Coffee Boom, Coffee Bust and Smallholder Response in Vietnam's Central Highlands

Data from smallholder coffee farms in the Central Highlands are used to examine responses to a drop in producer coffee prices.. Vietnam’s entry into the world coffee market, combined wit

Coffee Boom, Coffee Bust and Smallholder

Response in Vietnam’s Central Highlands

Dang Thanh Ha and Gerald Shively*

Abstract

This paper studies the recent boom and bust in Vietnam’s coffee economy Data from smallholder coffee farms in the Central Highlands are used to examine responses to a drop in producer coffee prices A multi-nomial logistic regression model is used to identify several factors associated with four specific patterns observed among coffee farmers: no response to price change, reductions in use of purchased inputs, changes

in crop mix, and responses aimed at enhancing liquidity through off-farm work or borrowing Patterns of response are shown to have differed systematically across sub-groups of smallholders Policy implications raised by the findings are discussed.

1 Introduction

Introduced by the French in the 1850s, coffee remained relatively unimportant in Vietnam until the 1980s (de Fontenay and Leung, 2001) In the decade between 1990 and 2000 Vietnamese farmers, approximately 80% of them smallholders (Greenfield, 2002), planted more than a million hectares of Robusta coffee, enabling Vietnam to surpass Colombia as the world’s second-largest coffee producer (ICO, 2002) In the 1990s coffee production increased at an annual average rate of 30% and by the end

of the decade coffee was providing as much as 10% of Vietnam’s annual export earn-ings A number of factors help to explain the rapid growth of the coffee sector in

Vietnam, among them a policy of privatization and economic liberalization (Doi Moi),

state-sponsored migration, and price spikes generated by frosts in Brazil During the mid-1990s the economic benefits of Vietnam’s coffee boom spread rapidly and were far reaching By mid-decade more than one million Vietnamese were participating directly or indirectly in the country’s coffee economy (Nhan, 2002)

Vietnam’s entry into the world coffee market, combined with rising stocks, inelastic demand and a shift toward low-cost Robusta for processing (Ponte, 2002) helped con-tribute to a precipitous decline in international coffee prices, to below 40 cents a pound

in late 2001—a three-decade low in constant dollar terms (Brown et al., 2001).1Figure

1 presents producer coffee prices and harvested coffee area in Vietnam over the period 1990–2002 The figure clearly illustrates Vietnam’s basic coffee story: a steady expan-sion of coffee area in the latter half of the 1990s and simultaneous eroexpan-sion in producer prices

Of course, fluctuating fortunes for smallholders is a phenomenon unique to neither coffee nor Vietnam For example, Figure 2 displays annual international price indices DOI:10.1111/j.1467-9361.2007.00391.x

* Ha: Nong Lam University, Thu Doc, Ho Chi Minh City, Vietnam Tel: 84-8-896-1708; Fax: 84-8-896-0713; E-mail: d.thanh.ha@hcm.vnn.vn Shively: 403 West State Street, West Lafayette, IN 47907, USA Tel: 765-494-4218; Fax: 765-494-9176; E-mail: shivelyg@purdue.edu For helpful research assistance we thank Jessica Perdew and Nam Anh Tran This research was made possible through support provided by the Office

of Agriculture and Food Security, Bureau for Global Programs, US Agency for International Development, under the terms of Award No PCE-A-00-98-00019-00 The opinions expressed herein are those of the authors and do not necessarily reflect the views of the US Agency for International Development.

for coffee, cocoa, tea, palm oil and rubber over the period 1960–2005 and illustrates the volatile nature of prices for these crops Coefficients of variation for these com-modities typically range between 30 and 70% when calculated using annual average prices and more than 100% when based on monthly data

Price fluctuations such as those in Figure 2 are important for several reasons On the supply side, many countries rely heavily on just a few primary commodities, including agro-industrial crops for export earnings, so that fluctuations in prices lead directly to fluctuations in incomes and account balances On the demand side, price cycles have been shown to affect trade flows as well as inflation rates in importing and exporting countries (Cashin et al., 2000; Barichello, 2000) Recent research documenting the

0

20

40

60

80

100

price

0 100 200 300 400

500 area

Price Area

Source: Prices for green Robusta beans (in US cents per kilogram) from International

Coffee Organization; harvested area (in 1000 hectares) from FAO

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Figure 1 Farm Gate Price and Harvested Coffee Area, Vietnam, 1990–2002

Source: United Nations Conference on Trade and Development (UNCTAD).

0

100

200

300

400

500

600

700

800

900

1960 1962 1964 1966 1968 1970 1972 1974 197

6

1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 199

8

2000 200 2 2004 Year

Cocoa Tea Palm oil Rubber

Figure 2 Historical Price Indices for Major Agricultural Export Commodities, 1960–2005

extent of agricultural price fluctuations and the detrimental effects thereof include Li (2002, for cocoa in Indonesia); Potter (2001, for oil palm in Indonesia); Gotsch and Burger (2001, for cocoa in Malaysia); McMillan (2003, for cashew in Mozambique); Burger and Smit (1997, for rubber worldwide); and Huff (2002, for rubber in pre-World War II Malaya)

The focus here is the collapse in coffee prices, which has attracted much attention Pérez-Grovas et al (2001), Varangis et al (2003), and Maluccio (2005), provide analyses of the “coffee crisis” from Latin American perspectives Charveriat (2001) and Mehta and Chavas (2004) offer global perspectives In the case of Vietnam, de Fontenay and Leung (2001), the World Bank (2004), and Lindskog et al (2005) con-clude that the collapse in coffee prices had important effects throughout the country, and reversed the fortunes of many smallholders For example, for three consecutive years beginning in late 1999, producer prices in Vietnam were insufficient for most farmers to cover the variable costs of coffee production (USDA FAS, 2002, 2003) Not surprisingly, one of the most acutely affected areas has been the largest coffee-producing area in Vietnam, Dak Lak province (Table 1) In Dak Lak, overproduction and subsequent price dampening had “a serious negative impact on the livelihoods of Vietnamese coffee producers” (ICARD and Oxfam, 2002, p 1) Although coffee prices recovered modestly in the 2002–03 production year, many smallholders in Dak Lak and elsewhere continue to struggle with low producer prices.Vietnam’s on-going coffee crisis is widely acknowledged within and outside the country, and finding ways to improve outcomes for coffee farmers remains a key component of Vietnam’s agricul-tural strategy.2

In this paper data from 1999 and 2003 surveys of smallholder coffee farmers in Vietnam’s Central Highlands are used to study how farmers coped with declining coffee prices Data from the 2003 survey are employed to estimate a multinomial logis-tic model to measure the factors associated with four specific patterns observed among smallholders: no response, changes in input use, changes in land use, and cash enhance-ment The role of ethnicity in conditioning these responses is studied, as well as the importance of irrigation access and farmers’ subjective outlooks regarding coffee prices We find farmers’ responses were fairly mixed, and that lowland Kinh migrants were more likely to change crops, borrow, or engage in off-farm employment than

Table 1 Coffee Production by Region in Vietnam, 1996

Source: Minot (1998).

members of ethnic minority communities The patterns suggest both farm- and farmer-specific constraints on smallholders’ ability to respond to declining agricultural prices

2 Study Site and Data

Coffee is grown in a variety of locations and by a variety of farmers in Vietnam Given this diversity generating a survey sample that broadly represents coffee growers is difficult Here, we focus on a sample drawn from one of the most important coffee districts of Vietnam, in the Central Highlands province of Dak Lak In 2001 the total coffee area of Vietnam was 566,800 hectares, of which the Central Highlands accounted for 476,800 hectares Within the Central Highlands, Dak Lak province accounted for 257,100 hectares and roughly 50% of total national coffee output Of Dak Lak’s 16 districts, Cu M’gar has the largest coffee area, representing roughly 14% of the total coffee area of Dak Lak Our data come from the Ea Tul catchment, in Dak Lak’s Cu M’gar district.3

The study area has undergone tremendous economic and social change in recent decades Following reunification in 1975, the government of Vietnam designated Dak Lak as a New Economic Zone (NEZ) and established hundreds of state farms and cooperatives in the area, most of them emphasizing industrial crops As a result of the NEZ designation, population density increased more than five-fold in just four decades (from 17 people per square kilometer in 1975 to more than 90 people per square kilo-meter in 2002) Most of this migration consisted of an influx of ethnic Kinh farmers (Vietnam’s majority) from densely populated lowland areas For example, although the Kinh originally represented a very small proportion of the population of Dak Lak, by

2000 their share in the provincial population had risen above 70% Migration was ini-tially induced by government programs, but beginning in the late 1980s and early 1990s substantial migration into Dak Lak occurred spontaneously By 2000, the province’s population stood at 1.8 million people, of which approximately 400,000 were members

of ethnic minority groups Smallholders control approximately 80% of the coffee area The remainder is controlled by the Vietnam National Coffee Coporation (VINACAFE)

The rise of Vietnam’s coffee economy nearly doubled GDP in Dak Lak within

agricultural growth has raised both environmental and social concerns On the envi-ronmental front, many observers note that coffee production in Dak Lak has tended

to follow an extensive, rather than intensive, path For example, in the decade between

1990 and 2000 an estimated two-thirds of coffee output growth was due to area expan-sion (ICARD and Oxfam, 2002) This rapid area expanexpan-sion led to substantial forest destruction Although Dak Lak still contains some of the most extensive forests in Vietnam, forest clearing has been severe (De Koninck, 1999) leading to habitat and biodiversity loss (UNEP-WCMC, 2000) Total land for agricultural production in Dak Lak doubled in the 1990s, with an average increase of 46,000 hectares per year A majority of newly claimed land has been used for cash crop production, notably coffee Due to the rapid expansion of land for agriculture, the forest-covered area decreased from 90% in 1960 to 57% in 1995, and then to less than 50% in the late 1990s Between

1980 and 2000 Dak Lak lost forest at an average rate of 20,000 hectares per year (ADB, 2003) In the Ea Tul catchment, forest area was reduced from 58% in 1980 to 20% in

2000 (Ha et al., 2001)

From a social perspective, a massive influx of Kinh farmers into areas of ancestral importance to ethnic minorities led to a series of land conflicts between migrants and

local ethnic minority people, and between various migrant populations (ADB, 2003) Although many minority farmers have benefited from Vietnam’s coffee expansion,

minority groups are widely perceived to be economically disadvantaged vis-à-vis their

Kinh counterparts This conjecture is partly supported by data from our sample indi-cating significantly higher per capita incomes among Kinh farmers (7.6 million VND

on average in 2003, compared with 4.4 million VND for non-Kinh respondents) The inability of ethnic minorities to benefit from and respond to general economic incen-tives is an issue that remains high on the stated agenda of policymakers in Vietnam, and significant challenges to reducing poverty among these groups remain (ADB, 2000; Doutriaux et al., 2006) Below we address the extent to which Kinh farmers were able

to respond in a different way from their non-Kinh counterparts to the collapse in coffee prices

random surveys were conducted by a team from Nong Lam University as part of a larger study of rural livelihood in coffee-growing areas of the Central Highlands The surveys cover agricultural production and household activities in 1999 and 2003 The observations do not represent a panel, but the sample households were drawn from the same area, namely the four communes of Ea Tul, Ea Kpal, Ea Pok and Quang Phu The two samples are similar in most respects The main exception is that rice

Table 2 Characteristics of Sample Farms, Dak Lak Province

Income a (1000 Dong) 25,273 29,097 35,501 31,673 Income per capita a (1000 Dong) 4,191 4,696 6,597 6,741

(# > 65 + # < 15)/# workers

Age of household head (years) 52.9 12.2 45.2 11.5 Education of household head (years) 7.7 3.5 7.0 3.7 Female headed household (%) 16.0 37.0 17.7 38.3 Year first planted coffee (year) 1990 4.3 1991 4.9

(% of farm area with title) b

Irrigation (% w/irrigation) 72.3 45.2 84.2 36.6

(% w/off-farm income > 0)

Rice cultivation (% w/rice area) 66.0 47.5 25.4 43.6

(% of farm area w/rice)

Source: survey data.

Notes: a nominal income in 1000 Dong; in 1999 1US$
14,000D; in 2003 1US$ = 15,553D.

b For definition of land ownership, see text.

production was far less prominent in the 1999 survey (66% of households vs 25% in 2003) and off-farm employment was far more widespread in the 2003 survey (43% vs 13% in 1999) All farmers in both rounds of the survey planted and sold coffee The economic rewards of producing coffee in Dak Lak have been very substantial for coffee-growing smallholders, on average For example, a disaggregated poverty map for Vietnam indicates that poverty rates in southern, coffee-growing districts of the Central Highlands are markedly lower than those in northern districts, where less coffee development has occurred (Minot, 2000) Our survey data indicate average annual household nominal incomes in the study areas of approximately 25 million

capita terms, these incomes are roughly equivalent to GDP per capita in Vietnam (US$430 in 2002), a remarkable pattern insofar as poverty in Vietnam tends to be highest in rural areas While this underscores the importance of coffee as an income-generating mechanism in the Central Highlands, our data also suggest that the crop has nevertheless diminished in importance in recent years Households derived 87%

of total income from coffee in 1999 and 77% in 2003 Thus while total household income rose slightly, on average, between the 1999 and 2003 surveys, the total share of income derived from coffee production fell This reflects lower prices and an ongoing process of crop and income diversification that began following the coffee price col-lapse in the mid-1990s

Table 3 contains data illustrating the broad patterns of activity diversification in the samples These data suggest widespread diversification of activities among the sample farms and a rising proportion over time of households reporting either planting multiple crops or engaging in off-farm employment For example, the proportion of households that reported planting two or more crops increased between 1999 and 2003,

Simi-larly, the proportion of households reporting off-farm employment increased from 13% in 1999 to 43% in 2003 Only the proportion of sample households that reported engaging in livestock sales was lower in the later round of the survey (53% compared with 39% in 1999) The remainder of the paper focuses on the extent to which these patterns and more specific reactions to the collapse in coffee prices can be explained

by features of the sample respondents

Table 3 Diversification of Activities by Sample Farms, Dak Lak Province

(% of households)

Livestock sales (cattle and pigs) 53.1 38.8

(% of households)

(% of households)

Planted three or more crops 7.9 14.4

(% of households)

Source: survey data.

3 Household Responses to the Price Collapse

We begin the analysis of household response to the collapse in coffee prices by exam-ining data on broad indicators of behavioral response, as reported retrospectively by households when interviewed in 2003 Table 4 contains data for the entire sample of farms, and separately for those below 1.5 hectares (small farms) and above 1.5 hectares (large farms) Note that 1.5 hectares corresponds to the average farm size in the sample This split places 66% of the sample in the small farm category The median farm size in the 2003 sample was 1.2 hectares

During the 2003 survey respondents were asked to report how they responded to the collapse in coffee prices We should note that we made no attempt to tie the drop in the producer price of coffee to a specific reference year because not all farmers perceived the downward trend in price the same way (in part due to some modest but short-lived price recovery in 1998) Instead, the survey simply aimed to record self-reported responses to falling prices As a result, some farmers in the sample may have reported how they responded to the price drop following the 1995 growing season, whilst others may have reported how they responded in the 1998–2002 period In all instances, we allowed farmers to report multiple responses We included specific categories of response and also allowed open-ended descriptions of changes

in household activities and practices As a result, column sums in Table 4 exceed 100% Most households, however, reported only a single major category of response to the collapse in price Nearly a third of the sample reported no change in strategy or activity

In the case of an annual crop, the theory of a profit-maximizing farmer would suggest that the primary response to a price reduction should be a reduction in planted area (Askari and Cummings, 1976) In the case of a perennial crop, however, optimal short-run responses to a price decline need not follow this pattern, since long-short-run

indicated in Table 4, in our sample the most common reaction to the drop in price,

Table 4 Smallholder Response to Collapse in Coffee Prices, Dak Lak Province, 2003 (%)

Note: multiple responses possible An asterisk indicates the difference between means is statistically

signif-icant at a 90% confidence level.

Source: survey data.

regardless of farm size, was a reduction in the amount of fertilizer applied to coffee.

A slightly larger proportion of small farms than large farms reported a reduction in fertilizer use (45% vs 35%), although this difference is not statistically significant While this reaction clearly reflects a desire on the part of households to reduce cash expenditures, it also reflects the tendency of households to shift resources out of coffee and into other crops, without abandoning coffee completely In fact, a number of house-holds reported that while they reduced fertilizer use in coffee, they simultaneously increased fertilizer use on other crops, especially corn and black pepper This basic pattern is further underscored by the complete elimination of fertilizer by 17% of households in the sample, and the reduction or elimination of irrigation (typically requiring the use of electric or gasoline pumps) in 19% of households.9Somewhat sur-prisingly, a statistically higher proportion of large farms reported a complete cessation

of irrigation or fertilization This may reflect a greater capacity to withstand a loss in income on larger farms and a tendency to maintain production on small farms, even

in the face of falling prices Nevertheless, a significantly larger proportion of small farms reported removing coffee trees (8.0% vs 2.8%) and a larger proportion of small farms reported planting new crops in response to falling coffee prices Roughly similar (and small) proportions of households reported borrowing as a direct result of falling prices Small farms were slightly more likely to seek off-farm employment or to sell livestock No household reported selling land Taken together, the data in Table 4 suggest that small farms were somewhat less responsive to falling coffee prices than large farms, and that a disproportionate share of large farms were sufficiently buffered from their reliance on coffee that they could treat their investment in trees with a certain degree of benign neglect (foregoing both irrigation and fertilizer) while waiting for coffee prices to recover

Questions that naturally emerge from Table 4 are whether patterns of response sys-tematically differed, and whether differences can be explained by household features

To answer these questions a more detailed analysis is undertaken using a multinomial logistic regression This approach requires collapsing the categories appearing in Table

4, so as to maintain the opportunity for econometric estimation of coefficients of interest Accordingly, we create four broad categories of response,q= {1, 2, 3, 4} Into

category 1 we place those households that reported no response to the price collapse Into category 2 we place households that reported input reductions, i.e reductions in

fertilizer or irrigation rates This includes households that eliminated fertilizer or

irrigation entirely Category 3 contains households that changed crops, either by

removing coffee trees or by planting new crops in combination with existing coffee

The final category includes households that directly enhanced liquidity, either by

borrowing money (formally or informally), seeking off-farm employment, or selling livestock When households reported multiple responses, a unique category was assigned, with changes in crops at the top of the hierarchy, followed by liquidity enhancements and reductions in inputs

Table 5 contains results for our four equation multinomial logistic regression where

no response is the omitted response category Normalizing the regression coefficients

on this category, the estimating equations take the form (Greene, 2001):

Pr q= 2

b

( )=

e

X

2

1

Pr(q= 1)= b b b

1

1 e X2 e X3 e X4

where X represents a vector of independent variables and bi represents a vector of

coefficients associated with response category i All coefficients are reported as

rela-tive risk ratios, that is, probabilities of response relarela-tive to the omitted category such

that:

As a result of this transformation, all estimated coefficients are positive by

construc-tion An estimated coefficient greater than 1 indicates that the independent variable is

associated with a higher probability of response under that category, compared with

the omitted response; an estimated coefficient less than 1 indicates that the

indepen-dent variable is associated with a lower probability of response under that category,

compared with the omitted response Absolute values of z statistics (reported in

paren-theses in Table 5) have been adjusted accordingly

Overall, several statistically meaningful patterns emerge from the data, with nine

statistically significant coefficient estimates Controlling for other features of the

house-holds, Kinh ethnicity is associated with a significantly lower relative probability of

reductions in inputs, and a significantly higher relative probability of seeking enhanced

Pr = 2

Pr = 1

q

q

b

( ) = e X2

Pr q= 4

b

( )=

e

X

4

1

Prq= 3

b

( )=

e

X

3

1

Table 5 Multinominal Logistic Regression of Smallholder Response, Dak Lak Province, 2003

Liquidity

Notes: Total sample size N = 209 Omitted response category (N = 64) is “no response.” See text for

defini-tion of response categories All coefficients reported as relative risk ratios Absolute values of z statistic

(reported in parentheses) are similarly transformed An asterisk indicates the estimated coefficient is

sig-nificantly different from zero at a 90% confidence level.

liquidity.The latter pattern likely indicates preferential access to borrowing and greater off-farm employment opportunities among Kinh respondents Some evidence points

to village differentials, especially in the case of reduced input levels, but no evidence

Several patterns emerge with respect to four fixed factors: irrigation source, farm size, number of workers, and security of tenure In each case, one would expect these factors to influence the range of possible response, and in each case statistical evidence supports this conjecture In the case of the irrigation source, which here is coded as a binary indicator of whether the household used a well as a source of irrigation water, households with an established well (a relatively reliable source of irrigation water compared with the alternatives, such as a stream or pond) were more likely to seek enhanced liquidity through non-agricultural means Within the study site, the water demands of coffee are quite high, and competition for water is frequently reported by farmers The regression results suggest farmers without a secure source of irrigation water were more likely to shift land or inputs away from coffee Results also suggest large farms were more likely than small farms to reduce or eliminate inputs (i.e fer-tilizer and irrigation) That the correlation between farm size and borrowing is weak

in this sample is somewhat surprising given Duong and Izumida’s (2002) evidence of

a link between land ownership and access to credit in Vietnam The pattern found here likely reflects the uniformly low rates of borrowing as a response to price reductions

in the sample

Household labor capacity emerges as an important explanatory factor in the regres-sions Not surprisingly, households with a larger number of workers per hectare were more likely to seek off-farm employment They were also more likely to reduce input levels, probably as a reflection of higher immediate requirements for cash and food in larger households.To gauge the importance of tenure security in conditioning response,

we include in the regression a measure of farm ownership During the survey, respon-dents were asked to identify tenure status for all of their plots based on three main categories: (1) area owned and documented by a government-issued land use

government-issued land use certificate (n= 14), and (3) area recently cleared and not documented

categories, several households held land under more than one category A variable

“percentage of land owned” was constructed using total farm size, net of area rented

in or out, as denominator and total area documented by government-issued land use certificates as numerator Land ownership was positively correlated with the likelihood

of reducing input levels The weak correlation of land ownership with other categories

of response indicates in part the small degree of variation in the land ownership vari-able, and in part the fact that households with high rates of ownership were the least likely to change their behavior in response to the price fall

The final variable included in the regressions is the respondent’s expectation regarding coffee prices During the survey each respondent was asked to forecast the minimum, expected, and maximum coffee price for the next growing season The expected price, expressed as a ratio of the respondent’s price to the sample mean response, is included as a conditioning variable in the regressions Other things equal, households that had above average expectations regarding the future coffee price,

vis-à-vis their sample cohorts, were less likely to change crops compared with the

reference group Within the context of this sample, crop changes seem to be largely driven by price expectations Changes in input use, in contrast, are conditioned by external factors and household characteristics