Abstract
Extensive land and market reform in Vietnam has resulted in dramatic increases in rice output and incomes. This is illustrated with measures of total factor productivity, net incomes, and net returns in rice production from 1985 to 2006. Results show considerable gains in major rice growing areas, but recent evidence of a productivity slowdown. The differences over time and region speak to existing land use practice, calling for further reform. Estimations detail the effects of remaining institutional and policy constraints, including existing restrictions on land use, ambiguous property rights, and inadequate markets for land and access to extension services and credit. (JEL O12, O13)
I. Introduction
Vietnam has achieved remarkable progress in reducing rural poverty in the last 30 years, due largely to a series of extensive market and land reforms in agriculture, along with impressive increases in economic growth at the national level. The land and market reforms in agriculture were pervasive, moving the system of rice production from commune-based public ownership and control to one with effective private property rights over land and farm assets, competitive domestic markets, and individual decision making over a wide range of agricultural activities. The substantial incentive effects created by these policy measures, inducing farmers to work harder and use land more efficiently, have been estimated to be as much as 50% of the increase in total factor productivity (TFP) during the peak of the reform period (Che, Kompas, and Vousden 2006). Overall, given these reforms, Vietnam has gone from being a large importer of rice from 1976 to 1980, to now the second- largest exporter of rice in the world, with considerable increases in farm profitability and rural incomes, and decreases in rural poverty rates by over 40% from 1993 to 2004 alone (Hansen and Nguyen 2007).
However, much still remains to be done to increase living standards in rural areas and enhance general rural development. Like many reform processes, the early rapid gains in economic activity have dissipated over time, with the suggestion now of a TFP slowdown in rice production in many areas in Vietnam. In addition, many of the poor still farm small areas of land, constrained in use, often with fragmented or noncontiguous plots, and with little or no human and physical capital accumulation or access to agricultural extension services and farm credit. Much of this is due to remnants from past institutional arrangements, but also to continued constraints in land use, credit availability, and the provision of rural services, all calling for further or renewed land policy and market reform.
This paper has two basic tasks. First, it assembles a data set from 1985 to 2006 to measure the changes in TFP, terms of trade (TOT), net incomes, and net returns in Vietnamese rice production, both in the principal rice growing areas and throughout the country. The results track the effects of the major land and market reform process and determine key differences in TFP and net returns over time and by region. All of this speaks directly to existing land use practices and is suggestive of needed policy response. The second task is to isolate the remaining institutional constraints and policy challenges that may be limiting increases in productivity and efficiency. For this purpose, two stochastic production frontier and inefficiency models, drawn from two different farm survey data sets, are estimated to determine the potential effects of ongoing issues over land use and sale, the provision of credit, land fragmentation, less than secure property rights, and the lack of rural education and support services.
II. Context
Rice continues to dominate agricultural production in Vietnam, with rice production accounting for nearly 90% of the output of food grains and almost two-thirds of rural farm income (SDAFF 2006). Although rice is produced in every one of the 60 (recently redefined as 64) provinces in Vietnam, the Red River Delta (RRD) and the Mekong River Delta (MRD) are the main rice growing regions. The smallest producers of rice (less than 100,000 tons per year) are in Binh Phuoc province, which is relatively small in area and contains the principal coffee growing (Gialai Kontum) and mining (Cao Bang, Bac Kan) areas. Provinces with the largest rice output (more than a million tons per year) are located in the MRD (Tien Giang, Soc Trang, Long An, Kien Giang and An Giang), which as a whole accounts for roughly half of Vietnam’s output of rice and most of its international exports. In terms of natural conditions, the MRD and the RRD are the most favorable for growing rice, with many areas naturally irrigated, producing up to three rice crops per year. Based on farm survey data for 2004 (as used below), the average farm size in the RRD (0.4 ha per farm) is much smaller than in the MRD (1.4 ha per farm). However, the number of threshing machines in the RRD is almost double that in the MRD. In the MRD, with a large volume of high-quality rice exports, rice processing takes place in mills rather than on the farm to maintain international standards.
Rice output has increased dramatically during the major land and market reform periods, from 10 million to nearly 34 million tons nationwide from 1980 to 2004 (Kompas 2004). After a period of “output share contracts” (from 1981 to 1987), where up to 80% of rice output had to be delivered to the state, with the remainder sold in competitive markets, a period of “trade liberalization” (1988–1994) brought major institutional and market change, allowing for effective private property rights over both land (initially 10- to 15year and later 20-year leases) and capital equipment. With reform, most production decisions were decentralized, all farm income (after tax) was retained by the farmer, and rice could be sold freely in competitive domestic markets. The result was an increase in rice prices (from state-controlled low prices prior to reform), added profitability, considerable increases in TFP, and dynamic gains from trade reform due to induced capital accumulation out of retained farm earnings (Che, Kompas, and Vousden 2001, 2006).
Since 1994, these dramatic market and land reforms have been solidified and in many cases extended in the postreform period.1 Nevertheless, a number of concerns remain and have been raised again recently in a Participatory Poverty Assessment (PPA), with over 240 focus groups and 1,450 survey participants, undertaken by the Vietnam Academy of Social Sciences (VASS 2009). Chief among these are issues surrounding land titles and their use, land fragmentation, and the lack of rural credit availability and supporting rural services.
Land title and use requirements are a good example of the challenges that remain. Although the Land Law of 1993 (amended and clarified in 1998, 2001, and 2003) allows “land use rights to be transferred, exchanged, leased, mortgaged and inherited” (Congress of Vietnam 1993), in practice, considerable constraints remain in place regarding both land conversion (i.e., land transferred or converted to other uses) and land accumulation (i.e., trades and accumulation of land plots). For any land conversion, for example, the commune authorities have to first develop a plan, often based on or as minor amendments to past historical blueprints for land use in that area, to submit to various levels of government for approval. The PPA reports that this process is often long and transactions costs are high, making it difficult for poor farmers in particular to participate (VASS 2009). In addition, although land consolidation in Vietnam is occurring, with a number of important benefits, there are still restrictions on overall land size (Ravallion and van de Walle 2008). In 2007, the Vietnamese government increased limits on the transfer of land use rights for annual agricultural land from 3 to 6 ha for the South East, the MRD, and Ho Chi Minh City, and from 2 to 4 ha for other cities and provinces. This is a welcome albeit modest change for many farmers, but in most cases rice farming outside the MRD still takes place on very small farms, at subsistence levels (GSO 2004; VASS 2009).
Part of the obstacle to land consolidation is the lack of fully secure property rights. The terms of land use titles were extended for agricultural land from 15 to 20 years with the Land Law of 1993, but in many cases even 20 years is too short to provide secure rights in the shift to larger farms, or allow farm land to be turned into use for small manufacturing or industry. Overall the process of land certification or entitlement itself has also been below expectations. Household survey data from 2006 (GSO 2006) suggests that only 76% of agricultural land parcels, 68% of urban land parcels, and 34% of forest land parcels have been granted land use right certificates (World Bank 2009). Even where land certificates do exist, there is a shortage of basic infrastructure for an effective operation of land administration, including cadastral mapping, transaction registrations, and record management in the provision of land administration services. Problems are compounded by lack of information or public awareness and the apparent limited capacity of land administration staff, especially at the commune and district levels (World Bank 2009). It is perhaps for this reason that real estate markets have been slow to develop, with recent data indicating that the role of land rental markets in agriculture in rural areas remains thin (GSO 2004), and that continued government restrictions often prevent lowcost and efficiency-enhancing transfers (Dein- inger and Jin 2008). Land rights that are not secure also have an impact on credit availability and capital accumulation. The PPA reports that farms without land use certificates, which would normally be used as collateral, are not able to obtain even short-term loans, much less transfer land use entitlements. This is especially true for many farms with land tenures that are due to expire soon, given the 20-year leases initiated in 1993 (VASS 2009).
Land fragmentation occurs when households have land use rights to a number of often small, noncontiguous plots. With reform and the resulting dissolution of the commune system in Vietnam, land was allocated to prior commune members in a roughly egalitarian manner: a more or less equal distribution of plots to each household, throughout the commune, so that no one household would have a concentration of plots in the most fertile parts of the commune, or near roads, water sources, or other essential services. Immediately after the major reform process (1988– 1994), there were as many as 75 to 100 million parcels or plots of land in Vietnam and on average about seven to eight plots per household (World Bank 2003; Hung, MacAulay, and Marsh 2007), of which about 10% had an area of only 100 m2 or less (Phien 2001). Evidence suggests that total plot numbers have been falling recently with land consolidation (nationwide, falling to 4.3 plots per household (GSO 2004)), but the problem is still common. Fragmentation levels, for example, continue to remain high in the RRD, the most populated region, with 90% of the households having rice farm sizes of roughly 0.2 to 0.5 ha (Chu 2008) and an average of 4.6 plots per farm (GSO 2004). The number of plots per farm in the MRD, by contrast, is only 1.6. In some cases (e.g., risk spreading), fragmentation may be an advantage (see Marsh et al. 2006). However, for the most part, small and scattered land holdings hamper mechanization, the use of buffaloes and tractors, and technological adoption and require additional time and labor for farming activities that must be carried out in geographically distant plots. The embankments that separate plots alone have been estimated to reduce total agricultural land for cultivation in Vietnam by 2.4% to 4% (Thanh 2008).
III. Data Sources, Variables, and Specification
Three different data sets are used in this paper. The first is a provincial-level data set on rice production in Vietnam, from 1985 to 2006, for 60 provinces, used to construct TFP, price, and quantity indices and net income and return measures, greatly improving on the basic TFP estimates (to 1994) provided by Che, Kompas, and Vousden (2006). This is an extensive data set on prices and quantities for all rice outputs and inputs, directly obtained as (or aggregated to) provincial averages. The key variables include paddy rice output, labor, land, material inputs (especially fertilizer, but also seeds and pesticide), and capital (a measure of tractors and buffaloes), as well as input prices for labor, land, fertilizer, pesticides, and capital used in rice production. (See the Appendix for detailed data sources, constructions, and adjustments.)
The second data set, used to construct a stochastic production frontier and inefficiency model, is a farm survey data set for 388 rice farms, complied in 2004 (by the authors) in the major rice growing regions (the MRD and RRD), designed specifically to isolate the potential effects on inefficiency from differing farm characteristics and land fragmentation. Key additional variables include measures of soil quality and irrigation, average plot size, as a proxy for land fragmentation, and the level of education of the household head of the farm.
The third data set is the 2004 Vietnam Household Living Standards Survey (VHLSS), which is used to confirm and extend the results of the smaller farm survey data set, providing added estimates of the effects of secure property rights, more precise measures of the effects of land fragmentation (using a Simpson index), and access to agricultural extension services and credit. The VHLSS is a household survey data set of roughly 9,000 households in 2,216 communes, with cluster-sampling techniques to cover the entire country, conducted by the General Statistical Office (GSO) in Vietnam in selected years (e.g., 2002, 2004, and 2006). The 2004 VHLSS data set, in particular, compared to other VHLSS data sets in Vietnam, has an extended module for land, with separate components for land use and agricultural production. Sample size is reduced to 3,671 households to isolate farms that are primarily rice producers. It is important to note that a Simpson index is not applicable to the 2004 farm survey data, since the exact size of each plot for the 388 farm households was not recorded. The measure of fragmentation is thus simply total rice land divided by the number of plots.
For all stochastic production frontiers log-likelihood specification tests were used to determine functional form and the presence of inefficiency effects. In all cases, standard ordinary least squares (OLS) estimates are seen to be inappropriate, and functional form tests reject a translog specification in favor of a more standard Cobb-Douglas production function. Tests also indicate that estimates of the stochastic frontiers using a random coefficients approach, allowing for nonneutral shifts in the production frontier, following Kalirajan and Obwana (1994), resulted in little difference in estimated coefficients, with the inefficiency term adequately represented by a truncated half-normal distribution. Frontier and inefficiency estimates are obtained using Frontier 4.1 (Coelli, Rao, and Battese 1998).
IV. TFP, TOT, and NET Returns
The change in TFP is a measure of outputs to inputs over any two time periods. Results for Vietnamese rice production are generated using Tornqvist quantity (and price) indices given by
[1]or
[2]for N quantities, q inputs or outputs (depending on context), periods s and t, and weights,
[3]for time period s, for example. TFP, for outputs y and inputs x is thus given by
[4]for input weighted shares υ i for periods s and t. For convenience, results are summarized across eight regions, as officially defined in Vietnam: RRD (1), the Northeast (2), the Northwest (3); the North Central Coast (4), the South Central Coast (5), the Central Highlands (6), the Southeast (7), and the MRD (8). As mentioned, the RRD and the MRD are the major rice growing regions in the country. Region 7 is largely industrial, and Regions 2, 3, and 6 are the poorest by conventional measures.
There is little doubt that the increase in rice production in Vietnam has been substantial, especially after the output share contracts period (1981–1987), or under the major land and market reforms (1988–1994 and forward) (Che, Kompas, and Vousden 2001). For the country as a whole, the indexed value of paddy rice output shows an average annual increase of 3.5%, as a fitted linear trend (Figure 1). The largest increases in rice output occur during the period of trade liberalization (1988–1994) and continue (virtually unabated) in the postreform period to 2006.
Paddy Rice Output (Indexed) in Vietnam (1985–2006), Average Annual Growth Rate by Fitted Linear Trend = 3.5%
However, trends in TFP vary markedly between regions in the country. Figure 2 shows that not only is TFP higher in the MRD, but also that the growth in TFP is substantially larger in the MRD compared to the RRD and all other regions. As a fitted annual trend, the growth in TFP in the MRD is 4.42%, while in the RRD it is 2.25%, and in all other regions 1.36% per year. This poor TFP performance is of special concern in the poorest regions of the country (regions 2, 3, and 6), where the average annual increase in TFP is less than 1.3%. In all cases, except for the MRD, there is also evidence of a slowdown in productivity after the year 2000. This is an added problem, again, in poor regions (which generally do not have a natural advantage in rice production, or sufficient water resources for wetland rice), but it is also a concern in the RRD, a major rice growing area, where farms remain relatively small and fragmented.
Total Factor Productivity (TFP) Indexes for the Mekong River Delta (MRD), the Red River Delta (RRD), and All Other Regions (Other) for Paddy Rice Production in Vietnam, 1985–2006, Average Annual Growth Rate in TFP by Fitted Linear Trend for the MRD = 4.42%, for the RRD = 2.15%, and for All Other Regions = 1.36%
With the reform process, the indexed price of rice increases throughout Vietnam, from the state-controlled low price in the communal period to a partially controlled price during the output share contracts period, and beyond. During the trade liberalization period (1988–1994), in particular, all controls over domestic prices were removed, and prices rose rapidly to world values, especially from 1993 to 1996. After 1996 the domestic price of rice generally tracks world prices for rice on international markets and fluctuates accordingly.
Input prices, on the other hand, increase more slowly (and uniformly) throughout the trade liberalization period, as the likely product of both increases in the output of rice and resulting patterns of economic development. Much of this increase is dominated by increases in the price of fertilizer (with some volatility), but farm wage rates also increase at an average annual rate of 1.44%.
The combination of output and input price changes, as the TOT, is summarized in Figure 3, showing a relatively stable trend until 1993, with improvement from 1993 to 1996, due mostly to increases in domestic rice prices, after which it declines sharply (save 1998) until 2001. Although the TOT improves after 2001, it still remains below its starting point throughout the remainder of the series, to 2006. This highlights the importance of TFP increases to partly offset this trend, since increases in productivity will generate proportionally more revenues for given input use.
Terms of Trade Indexes for Rice Production in Vietnam, as the Ratio of Indexed Paddy Prices to the Indexed Value of All Input Prices, 1985–2006 (Base Year 1985)
Figure 4 is the key graph, in effect combining all price, quantity, and productivity indices together. It shows both the indexed value of paddy rice output (i.e., the indexed price multiplied by the indexed quantities of rice) and the indexed value of input expenditures (i.e., the indexed input prices multiplied by the indexed quantities of inputs). The wedge between the two lines provides a measure of net income in rice production over time. For Vietnam as a whole, land and market reforms indicate substantial increases in net income from 1988 to 1999, and especially so in the years 1993–1999. During 2000–2001, both the domestic and international price for rice fall dramatically, and the wedge closes.
Net Income Measure or the Indexed Value of Paddy Rice Output Values (Indexed Output Prices Multiplied by Indexed Output Quantities) and the Indexed Value of all Input Values (Indexed Input Prices Multiplied by Indexed Input Quantities) in Rice Production in Vietnam, 1985–2006
The wedge for the years 1988–1999 provides an essential story of economic development and also coincides with a well-documented decrease in the rural poverty rate in Vietnam. Increases in the price of rice, output, and productivity, on the one hand, and increases in the farm wage rate on the other, result in substantial increases in farm and rural income. It is roughly during this period, specifically, as commonly measured, from 1993 to 2004, that the defined share of poor people in Vietnam “dropped by two thirds and approximately 24 million people were lifted out of Poverty” (Hansen and Nguyen 2007). Not all of this poverty reduction was due to rice production, of course, but given the large share of the population in rural areas and the predominance of rice production in rural agriculture, there is little doubt that the reform and postreform periods had a major impact on overall living standards.2
Nevertheless, it is also clear that these gains are not shared equally throughout the country. Figure 5 shows the indexed ratio of the value of revenues to the value of input costs for paddy rice production, for selected regions, as a measure of net returns. Relative to the starting point, all regions do reasonably well from 1992 to 1998, but after 1998 both the RRD and all other regions fall (in some cases) far below the starting point. For the years 1999–2004, net returns are even less than one for areas outside of the MRD and RRD. Overall, the MRD does consistently better than all other regions with relatively large net returns from 1989 forward.
Net Returns in Vietnam, as the Indexed Ratio of Revenues to Input Costs in Paddy Rice Production for Mekong River Delta (MRD), Red River Delta (RRD), and All Other Regions (Other), 1985–2006
V. Frontiers and Efficiency
The importance of the level of and changes in TFP and resulting changes in net income highlight the importance of potential efficiency gains that accompany further land and market reform. The following sections use stochastic production frontiers and inefficiency models to isolate the key constraints on efficiency gains (as a component of TFP), and what policy measures might be most suitable to increase efficiency.
Stochastic Frontiers and Inefficiency
Stochastic production frontiers were first developed by Aigner, Lovell, and Schmidt (1977) and Meeusen and van den Broeck (1977). The specification allows for a nonnegative random component in the error term to generate a measure of technical inefficiency, or the ratio of actual to expected maximum output, given inputs and the existing technology. The idea can be readily applied to both cross-section and panel data, following Battese and Coelli (1995). Indexing firms by i = 1,2, . . . , n the stochastic output frontier is given by
[5]for Yi output, Xia 1× k vector of inputs and β a k × 1 vector of parameters to be estimated. As usual, the error term vi is assumed to be independently and identically distributed as 
and captures random variation in output due to factors beyond the control of firms. The error term ui captures firm-specific technical inefficiency in production, specified by
[6]for zi a 1×m vector of explanatory variables, δ a m×1 vector of unknown coefficients, and wia random variable such that uiis obtained by a nonnegative truncation of 
. Input variables may be included in both equations [5] and [6] as long as technical inefficiency effects are stochastic (Battese and Coelli 1995; Forsund, Lovell, and Schmidt 1980; Schmidt and Lovell 1979).
The condition that ui ≥ 0 in equation [5] guarantees that all observations lie on or beneath the stochastic production frontier. A trend can also be included in equations [5] and [6] to capture time-variant effects. Following Battese and Corra (1977) and Battese and Coelli (1993), variance terms are parameterized by replacing 
with 
. The technical efficiency of the ith firm for the basic case can be defined as
[7]and clearly must have a value between zero and one. The measure of technical efficiency is thus based on the conditional expectation given by equation [7], given the values of vi- ui evaluated at the maximum likelihood estimates of the parameters in the model, where the expected maximum value of Yi is conditional on ui =0 (Battese and Coelli 1988). Efficiency can be calculated for each individual firm per year by
[8]for 
and Φ(⋅), the density function of a standard normal random variable (Battese and Coelli 1988). The value of gamma equals zero when there are no deviations in output due to inefficiency, and equals one when there are no deviations in output due to random effects, or variance in v .
Farm Survey Data, 2004
Econometric Specification
The first frontier estimate uses survey data obtained from a selection of 388 farms producing rice from 32 communes across eight provinces in the RRD and MRD, with a roughly equal split of farms and communes in each area. The survey, a targeted sample of farmers primarily producing rice, was carried out by the authors from August to December 2004, with detailed collection of rice output and input data, as well as farm-specific characteristics. The main areas from which farms were selected in the MRD are Soc Trang, Tra Vinh, Vinh Long, and Can Tho; and, in the RRD, from Ha Tay, Nam Dinh, Thai Binh, and Nam Ha. Summary statistics are provided in Table 1. The specification for the stochastic production frontier is given by
Summary Statistics for Paddy Rice Production in the Farm Survey Data Set, 2004
[9]for Y, the output of paddy rice in kilograms; K, capital in machinery hours, as the sum of hours a farm uses tractors in land preparation and transportation, as well as hours employed for pumps and threshing machines; LAB, working days, as the sum of family and hired labor; LAN, total cultivated rice land in hectares times the number of crops over the year; F, kilograms of fertilizer used; P, pesticides in kilograms; and MRD, a binary variable for MRD rice farms. The inefficiency model in this case is
[10]for SIZE, the area of rice land (both leased and directly controlled by the household) in hectares; PLOTSIZE, the average size of plots for rice land on a given farm, or the area of rice land in hectares, divided by the number of plots, as a proxy for land fragmentation; and SOIL, a measure of soil quality, ranked in decreasing order (from 1 to 6), based mainly on the chemical composition of the soil. This is an ordinal ranking, tied to land tax levies by land quality in Vietnam, and does not literally measure relative differences across land-quality types (e.g., type 2 land is not twice as good as type 4 land). IRR is a measure of water availability (natural and irrigated), ranked in decreasing order (from 1 to 4), obtained by asking farmers to rank their irrigation conditions, based on the level and difficulty of supplying water and drainage. The ranking from 1 to 4 is simply given by “very good,” “good,” “fair,” or “poor.” ED is the level of education of the farm decision maker, categorized by levels: primary, secondary, high school, and higher education.
Additional likelihood ratio (LR) tests to confirm specification are summarized in Table 2. Correct critical values from a mixed v 2 distribution (at the 1% level of significance) are drawn from Kodde and Palm (1986). The relevant test statistic is
Generalized Log-Likelihood Ratio Tests and Parameter Restrictions for the Stochastic Production Frontier and Technical Inefficiency Models, Farm Survey Data, 2004
[11]where L(H0) and L(H1) are the values of the likelihood function under the null and alternative hypotheses, respectively. The null hypothesis that technical inefficiency effects are absent (γ = δ0= δ 1= δ2= δ 3= δ4= δ 5=0) and that farm-specific effects do not influence technical inefficiencies (δ1= δ2= δ3= δ4=δ5 =0) in equation [10] are both rejected, as is δ0= δ1= δ2= δ3= δ4= δ5=0. Finally, the null hypothesis that 
, or that inefficiency effects are not stochastic, is rejected. All results indicate that stochastic effects and technical inefficiency matter, and thus that traditional OLS estimates are not appropriate in this study. Specifications with interaction terms and nonlinear effects of farm size on efficiency were also attempted, but these proved to be nonsignificant.
Results for Farm Survey Data
Results for the farm survey data set are reported in Table 3. The coefficients for capital, labor, land, fertilizer, and pesticide in the stochastic production frontier model are 0.15, 0.13, 0.41, 0.21, and 0.06, respectively. The value of the coefficient for the binary variable MRD illustrates the advantages of growing rice in the south, compared to the north. This value is 0.17 and is consistent with the measured difference in TFP between the RRD and the MRD, illustrated in Figure 2.
Parameter Estimates of the Stochastic Production Frontier and Technical Inefficiency Models for the Farm Survey Data, 2004 (Equations [9] and [10])
Of particular interest, however, are the inefficiency results. The coefficients on soil and irrigation are positive, as expected, since the ranking on soil and irrigation is in decreasing order of quality, implying that higher-quality soil and better irrigation increase efficiency. It is also clear that more-educated farmers are also more efficient. The coefficients on SIZE and PLOTSIZE indicate the loss in efficiency from current land use practice. The estimates indicate that larger farms and farms with larger average plot size are more efficient. The latter in particular indicates a potential issue with land fragmentation, and the constraints to consolidation. Admittedly, average plot size on a given farm is a crude indicator of fragmentation, since it lacks a measure of distance between plots or whether plots are contiguous. However, it is also clear from the discussion in Section II that the smaller the average plot size on a farm the more likely it is that these plots are not contiguous. This is especially so in the north, where, as indicated, small and highly fragmented farms predominate. Frontier estimates by Hung, MacAulay, and Marsh (2007), on a smaller survey data set for 188 farms in the north only, near Hanoi in the RRD, in the year 2000, show a comparable negative relationship between the number of plots and farm efficiency.
VHLSS Data, 2004
Econometric Specification
The second frontier estimate uses VHLSS data for 3,671 households largely engaged in rice production (from a total of more than 9,000 households surveyed) in 2004. The cluster survey sample applies to the entire country, not only to the main rice growing areas. Rice output of the households in this subsample of 3,671 farms accounts for more than 75% of total household annual crops in terms of quantity and more than 78% in terms of value. Summary statistics are listed in Table 4. Log-likelihood ratio tests (Table 5) generate a specification for the stochastic production frontier of the form
Summary Statistics for Paddy Rice Production in the VHLSS Data, 2004
Generalized Log-Likelihood Ratio Tests and Parameter Restrictions for the Stochastic Production Frontier and Technical Inefficiency Models, VHLSS Data, 2004
[12]with an inefficiency model given by
[13]for Y, the output of paddy rice in kilograms, produced over the 12 months prior to the survey date; KR, the value of rented capital in rice production (tractors, machines, tools, and implements in thousand Vietnamese Dong [VND]); and LAN, the amount of area in hectares that the household uses for annual crop production, regardless of its ownership. Labor comes from two sources: LAB is household labor (in hours), and HLAB is hired labor in rice production (in thousand VND). Expenditures on fertilizer (F) and pesticide and herbicide (P) in rice production are all measured in thousand VND. MRD is a binary variable for households in the MRD.
In the inefficiency model, ED is years of schooling of the household head, and FRAG is a measure of land fragmentation given by a Simpson index, defined as
[14]where ai is the area of the ith plot and 
is total household farm size. A value of zero for the Simpson index indicates that a household has only one parcel or plot of land. The closer the index is to one, the more fragmented is household farm land, or the more numerous are the number of plots. CERT designates that a farm household holds a land certificate title (measured as a ratio of land under title to total land size), allowing for the sale or lease of all or some plots of land, and SOIL is a measure of soil quality, as in the farm survey data set, ranked in decreasing order (from 1 to 6). Not surprisingly, the measure of soil quality varies more and the mean is higher (meaning less quality) in the VHLSS data, with its broader coverage across all types of land, and throughout the country, compared to the farm survey data, which concentrates only on rice farms in the MRD and RRD (see Tables 1 and 4). EXT is a binary variable indicating whether the farmer visited an extension services office, attended meetings to seek advice or guidance on cultivation practices or raising livestock, or was visited on the farm by an extension staff officer. In the summary data, 48% of those surveyed accessed these services. The value CRE is total farm household credit from banks and other financial institutions for agricultural production during the year in tens of millions of VND.
Results for VHLSS Data
Results for the VHLSS data set are reported in Table 6. Estimated input coefficients are comparable with the results for the farm survey data set for capital, land, fertilizer, and pesticide. The binary variable MRD again indicates the advantages of growing rice in the MRD. Results indicate that better-educated farmers and higher-quality soil increase efficiency across farms. The estimated coefficient on SOIL in the VHLSS estimates (0.48 compared to 0.05 in the farm survey data) reflects the importance of the contrast in soil quality outside of the major rice growing regions, and its effect on inefficiency.
Parameter Estimates of the Stochastic Production Frontier and Technical Inefficiency Models for VHLSS Data, 2004 (Equations [12] and [13])
The VHLSS estimates also show, more pointedly, the effect of land fragmentation on efficiency, using a Simpson index. The more fragmented is a household farm, or the larger the number of plots, the lower is efficiency, with a coefficient value of 0.25.
Of added interest here are coefficient estimates on land use certificate, access to extension services, and credit. As mentioned, a proper land use certificate is essential not only for the ease of acquiring, selling, or leasing land; it also provides the only ready source of collateral for farm loans. The estimates show that farms with a proper certificate are more efficient, as are those that have access to agricultural extension services and credit. The result for CRE, showing that farms receiving more credit are more efficient, along with the fact that only 20% of households have access to funds, is consistent with recent findings suggesting substantial credit constraints for all rural households in Vietnam (Dung and Izumida 2002).
VI. Closing Remarks
Extensive land and market reform in Vietnam has resulted in dramatic increases in rice output over the past 30 years. Results show that TFP increases considerably in the major rice growing areas (the MRD and RRD areas) during the early years of land and market reform, but with evidence of a productivity slowdown since 2000 in all regions except the MRD. TFP in the MRD remains much higher than in the RRD, and TFP in other regions (excluding the RRD and especially in poor areas) remains virtually unchanged throughout the entire period. TOT, net incomes, and returns are also favorable throughout the reform period, providing much of the explanation for increased incomes and poverty reduction during this time, noting that overall performance has worsened considerably since 2000–2001. The differences over time and by region speak directly to existing land use regulations and practices and suggest calls for further land and market reform. In this regard, additional frontier and efficiency model estimates illustrate the remaining institutional and policy constraints, including existing restrictions on land consolidation and conversion and poorly developed markets for land and capital. Estimates show that larger and less-fragmented farms and those that are better irrigated, with higher quality spoil, a clear property right or land use certificate, and access to agricultural extension services and credit are more efficient.
With this in mind, it seems clear that the mandate to grow rice in every province, at least in terms of narrowly defined efficiency criteria, is inappropriate. Productivity and efficiency are both substantially larger in the MRD and RRD areas, where rice production has a clear comparative advantage. This shows up repeatedly in both TFP and related measures, as well as in frontier and inefficiency models in terms of the magnitude of the binary variable for the MRD (and its effect on output). It is also indicated by coefficient estimates that measure the effects of irrigation and soil quality on efficiency. Land in the MRD and RRD is of much better quality in this regard, and naturally suited to rice production. Land policy (formal or in practice) that makes it difficult for land to be converted to other uses is thus difficult to justify.
The same can be said for land consolidation. If farms that are larger and less fragmented are more efficient, practical restrictions on land size need to be relaxed and a more active real estate market for land needs to be provided, encouraging low-cost and efficiency-enhancing land transfers. A necessary and straightforward prerequisite for this is well-defined land use certificates, covering every parcel of land, something that Vietnam has not yet been able to accomplish. This may also partly resolve problems with credit availability, as would a significant extension of the 20-year lease provisions on parcels of agricultural land. Without a land use certificate, or with limited remaining tenure, it is difficult, if not impossible, to secure a loan, much less convert and consolidate land. The original land and market reforms, as dramatic as they were, have not gone far enough to secure property rights or provide sufficient or suitable markets for land and credit.
There are at least three issues that warrant further research. First, it would be useful to have a more refined measure of land fragmentation than either the Simpson index or average land size and number of plots used here, one that includes distance and a spatial representation of noncontiguous plots. There is partial data available for this in Vietnam, but more needs to be collected. Second, the estimates would benefit from additional measures of rural services. The only variable used here, access to agricultural extension services, as a simple binary variable, matters to efficiency, but so too must variables like rural infrastructure (e.g., roads, water rights, and quarantine and surveillance measures) and specific cultivation practices, including the use of rice hybrids. Unfortunately, there is a lack of broad rice-farm survey data to provide such estimates. Finally, and perhaps most importantly, there needs to be a clear investigation into the precise nature and cause of the thin or poorly developed agricultural land and credit markets in Vietnam, and what specific policies might be best to help relax these constraints.
Acknowledgments
Funding for this research from the Ford Foundation, the World Bank, and AusAID is gratefully acknowledged, as is financial support from the Crawford School of Economics and Government in order to obtain farm survey data. Thanks to Thang Nguyen and participants at the “Workshop on Poverty Assessment” at the Centre for Analysis and Forecasting in Hanoi, the Editor, and two anonymous referees for very helpful comments.
Appendix Data Sources and Adjustments
Data for TFP and related measures (1985–2006) are drawn mainly from the Agriculture, Forestry and Fisheries), 1991–2006, data sources obtained from the General Statistics Office of Vietnam (GSO), including VHLSS data, related project investigations, studies and reports by Vietnamese organizations, such as the State Planning Committee (SPC), the Ministry of Agriculture and Food Processing Industry (MAFI), the Ministry of Water Resources (MWR), the Department of Prices and Markets (DPM) (formally known as the State Department of Price ), and international organizations such as the World Bank and the Food and Agriculture Organization (FAO). The details of the structure of rice production (especially for the early data series) are extracted from the Surveys of Rice Production in the RRD and the MRD by Cantho University, funded by the International Rice Research Institute (K. Nguyen 1995; Vo 1995).
It should be noted that from 1985 to 2002 there were 60 defined provinces in Vietnam, based on the GSO statistics and administrative units. However, beginning in 2003, provinces were redefined into 64 provinces, based on the GSO statistics and administrative units. In this study the new provinces are aggregated into the previous provinces in the data set before 2003, for consistency. In particular, Can Tho, Dak Lak, and Lai Chau refer to Can Tho and Hau Giang, Dak Lak and Dak Nong, and Lai Chau and Dien Bien provinces. Regions are as currently defined by the GSO. Primary data for 1985–1999 is obtained from Che, Kompas, and Vousden (2006) and Kompas (2004). The data set for 2000–2002 is from Kompas (2004). In general, prices are measured in constant 2006 U.S. dollars and converted to Dong where appropriate. Data assembly and construction is as follows.
1. Output quantity and prices. Paddy output is drawn from SDAFF (2001, 2006) and the GSO (2008) under the category of “production of paddy by province.” The time series of rice prices by province is computed from a number of sources, with recent data provided by the GSO. For the period 1985–2003, price data is based on that of Kompas (2004) and Che, Kompas, and Vousden (2006), most of which is obtained from the Department of Prices and Markets (DPM). A rice equivalent for output is chosen, rather than rice output alone, since in the same rice fields farmers usually overlap production with other short-term cereal crops, such as sweet potatoes and maize. There are multiple crops per year in many areas. Specific time-series data for rice output is from the SDAFF (1991) and MAFI (1991) for the period 1976–1990, from SDAFF (2001) for 1990–93, GSO (1995) for 1994, and the SDAFF (2001) for 1995– 1999. All measures were verified by alternative data sets contained in the SDAFF (2001) for the years 1975–1999. Updates were obtained from the SDAFF (2006).
2. Land quantity and prices. The time series for “planted area of paddy” is obtained from the SDAFF (1991, 1992, 2001, 2006), SPC (1995), and GSO (2008). The Vietnamese government divides the soil quality of land into seven levels and levies land tax depending on quality. A study by the World Bank (1994) distinguished the quality of soil into five levels in terms of cultivated area. Soil conditions and irrigation are generally much better in the RRD and MRD, compared to other regions (MWR 1994). Land use price variables are defined as the cost of land use, or the average tax levies per one sown hectare in terms of value. The tax levies are required to be paid to the government for the right of using land, and depend on land quality (by rank from type 1 to 5). Land taxes for rice land are based on the gross value of rice production (SDAFF 2006). It is assumed that the land price indices are coincident with the gross value of rice area (as the multiple of rice output overall crops per year and the price of rice).
3. Labor quantity and prices. Data for the quantity of labor is obtained by multiplying average man-days worked per hectare by the number of hectares in a given rice cultivation area. The rice cultivated area is obtained from the SDAFF (1991, 2001, 2006) and GSO (2008). Total labor for rice production is calculated from total rice planted (in area) and average labor used for rice production per hectare. Average man-day working requirements include work for land preparation, transplanting, weeding, and harvesting, originally based on the survey of rice production by Cantho University (1990–1995), as detailed in K. Nguyen (1995) and Vo (1995). The data on the price of labor for paddy production is estimated from average labor costs by the Survey of Rice Producers (SRP) by Cantho University (1990–1995), SDP (1995, 2002) and DPM (2005), and GSO (2006), for the RRD and MRD. For 2003–2006, the labor price variable is estimated using 2002 as a base year and the movement in the wage index for rice production, estimated as the average annual change in labor costs for rice cropping per hectare (SDAFF 2006).
4. Material inputs and prices. Materials for paddy production are largely composed of rice seeds and preparation, fertilizer, and insecticide (of these, fertilizer is the largest component, representing at least 30% to 40% of total costs (SDP 1995, 2002; DPM 2005; GSO 2006). For the period prior to 2002 material input quantities are partly measured in terms of a “urea-used equivalent,” or total planted paddy area multiplied by the rate of fertilizer used per hectare per rice crop. The rates of fertilizer use for paddy production per hectare per rice crop are obtained from the SRP for the RRD and MRD and SDAFF (2001, 2006). This rate is adjusted in some nonprincipal rice growing provinces, based on reports provided by the GSO. For the period 2000–2002, material inputs are estimated as a multiple of the growth indices of total fertilizer consumption used by Vietnam (FAO 2007), using 1999 as a base year, and the actual current fertilizer use by provinces in 1999 (provided by Kompas 2004).
In Kompas (2004), material inputs include the nutrition content of all fertilizers (organic and chemical), insecticides, and seeds. The conversion factor used to aggregate organic and chemical fertilizers is similar to that used by Tang (1980) and Sicular (1988). The amount of organic fertilizer for the rice industry is obtained from the total amount of organic fertilizer used for agriculture. Organic fertilizer for agriculture is assumed to be supplied from two main sources: night soil and large animal manure (buffaloes, cattle, and pigs). Population-adjusted night soil is estimated based on the size of the rural population (GSO 2008) multiplied by a rural utilization rate (0.9). The standard number of large animals equals the sum of buffaloes, cattle, and pigs (GSO 2008), for which the weighted ratios are 1, 1, and 0.33, respectively. Organic fertilizer for rice production is obtained by multiplying the amount of organic fertilizer for agriculture with the weighted ratio between food grain area sown to the total sown area for cultivation. The chemical fertilizer data used for rice production is derived directly by multiplying the average amounts of chemical fertilizer used in the north (165.4 kg/ha) and the south (193 kg/ha) (drawn from the SRP) and the rice area in every province (SDAFF 2006). The data set for insecticides is constructed by multiplying the average use of insecticide per hectare in the year 1992, or 5.8 kg and 7.6 kg in the north and south, respectively, and the total rice area (SDAFF 2006). In a similar manner, the data for seeds are calculated from the average use of seeds per hectare, or 140 kg/ha and 240 kg/ha in the north and south, respectively, multiplied by the total rice area (GSO 2008). The time series for chemical fertilizer is calculated from the average amounts of chemical fertilizer used per hectare multiplied by cultivated area in each year (SDAFF 2001, 2006). The time-series data for insecticides and seeds are calculated from the average use of insecticide and seeds per hectare (SDAFF 2001, 2006) multiplied by rice area for each year (SDAFF 2006). To verify, an updated measure of fertilizer (in terms of quantities) for 2003–2006 is estimated from the trend of average fertilizer use in Southeast Asia (FAO 2007), using the 2002 as a base year.
5. Capital quantity and prices. The capital variable for 1985–1999 is based on data from Kompas (2004), following a similar approach to that used by McMillan, Whalley, and Zhu (1989), and assumes that the physical capital can be represented by the capacity of tractors in combination with a buffalo-equivalent measure. The conversion from the number of draught animals to the capacity of tractors is based on well- known observations in Pakistan (Blomqvist 1986), indicating that a bullock-day (a pair of bullock working 8 hours) is approximately the same as a tractor-hour, with a typical tractor being between 15 and 25 horsepower. In the Vietnamese case, we assume that one cattle-day or buffalo-day is equivalent to roughly 0.6 bullock-days (or 14 hours of work by one pair of cattle or buffalo is roughly 8 hours of work by one bullock), with a typical tractor being 15 horsepower. The data sources for the capacity of tractors and number of buffaloes and cattle are provided from MAFI (1991, 1994), SDAFF (2001), and for recent series by SDAFF (2006) and the GSO (2008). The capital measure used for rice from 1985 to 1999 is drawn from Kompas (2004). The capital measure used from 2000 to 2002 is estimated from the planted area, and the average capital cost for rice from DPM (2005). The updated capital quantity variable for 2003–2006 is estimated and verified from the trend of tractors used in Southeast Asia (FAO 2007), using the 2002 as a base year. Capital prices for 1985–1999 are obtained from Che, Kompas, and Vousden (2001) and Kompas (2004), with additional details for the early part of this series provided by Che, Kompas, and Vousden (2006). An updated series is drawn from district-level data obtained from the GSO (2008).
Footnotes
The authors are, respectively, professor and director, Crawford School of Economics and Government, Australian National University, Canberra; adjunct associate professor, Crawford School of Economics and Government, Australian National University, Canberra; researcher, Centre for Analysis and Forecasting, Vietnam Academy of Social Sciences, and Ph.D. scholar, Crawford School of Economics and Government, Australian National University, Canberra; senior lecturer, Vietnam Forestry University, Hanoi.
↵1 For a detailed discussion of land reform policy in Vietnam, see Chu et al. (1992), Fforde (1996), and Marsh and MacAulay (2002).
↵2 See Glewwe, Gragnolati, and Zaman (2002) for the distributional effects of poverty reduction in Vietnam, based on early household surveys in 1993 and 1998, and Ravallion and van de Walle (2008) for the recent welfare impacts of land reform.
