown

the highest growth rates. In 2008, the total area of tea was 129,000 ha, more than double

the tea area in 1990. Over the 1990 to 2008 period, the coffee area increased 4.4 times

to over 500,000 ha in 2008. Pepper and cashew also increased significantly. In the same

period, the area of pepper rose 5.43 times, from 9200 ha in 1990 to 104,000 ha in 2008.


11

Prior to 1990, cashews were a minor crop, however, from 1992 rapid growth has led to

over 300,000 ha in 2008 (Table 2.3).

Rice is the mainstay of smallholders in the agricultural sector and is important for

ensuring food security. Between 1990 and 2008, rice area and production grew 1.22

times and 2.01 times, respectively. The rapid development of rice changed Vietnam

from a rice importer to an exporter; in 1999 Vietnam ranked third in rice exports behind

Thailand and the United States.

Some annual industrial crops such as sugar, cassava and soybean also increased

markedly. The area under some minor commodities (jute, cotton and tobacco) have

tended decline in the 1990-2008 period (see Table 2.3).

Table 2.3: Area and output of crops in Vietnam, 1990-2008

Crop

1990 2008 Change in area

2008/1990

(times) Area

(000 ha)

Output

(000 tonnes)

Area

(000 ha)

Output

(000

tonnes)

Paddy 6042.8 19225.1 7400.2 38729 1.22

Cassava 256.8 2275.8 543.8 9090.3 2.12

Cotton 19.2 3.1 5.8 8 0.30

Jute 11.6 23.8 3.3 7.8 0.28

Rush 11 63.6 11.7 84.8 1.06

Sugarcane 146.4 5405.5 270.7 16145 1.85

Peanut 217.4 213.2 255.3 530.3 1.17

Soybean 97.3 86.6 192.1 267.6 1.97

Tobacco 31.4 21.8 16.6 28.8 0.53

Tea 59.9 145.1 129.6 706.8 2.16

Coffee 119.3 92.0 525.1 996.3 4.40

Rubber 221.7 57.9 618.6 662.9 2.79

Pepper 9.2 8.6 50 104.5 5.43

Cashew 79*

23.7*

404.9 313.4 5.13

Agricultural land 6693 9436 1.4


Note: * data for cashew area and output are for 1992

Vietnam has become a major world exporter of several agricultural products. Vietnam is

the largest exporter of Robusta coffee and pepper and the second largest exporter of rice

and cashew. Rice is the largest export commodity in terms of value from Vietnam. In

2008, the export value of rice was over $2.7 billion. Coffee is the second largest export

commodity in agriculture, with an export value of $2.1 billion in 2008. Alongside


12

agricultural commodities, aquaculture and forestry contribute markedly to earnings from

international trade. In 2008, export value of aquaculture and forestry was about $7.4

billion (see Table 2.4).

Table 2.4: Agricultural commodity exports in Vietnam, 2007-2008 (mill. $)

2007 2008 % change 2008/2007

Agriculture 6153 8572 + 39.3

Coffee 1881 2116 + 12.5

Rubber 1296 1675 + 29.2

Rice 1472 2758 + 87.4

Tea 128 147 + 15.2

Cashew nut 642 914 + 42.4

Groundnut 30 14 - 45.8

Pepper 267 310 + 16.1

Fruit/vegetable 303 394 + 30.0

Sugar 5 9 + 60.7

Milk, Milk products 35 76 + 117.1

Oils 47 101 +114.9

Meat, meat products 46 57 + 25.7

Aquaculture 3752 4436 + 18.2

Forestry 2564 3004 + 17.2

Wood, wood products 2330 2764 + 18.6

Bamboo, Jute 218 223 + 2.3

Cinnamon 16 17 + 6.3

Total 12469 16012

Source: AGROINFO per com9

In 2008, the value of agriculture, forestry and fishery export accounted for over 25

percent of total export of Vietnam (Table 2.5)

Table 2.5: Value and share in import-export of agricultural commodity

Year Export Import

Value (bil. $) (%) Value (bil. $) (%)

2007 Total 48.56 100 62.68 100

Extended agriculture 12.47 25.68 7.02 11.20

Others 36.09 74.32 55.66 88.80

2008 Total 62.90 100 80.40 100

Extended agriculture 16.01 25.45 10.14 12.61

Others 46.89 74.55 70.26 87.39

Source: GSO (2007) and GSO (2009)

9 Agroinfo denotes Information Center for Agriculture and Rural Development.


13

The development of the agricultural sector has contributed greatly to improving

livelihoods of people in rural areas as well as the positioning of Vietnam in international

markets. However, despite the success of “Doi Moi”(“Innovation”), the agricultural

sector in Vietnam still has a number of economic problems, such as high labour surplus

in rural areas, limited land endowment, high poverty incidence, low labour productivity

and environmental degradation (Son, 2008).

2.3. Coffee Production

Coffee is an important part of Vietnam’s economy – even though the price collapsed in

early 2000s it is still the second largest export agricultural commodity after rice, and

employs over 600,000 workers, rising to nearly 800,000 workers at the peak of the

season or 2.93 percent of the agricultural labour force (The World Bank, 2002).

French missionaries introduced coffee to Vietnam in 1857. However, coffee production

developed after the unification of Vietnam in 1975. Following reunification and as a

result of resettlement programs to move people from densely populated provinces

towards sparsely populated provinces in the Central Highlands, the coffee area more

than doubled to reach nearly 45,000 ha in 1985 (GSO, 2000). Although ethnic

minorities predominated in the region, almost all of the immigrants and most of the new

coffee farmers were Kinh people. However, the growth rate of coffee area in 1975-1980

was not high, only 3.56 percent annum on average (see Table 2.6).

Coffee production has expanded rapidly since 1980, especially after the “Doi Moi”

(Innovation) policy in 1986 that transformed Vietnam from a centrally planned to

market-oriented economy. From 1980-1990 the rapid development of coffee was

fostered by a policy of land use reforms and relaxed government control. Before 1981,

farmers belonged to a cooperative and land belonged to the government. The production

contract (“Contract 100”) policy introduced in 1981 radically changed the role of

households. Contract 100, reallocated land to individual households, meaning they were

still members of cooperatives but it gave farmers more rights in the management of their

land. They were required to produce a predetermined quota by cooperatives but they

could sell output above the quota. However, maximum forestry land allocated to

farmers was only three ha. In addition, all input purchases were through the cooperative

(fertilizer, seed, pesticide and irrigation water).


14

In 1988, the Contract 10 policy introduced rights for households which had not been

covered by the Contract 100 policy. According to Contract 10, farmers had control over

the entire management of their land including the purchase and utilization of inputs. The

land, assigned for a maximum of 15 years gave households a strong incentive to invest

in their farms.

The land reforms (Contract 100 and Contract 10) and mass migration to the Central

Highlands are the main drivers of the rapid development of the coffee sector from 1980

to 1990. During that period, the coffee area in Vietnam grew on average by 23.2 percent

per annum. With considerable improvement in yield, the coffee output in that period

had a very high annual growth rate of over 40 percent (Table 2.6).

Table 2.6: Changes in coffee production in different periods (%)

Period 1975-1980 1981-1990 1991-1999 2000-2008

Area 3.56 23.29 17.27 -1.71

Output 9.51 41.58 22.73 7.95

Yield 6.38 13.06 5.72 9.81


In the following five-year period, coffee production continued expanding rapidly. The

area of coffee increased from around 100,000 ha in early 1990s to nearly 600,000 ha in

1999. A trade liberalization policy within Vietnam and relatively high coffee prices

since early 1990 drove the development of coffee during this period. The export price of

coffee in Vietnam increased from about $700 per ton in 1990 to over $2000 in the

middle of the 1990s (GSO, 2008). Furthermore, land reform continued improving

through the Land Law in 1993, strengthening the rights of households. According to the

Law, households can use, exchange, transfer, lease, inherit, and mortgage their land.

Furthermore, land allocates to households for long-term use (20 years for annual crops

and aquaculture, 50 years for perennial crops). In addition, the maximum arable area for

perennials was not limited. By removing restrictions to expansion, the Land Law in

1993 facilitated the development of the coffee sector. During the period from 1991 to

1999, the coffee area kept increasing, on average, 17.2 percent per year. The second half

of 1990s was a boom period for coffee in Vietnam. The coffee area rose from nearly

160,000 ha in 1993 to 600,000 ha in 1999 (see Figure 2.2).

Together with land policy reform, the re-settlement policy generated mass migration to

the Central Highlands and contributed significantly to the expansion of the coffee area


15

(D’Haeze et al., 2005). Besides large planned migration flows, spontaneous migration

was also significant. Attracted by high economic returns for Robusta coffee,

spontaneous migration increased rapidly between 1991 and 1995. According to the

Settlement Committee of Dak Lak province, about 100,000 people moved to this

province in 1991-1995 (see Figure 2.1). Immigrants to the Central Highlands exploited

uncultivated land to grow crops (mainly coffee) but the expansion of agricultural land in

Central Highland resulted primarily from the conversion of forestland. For example,

between 1976 and 2001 forest cover in Dak Lak province decreased by approximately

235,000 hectares to approximately 1 million hectares, approximately the increase in

coffee area (D’Haeze et al., 2005).

Figure 2.1: Migration to Dak Lak province, 1976 to 2000

Source: Settlement Committee of Dak Lak per com

The coffee price collapse in the early 2000s had a detrimental effect on coffee

producers, processors, traders and exporters. In response to the price fall, many coffee

farmers cut their trees and switched to other crops. At that time, the coffee area in

Vietnam reduced, from 600,000 ha in 2001 to about 491,000 ha in 2005. The

government response to the crisis was to reduce the coffee area. In 2001/2002, the

Vietnam government suggested a reduction of about 150,000 ha of coffee. Furthermore,

provincial authorities in the main coffee areas such as Dak Lak, Lam Dong proposed to

keep the existing area of coffee but not allow any expansion (The World Bank, 2004).

More recently, however, in response to an increase in the coffee price, many producers

have replanted coffee trees as the price has been gradually increasing (see Figure 2.2).

0

40000

80000

120000

160000

200000

1976-80 1981-1985 1986-90 1991-95 1995-99 2000

Nu

mb

er

of

pe

op

le

Official migration

Spontaneous migration


16

Alongside the rapid expansion of area, coffee yield has been increasing, especially in

the past 10 years. During 1998-2006, coffee yield had an average annual growth rate of

9.8 percent. In general, however, the coffee sector in Vietnam has been developed

extensively via area expansion, although yield improvements have contributed to

output increase.

Figure 2.2: Coffee area development in Vietnam, 1975-2008


There are two coffee varieties in Vietnam– Robusta and Arabica, of which Robusta is

the major one, accounting for more than 95 percent of the total coffee cultivation area.

Coffee has been planted mainly in the North West of Vietnam, Central Highlands (see

Table 2.7)10

. The Central Highlands region (including Dak Lak, Dak Nong, Gia Lai,

Kon Tum, Lam Dong province) is the main coffee production area, producing over 90

percent of the national coffee output. Dak Lak is the largest coffee province with total

area of about 180,000 ha (approximately 35 percent of national area).

10

The regions of Vietnam are mapped in Figure A1 in Appendix A. The coffee area and output maps in

Vietnam are presented in Figure A3 and Figure A4 in Appendix A.

0

100000

200000

300000

400000

500000

600000

700000

1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008

Are

a (

ha)

0

500

1000

1500

2000

2500

Yie

ld (

kg

/ha)

Area (ha)

yield(kg/ha)


17

Table 2.7: Coffee production by region in Vietnam, 2008

Area New plant

Harvested

area Yield Output

(ha) (ha) (ha) (quital/ha) (tonnes)

North West 4478 707 2 931 12.46 3651

Dien Bien 1029 575 374 16.63 622

Son La 3449 132 2 557 11.85 3029

North Central Coast 6621 472 5 491 14.75 8097

Thanh Hoa 102 82 15.61 128

Nghe An 1269 49 1 077 14.08 1516

Quang Tri 4335 209 3 681 16.65 6128

Hue 915 214 651 4.99 325

South Central Coast 1729 79 1 650 13.15 2170

Binh Dinh 278 278 7.01 195

Phu Yen 1174 70 1 104 14.00 1546

Khanh Hoa 277 9 268 16.01 429

Central Highlands 480774 10023 456862 21.69 990924

Kon Tum 10360 476 9626 22.61 21764

Gia Lai 76368 337 75788 17.76 134595

Dak Lak 182434 2946 173233 23.98 415494

Dak Nong 75470 2898 70341 19.40 136484

Lam Dong 136142 3366 127874 22.10 282587

North East South 37 306 3 011 33 246 15.32 50931

Binh Phuoc 11 130 556 10 215 12.92 13198

Dong Nai 17 729 2 189 15 516 16.30 25294

Binh Thuan 1 381 92 994 13.55 1347

Ba Ria-Vung Tau 7 066 174 6 521 17.01 11092

Source: ICARD per com.11

.

Despite its position as a major exporter, the quality of Vietnamese coffee is still

relatively low due to inferior harvest and post-harvest technologies. For instance, about

65 percent of Vietnam’s coffee is graded second class due to high proportions of black

and broken beans and high humidity. This affects the price of Vietnam’s coffee in the

world market (Chi et al., 2009).

11

ICARD denotes Information Center for Agriculture and Rural Development, Ministry of Agriculture

and Rural Development, Vietnam.


18

The normal production cycle of coffee trees is from 20 to 25 years, of which the first

two to three years is the gestation period. Normally, the coffee tree starts to produce

‘‘berries’ from the third year. From the eighth to sixteenth year, the tree reaches its

highest yield. The age distribution of coffee trees in 2007 is represented in Figure 2.3.

Figure 2.3: Coffee area structure by age group in Vietnam, 2007

Source: MARD, 2008

With a relatively large share of aging trees, there is a prospect that the quality and yield

of coffee in Vietnam will decline in coming years. Thus, policies for encouraging

farmers to replace old trees are a big concern of the Vietnam governments.

2.4. Coffee Export

More than 90 percent of coffee in Vietnam is exported. The export quantity has

increased rapidly since the early 1990s from less than 100 thousand tonnes in 1991 to

over 1 million tonnes in 2007. Similarly, export value rose to over $2 billion in 2008.

During the period 1991-2008, export quantity grew on average by 17 percent per year,

while value increased on average at 30 percent annum (see Figure 2.4 ).

0-4 years

5%

5-9 years22%

10-15 years40%

15-20 years24%

>20

years9%


19

Figure 2.4: Quantity and value of coffee export in Vietnam, 1991-2008

Source: MARD per com

Vietnam is currently the second largest coffee exporter after Brazil. In 2008, Vietnam

contributed over 40 percent of Robusta coffee and 13 percent of overall coffee trade in

the world market. During the 1995 to 2002 period, the export value of coffee has

stagnated due to the falling real price. The export price reduced from over $2000 per ton

to less then $500 per tonne in 2001. Since 2002, the coffee price has recovered

gradually, resulting in an increase in export value and quantity12

.

Figure 2.5: Vietnam coffee price export, 1988-2008 (FOB-$ per tonne)

Source: MARD per com.

12

The time series data of coffee exports from Vietnam is presented in Table B3 in Appendix B.

0

500

1000

1500

2000

2500

0

200

400

600

800

1000

1200

1400

mil

.$

000 t

on

s

Quantity (000 tons)

Value (Mil.USD)

0

500

1000

1500

2000

2500

3000

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

$/tonne


20

The number of countries to which Vietnam exports is increasing over time. In 2000,

Vietnam exported coffee to only about 50 countries. In 2005, the number rose to 80

countries and in 2008, Vietnamese coffee has been exported to about 100 countries

throughout the world. The main export markets are the EU (Germany, Switzerland,

England, Netherlands, Spain, Italy …), USA and Asia (Japan, Singapore, China,

Philippine, Malaysia and Indonesia); those markets accounts for 58 percent, 15 percent

and 21 percent of total export coffee value of Vietnam in 2008, respectively (see Figure

2.6).

In 2005 In 2008

Figure 2.6: Coffee export value of Vietnam by destination (%)

Source: General Custom Office of Vietnam per com.

Table 2.8 gives the main destinations of Vietnamese coffee in 2005 and 2008. The

United States and Germany, Italy, Japan, Spain are the main export markets.

Table 2.8: Main markets for Vietnamese coffee in 2005 and 2008

Destination In 2005

Destination In 2008

Quantity

(000 tonnes)

Value

(mil. $)

Quantity

(000 tonnes)

Value

(mil. $)

United States 117.7 97.5 Germany 138.5 274.1

Germany 92.1 76.1 United States 131.5 211.4

Italia 62.6 54.2 Italy 86.4 171.1

Spain 63.9 53.8 Belgium 88.5 168.1

United Kingdom 46.4 36.7 Spain 78.5 148.5

Japan 29.4 25.9 Japan 59.2 127.5

France 27.5 22.7 Korea 42.1 82.8

Switzerland 27.1 19.5 United Kingdom 35.2 69.3

Belgium 23.4 19.3 Switzerland 29.4 54.4

South Korea 23.0 18.2 Algeria 22.4 47.7

Total Export 892 735 Total Export 1000 2115

Source: General Custom Office of Vietnam per com

West European

59%

East

European4%

Latin America

18%

Asia

12%

Ocean

Cont.2%

Other

5%

West

European

51%

East

European

7%

Latin

America13%

Asia

21%

Ocean

Cont.

1%

Africa

7%


21

Vietnam has important advantages in coffee exports. The natural conditions in the

Central Highlands of Vietnam are suitable for growing Robusta coffee (D’Haeze, 2004).

The production cost of coffee in Vietnam is relatively low. According to Chi et al

(2009), the cost for exported coffee (which includes both production cost and ex-farm

gate cost, defined as the cost from farm gate to port) in Vietnam is only 25 cent/lb,

while exporting costs for India and Indonesia where Robusta coffee is the important

crop are 34 cent/lb and 37 cent/lb, respectively.

Figure 2.7: Exporting cost for coffee in some countries (cent/lb)

Source: PI-IPSARD (2008) quoted in Chi et al (2009)

The advantage of coffee production in Vietnam reflects the Domestic Resource Cost

index (DRC)13

. Generally, DRC ranges from 0 to 1, with a smaller value implying a

competitive advantage because it means Vietnam utilizes less domestic resource to

produce one unit of coffee for export. A study by the CAP (2006) on Competition under

AFTA shows that the DRC of coffee in 2006 is only 0.37, smaller than rice (0.59) and

much lower than rubber (0.7) or tea (0.79). This suggests Vietnam has a comparative

advantage in coffee production (see Figure 2.8).

13 Formally, the DRC is defined as

aijp j*

j k 1

n

pib aij p j

b

j 1

k, where j=1…k are traded inputs, j=k+1…n are

domestic resources and/or non-traded inputs, p* is the shadow price of domestic resources and non-traded

inputs, pib is the border price of traded output calculated at the shadow exchange rate and pj

b is the border

price of the traded input at the shadow exchange rate (Sadoulet and deJanvry 1995)

0

10

20

30

40

50

60

Vietnam India Indonesia Brazil

cen

t/lb

Extra cost

Production cost


22

Figure 2.8: DRC of coffee and other export commodities in Vietnam

Source: CAP (2006)

Son et al. (2005) summarizes strengths, weakness, opportunities and threats (SWOT) in

the context of export assessment of coffee. The core strengths of coffee in Vietnam are

high yield and low production costs. However, low/inconsistent quality, undeveloped

post harvest process technology and the shortage of storage capacity are the main

problems (see Table 2.9).

Table 2.9: SWOT analysis of coffee

Strengths

Suitable natural conditions for coffee

Low production cost

High yield

Good experience in coffee cultivation

Having concentration area

Large export market share , specially Robusta

Development of private export

Weaknesses Dominated by small households

Low/inconsistent quality

no brand name/mostly export coffee bean

over expansion of coffee area

Lack of storage facilities, marketing services

Export through intermediaries

Underdevelopment of future market, transaction floors

Vietnam standards are inconsistent with international standards.

Overuse of fertilizer and pesticides

Mainly dry process application

Lack of risk management measurements

Opportunities Export market diversification

Recovery of export market

Development of wet processing technique

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Rice Coffee Rubber Tea


23

Government supports to develop brand name , trade promotion

Threats Competition from other crops

Competition from other exporters

Unstable price

Over expansion of Robusta

Inefficient plan for Arabica development

Drought

Water resource limitation

Source: Son D.K. et al (2005)

2.5. Coffee Households

2.5.1. Farm Size and Distribution

Coffee is the most important perennial crop in Vietnam with over 477 thousand

households producing in Vietnam, accounting for 3.29 percent of the total number of

farm households in Vietnam. The share of coffee households in Vietnam is much higher

than rubber households (0.73 percent) and pepper households (1.24 percent) and slightly

larger than cashew household’s (3.14 percent). The average of coffee area is 8799 m2,

much higher than average farm-size of tea households and pepper households, but

smaller than rubber and cashew (see Table 2.10).

Table 2.10: Number of perennial crop households and size in Vietnam

Tea Coffee Rubber Cashew Pepper

Number of household 380751 477235 106139 456141 179478

% farm household 2.62 3.29 0.73 3.14 1.24

Average area (m2/household) 2414 8799 16906 10274 2470

Source: GSO (2007)

The coffee sector in Vietnam is dominated by small households. According to GSO

(2007), over 60 percent of coffee households have less than 1 ha of coffee land and 10

percent of households have more than 2 ha of coffee land (see Figure 2.9). Small size

and land fragmentation are problems for the coffee sector in Vietnam. Small farms are

unable to invest in improved harvest and post-harvest technology and this has led to

variable coffee quality.


24

Figure 2.9: Coffee Household Structure by farm-size (%)

Source: GSO (2007)

Most coffee households are concentrated in the Central Highlands. According to

Agrocensus_2006, nearly 90 percent of coffee households are located in the five

provinces in Central Highlands (Gia Lai, Dak Lak, Dak Nong, Kon Tum and Lam

Dong), of which nearly 40 percent of total households are located in Dak Lak province;

24 percent in Lam Dong and 7.5 percent in North East South. Recently, Vietnam has

tried to expand coffee in some provinces in the North Mountains of Vietnam. However,

programs for developing Arabica in the North Mountains could not achieve successful

results. Thus, coffee area in the region is still very limited14

.

Figure 2.10: Distribution of coffee households across Vietnam

Source: GSO (2007)

14

The structure of coffee households in Vietnam by farm size in all provinces in Vietnam is presented in

Table B6 (Appendix B).

0

5

10

15

20

25

30

35

under 0.5 ha 0.5-1 ha 1-2 ha 2-3 ha 3-4 ha over 5 ha

10,463 3,364

427,316

36,054

0

50000

100000

150000

200000

250000

300000

350000

400000

450000

North Central Coast South Central Coast Central Highlands North East South


25

Figure 2.11 gives the distribution of coffee households by number of plots farmed.

Nearly 70 percent of household own one plot of coffee land. Small size and land

fragmentation are general characteristics of agricultural production in Vietnam. In 2008,

the average agricultural land per person in Vietnam was approximately 0.11 ha (GSO,

2008). Due to the high population growth, this number is projected to be only 0.08 ha

by the year 2020 (D’Haeze, 2004). However, land for coffee as well as other perennial

crops is spatially concentrated compared to that for annual crops such as rice.

Figure 2.11: Coffee household structure by number of plots (%)


2.5.2. Starting Year of Coffee Production

Coffee became a major crop in Vietnam in the 1980s following its introduction during

the French colonization period. As mentioned in Section 2.3, this expansion of the

coffee area was due to stimulation by policy reform, migration into the Central

Highlands and the development of a coffee market. The most rapid growth of coffee

area occurred in the mid 1990s when coffee prices were high. According to the Coffee

Farm Survey 2007, about 30 percent of coffee households in Dak Lak province started

producing in 1994-1995.

1 plot69%

2 plots17%

3 plots11%

4 plots3%


26

Figure 2.12: Percentage of new coffee farmers and farm-gate price

Source: Thang (2008) and GSO (2009)

The expansion of coffee area in Central Highlands goes hand in hand with migration

into that region. Thus, many coffee households are migrants from other regions. They

came to the Central Highlands and exploited bare land or forest for coffee cultivation or

bought coffee land from ethnic people. The migration to the Central Highlands has

changed the ethnic distribution of the population. Indigenous minorities such as the E-

De and the H’Mong, who had made up 48 percent of Dak Lak’s population in 1975,

now only account for 20 percent of the population (D’Haeze et al., 2005). According to

Agrocensus_2006, in the Central Highlands about 70 percent of coffee households were

Kinh people who are not originally local people (see Table 2.11). The rest are local

people such as E-De (9.8%), Gia-Rai (5.1%), Co Ho (4.9%). The involvement of a large

percentage of ethnic households in coffee production shows the importance of this

sector in rural development and social stability. Because a disproportionally large

number of ethnic minority people are in low-income groups, they are vulnerable to

coffee price shocks.

Table 2.11: Distribution of coffee household by groups

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

0.0

2.0

4.0

6.0

8.0

10.0

12.0

14.0

16.0

18.0

20.0

pri

ce

% o

f new

gro

wer

proportion of new coffee growers (%)

farm-gate price (USD/kg)


27

Groups Number of households Percent

Kinh people 263292 70.44

Minor ethnic people 110491 29.56

In which:

E de 36570 9.78

Gia-Rai 19052 5.1

Co Ho 18250 4.88

Nung 7451 1.99

Ba Na 6365 1.7

Tay 6196 1.66

Ma 3475 0.93

Other 139349 3.52

Total surveyed households 373783 100

Source: Author’s calculation based on Agrocensus_2006

2.5.3. Income Sources

Besides coffee, farmers in Central Highlands also cultivate other annual crops such as

rice, maize, cassava and sugarcane. Some people also grow other perennial crops such

as rubber, pepper, cashew and durian. However, the area of those crops is very limited.

Table 2.12: Average crop area of coffee households by district (m2)

Cu Mgar Krong Pak Eakar

Rice 104 2529.0 2118

Maize 60 279.6 0

Cassava 0 20.4 20

Sugarcane 0 0 1600

Coffee 19376 9038.8 7354

Rubber 200 0 0

Pepper 101 0 234

Cashew 0 0 3270

Durian 0 20.4 0

Souce: Thang (2008)

Beside crop production, coffee households also participate in other economic activities

with livestock production and waged labour as the other main sources. The income and

production diversification are important to ensure food security. According to the

Coffee Farm Survey 2007, the average revenues of coffee households from livestock in

Krong Pak and Cu Mgar were nearly $300 (8.1%) and $200 (3%), respectively. Waged

labour is also a good source of income and it is stable across surveyed districts.

However, analysis of income sources shows coffee is the most important source of


28

income and coffee households have limited opportunities to diversify their income. This

makes coffee households vulnerable when the price of coffee goes down.

Ability to diversify income of coffee farmers depends on different factors. Agergaard et

al (2009) indentified four factors which influence the possibility for livelihood

diversification of coffee farmers in Daklak province such as (i) the ethnic background of

the inhabitants. (ii) the specific period in which settlements become coupled to the

dynamics of the global value chain. This relates to the timing of when smallholders

started to benefit from their investments in coffee (iii) organisation of coffee marketing

activities, (iv) constraints of remote and undeveloped areas.

Table 2.13: Earning sources of coffee households in 2006 ($)

Source of revenue Cu Mgar Krong Pak Eakar

1.Crop 5643 3042 2680

from coffee 5565 2906 1918

2.Livestock 204.7 299.9 29.9

3. Aquaculture 0 5.1 303.8

4. Wage/salary 295.7 144.2 263.2

5. Pension 2.1 39.8 0

6.Other income 37.5 169.2 37.5

Total 6182.9 3700.4 3314.8


Note: earning sources from activities are measured by sale revenue.

2.5.4. Profitability of Coffee Production

The production cost of coffee varies across districts. This depends on favorable natural

conditions for coffee such as soil, water, slope and the weather. According to a Coffee

Farm Survey in 2007, the average annual cost of coffee production in Cu Mgar district

and Krong Pak district was about $980 per ha. This cost was lower at $920 per ha in the

Eakar district (Table 2.14).

Fertilizer is the largest cost component with about 45-50 percent of total cost. Next is

labour, with about 35-40 percent. Expenditure for electricity and fuels are also

important, accounting for 15-20 percent of total annual production cost.

Yield varied across districts. This affects revenue and profit for coffee farmers. On

average, profit per ha achieved by coffee farmers in Cu Mgar was $1775, Krong Pak

($1638) and Eakar ($1395). These profits reflect the relatively high price of coffee in


29

that year. A simple simulation of coffee revenue earned by coffee household at the 2002

price with an assumption of the same cost in 2006 are presented in Table 2.14. This

assumption may not hold because farmers respond by reducing input use during periods

of low prices.

Table 2.14: Coffee farm performance in Daklak province, Vietnam 2006 ($/ha)

Cu Mgar district Krong Pak district Eakar district

Value ($) % Value ($) % Value ($) %

Annual cost ($/ha) 978.1 100 977.5 100 920.5 100

Fertilizer 290.9 29.7 475.7 48.7 367.6 39.9

Manure 67.0 6.9 38.6 4.0 0.5 0.1

Micro fertilizer 26.0 2.7 5.7 0.6 60.6 6.6

Labour 363.7 37.2 345.7 35.4 368.8 40.1

Electricity/fuel 199.5 20.4 111.8 11.4 153.5 16.7

Irrigation fee 0 0 0 0 2.6 0.3

Others 30.9 3.2 33.3 3.4 24.7 2.7

Yield (kg of coffee

bean/ha)

2165 2037 1867

Price ($/kg coffee bean) 1.3 1.3 1.2

Revenue ($/ha) 2753.3 2615.6 2316.3

Profit ($/ha) 1775.2 1638.1 1395.7

Price in 2002 ($/ha) 0.31 0.31 0.31

Revenue 2002 ($/ha) 671.1 631.4 578.7

Simulated profit in 2002

($/ha)

-307 346.1 341.8


Production cost varies among regions and households but most values are between $900

and $1200 per ha. Figure 2.13 shows the distribution of coffee production cost during

2006.


30

Figure 2.13: Distribution of coffee production cost, 2006 ($ per ha)


2.5.5. Source of Water

The coffee households use different water sources for coffee, namely surface water

resources (reservoirs and irrigation perimeters) and groundwater sources (i.e. from,

hand-dug and drilled wells). According to the Coffee Farm Survey, over 60 percent of

farmers used water from wells as their main source, and in Cu Mgar this proportion

reached 96 percent. About 30 percent of the households principally used water from a

nearby reservoir and lakes (see Table 2.15). This result is quite similar to a study by

D’haeze (2004) in which he estimated that 21 percent of irrigation water in Dak Lak

was extracted from surface water stored in artificial ponds and water reservoirs, 29

percent comes from natural rivers, streams and lakes and 57 percent is extracted from

ground resources.

Table 2.15: Main source of water (%)

District Wells Lake, dam Streams Total

Cu Mgar 96 0 4 100

Krong Pak 51.02 28.57 20.41 100

Eakar 34 66 0 100

Total 60.4 31.54 8.05 100


0

0.0005

0.001

0.0015

0 500 700 900 1100 1300 1500 1700 1900 2100 2300 2500 2700 2900

Cost per ha (USD)

Den

sit

y


31

According to Chi and D’haeze (2005), coffee producers in Dak Lak mainly use two

irrigation methods: basin and overhead sprinkler irrigations. Sprinkler irrigation is the

most widespread method in coffee growing countries because this irrigation system can

operate efficiently even in mountainous areas with uneven topography. Sprinkler

irrigation can also apply a uniform amount of water over the tree canopy. Nevertheless,

this method requires expensive irrigation facilities, high risk of water losses especially

in windy conditions, and high-energy consumption because of high pumping pressure

required for sprinklers. Coffee producers also use a basin irrigation method. This

method has several advantages such as low initial investment cost, inconsiderable water

losses, lower energy costs and low evaporative losses. However, basin irrigation method

is labor intensive (operation costs and basin maintenance). Chi and D’haeze (2005)

found that 85 % of their households in surveyed sites used the basin irrigation method,

while only 15% used sprinklers. All the interviewed households of ethnic minority

origin used basin irrigation, while sprinkler systems were only observed in Kinh

households.

Water supply is an important factor affecting the yield and quality of coffee cherries.

According to farmers’ assessment, only 72 percent of households reported that they had

enough water for coffee. Those farmers estimated that if there was a sufficient supply of

water for coffee the yield could increase by about 20 percent compared to the current

level.

Table 2.16: Is yield limited by water (%)

District Enough Not enough

% increase with

enough water

Cu Mgar 64 36 30

Krong Pak 89.8 10.2 9.64

Eakar 48 52 20.85

Total 72.97 27.03 21.06


However, coffee production in the Central Highlands is facing water scarcity. There was

a reduction of water flows in all rivers in the Central Highlands in 2003 – down by 20

and 50 percent on 2002 levels. The drought conditions resulted in a water supply

shortage for 100,000 households in the Central Highlands (The World Bank, 2004).

Similarly, in 2004 approximately 70,000 hectares of coffee was damaged or lost due to

the water shortage. However, previous studies (D’Haeze et al., 2003, D’Haeze, 2004)

pointed out that the amount of water presently used by coffee farmers exceeds the crop


32

water requirement and therefore endangers water resources in the region. To develop

sustainable coffee production, apart from an irrigation program for constructing water

reservoirs, training for farmers is also necessary.

2.6. Conclusion

Coffee is an important crop in Vietnam’s agricultural sector. It is the second largest

export agricultural commodity in Vietnam after rice. The coffee sector accounts for

about 3.29 percent of total households in Vietnam. Most coffee households are located

in provinces in the Central Highlands of Vietnam such as Dak Lak, Dak Nong, Gia Lai,

Kon Tum, Lam Dong. The coffee production in Central Highlands accounts for over 90

percent of national output.

The first planting of coffee in Vietnam occurred in 1857 but only became a

commercially significant crop after the “Doi Moi” policy of Vietnam in 1986. The area

of coffee increased exponentially, from 50,000 ha in 1986 to a peak level of about

600,000 ha in 2000. Policy reform (Contract 100, Contract 10, Land Law in 1993, re-

settlement policy, trade liberalization policy) and the development of a high price in the

international market was a major contributing factor in the development of the coffee

sector in Vietnam.

The rapid expansion of the coffee area has made Vietnam an important coffee exporting

country. At present, Vietnam contributes over 40 percent of Robusta and about 13

percent of the total world coffee market. In general, Vietnam has a comparative

advantage in coffee production with high yields and low cost.

With more than 90 percent of coffee output in Vietnam exported, the coffee price in

Vietnam depends heavily on the international price. Low prices in 2002 led many

farmers to cut coffee and switch to other crops. During three seasons (2001/2002 to

2004/2005), the coffee area in Vietnam reduced over 100,000 ha.

In addition, despite achieving this rapid expansion, the coffee sector in Vietnam faces

significant economic challenges. First, small farm sizes dominate the coffee sector.

Second, a high percentage of coffee households are minority ethnic people with low

levels of education. Third, a high proportion of coffee households are poor and have

limited opportunities to diversify their income.


33

With such characteristics, coffee farmers are vulnerable to price fluctuations. In the

following chapters, several models analyze the supply response of coffee and identify

the optimal rules for coffee farmers in Vietnam. Chapters 4, 5, 6 apply the Fixed Form

Optimization approach to analyze coffee farmer’s decision at individual or household

level. Chapter 8 uses a “positive” approach to analyze the coffee supply response at an

aggregate level.

Chapter 3.Stochastic Optimal Investment Decision for Perennial Crops: A Literature Review

34

CHAPTER 3. STOCHASTIC OPTIMAL INVESTMENT

DECISION FOR PERENNIAL CROPS: A LITERATURE

REVIEW

3.1. Introduction

As established in Chapter 2, coffee is vitally important to Vietnam’s economy. The

increased supply of coffee since the early 1990s led to the price reduction in the early

2000s that significantly reduced producers’ incomes. Due to the price collapse, a large

number of households cut down coffee trees and switched to other crops. Many farmers

lost money on their investments in coffee gardens.

The expansion of the coffee area by farmers in Vietnam occurred rapidly without the

intervention of the Government. Managing a farm is complex and the choice of farm

strategy may be influenced by the farmer's knowledge of crop husbandry, machinery

availability, economic/commercial factors, political events, legal constraints, historical

trends, climate/weather, environmental issues, personal circumstances and any number

of practical considerations (Pannell, 1996). In addition, coffee is a perennial crop, thus

deciding when to plant or when to clear coffee trees is much more complicated than for

annual crops. The investment decisions of farmers (including the planting, replanting,

and cutting decision) are determined by factors such as: (i) resource availability, land,

capital and labour; (ii) the age of orchards; (iii) profit expectations; (iv) relative

profitability of substitute crops, such as rice and (v) risk aversion (Ruf and Burger,

2001).

The main objective of this study is to solve the stochastic optimal control problem of

coffee farmers by identifying the optimal removal and replanting price for coffee

farmers to maximise the expected income from land use. Thus, the problem for the

representative coffee producer considered in this study is one of achieving an optimal

harvest, including planting and removing trees, under stochastic conditions.

This chapter reviews the literature on optimal planting and clearing decisions. Prior to

detailing stochastic optimal control methods, the next section begins with some basic

theoretical models for optimal harvest of perennial crops.


35

3.2. Theoretical Models for Optimal Investment Decision

The fundamental questions for perennial crop and forestry economics are: when should

farmers harvest or clear fell and when should they replant? This issue was first

considered by forestry economists when they tred to determine the optimal harvesting

rotation. Faustmann (1849) presents the first robust solution to the optimal rotation

problem (Faustmann, 1995, Samuelson, 1976, Brazee, 2007). In this model, all

parameters including stumpage price, replanting cost, the discount rate and timber

volume function are deterministic and constant over time. The forest owners maximise

the net present value (NPV) of their land. These assumptions imply that the optimal

harvesting age is the same in every rotation. The basic Faustmann optimal NPV is given

as (Samuelson, 1976, Faustmann, 1995, Buongiono, 2001, Brazee, 2007):

Maximise 2

2

( ) ( ) ( ( ) ) ( ( ) ) ...

( ( ) )(1 ...)

( )

1

iT iT iT iT iT

iT iT iT

iT

iT

V T e G T C e e G T C e e G T C

e G T C e e

e G T C

e

(3.1)

where C is replanting cost, i is discount rate, T is harvest age and ( )G T is the value of

stumpage harvested at age T.

In the basic Faustmann model, the opportunity cost of land is the present value of future

rotations. The opportunity cost may be the income from non-forestry crops. Let S be

the maximum NPV of land from either non-forestry uses (W ) or land expectation value

(site value). Thus, the objective function (3.1) becomes:

( ) ( )iT iTMaximizeV T e G T C e S

(3.2)

To find the optimal harvest time, (3.2) is differentiated with respect to T and the

derivative is set equal to zero

( )( ) 0iT dG T

e iG T iSdT

(3.3)

or

( )( )

dG TiG T iS

dT

(3.4)


36

After the first rotation, if non-forestry crops are more profitable than forestry uses, S is

replaced by W in (3.4). Otherwise, income from forestry uses is higher, then (3.4) is

rearranged as

( ) ( ) ( ( ) )( )

1 1

it

iT iT

dG T e G T C i G T CiG T i

dT e e

(3.5)

The results in (3.3) to (3.5) are the “Faustmann rule”.

The assumptions of the Faustmann model characterize its weaknesses. The assumptions

of the basic Faustmann model omits many important factors that affect the optimal

rotation age and decision of farmers. Providing a more complete analysis for optimal

decision of forestry growers, many economists extend the Faustmann model in various

ways: by adding tax on forest value (Klemperer, 1976, Chang, 1982), including

silvicultural efforts of the forest (Samuelson, 1976) and subtracting harvest cost, road

building and maintenance costs at the time of harvest from the revenue (Heaps and

Neher, 1979).

The Faustmann rule has been applied not only in defining the optimal harvest age for a

forest, where the yield is determined at the end of the economic life of the tree, but also

in investigating the investment decision of agricultural crops (perennial crops) where

there are continuous flows of benefits. Jayasuriya et al (1981) used a dynamic profit

maximization model for analyzing the long-term investment decision of Sri Lankan

rubber smallholders. The rubber farmers face the decision problem of when to replant

existing trees. The study presents an analysis of planting decision and investigates the

relevance of conventional investment decision criteria for rubber smallholders. The

analysis starts from the formulation of NPV of the crop sequence:

0

( ) ( )

T

iT iTNPV R t e dt S T e

(3.6)

where ( )R t is net revenue in the year when the age of trees is t year; ( )S T is a salvage

value at the end of its life in year T ; i is the discount rate


37

The optimal condition for maximizing NPV of the earning stream is:

( ) '( ) ( ) ( )1

T

iT

iT

o

iR T S T R t e dt S T

e

(3.7)

When data is available in discrete form, this equation can be expressed as

1

( ) ( ) (1 ) ( ) ( )1 (1 )

St

Tt

iR T S T i R t S T

i

(3.8)

where i is the interest rate, S is change in salvage value in year T

The optimal replacement age occurs when the marginal revenue from these trees (i.e

income in current period plus the increase in salvage value as a result of keeping it for

another period) falls below the highest perpetual annuity from the replacement crop.

Farmers select the crop with highest NPV to replace the standing rubber tree and they

replace the current trees when marginal revenue from these trees fall below the level of

highest perpetual annuity from the replacement crop. By analyzing two groups: rubber

replanters and non-rubber replanters (people who do not replant rubber), the study

identifies the main reasons to explain why they continue tapping the existing trees

(mainly due to low current income from current trees), why they decide on replacement

(the common reason is old trees). However, this study does not investigate the

relationship between cutting decision and age of rubber trees. Furthermore, the study

does not investigate the price levels at which rubber smallholders should cut or replant.

Applying the Faustmann formula, Kearnev (1994) used a dynamic LP approach to

analyze the planting and replacement decision of farmers for pip fruit in New Zealand.

The objective of the model is to explore the optimal variety mix for an individual apple

orchard. The choice of variety mix within the orchard is an important strategic decision

because the trees take 10 years to reach the maturity and consumer’s preferences change

over time. The decision variables are t

jNP1 (wheret

jNP1 is area of new planting of

variety j at the beginning of year t in age class 1) and t

ijA (wheret

ijA is area of age group

i of variety j at the end of year t after removals have been deducted)

The objective function is given by:


38

Maximise Z = 1

1

(1 )

Tt

tt

Ri

in which t t t t

kj kjR A GM FC

where tR is net return from t

kjA andt

kjA is area planted for each age group k and variety

j, t

ijGM is gross margin for age class k for variety j in year t, tFC is fixed costs in year t,

i is the discount rate.

The constraints of the model include land limitation, a cash constraint and harvest

capacity constraint. The apple prices in the model were the average national price in the

first year. The prices, assumed to reduce over the following 10 years at different rates

for each variety and become stable for the remaining 10 years. The results from the

model show the maximised profit, the removal of each variety and area change from

1991 to 2000. However, the model is deterministic with the impact of age on the cutting

rule and possible input response of farmer in the short-run when ignoring price changes.

The determination of the optimal harvest age for a growing forest has received a great

deal of attention by economists. However, in the Faustmann model, the only economic

value of a forest is through its wood production. However, other values of the standing

trees ignore issues such as flood control, recreation, and other services. To incorporate

those benefits into the basic Faustmann model, Hartman (1976) develops a model to

analyse the optimal harvesting time when a standing forests has a additional non-timber

values (denoted by ( )F t ). The additional non-timber values are assumed to increase

with the age of trees.

Let ( )G t be still the stumpage value at age t. In the simple model of one cutting forest,

the objective is to maxinise the sum of the integral of discounted benefits ( )F t plus the

discounted value of timber at harvest time. Mathematically, the problem is to find t to

maximise (Hartman, 1976):

0

( ) ( ) ( )

t

it itV t e F t dt e G t

(3.9)

where i is the discount rate and t is the harvest age.


39

To find the optimal age T , one needs to solve the first order condition for a maximum

' '

'

'

( ) ( ) ( ) ( ) 0

( ) ( ) ( )

( ) ( ) ( ) ( )

itV t e F t G t iG t

F t G t iG t

G t G t i F t G t

(3.10)

and the second order condition is

'' ' ' '' '( ) ( ) ( ) ( ) ( ) ( ) ( ) 0it itV t ie F t G t iG t e F t G t iG t

(3.11)

After replacing (3.10) into (3.11), the second order condition simplifies to:

' '' '( ) ( ) ( )F t G t iG t

(3.12)

The first order condition for optimality shows that the marginal loss by delaying cutting

the trees one period is equal to the gain from postponing the harvest (it is the summation

of recreational value and timber value over one period). In the absence of recreational

value ( ( ( ) 0)F t , the landowner should harvest the forest if its growth is equal to the

discount rate. With recreational values, the landowner should harvest at a later age,

when the growth rate is less than the discount rate.

The conditions in (3.10)-(3.12), as mentioned earlier, are applied for the first harvest.

For indefinite sequence of harvests, the objective now is to maximise

2 3 2

0

0

( ) ( ) ... ( ) 1 ...

( ) ( )

1

T

it it it it it it

T

it ix

it

V t G t e e e e F t dt e e

G t e e F t dt

e

(3.13)

The first order condition for maximizing ( )V t is

' '

2

0

( ) ( ) ( ) ( ) (1 )

( ) ( ) (1 )

it it it it

T

it ix it it

V t ie G t e G t e F t e

G t e e F t dt ie e

(3.14)


40

ix

'

0

( )( ) 1 ( )

( ) 1 ( )(1 ) ( )

T

it it

e F t dtG t F t

iG t e G t e G t

(3.15)

The difference between (3.15) and (3.10) is the term in the brace. Because ix

0

( )

T

e F t dt

and ( )G t are positive, thus

ix

0

( )1

1 ( )(1 )

T

it it

e F t dt

e G t e is greater than 1, so

ix

0

( )1

1 ( )(1 )

T

it it

e F t dt

ie G t e

(which is referred to as the effective discount rate) is greater

than the normal discount rate i. This change makes the optimal harvest age of the

infinite time horizon harvest decision different from the one harvest horizon.

3.3. Faustmann Model with Risk

In the optimal Faustmann model, a standing tree will grow until it reaches maturity

unless cut down by the landowners. However, it is assumed that this is known with

certainty - ignoring risk factors that affect optimal harvesting. Reed (1984) investigates

risk of fire on the optimal rotation of a forest. Based on the basic Faustmann formula,

Reed investigated the optimal harvest age for maximizing the expected return of a forest

under the risk of fire.

According to Reed’s model, the optimal cutting age to maximise the long-run expected

yield is 15

' ( ( ) )( )

1 T

V T CV T

e

(3.16)

where C is replanting cost and is the probability of fire occurrence per unit of time

in a Poisson process. ( )V t is the stumpage value of a stand of trees at age t. This is the

15

To see how to get the optimal rules with risk of fire in detail, see Reed (1984)


41

same as Faustmann formula in (3.5) with the discount rate i replaced by the probability

of fire occurrence .

The optimal cutting age to maximise the expected discounted yield is

'

( )

( )( ( ) )( )

1 i T

i V T CV T

e

(3.17)

This is also as the same as Faustmann formula in (3.5) but the discount rate i is

replaced by ( )i . Thus, the effect on optimal rotation of fire risk is the same as that of

an increase in discount rate by an amount of average rate at which fire occurs. This

shortens the rotation length.

In general, theoretical models for optimal harvest decision by Faustmann (1995),

Hartman (1976) and Reed (1984) are useful to identify the optimal age for cutting the

standing trees. However, they do not address the price uncertainty problem. In this case,

there is no fixed output price as in the Faustmann model, and it varies stochastically. In

addition, those models have not developed the detailed analytical framework to analyse

the optimal cutting and replanting decision. When adding such factors into a problem of

the optimal cutting decision, the problem becomes much more complicated. However,

different mathematical methods may solve these problems. The next section will review

different methods for solving the stochastic optimal control problem.

3.4. Stochastic Optimal Control Methods

This section provides a literature review of different methods to solve the problem of

stochastic optimal control. The discussion summarizes different methods including

Dynamic programming, Real option approach, and other techniques for solving the

complex dynamic problem.

3.4.1. Dynamic Programming (DP)

Dynamic programming (DP) is a numerical and analytical method to solve dynamic

optimization problems. It is based upon the principle of optimality which states that an

optimal policy has the property that whatever the initial state and initial decision are, the

remaining decisions must constitute an optimal policy with regard to the state resulting


42

from the first decision (Bellman, 1957). This principle gives the recurrence relation

function or Bellman equation as (Dixit and Pyndyck, 1994, Kennedy, 1986):

1 1

1( ) ( , ) ( )

1tt t u t t t t t tF x Max x u F xi

(3.18)

where ( )t tF x is the outcome - the expected net present value, tu is a control variable,

tx is a state variable, ( , )t t tx u is intermediate profit flow, (1 1 )i is the discount

factor and i is the discount rate. The aim is to choose the sequence of controls tu

over time to maximise the expected net present value.

DP is a technique ideally suited for use in finding the optimal sequencing problems of

inputs and harvesting outputs in many types of agricultural products. Those problems

entail decisions which are sequential, risky and irreversible (Kennedy, 1986).

Furthermore, Kennedy (1986) also points out that DP is a technique particularly suited

for obtaining numerical solutions to problems that involve functions which are non-

linear, stochastic or models in which state and decision variables are constrained to a

finite range of values.

Dynamic programming has found wide use in pest management, water resources,

fisheries, and in the management of animal populations. However, DP can analyze crop

rotations, and find the optimal cutting time for forest trees.

Burt and Allison (1963) applied stochastic dynamic programming to analyze the

decision to leave land fallow or plant wheat with soil moisture as the state variable. The

model answered the question “when should the farmers fallow land?” The study

indicates that an optimal policy based on soil moisture at wheat planting time will give

an expected return per year of about 13 percent higher than a policy of continuous

wheat and 30 percent than the fallow and wheat.

Dynamic Programming applies in farm forestry analysis to identify the optimal cutting

time. As noted previously, the question of optimally deciding when to cut down a tree is

a major concern for forestry economics. Matheson (2007) applied DP for finding

optimal harvest length when replanting is addressed. This is different from the previous

standard model without replanting decision. Without replanting, farmers simply cut the

timber if the growth rate is equal to the discount rate. However, with the presence of

replanting and cutting costs, the tree-cutter has three options in each period: (i) leaving


43

the trees for later harvest (ii) cutting the trees and replant; (iii) cutting the trees and

leaving land idle.

The decision is based on the objective of present value maximization. The functional

equation is given by:

( ) max ( ) , (0) ,v k v h k k v RC k

(3.19)

In this equation, v k is the maximum value of the tree given the three alternative

decision options. is the discount factor { 1/ (1 )i }; h is the growth function of

the tree and represented by a third-degree polynomial form of time; k is the size of tree

and 1 ( )t tk h k . RC is the replanting cost and is assumed constant overtime. Price of

timber is exogenous and assumed to be 1.

This equation is not solved analytically. The author uses a numerical method to find the

optimal time for cutting. In this study, the author also simulated to find the optimal time

when the discount rate was changed.

Different factors determine the forestry yield such as variety, resource management and

more importantly climate. To analyze the decision of the forester with weather

uncertainty, Jia (2006) applied DP to find out the optimal forest rotation decision for

loblolly pine in North Carolina (USA) under climate fluctuations. Two factors

determine the growth of trees: genetics and climate in which the genetic effect is

deterministic, and climate fluctuation is stochastic. A quadratic function of tree age

expresses the growth of timber.

The climate effect is introduced by using a multiplicative factor L, and in the paper, Jia

investigated the optimal rotation in two possible cases: (i) L is assumed to follow a first-

order autoregressive process and (ii) L is a random walk model.

The growth function with both effects would be:

1( ) ( ) ( ) ( )R R

t t t tY n Y n L n Y n

(3.20)

where 1( )R

tY n is realized yield in year 1tn

, ( )R

tY n is realized yield in the previous

year nt, ( ) ( )t tL n Y n is the realized new growth in year tn , which is the product of


44

natural growth and the climatic fluctuation effect.

The revenue function from timber is:

( , ) ( , )t t t tR Y n L PY L n K

(3.21)

where ( , )t tY L n is saw timber yield (mbf/ac) obtained from a t year-old forest-stand, K

is the given value of bare-land.

The Bellman’s equation for a planning horizon of T years is:

1 1 1( , , , ) max ( , ), ( , )t t t t t t t t t tV Y L n X R Y L ER Y L

(3.22)

where is the discount factor, V is a reward function, and E is the expected value of

timber.

With the decision of cutting or keeping to maximise present value of forest over a

certain time horizon, the author used DP to solve the problem of finding the optimal

rotation age. The results of the study could identify the relationship between the climatic

effect and age of tree and the expected rotation ages in both climate simulations.

However, the model assumed a constant price so it did not identify at what price farmers

should cut and replant.

With biological features of forestry crops, DP has been applied to analyze tree

production cycle/cutting time under the different impacts and conditions such as timber

stock and resource management (Dixon and Howitt, 1980), stochastic price (Penttinen,

2006, Chladná, 2007), change in environment (Chladná, 2007), and interest rate

variability (Alvarez and Koskela, 2004).

The application of DP is very popular to solve optimization problems for farmers’

planning horizon in number of years. However, the increase in number of state and

decision variables brings a computation burden when solving the DP problem. This

issue has been termed “the curse of dimensionality” (Bellman and Dreyfus, 1962),

often causing it to be dismissed. With the support of modern computation techniques,

the capacity of current computers limits the maximum number of state variables to

three. However, in many cases it is possible to use other approaches to solve

approximately a DP problem with many state variables (Kennedy, 1981). One of the

important approaches refers to Reinforcement Learning (RL) or Neuro- Dynamic


45

Programming (NDP) or Approximate Dynamic Programming (ADP). In this approach,

they use basis functions or neural networks to retain an estimate of the value function at

each iteration within a dynamic programming algorithm. ADP primarily solves “the

curse of dimensionality”. ADP avoids the exponential increase of computations by

using parametric approximate representations of the cost-to-go function. Compared to

traditional DP, which performs exhaustive sampling of the entire state space in solving

the stage-wise optimization, these approaches sample only a small, crucial fraction of

state space and thus require less computation. The detailed descriptions of RL, NDP or

ADP is presented in Bertsekas and Tsitsiklis (1996), Sutton and Barto (1998), and

Powell (2007).

3.4.2. Real Option Approach

The conventional approach in selecting investment projects where there is uncertainty

about future market conditions is based on the comparison of expected net present

value. However, this basic principle is often erroneous because the expected net present

value is built on faulty assumptions. Either the investment is reversible (it can somehow

be undone and the expenditure recovered should market conditions turn out to be worse

than anticipated), or if the investment is irreversible, it is a now or never proposition: if

the firm does not make the investment now, it will not be available in the future (Dixit

and Pyndyck, 1994).

In most cases, one can delay investment. Thus to analyze the investment decision one

needs to develop a better framework to address the issues of irreversibility, uncertainty

and time. A firm that has an opportunity to invest is holding something like a financial

option. The development of the option approach brings a richer framework for

investment analysis. There are a number of detailed introductions to the options

approach to investment (Dixit and Pindyck, 1994, Smith, 2004, Gilbert, 2004). An

option exists when a decision maker has the right, but not the obligation, to perform an

act. For example, financial options, the mostly common option in economics, give the

owners the right, but not an obligation, to buy or sell financial assets at a predetermined

price before a particular event. According to Gilbert (2004) and Mauboussin (1999), the

real-options approach applies financial options theory to real investments, such as

manufacturing plants, product line extensions, and research and development.

Analogously, companies that make strategic investments have the right, but not the

obligation, to exploit these opportunities in the future.


46

Real options refer to the fact that firms have similar rights with regard to real (non-

financial) assets. Options add value as they provide opportunities to take advantage of

an uncertain situation as the uncertainty resolves itself over time. The combination of

two things need to be in place for a real option to exist: there must be uncertainty in

terms of future project cash flows and management must have the flexibility to respond

to this uncertainty as it evolves.

Two well-known approaches for valuing an investment using real options theory are

dynamic programming (DP) and contingent claims (CC). DP, presented in the previous

section, is an older approach developed by Bellman and others in the 1950s and used

extensively in management science. The Contingent Claims approach assumes the

existence of a sufficiently rich set of markets in risky assets so that the stochastic

component of the risky project under consideration can be exactly replicated (Insley and

Wirjanto, 2010).

Both CC and DP get a mention in the natural resources literature, especially in forestry

economic modeling to determine the optimal harvesting decision. In the forestry

economics literature, however, the DP approach has generally dominated (Insley and

Wirjanto, 2010). There are numerous studies on forestry investment analysis by

applying the real option approach and using DP. Most recent studies have focused on

the optimal harvesting time of standing trees under price uncertainty. Thomson (1992)

compares the optimal rotation ages of the Faustmann model with fixed timber prices

with a binomial option-pricing model when prices follow a diffusion process. The study

shows that the stand NPV from diffusion model is generally higher than Faustmann

NPV and the rotation lengths are longer except at high prices where they are the same as

the Faustmann rotation. Most studies of forestry investment analysis using option

models have tried to expand the conventional approach by looking at more complicated

variation process of timber price. Gjolberg and Guttormsen (2002) also investigate the

impact of price variability on cutting decisions of forestry owners by looking at real

option valuation of forest when prices are mean reverting. They indicate that the mean

reverting price may significantly increase the option value in the forest investment as

compared to the Faustmann rules. Insley (2002) applies DP and the real option approach

to determine the value of the option to harvest standing trees in Canada and the optimal

cutting time when lumber price is assumed to follow a known stochastic process (mean

reversion and geometric Brownian motion). This study found that the mean reverting

process had a significant impact on the optimal cutting decision and on the value of the


47

forestry investment rather than geometric Brownian motion process. In addition, for

prices below the mean, a process of mean reversion has an option value higher than that

of geometric Brownian motion.

Kaakyire and Nanang (2004) also compare the forestry investment using the static

Faustmann model and the real options approach but they use the binomial option-

pricing model, where timber values are assumed to follow a multiplicative binomial

process. By looking at four different management options (reforestation delay,

expansion of the wood processing plant, abandon of the processing plant because of low

timber prices and multiple options), the model results show that option analysis

supported the reforestation investment although the Faustmann model rejected it. All

options show that the reforestation investment was highly valuable for the owner. Insley

and Rollins (2005) extend the model of Insley (2002) to a multi-rotational framework

using a linear complementary formulation to estimate the value of a representative stand

in Ontario’s boreal forest (Canada)16

. The multi-rotation model can represent the “path-

dependent option”. For the multi-rotational optimal harvest problem, the value of a

stand today depends on the quantity of lumber, and thus depends on when harvesting of

the stand took place. The important improvement of the linear complementary

formulation is that it can assure that the solution will converge to a correct answer and

the accuracy can be checked easily.

Most recent studies using a real options approach and DP technique just focus on

identifying the optimal harvesting decision for the stand of trees under different random

price processes. They are rarely concerned about the price level at which cutting and

replanting should occur. Furthermore, in many projects, the investment sequence covers

many stages and control variables. Sometimes a firm in a sequential investment using

DP cannot compute the identification of entry and exit points because of “the curse of

dimensionality”. As mentioned earlier, however, in many cases, researcher can use

Approximate Dynamic Programming approach for solving the problem of the curse of

dimensionality (Powell, 2007).

Although application of real options theory to study “entry and exit” decision of farmers

is limited, it is more popular in financial and other sectors (Dixit and Pindyck, 1994).

The expected output of entry and exit model using option theory is very similar to the

coffee model in this study when identifying the optimal cutting (actually it is the exit

16

linear complementary formulation is described detail in Insley and Rollins (2005)


48

point) and replanting (entry point) prices. In the entry and exist model, the solution is a

pair of trigger prices for entry and exits. Mathematically, the model can be described as

follows (Dixit, 1989, Dixit and Pindyck, 1994):

Let P be market price, determined exogenously. P follows the geometric Brownian

motion as:

ordP

dt dz dP Pdt dtP

(3.23)

Where dz is the increment of a standard Wiener process, uncorrelated across time and

2( ) 0, ( )E dz E dz dt . is trend growth rate of market price P and where

is the firm’s discount rate. is a random number.

Let 0 ( )V P be the expected net present value of an idle investment. Similarly 1( )V P

defines for the active state; is the variable cost of a unit; k is the cost of investment

per unit of output; l is the cost of investment suspension per unit of output; HP is the

market price level at which investment occurs and LP is the market price level at which

abandonment occurs.

The value of investment is ( , )V P t . By a second order Taylor series, dV can be

approximated as

2 2 22 2

2 2

1 1( ) ( )

2 2

V V V V VdV dP dt dP dPdt dt

P t P P t t

(3.24)

In the limit, ,dP dt go to zero but 2 2 2( )dP P dt . So (3.24) becomes

22 2

2

1

2

V V VdV dP dt P dt

P t P

(3.25)

Replacing dP Pdt dt , it yields:

22 2

2

1

2

V V V VdV P P dt P dt

P t P P

(3.26)

Since this is an infinite horizon problem, the derivative V

tcan be deleted.


49

' '' 2 2 '1( ) ( ) ( )

2dV V P P V P P dt V P P dt

(3.27)

Taking the expected value of both sides, because ( ) 0E dt , so we have

' '' 2 21( ) ( ) ( )

2E dV V P P V P P dt

(3.28)

In asset equilibrium conditions, the expected capital gain of an idle project ( 0 ( )dV P ) is

equal to normal return ( 0 ( )V P dt ), so

' '' 2 2

0 0 0

1( ) ( ) ( )

2V P P V P P dt V P dt

(3.29)

' '' 2 2

0 0 0

1( ) ( ) ( ) 0

2V P P V P P V P

(3.30)

Similarly, one can calculate the return on assets of an active project. The only difference

is that there is dividend added to the expected capital gain.

' '' 2 2

1 1 1

1( ) ( ) ( ) 0

2V P P V P P V P P

(3.31)

The general solutions for (3.30) and (3.31) are easy to obtain. The solution (3.30) can

write as:

0 0 0( )V P A P B P

(3.32)

and for (3.31) as

1 1 1( ) ( )P

V P A P B P

(3.33)

where 0 0 1 1, , ,A B A B are constants to be determined and , is formulated as:

12 2 2 2 2

2

2 (( 2 ) 8 )0

2 and

(3.34)


50

12 2 2 2 2

2

2 (( 2 ) 8 )1

2

(3.35)

For the idle project, the value of an investment should go to zero as price P goes zero.

Because 0, 1 so 0 0 0( )V P A P B P goes to zero when P goes to infinity only

if 0 0A , so the function form value of the idle project now become:

0( )V P BP

(3.36)

Similarly, as price goes to infinity, option value of abandonment goes to zero. Because

0, 1 so 1( )V P goes to zero when P goes to infinity only if 0B . Thus,

1 1( ) ( )LPV P A P

(3.37)

Because HP is the price that triggers entry. The firm has to pay K to get 1( )V P . Thus,

HP must satisfy the value matching condition and the higher contact or smooth pasting

condition:

0 1( ) ( )H HV P V P k

(3.38)

' '

0 1( ) ( )H HV P V P

(3.39)

Similarly, LP must satisfy:

0 1( ) ( )L LV P V P l

(3.40)

' '

0 1( ) ( )L LV P V P

(3.41)

Replacing 0V and 1V in (3.36) - (3.37) into (3.38)-(3.41), we get the system of 4

equations:

LL L

PAP BP l

(3.42)

HH H

PAP BP k

(3.43)


51

1 11L LAP B P

(3.44)

1 11H HA P BP

(3.45)

The parameters 2, , are estimated from empirical data. Then, , can be easily

calculated from (3.34) - (3.35). Finally, , , ,H LA B P P can be obtained numerically. The

determination of ,H LP P is illustrated in Figure 3.1.

Figure 3.1: The determination of PH and PL

Note: 1 0( ) ( ) ( )G P V P V P

So far, applying real option theory using entry and exit decision model for analyzing the

cutting and replanting decision of perennial crops is very limited. To our knowledge,

only one study implemented by Luong and Loren (2006) analyses the optimal decision

for coffee farmers. The objective of this study is similar to one of objectives in the

present thesis. In the study, Luong and Loren used a real option model to examine

Vietnamese coffee farmers’ investment decisions. Starting with the role of fixed assets

in agricultural production, the authors point out that the coffee production investment

and disinvestment decision depends on the difference between the acquisitions and

salvage price. This approach permits the authors to build a model of investment under

uncertainty and captures the response in investment decision. Luong and Loren (2006)

applied the same entry-exit model as described above to identify the entry/exit points

for different groups of coffee farmers in Vietnam. There are three steps in their model:


52

Step 1: Determine the value of an idle project. At this stage, the value of the investment

is equal to the value of the option to invest.

Step 2: Specify the value of an active project, the value of the investment now

comprises both the present value of the net revenue generated by the project and the

value of the option to abandon the project.

Step 3: Determine the entry and exit points. At the investment entry/exit points, the

investor must be indifferent between being “idle” or “active”. This may result in two

equilibrium conditions: (i) the value of an idle project is equal to the value of an active

project, and (ii) the rate of change of an idle project’s value is equal to the rate of

change of an active project’s value. So, equating the values of the idle and active project

as well as their derivatives produces a system of four equations ((3.42)- (3.45)).

By solving the system of equations simultaneously, the model finds the entry and exit

points. The entry-exit model found that a small farmer would enter coffee production

when the farm-gate price was above 47.2 cents/lb and exit the business if price dropped

below 14.2 cents/lb. When price fluctuates between 14.2 and 47.2 cents/lb, no entry or

exit would occur. The exit and entry points were calculated for three groups of farmers

by production cost. The low cost/more efficient producers will decide to plant/or cut

coffee at lower prices, while the less productive producers have to wait for better prices.

The efficient farmers (with average yield of 3 tonnes per hectare) enter coffee

production at price level of 38.8 cents/lb and exit at a price of 10.2 cents/lb. Meanwhile

the entry and exit prices for average cost farmers (average yield of 2.08 tonnes per

hectare) are 47.2cents/lb and 14.2 cents/lb, respectively. The low advantage producers

who achieved only 70 percent of the average yield of 2.08 tonnes per decide price levels

for entry and exit at 58.4 cents/lb and 20 cents/lb.

The real option model by Luong and Loren (2006) identifies the optimal entry and exit

points for coffee farmers. However, coffee is a multi-year crop in which yield and

production cost vary by age of the trees. Thus, the age of the trees may influence the

cutting decision. Other factors such as a cash constraint might influence coffee owners’

cutting and planting decisions. In addition, real options models assume that price

follows a continuous time stochastic process while in many cases the problem is a

discrete time process.


53

3.4.3. Other Techniques of Solving the Complex Dynamic Stochastic Models

These include a range of techniques developed to solve complex dynamic stochastic

optimization models not solved by standard DP because of “the curse of

dimensionality”. Two primary techniques popularly used for solving the problem of

“the curse of dimensionality” are: Neurodynamic Programming –NDP (or

Reinforcement Learning -RL or Approximate Dynamic Programming-ADP (Bertsekas

and Tsitsiklis, 1996, Sutton and Barto, 1998, Powell, 2007)) and Simulation-based

optimization (Carson, 1997, Azadivar, 1999, Ólafsson and Kim, 2002).

As mentioned earlier, NDP/RL/ADP is primarily viewed as a way to solve optimal

problems using the traditional DP because of “the curse of dimensionality”. In the

NDP/RL/ADP they use basis functions or neural networks to retain an estimate of the

value function at each iteration within a dynamic programming algorithm. The potential

benefit of NDP/RL/ADP can be summarized as follows (Powell, 2007). First, in

general they do not require an explicit model of the system that is to be controlled. The

controller can learn to control ‘on the fly’. Second, they may avoid the ‘curse of

dimensionality’ by providing approximate solutions. Third, they may not require an

explicitly defined system performance measure, which is usually a function of the

system states and the control actions in the classical optimal control theory. Some

examples of NDP/RL/ADP are Bertsekas and Tsitsiklis (1996) Roy et al. (1997),

Schutze and Schmitz (2007), Castelletti (2007), Powell (2007).

Simulation optimization provides a structured approach to determine optimal input

parameter values, where optimal is measured by a function of output variables (steady

state or transient) associated with a simulation model (Swisher and Hyden, 1998).

Simulation optimization can be seen as a process of finding the best values of some

decision variables for a system where the performance is evaluated based on the output

of a simulation model of this system (Ólafsson and Kim, 2002). Thus, the techniques of

simulation optimization vary greatly depending on the exact problem setting. A survey

of techniques for simulation optimization are described in Andradottir (1998), Swisher

and Hyden (1998) Azadivar (1999) and Ólafsson and Kim (2002). Some recent

examples of application of simulation optimization are L'Ecuyer et al (1994), Marbach

and Tsitsiklis (2001), Konda and Tsitsiklis (2003) and Barton (2009).


54

Fixed form or policy space optimization is a special case of these techniques that

obtains near-optimal feedback policies for complex ecosystem problems. It is a

multidimensional extension of control-space optimization. In control-space evaluation,

the analyst is largely limited to two control dimensions due to the curse of

dimensionality17

. However, when a objective function is defined with many control

variables, it is still possible to find the optimization by adding additional parameters

which make this function more flexible and general (Walters and Hilborn, 1978).

According to Walters and Hilborn (1978), there are two basic steps in the development

of fixed form optimization. The first is to find the algebraic form of the control function.

Commonly, one can intuitively guess the form, and in systems with a few state variables

and controls, one can simply make the control a polynomial function of the state

variables.

The second step in fixed-form optimization is to find the optimal values of the control

parameters (Walters and Hilborn, 1978). There are two alternative approaches to this

problem. The most elaborate is to use one of the many general gradient search

algorithms developed for nonlinear optimization. However, each evaluation of a set of

parameters involves a large number of numerical simulations. A second approach is

much simpler: by testing a large set of randomly chosen values for the control

parameters. Such random searching methods can work as well as gradient search

methods for problems that involve discontinuous response surfaces, or ones with several

peaks. In this thesis, to find the parameters of the optimal rule for cutting and replanting

of coffee trees in Vietnam, the GRID method is used. The GRID method is presented in

Section 4.3.7.

Peterman (1977) applied the fixed form method for hazard index function (H) of

budworm. He determined the optimal threshold value of H at which spraying should

happen. Generally, spraying and tree harvesting are the two primary management

options present for the budworm-forest system in eastern Canada. The paper

investigated the "rules" for these options: the age above which trees were harvested and

the ‘threat state’ above which insecticide should be applied. ‘Threat state’ measured by

the hazard index was dependent upon egg density and amount of defoliation of both old

and new foliage. A simple fixed-form optimization for spraying was defined as follows

(see more in Peterman 1977; Walters and Hilborn 1978):

17

See more about control space optimization in Walters and Hilborn (1978)


55

1 2( ) ( )H defoliation eggs Spray if H>1

Then, by seeking the values for 1 and 2 , the model can maximise the objective

function. This form can extend by including the product of defoliation and eggs to

account for potential interaction between these variables.

When searching for the algebraic form of the control function, in a system with few

state variables, authors can make the control a polynomial function of state variables.

Walters (1975) explored the optimal harvest strategies for salmon in relation to

environmental variability and uncertainty about production parameters. To meet the

objective, Walters applied the fixed-form for exploitation rate as a function of total

population (N) as follows:

3

4

2

321 NNNrateonExploitati

(3.46)

After the different steps, Walter found the αi to give the best overall return and the

relationship between harvest rate and population is the optimal control law.

In a further study of budworm management, Sonntag and Hilborn (1978)18

used fixed

form optimization for spruce budworm to decide whether farmers should spray or cut

the trees. The fixed form is given by:

Using this form of control law, they applied a random searching algorithm to optimize

the objective function. The process used by Walters (1975) and Sonntag and Hilborn

(1978) could more accurately be described as solving DP by approximation. They used

“approximate” fixed forms to identifying the relations between variables in their

models, from which they can simply solve the problem.

3.5. Conclusion

In conclusion, there have been many studies that analyse the stochastic optimal control

problems faced by farmers using three main methods: Dynamic Programming, Real

18

Quoted in Walters, J.C. and Hilborn, R. (1978)


56

Options Approach and other techniques for solving the complex DP (including the

Fixed form optimization). The model in this study aims to identify the optimal cutting

and replanting rules for coffee farmers. The problem for optimal cutting and replanting

decision of coffee in Vietnam is a stochastic dynamic problem. Coffee is a perennial

crop, thus the decision made in any period will affect the state of following periods and

in turn influence the total income from land. The model includes only two state

variables (age of coffee tree and price) but a number of different control variables.

There are five control variables in the coffee model reflecting the decision of farmers at

each stage. They are: (i) keep coffee if land is occupied by coffee, (ii) cutting standing

trees for other crop (maize) if land is occupied by coffee, (iii) cutting standing trees and

replanting new trees, (iv) keeping other crop (maize) if land is not occupied by coffee,

and (v) going back to coffee if land is occupied by maize. In addition, the model is

stochastic dynamic because the price of coffee is uncertain.

With such characteristics, it is not possible to apply the standard DP for solving the

coffee model because of the problem of the “Curse of dimensionality”.To solve the

coffee optimal rule problem, the study applies the fixed form approach. By assuming

the fixed functional forms for cutting price and replanting price, the model can be

solved with DP by approximation. The next chapter will describe the model structure,

content and its results in detail.

Chapter 4.Optimal Replanting and Cutting Rules for Coffee Farmers in Vietnam: Fixed Yield Model

57

CHAPTER 4. OPTIMAL REPLANTING AND CUTTING

RULES FOR COFFEE FARMERS IN VIETNAM: FIXED

YIELD MODEL

4.1. Introduction

The main objective of the study is to solve the stochastic decision problem for coffee

farmers in Vietnam. More specifically, the study builds a main model (and some

extended models) to identify the optimal cutting and replanting prices for coffee farmers

to achieve the maximum net present value that farmers can earn from their land. Given

that coffee is a multi-year crop, the application of DP seems an appropriate method to

solve the dynamic optimization problem of coffee farmers. However, this study is not

using stochastic DP because of "the curse of dimensionality”. The models in this study

use a horizon of 50 years for farmers and they have four decision options in each period:

(i) keep growing coffee, (ii) cut and grow maize, (iii) cut and replant coffee; and (iv)

switch to coffee if they are growing maize. In addition, models in this study also cover

different groups of farmers categorized by the age of their coffee trees. Moreover, the

coffee price in the model fluctuates stochastically. Thus, the number of decision and

state variables becomes very large so one cannot use DP to solve them. In addition, the

objective of the models in this study is a little different from traditional DP. The

objective of the present model is to find the maximum net present value for coffee

farmers earned from their land by identifying the optimal rules of cutting and replanting.

There are only two previous studies that have examined the optimal ‘trigger’ prices at

which farmers should change their coffee plantings in Vietnam (Luong and Loren,

2006; Oxfam, 2002). These studies are limited to presenting single period cost-benefit

analysis and do not investigate the relationship between tree age and farmer’s decisions.

To identify the optimal cutting and replanting price for coffee farmers in Vietnam under

price uncertainty, this study applies the fixed-form optimization approach (Walters and

Hilborn, 1978). The fixed form approach is applied to a specific functional form of the

cutting and replanting price. Farmers will cut coffee trees to switch to other crops if

output price is very low. However, the coffee yield and production cost of coffee

normally varies by the age of coffee trees. Thus, replanting and cutting prices are related


58

(via an appropriate polynomial functional form) to tree age. With the assumption of

fixed forms for cutting and replanting price, identifying the solutions for the optimal

rules is much simpler as compared to conventional DP.

This chapter starts with the Fixed Yield Optimal Model (FY model). The core objective

of the FY model is to identify the optimal cutting and replanting rules for coffee farmers

with the assumption of a fixed coffee yield function. In the FY model, yield of coffee

varies via the age of the coffee tree but otherwise cannot be altered through

management. The model is extended in the following chapters by adding a cash

constraint and the possibility of a short-run yield response.

Before moving to the detailed FY model, it is useful to understand the coffee farmers

and their production system in Vietnam. The coffee farm system in Vietnam will be

described in Section 4.2. The FY model structure is described in Section 4.3. Section

4.4 presents the results of the model. Section 4.5 gives some conclusions.

4.2. Coffee Farm System in Dak Lak

Prior to discussing the practical optimization model, this section describes the farm

system of coffee households based on a Coffee Farm Survey in Dak Lak province in

2007. Dak Lak is located in the Central Highlands of Vietnam. With favorable climate

and land, Dak Lak (including Dak Nong19

) is the principal coffee producing area in

Vietnam, accounting for about 50 percent of national output. In the 1990s, the area for

coffee in the province increased rapidly with an annual area growth rate of 14.1 percent.

In 2000, the coffee area in Dak Lak reached the peak level of 260,000 ha, accounting

for approximately 60 percent of cultivated land and 86 percent of the area of multi-year

industrial crops in the province.

19

Dak Lak province was divided into two provinces in 2003: Dak Lak and Dak Nong. The map of Dak

Lak location is presented in Appendix A.


59

Figure 4.1: Development of coffee area in Dak Lak, 1986-2008

Source: MARD per com.

After 2000, due to the price reduction, the coffee area in Dak Lak fell sharply. Since

2004, when prices started to increase, most cutters replanted again. According to a

report of Department of Agriculture and Rural Development of Dak Lak province, to

the end of 2007 most of the farmers who had cut coffee grew coffee again (DARDD,

2008).

As presented in Chapter 2, coffee households in Dak Lak are highly specialized, with

the major land use being for coffee cultivation. Beside coffee, households utilize flat

land to grow annual crops such as rice for both home consumption and cash. Maize,

rubber, cashew and sugarcane are also the main alternative crops cultivated by coffee

farms (see Table 2.12).

The farm size varies highly among households and districts. In three surveyed districts

in Coffee Farm Survey, coffee households in Cu Mgar have the largest land area with an

average of over 1.9 ha while farmers in Krong Pak and Eakar have a smaller scale with

the average of 0.9 ha and 0.73 ha, respectively.

0

50000

100000

150000

200000

250000

300000

1986 1989 1992 1995 1998 2001 2004 2007

ha


60

Table 4.1: Percentage of household with other activities excluding cropping (%)

District Cu Mgar Krong Pak Eakar

Chicken 20 8.2 8

Pig 20 49.0 4

Cattle/buffalo 14 16.3 0

Other animal 0 2.0 0

Aquaculture 0 4.1 58

Wage 10 26.5 50

Other activities 2 30.6 4

Source: Thang (2008).

When the price of coffee is depressed, most farmers had to reduce their inputs and

labour cost to save money. Some farmers had to cut coffee and change to other crops,

with maize as the main substitute crops for farmers in Dak Lak. According to the Coffee

Farm Survey 2007, over 6 percent of coffee farmers in Dak Lak had to cut trees early

because of the price fall (see Table 4.2). The majority of cut trees were in low yield

areas and where farmers were relatively poor. These farmers replanted coffee land to

other crops. The largest proportion cut coffee to grow maize (29.2 percent) and paddy

(25 percent) (see Table 4.3).

Table 4.2: Percentage of households reducing coffee area

District Yes No

Cu Mgar 0 100

Krong Pak 4.0 95.9

Eakar 14 86

Total 6.0 93.9


Table 4.3: Percentage of farmer switched to other crops

Percent Cumulative

Paddy 25.0 25

Maize 29.2 54.2

Durian 2.1 56.3

Sugarcane 6.3 62.5

Cassava 2.1 64.6

Pepper 14.6 79.2

Bean 2.1 81.3

Cashew 18.8 100.0


A large proportion of farmers (29.2%) switched to maize due to the price fall. Hence, to

simplify the model it is assumed that coffee is always replaced by maize.

.


61

4.3. Model Structure

This section describes the structure of the Fixed Yield model (FY model) and introduces

the procedure to get the solution. The FY model includes a profit function, a coffee

yield function, a production cost function and a revenue function. The objective of the

model is to identify the optimal cutting price and replanting price for coffee farmers to

maximise expected net present value (ENPV).

The model is similar to the option model by Luong and Loren (2006) because both aim

to identify the cutting and replanting prices. However, their model does not cover the

impact of tree age on the cutting decision of the farmers. Nor does it capture the

presence of a competitive crop, while the FY model investigates the farmer’s decision

when the profit of the replacement crop (maize) changes.

4.3.1. Objective Function

The FY model aims at identifying the optimal cutting and replanting decisions that

maximise the ENPV from “land use choice” per one unit of land (1 hectare) over the

entire planning horizon. The representative farmer can either produce coffee on the land

or switch the entire area to maize.

The ENPV depends on the current tree age, coffee price and production costs. Thus, the

ENPV of a block of land with coffee trees aged a at year 1, evaluated over a 50 years

planning horizon given N possible random price sequences is given by:

, ,

1 1

1( )

N Te t T

a t r a

r t

ENPV V TN

(4.1)

where aENPV is expected NPV given coffee trees at starting age a , for the next 50

years given N possible random price sequences; , .

e

t r a is profit per ha in year t for price

trajectory r and starting age a ; and is the discount factor. V T denotes the terminal

value of the coffee garden. However, given the model is defined over a long period of

50 years, at the end of the period the terminal value {TV T } will be insignificant and

hence is set to zero in the model; and r identifies the replicate number for one age

group. In this model, one hundred replicates are employed for each starting age group.


62

For each replicate, a separate random price trajectory is simulated. The method used to

generate price simulations over 50 years are presented in Section 4.3.6

In the FY model, it is assumed that the farmer controls one hectare of land and the

planning decision applies to the whole area i.e. they cannot make decisions on a fraction

of the land area. This means farmers can produce either coffee or maize in a given year

but not both. The model focuses on finding the price at which farmers switch to grow

maize and investigating the age dependent cutting price. It does not allow for the case

where farmers cut the coffee down and leave the land bare, even though this is a

technically feasible option for farmers.

Equation (4.1) assumes a particular age of the coffee tree at the start of the simulation.

Although it would be possible to solve for the optimal cutting and planting assuming an

arbitrary initial age (i.e. 1 year old), the impact of discounting would mean that the rule

developed for trees at the end of the cycle would be relatively imprecise, as changes in

the rule would have relatively little impact on ENPV. Therefore, it is important to solve

for the optimal rule using a simulation that has all ages of trees represented in the initial

period. Thus, it is necessary to develop a model that covers all different starting ages,

such that the optimal cutting/replanting rule is the one that maximises the average

expected NPV across all starting ages. Thus, the functional form and parameters of the

optimal rule will be independent of any assumption about the starting age of tree used in

the solution algorithm. In all optimal models in this study, the life cycle of coffee trees

is assumed to be 22 years. Hence, the final objective function of the FY model and

following optimal models is defined as the ENPV, averaged across all 22 initial ages:

22

1

1

22a

a

ENPV ENPV (4.2)

Thus, the ENPV in the FY model is average ENPV of all ENPV attained from 2200

random price trajectories (100 replications for each of 22 initial aged groups), over 50

years for each trajectory. This is used as the criterion for assessing the optimum when

evaluating parameters in the decision rule function described in Section 4.3.3. From

now when the maximum or optimal ENPV is mentioned in optimal models, it means the

ENPV as given in (4.2)


63

4.3.2. Profit Function

To develop the profit function from land use choice, whether coffee trees exist on the

land or not is denoted by a binary variable. According to the model’s assumption if

coffee is not on the land, maize is grown. Thus, profit at time t is as follows:

, . ( ) ( )(1 )e c c c m m m

t r a t t t t t t t tP Y v S P Y v S

(4.3)

where tS denotes the existence of coffee and equals 1 if coffee is growing and 0

otherwise; c

tP is the price of coffee in year t; c

tY is the coffee yield in year t; m

tP is price

of maize in year t and ,c m

t tv v are annual costs of coffee and maize, respectively.

In the model, the profit of maize is assumed constant and estimated from the Coffee

Farm Survey 2007 (Thang, 2008). The annual profit of maize is fixed at $440 per ha.

However, a sensitivity analysis is undertaken to investigate the impact of efficiency of

maize on the farmer’s decision by changing maize profit to see how the optimal cutting

and replanting rules respond.

The decision of the farmer on whether to cut/plant coffee depends on the interaction

between the price of coffee and the rule for keeping or cutting the existing trees. The

farmer’s decision will be described in the following section.

4.3.3. Decision Rule

The planting and replanting of coffee trees represent long-term investment problems for

farmers, with a number of control variables. The first decision is the (re)planting

decision; should the farmer (re)invest in coffee production, given current land area is

not in coffee. Their second decision is when, within the tree’s life cycle, should they

cut or replace coffee or leave the land idle. The cutting or replanting decision of farmers

are based on future or expected prices, which are unknown.

Yield of coffee trees relates closely to the age of the tree. Generally, after reaching the

peak level, yield will start decreasing gradually. With coffee, yield usually attains the

maximum level after the 7th

year and the mature period generally lasts about 8-9 years.

After that, the coffee yield declines. Thus, the cutting decision not only depends on the

expected price of coffee but also depends on the current age of the tree. Within the


64

model, it is assumed that the maximal age of a tree is 22 years, at which point yield falls

to zero.

Conceptually, the farmers may cut existing coffee trees before they reach the maximum

age for two main reasons. Firstly, the output price is relatively low so they may incur

losses from coffee production and they will switch to other crops (as maize in FY

model). The price at which farmers cut down the trees and switch to maize because it is

too low is denoted as CP. In this study, whenever we refer to cutting price or replanting

price, it means the prices at farm-gate and not the international or other prices.

Secondly, however, one hypothesis also considered by the model is that if output price

is very high and existing coffee trees are quite old, farmers may cut coffee trees earlier

and replant new trees. The expected net benefit from bringing the future profit stream

forward exceeds the loss associated with cutting trees before the net annual profit falls

to zero. The price at which farmers cut coffee earlier and replant new trees are denoted

as CP .

In other cases, farmers have to cut trees because of cash flow concerns i.e. expected

profit over the remaining lifetime of the tree is positive, but transitory losses mean that

the tree has to be removed early. However, this issue is not analysed in the FY model.

The cash constraint is analyzed later in the Fixed Yield – Cash Constraint model (FY-

CC model) in Chapter 5.

When the coffee price is less than the CP, farmers cut trees down and switch to maize.

However, they will grow coffee again if the price increases to the replanting price (RP).

If the price is higher than CP , farmers with old trees may cut down the existing trees

and replant new trees. Both CP and CP are expected to be age dependent.

The identification of the optimal cutting (CP,CP ) and replanting rule (RP) based on

the age of coffee trees with random prices are the outcomes of the FY model. The

model covers both the stochastic and dynamic aspect of the problem. To deal with those

problems, a sensitive fixed-form optimization technique is used for describing the

decision of farmers. This approach is similar to the method applied by Sonntag and

Hilborn (1978).

To describe the decision rules of farmers, whether coffee trees exist on the land or not is

denoted by a (0, 1) variable (St). The choice of the farmers is as follows:


65

St = 1 if St-1=1 and Pt > min (CP, RP) keep growing coffee

St = 0 if St-1=1 and Pt ≤ min (CP, RP) cut coffee and switch to maize

St = 1 if St-1=1 and Pt ≥ CP cut coffee and replant new trees

St=1 if St-1=0 and Pt>=RP switch back from maize

Because the life cycle of coffee is 22 years, all trees are cut down during their 22nd

year

regardless of the price level. In the above decision, farmers will cut coffee if Pt is

greater than the minimum value of CP and RP. This condition means farmers never cut

down the trees and switch to maize if price is above RP.

In its simplest formulation, the model will identify at what prevailing price should

coffee producers cut and when they should replant. The main fixed-forms for CP in the

FY model are defined as follows:

The quadratic form CP:

2

1 2oCP age age (4.4)

The quadratic form of CP with price change effect:

2

1 2 1( )t o t tCP age age P P (4.5)

In the “quadratic form”, CP is a quadratic function of the age of coffee trees (“age”).

Hence, CP of trees at different age may not be similar. It is anticipated that younger

trees that have a longer remaining productive life will be retained at lower prices than

those closer to their maximum age.

In the “quadratic form CP with price change effect”, the critical value for the current

coffee price at year t depends not only on the age of trees but also on the change in

coffee price between year t and 1t . The price change effect allows for information

on the direction of change in prices to influence decisions. Thus, if there is structure in

the price series, one might expect that the cutting decision will be different - for the

same current price level - if the change in price implies future increases, as compared to

future falls.

The fixed form of CP in FY model is defined as:


66

age ageCP (4.6)

As mentioned earlier, the appearance of CP in the model is to test whether farmers

will cut and immediately replant with new trees, before the maximal age. The model

does not identify the CP for all ages of coffee, it only identifies the CP for coffee

trees over 18 years old, the ages at which yield tends to go down. Hence, age+ takes the

values 18, 19, 20, 21.

According to the decision rules, farmers will cut coffee to grow maize if the coffee price

at time t is lower than age dependent CP. Hence, they cut to replant new trees if price at

time t is greater than age independent CP .

The replanting decision does not depend on the age of trees so RP function can be

defined more simply as follows:

3RP (4.7)

If the land is planted to maize (or left bare) or the coffee tree are in the last year of the

cycle (age of tree =22 year old), the farmer will replant if the price of coffee is greater

than 3

A search procedure is implemented to estimate the values of i , i and which

maximise the ENPV. All steps of the estimation procedure are described in Section

4.3.7.

The profitability of substitute or competitive crops may change the decision of farmers.

The CP or RP functions in the FY model do not include any measure of profit of the

substitute crop. However, as mentioned earlier, a sensitivity analysis is conducted to see

how the profit of maize influences the farmer’s decision.

A hypothetical example of the optimal cutting and replanting rule is illustrated in Figure

4.2. As seen in the figure, farmers will cut coffee to grow maize if the coffee price is

under the CP curve. If the coffee price is greater than RP, farmers will grow coffee

again if they have maize or bare land. In another case, if the price is above the CP

curve, farmers will cut older stands and replant new trees.


67

There is a hysteresis effect in farmer’s decision in Figure 4.2. If price drops below the

RP but it does not go below the CP, farmer will continue to maintain existing trees.

Figure 4.2. Example of optimal cutting and replanting rule

4.3.4. Yield Function

The yield of a crop depends on various factors such as natural conditions (land quality,

weather, and water supply), variety, level of intensive farming, farmer characteristics

(such as experience, farm size, education) and so on. With perennial crops such as

coffee, rubber, cashew or forestry, tree age strongly affects the yield.

In the current model, yield is assumed fixed for any given age of coffee tree but variable

with age (Figure 4.3). The first two years of the tree’s cycle are unproductive. Farmers

can start to harvest coffee in the 3rd

year, although the yield is still very low (only about

500 kg of coffee bean per hectare). After that, the coffee yield increases as the age of

the tree increases and it gets to the peak level at age 8. Once hitting the mature yield,

generally coffee yield becomes stable until it starts falling at around age 15-16. During

the mature period, an average farmer can attain approximately 2500 kg per ha. The

productivity of coffee starts declining when the trees are in its 16th

year. In general, the

coffee cycle is about 20 to 25 years. In the model, the coffee life cycle is 22. After that,

farmers will cut down their trees and if price is profitable, they will replant new coffee

trees.

0

0.5

1

1.5

2

2.5

3

1 3 5 7 9 11 13 15 17 19 21

age of trees

co

ffee p

rice (

$/k

g)

CP RP CP+

keep coffee

keep

coffee

CP+

CP

cut and

replant


68

Figure 4.3: Coffee yield by age of tree


The yield cycle by age in Figure 4.3 is based on the Coffee Farm Survey 2007 and

experience of coffee experts. In practice, however, yield of coffee is affected

considerably by input use, with some who farm intensively reporting yields of 2200 kg

per ha for trees more than 17 year old, according to the Coffee Farm Survey 2007. This

is the motivation for the study to develop the Variable Yield Optimal Model (VY

model) in Chapter 6.

4.3.5. Production Cost

Production costs by age group were estimated from the Coffee Farm Survey 2007.

Production costs by age of tree are summarized in Table 4.4. The initial investment for

(one-year old) coffee trees (replanting cost) is very high ($1440) because of expenditure

on new trees, fixtures and land preparation and planting. In the model, it is assumed that

the annual production cost in the 5-20 year age range are the same, about $930 per ha.

In the last two years of the coffee cycle, the cost reduces to just over $600 per ha

because of the reduction in labour cost for harvesting, and lower level of input

application.

0

500

1000

1500

2000

2500

3000

1 4 7 10 13 16 19 22

age of trees

yie

ld (

kg

co

ffee b

ean

/ha)


69

Table 4.4: Coffee production cost by age of tree (US$/ha)

Items Year 1 Year 2 Year 3 Year 4 Year 5-20 Year 21-22

Seedlings 179.7 18.8 9.4 0.0 0.0 0.0

Labour cost 517.5 480.1 420.0 515.0 602.5 301.3

Chemical fertilizer 106.9 118.4 134.7 168.8 181.3 90.6

Manure/organic 312.5 0.0 312.5 0.0 0.0 0.0

Pesticide 2.5 6.5 7.5 10.0 15.0 7.5

Lime 34.4 0.0 0.0 0.0 0.0 0.0

Fixed asset 156.3 0.0 0.0 0.0 0.0 0.0

Fuel/electricity 114.3 149.3 114.3 114.3 114.3 114.3

Others 16.3 21.2 16.3 16.3 16.3 16.3

Total cost 1440.2 794.3 1014.6 824.3 929.3 613.0


Based on the yield function and production cost data, we use a simple model to identify

the rotation length. The results of the model determine that the optimal time for cutting

and replanting is around 21 to 23 years depending on price levels. This strongly

supports the assumption of a 22-year cycle for coffee trees used in the optimal models.

4.3.6. Price Simulation

As mentioned earlier, the model operates over a period of 50 years. In order to get

revenue and profit of coffee production, series of coffee prices for 50 years are

simulated. This is an important step in the development of the optimal models in this as

well as subsequent chapters.

To attain the required age-structure, the model is based on 2200 price trajectories over

50 years, 100 series for each starting-age group. The simulation of these price

trajectories was based on the historical price data and a price simulation model. The

functional form of the statistical price model will generate different price structures that

may change the optimal cutting and replanting rules. Two price models are estimated

using the annual international Robusta price series from 1964 to 2006. These prices are

taken from International Coffee Organization and measured in USD per kg20

. The first

model estimates the current price as a function of lagged prices, which is called the

Lagged price model. Alternatively, the second model, Cycled price model, is based on a

9-year price cycle, and a trend. These estimated price functions for each model are

used, in conjunction with a stochastic error term to produce the required coffee price

trajectories. These price trajectories are used as exogenous variables in the model.

20

www.ico.org


70

It should be noted that in this study we are not trying to find the best price forecast

model based on historical price series. We are just generating the price simulations from

estimated models. These price simulations are expected to be consistent with the real

price in history in terms of mean, variation or time series property.

The following part of this step will provide the functional forms and results of two price

models.

4.3.6.1. Lagged Price Model

The dependent variable in this model is current price and the explanatory variables are

prices in previous years. By testing various functional forms with different numbers of

lag, a parsimonious model is identified in which the logarithm of price lagged one and

two years. The regression equation is specified as:

lnPt = 0.093 + 1.15lnPt-1 – 0.334lnPt-2 (4.8)

p- value (0.08) (0.000) (0.000)

se (0.05) (0.15) (0.15)

R2

= 0.78%

Prob. of Portmanteau =0.80

The model results show a good fit with high R2 (0.78), and all coefficients are

statistically significant. In this model, current price increases when the previous price

goes up, but with some reversal from the second lag. The model also tests the

autocorrelation property using the Portmanteau test or Q test (Ljung and Box, 1978,

Pena and Rodriguez, 2002). The Portmanteau probability value is 0.80, meaning that the

null hypothesis of no autocorrelation is accepted. The fitted logarithm of price and

observed logarithm of price since 1964 are shown in Figure 4.4


71

Figure 4.4: Fitted and actual value of logarithm of price ($/kg)

When conducting simulations, the price in year t is specified as:

t 1 t 2 0.093 1.15lnP – 0.334lnP t

tP e (4.9)

In (4.9), μt is a random variable. μt follows a normal distribution with a mean of zero

and a variance equal to that of the regression error from (4.8).

To get price trajectories over 50 years, it is necessary to have initial prices. The initial

price is drawn from a random distribution with a mean and standard deviation of

historical coffee price from 1964 to 2006.

Figure 4.5 provides some examples of international price simulation from the lagged

model. These simulated international prices are converted to a farm gate price in the

model, represented as a proportion of the international price. This proportion estimated

from simple regression between farm-gate prices and international price is equal to

0.52521

.

21

The farm gate coffee price in Vietnam is not sufficiently available in all provinces; the series is quite

short which is not representative thus the author has to use the international prices for estimating price

function. An attempt is made to estimate the relation between farm-gate price and international price.

-1

-0.5

0

0.5

1

1.5

2

1965 1970 1975 1980 1985 1990 1995 2000 2005

log of price

fitted values


72

Figure 4.5: Examples of price trajectories predicted from Lagged price model

Source: Simulated from Lagged Price model

4.3.6.2. Price Cycle Model

The Price Cycle model was motivated from observing the historical trend of coffee

prices. By analyzing coffee price over the last 30 years, results show the coffee price

seemingly follows a 9-year cycle. Although the Lagged price model does show some

degree of lagged structure, it will not generate cycles within the simulated series.

Figure 4.6: Price cycle of coffee in the world market (UScent/lb)

Source: ICO per com.

Thus, an alternative model that simulates a 9-year price cycle is estimated as follows:

0

2

4

6

8

10

12

1 5 9 13 17 21 25 29 33 37 41 45 49

time period (years)

US

D/k

g

0

50

100

150

200

250

300

350Robustas Group (Dry and wet processed)

Composite Indicator Price

1995

1977

2004

1986


73

0 1 2 3sin(2 / 9) cos(2 / 9) _tP a a year a year a year error term (4.10)

where is the ratio of the circumference to the diameter of a circle; approximately

equal to 3.14, year is a trend variable.

Equation (4.10) imposes a nine-year cycle, but the amptitude of the cycles and the trend

effect is estimated. Using price series from 1964 to 2006 and applying a regression

method, the estimated price cycle model is:

175.5 0.43sin(2 / 9) 0.33cos(2 / 9) 0.085tP year year year (4.11)

p-val (0.00) (0.02) (0.00) (0.00)

se (21.7) (0.13) (0.14) (0.01)

Prob > F = 0.0000

R-squared = 0.74

Prob. of Portmanteau = 0.83

The results give a strong goodness of fit and the coefficients are statistically significant.

The negative coefficient on the year variable captures the general downward trend in the

price series. The Portmanteau probability value of 0.83 means that it accepts the null

hypothesis of no autocorrelation in this model (Pena and Rodriguez, 2002, Ljung and

Box, 1978). Based on (4.11), alternative series of price data for the model are generated.

Similar to the Lagged price model, it is necessary to identify the starting price.

However, rather than select an initial price at random, here the model selects an initial

point in the cycle at random (each simulation employs a time trend yeart where the

value of year0 is selected with uniform probability from 1-9)

175.5 0.43sin(2 / 9) 0.33cos(2 / 9) 0.085*2000t tP year year (4.12)

where μt is a random variable. Values of μt distribute normally with a mean of zero and

a variance equal to that of the regression error from (4.11). The trend is held at 2000

because the coffee price in 2000 is closest to the mean of coffee price series from 1964

to 2006.

Figure 4.7 gives an example of the international price trajectories predicted from the

price cycle model.


74

Figure 4.7: Example of price trajectories predicted from Price Cycle model

Source: Simulated from the Price Cycle Model

Table 4.5 compares the distribution of historical international price series from 1964 to

2006 and simulated price data sets from the Lagged model and the Price Cycle model.

The mean of price distribution generated from the two models is very close but variation

of price distribution from the Lagged model is much larger. The distribution of price

data set from the Lagged model gives a better fit to the historical data.

Table 4.5: Distributions of actual international price and price data set simulated

from two models

Actual

pricesa

Lagged

modelb

Price Cycle

Modelb

International price

Number of observations 53

110,000

110,000

Mean 1.76 1.91 1.98

Standard deviation 1.02 1.14 0.66

Farm-gate average pricec

0.92 1.0 1.04

Note: a data price series from 1964 to 2006;

b 2200 simulated price trajectories for 50 years

c farm-gate price is derived from international price

In the FY model, the identification of the optimal cutting and replanting price for coffee

farmers is implemented using the price data set simulated from both price models

(Lagged price model and Price cycle model). The price data set simulated from each

model has its own characteristics. The price data in the Lagged model are quite quick to

reverse to the mean but in the Cycled price model, price trajectories have a strong

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

1 6 11 16 21 26 31 36 41 46 51


75

structure with a 9-year cycle. The difference of price data simulations may affect the

optimal rules as well as farmers’ income.

4.3.7. Procedure for Estimation

To identify the optimal CP (and CP ) and RP that maximises the expected net profit,

the model applies a search procedure for retrieving αi, γ and age

in (4.4), (4.5), (4.6)

and (4.7). The procedure for estimation includes the steps as follows:

Step 1: Preparing Input-Output data

In this step, data on coffee production cost, yield for 22 groups of coffee by tree ages,

maize profit were calculated from survey data and other sources.

Step 2: Price simulation

This step generates the price simulations as presented in Section 4.3.6 and puts them

into the model. In total, 2200 price trajectories for 50 years are generated by two

models- Lagged price model and Price Cycle model.

Step 3: Setting decision rule

Select the fixed-form for the cutting rule and replanting decision of coffee farmers as

presented in “decision rule” section of the model.

Step 4: Search for the best parameters in CP (and CP ) and RP

This step uses the searching method to find the cutting price rule (CP and CP ) and

replanting price (RP) to get the maximum ENPV. As described above, the CP was

expressed as a fixed form of coffee age, and change in price. Thus, the final objective of

searching is to find the αi and γ to get optimal CP, CP and later RP to achieve the

maximization of ENPV.

To get optimal value for CP, CP and RP, the model applies the one-at-a time method.

This is one of the simplest optimum seeking technique which may be applied to a

function of any number of decision variables (Taylor et al., 1973). To apply the one-at-a

time method, RP is first fixed by assigning an initial value and then finding the optimal


76

CP and CP . When having optimal CP and CP in Step 1, the cutting rule is now fixed

and RP is searched over until its optimal value is determined. The entire procedure is

repeated until CP, CP and RP converge, and the ENPV gets the maximum value.

The strategy used is to start with a relatively coarse grid of g values, giving a total

search space of gn (where n is the number of parameters) and then progressively refine

the search with a smaller grid size around the maximal values.

Excel software is used for running the model. The structure map of the spreadsheet

model in Excel is presented in Figure 4.8. The model consists of different sheets: coffee

price, decision sheet, coffee yield, production cost, revenue and profit. When the cutting

and replanting rule changes, the farmer’s decision will change which in turn brings

about the new cost, yield, profit and finally ENPV.


77

Figure 4.8: Model Structure Map

Coffee Price sheet:

2200 price series in 50 years for 22 age group of coffee trees

Decision sheet: This sheet expresses the decision of farmers: cutting for maize or keeping; replanting through the age of coffee tree. This sheet covers 22 starting age groups of coffee x 100 replications for one group (total of age replication is 2200, equivalent to 2200 price series in Sheet “Coffee price”. Changes in decision rule will vary age of tree, thus change cost, yield and profit

Decision sheet

Age-Year 1 Year 2 Year 3 ………….. ………….. ………….. Year 49 Year 50

1 2 3

1 2 3

……

1 0 0

2 3

2 3

……

2 3

……

22 1

22 0

……

22 0

100 replicants

Coffee cost: The production cost is derived from the age of coffee trees in Decision sheet.

Yield The yield of coffee is also derived from the age of coffee trees in Decision sheet.

Revenue The revenue of coffee is calculated by multiplying Yield and Coffee Price

Farm profit = coffee profit +maize profit To get the ENPV of farm profit for 2200 replications in 50 years, first we calculate the NPVa of farm profit for each starting age a for 50 years, and take

the average of 2200 NPV. Changes in decision rule ( i , ) will produce a

particular NNPV

Coffee Profit The revenue of coffee was calculated by taking the difference between coffee cost and revenue


78

4.4. Results of the FY Model

This section presents results of the FY model. As mentioned above, the optimal rule

may depend on the way in which the price of coffee is simulated. Thus, the presentation

of results of the FY model is split into two parts. Section 4.4.1 discusses the results

based on price simulations generated from the Lagged price model. The results based on

price simulations from the Price Cycle model will be presented in Section 4.4.3.

4.4.1. Optimal Rule with Lagged Price Model

With price trajectories simulated from the lagged price model and application of the

searching procedure, the FY model finds the optimal quadratic CP and replanting rule

for coffee as follows:

CP = 0.402 - 0.0509age + 0.00367age2 (4.13)

RP =0.74 ($/kg of coffee bean)

Optimal ENPV = 9226 ($)

The FY model finds that the ENPV achieves a maximum when CP for the 18-21 age

group is infinitive (very high price). This means that it is never optimal within the model

to cut trees early and immediately replant, in an effort to bring the future benefit stream

from subsequent trees forward. However, in other cases if the replacement cost was

relatively low and discount rates close to zero, CP would be finite.

Because CP does not influence farmer’s decision, thus from now the FY model and

other optimal models in following chapters eliminates the hypothesis which states that

farmers cut very old trees and replace by new plantings if the coffee price is very high.

Hence, only cutting rules of CP in which farmers cut coffee trees and switch to maize if

price is low are considered hereafter by the FY model and other optimal models.

Figure 4.9 depicts the optimal rule with quadratic CP for all age groups of coffee. The

results shows that the optimal CP for one year old coffee is 0.34 $/kg. In initial years of

the life cycle, the CP decreases slightly when tree age increases and CP is smallest for

7-year old trees, at only $0.23 per kg.


79

The replanting price of $0.74 per kg indicates that if farmers have bare land or land is

occupied by maize production, they should switch back to coffee if the coffee price is

equal to or greater than $0.74 per kg.

Figure 4.9: Optimal cutting and replanting rule from the FY model

This optimal replanting price ($0.74 per kg) is significantly smaller than the optimal

replanting price ($0.91 per kg) when the model is solved assuming constant

deterministic price (i.e the price at which the NPV of coffee production is equal to NPV

from maize only). This is because the average of simulated farm-gate prices is equal to

$1 per kg of coffee bean and hence it is optimal to replant at lower price given that one

expects the price to rise on average over the life of the tree.

The model was also solved with a cubic form CP ( 2 3

1 2 3oCP age age age ).

However, the results are almost the same in terms of the cutting rule per age of tree, and

ENPV generated, but the cubic model takes longer to determine the optimal rule

because of the additional parameter ( 3 ). Thus, to save searching time with different

scenarios, the model uses the quadratic function as the optimal form of CP.

The optimal rule of the FY model is presented in Figure 4.9. However, the optimal rules

do not show the frequency cutting decision. Thus, it is useful to see how many times the

cutting rule is invoked at each age or the percentage of cases in which farmers actually

cut their coffee at optimal rules. Perhaps the cutting rule is not invoked for some ages of

trees due to the range of prices, and hence little reliance can be placed on the precise

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1 3 5 7 9 11 13 15 17 19 21

age of tree

farm

-gate

pri

ce (

$/k

g)

Optimal CP

Optimal RP

RP=0.74

cut and switch to maize

keep coffee

keep coffee


80

value given for the cutting rule. Figure 4.10 presents the proportion of times a tree of a

specific age is cut due to the optimal cutting rule being invoked. As shown in this

figure, with price trajectories generated by the lagged model, farmers rarely cut if the

age of the coffee tree is less than 11 year old. The cutting frequency is increasing

continuously for trees of ages from 11 to 20. The implication is that, for the simulated

price series used, which is based on historical data, it is seldom optimal to cut in this

early period. This result may not hold in alternative circumstance i.e. where the price

series follow different distributions.

Figure 4.10: Proportion of actual cut in FY model with Lagged price simulation

The optimal RP is quite consistent with stated replanting prices from the Coffee Farm

Survey 2007. The optimal RP from the FY model ($0.74 per kg) is very close to the

replanting price reported by farmers in Krong Pak district (Dak Lak province, Central

Highlands of Vietnam). Krong Pak has a medium comparative advantage in coffee

production. Coffee yield in 2006 in Krong Pak district was about 1920 kg coffee bean

per ha (Thang 2008). The replanting price reported by farmers in the more productive

area, Cu Mgar district with average coffee yield of 2100 kg per ha, is lower, only $0.706

per kg. By contrast, the replanting price of farmers in Eakar district was reported at

$0.809 per kg. This means farmers in lower yield areas are less likely to replant coffee,

for any given price.

0

2

4

6

8

10

12

14

16

18

20

1 3 5 7 9 11 13 15 17 19 21

age of trees

actu

al c

ut (%

)


81

Figure 4.11: Comparison of optimal RP of FY model and farmer’s estimates

The decision to cut early is implicitly dependent on the prices that are expected to hold

over the remaining lifetime of the trees. According to the cutting rule, the current price

of coffee summarizes all possible information about that future trajectory. However, it

is possible that this is not the case, and that additional information may be of value in

making the optimal decision. The FY model also specifies the fixed-form cutting rule

as a quadratic function of age with an additional price change effect. The hypothesis is

that information on the most recent change in price may moderate the cutting decision

(i.e. if the previous price change was positive, one may be less willing to cut than if it

were negative, for any given price level). The results from this FY model show that the

difference in price does not seem to influence the cutting rule (i.e =0.002). Similarly,

the optimal ENPV and RP are almost unchanged.

CPt = 0.40 - 0.050age + 0.0036age2 + 0.002 (Pt –Pt-1) (4.14)

RP=0.74 ($/kg)

Optimal ENPV =9228 ($/ha)

The FY model was also solved with the constant CP form (i.e. CP= 0 ). This means CP

does not depend on the age of trees. With such constant form of CP, the results show

that the coffee farmer will get the maximum income if they cut trees at a price of $0.36

per kg. The optimal RP when there is a constant CP was found to be the same as RP

when CP is a function of age (see Figure 4.12).

0.706

0.743

0.809

0.640

0.660

0.680

0.700

0.720

0.740

0.760

0.780

0.800

0.820

Cu Mgar Krong Pak Eakar

district

$/k

gOptimal RP =0.74


82

Figure 4.12: Optimal quadratic CP and best constant CP from the FY model

Figure 4.13 presents the optimal income per ha attained by applying the three cutting

rules: (i) the constant CP, (ii) the quadratic CP and (iii) no cutting rule (CP=0). The

results show that ENPV per ha gained from the optimal quadratic CP rule is about 3

percent higher than income with a constant CP (the best constant CP =0.36), and nearly

5 percent higher than income if farmers never cut early.

Figure 4.13: The maximum ENPV per ha among different CP rules

The results of different cutting rules are summarized in Table 4.6

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1 6 11 16 21

age of tree

pri

ce (

$/k

g)

Optimal CP

Best contant CP

RP RP=0.74

CP =0.36

8600

8700

8800

8900

9000

9100

9200

9300

NPV with constant CP Optimal NPV NPV with no cutting rule

NP

V (

$)

100%

97%

95%


83

Table 4.6: Summarized results of different cutting rules of FY model

Cutting rule (CP) Optimal cutting price Optimal

RP

ENPV

($/ha)

Quadratic form CP = 0.402 -0.0509age + 0.00367age2 0.74 9226

Quadratic with

price change effect CPt = 0.4 -0.05age + 0.0036age

2 + 0.002(Pt –Pt-1) 0.74 9228

Constant 0.36 0.74 8950

never cut early n.a 0.74 8760

Source: optimal model results

To explore the benefit from the application of the optimal quadratic CP against “no

cutting” rules, the FY model is used to identify the ENPV for different starting age of

trees in two cases. Figure 4.14 shows the results of ENPV (per ha) for “never cut early”

and quadratic CP by the starting ages (the age of the tree at first year of period in the

model). With the younger trees, ENPV does not show a big change but the difference

gets larger with the increasing starting age. This is consistent with the evidence

presented earlier that the optimal cutting rule was seldom invoked for young tree ages.

As a result, for blocks with initial trees of young ages, divergences in behavior will only

occur after a significant number of years have lapsed, and the impacts of these will be

discounted. On the other hand, blocks with trees that are of older age will see

divergences in behavior more quickly in the time sequence.

Figure 4.14: ENPV with different starting ages for quadratic CP and no cutting

rule

5000

6000

7000

8000

9000

10000

11000

12000

13000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

initial age of tree

NP

V (

$)

no-cut

Optimal rule


84

4.4.2. Impact of Substitute Crop on Coffee Farmer’s Decision

In the FY model, maize is assumed to be the sole replacement for coffee. If the coffee

price is too low, farmers will cut coffee and switch to maize until the price comes back

to the replanting price. In addition, maize profit is also assumed fixed. However, the

change in profit for the substitute crop may affect the cutting and replanting decision of

coffee farmers. To examine how the profit of the substitute crop affects the farmer’s

decision, the FY model has been resolved for a number of alternative values for the

maize profit. The model is only solved with the quadratic form of CP (

2

1 2oCP age age ).

First, maize profit is assumed to increase by 20 percent. With young coffee groves, the

CP is seemingly unchanged when maize profit rises by 20 percent. However, farmers

are more likely to cut older trees. With 20 percent increase in maize profit, total income

per ha is $9405, approximately 2 percent higher than the previous ENPV.

With a reduction in maize profit, coffee farmers are less likely to cut and the ENPV is

also lower (see Figure 4.15 and Figure 4.16). This happens because with lower income

from replacing crop, farmers will not cut earlier and they should continue to grow

coffee trees if the expected profit from coffee production is still greater than that from

maize.

Figure 4.15: Changes in optimal rule when maize profit varies

0

0.2

0.4

0.6

0.8

1

1.2

1 3 5 7 9 11 13 15 17 19 21

price ($/k

g)

age of coffee trees

Optimal CP 20% reduction in maize prof it CP

Optimal RP 20% reduction in maize prof it RP

-0.1

0.1

0.3

0.5

0.7

0.9

1.1

1.3

1.5

1 3 5 7 9 11 13 15 17 19 21

age of coffee trees

pri

ce

($

/kg

)

Optimal CP 20% increase in maize profit CP

Optimal RP 20% increase in maize profit RP


85

Figure 4.16: Changes in the maximum ENPV when maize profit varies

Another simulation in which maize profit is increased to $1500 per year has been

solved. The results shows that in this case all coffee farmers will cut their coffee trees

and switch to maize. They will never grow coffee trees except under the circumstances

that the farm-gate price of coffee bean exceeds $7.6 per kg.

4.4.3. Optimal Rules with Price Cycle Simulation Model

In the previous section, the FY model identifies the optimal rule for coffee farmers

based on price series simulated by using an autoregressive model. The estimated

equation implies some structure in the price series, but no structural cycles. If a coffee

cycle exists, then it may make prices more predictable. In this case, the decision to cut

and replant trees may depend on not only the level of the price, but also where price is

in the cycle. However, the lack of significance of the change in prices term in the fixed

form rule may be accounted for by the relatively weak structure within the price

simulation (i.e. it closely approximates a random walk). This may not be the case if

there are clearly predictable cycles in prices. This section reports the results from

solving the FY model with the same procedures but with price data generated from the

Cycle model.

With the Price Cycle model, the optimal quadratic CP and RP are identified as follows:

CP = -0.07 - 0.087age + 0.0065age2

(4.15)

RP = 0.61 ($/kg) ENPV= 9659 ($/ha)

9226

9080

9405

8900

9000

9100

9200

9300

9400

9500

Optimal 20% reduction in maize profit 20% increase in maize prof it

($/h

a)


86

The rules are depicted in Figure 4.17 below. The results show that if price follows the

cycles with the same distribution as occurred in the past, farmers should never cut

coffee if trees are younger than 14 years. If land is either bare or grows maize, they

replant earlier, and the optimal RP is only $0.61 per kg of coffee bean. The optimal

ENPV with the Price cycle model is similar to the maximum ENPV from the Lagged

price model simulation.

Figure 4.17: Optimal rule of the FY model with Price cycle model

Figure 4.18: Optimal rules of Price cycle and Lagged price simulations

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1 3 5 7 9 11 13 15 17 19 21

age of trees

pri

ce (

$/k

g)

Optimal CP optimal RP

cut and

switch to

maize

keep coffee

keep coffee

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1 3 5 7 9 11 13 15 17 19 21

pri

ce (

$/k

g)

age of trees

CP-Price cycle simulation RP-Price cycle simulation

CP-Lagged price simulation RP-Lagged price simulation


87

The difference between the optimal rules in the two price simulations is in part due to

the distribution of price trajectories predicted from the two models. Figure 4.19 graphs

the distribution of farm-gate price data sets simulated from the Lagged price model and

the Price cycle model. The mean of the two data sets is similar but the distribution is

quite different. Price data generated from the Lagged price model is skewed to the left

with a long tail to the right and a higher standard deviation (0.6) whereas the Price cycle

data is symmetric with similar mean (1.04) and smaller standard deviation (0.34).

Figure 4.19: Distribution of farm-gate price data set simulated from two models

From the optimal rule, the model identifies the actual percentage of trees at each age

that is cut in this case. Figure 4.20 compares the percentage of times trees are cut at

each age under the two alternative models. As shown in the histogram, with the Lagged

price model simulation farmers are more likely to cut earlier. With the Price Cycle

model, the cutting percentages of trees increase rapidly in the 17 to 19 age group.

Although the two rules look visually quite different, in terms of economic implications,

both predict very low levels of cutting up to age 10.

0.2

.4.6

.81

Den

sity

0 1 2 3 4Lagged price model

0.5

1

Den

sity

0 .5 1 1.5 2 2.5Cycle price model


88

Figure 4.20: Simulated percentage cut at each age of trees from two data sets

As was undertaken with the Lagged price model, the Price cycle model was simulated

allowing for a price change effect in CP. In this case, the form of CP can be repeated as

follows:

2

1 2 1( )t o t tCP age age P P

With the above cutting function, the model was solved again and the new optimal rule is

identified as follows:

CPt = -0.16 - 0.083age +0.0063age2 -0.24(Pt-Pt-1) (4.16)

RP = 0.61 ($/kg)

NPV = 9660 ($)

The negative sign of price difference coefficient (-0.24) in (4.16) is expected and it

shows that if the current price follows a downward trend, farmers should cut earlier and

vice versa. The coefficient (-0.24) also shows the more significant impact of price

differences in the Price Cycle model simulation compared to the Lagged Price model.

However, the new results from (4.16) give an almost unchanged optimal ENPV and the

same RP when compared with those from (4.15), suggesting that allowing for this

additional information does little to improve the economic performance of the farmer.

Figure 4.21 presents the optimal rules for the quadratic model with price change effect

CP assuming that the price difference is equal to zero. Similarly, the actual cutting

percentages at different age of coffee trees are illustrated in Figure 4.22. According to

0

2

4

6

8

10

12

14

16

18

20

1 3 5 7 9 11 13 15 17 19 21

age of coffee

% a

ctu

tal

cu

tLagged price model

Price Cycle model


89

the results, with the CP form of quadratic and price change effect, farmers are less

likely to cut. The cutting percentage in this case is a little bit lower than in the Price

cycle model and quadratic form of CP.

Figure 4.21: Different optimal rules of FY model with Price cycle simulation

Figure 4.22: Actual cut of the FY with Price cycle model and different CP forms

4.5. Conclusion

There are many approaches to analyze farmer’s decisions and identify the optimal

cutting and replanting rules. By using fixed form optimization, this chapter develops the

-0.5

-0.3

-0.1

0.1

0.3

0.5

0.7

0.9

1.1

1.3

1 3 5 7 9 11 13 15 17 19 21

age of trees

pri

ce (

$/k

g)

Quadratic CP

RP

Quadratic with price change effect CP (assume price change =0)

0

2

4

6

8

10

12

1 3 5 7 9 11 13 15 17 19 21

age of trees

% a

ctu

al

cu

t

With cycled price model, quadratic CP

With Price cycle model, quadratic with price

change effect CP


90

FY model to identify the optimal cutting rule and replanting rule for coffee farmers in

Vietnam.

Results from the FY model found the optimal CP is dependent on the age of coffee

trees. In addition, the FY model found that the optimal RP is $0.74 per kg of coffee

bean. The maximum ENPV earned from one hectare of land is $9226.

With optimal rules, farmers rarely cut if the age of coffee trees is less than 11 year old.

The cutting frequency increases continuously for the coffee trees from the 12th

year.

The FY model was also solved with other fixed forms of CP (i.e. age dependent cubic

CP, and quadratic CP with price change effect). However, the results are very similar.

Solving the FY model with constant form CP, the results shows that the maximum

ENPV from optimal quadratic CP is higher than the ENPV from the optimal constant

CP (CP is not a function of age) by 5 percent.

The coffee farmer’s decision changes when the profit of maize (substitute crop) varies.

The results from the FY model indicate that if the profit of maize increases, coffee

farmers are more likely to cut, and then only replant coffee at a higher price. By

contrast, coffee farmers are less likely to cut with a lower profit of maize and they will

plant coffee again at lower prices.

In general, the FY model identified the optimal CP and RP for achieving the maximum

ENPV for coffee farmers. Furthermore, the model can investigate the change of coffee

grower’s decision when the price of the substitute crop changes. However, farmers in

the FY model are assumed to have no cash constraint i.e. they can take on high levels of

debt in initial years in anticipation of profit streams in the future. In practice, coffee

farmers, especially poor farmers may not be able to follow this optimal decision. Based

on the FY model, the following chapter will develop the Fixed Yield- Cash Constraint

model (FY-CC model) to investigate changes of the coffee farmer’s decision if they

face cash constraint, and the impact on their income.

Chapter 5. Optimal Coffee Planting Decisions under a Cash Constraint

91

CHAPTER 5. OPTIMAL COFFEE PLANTING DECISIONS

UNDER A CASH CONSTRAINT

5.1. Introduction

In the previous chapter, the Fixed Yield (FY) model was developed to determine the

optimal cutting and replanting price for coffee farmers in Vietnam to get the maximum

ENPV per ha of land. In the FY model, it was assumed that the farmer’s decision only

depends on the price of coffee and returns to maize. However, if farmers do not have

enough working capital from either retained profits or credit to pay for new coffee

gardens they may be unable to replant optimally. In addition, they have to wait several

years to get income from coffee. During those years, they need money to cover their

living expenditure and pay for inputs. This chapter will look at the coffee farm as a

household and analyze the impact of cash constraints on the farmer’s decision. With that

goal, this chapter focuses on relatively poor coffee farmers, who are presumed to have a

shortage of cash for farm investment and living expenses.

This chapter aims to identify to what extent the expected profit from coffee/maize is

reduced under a cash constraint and examine the impact of credit policy on the income

and planting decisions of poor coffee farmers;

To achieve the above objectives, this chapter develops the Fixed Yield- Cash Constraint

model (FY-CC). This differs from the FY model in Chapter 4, as the model includes

other aspects of the household such as expenditure, savings, loans and household size.

However, yield and production costs remain fixed in the same manner as in the FY

model.

The remainder of the Chapter is organized as follows. Section 5.2 briefly reviews the

theoretical literature on the impact of cash constraints on farmer decisions. Section 5.3

describes the main characteristics of poor coffee households in Vietnam including

income, expenditure and savings. Section 5.4 describes the FY-CC model and its

decision rules. The final section presents results from the FY-CC model.

5.2. Impact of Cash Constraints on Farmer’s Decision


92

Rural households in developing countries like Vietnam are characterised by low and

variable incomes. These households suffer from income variability due to fluctuations

in weather and output prices. Farmers are also vulnerable to other risks associated with

small businesses. During periods of low income, farm households have to use their

savings and borrow money to continue farming and cover living expenses. Agricultural

investments tend to be funded by credit from banks and other organizations. Thus the

development of financial markets has become more important for households as well as

the agricultural sector, especially for poor households (Gutierrez, 2002). A shortage of

cash results in many problems for households. According to Mendola (2007), having

poor initial asset endowments means that poor households may not be able to use their

existing resources as efficiently as better-off households. In other words, poverty

contributes to deviations in the behavior of farm households from full efficiency.

Similarly, Winter-Nelson and Temu (2005) indicate that small farmers in developing

countries are trapped in poverty for lack of cash needed to make profitable investments.

Thus, increased access to credit could generate economic growth amongst poor

households.

There have been many studies to evaluate the role of credit and the impact of cash

constraints on household’s behaviors and income. It is widely believed that farm

households in developing countries are credit constrained and the provision of credit

would lead to an increase in production and income (Simtowe et al., 2006, Freeman et

al., 1998).

Credit access may affect the household production and income in various ways.

Through access to credit markets, households can move away from risk reducing but

low return diversification strategies and concentrate on risky investment that gives

higher returns (Simtowe et al., 2006). Similarly, by using a dynamic inter-temporal

model for the analysis of the rate of investment in the agricultural sector in Italy,

Gutierrez (2002) pointed out the importance of financial constraints in capital markets

in determining the rate of investment. He showed that when credit constraints hold, the

expected marginal profit per unit of capital is reduced. In a similar vein, through looking

at the role of a credit constraint on dairy households, Rosenzweig and Wolpin (1993)

found that low incomes combined with borrowing constraints are the primary reasons

for underinvestment in bullocks in India, with improvements in earnings increasing

agricultural profitability by permitting farmers to accumulate larger capital stocks. With

better access to credit, farm productivity increases. By comparing the impact of credit


93

programs on capital constrained and non-constrained dairy small holders in Ethiopia

and Kenya, Freeman and Ehui (1998) showed the significant impact of loans on farm

productivity. According to the study, the marginal contribution of credit to milk

productivity is relatively high on liquidity-constrained farms compared to liquidity non-

constrained farms: one percent increase in credit used to purchase crossbred dairy cows

leads to 0.6 percent increase in milk productivity on credit-constrained farms and 0.4

percent increase on non-credit constrained farms in Ethiopia. In Kenya one percent

increase in credit for investment in crossbred dairy cows leads to 1.6 percent increase in

milk productivity on credit constrained farms and 0.9 percent increase on non-credit

constrained farms.

Issues related to poor households and poverty in Vietnam have been extensively

studied. However, most of them focused on the causes of the poor and suggesting

solutions for attacking the household poverty (SRV, 2002, MOLISA, 2003, Svendsen,

2003, The World Bank, 1999, The World Bank, 2003, The World Bank, 2005, The

World Bank, 2007, Inter-Ministerial Poverty Mapping Task Force, 2003 , Shenggan et

al., 2003). Very few studies were concerned about the poor coffee households, and

mostly focused on the Central Highlands region, especially the Dak Lak province. For

example, ICARD and Oxfam (2002) implemented a study to evaluate the impact of the

collapse in the global coffee trade on Dak Lak province. This study described some

problems of coffee farmers and poor households under the price crisis in early 2000.

Similarly, ADB and ActionAid Vietnam (2003) investigated determinants of poverty in

Dak Lak and analyzed solutions that can support poor households. Those studies on the

poor coffee farmers did not investigate the optimal cutting and replanting decision of

the poor coffee farmers as well as the impact of credit policy on their income. These

issues will be analysed in the FY-CC model. Before moving to the model section to

estimate the impact of cash constraints and identify behaviors of coffee household under

a cash shortage for investment, the next section will review the poverty trend in

Vietnam and investigate some characteristics of poor coffee households in Vietnam,

especially in the Central Highlands where most of coffee producers are located.

5.3. Poverty Trends in Vietnam

Economic growth over the 1990s generated significant improvements in living

standards in Vietnam. The national trend shows strong poverty reduction in the past 15


94

years throughout Vietnam. Poverty incidence22

was reduced from nearly 60 percent in

1993 to only 16 percent in 2006. The rate of decline in poverty was faster in urban

areas, but rural populations also saw improvements in well-being.

Figure 5.1: Poverty trend in Vietnam 1993-2006

Source: GSO per com

The declining trend in poverty is evident in all regions. Poverty rates and the speed of

poverty reduction vary from region to region. The North West is the poorest region in

the country, followed by the Central Highlands and the North Central Coast. Poverty

rates are also relatively high in the two deltas, and in the South Central Coast, but are

much lower than in the Central Highland. The poverty map and poverty depth map by

provinces in Vietnam are presented in Figure A5 and Figure A6 in Appendix A.

Figure 5.2 shows the poverty trend in all regions in Vietnam since 1993. It is noted that

in the period 1998-2002, the fraction of the population deemed poor was declining in all

regions, excepting the Central Highlands. There was no reduction in poverty in the

Central Highlands from 1998 to 2002. Coffee is the main crop in the Central Highlands

and the slow rate of poverty reduction in this period could be explained by the sharp fall

in the coffee price during the same period. The nominal farm gate price in 2002 was one

fourth of that in 199823

. Consequently, the percentage of coffee producers with incomes

below the poverty line in Vietnam remained at around 35 percent in 1998-2002.

22

This is based on the poverty line of per capita income of less than $12.5 a month in rural areas and

$16.25 a month in urban areas 23

Calculation based on VHLSS1998 and VHLSS2002

16

19.5

28.9

37.4

58.1

25

97

44

20

25

66

46

36

0

10

20

30

40

50

60

70

1993 1998 2002 2004 2006

povert

y in

cid

ience (

%)

Vietnam

Urban

Rural


95

Figure 5.2: Poverty trend in Vietnam by regions 1993-2004

Note: RRD: Red river delta; NCC: North Central Coast; SCC: South Central Coast, CH: Central

Highlands, NES: Northern East South; Mekong River Delta

The decline in the coffee price in the early 2000s forced many farmers to reduce their

coffee area, as revenue did not cover variable costs. Many producers, mainly poor

farmers, could not cover their expenditures (living costs and input investment) so they

had to shift to other crops such as rice or maize for food security. Poor farmers are

especially vulnerable and influenced strongly by external shocks such as price

reductions and natural disasters. The causes of poverty are diverse. Table 5.1 presents

the main causes of poverty from the perception of poor people and local authorities in

Dak Lak province (the largest coffee production province in Vietnam) with lack of

capital, shortage of land, uncertainty of market, and poor infrastructure are the main

reasons.

Table 5.1: Perceived causes of poverty in Dak Lak Province

Perceptions of Poor People Perceptions of Local Authorities

Poor infrastructure: irrigation systems,

roads

Poorly developed markets

Ineffectiveness of Government

policies and programs at grass-root

level

Lack of transparency, accountability,

resulting in corruption; lack of

Lack of capital

Shortage of land

Many dependents to support/subsidy

Lack of experience, and inability and

incapability to apply new farming

techniques

Investment failure, risks in agriculture

(coffee price dropped)

0

10

20

30

40

50

60

70

80

90

100

Vietnam North

East

North

West

RRD NCC SCC CH NES MRD

1993 1998

2002 2004


96

people’s participation in decision

making

Inability and weakness of grass-roots

authorities and cadres

Villagers’ inability to apply new

farming techniques and low level of

education

Shortage of land

Lack of capital

Free in-migration

Poor health and lack of labour

Harsh climatic conditions: drought

Poor health, disability, getting old

Lack of labour

Committed to social diseases (drug

addicted), and laziness

Harsh climatic conditions: drought,

flood

Source: ActionAid Vietnam & Asian Development Bank (2003)

Of the eight ecological regions in Vietnam, the Central Highlands is the major coffee

area with nearly 90 percent of coffee output produced in this region24

. Farmers in the

Central Highlands invested heavily in coffee over the mid to late 1990s and the

subsequent fall in coffee prices left many of them with low incomes. About 40 percent

of households in the Central Highlands produce coffee. This proportion does not vary

across the population, except for the richest fifth who were much less involved in coffee

growing. According to ActionAid Vietnam & Asian Development Bank (2003), the

number of trees planted, on the other hand, varies substantially. Coffee farmers in the

poorest population quintile have, on average, 6500 trees. Those in the second-richest

quintile have nearly doubled this number (see Table 5.2).

Table 5.2: Coffee farming in Central Highlands

Expenditure quintile

Central

Highlands I-lowest II III IV

V-

highest

Households growing

coffee (% of total

household) 38 43 40 44 24 39

Average area (m2) 6539 9499 9184 12820 11487 8881

Source: ActionAid Vietnam & Asian Development Bank (2003)

Generally, the coffee farm size in Vietnam is quite small with less than 1 ha per

household. The result provided by ActionAid Vietnam & Asian Development Bank

24

GSO (2006)


97

(2003) is consistent with data from Agrocensus_2006. According to Agrocensus_2006,

the farm size of poor coffee farmers is much less than higher income households, only

5287 m2. Even poor farmers in Kon Tum province have nearly 3500 m

2 land of coffee

(see Figure 5.3). The expansion of coffee area in Central Highlands has been limited by

land availability. The development of the coffee area in Central Highlands and in

Vietnam has moving together with migration and forest exploitation. The Central

Highlands region has become an important destination for migrants since 1980, with the

population of Dak Lak increasing from 35,000 people to more than 2 million in 2003.

According to provincial statistical authorities, sixty percent of current population is

migrants (ActionAid Vietnam & Asian Development Bank, 2003). The Statistical

Office of Dak Lak indicates that one million hectares of forestry land has been

converted to other uses (especially coffee) since 1975.

Figure 5.3: Coffee area of poor farmers in Central Highlands by provinces

Source: Calculation based on Agrocensus_2006

Within coffee farmers in the Central Highlands, there are approximately 25 percent of

households living under the poverty line. In four provinces in Central Highlands, Kon

Tum province has the largest proportion of poor coffee farmers with over 39 percent;

followed by Gia Lai (24.4 percent)25

.

Table 5.3: Poverty incidence of coffee farmers in Central Highlands, Vietnam

Province Poor Non-poor Total

Kon Tum 39.9 60.0 100

Gia Lai 24.4 75.5 100

25

Calculation based on Agrocensus_2006.

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

Kon Tum Gia Lai Dak Lak Lam Dong Average

Overall

Poor household


98

Dak Lak 24.0 75.9 100

Lam Dong 20.4 79.5 100

Overall 24.58 75.4 100


There is not much difference in family size between poor and non-poor farmers. Most

households have four or five people. On average, there are 5.23 people in poor

household while this number in non-poor household is 4.11 persons. Only 15 percent of

coffee household have more than six people

Figure 5.4: Structure of family size of poor and non-poor coffee household


In general, poverty in Vietnam is still relatively high, especially in rural areas and

amongst coffee households. The next section will look at the relationships between

income, expenditure and savings, as savings are the principle way in which households

overcome variability in income.

5.3.1. Saving and Income Level in Vietnam

The analysis of income and savings is important for modeling the decisions of poor

coffee households in the FY-CC model because savings determines the investment

capability of households. The cutting and replanting decision of farmers depends on

various factors, but mainly based on (i) coffee and replaced crop price levels and (ii)

cash available to the household. Farmers cannot replant coffee if they do not have

enough money for new investment and for household expenditure especially while

coffee trees are unproductive. In addition, analysing the relationship between household

0

5

10

15

20

25

30

1 2 3 4 5 6 7 8 9

people per household

pe

rce

nta

ge

Poor

Non-poor


99

expenditure and income of household helps us understand how much farmers spend on

living costs and other expenditure, and how much they save for production investment.

Unfortunately, data on expenditure and income for poor coffee farmers is not available.

Hence, in this study, the relationship between savings and income of poor household is

investigated using the VHLSS2006 data26

. One important issue should be noted that the

poverty line is usually based on household expenditure rather than income. Poverty is

measured using expenditure because income data is less reliable than expenditure data

in the VHLSS. In this section, per capita expenditure is the criteria for distinguishing

between poor and non-poor households and applying the poverty line set by the General

Statistical Office of Vietnam and Ministry of Labour and Invalid Social Affair

(MOLISA). According to this line, households in rural area are poor if their per capita

expenditure is less than US$12.5 a month. This number increases to US$16.25 a month

for households in urban areas.

According to the VHLSS2006, the average total income of non-poor households in rural

areas was $2420 per year while poor household earnt $1168 per year. Non-poor

households expenditure was on average about $1300 per year while for poor households

it was $628 per year (see Table 5.4). On average, poor farmers in rural areas could save

$540 per year.

Table 5.4: Household income and expenditure in rural area in 2006 ($/year)

Type of household Total household income

Total household

expenditure

Household

savings

Non-poor 2422 1310 1111

Poor 1168 628 540

Source: calculation from VHLSS2006

There is a big gap in household income between urban and rural areas. According to

VHLSS2006, per capita income of households in urban areas in 2006 was

approximately $900 whereas in rural area it was only $550. However, the expenditure

of households in urban area is much higher than those in rural area. The average per

capita expenditure in urban areas was over $600, doubled that of rural areas (see Figure

5.5)

26

Actually, data on poor coffee farmers can be extracted from VHLSS2006 but the sample is very small

which cannot represent poor coffee household’s behaviours.


100

Figure 5.5: Per capita income and expenditure by area in 2006 ($/year)

Source: calculation from VHLSS2006

Rural household income and savings vary slightly among regions. Surprisingly, even

being one of poorest region in Vietnam, rural households in Central Highlands (CH) has

the highest average income as well as saving. The average household’s income and

saving in Central Highlands in 2006 were $1450 and $790, respectively. The higher

saving of household in Central Highlands in 2006 was mainly due to relatively high

profit from coffee in that year. The farm gate price of coffee in 2006 was over $1 per

kg, triple the price level of 200127

Table 5.5: Household income and saving in rural by region in 2006($)28

Region

Poor household Non-poor household

Household

income

Household

saving

Household

income Household saving

Red River Delta 1066 530 2053 868

North East 1170 509 2176 1008

North West 1162 473 1886 738

North Central Coast 1051 487 1880 800

South Central Coast 929 360 2005 759

Central Highlands 1452 790 3343 1905

North East South 1329 674 3261 1368

Mekong River Delta 1234 617 2878 1474

Source: calculation based on VHLSS2006

27

Coffee Farm Survey 2007 28

Vietnam is currently divided into 8 regions: Red river delta (RRD), North East (NE), North West

(NW), North Central Coast (NCC), South Central Coast (SCC), Central Highlands (CH), Northern East

South (NES) and Mekong River Delta (MRD). The regional map of Vietnam is presented in Figure A1 in

the Appendix A.

0

100

200

300

400

500

600

700

800

900

1000

per capita income per capita expenditure

Urban area

Rural area


101

Although having the highest household income and savings, due to the large household

size, the average per capita income of poor households in the rural area of Central

Highlands (CH) was only $ 230. This level is similar to per capita income in Northern

East South (NES) and slightly lower than per capita income of poor households in rural

areas of the two deltas (Red River delta and Mekong River delta).

Figure 5.6: Per capita income and expenditure of the poor in rural areas by

regions, 2006 ($)

Source: calculation based on VHLSS2006

Households usually keep their savings in several different forms. A study by CAP

(2008)29

shows that the main types of total savings are cash (accounting for nearly 40

percent of total saving) and gold, silver (23%). The study also pointed out that

agricultural households generally have the lowest saving level ($526), much smaller

than that of households in the service sector ($927).

Table 5.6: The saving flows of household by types

29

CAP: Center for Agricultural Policy

0

50

100

150

200

250

300

RRD NE NW NCC SCC CH NES MRD

US

D

per capita expenditure

per capita income


102

Total

saving

($)

% by type of saving

Total Assets Deposit Cash

Gold,

silvers…

Holdings,

Others

Household

head’s gender

Male 602 14.5 10.3 38.4 22.6 14.2 100

Female 551 7.3 18.8 37.5 23.9 12.4 100

Household

types

Agriculture 526 18.4 10.3 35.2 22.0 14.1 100

Industry 588 7.7 5.1 51.8 21.8 13.7 100

Service 927 5.4 4.6 42.5 23.0 24.5 100

Gov. officer 787 9.9 8.8 35.9 29.1 16.2 100

Others 696 7.1 31.0 33.6 23.6 4.6 100

Income group

Poorest 143 38.7 0.0 36.7 14.7 10.0 100

2 263 21.4 15.7 37.9 20.3 4.7 100

3 483 19.9 6.3 32.8 20.2 20.7 100

4 648 14.1 9.3 36.0 22.5 18.0 100

Richest 1344 6.9 14.9 41.3 24.9 12.0 100

Average 594.3 13.4 11.7 38.2 22.8 13.9 100

Source: CAP (2008)

5.3.2. Relationship between Income and Expenditure of Poor Farmers

Because of the dynamic nature of the household model used, one needs a model of how

savings are accumulated over time. This requires an understanding of the relationship

between income, expenditure and savings.

The relationship between income and expenditure is estimated using the econometric

method. These estimates are important for the FY-CC model when identifying the

saving level of household, a key factor determines the coffee farmer decision to cut,

keep or replant coffee. As mentioned earlier, data on expenditure and income for poor

coffee farmers are not available. Thus, data on the poor households in VHLSS2006 is

used as a basis for this analysis.

The total number of poor households in VHLSS2006 is 1038 of which 912 households

are located in rural area. Table 5.7 presents the number of poor households extracted

from VHLSS2006 by regions to estimate the relationship between income and

expenditure.


103

Table 5.7: Number of poor households by region in VHLSS2006

Region Frequency Percent

Red River Delta 92 8.86

North East 209 20.13

North West 191 18.40

North Central Coast 175 16.86

South Central Coast 60 5.78

Central Highlands 119 11.46

North East South 58 5.59

Mekong River Delta 134 12.91

Total 1,038 100

Source: Summarised from VHLSS2006

There is expected to be a non-linear relationship between expenditure and income. By

testing different functional forms it was found that a simple linear spline function

provided the best fit, with two sections and a ‘knot’ at $180 per capita (Table 5.8). The

dependent variable is per capita income while household size, income group 1 (less than

$180 per person per year), income group 2 (greater than $180 per person per year) are

explanatory variables. The negative coefficient of family size (-3.53) means per capita

expenditure decreases as the number of people in the household increases, possibly

reflecting economies of size in consumption or a reflection of poverty in particular. The

positive coefficients of both income groups indicate the per capita expenditure will

increase with higher income. However, the much higher coefficient of income group 1

(0.42) compared to group 2 (only 0.02) means that with lower per capita income,

farmers tend to spend a larger share of income. As income increases farmers tend to

save more for other activities. This pattern can be seen more clearly in Figure 5.7 with

the variation of fitted expenditure per person versus per capita income for different

family sizes and Figure 5.8 with observed values.


104

Table 5.8: Regression between per capita income and expenditure of poor HHs

Type of

variables

Variable

name Description Coefficient

Standard

Errors P>t

Dependent

variable

Pcexpend Per capita expenditure

Independent

variables

Hhsize Family size -3.53 0.34 0

Pcincome1 Income per capita if income

< $180, 180 otherwise 0.42 0.03 0

Pcincome2 (Income per capita -180$)

if income > $180, 0

otherwise 0.02 0.01 0.01

_cons Constant term 73.53 4.49 0

Source: Estimated from VHLSS2006

Figure 5.7: Fitted per capita income and expenditure by family size

household size =6

household size=4

household size=2

80

100

120

140

160

fitte

d p

er

ca

pita

expe

nd

iture

($

/ye

ar)

100 200 300 400 500per capita income ($/year)


105

Figure 5.8: Plot of per capita income and expenditure of the poor

Source: based on VHLSS2006

Next section will describe the structure of the Fixed Yield–Cash Constraint Model (FY-

CC model) with its objectives, functions and decision rules.

5.4. Structure of the FY-CC Model

The FY-CC model is a representative cash-constrained farm based on the characteristics

of the household set out in Table 5.9. The average land holding of poor coffee farmers is

5287 m2. The annual average loan and saving of poor households are $625 and $567,

respectively. These values of loan and saving are used as base loan and base saving.

Table 5.9: General data of poor coffee household

Indicators Value Sources

Farm size (m2) 5287 Agrocensus_2006

Non-coffee maize income ($) 200 Thang (2008)

% family labour 30 Thang (2008)

Family size (persons) 4.7 Agrocensus_2006

Average loan (base loan) ($/year) 625 Thang (2008)

Average saving (base saving) ($ per year) 567 VHLSS2006

Replanting cost ($/ha) 1440 Thang (2008)

Note: Non-coffee maize income or other income from other activities of households

50

100

150

per

cap

ita e

xpen

ditu

re (

$/y

ea

r)

0 100 200 300 400 500per capita income ($/year)


106

Before moving to look at the detailed functions and decision rule of the FY-CC model,

it is important to note the main differences between the FY model and the FY-CC

model. The differences include:

First, the FY model identified the optimal rule to maximise the ENPV per hectare of

land. The FY model did not consider other issues such as farm size, expenditure, saving,

credit, other income, and family labour return. The FY-CC model tries to integrate all

the above issues to investigate the optimal rules of resource poor coffee households, so

the FY-CC is a resource poor coffee household model. In the FY-CC model, the

objective function is not the ENPV per one hectare, it is replaced by the ENPV per

resource poor farm (only 5287 m2).

Second, farmers in the FY model are assumed unconstrained by liquidity issues. Thus,

the household’s budget does not restrict decisions to keep or replant the coffee trees.

However, in the FY-CC model, resource poor coffee households may operate under a

cash constraint and this may affect their decision. Thus, decision rules of the FY-CC

model are different from the FY model.

Third, the FY model considered several fixed form equations of age dependent CP

(quadratic, cubic and quadratic with price change effect) solved to find the

corresponding optimal rules and ENPV. However, the results of different age dependent

CP were quite similar. Thus, the FY-CC model assumes a quadratic form of CP (

2

1 2oCP age age). The CP is found infinite in the FY model thus it is ignored

in the FY-CC model

Fourth, the FY model in the previous chapter identified the optimal cutting and

replanting rules for coffee farmers with two price simulation models (Lagged price

model and Price Cycle model). However, the price data simulations generated by the

Lagged price model are more realistic in terms of the underlying stochastic process.

Besides that, the estimated price function using Lagged prices had a more suitable

match with higher R2 and F-value. Thus, this chapter only considers the case where

prices are generated using the Lagged price structure.

The following sections will describe the main functions in the FY-CC model such as the

objective function, profit function, yield and production cost function, expenditure and


107

saving, decision rules. The general structure of the model is similar to the FY model, but

there are significant differences.

5.4.1. Objective Function

The FY-CC model aims to identify the optimal rules to maximise the ENPV for a

resource poor farm. Thus, the objective function of the FY-CC model is the same as the

objective function of the FY model in (4.1) and (4.2).

The ENPV function is given by:

22

1

1

22a

a

ENPV ENPV

(5.1)

in which

, ,

1 1

1( )

N Te t T

a t r a

r t

ENPV V TN

(5.2)

where aENPV is expected NPV given coffee trees at starting age a , for the next 50

years given N possible random price sequences; , .

e

t r a is profit per ha in year t for price

trajectory r and starting age a ; and is the discount factor. V T denotes the

terminal value of the coffee garden and it is set to zero in the model; and r identifies

the replication number for one age group. As same as in the FY model, one hundred

replications are employed for each starting age group. For each replication, a separate

random price trajectory from the Lagged Price model is simulated.

The only difference in the objective function between the FY model and the FY-CC

model is the size of land area managed by farmers. In the FY model, it is assumed that

the farmer controls one hectare of land and the planning decision applies to one hectare.

However, the poor coffee farmers in the FY-CC only control 5287 m2

(the average

coffee area of poor coffee farmers). Thus, the ENPV in the FY-CC model is achieved

for that farm size.


108

5.4.2. Profit, Yield and Production Cost Function

The yield and production cost function in the FY-CC model are the same as those used

in the FY model. The coffee yield and production cost are fixed for any given age of

coffee tree but variable with age. In reality, resource poor coffee farmers usually own

the less productive coffee land so the yield of poor households is generally lower than

that of non-poor households. However, the FY-CC model assumes no yield difference

between poor and non-poor households. With this assumption, it makes the comparison

of results between the FY model and FY-CC model possible.

The profit function in the FY-CC model here has a very small change with the inclusion

of the “other income” component.

The profit of coffee poor household at time t is as follows:

, , [( . ). ( . ).(1 )]e c c c m m m

t r a t t t t t t t t oP Y v S P Y v S I (5.3)

where tS denotes for the existence of coffee. tS is equal to1 if coffee is growing and 0

otherwise; c

tP is price of coffee at year t; c

tY is coffee yield at year t; m

tP is price of

maize at year t; m

tY is coffee yield at year t; oI is the other household income that is

besides profit from coffee and maize.

In common with the FY model, in the FY-CC model the profit of maize and other

income are assumed constant. The profit of maize is fixed at $440 per hectare or $232

per poor farm. The other income of poor coffee households is estimated by Thang

(2008) and is fixed at $200 per year.

5.4.3. Expenditure, Saving and Loan

The household expenditure (HE) is estimated from income based on the regression

result in Table 5.8. HE is given by:

HE = hhsize (73.53 -3.53hhsize +0.42pcincome) if per capita income

of household is less than $180 per year

(5.4)

HE = hhsize (73.53 -3.53hhsize +0.42*180 +0.02(pcincome-180) if

per capita income of household is greater than $180 per year

(5.5)


109

where hhsize is number of persons in household; pcincome is per capita income of poor

household.

The annual saving of the household is measured as the difference between household

income and expenditure. Thus, aggregate savings available in year t will be given by:

t t ă t tSaving Saving HE (5.6)

Farmers cannot invest in a new coffee garden if they do not have enough money, even if

the expected profit from the investment is positive. In general, farmers can mobilize

capital from two sources: household saving and loans. Thus, the capital function in the

model is given by:

Capitalt = Savingt + Loant (5.7)

According to Coffee Farm Survey 2007 in Dak Lak province, the average loan of coffee

farmer was $837. This level is slightly larger than the base loan borrowed by poor

coffee household in the FY-CC model ($625 per year). Farmers in Cu Mgar district

borrowed larger amounts because they have a larger farm size and more productive

land. Most loans have a term of one or two years (see Table 5.10).

Table 5.10: Loan amount and duration

District Average Loan ($) Duration (months)

Cu Mgar 921.5 18.3

Krong Pak 848.5 16.2

Eakar 741.1 13.1

Overall 837.0 15.9

Source: Coffee Farm Survey 2007

The main proportion of loans is used for buying inputs and hiring labour (over 70%). In

Cu Mgar district, farmers spend 90 percent of loan value for input purchases and labour

payment. Only 2 percent of loans borrowed by farmers in Eakar are used for hiring

labour. This can be explained because farmers in Eakar are quite poor so they employ

mainly family labour. About 17 percent of loans are spent for other economic activities

of households (see Table 5.11).


110

Table 5.11: Main loan purpose (% respondent)

Loan purpose Cu Mgar Krong Pak Eakar Total

Buy inputs 82.0 57.5 66.6 69.8

Hire labour 7.6 6.0 2 6.3

Other economic activities 10.2 27.2 14.2 16.2

Other 0 9.1 17.0 7.5

Total 100 100 100 100

Source: Coffee Farm Survey 2007

The main sources of loans for farmers are banks. According to the Coffee Farm Survey

2007, over 70 percent of loan value was borrowed from banks. In Krong pak, this

number reached nearly 85 percent. Private lenders are also important credit providers. In

Eakar, over 50 percent of loans came from private lenders.

Table 5.12: Percentage of loan by different sources by districts

Sources of loan Cu Mgar Krong Pak Eakar Total

Banks 76.9 84.8 42.8 72.0

Private lenders 2.5 0 52.3 12.9

Relatives 10.2 0 4.7 5.3

Women's association 0 3.0 0 1.0

Commune Committee 0 9.0 0 3.2

Other support

programs 10.2 3.0 0 5.3

Total 100 100 100 100


5.4.4. Decision Rule

The decision rule for deciding when to replant/cut coffee in the FY-CC model is slightly

different from the FY model. To describe the decision rule, St is denoted for the coffee

area at year t. Whether coffee trees exist on the land or not is denoted by a (0, 1)

variable. Thus, St is equal to 1 if coffee is planting, otherwise it takes 0. The decision of

poor coffee household in the FY-CC model is as follows:

(i)

St = 1 if St-1=1 and Pt > min [(CP, RP) and (Capitalt + Io>= Costt+1 + Minexpend)]

where Io is other income of household, Minexpendt is the minimum expenditure for

household in year t and it is about $270. The minimum expenditure is estimated from

regression results of income and expenditure in Table 5.8; Pt is the price of coffee,

Costt+1 is production cost in year t+1, Capitalt is capital of household in year t

(Capitalt= savingt + loant)


111

This condition means that if farmers are growing coffee, they will keep coffee if price is

greater than the minimum level of CP and RP and the household’s budget can at least

cover the minimum household expenditure and the production cost in the coming year.

(ii)

[St = 0 if St-1=1 and Pt ≤ min (CP, RP)] or

St = 0 if St-1=1 and [Pt > min (CP, RP) but (Capitalt + Io< Costt+1 + Minexpend)]

This constraint indicates that farmers will cut coffee and switch to maize if (i) price is

below the minimum level of CP and RP or (ii) despite price being higher than the

minimum level of CP and RP if the household’s budget cannot cover the minimum

household’s expenditure and production cost in next year.

(iii)

St=1 if St-1=0 and Pt>=RP and (Capitalt + Io< replanting cost + Minexpend)

This relationship gives the condition for replanting coffee if farmers are growing maize.

The farmers grow coffee again if price is above the RP and the household budget can at

least cover the minimum expenditure and replanting cost.

As mentioned earlier, the FY-CC model is limited to only the quadratic CP. Thus, the

fixed-form law for cutting price in the model will be defined as follows:

2

1 2oCP age age (5.8)

The replanting price is again specified as:

RP = 3 (5.9)

To find the optimal CP and RP, it is necessary to solve the FY-CC identifying

1 2 3, , ,o for the maximum ENPV.

5.5. Results of the FY-CC Model


112

5.5.1. Impact of Cash Constraints on Income

Before re-solving for the optimal cutting/replanting rules under cash constraints, it is

informative to identify the extent to which the poor farmers lose income because of cash

constraints if they still apply the optimal rules of the FY model as in the previous

chapter. Recall, the optimal rules in the FY model are identified as follows:

CP = 0.402 - 0.0509age + 0.00367age2 (5.10)

RP =0.74 ($/kg of coffee bean) (5.11)

With optimal rules, the maximum ENPV per hectare from FY model is $9226. This

number falls to $4878 if farm size is reduced from one hectare to the average farm size

of poor coffee households (5287 m2).

Now, this rule applies to household in the FY-CC model in conjunction with

characteristics of poor coffee households presented in Table 5.9. The result shows that

the maximum ENPV of farm income is about $4300 per poor farm size, about 15

percent lower than the ENPV achieved by the FY model for the same coffee area (see

Figure 5.9). The income reduction is due to cash shortages that prevent farmers

replanting coffee even if the current coffee price is greater than the RP ($0.74 per kg).

Furthermore, income is also reduced because in some simulations farmers may not have

enough money to sustain both their production and living costs so they have to cut and

switch to maize.

Figure 5.9: ENPV from FY-CC if imposing optimal rule of FY and optimal ENPV

from FY ($/poor farm)

4000

4100

4200

4300

4400

4500

4600

4700

4800

4900

5000

ENPV from FY-CC if imposing optimal rule of

FY model

Optimal ENPV from FY at poor farmsize

Exp

ecte

d N

PV

($)


113

5.5.2. Effect of Loans and Savings

The income loss when applying the optimal rule of the FY model into the FY-CC model

shows the important role of saving and credit to support farmers. To investigate the

importance of saving and loan on income, the FY-CC model can be solved by changing

the saving and loan levels while still keeping the optimal rules of the FY. Figure 5.10

presents the results for different levels of initial savings, holding all other factors

constant. As expected, with high initial savings levels, the maximum ENPV from the

FY-CC model is close to the ENPV from the FY model. However, as initial savings fall,

so does ENPV. The results of the FY-CC model give an interesting point with respect to

the contribution of initial savings in the range of $500-$1100. It would appear that in

this range, changes in initial savings have little impact on ENPV, suggesting threshold

effects within the model. The saving range of $500-$1100 cannot cover the production

cost of coffee in non-productive period and other living expenditure of household. Thus,

it may not help the poor people optimize their investment decision, so the ENPV

remains unchanged.

Figure 5.10: ENPV from FY-CC at different initial savings at annual loan of 625$

Similarly, to investigate the impact of limits on credit on income, the FY-CC model is

solved with a fixed initial saving value and exploring the impact of changes in the

maximum level of annual loan. The results are depicted in Figure 5.11. As shown in

Figure 5.11, the increase in annual loan strongly improves ENPV if farmers do not have

any savings. The ENPV increases from $3300 (with loan = 0) to about $4400 (with

4200

4300

4400

4500

4600

4700

4800

4900

0 500 1000 1500 2000 2500

EN

PV

($)

initial saving ($)

loan =$625


114

annual loan of $1100). The growth of ENPV is much lower once the loan is over $1200

and mostly unchanged with loans over $2500.

Figure 5.11: ENPV of farm income at different annual loans and savings

With initial savings of $ 567 (base saving), the increase in loan has much less impact on

farm income. The ENPVs with two different initial saving levels converge when the

available loan is over $2500. This means the initial savings does not affect ENPV once

the annual loan exceeds $2500 and if farmers can get enough credit they will achieve

the optimal decision regardless of the initial saving.

The small increase in ENPV with base saving ($567) explains the way in which ENPV

is calculated. The ENPV here is the mean of ENPV from 22 starting age groups.

However, it is anticipated that the importance of savings will vary according to age of

coffee trees. Figure 5.12 reports the results for ENPV over a 50-year horizon time for

farmers with different initial ages of coffee trees. As shown in Figure 5.12a, the ENPV

with different annual loans varies significantly by the starting ages of coffee trees. For

farmers who have just replanted coffee trees, loans under $500 do not help them change

their income. More interestingly, even with a loan from $500 to $1100, the average

income of farmers tends to be low. This can be explained by the fact that with this level

of loan, farmers can only afford to replant coffee but they do not have enough money

for keeping coffee in the following years. The amount of loan is effective for farmers

who want to replant coffee when it is over $1100. With over $1100, household’s farm

income increases steadily and nears the maximum level at a loan of about $2000.

3000

3200

3400

3600

3800

4000

4200

4400

4600

4800

0 500 1000 1500 2000 2500 3000 3500

annual loan ($)

Avera

ge N

PV

($)

NPV with initial saving = 567$

NPV with initial saving =0


115

With the starting age of 2-year old coffee trees, the variation of ENPV by loans is less

complicated. After reducing slightly with loan under $400, average ENPV increases

quickly with loan amounts and get close to the maximum at loan of $1500. The results

in Figure 5.12a also indicate that the ENPV of farm income with starting age of coffee

from 3 to 5 year old are getting higher with bigger loans. However, the amount of loans

that helps income to reach maximum level at different starting ages is not similar. For

the farmers with trees of an initial age of 5, the amount of loan does not significantly

affect the ENPV. This suggests that once trees are established the income stream from

coffee is sufficient for even poor farmers to follow the optimal decision rules.

The variations of ENPV for starting age of trees in the mature period (from 6 to 15) are

quite similar and increase slightly with higher loans. However, the higher amount of

loan does not produce a large impact on farm income for those groups. This happens

because with mature coffee garden, farmers may have a good income that helps them

save enough for keeping coffee or replanting new trees (see Figure 5.12b).

However, when trees are in the last years of the life cycle with downward yield, the size

of loan becomes more important and has significant effects on farm income. Figure

5.12c presents the change in ENPV by loan for different starting ages from 17 year old

to 22 year old. With trees from 20 to 22 year old, the impact of loan is quite similar to

young trees in gestation period. The ENPV rises with greater credit and the role of loans

are much more significant when trees are getting older. This is because, at this stage in

the cycle, landholders are approaching the point at which they will replace trees, and if

they have not built sufficient savings over the productive portion of the trees lifecycle,

they will find it difficult to re-establish the coffee trees, if the replanting rule suggests

that that is appropriate.


116

(a) with starting age from 1 to 5 year old

(b) with starting age from 8 to 12 year old

(c) with starting age from 17 to 22 year old

Figure 5.12: ENPV of FY-CC with different starting age of trees and loans

0

1000

2000

3000

4000

5000

6000

7000

020

040

060

080

010

0012

0014

0016

0018

0020

0022

0024

0026

0028

0030

00

annual loan ($)

avera

ge N

PV

($)

starting age of trees =5 years old

starting age =4

starting age =3starting age =2

starting age =1

5000

5200

5400

5600

5800

6000

6200

6400

0 200 400 600 800 1000 1200 1400 1600 1800 2000

annual loan ($)

Av

era

ge

NP

V (

$)

starting age =12

starting age of tree =8

starting age =9

starting age =10

starting age =11

2000

2200

2400

2600

2800

3000

3200

3400

3600

3800

4000

0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600

Annual loan ($)

Ave

rag

e N

PV

($) starting age =19

starting age of trees =17 years old

starting age =18

starting age =20

starting age =21

starting age =22


117

5.5.3. Optimal Rule for Poor Coffee Farmers

The previous analysis has taken the estimated optimal decision rules, estimated without

liquidity constraints, and simulating behavior in the presence of such constraints.

However, it is also of interest to see if the behavioral rules themselves would change if

the fixed form optimization process were repeated, with explicit consideration given to

the presence of liquidity constraints.

To find the optimal rules for poor coffee households, the FY-CC model repeats the

searching procedures for the FY model.

As mentioned earlier, The FY-CC model only investigates the optimal rule for poor

coffee households with the Lagged price simulation model and quadratic form of CP.

With application of the grid searching procedure, the FY-CC identifies the optimal rules

for poor coffee households as follows:

CP =0.458 -0.46age +0.00325age2 (5.12)

RP=1.4 ($/kg) (5.13)

ENPV =4375 ($) (5.14)

The optimal rules of the FY-CC model are illustrated in Figure 5.13. The optimal CP

for one year old trees is about 0.4($/kg). The CP reduces slightly when the age of trees

is getting close to their mature period. After that, the CP increases gradually with older

ages.

The comparison of optimal rules between the FY model and the FY-CC model is

presented in Figure 5.14. As shown in the Figure 5.14, the CP of poor coffee growers is

generally higher than CP in the FY model. This means that poor households in the FY-

CC model are more likely to cut compared to farmers without a cash constraint in the

FY model. This may happens because of cash shortage for buying inputs and for

household consumption. That is why the poor farmers have to cut earlier and switch to

maize.


118

Figure 5.13: Optimal Rules of the FY-CC model

Figure 5.14: Optimal rule of FY model and FY-CC model

The optimal RP in the FY-CC model ($1.4 per kg of coffee bean) is much higher than in

the FY model ($0.74 per kg of coffee bean). This suggests a much more cautious

approach to replanting which is logical: because of the period when trees are

unproductive, it is highly disadvantageous to plant trees and then remove them within a

couple of years because of cash flow shortages. Thus, the poor farmers usually wait for

significantly higher prices before making replanting decisions.

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1 3 5 7 9 11 13 15 17 19 21

age of trees

pri

ce

($

/kg

)CP of FY-CC model

RP of FY-CC modelkeep coffee

cut and switch to maize

replanting

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

0 3 6 9 12 15 18 21

age of trees

pri

ce (

$/k

g)

CP of FY model CP of FY-CC model

RP of FY model RP of FY-CC model


119

Due to the cash constraint, poor farmers in the FY-CC model cannot make the optimal

decision as in the FY model. Thus, the optimal ENPV of the FY-CC model is only

$4370 per poor coffee household size (5287m2). This value is still much lower than

optimal ENPV of the FY model ($4878) at same area of coffee.

With optimal rules, the actual cutting percentage of coffee farmers in the FY-CC model

is quite different from the FY model. In general, poor farmers in the FY-CC model are

more likely to cut than coffee farmers in the FY model, especially when the trees are

still young (less than 4 year old) and very old (over 20 year old). The cutting frequency

is not much different when trees are 6-19 year old (see Figure 5.15).

Figure 5.15: Actual cutting percentages by age of trees

The higher cutting frequency of poor coffee farmers can be derived from the optimal

cutting rule and cash constraint. Hence, it is much better to split the impact of those two

factors on cutting decision of poor coffee farmers. Figure 5.16 below presents the

separated cutting percentage by age of coffee trees under the impact of CP rule and cash

constraint. When trees are young (less than 4 year old), cash constraint influences

significantly on cutting decision of poor farmers. However, when trees are older, the

cutting decision is mostly not affected by cash constraint.

0

5

10

15

20

25

30

1 3 5 7 9 11 13 15 17 19 21

age of coffee trees

% a

ctu

al c

ut

FY model

FY-CC model


120

Figure 5.16: Impact of cutting decision by CP rule and by cash constraint in the

FY-CC model

To investigate the change in the optimal rules of the FY-CC when initial savings varies,

a simulation in which $1500 replaces the base initial saving ($567) is solved. The model

output shows that with higher initial saving, poor coffee farmers are less like to cut and

more likely to replant than with the base initial saving. Furthermore, the new maximum

ENPV increases to $4712. This level is higher than the maximum ENPV from the FY-

CC model with base saving and near to optimal ENPV of the FY model with poor

farms. In addition, the optimal CP in the FY-CC model with initial saving of $1500 is

very close to the optimal CP in the FY model. However, the optimal RP in the FY-CC

model ($1.07 per kg of coffee) with initial savings of $1500 is still relatively high

compared to the FY model (RP=$ 0.74) (see Figure 5.18).

0

5

10

15

20

25

30

1 3 5 7 9 11 13 15 17 19 21

age of coffee trees

actu

al cu

ttin

g (

%)

by CP rules

by cash constraint


121

Figure 5.17: Optimal rule of the FY-CC model with different initial savings

Figure 5.18: Optimal rules of FY model and FY-CC with initial saving of $1500

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0 3 6 9 12 15 18 21

age of coffee trees

co

ffe

e p

ric

e (

$/k

g)

CP with initial base saving CP with initial saving of $ 1500

RP with initial saving of $ 1500 RP with initial base saving

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

0 3 6 9 12 15 18 21

age of coffee trees

price (

$/k

g)

CP of FY-CC, initial saving=$1500 RP of FY-CC, initial saving=$1500

CP of FY model RP of FY model

RP =1.07

RP = 0.74


122

5.6. Conclusion

Rural households in general and coffee farmers in Vietnam have volatile and low

incomes. Many coffee farmers are poor and they often have cash constraints on

investment. Due to the cash shortage, poor farmers may not be able to use their own

assets as efficiently as the non-poor’. They are disadvantaged when trying to optimize

their decisions. This is a common finding in other developing countries where

agricultural investments tend to rely heavily on credit from banks and other

organizations.

The FY-CC model in this chapter investigates the changes of coffee farmer’s decision

when they face a cash shortage. The model found that the poor coffee farmers could not

be optimal compared to whose without a cash constraints. Compared to the FY model,

due to cash constraint, poor coffee households in the FY-CC model have to cut coffee

earlier and switch to maize. Furthermore, they replant coffee at a higher price as

compared to the FY model.

Savings and loans play an important role in helping farmers allocating their resource

optimally. In general, poor coffee households can get higher income with access to

larger loans. However, the importance of loans and saving varies according to the initial

age of coffee trees. The amount of loan plays a more important role in improving

household’s income when trees are younger. For farmers who have just replanted coffee

trees, loans under $1000 do not help them change their income but the ENPV increase

considerably with loans above that. The amount of annual loan is also important for

starting age of trees under 4 years old. With the initial age of trees in mature period, the

increase in loans does not have a substantial impact on household income.

The result of the analysis of actual cutting decisions when optimal rules are invoked

show that poor farmers in the FY-CC model are more likely to cut than the coffee

farmers in the FY model, especially when the trees are less than 4 year old and over 20

year old. The cutting frequency is not much different when trees are in the 6-19 year old

group. The model result also indicates that the liquidity constraint affects considerably

on the cutting decision for young trees in the gestation period.

The FY-CC model identifies the change in the decision rules needed when considering

poor coffee household with the cash constraint. The impacts of savings and loans on


123

different coffee groups are also investigated using the FY-CC model. However, in both

the FY model and the FY-CC model, the cost and yield of coffee are fixed according to

the age of trees. In practice, coffee yield is influenced by input use (including labour). In

addition, farmers often change their input application in response to the output price.

The response of production cost to output price and the yield response to input use are

the possible short-run responses of farmers. The presence of short-run responses reflects

closely to coffee farmer’s behavior in practice. The decision of farmers and the ENPV

may change considerably with the appearance of short-run responses. In the next

chapter, the yield function of coffee will be estimated based on the age of coffee trees

and the production cost. After that, the estimated yield function will be integrated into

the FY and the FY-CC model to see how farmers’ decisions change with the possibility

of the short-run response.

Chapter 6. Short-run Response and Optimal Rules for Coffee Farmers in Vietnam

124

CHAPTER 6. SHORT-RUN RESPONSE AND OPTIMAL

RULES FOR COFFEE FARMERS IN VIETNAM

6.1. Introduction

This chapter investigates the farmer’s decision when it is possible to make short-run

‘tactical’ changes to coffee yields by adjusting input levels. In the previous models FY

and FY-CC, the farmer’s decision merely determines the coffee area. In this chapter,

the farmer adjusts variable inputs within the production season. However, the response

to yield, cost and price in the short-run, can also influence the optimal decision of

farmers in terms of cutting and replacement rules. Thus, input use is included as a

decision variable in the simulation model. The model is then re-solved to identify

optimal rules and farmer’s choices under an extended choice set. To this end, this

chapter develops two extended models: Variable Yield Optimal Model (VY model) and

Variable Yield- Cash Constrained Model (VY-CC model)

This chapter consists of five sections. Section 6.2 will review previous studies on yield

response functions. Section 6.3 will estimate the relationship between coffee yield and

variable inputs in Vietnam. An analysis of the supply elasticity with respect to price

based on the estimated yield function is reported in Section 6.4. The optimal rule and

income of coffee farmers under these conditions will be analyzed in Section 6.5 and

Section 6.6.

6.2. Review of Literature on Yield Response Functions

There have been numerous studies on crop yield response functions. Much of the work

has been focused on the best functional form to identify crop yield response to fertilizer

as well as using these models to identify the optimal level of fertilizer (Reeder and

McGinnies, 1989, Wight and Godfrey, 1985, Ackello-Ogutu et al., 1985, Taylor and

Swanson, 1973, Mendelssohn, 1979, Rajsic et al., 2009).

The majority of studies have applied polynomial functions (quadratic or square root) to

represent the relationship between fertilizer and yield (Mendelssohn 1979; Reeder and


125

McGinnies 1989; Rajsic et al. 2009). However, other researchers have pointed out the

inappropriateness of polynomial crop response functions because they allow

substitution between nutrients and tend to overestimate the maximum yield and optimal

fertilizer application (Anderson and Nelson, 1975). Thus, some have applied different

forms when estimating the response of the crop yield. For instance, Anderson and

Nelson (1975) used linear-plateau models to estimate a Tennesse corn yield function in

North Carolina. Similarly, Tembo et al (2008) developed a method of estimating a

response function with a stochastic plateau that can capture random effects and

determine economically optimal levels of nitrogen fertilizer for wheat in the Southern

Great Plains of the United States. Ackello-Ogutu et al. (1985) used the von Liebig

response function instead of a polynomial function to analyze fertilizer and yield

response for corn, soybean, wheat and hay rice using data from a thirty year experiment

conducted on the agronomy farm at Purdue University (USA).

Water and fertilizer management are crucial to high yields. Irrigation water is becoming

an increasingly limited resource in many areas in different countries, and as a result, an

appropriate choice of irrigation is needed. Furthermore, optimal combinations of

fertilizer and water can increase crop yield and reduce groundwater pollution. Thus,

previous papers have tried to estimate the relationship between yield and fertilizer

levels, in combination with irrigation (Di Paolo and Rinaldi, 2008, Reid et al., 2002,

Pandey et al., 2000, Lovelli et al., 2007) or evaluate the impact of water deficit as well

as irrigation method on crop yield (Dagdelen et al., 2006, Oktem, 2008, Karama et al.,

2003, Pandey et al., 2000, Panda et al., 2004, Topcu et al., 2007, Melgar et al., 2008,

Jalota et al., 2009, Karam et al., 2009, Kunzová and Hejcman, 2009, Li et al., 2009).

Literature on measurement and effects of fertilizer-water use efficiency are reported in

Zwart and Bastiaanssen (2004) and Oktem (2008).

In agricultural production, the effects of input use can carry over from season to season.

Input carryover effects have been included in a number of studies. Akbar (2003) used

field studies to estimate residuals of input (N, NPK) from cereal and legume cultivation.

Segarra (1989) applied a dynamic optimization model in which an intertemporal nitrate-

nitrogen residual function was used to derive and evaluate nitrogen fertilizer optimal

decision rules for irrigated cotton production in the Southern High Plains of Texas

(United States). Ackello-Ogutu (1985) estimated an econometric model for phosphorus

carryover in United States based on the geometric distributed lag form, prices and yields

of hay, wheat and corn.


126

Studies on yield response have been extended to include the impact of weather and price

risks. Rötter and Van Keulen (1997) analysed risks and opportunities of small farmers

in Kenya when they assessed the variation in yield response to fertilizer application for

maize. The risk assessment approach in this paper is based on crop growth modeling in

which risk is assessed based on yield probability distributions, product prices, costs of

inputs and the level of the most important economic and environmental risks.

Stochastic weather and soil conditions explain why farmers tend to apply more than the

recommended levels of nitrogen. Rajsic et al (2009) examined the effect of temporal

uncertainty on the optimal level of nitrogen application for both risk-neutral and risk-

averse corn producers in Haldimand-Norfolk County, United States. The authors found

that uncertainty plays a role in the application decision of farmers but not in the manner

typically assumed. While uncertainty can justify farmers applying more than the

recommended inputs for risk neutral farmers, it does not for risk-averse farmers.

Most studies on yield response to inputs have focused on annual crops. The number of

papers looking at the response of perennial crop is limited. Salardini (1978) investigated

the response of tea to fertilizer in Iran. However, this study is only based on

experiments on different sites with a single treatment of fertilizer. Similarly, Cong

(2001) estimated the response of some crops (rice, coffee, cabbage, and rambutan) on

different types of land in the south of Vietnam. This study identified that the coffee

yield increases with nitrogen application as well as potassium. However, this study only

did an experiment with three levels of input application, and it did not estimate a

relationship between yield and fertilizer. Garcia and Sively (2001) used DEA method to

measure the technical efficiency of coffee producers in Daklak province, Vietnam. They

studied the effect of different inputs on technical efficiency and found that 30% of farms

are identified as efficient under an assumption of CRTS (Constant returns to scale) and

39% are identified as efficient under an assumption of VRTS (variable returns to scale).

The following section reports a yield response function of coffee based on Vietnamese

Agrocensus_2006 data. After estimation, the yield function is incorporated into the FY

model and the FY-CC model to investigate whether the possibility of a short-run

response changes the optimal cutting/replanting decisions, and incomes. In previous

models (FY, FY-CC), yield of coffee trees is assumed to be constant for a given age.

However, in this model (VY, VY-CC), the yield of coffee trees will be determined as a

function of production cost.


127

6.3. Coffee Yield Function in Vietnam

6.3.1. Yield Coffee Function Estimation

This section estimates a coffee yield function based on two explanatory variables:

annual production cost and age of trees. Unlike the yield response functions for inputs

(such as fertilizer inputs or water) reviewed in Section 6.2, this study does not use yield

response as a function of physical inputs (nitrogen, potassium or phosphate, and labour

or water), but instead uses variable input costs as an index of aggregate inputs. To

investigate the impact of short-run response on coffee farmer’s behaviors and ENPV,

the assumption is that coffee farmers will optimize input use to maximise profit, and the

relationship between inputs and yield will be incorporated into the FY model and the

FY-CC model to develop the VY model and the VY-CC model, respectively.

The data used for estimating the coffee yield function are based on a sub-set of the

coffee production efficiency survey reported in Agrocensus_2006. In the data set, 500

coffee farmers in four provinces in the Central Highlands were interviewed. The sample

of this survey is presented in Table 6.1. Variables in the data set relate to coffee area,

age of coffee, production cost, selling price and quantity sold.

Table 6.1: Sample distribution of coffee households in Agrocensus_2006

Province Number of surveyed

coffee households*

Average coffee

area/household (m2)

Average age

of trees

Average

sale price

($/kg)

Dak Lak 200 9471 7.6 1.0

Kon Tum 100 12894 9.7 1.1

Gia Lai 100 12080 10.2 1.0

Lam Dong 100 7541 8.5 1.1

Source: calculated from Agrocensus_2006

A number of potential functional forms could be used for the relationship. One

restriction is that the functional form has to be amenable to the solution for the optimal

input use, and to be easily implemental within the optimization model. The quadratic

form has a satisfactory fit (high R2, F_value and the significant level of estimates) while

allowing a simple expression for the optimal input use, conditional upon coffee price.

The estimated equation takes the form of:

2 2

0 1 2 3 4 i iq C C A A D (6.1)


128

where q is the yield of coffee (kg per ha), C is production cost of coffee per ha ($), A

denotes the age of coffee trees and Di is dummy variable for district i30

.

The regression results are presented in Table 6.2. As shown in the table, R2 value is

quite high, 75 percent. The negative sign of cost per ha squared and age squared give a

concave function. Coefficient values of the region dummy variables are highly

statistically significant. This means that coffee yield varies among selected districts.

Table 6.2: Regression Results of Coffee Yield Function

Number of obs : 494

F( 16, 477): 91.81

Prob > F : 0.0000

Adj R-squared : 0.75

Variable Coefficient T-value

Dependent variable

yield kg/ha

Independent variables

District Dummies

Kon Tum, Dak To -462.79 -4.88

Kon Tum -121.71 -1.35

Kon Tum, Dak Ha -13.74 -0.14

Gia Lai, Ia Grai -303.81 -3.04

Gia Lai, Chu Se -543.16 -4.99

Dak Lak, C M'gar -403.14 -4.41

Dak Lak, Krong Buk 367.58 4.35

Dak Lak, Krong A Na -408.33 -4.21

Lam Dong, Da Lat city -213.71 -2.03

Lam Dong, Lam Ha 12.41 0.13

Lam Dong, Di Linh -126.23 -1.04

Lam Dong, Bao Lam -13.07 -0.11

C 2.55 12.17

C2 -0.00043 -5.84

A 62.05 2.41

A2 -2.60 -2.18

_Cons -665.72 -3.63

Source: Estimated from Agrocensus_2006 data

Note: C is annual cost per hectare of coffee ($), C2 is square of cost per hectare, A is age of

coffee trees and A2 is square of age of coffee trees.

30 The extension of (6.1) by adding the interaction term between age of tree and

production cost (CA) was also tested but was not statistically significant.


129

To see the impact of age of trees on coffee yield, the cost is fixed at the average cost

from Agrocensus_2006 ($1230 per ha). The relation between coffee ages and yield is

presented in Figure 6.1. In the early years, after the gestation period, coffee yield

increases as trees get older. The coffee trees reach a maximum yield at 11 year and then

reduce gradually. However, the variation in the coffee yield by age is not large

(although statistically significant). The difference of yield by age in the estimated yield

function is much smaller than that in the yield function used in the FY model in Chapter

4 and the FY-CC model in Chapter 5, which was based on judgments from a coffee

expert focus group and Coffee Farm Survey 2007. A change in yield by coffee age is

small, possibly because the age of trees in the sample are mainly mature, with only a

few households with young and very old trees.

Figure 6.1: Coffee yield – age relationship at average cost

Figure 6.2 illustrates the cost-yield relation for 11-year old trees. According to the

estimated results, the yield of coffee increases as variable inputs increase and reaches a

maximum yield of 3600 kg per ha when the production cost is approximately $3000 per

ha. After reaching the maximum level, coffee yield tends to reduce. However, the

maximum cost in the sample is $2880 per hectare, implying that, within the data range,

there is a positive relationship between yield and inputs.

2100

2150

2200

2250

2300

2350

2400

2450

2500

2550

2600

0 5 10 15 20 25

age of coffee

yie

ld (

kg

/ha)


130

Figure 6.2: Cost –yield relation for 11 year old coffee trees

The result of the estimated yield function will be used (with some modification) in the

farm model to investigate changes in coffee farmer’s behavior and income if they have

optimal responsiveness to price. The next section will identify the optimal variable cost.

6.3.2. Optimal Cost Specification by Output Price

This section estimates the optimal cost as a function of price based on the yield function

estimated in the previous section.

Starting from the yield equation based on cost per hectare of coffee land and age

2 2

0 1 2 3 4 i iq C C A A D

The profit of coffee per ha is the difference between revenue and production cost

pq C

Profit reaches a maximum level when its first derivative equates to zero, and so

(6.2)

In the estimated coffee yield model, 1 =2.55 and 2 =0.00043. Figure 6.3 graphs the

cost and yield relationship base on the estimated function for Ia Grai district (Gia Lai

0

500

1000

1500

2000

2500

3000

3500

4000

4500

0 1000 2000 3000 4000 5000

cost (USD/ha)

yie

ld (

kg

/ha

)

339

2883

range of surveyed cost

1

2

1

2optimal

pC

p


131

province) and the price of coffee in 2006 for 11-year old trees. As shown in the figure,

farmers can get the maximum yield of 3600 kg per ha when they invest a cost of about

$2800.

Figure 6.3: Simulation of cost and yield relationship (age of tree =11 year old,

medium yield level district)

6.3.3. Supply Price Elasticity

To get the supply price elasticity, optimal cost in (6.2) is substituted into the estimated

yield function:

2 2

0 1 2 3 4q C C A A (6.3)

2

21 10 1 2 3 4

2 2

1 1

2 2

p pq A A

p p

22 1

0 3 4 2

2 2

1

4 4q A A

p

'

( ) 3

2

1

2Pq

p (6.4)

'

( ) 2

2

1

2

p

q P

pq

q p q (6.5)

With the average sale price and average yield from data set of $1.05 per kg and 1932.8

kg per ha respectively, the supply price elasticity is 0.54. This means that if price

0

500

1000

1500

2000

2500

3000

3500

4000

4500

0 500 1000 1500 2000 2500 3000 3500 4000 4500

Cost per ha (USD)

Yie

ld (

kg

per

ha)

Optimal yield

Maximum yield

Surveyed yield


132

increases by 1 percent, yield of coffee increases by 0.54 percent. The relationship

between price and yield of coffee is given in Figure 6.4. As p increases, the price

elasticity is getting smaller. This happens due to a declining marginal yield.

Figure 6.4: Simulation of price and coffee yield

6.4. Variable Yield Model (VY model)

6.4.1. Model Structure

The model structure of the VY model is the same as the FY model presented in Chapter

4. The objective function, profit function and decision rule are unchanged. The VY

model as well as the VY-CC model in this chapter only investigate the optimal rules

with lagged price simulation and the quadratic CP (2

1 2oCP age age ).

The only difference between the FY model and the VY model is the definition of the

yield and cost function. As presented in the FY model, yield and cost depends only on

the age of coffee trees and these functions derive from the Coffee Farm Survey data and

estimation of coffee experts. Figure 6.5 repeats the yield-age function used in the FY

model. However, in the VY model, the yield of coffee is a function of optimal

production cost and age of trees in which optimal cost is a function of coffee price as

shown in (6.2).

price

Yield

Ymax


133

Figure 6.5: Coffee yield by age of tree in FY model

However, there is a difference between the estimated yield function in Table 6.2 and

yield in Figure 6.5. Thus, to compare farmer’s behavior for optimizing profit in both

cases with and without short-run response, it is necessary to adjust the estimated yield

function to make it consistent with yield function in the FY model. The next section will

present the adjustment of the yield function used in the VY model. In addition, the

optimal cost in the VY model is a function of price.

6.4.2. Adjustment of Yield function

The estimated yield function from Agrocensus_2006 is different from the yield function

used for the FY model in Chapter 4. To compare farmer’s optimal decisions in the FY

model (where yield is only a function of age), and in the VY model (where yield is a

function of age and production cost), it is necessary to adjust the yield function so as to

give the same yield at mean level of input, while at the same time reflecting the short-

run response to prices. This means when the average production cost of coffee used in

the FY model is used in the yield function and for the VY model, yields should be

similar.

The adjustment changes the intercept term in the estimated yield function in Table 6.2

and fixing yield for trees of age 7 to 17 years at a constant level. The adjusted yield

function for coffee in the mature period is:

-500

0

500

1000

1500

2000

2500

3000

0 5 10 15 20 25

age of tree

yie

ld (

kg

be

an

/ha

)


134

2 2max max 89 2.55*cos 0.00043cos 62 2.6mY tpha tpha age age (6.6)

where Ym denotes for yield at mature period; agemax is age at which coffee attains the

maximum yield and costpha stands for cost of production per hectare ($/ha). This

function describes yield of trees in the mature period only.

The yield of coffee in non-mature years is rescaled by yield in the mature period to

make yields in both cases (yield is constant at each age of coffee trees and yield is

function of production cost and age of coffee trees) consistent, and shows as follows:

If age of coffee trees is smaller than 3 years (age <3), yield of coffee trees is equal to

zero

If age coffee trees is greater than 2 and smaller than 7 (2<age<7),

2

5m

ageYield Y

If age of coffee trees is greater than 16 year old,

27

11m

ageYield Y

The yield function used in the FY model and the adjusted yield function in the VY

model are presented in Figure 6.6, for the case where the average production cost is

$930 per ha. They are very close to each other at all ages. In summary, the adjusted

yield function replicates the age-dependent yields based on expert judgment, while at

the same time incorporating the estimated impact of variable inputs.


135

Figure 6.6: Yield in the FY model and Adjusted Yield in the VY model at cost of

$930/ha

The adjusted yield function moves up and down based on production cost as shown in

Figure 6.7.

Figure 6.7: Yield variation at different production costs

0

500

1000

1500

2000

2500

3000

1 3 5 7 9 11 13 15 17 19 21

age of tree

kg

per

ha

yield in optimal model

Adjusted yield function

0

500

1000

1500

2000

2500

3000

3500

1 3 5 7 9 11 13 15 17 19 21

age of coffee tree

kg

per

ha

Adjusted yield (Cost =930USD)




136

6.4.3. Optimal Rule of the VY model

The optimal cutting and replanting rules are re-solved in the VY the model, given the

new opportunity for short run yield response. The optimal rule for the VY model is as

follows:

CP = 0.14 - 0.029age+0.0034age2 (6.7)

RP = 0.51 ($/kg) (6.8)

Maximum ENPV = 14369 ($) (6.9)

The optimal rule for the VY model, sketched in Figure 6.8

Figure 6.8: Optimal cutting and replanting rules in the VY model

More importantly, the profit of farmers with a short-run response increases significantly

compared to the FY model. The maximum ENPV achieved by the VY model is

approximately $14380, an increase at over 50 percent compared to the maximum ENPV

in the FY model (see Table 6.3 and Figure 6.9).

0

0.2

0.4

0.6

0.8

1

1.2

0 3 6 9 12 15 18 21age of trees

pri

ce

($

/kg

)

CP of VY Model RP of VY Model

Replanting price

keep coffee

keep coffee keep coffee

Cut and grow

maize


137

Table 6.3: Comparison of optimal rule between FY and VY model

Models Results Unit

FY model

Cutting rule CP =0.402 -0.05age +0.0036age2 $/kg

Replanting price RP =0.74 $/kg

maximum ENPV NPV =9226 $

VY model

Cutting rule CP =0.14 - 0.029age+0.0034age2 $/kg

Replanting price RP=0.51 $/kg

maximum ENPV NPV=14369 $

Figure 6.9: Optimal rules for FY model and VY model

As illustrated in Figure 6.9, the optimal replanting price in the VY model is only 0.51

($/kg), much smaller than that in the FY model (0.74 $/kg). In addition, the results from

the models show that in the FY model coffee farmers are more likely to cut than in the

VY model. However, the calculation of the percentage of farmers who really cut the

coffee when the optimal rule is invoked with our price simulation gives alternative

trends. The percentage of cutting cases in both models is not much different for trees

under 15 years old. However, with older trees, the percentage cutting coffee in the FY

model is much higher than in the VY model (see Figure 6.10). This implies the

difference in the cutting prices at low ages is not binding on behavior, but at higher ages

is.

0

0.2

0.4

0.6

0.8

1

0 2 4 6 8 10 12 14 16 18 20 22

age of coffee trees

co

ffe

e p

ric

e (

$/k

g)

CP of FY Model CP of VY Model

RP of FY Model RP of VY Model


138

Figure 6.10: Percentage of cases in which farmers cut coffee at optimal rule

The comparison between profit levels generated from the FY model and the VY model

may not be fully accurate, as the yield and cost in the FY model are not representative

for the average cost yield in the VY model. This is because the optimal variable input

costs, as solved within the VY model is much higher, and hence gives a higher yield,

than in the FY model (see Table 6.4). Thus, to see more clearly the change in benefit

gained by farmers with the short-run response, the FY model is re-solved for the

optimal rule, with the yield replaced by the higher average yield in the VY model. The

results of both models are presented in Table 6.5. Farmers in the FY model are still

more likely to cut than in the VY model. However, the replanting price is the same in

both. In addition, the maximum ENPV in the re-solved FY model is still much lower

than in the VY model. This shows that if farmers have the opportunity for a short-run

response of cost and yield to price, they can greatly improve their profit.

0

2

4

6

8

10

12

14

16

18

20

0 2 4 6 8 10 12 14 16 18 20 22

age of coffee trees

% c

utt

ing

FY model

VY model


139

Table 6.4: Average cost and yield from the VY model and the FY model

Age of

trees

Average cost

from the VY

model

Annual cost in

the FY model

Average yield

from the VY

model

Yield in the

FY model

1 1440 1440 0 0

2 800 800 0 0

3 296 1015 987 500

4 592 824 1603 1200

5 877 929 2102 1500

6 1166 929 2519 2000

7 1466 929 2870 2300

8 1466 929 2874 2500

9 1474 929 2886 2500

10 1476 929 2889 2500

11 1471 929 2882 2500

12 1470 929 2876 2500

13 1465 929 2867 2500

14 1466 929 2865 2500

15 1469 929 2871 2500

16 1470 929 2878 2300

17 1344 929 2743 2100

18 1204 929 2573 2000

19 1074 929 2403 1800

20 940 929 2204 1600

21 808 613 1989 1400

22 672 613 1752 1200

Table 6.5: The results of FY model with average cost and yield from VY model

Models Results Unit

FY model with average cost and yield from VY model

Cutting rule CP =0.3 -0.025age +0.0034age2 $/kg

Replanting price RP =0.51 $/kg

maximum ENPV NPV =10966 $

VY model

Cutting rule CP =0.14 - 0.029age+0.0034age2 $/kg

Replanting price RP=0.51 $/kg

maximum ENPV NPV=14369 $

The next section will incorporate the short-run response in the FY-CC model which was

presented in Chapter 5 and investigate how the poor household’s decision changes.


140

6.5. The Variable Yield – Cash Constraint Model (VY-CC model)

6.5.1. Model Structure

The structure of the VY-CC model is almost the same as the FY-CC model but the

coffee yield is replaced by the adjusted yield function as presented in Section 6.4.2 and

a small change in decision rules is introduced. The objective function is still the

maximum ENPV from land choice per poor farm size (5287 m2). The coffee yield

function in the VY-CC is the same as in the VY model. Similar to the VY model,

production cost of coffee in the VY-CC model is determined by the coffee price as

follows:

1

2

1

2optimal

pC

p

(6.10)

where 1 =2.55 and 2 =0.00043

The yield is a function of production cost and age of coffee trees as presented in (6.6).

However, the production costs for the first two years of gestation period are not

dependent on price; they are estimated from Coffee Farm Survey 2007. This is the same

as the VY model.

There is a small change in decision rule in the VY-CC model compared to the FY-CC

model. With the FY-CC model, production cost is fixed for a given age so farmers keep

growing coffee if:

coffee price >min (CP,RP) and

Capitalt + Io>= Costt+1 + Minexpend

In the VY-CC model, only production costs for the first year coffee trees (replanting

price) and the second year are fixed. The production costs for older trees are a function

of output price. If prices are too low, farmers may not apply inputs for coffee

production. This means the variable cost for trees greater than 2 years can be zero.

However, this decision will affect the coffee output through the yield function. Thus, in

the VY-CC model when farmers are growing coffee, they will keep coffee if:

coffee price >min (CP,RP) and


141

Capitalt + Io>= Cost2 (if aget-1 =1) + Minexpend

where Io is other income of poor household, Cost2 is production cost of 2 year old

coffee trees, Minexpend is the minimum expenditure of household.

If farmers are growing maize or they want to replace old coffee trees (if coffee age is 22

year old), price and household budget must satisfy two conditions:

coffee price >= replanting price, and

Capitalt + Io>= replanting cost + Minexpend

The VY-CC only investigates the optimal rules using the lagged price simulation and

the quadratic fixed form of CP (i.e.2

1 2oCP age age )

6.5.2. Optimal Rule of the VY-CC model

Using the grid search method to solve the VY-CC model, the optimal rules are

identified as follows:

CP = 0.19 – 0.067age + 0.0044age2 (6.11)

RP = 0.59 ($/kg)

Optimal ENPV per poor farm size = 7086 ($)

The output from the VY-CC model shows significant changes compared to the FY-CC

model in Chapter 5. Model results indicate that poor coffee farmers with short-run

response in the VY-CC are much less likely to cut. Poor farmers with 4 year old to 11

year old coffee trees should not cut coffee even if price reduces to 0 ($/kg). A farmer

with trees over 12-year old is much more likely to cut. Model output also shows that

farmers should replant coffee when price reaches 0.59 ($/kg). The optimal cutting and

replanting rules for poor farmers with short-run response is illustrated in Figure 6.11. A

comparison of optimal rules between the FY-CC model and the VY-CC model is

presented in Figure 6.12. As shown in the figure, with the option of a short-run

response, poor coffee farmers replant much earlier and are much less likely to cut for

growing maize.


142

Figure 6.11: Optimal rule of the VY-CC model

Figure 6.12: A comparison of optimal rule between FY-CC and VY-CC model

The difference between the FY-CC model and the VY-CC model is shown clearly by

the percentage of cases in which farmers cut their coffee trees when optimal rules and

the cash constraint are invoked. As presented in Figure 6.13, the actual cutting

frequency in the FY-CC model is always higher than that in the VY-CC model,

especially for 2-year old trees. As shown in the FY-CC model, some farmers have to cut

their coffee trees because they cannot afford to keep coffee when the price goes down.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1 3 5 7 9 11 13 15 17 19 21

age of coffee trees

pri

ce (

$/k

g)

CP of VY-CC Model RP of VY-CC Model

keep growing coffee

keep growing coffee

cut and grow

maize

replanting

price

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1 3 5 7 9 11 13 15 17 19 21

age of trees

pri

ce (

$/k

g)

CP of FY-CC model RP of FY-CC Model



143

According to the VY-CC model output, the percentage of farmers who cut their trees in

the first year remains high. The reason is the same as in the FY-CC model in which

farmers have to cut trees because of the cash constraint. This is shown clearly in Figure

6.14. Similar to the FY-CC model, the liquidity constraint is the main cause that force

farmers with young coffee trees to cut down. Meanwhile, farmers cut old trees when the

CP rule is invoked.

Figure 6.13: Percentage of cases in which farmers cut coffee at optimal rules

Figure 6.14: Percentage of actual cut of the VY-CC model by cutting rule and by

cash constraint

0

5

10

15

20

25

30

0 2 4 6 8 10 12 14 16 18 20 22

age of coffee tree

% c

ases o

pti

mal

rule

in

vo

ked

FY-CC model

VY-CC model

0

2

4

6

8

10

12

14

1 3 5 7 9 11 13 15 17 19 21

age of coffee trees

% a

ctu

al

cu

t

by CP

by cash constraint


144

Figure 6.15 compares maximum ENPV gained by farmers for different models with a

farm size of 5287m2. As is shown clearly in the figure, with the short-run response, poor

coffee farmers can get much higher income compared to the FY-CC model. The

maximum ENPV from the VY-CC model is 7086 ($/poor farm-size), much higher than

income from the FY-CC model in Chapter 5 (only $ 4375). However, due to the cash

constraint, poor farmers in the VY-CC model cannot optimize their decision, thus the

maximum ENPV in the VY-CC model is still lower than the ENPV in the VY model.

Figure 6.15: Comparison of ENPV from different models at poor farm-size

The comparison between incomes from two models (FY-CC and VY-CC) is to some

extent incomplete because production cost and yield from the FY-CC model is not

compatible with those from the VY-CC model. Thus, to see the ENPV gained by poor

farmers from being able to utilize the short-run response, the FY-CC model is re-solved

and cost and yield are replaced by average cost and average yield from the VY-CC

model. The re-solved optimal rules of the FY-CC model with average cost and yield

from the VY-CC model are presented in (6.12). The model output shows the

improvement of maximum ENPV, but it is much lower than maximum ENPV from the

VY-CC model. Again, this result confirms the importance of short-run response and its

impact on coffee farmers’ decision.

4376

7086

7600

0

1000

2000

3000

4000

5000

6000

7000

8000

FY-CC model VY-CC model VY Model

EN

PV

-$/p

er

po

or

farm

siz

e


145

CP = 0.38 - 0.026age + 0.0018age2 (6.12)

RP = 0.74 ($/kg)

Optimal ENPV at poor farm size=5323 ($)

6.6. Conclusion

This chapter investigates the coffee farmer’s decision when it is possible to change

yields in the short-run. By integrating the coffee yield function in the short-run into the

FY model and the FY-CC model, this chapter develops two corresponding models: the

VY model and the VY-CC model.

In general, in the presence of a short-run response, farmers in both the VY model and

the VY-CC model are much less like to cut as compared to the FY model and FY-CC

model. Farmers in the VY model seem never cut if the coffee trees are less than 11

years old.

With the short-run response, the actual cutting percentages in the VY-CC model at

optimal rules reduces significantly compared to the FY-CC model. The VY-CC model

shows that the poor farmers never cut when coffee trees are in 4-16 age groups. They

only cut in the early years of the cycle or when the trees are getting quite old. Similar to

the FY-CC model, farmers with young trees in the VY-CC model have to cut coffee

trees because of the liquidity constraint.

Furthermore, with the short-run response, farmers are more likely to replant coffee if

they are growing maize or having bare land. The VY model found that farmers should

replant when the price is $0.51 per kg. The optimal replanting price in the VY-CC

model is only $0.59 per kg of coffee, much lower than that in the FY-CC model ($1.4

per kg).

The short-run response of farmers improves the value of ENPV considerably. The

maximum ENPV from the VY model increases by over 50 percent compared to the

ENPV in the FY model. Similarly, the maximum ENPV of the VY-CC model is about

60 percent higher than that in the FY-CC model. The ENPV of the FY model and the

FY-CC model is changed when the yield and cost functions in these models are replaced


146

by the average cost and average yield from short-run response, however it does not

improve significantly.

In chapters 4, 5 and 6, different optimal models are developed to investigate the coffee

farmer’s decision in different scenarios. The following models are expanded from the

FY model. The objective and general structure of models are similar but the detail

functions and decision rules are quite different. Thus, it would give a much better

understanding when comparing and summarizing main ideas, structure and output of all

models. This will be done in next chapter.

Chapter 7. Summary of the optimal models

147

CHAPTER 7. SUMMARY OF THE OPTIMAL MODELS

7.1. Introduction

In the three previous chapters (4, 5 and 6), four version of a simulation model were

developed to identify the optimal cutting and replanting rules for coffee farmers in

Vietnam. The four model version are:

Fixed yield optimal model (FY model)

Fixed yield optimal model with cash constraint (FY-CC model)

Variable yield optimal model (VY model), and

Variable yield optimal model with cash constraint (VY-CC model)

Latter models have a similar structure and specification to the FY model but with

additional constraints added to explore different aspects of the economic behavior of

coffee farmers. This chapter is a synthesis of the previous optimal model discussion to

compare results across the models.

The purpose of this chapter is to provide a synthesis of the results from these models,

and provide some interpretation as to their implications. Following the introductory

part, Section 7.2 will summarize the differences of objective, structure and particular

constraints among models. The change in farmer’s decision will be presented in Section

7.3.

7.2. Model Development

7.2.1. Objectives of Models

In general, all models aim to identify the cutting price (CP) and replanting price (RP)

for maximizing the expected net present value from “land use choice” (ENPV).

However, each model investigates the optimal rules in different contexts. The FY and

VY model identify the optimal rules to maximise ENPV per ha of farm land. The land

can be used to grow coffee or maize (as a substitute crop).


148

Similarly, the main objective of the FY-CC and the VY-CC model is to investigate the

change in the poor coffee farmer’s decision, when they face a cash constraint. These

latter models concern a household as a whole, and cover other aspects of a household

such as living expenditure, other income, loans and family labour. In addition, the farm

size in those models is fixed at an appropriate size (5287 m2). The aims and objective

function of the models are summarized in Table 7.1

Table 7.1: Main objective of models

Models Core objective

FY model Identifying the optimal CP and RP to maximise the ENPV for

1 ha of land

Objective function: maximum ENPV per ha

FY-CC model Investigating the optimal CP and RP to maximise the ENPV

for poor coffee farmers in the presence of a cash constraint

Objective function: maximum ENPV per poor coffee farm (size

5287 m2)

VY model Finding CP and RP to get the maximum ENPV with the

incorporation of short-run response in which coffee yield is a

function of production cost and tree age

Objective function: maximum ENPV per ha

VY-CC Model Specifying the optimal cutting and replanting rule to maximise

ENPV for poor coffee farmers in the presence of a cash

constraint and short-run response

Objective function: maximum ENPV per poor coffee farm (size

5287 m2)

7.2.2. Rules and Constraints

The models apply the fixed form optimization approach to specify the cutting and

replanting prices for obtaining the maximum ENPV. The fixed form used for CP and

RP is as follows:

2

1 2oCP age age

3RP


149

To find CP and RP, models finally need to identify 1 2, ,o and 3 .

Each model contains particular rules and constraints. In the FY and the FY-CC model,

coffee yield and production cost vary by age of tree, but cannot be altered by the

farmers. In FY model, coffee farmers are assumed to not have a cash constraint. Thus,

the cutting and replanting decision are only dependent on the price of coffee at which

farmers will cut coffee and switch to maize (as a substitute crop) if coffee price falls

below CP. They will decide to grow coffee again if price increase to RP (see Table 7.2).

The cutting and planting decision rules of the VY model are the same as the FY model.

The structure of the FC-CC and the VY-CC model are different from that of the FY and

the VY model. The FC-CC and the VY-CC model concern the farm as a household and

focus on the decision of the poor coffee household. Some additional equations are added

into these models such as living expenditure, family labour return, saving and credit.

However, coffee yield in the FY-CC model is the same as in the FY model: yield is

fixed at a given age of coffee trees. The decision of poor coffee households in the FY-

CC model and the VY-CC model depends on both the price and the cash availability of

the household. Decision rules can be presented as follows:

Keep growing coffee if (i) the price is greater than the CP and (ii) the household budget

(savings + borrowings + other income) can at least cover the minimum household

expenditure and production cost in the following year. Otherwise, they will switch to

maize.

Replant coffee if the price is greater than or equal to the RP and the budget of

households is greater than the sum of replanting cost and minimum household

expenditure.

Table 7.2 summarizes the decision rules and constraints in the four optimal models.

Table 7.2: Decision rules and constraints

Models Relaxed constraints

FY model o Yield is fixed by age of coffee trees

o yield =f(age);cost =f(age)

o no cash constraint

o decision rule: cut and replace with maize if price <CP otherwise

keep growing; and replant if price >=RP

VY model o Yield is now a function of age and production cost


150

o yield =f(age, production cost); cost =f(coffee price)

o no cash constraint

o decision rule: cut and replace with maize if price <CP otherwise

keep growing; and replanting if price >=RP

FY-CC model o include expenditure function, saving, other income, loan

o yield =f(age); cost =f(age) (as for FY model)

o New rule:

keep coffee if price > CP and household’s budget >=

minimum household expenditure+ production cost in next

year. Otherwise, cut for maize.

Replant coffee if (price >=RP) and (household’s budget>

replanting cost + minimum household expenditure)

VY-CC Model o including expenditure function, saving, other income, loan

o yield =f(age, production cost);optimal cost =f(coffee price)

o rule:

keep coffee if price > CP and household’s budget >=

minimum household expenditure+ production cost in next

year. Otherwise, cut for maize.

replant coffee if (price >=RP) and (household’s budget>

replanting cost + minimum household expenditure

7.3. Changes in Coffee Farmer’s Decision

The main outputs of all models are summarized in Table 7.3. The maximum ENPV

from the VY model is highest with about $14370 per hectare (or equivalent to $7600

per poor farm size). This means that if coffee farmers are not restricted by a cash

constraint and input use is responsive to the output price, they can achieve the

maximum returns from investment in coffee and optimize their decision. The maximum

ENPV in the VY model is about 50 percent higher than in the FY model (see Figure

7.1). The difference reduces to 31 percent if the yield function in the FY model is

replaced by the average yield of the VY model.

Table 7.3: Main results of simulation models

Models Main output

FY model CP = 0.402 - 0.0509age + 0.00367age2

RP = 0.74 ($/kg of coffee bean)

ENPV = 9226 ($/ha) ~= 4878 ($/poor farm size)

VY model CP = 0.14 - 0.029age+0.0034age2

RP = 0.51 ($/kg)

ENPV = 14369 ($/ha) ~=7600 ($/poor farm size)


151

FY-CC model CP = 0.458 -0.46age +0.00325age2

RP = 1.4 ($/kg)

ENPV = 4375 ($/poor farm size)

VY-CC Model CP = 0.19 – 0.067age + 0.0044age2

RP = 0.59 ($/kg)

ENPV = 7086 ($/poor farm size)

Note: poor farm size is 5287m2

Figure 7.1: Maximum ENPV achieved from models

The hypothesis behind the FY-CC and the VY-CC models is that in some cases farmer’s

efficiency may be altered by the cash constraint. They may have to cut earlier than the

original optimal point or they cannot replant because of the cash shortage. The results of

models are consistent with farmer’s expected behaviors. A comparison between the FY

and the FY-CC shows that farmers in the FY model are less likely to cut and they

replant earlier (see Figure 7.1 and Figure 7.2). Households represented by the FY-CC

model cannot achieve the oreginal optimal land use choice and earn less than those who

are represented by the FY model. The income of poor farmers reduces by 15 percent if

they follow the optimal decision of the non-poor farmers. However, despite the cash

constraint if coffee farmers adjust input use efficiently to coffee price, they can greatly

improve their income. This explains why the ENPV in the VY-CC is much higher than

in the FY-CC.

100%92%

57%64%

0

1000

2000

3000

4000

5000

6000

7000

8000

FY model FY-CC model VY model VY-CC model

Exp

ecte

d N

PV

($/p

oo

r fa

rm s

ize)


152

Figure 7.2: Optimal cutting and replanting rules in different models

The effect of a cash constraint on farmer’s decision is expressed more clearly through

the actual cutting farmers in models when optimal rules are invoked. The cutting

frequency in the FY-CC is higher than that in the FY model, especially for young trees

and very old trees (see Figure 7.3). The cutting decision of poor coffee households is

influenced by both the optimal rules and cash constraint. This can also be seen in Figure

7.3 where the cutting percentage is affected by the combined impact of the CP rule and

the cash constraint. The result shows that cash problems have a significant effect on the

cutting decision of farmers with young trees.

Figure 7.3: Actual cutting percentage at optimal rules in FY and FY-CC model

The cutting percentage also changes if farmers have an efficient short-run response.

Figure 7.4 presents the cutting percentage at optimal rule in both the FY and the VY

models. There are two points to note in this figure. First, the optimal rule shows a

positive value of CP for all trees, but the real cutting percentage under the price

simulation in the FY and the VY models for young trees is very small. This means

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

0 5 10 15 20

age of trees

pri

ce (

$/k

g)

CP of FY model CP of FY-CC model

RP of FY model RP of FY-CC model

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1 3 5 7 9 11 13 15 17 19 21

age of trees

pri

ce (

$/k

g)

CP of FY-CC model RP of FY-CC Model


0

5

10

15

20

25

30

1 3 5 7 9 11 13 15 17 19 21

age of coffee trees

% a

ctu

al c

ut

FY model

FY-CC model

0

5

10

15

20

25

30

1 3 5 7 9 11 13 15 17 19 21

age of trees

% a

ctu

al

cu

t

by CP rules

by cash constraint

FY-CC model


153

without cash problems, farmers mostly never cut if trees are less than 11 year old. In

addition, the actual cutting percentage under 15-year old trees is also quite small. This

pattern is almost the same for both the FY and the VY models. Secondly, the actual

cutting percentage in the VY model is generally smaller than in the FY model. The

difference between models becomes significant for 17 year old and older trees.

Figure 7.4: Cutting percentage at optimal rule in FY and VY model

The optimal replanting prices vary across different models. The replanting price reflects

the expected income earned by farmers from coffee production. According to model

output, farmers in the VY model decide to replant at a relatively low price of $0.51 per

kg. Due to the cash constraint, farmers in the VY-CC model wait for a higher price to

replant coffee trees and they grow coffee again at a price of $0.59 per kg (Figure 7.5).

The poor households in the FY-CC model optimize their decision to replant coffee at

$1.4 per kg. In this case, farmers do not have the response of input use to the output

price. In addition, their decision is constrained by a cash constraint so they wait until

there is a high price to reduce the chance of low future price occurring. This explains

why poor farmers without short-run response are much less likely to replant.

0

2

4

6

8

10

12

14

16

18

20

0 2 4 6 8 10 12 14 16 18 20 22

age of trees

actu

al cu

ttin

g (

%)

FY model

VY model


154

Figure 7.5: Optimal replanting prices for different models

The optimal cutting and replanting prices for the fixed yield models in this study differ

from those of Luong and Loren (2006). Luong and Loren (2006) found that a farmer

would enter into coffee production when prices are above 1.04 $/kg and exit if the price

dropped below 0.32 $/kg. Hence, farmers without cash constraints in optimal models

replant earlier. An important improvement in decision analysis in this study compared

to the model by Luong and Loren (2006) is that, cutting prices in fixed form are a

function of the age of trees. It is not a fixed number for all coffee groups as presented in

Luong and Loren (2006).

7.4. Conclusion

The application of simulation models helps to understand the replanting and cutting

decisions of individual farmers. More clearly, the models identify at what price farmers

should cut their trees down and switch to other crops. In addition, the optimal models

point out when farmers should replant coffee if they currently have bare land or are

growing other crops.

One conclusion in all models is that coffee farmers optimized their decision at different

‘trigger’ prices for cutting and replanting. This asymmetric response of coffee

households may be reflected in an asymmetric response of coffee area at aggregate

levels to price changes. To test this hypothesis, the next chapter will analyze the supply

response of the aggregate coffee area.

0.74

1.4

0.510.59

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

The FY model The FY-CC model The VY model The VY-CC model

pri

ce (

$/k

g c

off

ee b

ean

)

Chapter 8. Coffee Supply Response in Vietnam

155

CHAPTER 8. COFFEE SUPPLY RESPONSE IN VIETNAM

8.1. Introduction

The estimation of supply response functions could improve the understanding of the

price mechanism and the responsiveness of supply to price changes (Nerlove and

Bachman, 1960). Understanding supply responsiveness can assist policy makers in

achieving production targets in markets where price is considered as a policy tool. In

addition, estimated supply functions can be used for forecasting.

Estimating supply functions for perennial crops such as coffee is more complex than for

annual crops due to the time lags associated with the decision to increase production and

production capacity becoming available. Furthermore, supply decisions are a function

of expected cost and return over the whole life cycle of coffee trees. The input

requirements and yields of perennial crop vary as a function of tree age, implying that

annual production depends on the age composition of the tree stock. Furthermore, the

age composition of trees influences plantings and removals. The pioneering work of

Nerlove (1956) on supply response was concerned with annual crops (wheat, cotton and

corn). Nerlove’s model has been adopted by later authors to represent the supply

response of perennial crops.

From the output of the simulation models in previous chapters, it is optimal for different

farmers to cut and replant at different prices. The cutting/replanting gap may be

reflected in an asymmetric response of the coffee area at the aggregate level to price

changes.

This chapter investigates a supply function based on time series data to see if the

behaviour of farmers identified in previous chapters can be observed in the time series

data. The supply function is used to estimate short and long-run elasticities. The

function tests the hypothesis that the coffee supply in Vietnam shows an asymmetrical

price response, i.e. is relatively responsive to price rises but is unresponsive to price

falls.


156

Section 8.2 reviews previous methods and studies on supply response using time series

data. The application to coffee supply response in Vietnam is introduced in Section 8.3.

Some conclusions are highlighted in Section 8.4.

8.2. Literature Review on Supply Response Analysis using the

Econometric Approach

This section begins with standard models where supply response is symmetric or

reversible and then reviews studies of asymmetric supply responses.

The structural equation approach (Sadoulet and deJanvry, 1995) is based upon

production economics and the theory of the firm and includes primal approaches, based

around the production function and dual approaches using the profit function and the

cost function.

Alternatively, the reduced form supply function approach simply explains supply of a

commodity as a function of selected commodity prices, with relatively limited

constraints derived from theory. The Nerlovian model (Nerlove 1956b) is a reduced

form model and this approach will be applied here.

When looking at the crop supply response, researchers often study the area grown of a

crop, but not the output. For annual crops, the supply response model looks at the

change in area but for perennial crops such as coffee, the cutting and replanting area

should also be included. Thus, the area equation of perennial crops is identified as:

1t t t tA A N R

where tA is crop area in year t, 1tA is the lagged area, tN is new planting area in year t

and tR is the removal area in year t.

For perennial crops, the analysis of supply by decomposing the response into a response

for plantings and a response for removals would give understanding of the response to

price changes. However, time series data on new plantings or removals is often not

available, instead only an aggregate area is available.

.


157

8.2.1. Nerlovian Approach

Coffee is a perennial crop with a three to four year establishment period between initial

planting and the first harvest. Thus, when deciding to replant, coffee producers base

their decision on expected, not observed, prices. Because of the time lag in crop

production in general, modeling with expected prices has been an important

consideration in the analysis of crop supply response. The time lag in production means

that there is often a divergence between desired and actual output (Sadoulet and

deJanvry, 1995). To address these issues, a number of models have been developed in

the literature.

The Nerlovian model of crop supply response is formulated in terms of desired area,

yield and output. The generalized Nerlove’s model basically includes three equations

(Nerlove, 1956, Askari and Cummings, 1977, Sadoulet and deJanvry, 1995)

1 2 3

d e

t t t tA P Z u (8.1)

10,.........)( 11 tt

d

ttt vAAAA (8.2)

t

e

tt

e

t

e

t wPPPP )( 111

t

e

tt

e

t wPPP 11 )1( (8.3)

where tA is the actual area under cultivation at time t , d

tA is the area desired to be

under cultivation at the time t , tP is the actual price at time t , e

tP is the expected

prices at time t , tZ is the other exogenous factors affecting supply at the time t , and

γ are termed the expectation and adjustment coefficients, respectively, ,t tu v and tw are

error terms.

In (8.1), desired area is a function of expected prices, own price and price of a

competing crop and other exogenous factors affecting supply (such as weather).

Equation (8.2) is a land adjustment equation. Full adjustment cannot be achieved in the

short run, thus the actual change between year t and 1t is only a fraction ( ) of

desired adjustment. Equation (8.3) is an adaptive expectation price equation. Because

the expected price cannot be observed, the model expresses expected prices based on


158

actual/observed prices. This equation represents a price learning process where farmers

adjust their expectation as a fraction of the forecast error in the previous year.

e

tP and d

tA are not observable, but by substituting e

tP and d

tA from (8.2) and (8.3) into

(8.1), e

tP and d

tA are eliminated and the reduced form of the area equation is:

1 2 1 3 1 4 2 5 6 1t t t t t t tA P A A Z Z e (8.4)

where:

tttttt wuuvve 2111

36

35

4

3

22

11

)1()1(

)1(

)1)(1(

)1()1(

There are six coefficients ( 1 6 ) in the reduced form but only five parameters in the

original equation system: 1 , 2 , 3 , , . Hence, to get the unique solution for

parameters in original equation system, the following constraint has to be imposed on

the coefficients in the reduced form:

0653

2

54

2

6

After estimating 1 6 from the reduced form, we can identify the parameters in the

original system.

/

/

/

)1/(1

01)2(

55

22

11

4

433

2

The general Nerlovian model described above has been applied in numerous crop

response studies (see Askari and Cumming (1977). Nerlove’s model has been modified

to represent livestock supply and perennial crops. Studies on the supply of perennial


159

crops are more common but are challenging due to the characteristics of perennial

crops, summarized by French and Matthews (1971) as follows:

(i) The long gestation period between initial input and first output or time of

establishment period

(ii) The extended period of output flowing from initial production or investment

(iii) The deterioration of plants over time

A supply response model for perennial crops has to explain not only the planting

process but also replanting and removal. Knapp (1987) also pointed out that perennial

crops pose additional challenges compared to annual crops because production extends

over several years. The planting decisions have to reflect expected costs and returns

over several years.

Despite the difficulties in studying supply response for perennial crops, a number of

researchers have used the Nerlovian model. One of the earlier applications of the

Nerlovian model to perennial crop supply response was Bateman (1965) who analyzed

the case of cocoa in Ghana. He assumed that the farmer’s objective was to maximise the

discounted value of the future stream of net returns from cocoa. The planting area of

cocoa was a function of its own expected real price and coffee price (as a substitute

crop) and the output was a function of yield. After taking the differences of output and

combining output and area equation, Bateman obtained the final reduced form function

in which output is a function of lagged prices of cocoa and coffee, rainfall, humidity and

lagged output. Similarly, Behrman (1968) applied a Nerlovian model to estimate the

supply function of cocoa for leading producing countries. In contrast to Bateman, he

started with an area function in which desired area is a function of expected cocoa price

and coffee price. After that, Behrman transformed this function to output function by

applying a yield factor. Similar to Bateman, by taking the first difference of output, the

change of cocoa output finally became a function of lagged cocoa price difference,

coffee price difference and the second and third difference of output.

Saylor (1974) applied Nerlovian equation systems to measure the supply elasticities of

coffee in Sao Paulo (Brazil). The data used in the model were coffee area in Sao Paulo

for the years 1947-1970 and farm-gate coffee price in the 1945-1969 period. In the

study, Saylor estimated several alternative models, with the main explanatory variables

being lagged price, lagged area, time trend and price index of 20 leading agricultural


160

commodities in Sao Paulo31

. Saylor pointed out that the Nerlove model could explain

most of the variation in coffee supply and found that the price elasticities are relatively

low for both short-run and long-run, with the long-run values (ranging from 0.5 to 0.73

depending on particular supply equations) much higher than the short-run (from 0.1 to

0.19).

8.2.2. Extended Nerlovian Approach

For perennial crops, removal and replanting are influenced by expected price and costs.

Replanting also allows new technology to be adopted. For these reasons, it is much

better for supply response analysis to explicitly represent planting and removal

decisions. Thus, to develop the Nerlovian model for analyzing total area response,

researchers have estimated new planting and removal equations. French and Bressler

(1962) used this approach to develop a supply model of lemons in California (USA) in

which they estimated new plantings and removals. Originating from the relationship in

which the acreage of new lemon trees depends on expected long run profitability, age

distribution and expected profitability of other activities, the authors tried to

approximate the relationship by a linear function of long-run profit expectation. The

profit expectation in this paper was calculated using five years of past net returns.

Similarly, tree removals were expressed as a function of expected current profit and

proportion of fruit bearing trees over 25 years. However, they found that the proportion

of fruit bearing trees over 25 years was insignificant due to its small variation during the

observed period, and expected profit did not give statistically significant results. Thus,

the estimate of the proportion of removals is simply the mean value of the ratio of

bearing acres divided by acres of trees removed.

To describe the characteristics of perennial asparagus crops, French and Matthews

(1971) provided a model of supply response with 5 major components: (1) functions

explaining quantity of production and crop bearing acreage desired by growers; (2) a

new plantings function, (3) removed acreage each year (4) relationships between

unobservable expectation variables and observable variables and (5) an equation

explaining variation in average yield. However, because of data constraints, they tried to

simplify their model by constructing an equation system to express those relationships

based on the expected profit function with explanatory variables such as price, a wage

index and a supply equation. The expected profit is:

31

This variable attempts to see whether the price of competing activities influence coffee area.


161

0 1( / )e ec c P M u

where e denotes expected profit, P is grower price, W is wage rate. The e means an

expected value.

The supply function consists of change of acreage (At-At-1) and a number of independent

variables such as previous acreage and (P/M) ratio and dummy variables32

. This

equation was estimated by the Ordinary Least of Square method (OLS). However, the

estimated coefficients could not be used to recover the structural parameters because the

model was under-identified. Thus, the effect of harvest and investment decisions could

not be measured separately.

8.2.3. Wicken - Greenfield Approach

Wicken and Greenfield (1973) criticized the Nerlovian model, because the model fails

to distinguish between the investment decision regarding the stock of trees and the

harvesting decision. Thus, they have attempted to develop a vintage production,

investment and supply response model for the coffee crop in Brazil. Their structural

equations are:

0

,n

P

i t it

i

Iq

1 t-1 2 tI a Pt oI a a

t 0 1 2 1

0

qm

P

t i t i t

i

q P q

where P

tq is the potential output, tq is the actual output, tI denotes investment and tP is

producer price.

Wicken and Greenfield obtained the reduced form of supply in which output is a

function of the distributed lag of price and area as follows:

t 1 1 1 2

0

q ( )m

i t i t t

i

P q q cons (8.5)

Where:

32

See more detail in French and Matthews (1971).


162

0122i i = 0

110122 tii i = 1,…m

11112 mmi a i = m+1

ii a 12 i=m+2,…..n.

cons is the constant term

The Wicken-Greenfield model has been applied to a number of crops. Dowling (1979)

used the Almon Lag model to analyze of the supply response of rubber in Thailand.

Hartley et al (1987) looked at a similar supply response of rubber in Sri Lanka. In their

models, new planting was a negligible component of total area so the researchers

focused on modeling the uprooting and replanting decision. Both authors conclude that

the relationship between production and the stock of trees was considerably more

complex than specified by the Wickens-Greenfield model. After estimating removal,

supply and new planting equations, they conclude that the Wicken-Greenfield model is

not appropriate for the rubber sector in Sri Lanka. Some coefficients in the estimated

equations had wrong signs: price was estimated to have a significant negative effect on

new planting, and the wage is significantly positive.

Akiyama and Trivedi (1987), provide an extended critique of the Wicken-Greenfield

approach. First, the model is over-identified. Second, it is difficult to add non-price

variables in the planting equation (because they appear as a distributed lag in the

reduced form). Akiyama and Trivedi (1987) developed a Vintage production model for

perennial crops and applied this to the tea sector in three countries (India, Sri Lanka and

Kenya). This model included new plantings, supply, replanting and uprooting based on

different explanatory variables such as moving average price, capacity for new

plantings, extension service (for India), expenditure per hectare for extension and

service, new plantings, real price (for Kenya), and tea production cost, tea price, new

plantings and uprootings, and a replanting subsidy (for Sri Lanka).

Akiyama and Trivedi’s vintage production model requires reliable time-series data for

production, area planted and credit availability. For many countries, these data are not

available. Furthermore, econometric models of commodity markets are valid only when

the relationships among variables are stable over time without any significant structural

changes. These are also problems raised by Nerlove (1979). According to Nerlove

(1979), there are four main problems when studying supply response for perennial crops


163

in developing countries using time series data. First, time series data needs to be

available, especially new planting data and current age structure data. Second,

government intervention is wide-spread and affects the supply response. Third, there is

frequently an imperfect relationship between output and stock of the perennial crop. The

depletion of stock varies not only because of cutting but also because of weather and

disease. Fourth, technical change (new varieties) produces an additional element of

uncertainty.

8.2.4. Price Asymmetric Response

When estimating the response of supply to price, the standard approach is to assume

that supply responds symmetrically to price increases and decreases. Based on Deaton

and Laroque (2003), Olsen (2005) built a model for world coffee supply response and

showed that the supply response is asymmetric. In particular, supply responds to price

increases but is unresponsive to price decreases. There is even some evidence of a

perverse supply response where decreasing prices may cause coffee farmers to increase

production. Olsen (2005) explains these findings by farmers being able to survive

through subsistence crops while they wait for the coffee price to increase; and a lack of

alternative income sources. Olsen (2005), proposes that the existence of a “fixed asset”

causes an asymmetric supply response. Low salvage values cause producers to continue

production even though prices are low because the acquisition costs are high.

A method for studying asymmetric supply response was introduced by Tweeten and

Quance (1969) when looking at crop and livestock supply in the United States. They

analyzed the supply response to price change by splitting the price variable into two

variables one each for price increases and price decreases, thus:

0 1t t t tQ p p (8.6)

in which

21 1

1

, 1, 0p

p p if otherwisep

1

, 1, 0tt t

t

pp p if otherwise

p

,t tp p if 0tp and

0tp , if t tp p

However, according to Wolffram (1971), the price split by Tweeten and Quance causes

incorrect solutions for irreversible supply reaction and differentiation of partial


164

influence because the supply in Tweeten and Quance’s method cannot reflect by change

of price (decrease or reduce). Thus, Wolffram splits the price series by using price

differences. According to Wolffarm’s method the supply function can be estimated by:

0 1 2t t tQ WR WF (8.7)

Where tWR is the sum of all period to period increases in expected price from its initial

value up to period t and tWF is the sum of period to period decreases. Mathematically,

the split price series can be expressed as follows:

1 1;WR p 1 1( )t t t tWR WR p p

1 1;WF p 1 1(1 )( )t t t tWF WF p p

t t-1α =1 if p -p 0, else=0

Houck (1977) modified Wolffram’s approach by cumulating positive and negative first

differences, beginning with zero and not with the actual first observation. Furthermore,

to focus on the identification of starting points and measurement through levels, Houck

(1977) changed the dependent variable to Qt – Q0, where Q0 is quantity in the first

period. Traill et al (1978) criticized the Wolffram function because it implies that for

given starting and finishing prices, the greater are price changes in intermediate periods,

the larger is output at the end of the period. However, in practice highly variable prices

would lead to an output reduction due to risk concerns. Traill et al (1978) pointed out

that the response of supply to price will only become elastic once it has risen beyond the

previous maximum price. Thus, Traill et al (1978) modified Wolffram’s method so that

when the price increases but remains below the previous maximum level, price change

is added to the price fall series (modified Wolffram fall – MWF ) rather than to the price

rise series (modified Wolffram rise – MWR ). Following Wolffram’s model, the

modified Wolffram supply equation is

0 1 2t t tQ MWR MWF (8.8)

The responses of supply to price changes in both models Wolffram and Modified

Wolffram are presented in Figure 8.1. It should be noted that the coefficient of tMWR

no longer presents the response of output to every price rise, but the response to price

increase beyond the previous maximum.


165

Figure 8.1: Hypothetical response overtime of Wolffram model and Modified

Wolffram model

Source: Traill et al (1978)

Traill et al (1978) point out that using the maximum price in all previous periods to

specify rise and fall price series may cause the “eternal asset” problem. The method of

splitting the price data are only relevant in the short-run, as in the long-run depreciation

of crop specific assets will erode the asset fixity. However, when generating the

maximum price empirically, there is no historical limit because price has to rise above

its previous maximum before there is an elastic response. This implies that, once

bought, the asset exists eternally. This specification will cause estimation problems if

there is a high price early in the series that is not surpassed and the price rise effectively

becomes a constant. To overcome this difficulty, Burton (1988) suggested using the

“window” technique. This technique defines the previous maximum price as the

maximum level occurring in only n previous years, not all previous years and n is

determined empirically. This method ensures that at some point historically high levels

of investment cease to have an effect on current output decision. With the window

technique, the price difference can be expressed as:

max, max, max, max,

1 1

max,

n n n n

t t t t

t n

t t

p p if p pp

p p otherwise (8.9)

And max, max, max,

1

0

n n n

t t t t

t

p p if p pp

otherwise (8.10)

Where

max,

1max , , ,n

t t t t np p p p

In addition, Burton (1988) indicated that the introduction of a dynamic response makes

the price partitioning method used in the modified Wolffram technique invalid, as a

peak in the price series can only specify a maximum desired asset level, and not

necessarily the capital assets held on the farm. Thus, the modified Wolffram technique

should not be used in conjunction with a distributed lag on the price, as it produces


166

incorrect signals of excess capacity. In addition, Burton suggested a new model based

on partial adjustment with a comparison between desired output and maximum level

output in previous periods. The model is:

*

3 4t tS P (8.11)

1 2 1( )adj

t t t tS S S S (8.12)

max * max

1(1 )( )adj

t t t tS S D S S (8.13)

where *

tS is the desired output at time t, tP is the price at time t, tS is the output level at

time t, max

tS is the maximum level in n previous period. adj

tS denotes adjusted desired

output, tD is dummy variable taking a value of 1 if *

tS < max

tS and 0 otherwise.

By substituting (8.11) and (8.13) into (8.12), the supply response function is given as:

max

2 3 2 4 2 1 2 1 3 4(1 ) ( )t t t t t tS P S D S P (8.14)

Granger and Lee (1989) introduce an alternative approach using the error term. This

model is called the Error Correction Model (ECM) and instead of partitioning price, the

ECM splits the error correction term 1tu into 1 1max( ,0)t tu u and

1 1min( ,0)t tu u .

The error correction term ( 1tu ) is derived from the regression:

0 1t t ty x u (8.15)

1 1 0 1 1t t tu y x (8.16)

The reduced structure of an asymmetric error correction model is given by:

0 1 2 1 3 1 4 1 5 1t t t t t ty x u u y x (8.17)

However, according to Wolffram (2005b), the asymmetric relations cannot be estimated

with positive and negative component of the split error correction term ( 1tu ) from the

ECM proposed by Granger and Lee (1989). The reason is that the sign-separation of

1tu based on its sign does not correspond with asymmetry changes in 1ty . Besides,

Wolffram (2005b) criticizes the symmetric as well as asymmetric ECM proposed as not

completely specified if the x-variable with level data and the time lag t-1, on which the

co-integrating regression is based, is not included in the function. Neglecting these


167

variables causes the parameters to be biased. Alternatively, Wolffram (2005b) suggests

the calculation of error correction terms for 1ty and

1ty derived from 1ty enabling the

quantification of asymmetric relations within ECM. The 1ty and

1ty variables are

given as

1ty = 1ty

if 1 2 0t ty y until 1 2 0t ty y

, otherwise 0

1ty = 1ty

if 1 2 0t ty y until 1 2 0t ty y

, otherwise 0

Thus,

1 1 1t t ty y y (8.18)

According to the calculation of error correction term 1tu , the following co-integrating

relations are defined for the 1ty and

1ty as:

1 0 1 1 1

new

t t ty a a x u (8.19)

1 1 0 1 1

new

t t tu y a a x (8.20)

Similarly

1 0 1 1 1

new

t t ty a a x u (8.21)

1 1 0 1 1

new

t t tu y a a x (8.22)

Substitute 1ty and

1ty from (8.19) and (8.21) to (8.18), the transformed equation is

given as:

1 0 1 1 1 0 1 1 1

new new

t t t t ty a a x u a a x u (8.23)

This is similar to the Granger and Lee approach but with a different error correction

term. The difference between the two approaches is clarified by Wolffram (2005a)

when he applied different methods to analyze price transmission between a wholesale

price for pork in north-western Germany and the producer price.

8.2.5. Previous Studies on Supply Response of Coffee in Vietnam


168

To date there have been only few econometric analyses of the supply response of coffee

in Vietnam. Ha and Shively (2008) used multinomial logistic regression model to

examine the responses to a drop in producer coffee prices of smallholder coffee farms in

the Central Highlands. They found that farmers responded to price drop by different

ways: no response, reductions in use of purchased inputs, changes in land use, and

responses aimed at enhancing liquidity through off-farm work or borrowing.

Tien (2006) analysed supply response in the Central Highland region and concluded that

the Nerlovian model was the best representation. The estimated supply equation for the

Central Highland region is:

1 1ln 1.3 0.28 0.91t t tA P A

where tA is area of coffee in year t, and tP the farm gate price in year t.

However, due to data limitations, Tien only used a time series data for one region in

Vietnam from 1990 to 2004, so the number of observation is very small.

The other study to estimate the supply function of coffee was prepared by EDE-

IPSARD (2007). This study estimates the coffee supply function to forecast supply. The

study used time series data for five main coffee producing provinces (Dak Lak, Kon

Tum, Gia Lai, Lam Dong and Dong Nai) and “all other” provinces in Vietnam over 20

years (1986-2005). The area model was given as:

1 2ln( ) 2.12 0.69 0.17 0.13T

t t t tA A E P P

where tA is area of coffee in year t; 1tA is lagged area; tP is the real FOB price of

coffee in year t; and TE is the exponential trend.

This model has a high level of fit and estimates are statistically significant. However,

the exponential trend term forces a trend on the data.

More importantly, both studies (Tien 2006; EDE-IPSARD 2007) assumed that supply

response of coffee area was symmetric. The following section provides an empirical

study on the supply response of coffee in Vietnam. Both reversible and irreversible

supply response of coffee will be analyzed.


169

8.3. Empirical Model of Coffee Supply Response in Vietnam

8.3.1. Data

Table 8.1 gives the data used in this chapter. Coffee production data are collected by

General Statistical Office of Vietnam (GSO). Data on prices (coffee and fertilizer) are

from Ministry of Agriculture and Rural Development (MARD) and Institute of Policy

and Strategy for Agriculture and Rural Development (IPSARD).

Table 8.1: Data series and source

Data series Time Source

National coffee area & production 1985 - 2006 GSO

Coffee area & production in Dak Lak province 1985 - 2006 GSO

Coffee area & production in Gia Lai province 1985 - 2006 GSO

Coffee area & production in Lam Dong province 1985 - 2006 GSO

Coffee area & production in Kon Tum province 1985 - 2006 GSO

Coffee area & production in Dong Nai province 1985 - 2006 GSO

Coffee area & production in other provinces 1985 - 2006 GSO

FOB coffee price 1986 - 2006 MARD-IPSARD

Domestic urea price 1986 - 2006 MARD-IPSARD

National consumer price index 1986 - 2006 GSO

Agricultural land by provinces 1986 - 2006 GSO

US_CPI 1986 - 2006 World Bank

Source: Author’s summary

8.3.2. Model Results

The supply function for coffee in Vietnam is estimated in logarithmic form with the

logarithm of coffee area as the dependent variable. The use of the logarithmic form can

help to infer directly the impact of explanatory variables on coffee area as estimated

coefficients are elasticities. Several approaches reviewed in the previous section are

applied to investigate whether coffee supply response to price is symmetric (reversible

supply) or asymmetric (irreversible supply).

Because the replanting time series data in Vietnam is unavailable, application of models

with replanting area cannot be used. Instead, area is used. To analyze the coffee supply

response in Vietnam, a ‘mother’ model (or full model) which nests potential different

models, including Nerlovian, Burton and Modified Wolffram/or Wolffram models is

developed. This ‘mother’ model for coffee supply is:


170

3 4 6 5 2 1 7 2

max

1 3 4 5( )

t t t t t t

t t t t i i

S a a MWR a MWF a YMA a S a S

a D S a a P a YMA D

(8.24)

where tS denotes logarithm of coffee area in year t . tMWR and tMWF are log of price

rise and log of price fall in Modified Wolffram model. tYMA is the yield of mature

coffee area in year t . max

tS is maximum logarithm of coffee area in a window of length

n years. In the empirical model of supply response in Vietnam, the maximum coffee

area is defined as area in previous years. However, max

tS can be imposed by different

lengths n . tD is a dummy variable and equal to 1 if max

3 4 5t tS a a P a YMA and 0

otherwise. tD shows the impact of the difference between the maximum area and the

desired area on variation of coffee area. iD is the provincial dummy variable. Note that,

this `mother’ model is not consistent with any individual model, but it provides a basis

for comparing alternative specifications, which can be identified by parameter

restrictions.

In this model, yield of mature area (YMA) is added into the area response model. The

YMA is a measure of yield per hectare corrected for the area of immature trees. It is

measured by the ratio of coffee output and equivalent mature area (EMA). The coffee

tree achieves mature yield (maximum yield) from 8th

year to 16th

year. Thus, the mature

area equivalent for other coffee groups out of 8-16 age groups is calculated as follows:

aa a

m

YEMA Area

Y

where aY and aArea are the yield and area of coffee at age a; mY is yield at mature age.

This ‘mother’ model in (8.24) can be restricted down to the Nerlovian model, Burton

model, Modified Wolffram model. The ‘mother’ model becomes the Nerlovian model

by imposing 4 6a a and 1a =0. In this case, model is given as:

3 4 4 5 2 1 7 2t t t t t t i iS a a MWR a MWF a YMA a S a S D (8.25)

The Burton model occurs if 4 6a a and 7a =0. The expression of the Burton model is

now:


171

3 4 4 5 2 1

max

1 3 4 5( )

t t t t t

t t t t i i

S a a MWR a MWF a YMA a S

a D S a a P a YMA D (8.26)

The case of 1a =0, the full model becomes the Modified Wolffarm model. In this case,

the model is:

3 4 6 5 2 1 7 2t t t t t t i iS a a MWR a MWF a YMA a S a S D (8.27)

The Wolffram model has the same form to the Modified Wolffram model, except that

the price rise and fall by Modified Wolffarm model presented in (8.8) are replaced by

price rise and price fall in (8.7)

The estimation of different models is tested by non-linear regression. The procedure for

estimating the full models includes several steps.

Step 1: Defining n. Defining the window length determines the relevant maximum area

in that period,

Step 2: Estimating the function 3 4 5t t t tS a a P a YMA e and predicting the fitted

value *

3 4 5t t tS a a P a YMA from the regression

Step 3: Generating dummy variable tD . D takes 0 value if St* <St-1, otherwise D is

equal to 1

Step 4: Running the full model using non-linear regression method:

3 4 6 5 2 1 7 2

max

1 3 4 5( )

t t t t t t

t t t t i i t


a D S a a P a YMA D u

Step 5: Substituting the new value of 3a, 4a

and 5a from Step 4 into fitted equation

3 4 5t t tS a a P a YMA to get new value of

*

3 4 5t t tS a a P a YMA

Step 6: Calculating tD again and substituting it into the equation

3 4 6 5 2 1 7 2

max

1 3 4 5( )

t t t t t t

t t t t i i t


a D S a a P a YMA D u


172

and run this regression again

Step 7: Iterating the steps above until 3a, 4a

, 5a and i converge

The procedure for estimating other models are the same as for the full model. However,

some parameters in the full model are restricted in particular models.

The estimated results of different models are presented in Table 8.2

Table 8.2: Estimated results from different models

Parameters

Variable

names ‘Mother’

model

Nerlovian

model

Burton

Model

Wolffram

model

Modified

Wolffram

modela

/a3 Constant 1.00* -0.31 -0.18 0.97* 1.58*

/a4 Price rise 1.06* 0.23* 0.14** 0.35* 0.87*

/a6 Price fall 0.17* 0.23* 0.14** 0.13* 0.21*

/a5 Yield 0.20* 0.12* 0.05 0.23* 0.12*

/a2 Lag(1) 0.73* 1.02* 1.04 0.87* 0.84*

/a1 Asy. Adjb

0.03 ---- 0.02 ---- ----

/a7 Lag (2) 0.2 0.04 ---- ---- ----

/D2 Gia Lai 0.21 -0.04 -0.02 0.20* 0.26*

/D3 Dak Lak 0.41 -0.15 -0.09 0.42* 0.45*

/D4 Lam Dong 0.29 -0.09 -0.05 0.29* 0.34*

/D5 Dong Nai 0.14 -0.13*** -0.08 0.17** 0.12**

/D6 Other 0.18 -0.07 -0.04 0.19* 0.27*

Adjusted R2 0.99 0.98 0.98 0.98 0.99

P_value of autocorrelationc

0.013 0.01 0.01 0.01 0.06

Adjusted R2 0.99 0.98 0.98 0.98 0.99

Note: *significant at 1%; ** significant at 5%; *** significant at 10%

a with window length of 6 years;

b asymmetric adjustment

c: autocorrelation test for panel time series data using Wooldridge test. H0: no first order

autocorrelation

The ‘mother’ model results show that the estimate of 1a is not significant. This means

that the asymmetry variable can be dropped from the model. Similarly, the estimate of

7a is not significant as well. This indicates the area lagged by two years does not affect

to the current coffee area in the full model.

The results of the Nerlovian model and the Burton model violate economic theory with

the estimates of 2a in both Nerlovian (1.02) and Burton models (1.04) being greater

than 1. This means that the current area of coffee will increase continuously holding


173

other variables constant. In addition, the Burton model does not show the impact of area

adjustment to the previous maximum because the estimate of 1a is not significant.

The results of the Wolffram model and Modified Wolffram model (with window length

of six years33

) indicates statistically significant estimates. According to the Wolffram

model, if price increases by one percent, the coffee area will increase correspondingly

by 0.35 percent. However, coffee area reduces only 0.13 percent in response to a one

percent reduction of output price. However, as mentioned earlier, the Wolffram model

is not consistent with economic theory. In addition, the regression results indicate the

autocorrelation problem.

The modified Wolffram model with window length of six years gives the best estimate.

The difference between price rise and fall coefficients in the Modified Wolffram model

is positive. The response to a price rise (with coefficient of 0.87) above the previous

maximum is about five times as large as the response to price fall (coefficient of 0.21

only). The Modified Wolffram model does not have autocorrelation problems.

The t-test proves that the difference between coffee acreage responses to a price rise

against a price fall is statistically significant34

. All coefficients of the provincial dummy

variables are also different from zero and statistically significant with confidence level

of 10 percent. Given differences in size of the provinces, this is not surprising.

However, the response of coffee area to price rises and falls and to the lagged dependent

variables may or may not be the same. To test the hypothesis of the same response to

price as well as lagged dependent variables, we run unrestricted Modified-Wolffram

model in which provincial dummy variables are included in the Modified-Wolffram

equation with window length of 6 years for all independent variables. The results of this

model are presented in Table 8.3 and shows that coffee acreage responses to price rise

are not different among provinces while there is a significant difference of response to

price fall between Dak Lak and Lam Dong with Kon Tum.

33

Ihe study also estimated coefficients in the Modified Wolffram model with different window lengths.

The model with window length of six years gave the best estimates. The results of other Modified

Wolffram models are presented in Table C1 in the Appendix C. 34

To test the difference, author uses the “test” command in STATA after regression

test mwr =mwf

( 1) mwr- mwf = 0

F( 1, 115) = 36.05

Prob > F = 0.0000

With Prob>F =0.00, it strongly indicates the significant difference between the coefficients of price rise

and price fall.


174

Table 8.3: Results for testing the difference of coefficients among provinces,

Modified Wolfram model with window=6

Independent variables Coefficients P-value

Constant 0.45 0.74

Constant dummy vars.

Gia Lai 2.54 0.14

Dak Lak 2.86 0.18

Lam Dong 3.73 0.05

Dong Nai 1.61 0.37

Other 1.03 0.54

Lnareat-1 0.99 0.00

Lnareat-1 dummy vars.

Gia Lai -0.30 0.16

Dak Lak -0.28 0.22

Lam Dong -0.40 0.07

Dong Nai -0.18 0.38

Other -0.12 0.56

lnYMA 0.08 0.47

lnYMA dummy vars.

Gia Lai 0.03 0.81

Dak Lak 0.07 0.65

Lam Dong -0.01 0.95

Dong Nai 0.05 0.70

Other -0.02 0.87

MWR (n=6) 1.03 0.00

MWR dummy vars.

Gia Lai 0.60 0.27

Dak Lak -0.22 0.67

Lam Dong 0.96 0.07

Dong Nai -0.38 0.37

Other -0.42 0.36

MWF (n=6) 0.37 0.00

MWF dummy vars.

Gia Lai -0.26 0.12

Dak Lak -0.32 0.04

Lam Dong -0.30 0.05

Dong Nai -0.17 0.22

Other -0.21 0.16

R2

98.7

Prob > F 0.00

Note: In the regression, Kontum is the reference province

A general F-statistic is calculated to test the overall significance of this unrestricted

Modified-Wolffram model and the restricted Modified-Wolffram model (as its results

are presented in Table 8.2). The general F-statistic is given by

( ) /

/ ( )

R U

U

SSE SSE JF

SSE T K


175

where J is the number of hypotheses, T K is denominator degrees of freedom, RSSE

is the restricted sum of squared errors, USSE is the unrestricted sum of squared errors.

From regression results of both models, the F-test statistic value is only 1.07, smaller

than F critical value (F(20,84) ~=1.7). This concludes that the overall significance of

restricted Modified-Wolffram model is not statistically different from the unrestricted

one.

The elasticities of coffee area with respect to price fall and rise are summarized in Table

8.4 for three models. The estimates of short-run elasticities to a price fall produced by

the three models are not much different. In contrast, the estimates for short-run

elasticities with respect to price increases are quite different, increasing from the

Wolffram model (0.35) to the full model (1.06).

Table 8.4: Elasticities of coffee acreage to price

Model Short-run price elasticities Long-run price elasticities

Price fall Price rise Price fall Price rise

‘mother’ model 0.17 1.06 3.9 3.9

Wolffram 0.13 0.35 2.7 2.7

Modified Wolffram (n=6) 0.21 0.87 5.4 5.4

Source: Summary from regression results

From previous discussion, it was concluded that the Modified Wolffram (n=6) is the

best choice of the three approaches for analyzing the asymmetric response of coffee

area in Vietnam in 1985-2006. The model is consistent with economic theory and

produces very high goodness of fit and statistically significant levels. Figure 8.2

presents the fitted and actual coffee area by provinces and it shows that the fitted value

and actual area are very similar in all provinces.


176

Figure 8.2: Fitted and actual area from Modified Wolffram model (ha)

Kon Tum province

0

500

01

00

00

150

00

1985 1990 1995 2000 2005year

actual area fitted value

Gia Lai province

0

200

00

400

00

600

00

800

00

1985 1990 1995 2000 2005year


Dak Lak province

0

100

00

02

00

00

03

00

00

0

1985 1990 1995 2000 2005year


Lam Dong province

0

500

00

100

00

01

50

00

0

1985 1990 1995 2000 2005year


Dong Nai province

100

00

200

00

300

00

400

00

500

00

1985 1990 1995 2000 2005year


other provinces

0

200

00

400

00

600

00

800

00

1985 1990 1995 2000 2005year



177

8.4. Conclusion

Studies of supply response play an important role for farmers and policy makers. They

can help farmers use their resource more efficiently. More importantly, the

understanding of supply response can support policy-makers to allocate production

resources and achieve targets. In addition, the supply response equation is useful in

forecasting future supply. However, studies on coffee supply response in Vietnam are

still limited and all of them have assumed a symmetric response when estimating the

coffee supply function.

This chapter uses the “positive approach” to estimate the supply response of coffee in

Vietnam. Both symmetric and asymmetric responses of coffee area in Vietnam are

estimated with different econometric models.

The Nerlovian model with reversible supply response was tested but it had the non-

stationary problem. The estimates are not statistically significant. The symmetric model

cannot explain the variation of coffee area in Vietnam in the past.

Application of the Wolffram model and the Modified Wolffram model investigated

whether the coffee area in Vietnam has responded asymmetrically. The Modified

Wolffram model with window length of six years gave the best estimate. The output of

the model shows that if the price rises by one percent, the coffee area will increase by

0.87 percent, while if the price falls by one percent, the coffee area reduces by only 0.21

percent. However, the elasticity of price is much larger in the long-run (5.4). The

results of testing the overall significance of the model indicate that the response of

coffee area is not different among provinces. The estimated model of coffee predicts an

appropriate value for all provinces.

The optimal models in previous chapters provide insights into individual farmer’s

decisions, especially replanting and cutting decisions. The output of optimal models

showed that individual farmers decide to cut and replant at different “trigger” prices.

This cutting/replanting gap in farmer’s decision explains the asymmetry of coffee

supply response at the aggregate level.

Chapter 9.Conclusions

178

CHAPTER 9. CONCLUSIONS

This final chapter comprises four main sections. The first section provides a brief

summary of the background and objectives of the thesis. The second section presents

the main findings derived from the different models in the study. The third section

discusses some limitations of the study and other complexities not addressed in this

thesis. Finally, the opportunities for further work are mentioned.

9.1. Background

Coffee is an important crop in Vietnam’s agriculture sector. It is the second largest

export agro-commodity in Vietnam after rice. In addition, coffee plays an important role

in labour absorption in rural areas. In the peak season, the coffee sector employs about

800,000 workers (The World Bank, 2002).

Following the implementation of the “Innovation” policy in 1986, the coffee area in

Vietnam increased rapidly, from only 50,000 ha in 1986 to about 600,000 ha in 2000

(GSO, 2001). The rapid expansion of coffee area and production has made Vietnam

become a significant exporter: currently, Vietnam contributes over approximately 40

percent of Robusta and 13 percent of all coffee traded on the world market.

Despite its rapid expansion, the coffee sector in Vietnam faces a number of issues. First,

the coffee sector is dominated by small households with over 60 percent of households

having less than one ha of coffee land. Secondly, about one-fourth of coffee households

are poor and 30 percent of farmers are from one of Vietnam’s ethnic minorities. In

addition, coffee households are highly specialized and thus it is not easy for them to

diversify their income. Furthermore, the coffee price is highly volatile and depends

heavily on the international market. The price crisis in early 2000 adversely affected the

whole coffee sector in Vietnam, but especially coffee farmers. The coffee price received

at that time by coffee farmers did not cover the variable costs, thus many farmers had to

cut down coffee trees and switch to other crops such as maize. During three seasons

(from 2001/2002 to 2004/2005), over 100,000 ha of coffee trees in Vietnam were

uprooted.


179

The reduction of coffee area in the early 2000s was in line with policies from both

Central and Local Government. At that time, the Government advised and supported

farmers with poor lands or households with old coffee gardens to clear trees and switch

to other crops. The Government partly subsidized uprooting costs and provided

substitute crops. Cutting coffee trees in response to price reductions is a complex

decision as establishment costs are high and a farmer may regret a decision to cut if a

price fall is temporary. To avoid the costly practice of cutting too early and to support

farmers in their decision-making process and planners in policy formulation, it is

necessary to have a thorough understanding of the optimal behavior for coffee

households with respect to price variation. To this end, this study examined two main

problems: (i) identifying the optimal price for cutting and replanting coffee trees so that

farmers can attain the maximum expected net present value (ENPV) when price varies

randomly and (ii) estimating the aggregate supply response function for coffee in

Vietnam.

Solving the first problem employed a number of models to:

(i) determine the optimal cutting and replanting rule for coffee farmers in Vietnam to

maximise their expected NPV from “land use choice”,

(ii) investigate to what extent poor farmers lose income from deviating from the optimal

rules because of cash constraints and

(iii) analyze how much farmers can improve their income if they follow an optimal

short-run yield response.

To identify the optimal cutting and replanting price of coffee farmers, the study

develops optimal models using the fixed form optimization approach. The fixed form

method specifies particular functional forms for cutting price and replanting price rules.

The purpose of the models is to identify the cutting and replanting price to maximise the

expected NPV from “land use choice” of coffee farmers. The term “land use choice”

refers to the planting decision of coffee farmers on their land: they can grow coffee or

switch to maize if the price of coffee is too low.

To solve this problem, the study develops four main optimal models. The first model is

the Fixed Yield model (FY model). This model identifies the optimal cutting and

replanting prices to achieve the maximum expected net present value (ENPV) while


180

coffee yield is assumed to vary by age of trees but fixed at a given age. The second

model, Fixed Yield- Cash Constraint Model (FY-CC model), is an extension of the

Fixed Yield model. This model integrates the liquidity constraint of coffee households

when they make their decision. In the FY-CC model, the cutting/keeping or replanting

decision of coffee farmers does not only depend on price levels but is also based on the

availability of cash after living costs are subtracted. The third model, Variable Yield

Model (VY model), is based on the FY model but it goes a further step to investigate the

farmer’s decision when the yield in the short-run can change. The change in inputs

affects the yield of coffee as well as production cost and thus influences the optimal

decision of farmers in terms of the cutting and replacement rules. The fourth model is

the Variable Yield–Cash Constraint Model (VY-CC model). The structure of the VY-

CC model integrates the two previous variants: both cash constraints and variable yields

are included.

Coffee is a multi-year crop and the cutting/keeping or replanting decisions in the current

year affect the expected income of households in subsequent periods. In addition, the

objective function of the optimal models is to maximise the expected NPV under price

uncertainty. Thus, the conventional stochastic dynamic programming (DP) is seemingly

a useful technique to solve this problem. However, the application of DP for coffee

models in this study faces the problem of “the curse of dimensionality” because the

optimal models cover a period of 50 years. At a given stage, coffee farmers have to

choose different options: (i) keeping coffee, (ii) cutting standing trees and replace by a

substitute crop (maize), (iii) replanting coffee or keep growing maize if land is being

used for maize. In addition, the objective of the present modeling approach is more

complex because the aim is to also consider the effect of cash constraints and short-run

responses on optimal cutting and replanting rules. Thus, application of the fixed form

approach with a grid search method is a simpler way to solve the coffee models to find

out the optimal rules and the maximum expected NPV. Because of the impact of age of

the tree on current and expected future yield, we estimate optimal rules as a function of

the age of the tree.

The estimation of the coffee supply response function is based on historical data at

provincial levels of Vietnam from 1986 to 2006. Different symmetric and asymmetric

forms of supply function were estimated to find the best-fit function.


181

9.2. Key results

9.2.1. Response at the Farm Level

The analysis of response at the farm level explains the optimal decisions for an

individual farmer, especially the replanting and cutting decisions.

9.2.1.1. Cutting and Replanting Decision

Although details of the optimal rules vary among different models, the results of all

optimal models indicate that farmers have different trigger prices for cutting and

replanting. This asymmetric response of individual households leads to an asymmetric

supply response at the aggregate level. The cutting/replanting gap means farmers

usually switch out of coffee production more slowly than commencing and expanding

production. The asset fixity problem will lock them into the coffee sector and, should

prices go down, they may lose money. By contrast, farmers should not be in a hurry to

cut the coffee down and switch to other crops.

In addition, there is an obvious relationship between the cutting rules and age of coffee

trees. In general, optimal cutting price for trees which are at the age of starting to

produce cherries (5-6 year old) are lowest. However, results from the optimal models

indicate that farmers should not cut their trees down even if they are mature (up to 12

year old), even if the price is very low. Furthermore, farmers should never cut their

coffee earlier than its biological limit if the price of coffee at that time is very profitable.

That is because they would have to forgo yield for some years, and given the price

volatility of coffee it is better to get the benefit of existing yield at high prices, than to

wait for the new tree to mature because prices might not be as high then.

The optimal cutting price is significantly influenced by the age of the coffee tree. Thus,

the model with age-dependent optimal cutting price generates higher income as

compared to a constant cutting price. This finding lends support to the type of approach

used here, compared to a real options approach applied by Luong and Lorrent (2006)

which focuses on a fixed cutting and replanting price. However, the model results are

almost unchanged among different fixed forms of CP (cubic CP, quadratic CP and

quadratic with price change effect CP)


182

9.2.1.2. Impact of Cash Constraints on Farmer’s Behaviors

Farmers in developing countries like Vietnam have volatile and low incomes. They

often suffer from different risks caused by bad weather, agricultural price shocks,

business failure and illness. A shortage of cash for investment causes many problems

for rural households. About 25 percent of coffee households in Vietnam are poor and

they often need support from credit organizations. When farmers have cash problems

they cannot optimize their long-run income and they have to change cutting or

replanting decisions to satisfy their short-term cash needs. The FY-CC model results

indicated that when poor coffee farmers have a cash-constraint, they lose about 15

percent of their income if they follow the optimal rules of non-poor farmers. This is

because they do not have the cash to replant coffee when the price rises. If they use

rules derived explicitly for accounting for the constraint, there is a small change in CP

but a significant increase in RP. Generally, the poor households are more likely to cut as

compared to the non-poor. Furthermore, the poor farmers usually wait for significantly

higher prices before deciding to replant. In addition, under the new optimal rules to the

constraint, the income of poor farmers is about 11 percent lower than that of non-poor

farmers derived from the FY model.

From the optimal rule, optimal results indicate the frequency cutting decision or cutting

percentage of coffee farmers with respect to different age groups of trees. The cutting

percentage is the percentage of cases in which farmers actually cut their coffee trees

down at optimal rules. The cash constraint problem has a remarkable impact on the

cutting percentage of poor households, especially for those whose coffee trees are still

young (less than 4 year old). As indicated by the FY-CC model, the actual cutting

percentage of coffee trees under 4 year old due to the cash constraint is about 10

percent, and this number reduces to only 2 percent when trees become older. Due to the

cash constraint, farmers cannot optimize their decision so they earn less than they could.

This implies that credit supports are important for coffee farmers but the priorities

should go into new households or farmers with young trees.

The amount of loan available for the poor household has a significantly impact on the

farmer’s income and behavior. In general, poor farmers can get a higher income with

bigger loans. In many cases, it is inefficient to support poor households with a loan that

is too small, as it has little impact on incomes, as there appear to be threshold effects.

Moreover, the model output also shows that, if the annual loan increases to $1500, poor


183

households can nearly optimize their investment, and their decision and expected

income is very close to the non-poor farmers.

The importance of loans varies depending on the age of coffee trees the farmer has at

the start of the period. The amount of annual loan is more important for farmers with the

young trees, especially for non-productive trees. With the mature trees, the impact of

loans becomes less important.

9.2.1.3. Change of Farmer’s Decision with Short-run Response

The liquidity constraint has a significant impact on the optimal decision of the coffee

farmers, especially for those whose trees are still young. Farmers cannot optimize their

long-run income if they are poor. The poor farmers often cut earlier and wait longer

before replanting.

The output of the optimal models shows that the optimal decisions of farmers

significantly change if they make efficient, short-run changes in input use in response to

output price changes. The change of input influences the yield of coffee trees. The

ability to respond in the short-run has a significant impact on the farmer’s planting and

cutting decision and on income. The income of coffee farmers increases significantly

when farmers can optimize their response of input use to output price. This is true

because they do not apply as much input in low-price years. With the short-run

response, coffee farmers can increase their expected income by over 30 percent of their

expected income when compared to the case without a short-run yield response at the

same average yield. In addition, with the presence of a short-run response, coffee

farmers are much less likely to cut and more likely to replant coffee. With short-run

response, the non-poor farmers optimize their decision to replant at a price of $0.51 per

kg. The poor farmers wait for a higher price and they decide to replant at a price of

$0.59 per kg of coffee. These replanting prices are much lower when compared to the

optimal RP of the poor and non-poor cases without the short-run response ($0.74 and

$1.4 per kg, respectively).

Significant improvements in profit that can be achieved as the result of being able to

change input use in the short-run implies that it would be valuable for farmers to be

educated about the benefit of short-run response.


184

9.2.1.4. Impact of the Profitability of the Substitute Crop

In the optimal models, maize is assumed to be the substitute crop when coffee trees are

cut down. The maize profit is constant in the optimal models. The sensitivity analysis

from the FY model in Chapter 4 concluded that if the replacement crop profit increases,

farmers are more likely to cut and less likely to replant and vice versa, as one would

expect. However, the model simulation shows that if the annual maize profit increases

by 20 percent, the ENPV will increase by only 2 percent.

The switching decision of coffee farmers to annual crops such as maize or rice is mainly

carried out to overcome a cash shortage, when there is insufficient cash for food. Thus,

this issue relates closely to credit and other subsidy policies. If the price is too low but

the household budget (including loans and household income) is sufficient for farmers

to cover living expenditure, they can continue producing coffee while waiting for higher

prices.

9.2.2. Coffee Supply Response at Aggregate Level

Empirical results in this study show that the supply response of the coffee area in

Vietnam is asymmetric. Farmers respond much more quickly to price rises than price

falls. The short-run elasticity of coffee area to output price rise (0.87) is much higher

than to the price fall (0.21). The irreversible supply response of coffee is consistent with

the asymmetry of individual farmer’s behavior in cutting/replanting found in the

individual-farmer decision models.

The asymmetric response of coffee in Vietnam is similar to the pattern of world coffee

supply found by Olsen (2005). According to Olsen, the supply response of world coffee

to price does not comply with standard economic theory, as it is asymmetric. The

reasons that coffee farmers in developing countries do not want to move away from the

coffee sector, even when prices are low are:

(i) possibility of subsistence farming in conjunction with coffee enables farmers to

survive, with an expectation to earn higher income from coffee,

(ii) lack of alternative income sources, and

(iii) low education


185

Olsen (2005), also identified that the existence of a “fixed asset” is an important factor

which leads to the asymmetric supply response. Low salvage values motivate producers

to keep growing coffee even when output prices are low because of high acquisition

costs, which again is consistent with farm-level models developed in this thesis.

9.3. Policy implications

The analysis and results from the farm level and supply response models provides

insights into policy implications for Vietnam’s Government

Coffee farmers in Vietnam and other coffee producing countries clearly perceive that

the price of coffee is very volatile. Sometimes, the price of coffee drops below the

production cost. However, the government can provide the information about when its

optimal to switch out. At some times it is optimal but it might be difficult for farmers to

identify the optimal triggers. The intervention could either be information as to when it

should happen, or direct incentives such as grants for removal if that is seen as needed.

So assistance of Government to provide such information would be good for helping

farmers to optimize their decision.

In addition, the government could encourage coffee farmers with young trees under 10

years old to not cut coffee trees even if the price of coffee is very low. Historically,

coffee prices increase after several years of reduction thus if farmers cut down the

young coffee trees they will lose money from future coffee sales. In addition, the

models shows that farmers can make better decisions based on expected changes in

prices (i.e. they have better information about what is likely to happen) and if they do

not have that then presumably they will be in difficulties. Thus, Government could

provide information about price forecast for coffee growers so the growers can base

their investment decision on that information.

The aggregate supply model shows that response of coffee to coffee prices is

asymmetric, which is consistent with the farm level models. So the government should

advise farmers to not cut down their trees or hurriedly switch to other crops if coffee

price is declining, because the price of coffee will recover so farmers will lose their

income from coffee production.

Coffee farmers should perceive clearly that the yield of coffee trees do not always

increase accordingly to greater input use. Overuse of chemical fertilizer not only causes


186

high expenditure but also lowers the yield of coffee trees. Furthermore, the model

shows that farmers can save their cost and increase their profit with better response in

input use to changes in output prices. So the diffusion of information by the

Government about the benefits of short run response in use of fertilizer is also needed.

The price of coffee is very volatile, and periods of low prices may induce farmers to cut

trees, especially if they have cash constraints. This suggests that policies to support

farmers via a stabilized market price would help make farmer’s income stable. A floor

price, or subsidized credit policy for coffee exporters can be good measures but it is

experienced in the past that farmers could not get much benefit from such policies. The

results from the optimal models with cash constraints show that credit policy for coffee

smallholders plays a very important role to help them overcome low price years and

optimise their investment decision. The provision of credit facilities would assist

farmers in avoiding being forced to cut because of cash flow problems. However, it is

necessary to be concerned about the size of loans for farmers in a credit program.

Small/inefficient loans are not sufficient to allow them to improve their decision and

cannot improve their livelihood. The model shows that there is a need to make sure that

the loan for coffee farmers is provided at the efficient level. It means that there is no

point if it is too small because farmers cannot afford to invest, but also there comes a

point where larger loans provide little additional benefit.

9.4. Limitations

Despite numerous insights provided by this study, it has a number of limitations.

First, the optimal rules being identified in the optimization models are only valid for the

existing price simulation. This does not refer to the specific price series themselves, but

the assumptions about the mean and variance of the price distributions. If there was a

change in the behavior of these time series then it is likely that the rules would alter.

Second, all fixed forms for cutting prices and replanting price in optimal models (age

dependent quadratic CP, quadratic CP with price change effect, age dependent cubic

CP) are specific forms to present the relationship between age of coffee trees and CP.

However, those forms may not be the best/optimal form for CP in reflecting the cutting

decision and age of trees. There may exist some alternative specifications that could

improve expected returns for farmers. This is equivalent to the issue of identifying a


187

local optimal in a conventional programming problem: although the search through the

existing set of functions suggests that the optimal has been identified, it cannot be

guaranteed.

Third, due to the unavailability of time series farm-gate price data by different

provinces, the study used FOB price as a proxy to estimate the supply function of coffee

at provincial level. In addition, FOB prices used in the study were average price at

national level. Thus, this may not reflect accurately the regional price received by

farmers.

Fourth, the optimal models assumed that all trees on a property are of a single age, so

when cutting coffee farmers switch all land to maize. The limitations of this assumption

will impinge on the results for the cash-constrained poor farmers, where their behaviour

may be different if they have the capacity to run mixed-age plantations, or remove only

part of their tree stock.

Fifth, the cutting and replanting decision of coffee farmers depends on the profit of

substitute crops. Thus, the replacement decision can be a function of certain maize

returns. In the optimal models, the sensitivity analyses were undertaken to see how the

optimal rule changed when maize profits varied. More realistically, the cutting price

might be influenced by the mean and variance of the maize price.

Sixth, the price of inputs may change the decision of farmers. In the optimal models,

prices of input (including fertilizer and labour) are assumed unchanged. In practice, the

price of inputs will vary over time, and again, have some degree of uncertainty

associated with them.

Seventh, the estimation of income and expenditure in this study for the poor coffee

households is based on data of all poor households from the VHLSS2006. The data for

income and expenditure for poor coffee households is not available. This result may

bring some bias when analyzing the structure of income distributed to investment and

household expenditure.

Eighth, the estimation of coffee yield neglected factors that significantly influence the

yield of coffee such as rainfall, type of land, education of households. Furthermore, the

same yield response function in this study is used for both poor and non-poor farmers.


188

However, in reality, poor farmers usually have poorer land and their yield will respond

differently to the non-poor farmer.

9.5. Further Studies

There are some complexities which have not been addressed in this study. In the area of

optimal decision for coffee farmers, optimal models could be further improved in some

ways.

a. It would be useful to identify better empirical data on the productivity vs. age

relationship for the Robusta varieties in Vietnam. In this study, the author's assumption

of the production pattern over time is in line with suggestions in the literature. This

information is crucial to the models, since the question remains "should the government

stimulate replanting when plantations reach 22 years, or is production still economically

viable after 22 years and for how long" i.e is the assumption of a 22 year life span for

the trees appropriate.

b. In the optimal models, maize price was assumed to be unchanged. The sensitivity

analysis showed that farmer’s behavior would be changed when profit of maize varied.

An improvement to the model would be to see how maize price influences the cutting

and replanting decision by adding maize price into the function of CP and RP. This

would make the optimal models much larger and would take longer to solve.

c. The price simulations in the model were generated from a lagged price model using

the time series data. The distribution of such price simulation has an extended range.

The change of the distribution of price will change the farmer’s decision rules. It would

be useful to investigate the optimal rules with different limits of the price distribution.

d. The optimal models in this study identified the cutting and replanting price based on

the normative approach. However, it would be interesting to develop a model based on

actual data for individuals for cutting and replanting decisions, possibly derived from

farm surveys. This model would be based on the farmer’s reported data on the time

they cut down the trees, the price at which they cut, input price, other crop price and

other household’s characteristics. The output of this model could be compared with the

simulated behaviors of the normative models.


189

e. When developing the yield response function to study the impact of short-run

response to farmer’s decision, this study did not address the dynamic effect of fertilizer

use. The application of inputs identifies only the yield of coffee in that year. In practice,

use of fertilizer has carry-over effects. Another improvement to the model would

include the carry-over effects in input use.

f. This study estimated econometrically the response of the coffee area. It would be

useful to estimate the yield response function at the aggregate level based on other

variables such as output price, input price, rainfall, humidity, price of competing crops.

The integration of area function and yield function would be useful to estimate and

forecast the output function. Ideally, such a model would be developed into a partial

equilibrium model for the coffee sector in Vietnam. The partial equilibrium model

would be helpful in evaluating the impact of government policies, price change and

other factors on farmers.

References

190

References

ACKELLO-OGUTU, C., PARIS, Q. & WILLIAMS, W. A. (1985) Testing a von Liebig

Crop Response Function against Polynomial Specifications. American Journal

of Agricultural Economics, 67, 873-880.

ACTIONAID VIETNAM & ASIAN DEVELOPMENT BANK (2003) Dak Lak: A

participatory poverty and governance assessment.

AGERGAARD, J., FOLD, N. & GOUGH, K. V. (2009) Global-local interactions:

socioeconomic and spatial dynamics in Vietnam's coffee frontier. Geographical

Journal, 175, 133-145.

AKBAR, H. (2003) Carry over effect of crop management techniques and nitrogen on

the yield of maize Faculty of Crop Production Sciences., NWFP Agricultural

University.

AKIYAMA, T. & TRIVEDI, P. K. (1987) Vintage production approach to perennial

crop supply : An application to tea in major producing countries. Journal of

Econometrics, 36, 133-161.

ALVAREZ, L. H. R. & KOSKELA, E. (2004) On Forest Rotation under Interest Rate

Variability International Tax and Public Finance, 10.

ANDERSON, R. L. & NELSON, L. A. (1975) A Family of Models Involving

Intersecting Straight Lines and Concomitant Experimental Designs Useful in

Evaluating Response to Fertilizer Nutrients. Biometrics, 31, 303-318.

ANDRADOTTIR, S. (1998) A Review of Simulation Optimization Techniques. IN

MEDEIROS, D. J., E.F.WATSON, CARSON, J. S. & M.S.MANIVANNAN

(Eds.) Proceeding of the 1998 Winter Simulation Conference, 13-16 Dec, 1998.

Washington D.C, USA,.

ASKARI, H. & CUMMINGS, J. T. (1977) Estimating Agricultural Supply Response

with the Nerlove Model: A Survey. International Economic Review, 18, 257-

292.

AZADIVAR, F. (1999) Simulation Optimization Methodologies. IN FARRINGTON, P.

A., NEMBHARD, H. B., STURROCK, D. T. & EVANS, G. W. (Eds.)

Proceeding of the 1999 Winter Simulation Conference, 5-8 Dec, 1999. Arizona,

United States.

BARTON, R. R. (2009) Simulation Optimization using MetaModels. IN ROSSETTI,

M. D., HILL, R. R., JOHANSSON, B., DUNKIN, A. & R. G. INGALLS (Eds.)

Proceedings of the 2009 Winter Simulation Conference, 13-16 Dec, 2009.

Texas, United States.

BATEMAN, M. J. (1965) Aggregate and Regional Supply Functions for Ghanaian

Cocoa, 1946-1962. Journal of Farm Economics, 47, 384-401.

BEHRMAN, J. R. (1968) Monopolistic Cocoa Pricing. American Journal of

Agricultural Economics, Vol. 50, 702-719.

BELLMAN, R. (1957) Dynamic Programming, Princeton, Princeton University Press.

References

191

BELLMAN, R. E. & DREYFUS, S. (1962) Applied Dynamic Programming, Princeton,

Princeton University Press.

BERTSEKAS, D. & TSITSIKLIS, J. N. (1996) Neuro Dynamic Programming, Athena

Scientific, Belmont, MA.

BRAZEE, R. (2007) The Faustmann face of optimal forest harvesting. IN

TIETENBERG, T. & FOLMER, H. (Eds.) The International Yearbook of

Environmental and Resource Economics 2006/2007: A Survey of Current Issues

New Horizons in Environmental Economics. Edward Elgar Publishing 2007.

BUONGIONO, J. (2001) Generalization of Faustmann's formula for stochastic forest

growth and prices with Markov decision process models. Forest Science, 47,

466-474.

BURT, O. R. & ALLISON, J. R. (1963) Farm Management Decisions with Dynamic

Programming. Journal of Farm Economics, 45, 121-136.

BURTON, M. P. (1988) Irreversible Supply Functions Revisited. Journal of

Agricultural Economics, 39, 113-120.

CAP (2006) Competition under AFTA. Hanoi, Vietnam.

CAP (2008) Capital accumulation of rural households in Vietnam. Hanoi, Vietnam.

CARSON, Y. (1997) Simulation Optimization: Methods and Applications. IN

S.ANDRADOTTIR, K.J.HEALY, D.H.WITHERS & B.L.NELSON (Eds.)

Proceeding of the 1997 Winter Simulation Conference, 7-10 Dec, 1997. Georgia,

USA,.

CASTELLETTI, A., RIGO, D. D., RIZZOLI, A. E., SONCINI-SESSA, R. & WEBE, E.

(2007) Neuro-dynamic programming for designing water reservoir network

management policies. Control Engineering Practice, 15, 1031–1038.

CHANG, S. J. (1982) An Economic Analysis of Forest Taxation's Impact on Optimal

Rotation Age. Land Economics, 58, 310-323.

CHI, T. T. Q. & D’HAEZE, D. (2005) Assessment of Water, Fertilizer And Pesticide

Use For Coffee Production In Dak Lak Province. Hanoi, Vietnam.

CHI, T. T. Q., TUAN, N. D. A. & LAN, N. N. (2009) Coffee Market in 2009. Hanoi,

Vietnam.

CHLADNÁ, Z. (2007) Determination of optimal rotation period under stochastic wood

and carbon prices. Forest Policy and Economics, 9, 1031-1045.

CONG, P., SAT, C. & HARDTER, R. (2001) Response of Selected Crops to K

Fertilization on Major Soil Types in South Vietnam. IN HORST, W., SCHENK,

M. K. & BÜRKERT, A. (Eds.) Plant nutrition- Food security and sustainability

of agro-ecosystems. Kluver Academic Publishers.

D’HAEZE, D., DECKERS, J., RAES, D., PHONG, T. A. & CHANH, N. D. M. (2003)

Over-irrigation of Coffea canephora in the Central Highlands of Vietnam

revisited:Simulation of soil moisture dynamics in Rhodic Ferralsols.

Agricultural Water Management, 63, 185-202.

D’HAEZE, D., DECKERS, J., RAES, D., PHONG, T. A. & LOI, H. V. (2005)

Environmental and socio-economic impacts of institutional reforms on the

agricultural sector of Vietnam Land suitability assessment for Robusta coffee in

the Dak Gan region. Agriculture, Ecosystems and Environment, 105, 59-76.

References

192

D’HAEZE, D. A. (2004) Water management and land use planning in the Central

Highlands of Vietnam: The case of Coffea canephora in Dak Lak province.

Landbouwkundige en Toegepaste. Katholieke.

DAGDELEN, N., YILMAZ, E., SEZGIN, F. & GÜRBÜZ, T. (2006) Water-yield

relation and water use efficiency of cotton (Gossypium hirsutum L.) and second

crop corn (Zea mays L.) in western Turkey. Agricultural Water Management,

82, 63-85.

DARDD (2008) Coffee production situation in Dak Lak province. Dak Lak, Vietnam.

DEATON, A. & LAROQUE, G. (2003) A model of commodity prices after Sir Arthur

Lewis. Journal of Development Economics, 71, 289-310.

DI PAOLO, E. & RINALDI, M. (2008) Yield response of corn to irrigation and

nitrogen fertilization in a Mediterranean environment. Field Crops Research,

105, 202-210.

DIXIT, A. (1989) Entry and Exit Decision under Uncertainty. Journal of Political

Economy, 97, 621-638.

DIXIT, A. K. & PINDYCK, R. S. (1994) Investment under Uncertainty, New Jersey,

United Kingdom, Princeton University Press.

DIXIT, A. K. & PYNDYCK, R. S. (1994) Investment under Uncertainty, Princeton,

Princeton University Press.

DIXON, B. L. & HOWITT, R. E. (1980) Resource Production under Uncertainty: A

Stochastic Control Approach to Timber Harvest Scheduling. American Journal

of Agricultural Economics, 62, 499-507.

DOWLING, J. M. (1979) The Supply Response of Rubber in Thailand. Southern

Economic Journal, 45, 795-805.

EDE-IPSARD (2007) Developing a Production Forecast Model for Coffee in Vietnam.

FAUSTMANN, M. (1995) Calculation of the Value which Forest Land and Immature

Stands Possess for Forestry Journal of Forest Economics, 1, 7-44.

FREEMAN, H. A., EHUI, S. K. & A. JABBAR, M. (1998) Credit constraints and

smallholder dairy production in the East African highlands: application of a

switching regression model. Agricultural Economics, 19, 33-44.

FRENCH, B. C. & BRESSLER, R. G. (1962) The Lemon Cycle. Journal of Farm

Economics, 44, 1021-1036.

FRENCH, B. C. & MATTHEWS, J. L. (1971) A Supply Response Model for Perennial

Crops. American Journal of Agricultural Economics, 53, 478-490.

GARCIA, A. F. & SHIVELY, G. E. (2001) How Might Shadow Price Restrictions

Reduce Technical Efficiency? Evidence from a Restricted DEA Analysis of

Coffee Farms in Vietnam. Journal of Agricultural Economics, 62 (1), 47-58.

GILBERT, E. (2004) Investment Basics XLIX. An introduction to real options.

Investment Analysts Journal, 60, pp 49-52.

GJOLBERG, O. & GUTTORMSEN, A. G. (2002) Real options in the forest: what if

prices are mean-reverting? Forest Policy and Economics, 4, 13-20.

GSO (2001) Statistical Yearbook 2000, Hanoi, Statistical Publishing House.

GSO (2007) Results of Agricultural Census 2006, Hanoi, Statistical Publishing House.

References

193

GSO (2009) Statistical Yearbook 2008, Hanoi, Statistical Publishing House.

GUTIERREZ, L. (2002) Borrowing Constraints and the Agricultural Investment

Decision Process. Agribusiness, 18, 101–114.

HA, D. T. & SHIVELY, G. (2008) Coffee Boom, Coffee Bust and Smallholder

Response in Vietnam's Central Highlands. Review of Development Economics,

12(2), 312-326.

HARTLEY, M. J., NERLOVE, M. & PETERS, R. K., JR. (1987) An Analysis of

Rubber Supply in Sri Lanka. American Journal of Agricultural Economics, 69,

755-761.

HARTMAN, R. (1976) The Harvesting Decision when a Standing Forest has Value.

Economics of Inquiry, 14, 52-58.

HEAPS, T. & NEHER, P. A. (1979) The Economics of Forestry when the Rate of

Harvest is Constrained. Journal of Environmental Economics and Management,

6, 297-319.

HOUCK, J. P. (1977) Approach to Specifying and Estimating Nonreversible Functions.

American Journal of Agricultural Economics, 59, 570-572.

INSLEY, M. (2002) A Real Options Approach to the Valuation of a Forestry

Investment. Journal of Environmental Economics and Management, 44, 471-

492.

INSLEY, M. & ROLLINS, K. (2005) On Solving The Multirotational Timber

Harvesting Problem With Stochastic Prices: A Linear Complementarity

Formulation American Journal of Agricultural Economics, 87, 735 - 755.

INSLEY, M. C. & WIRJANTO, T. S. (2010) Contrasting two approaches in real

options valuation: Contingent claims versus dynamic programming. Journal of

Forest Economics, 16, 157-176.

INTER-MINISTERIAL POVERTY MAPPING TASK FORCE (2003 ) Poverty

mapping and market access in Vietnam.

JALOTA, S. K., SINGH, K. B., CHAHAL, G. B. S., GUPTA, R. K.,

CHAKRABORTY, S., SOOD, A., RAY, S. S. & PANIGRAHY, S. (2009)

Integrated effect of transplanting date, cultivar and irrigation on yield, water

saving and water productivity of rice (Oryza sativa L.) in Indian Punjab: Field

and simulation study. Agricultural Water Management, 96, 1096-1104.

JAYASURIYA, S., BARLOW, C. & SHAND, R. T. (1981) Farmers' long-term

investment decisions: A study of Sri Lankan rubber smallholders Journal of

Development Studies, 18, 47-67.

JIA, H. (2006) The optimal single forest rotation under climatic fluctuation effect.

Selected Paper prepared for presentation at the Southern Agricultural

Economics Association Annual Meetings Orlando, Florida, February 5-8, 2006.

KAAKYIRE, A. D. & NANANG, D. M. (2004) Application of real options theory to

forestry investment analysis. Forest Policy and Economics, 6, 539-552.

KARAM, F., KABALAN, R., BREIDI, J., ROUPHAEL, Y. & OWEIS, T. (2009) Yield

and water-production functions of two durum wheat cultivars grown under

different irrigation and nitrogen regimes. Agricultural Water Management, 96,

603-615.

References

194

KARAMA, F., BREIDY, J., STEPHAN, C. & ROUPHAEL, J. (2003)

Evapotranspiration, yield and water use efficiency of drip irrigated corn in the

Bekaa Valley of Lebanon. Agricultural Water Management, 63.

KEARNEV, M. (1994) An intertemporal linear programming model for pip fruit

orchard replacement decisions. Technical paper No 94/6, Ministry of

Agriculture and Fishers, Wellington, New Zealand.

KENNEDY, J. O. S. (1981) Applications of Dynamic Programming to Agriculture,

Forestry and Fishery: Review and Prognosis. Review of Marketing and

Agricultural Economics, 49, 141-173.

KENNEDY, J. O. S. (1986) Dynamic programming: applications to agriculture and

natural resources, London, Elsevier Applied Science.

KLEMPERER, W. D. (1976) Impact of Tax Alternatives on Forest Values and

Investment. Land Economics, 52, 135-157.

KNAPP, K. C. (1987) Dynamic Equilibrium in Markets for Perennial Crops. American

Journal of Agricultural Economics, 69, 97-105.

KONDA, V. R. & TSITSIKLIS, J. N. (2003) Linear stochastic approximation driven by

slowly varying Markov chains. Systems & Control Letters, 50, 95-102.

KUNZOVÁ, E. & HEJCMAN, M. (2009) Yield development of winter wheat over 50

years of FYM, N, P and K fertilizer application on black earth soil in the Czech

Republic. Field Crops Research, 111, 226-234.

L'ECUYER, P., GIROUX, N. & GLYNN, P. W. (1994) Stochastic Optimization by

Simulation: Numerical Experiments with the M/M/1 Queue in Steady-State.

Management Science, 40, 1245--1261.

LI, T., FENG, Y. & LI, X. (2009) Predicting crop growth under different cropping and

fertilizing management practices. Agricultural and Forest Meteorology, 149,

985-998.

LJUNG, G. & BOX, G. (1978) On a measure of lack of fit in time series models.

Biometrika, 65, 297-303.

LOVELLI, S., PERNIOLA, M., FERRARA, A. & DI TOMMASO, T. (2007) Yield

response factor to water (Ky) and water use efficiency of Carthamus tinctorius

L. and Solanum melongena L. Agricultural Water Management, 92, 73-80.

LUONG, Q. & LOREN, T. (2006) A Real Options Analysis of Coffee Planting in

Vietnam. Agricultural Economics, International Association of Agricultural

Economists, 35, 49-57.

MARBACH, P. & TSITSIKLIS, J. N. (2001) Simulation-Based Optimization of

Markov Reward Processes. IEEE Transactions on Automatic Control, 46, 191-

209.

MARD (2008) Agricultural Sector Report. Hanoi, Vietnam.

MATHESON, V. A. (2007) Alternative Methods of Calculating Optimal Timber

Rotations: A Critique of the Stokey/Lucas/Prescott Tree-Cutting Problem. IN

COLLEGE OF THE HOLY CROSS, D. O. E., FACULTY RESEARCH

SERIES, PAPER NO. 07-01 (Ed.).

MAUBOUSSIN, M. J. (1999) Get real :Using Real Options in Security Analysis.

References

195

MELGAR, J. C., MOHAMED, Y., NAVARRO, C., PARRA, M. A., BENLLOCH, M.

& FERNÁNDEZ-ESCOBAR, R. (2008) Long-term growth and yield responses

of olive trees to different irrigation regimes. Agricultural Water Management,

95, 968-972.

MENDELSSOHN, I. A. (1979) The Influence of Nitrogen Level, Form, and

Application Method on the Growth Response of Spartina alterniflora in North

Carolina. Estuaries, 2, 106-112.

MENDOLA, M. (2007) Farm Household Production Theories: A Review of

“Institutional” and “Behavioral” Responses. Asian Development Review, 24, 49-

68.

MOLISA (2003) Study on corporate social responsibility: synthetic report of segment II

study on footwear and textile enterprises.

NERLOVE, M. (1956) Estimates of the Elasticities of Supply of Selected Agricultural

Commodities. Journal of Farm Economics, 38, 496-509.

NERLOVE, M. (1979) The dynamics of supply: Retrospect and prospect. American

Journal of Ag- ricultural Economics 61, 874-888.

NERLOVE, M. & BACHMAN, K. L. (1960) The Analysis of Changes in Agricultural

Supply: Problems and Approaches. Journal of Farm Economics, 42, 531-554.

OKTEM, A. (2008) Effects of deficit irrigation on some yield characteristics of sweet

corn Bangladesh J. Bot., 37, 127-131.

ÓLAFSSON, S. & KIM, J. (2002) Simulation Optimization. IN YÜCESAN, E., CHEN,

C.-H., SNOWDON, J. L. & CHARNES, J. M. (Eds.) Proceeding of the 2002

Winter Simulation Conference, 8-11 Dec, 2002. California, United States.

OLSEN, T. S. (2005) Asymmetric supply response in coffee production. PhD

Conference in Economics and Business, 2-4 November 2005. The University of

Western Australia,Perth

OXFAM (2002) The impacts of trade liberalisation on coffee farmers in Dak Lak

province. Hanoi, Vietnam.

PANDA, R. K., BEHERA, S. K. & KASHYAP, P. S. (2004) Effective management of

irrigation water for maize under stressed conditions. Agricultural Water

Management, 66, 181-203.

PANDEY, R. K., MARANVILLE, J. W. & ADMOU, A. (2000) Deficit irrigation and

nitrogen effects on maize in a Sahelian environment. Agricultural Water


PANNELL, D. J. (1996) Lessons from a decade of whole-farm modelling in Western

Australia. Review of Agricultural Economics 18, 373-383.

PENA, D. & RODRIGUEZ, J. (2002) A Powerful Portmanteau Test of Lack of Fit for

Time Series. Journal of the American Statistical Association, 97, 601-610.

PENTTINEN, M. J. (2006) Impact of stochastic price and growth processes on optimal

rotation age European Journal of Forest Research, 125, 335-343.

PETERMAN, R. M. (1977) Graphical Evaluation Of Environmental Management

Options: Examples From A Forest--Insect Pest System. Ecological Modelling,,

3, 133--148.

PI-IPSARD (2007) Evaluation of strategy options for the Vietnamese coffee sector

References

196

POWELL, W. P. (2007) Approximate Dynamic Programming: Solving the curses of

dimensionality, Canada, John Wiley & Sons, Inc, New Jersey.

RAJSIC, P., WEERSINK, A. & GANDORFER, M. (2009) Risk and Nitrogen

Application Levels. Canadian Journal of Agricultural Economics, 57.

REED, W. J. (1984) The Effects of the Risk of Fire on Optimal Rotation of a Forest.

Journal of Environmental Economics and Management, 11, 180-190.

REEDER, J. D. & MCGINNIES, W. J. (1989) Response of Established Forages on

Reclaimed Mined Land to Fertilizer N and P. Journal of Range Management,

42, 327-332.

REID, J. B., STONE, P. J., PEARSON, A. J. & WILSON, D. R. (2002) Yield response

to nutrient supply across a wide range of conditions: 2. Analysis of maize yields.

Field Crops Research, 77, 173-189.

ROSENZWEIG, M. R. & WOLPIN, K. I. (1993) Credit Market Constraints,

Consumption Smoothing and the Accumulation of Durable Production Assets in

Low-Income Countries: Investments in Bullocks in India. Journal of Political

Economy, 101.

RÖTTER, R. & VAN KEULEN, H. (1997) Variations in yield response to fertilizer

application in the tropics: II. Risks and opportunities for smallholders cultivating

maize on Kenya's arable land. Agricultural Systems, 53, 69-95.

ROY, B. V., BERTSEKAS, D. P., LEE, Y. & J.N.TSITSIKLIS (1997) A neuro-

dynamic programming approach to retailer inventory management, in

‘Proceedings of the IEEE Conference on Decision and Control’, Vol. 4, pp.

4052-4057.

RUF, F. & BURGER, K. (2001) Planting and Replanting Tree Crops – Smallholders’

Investment Decision. paper presented at the international conference on The

Future of Perennial Crops. Yamoussoukro.

SADOULET, E. & DEJANVRY, A. (1995) Quantitative Development Policy Analysis,

London, The Johns Hopkins University Press.

SALARDINI, A. A. (1978) Yield response of tea (mamelia sinensis) to fertilizer and

irrigation in two locations of Iran. Plant and Soil, 50, 113-123.

SAMUELSON, P. A. (1976) Economics of Forestry in an Evolving Society. Economics

of Inquiry, 4, 466-492.

SAYLOR, R. G. (1974) Alternative Measures of Supply Elasticities: The Case of Sao

Paulo Coffee. American Journal of Agricultural Economics, 56, 98-106.

SCHÜTZE, N. & SCHMITZ, G. H. (2007) Neuro-dynamic programming as a new

framework for decision support for deficit irrigation sytems. In International

Congress on Modelling and Simulation (ModSim 2007). Christchurch, New

Zealand.

SEGARRA, E., ETHRIDGE, D. E., DEUSSEN, C. R. & ONKEN, A. B. (1989)

Nitrogen Carry-over Impacts in Irrigated Cotton Production, Southern High

Plains of Texas. Western Journal of Agricultural Economics,, 14, 300-309.

SHENGGAN, F., HUONG, P. L. & LONG, T. Q. (2003) Government Spending and

Poverty Reduction in Vietnam. A report prepared for the World Bank funded

project “Pro-poor spending in Vietnam”, by the International Food Policy

Research Institute and the Central Institute for Economic Management.

References

197

SIMTOWE, F., PHIRI, A., THOMAS, A. & ZELLER, M. (2006) Can Risk-aversion

towards fertilizer explain part of the non-adoption puzzle for hybrid maize?

Empirical evidence from Malawi. Journal of Applied Sciences 6, 1490-1498. .

SMITH, R. (2004) Applying Options Theories to Technology Management Decisions

[online] available at http://www.ctonet.org/ [Accessed on 7 July 2008]

SON, D. K. (2008) Vietnam's Agriculture: 20 years after "Doi Moi". Hanoi, Vietnam.

SON, D. K., THANG, T. C. & GIANG, P. H. (2005) Export Potential Assessment of

main agro-commodity in Vietnam. Hanoi.

SONNTAG, N. & HILBORN, R. (1978) Beyond dynamic programming: determining

improved management policies for budworm-forest interaction. Uni.British

Columbia, In press.

SRV (2002) The comprehensive poverty reduction and growth strategy (CPRGS)

Hanoi, Vietnam.

SUTTON, R. S. & BARTO, A. G. (1998) Reinforcement Learning London, England,

MIT Press, Cambridge, MA.

SVENDSEN, M. (2003) Operational governance and financing in Vietnamese

irrigation. Paper prepared for the World Bank, Hanoi.

SWISHER, J. R. & HYDEN, P. D. (1998) A Survey of Simulation Optimization

techniques and Procedures. IN JOINES, J. A., BARTON, R. R., KANG, K. &

FISHWICK, P. A. (Eds.) Proceeding of the 2000 Winter Simulation Conference,

10-13 Dec, 2000. Florida, United States.

TAYLOR, C. R. & SWANSON, E. R. (1973) Experimental Nitrogen Response

Functions, Actual Farm Experience and Policy Analysis. Illinois Agricultural

Economics, 13, 26-32.

TAYLOR, R. E., SCHMIDT, J. W. & CHACHRA, V. (1973) Optimization of

simulation experiments. The 6th conference on Winter simulation San Francisco,

CA ACM New York, NY, USA

TEMBO, G., BRORSEN, B. W., EPPLIN, F. M. & TOSTAO, E. (2008) Crop input

response functions with stochastic plateaus. Amer. J. Agr. Econ. , 90, 424-434.

THANG, T. C. (2008) Daklak coffee farm survey 2007. Hanoi, Vietnam.

THE WORLD BANK (1999) Vietnam development report 2000: Attacking poverty.

Hanoi: World Bank.

THE WORLD BANK (2002) Vietnam: Agricultural Price Risk Management.

THE WORLD BANK (2003) World development report 2004: Making services work

for poor people. . Washington, DC: World Bank and Oxford University Press.

THE WORLD BANK (2004) Coffee Sector in Vietnam.

THE WORLD BANK (2005) Viet Nam: Poverty assessment and strategy. Washington,

DC: Worldbank.

THE WORLD BANK (2007) Vietnam Development report 2008: Poverty. Hanoi:

World Bank.

THOMSON, T. A. (1992) Optimal Forest Rotation When Stumpage Prices Follow a

Diffusion Process. Land Economics, 68, 329-342.

http://www.ctonet.org/

References

198

TIEN, T. V. (2006) An analysis of the supply response of robusta coffee in Vietnam.

Agricultural and Resource Economics. Kasetsart University.

TOPCU, S., KIRDA, C., DASGAN, Y., KAMAN, H., CETIN, M., YAZICI, A. &

BACON, M. A. (2007) Yield response and N-fertiliser recovery of tomato

grown under deficit irrigation. European Journal of Agronomy, 26, 64-70.

TRAILL, B., COLMAN, D. & YOUNG, T. (1978) Estimating Irreversible Supply

Functions. American Journal of Agricultural Economics, 60, 528-531.

TWEETEN, L. G. & QUANCE, C. L. (1969) Positivistic Measures of Aggregate

Supply Elasticities: Some New Approaches. American Journal of Agricultural

Economics, 51, 342-352.

WALTERS, C. J. (1975) Optimal Harvest Strategies For Salmon In Relation To

Environmental Variability And Uncertainty About Production Parameters.

WALTERS, C. J. & HILBORN, R. (1978) Ecological Optimization And Adaptive

Management. Ann. Rev. Ecol.Syst., 9, 157-188.

WICKENS, M. R. & GREENFIELD, J. N. (1973) The Econometrics of Agricultural

Supply: An Application to the World Coffee Market. The Review of Economics

and Statistics, 55, 433-440.

WIGHT, J. R. & GODFREY, E. B. (1985) Predicting Yield Response to Nitrogen

Fertilization on Northern Great Plains Rangelands. Journal of Range


WINTER-NELSON, A. & TEMU, A. A. (2005) Liquidity constraints, access to credit

and pro-poor growth in rural Tanzania. Journal of International Development,

17, 867–882.

WOLFFRAM, R. (1971) Positivistic Measures of Aggregated Supply Elasticities -

Some New Approaches - Some Critical Notes. American Journal of Agricultural

Economics, 53, 356-359.

WOLFFRAM, R. E. (2005a) Estimation of short- and long-term irreversible Relations

[online] available at http://www.ilr1.uni-

bonn.de/mafo/publ/estimation%20of%20short-%20and%20long-

term%20irreversible%20Relations.pdf [Accessed on 8 July 2008].

WOLFFRAM, R. E. (2005b) Asymmetric Error Correction Models – some Comments

and Improvements [online] available at http://www.agp.uni-

bonn.de/mafo/publ/Asymmetric_Error_Correction_Models_5.pdf [Accessed on

8 July 2008].

ZWART, S. J. & BASTIAANSSEN, W. G. M. (2004) Review of measured crop water

productivity values for irrigated wheat, rice, cotton and maize. Agricultural

Water Management, 69, 115–133.

http://www.ilr1.uni-bonn.de/mafo/publ/estimation%20of%20short-%20and%20long-term%20irreversible%20Relations.pdf



http://www.agp.uni-bonn.de/mafo/publ/Asymmetric_Error_Correction_Models_5.pdf

http://www.agp.uni-bonn.de/mafo/publ/Asymmetric_Error_Correction_Models_5.pdf

Appendix A

199

Appendix A

Figure A1. Regions in Vietnam

Name of country Vietnam

Surface area 329,241 km²

Population 85.2 million(2007)

Population density 247/km²(2004)

Percentage of urban population 26.4%(2005)

GDP (nominal) $ 71.6 billion(2007)

GNI per capita $ 656(2006)

Appendix A

200

Figure A2. Vietnam and Dak Lak province

Appendix A

201

Figure A3. Coffee Area Map in Vietnam

Appendix A

202

Figure A 4. Coffee Output Map in Vietnam

Appendix A

203

Figure A 5. Poverty map of Vietnam

Appendix A

204

Figure A 6. Depth of Poverty Map in Vietnam

Appendix B

205

Appendix B

206

Appendix B

Table B1. Vietnam GDP at current price in 2008 by sector

GDP at current price (bill. VND) (%)

2007 2008 2007 2008

Vietnam 1144015 1478695 100 100

Agriculture, forestry and aquaculture 232188 325166 20.3 22.0

Construction and Industry 475681 590075 41.6 39.9

Services 436146 563454 38.1 38.1

Source: GSO per com

Table B2. Area and output of selected perennial crops in Vietnam, 2007-2008

Change 2008/2007

2007 2008 Level %

Fresh tea

Cultivated area (1000 ha) 126.6 129.6 3.0 102.4

Harvested area (1000 ha) 107.4 110.7 3.3 103.1

Yield (quita/ha) 65.8 68.6 2.8 104.3

Output (000 tonnes) 706.8 759.8 53.0 107.5

Coffee



Yield (quita/ha) 19.7 19.9 0.2 101.0

Output (000 tonnes) 961.7 996.3 34.6 103.6

Rubber


Harvested area (1000 ha) 377.8 399 21.2 105.6

Yield (quita/ha) 16.1 16.6 0.5 103.1

Output (000 tonnes) 609.8 662.9 53.1 108.7

Pepper

Cultivated area (1000 ha) 48.4 50 1.6 103.3

Harvested area (1000 ha) 41.1 43 1.9 104.6

Yield (quita/ha) 21.7 24.3 2.6 112.0

Output (000 tonnes) 89.3 104.5 15.2 117.0

Cashew

Cultivated area (1000 ha) 440.1 404.9 -35.2 92.0


Yield (quita/ha) 10.3 10 -0.3 97.1

Output (000 tonnes) 312.5 313.4 0.9 100.3

Source: GSO per com

Appendix B

207

Table B 3. Vietnam coffee export, 1991-2008

Year Quantity (000 tonnes) Value (Mil.$)

1991 93.50 76.30

1992 116.20 91.50

1993 122.60 110.80

1994 176.40 330.30

1995 248.10 598.10

1996 283.70 400.26

1997 391.60 493.71

1998 381.80 593.80

1999 482.46 585.30

2000 733.94 501.45

2001 910.00 385.00

2002 719.00 317.00

2003 749.24 504.81

2004 974.80 641.02

2005 892.37 735.48

2006 775.46 826.99

2007 1200.00 1800

2008 1000.00 2115.00

Source: MARD per com

Table B 4. A comparision of coffee export cost in some countries (cent/lb)

Countries Production cost extra cost

Vietnam 20.5 4.5

India 30 4

Indonesia 29 8

Brazil 55 0

Source: PI-IPSARD per com

Table B 5. DRC of main commodities of Vietnam

Commodity DRC

Rice 0.59

Coffee 0.37

Rubber 0.7

Tea 0.79

Source: CAP (2006)

Appendix B

208

Table B 6. Number of coffee households by region and age structure in Vietnam

2006

Number of coffee

households

By farm-size

Less than 0,5 ha 0.5-1 ha 1-2 ha 2-3 ha 3-5 ha 5-10 ha

Over 10 ha

Vietnam 477,235 156,611 146,097 128,491 32,093 11,638 2,113 192

+ Red River Delta 7 5 2

Ha Tay 7 5 2

+ North East 20 15 5

Ha Giang 8 6 2

Lao cai 12 9 3

+North West 11 9 2

Lai Chau 11 9 2

North CenTral Coast 10,463 5,617 2,366 2,050 332 79 18 1

Thanh hoa 405 217 91 79 14 4

Nghe An 2,189 808 757 574 40 9 1

Ha Tinh 5 4 1

Quang Binh 245 184 59 2

Qung Tri 6,404 3,525 1,247 1,284 268 64 15 1

Thua Thien – Hue 1,215 879 212 110 10 2 2

South Central Coast 3,364 657 966 1,307 309 101 19 5

Binh Dinh 1,798 211 551 824 166 35 9 2

Phu Yen 1,035 168 280 400 119 55 10 3

Khánh Hoa 531 278 135 83 24 11

Central Highlands 427,316 139,393 131,797 114,927 28,696 10,450 1,888 165

Kon Tum 9,877 4,327 2,138 2,385 689 276 49 13

Gia Lai 69,370 28,147 19,629 16,600 3,596 1,165 205 28

Dak Lak 180,434 62,224 60,770 44,195 9,650 3,054 507 34

Dak Nong 53,534 9,447 14,059 19,763 6,894 2,758 580 33

Lam Đong 114,101 35,248 35,201 31,984 7,867 3,197 547 57

North East South 36,054 10,915 10,959 10,207 2,756 1,008 188 21

Ninh Thuan 7 3 2 2

Binh Thuan 1,144 186 328 418 140 63 9

Binh Phuoc 8,000 1,226 1,910 3,156 1,081 493 118 16

Binh Duong 221 84 39 47 21 17 11 2

Dang Nai 18,352 5,638 6,074 5,068 1,197 336 37 2

Ba Ria - Vung Tau 8,330 3,778 2,608 1,516 315 99 13 1

Source: GS0, 2007

Appendix B

209

Table B 7. Area of coffee in main provinces in Vietnam, 1986-2006 (ha)

Year Kontum Gia Lai Dak Lak Lam Dong Dong Nai Vietnam

1986 2536 5438 29949 10159 10890 65630

1987 3695 7926 39014 15854 17588 92300

1988 5152 11051 63265 23009 22270 135940

1989 5088 10911 61233 23909 26306 139568

1990 4660 9996 70242 24883 28326 153627

1991 4782 10381 69320 23120 29686 153060

1992 3006 8857 73546 23391 26553 153060

1993 3195 10548 82147 22031 21984 158729

1994 3046 11387 96768 25508 21358 176302

1995 4219 23999 112477 49561 23049 232424

1996 6479 29796 165694 67777 28823 323694

1997 7177 40274 216964 98435 31576 436504

1998 10667 58506 257504 119492 47089 549882

1999 14112 66005 267994 128734 49364 599768

2000 15492 87156 278596 122805 38516 604304

2001 15234 86898 277323 132922 35618 606571

2002 14110 85961 257558 129159 29305 566889

2003 12368 77571 232416 118229 25072 510200

2004 11513 76063 232961 116739 22471 496800

2005 10594 75854 241800 117428 20288 491400

2006 9844 75910 242250 118788 16857 497000

Note: the data for Dak Lak province includes Dak Nong

Appendix C

210

Appendix C

Table C 1. Estimated results from Modified Wolffram model with different

window length

Dependent variable: lnarea (logarithm of coffee area)

Independent

variables MWM (window =3 years)

MWM

(n =4 years)

MWM

(n =5 years)

Coef. t ratio Coef. t ratio Coef. t ratio

Lnareat-1 0.86 18.31 0.85 18.91 0.85 18.95

lnYMA 0.14 4.24 0.14 4.26 0.14 4.27

MWR (n=3) 0.35 6.82

MWF (n=3) 0.22 5.47

MWR (n=4) 0.42 6.31

MWF (n=4) 0.22 5.62

MWR (n=5) 0.45 6.19

MWF (n=5) 0.22 5.62

MWR (n=6)

MWF (n=6)

Provincial dummy

variables

Gia Lai 0.23 3.02 0.23 3.22 0.24 3.34

Dak Lak 0.37 2.73 0.39 2.97 0.40 3.11

Lam Dong 0.29 2.93 0.30 3.16 0.31 3.29

Dong Nai 0.08 1.12 0.09 1.28 0.10 1.39

Other 0.22 2.78 0.23 2.97 0.24 3.07

constant 1.44 3.95 1.49 4.22 1.53 4.36

R2 0.98 0.98 0.98

F-value 790 804 912

Appendix D

211

Appendix D

COFFEE FARM SURVEY QUESTIONNAIRE

Name Code

Province

District

Commune/town

Village

Interviewer

Respondent

Address of respondent

Phone

Date of interview

Appendix D

212

A. HOUSEHOLD CHARACTERISTICS

1. Name of household head ? ……………………………………………………………………………

2. Gender of household head ? ___________ 1. Male 2. Female

3. Year of household head’s birth ?__________ ……………………………………………………………………………

4. Ethnic origin ?____________ 1. Kinh 2. Other (specify) ………………………….

5. Geographical origin ?_____________ 1. Local resident 2. Migrant

6. [If migrant] When did you come here ? ………...(year)

7. [If migrant] Where do you come from ? province name__________ provinve code________

8. Total No. of HH members ? ……………………………………………………………………………

9. No of adults (>=15 year old and <= 65) ? ……………………………………………………………………………

No of children (< 15 year old) ? …………………

10. How many people are usually working in agriculture? ____________

Appendix D

213

B. LAND AND LANDUSE

rownum 1 Crop

name

2. Crop

code

3.Plant

area (m2)

4. Harvested

area (m2)

5. Harvested

output (kg)

6. sale

output (kg)

7. Sale price

(d/kg)

8. Sale value

(000 D)

Note: if forestry trees, “Harvested output” can be blank

Appendix D

214

C. OTHER SOURCES OF INCOME

Source of revenue Revenue (000 D)

1.Livestock

from pig

from chicken

from cattle/buffalo

from other animal

2. Aquaculture

3. Wage/salary

4. Pension

5.Other income (specify____________________)

Appendix D

215

D.COFFEE PRODUCTION

1. When did you start growing coffee?____________(years)

2. Total area for coffee cultivation last years? ……………………… (m2). # of plots for coffee cultivation ?........

3. Coffee distribution per plot and land quality ?

# Plot Planting

year

Area

(m2)

% rented Registered with red

certificate

1. Yes 2. No

% irrigated Soil type (see

code)

Harveted

ouput

(kg)

Yield

(kg/ha)

1 Plot 1

2 Plot 2

3 Plot 3

4 Plot 4

Soil type code: 1. Ferralsol (đất đỏ) 2. Arcrisol (đất xám) 3. Luvisol (đất đen) 4. Other (specify)

5. % removal and replantings in each year

Age of tree (years) % removed % replantings

0-3

3-8

8-15

15-20

20-25

>25

6. Yield of coffee by age (kg bean/ha) ?

Age 3 4 5 6 7 8-15 15-20 20-25 >25

Yield

Appendix D

216

E. THE LARGEST PLOT

How large the biggest plot is ________________(m2)….. Out put………..(kg) and how old is it?________________(years)

Could you please tell us the cost in last year for the biggest plot

# Cost items Quantity Price(d/unit)

value (000VND)

A Preparation cost

1 seedlings (trees)

2 land rent

3 well

4 water system (pumb, tube…)

B Fertlizer/pesticide

5 Ure (kg)

6 KCL(kg)

7 Nitro(kg)

8 DAP(kg)

9 SA(kg)

10 NPK

11 Manure

13 Pesticide

C Labour

14 Hole, design

15 Growing

16 Weeding

17 Water

18 Prunning

19 Fertilize, pesticide

20 Harvest

D Energy

21 Electricity, Petrol

22 irrigation

Appendix D

217

F. GROW AND CUT DECISION

1. Have you ever reduced coffee area? __________1.yes 2.no

When was the most recent reduction?_______(year)

Why did you redude?_________1. fall in coffee price 2.Other ___________

If (1), At what price did you cut?_________(d/kg)

Maximum area did you cut? __________(m2)

3. When you decide grow coffee, what do you base on most?_________

1. Price of last year

2. Price of last several years

3. Advice of local authority

4. Price prediction

4. What was the received price in last year?___________(d/kg coffee bean)

5. At what minimum price/and how long it last do you intend to cut ?

Price level

(VND/kg)

If price in one year

% reduction

If price in 3 years

% reduction

If price in 5 years

% reduction

If price in 10 years

% reduction

8000

7000

6000

5000

4000

<4000

6. After cut, do you intend to grow coffee again when price up?___________ 1. Yes 2.No

7. At what minimum price do you intend to grow coffee again ?_________(d/kg)

Appendix D

218

G. FARMER’S RESPONSE TO COFFEE PRICE REDUCTION

When price reduced, did you reduce input application?

1. Have you ever switched to other crops due to price fall? __________1.YES 2.NO

If yes, please list two main crops which were replaced for coffee _________ __________

If not, why did not you change to other crops

# Reasons 1. True 2. False

1 Keep coffee and expect higher price

2 Lack of capital to grow other crop

3 Do not know which crop should be

replace

4 Risk afraid

5 Other (specify)________________

2. In bad price years, could household income cover the household expenditure and annual cost? _______ 1.YES 2. NO

If not, what did you have to do for overcoming the problem?

# Solutions 1. True 2. False

1 Using saving stock

2 Borrowing money

3 Reducing expense of household

4 Selling asset, animals

5 Finding other job

6 Selling coffee garden

Appendix D

219

H. CREDIT

1. When did you borrow last loans (year)?________

Amount of loan?___________(000 VND) interest rate ?_____(%/year) loan duration_____(months)

4. Main purpose of loans?________

1. Buying input for coffee production 2. Paying for labour cost for coffee production 3. For other activities

4. Buy food 5. Other

5. If you want to get loan for coffee production, do you get sufficient loan ? _________1.YES 2.NO

If not, How many percent did loan account for total requirement ?_______(%)

6. Main source of loan?___________

1.Bank 2.Private lenders 3.Relatives 4.Women’s association 5.Commune Committee 6.Other credit

programs 7.Other

I. HIRING LABOUR

1. In last season (2005/2006), did you hire labour for coffee production? _____1.YES 2.NO

2. If yes, percentage of hired labour in total?________ (%)

3. Did you get any problems to hire labours?________1.YES 2.NO

If yes, which problems?_____________

K.WATER

1. What is main source of water for coffee plantation?__________1. Wells 2. Lake, reservoir 3. Streams 4.Other

2. Are your coffee plantation provided enough water?_____1.YES 2.NO

3. According to your estimation, Is yield of your coffee plantation limited by water scarcity ?________1.YES 2.NO

4. If yes, how much can coffee yield increase with enough water ?_________(%)

OPTIMAL INVESTMENT DECISIONS OF COFFEE … · COFFEE FARMERS IN VIETNAM ... Nguyen Le Hoa, Truong...

Documents

Transcript of OPTIMAL INVESTMENT DECISIONS OF COFFEE … · COFFEE FARMERS IN VIETNAM ... Nguyen Le Hoa, Truong...