Non-farm Income, Diversification and Welfare: Evidence from Rural Vietnam

Transcription

1 Non-farm Income, Diversification and Welfare: Evidence from Rural Vietnam Luu Duc Khai Central Institute for Economic Management Christina Kinghan and Carol Newman Trinity College Dublin Theodore Talbot University of Copenhagen Abstract Many emerging economies are characterized by an on-going structural transformation in economic activity that involves graduation from traditional forms of agriculture. This paper contributes to the active area of research on the household-level determinants of diversification and its welfare consequences for household members. We use data from the Vietnam Access to Resources Household Survey (VARHS) from 2008, 2010, and 2012 to study the extent to which rural households in our sample diversify away from own-farm agriculture into waged employment and operating a household enterprise. Consistent with macroeconomic aggregates indicating a significant and on-going shift in the shares of labor allocated, and national value attributed, to agriculture, we observe significant levels of diversification. Moreover, we find this diversification is welfare enhancing, albeit with significant heterogeneity across geographic areas and by household characteristics. In addition to evaluating observable household-level information that is associated with the observed transition out of wholesale agricultural specialization, we provide policy recommendations to enable more households to diversify and increase their welfare gains from diversification. Acknowledgements We are grateful to participants in various seminars at the Central Institute for Economic Management (CIEM) in Hanoi for useful comments and suggestions. We would like to sincerely thank the effort invested by the survey teams of the Vietnamese Institute of Labor Science and Social Affairs (ILSSA). Financial support under the Business Sector Programme Support (BSPS) by Danida is gratefully acknowledged. The usual caveats apply. 1

2 1. Introduction Many developing countries continue to struggle to raise growth rates of real income, but a significant minority of emerging economies have succeeded in generating significant (albeit not always stable) growth. A common feature of the convergence of these low-income countries is a fundamental change in the pattern of economic activity, as households reallocate labor from traditional agriculture to more productive forms of agriculture and modern industrial and service sectors. The combination of this large-scale shift in labor allocation and the resulting change in the composition of economic output are collectively referred to as the structural transformation of an economy. Vietnam s economy is undergoing a large-scale transformation of this kind. Nearly 45% of the country s exports in 2002 were crude oil and rice; this share had fallen to under 20% by Similarly, agriculture s share of value added went from around 25% in 2000 to less than 22% in 2012, while over the same time frame the share of GDP attributed to the industrial sector grew from 36% to more than 40% (WDI, 2012). This transformation manifests itself at the household level in a diversification away from traditional own-farm agriculture towards other activities (which, of course, may still be part of the agricultural value chain). Despite achieving an average 6% growth over the last decade in PPP-adjusted USD, in 2012 the agricultural sector (defined to include forestry and fisheries) contributed about a fifth of national value-added and employed 47% of the overall labor force. Examining just the allocation of labor in rural areas, table 1 shows a significant shift in the patterns of employment across Vietnam over the last decade. Table 1: Pattern of employment in Vietnam Agriculture, Forestry, and Fisheries 2 Industry and Construction Services National Average Red River Delta Northern Midland and Mountain Areas North Central and Central Coastal Areas Central Highlands South East Mekong River Delta Source: Authors calculations from General Statistics Office: Agriculture, Forestry, and Fisheries Census, 2001, 2006, and Indeed, Vietnamese policymakers prioritized this shift away from own-farm agriculture into modern productive sectors: the national five-year Socio-Economic Development Plan ( ) endorsed by the Communist Party of Vietnam at the 10 th Party Congress, for example, calls for [promoting] the restructuring of rural labor to rapidly reduce the proportion of employees in agriculture and increase the proportion of employees in industry and services. Despite rapid urbanization, rural areas remain the focus of this transition: with national population crossing the 90 million mark in 2013, the number of rural households increased by 1.58 million between 2006 and The combination of large-scale macroeconomic structural transformation and microeconomic diversification make Vietnam a unique environment in which to study the dynamics and

3 effects of diversifying labor allocation and investment away from traditional own-farm agriculture. Studying the extent of diversification and the effects of diversification on welfare is immediately relevant to national economic policy. It also provides evidence that may be useful to policymakers in several other structurally-similar countries that are now negotiating the process of industrial transition while remaining primarily agricultural and rural. This paper explores the pattern of diversification in rural Vietnam between 2008 and We examine whether households that have diversified their economic activities are better off than households who have remained in specialized agricultural production. We focus on the extent to which there are heterogeneous welfare effects for households with different characteristics and examine how exposure to shocks impacts on the choice to diversify and its effects. We analyze factors that determine the transition out of specialized agriculture and the choice of economic activities of diversified households. Our primary objective is to provide an evidence base for informing policy on the transformation process now occurring throughout rural Vietnam. The paper is structured as follows. In Section 2 we provide an overview of the related literature focusing in particular on experiences of diversification into non-farm activities in other developing countries. Section 3 describes the data while our empirical analysis is presented in Section 4. Section 5 concludes the paper with a summary of our key findings and recommendations for policy and future research. 2. Related Literature The impact of income diversification on the livelihoods of the rural poor has been well documented in the literature. In what follows, we emphasize some key findings of relevance to the discussion presented in this paper. In general, research into the role of diversification into off-farm income generating activities has concluded that: i) diversification has the potential to increase the incomes and welfare of the poorest and most vulnerable; and ii) that it also has the potential to increase income inequality given that the least wealthy households tend to diversify into low return activities in contrast to wealthy households who invest in more productive activities. These key findings help to motivate the analysis of rural households in Vietnam presented in this paper. Owusu et al. (2010) highlight the potential for diversification to improve the living standards of households in Northern Ghana due to the low productivity of agriculture resulting from poor agro-ecological factors. They empirically examine the impact of non-farm activity on household income and food security status and find that non-farm work has a positive and statistically significant effect on both. However, as highlighted by Abdulai and Crole-Rees (2001), poorer households tend have fewer opportunities to participate in non-farm activity and hence have less diversified incomes. In their study of rural households in Mali they find that a lack of capital seems to act as a constraint in the development of non-farm activities and makes it difficult for poorer households to diversify away from subsistence agriculture. Those living nearer markets have higher levels of education or those with greater land holdings are more likely to diversify their income, highlighting the significance of entry constraints in explaining households diversification decisions. Bezu et al. (2012) examine whether non-farm employment leads to higher consumption expenditure growth in Ethiopia. They also examine whether non-farm employment is propoor by estimating its impact on expenditure growth separately for rich and poor households. 3

4 Using household data over a 10-year period they find a strong positive relationship between a household s non-farm income share and its subsequent expenditure growth, highlighting that households engaged in non-farm activities experience higher growth in expenditure and that this growth rate increases as the level of investment in non-farm employment increases. The positive relationship between the share of non-farm income and expenditure growth holds for both poor and well-off households. However, they find that relatively wealthier households benefit more from off-farm activity than poorer households concluding that diversification into non-farm enterprises may increase rural income inequality. Lay et al (2008), focus further on the diversification of household resources into low or high return activities by examining whether a dichotomous non-agricultural sector exists in a poor rural area of Western Kenya. They define high return activities as those requiring special skills, particular assets and a clean and healthy appearance such as vehicle repair, hair dressing and beauty, hotels and restaurants and NGO s or international organizations. All remaining non-agricultural activities are classified as low return, such as enterprises run by one household member only. This classification of activity is done to reflect survival-led versus opportunity-led income diversification. They find that 55% of households diversify into some form of non-farm activity and interestingly highlight how more than a third of households undertaking a high-return activity in the non-agricultural sector also pursue some low-return activity. They find that the share of low-return non-farm income falls with the education level of the household head and the size of landholding. Moreover, they find that wealth barriers to high-return activities do exist. Lay et al. s (2008) findings also reinforces the conclusion of Bezu et al (2012), that given the concentration of high-return activity among already wealthy households, it is possible that diversification may aggravate inequality. Further support for this argument is provided by the evidence in the literature on the link between diversification and shocks. Kijima et al. (2006) find evidence in Uganda that poorer households respond to shocks by being pushed into low return non-farm jobs. Wealthier households, in contrast, are better able to cope with shocks as they are more likely to earn income from salaried employment or high-return self-employed business opportunities thus exacerbating income differentials in the longer run. In contrast, Van Den Bery and Kumbi (2006) find in their study of diversification in Oromia, the largest state in Ethiopia, that if barriers to entry to the non-farm sector are low, diversification will have an equalizing effect on the distribution of income. The authors highlight the positive role that diversification into non-farm activities can have on income levels in poor rural areas. Similarly, Schindler and Giesbert (2012) examine welfare dynamics among rural households in Mozambique. They find that drought has a negative impact on a household s asset accumulation but households in which at least one member has regular non-farm work experience less adverse asset growth from a drought than those without non-farm wage opportunities suggesting that income diversification has a positive impact in the aftermath of an exogenous shock. The authors conclude that non-farm labor market opportunities are an important means of mitigating the effects of drought in the short term. Barrett et al (2005) undertake an analysis in Cote d Ivoire, Kenya and Rwanda with the aim of informing the debate on the determinants of patterns of asset allocation and income sourcing observed across individual studies of diversification. The authors emphasize the role of household heterogeneity in both constraints and incentives and how this impacts on the 4

5 income diversification patterns of households, with constraints not only restricting certain types of diversification but also compelling diversification into low return activities. They emphasize the role of shocks in particular in driving households into non-farm employment as an ex-post response to risk. In general, however, the authors find that livelihood strategies that include non-farm income sources, and in particular activities that exclude unskilled labor, are associated with higher income and greater upward earnings mobility. Households that do not have access to non-farm activity and also do not have productive non-labor assets often derive their full income from on-farm agriculture and this dependence often leads to these households being trapped in poverty. This again highlights the importance of diversification into non-farm activities for incomes in rural areas. However, they also find evidence of rural market failures that force the poorest households to undertake non-farm activity with low returns, whereas wealthier households are more able to enter high return activities. In summary, the literature exploring the diversification of economic activities in low income countries highlights the importance of the rural non-farm economy for income generation that benefits the least well off. The evidence for inequalities, however, whereby less well-off households tend to diversify into low-return activities while those with better endowments of assets and higher incomes are able to take advantage of high-return non-farm opportunities, highlights the need for country specific research to establish the type of diversification taking place and its impact on households. In this paper we focus on the livelihoods of rural households in 12 provinces of Vietnam and document how they have evolved over the 2008 to 2012 period. Our aim is to establish the extent to which structural transformation is occurring and whether the diversification of livelihoods out of agriculture into other economic activities enhances household welfare. We also examine which households diversify and whether there are heterogeneous effects in the impact on welfare of the transformation process. 3. Data and Summary Statistics Our data come from three rounds (2008, 2010 and 2012) of the Vietnam Access to Resources Household Survey (VARHS). 1 The survey instrument provides detailed information about the incomes, assets, and access to public services and other resources of rural households in 12 provinces in Vietnam. 2 While the full dataset includes over 3,000 households for the purpose of this paper we focus on the balanced panel (i.e. the same households surveyed in each year) of 1,873 households. 3 Table 2 documents the economic activities that households in our sample were engaged in during the 2008 to 2012 period. We consider three types of economic activities, agriculture, waged employment and household enterprises, and their interactions. In 2008, agriculture is the main activity with the majority of households specialized in agricultural production. Between 2008 and 2010, however, a major structural transformation took place with 1 The survey was developed in collaboration between the Development Economics Research Group (DERG), Department of Economics, University of Copenhagen and the Central Institute of Economic Management (CIEM), the Institute for Labor Studies and Social Affairs (ILSSA) and the Institute of Policy and Strategy for Agriculture and Rural Development (IPSARD), Hanoi, Vietnam. 2 The survey is implemented at the same time of year in each round to ensure that responses are internally consistent and to avoid issues relating to the seasonality of income and other types of household behavior. 3 See CIEM (2009), CIEM (2011) and CIEM (2012) for a comprehensive descriptive report of the data gathered in each round of the survey. 5

6 diversification of economic activities into waged employment. While the majority of households continued to engage in the agricultural sector, many expanded the range of activities they were engaged in. Few switched out of agriculture completely. This is evident from the tiny proportion of households that engage in waged employment and household enterprises with no agricultural activities. Table 2: Economic activities of households % hhs Agriculture only Agriculture + Labor Agriculture + Enterprise Agriculture + Labor + Enterprise 6 Labor only Enterprise only Labor + Enterprise n = 1,873 No activity Table 3 shows the change in the proportion of income coming from each type of economic activity. Of particular note is the decline in the proportion of income from agricultural sources. Between 2008 and 2010 a notable increase in the importance of waged income is observed but this declines slightly in importance between 2010 and 2012 when household enterprise income makes a higher contribution to the total income of households. Table 3: Proportion of income by economic activities of households % Agriculture Labor Enterprise n = 1,481. Households that report zero positive total income from these sources in any given year are excluded from this table leading to a reduced sample size. Table 4 presents an occupation transition matrix for the period 2008 to 2010 and from 2010 to It illustrates the movement of households between different occupation types over time. For example, in ,029 households were in the agricultural sector with no other economic activities. Of these households only 32.65% stayed in agriculture in 2010 while the rest diversified. Slightly over half (50.44%) diversified by combining agriculture with waged employment while just over 8% combined agriculture with some form of non-farm non-wage activity. In contrast, of the 220 households that were involved in agriculture and waged employment in 2008, over half (54.55%) stayed in this occupation grouping while over 11% diversified further by adding an enterprise activity. An additional 25.45% specialized to focus on agricultural production only. Of those households in agricultural production and with an enterprise activity in 2008 (239 households in total) over a third continued with these activities in 2010 while 24.27% diversified further by adding a waged activity. Another 24.27% switched from a household enterprise to waged employment. Similar transitions are observed for households that have all three activities in 2008 (166 households in total). Most shifted between different combinations of agriculture, waged employment and enterprise activities. For the other categories, those without agricultural activities to begin with, there was less transition between groups but it should be noted that the sample size is much smaller for these groups. Between 2010 and 2012 there is less transition between different types of activities although a move away from agriculture is still evident. In 2010 only 417 households were exclusively engaged in agricultural activities. Of these, half continued to exclusively engage in agricultural production while 36% diversified by engaging in waged employment; only a small proportion of households diversified by engaging in household enterprise activities.

7 Most households involved in agricultural activities and waged employment in 2010 (771 households) continued with activities in both sectors in 2012 (69.39%). In contrast, households involved in some form of enterprise activities appear to have been more mobile in transitioning between occupation categories. This suggests that there may be greater stability associated with waged employment that is less evident for households that are engaged in enterprise activities leading them to shift more between economic activities. This is not surprising given the additional risks that are incurred when households are self-employed and the higher variance in returns to enterprise activities. Table 4: Occupation transition matrix Ag only Ag + Lab Ag + Ent Ag + Lab + Ent Lab only Ent only Lab + Ent Ag only (n=1,029) Ag + Lab (n=220) Ag + Ent (n=239) Ag + Lab + Ent (n=166) Lab only (n=18) Ent only (n=60) Lab + Ent (n=10) Ag only Ag + Lab Ag + Ent Ag + Lab + Ent Lab only Ent only Lab + Ent Ag only (n=417) Ag + Lab (n=771) Ag + Ent (n=229) Ag + Lab + Ent (n=186) Lab only (n=62) Ent only (n=50) Lab + Ent It is clear from the descriptive statistics presented in Tables 2 to 4 that rural households in Vietnam increasingly rely on different activities to generate income. The aim of this paper is to explore the extent to which this leads to improvements in welfare. We consider three different indicators of welfare: income, food consumption and wealth. Income is calculated as the total income of households from all sources and expressed in real 2012 prices. Adjustments are also made for provincial price differences. Food consumption is measured as the value of food items consumed in the previous month. It is based on a select number of representative items. Food consumption is a more widely used measure of household wellbeing due to measurement errors in income data gathered from household surveys. This is particularly the case when income is generated from a number of different sources as in our case (Deaton, 1997; Fox, 2013). Household wealth is an important covariate of income measures and determinant of consumption. Because this wealth is generally embodied in a large number of physical assets, we follow the established practice (popularized by Filmer and Pritchett (2001)) of constructing an asset-based wealth measure that summarizes households asset holdings based on weights calculated by Principle Component Analysis (see McKay and Tarp (2011) for details of the procedure in the context of VARHS data; here, we use a subset of the assets used in the calculation of the index presented in that paper). The resulting index is a stock measure rather than the flow represented by real income captured directly by the survey instrument and included in the set of conditioning information. The specific assets used in the calculation of this index are: number of plots of land, total land area owned (square meters), total irrigated area owned (square meters), area of main dwelling (square meters) and whether the dwelling has a good light source, toilet, or water source, and the number of cows, buffalo, 7

8 pigs, chicken, color TVs, DVD players, telephones, motorcycles, bicycles, pesticide sprayers, and cars the household owns. Summary statistics for each of these variables are provided in Table 5 along with a disaggregation across households engaged in different types of income generating activities. Overall, household income increased significantly between 2008 and 2010 in real terms. However, between 2010 and 2012 we observe very little increase overall. This can be explained by the very high rate of inflation in Vietnam during this period. Despite the stagnation in incomes, average monthly food expenditure increased steadily over the sample period as did overall wealth. The better off households in all time periods are those with a household enterprise. This holds across all welfare measures. Households that are specialized in agriculture do the worst on each measure in each year. This suggests that diversification out of agriculture is associated with higher levels of welfare, although causal inferences cannot be made on the basis of these descriptive statistics. Identifying the nature of the relationship between diversification and welfare outcomes is the focus of our empirical analysis presented in Section 4. Table 5: Welfare measures, group-wise means Income Food consumption Assets 2008 Total 56,574 1, Ag Only 49,650 1, Ag + Lab 47, Ag + Ent 68,930 1, Ag + Lab + Ent 80,825 1, Lab Only 50, Ent Only 80,232 1, Lab + Ent , Total 76,596 1, Ag Only 52, Ag + Lab 69,243 1, Ag + Ent 103,568 1, Ag + Lab + Ent 99,635 1, Lab Only 81,448 1, Ent Only 177,142 1, Lab + Ent 106,794 1, Total 76,762 1, Ag Only 60,622 1, Ag + Lab 72,769 1, Ag + Ent 88,126 1, Ag + Lab + Ent 87,915 1, Lab Only 101,972 1, Ent Only 175,844 1, Lab + Ent 103,865 1, n = 1, Empirical Analysis 4.1 The welfare effects of diversification Identifying a causal relationship between diversification of economic activities and welfare is complicated by a number of confounding factors, particularly when income or wealth is the outcome variable of interest. First, there may be self-selection of households into more productive activities. In other words, richer or wealthier households may choose to diversify rather than diversification in itself leading to higher levels of incomes or wealth. Second, as highlighted above, measurement errors in income are commonplace in household surveys of 8

9 this kind, particularly with multiple income sources. This will further complicate the identification of unbiased coefficient estimates in a standard regression model. Given these constraints we focus on consumption as our outcome measure of interest and control for past wealth to address the self-selection problem. As such to determine the impact that diversification has on welfare we consider the following regression model: C it β α τ e (1) 1Sit β2xit β3xit 1 β4incomeit 1 β5wealthit 1 i t it The key variables of interest are the sources of income of households. They are included in the vector S it in the form of dummy variable indicators of the various categories described above with households that are involved in agriculture only (i.e. specialized agriculture) forming the base category. The vector X it includes time varying household characteristics, namely, household size, household size squared, whether the household head is female, age of the household head, age squared, the education level of the household head, the number of children in the household, whether the household is of Kinh ethnicity, whether the household is born in the commune and whether the household is classified as poor by the authorities. Current period wealth is also included as a control variable within this vector. An additional complication with this specification is the need to control for current period income of households which is collinear with the sources of income and with the other control variables. If we assume that the generation of income is a dynamic process, in that past values will determine future values, the lag of income and the lag of other time varying household characteristics (included in X it 1 ) should serve as adequate controls. The model includes household fixed effects, α i, which means that identification of the impact of diversification of livelihoods on welfare comes from within-household variations overtime controlling for income, wealth, time varying household characteristics and past values of income and wealth. The latter is included as a control for selection. Time dummies, τ t, are also included and e it is the statistical noise term. Summary statistics for each all variables included in the analysis are presented in Table A1 of the Appendix. The results for the core variables of interest from the main model given in equation (1) are presented in Table 6. 4 Table A2 of the Appendix details the full set of results. Our dependent variable is the log of real consumption per capita. Making the per capita adjustment is particularly important in this model given that diversification and food consumption will be related to the size of the household. Adjusting to per capita consumption overcomes potential bias in the estimates of the impact of diversification on welfare. We also include household size and its square to control for the fact that there may be economies of scale associated with food consumption in larger households. A log transformation is used to reduce the impact of outliers and for ease of interpretation of the parameter estimates. Columns (1) to (4) show that households that are diversified are better off than households that are specialized in agriculture. In particular, when all control variables are included (column (4)) we find that households that are engaged in agriculture with some other type of activity, waged employment, a household enterprise or both, have higher levels of consumption per capita than those that are engaged in agricultural production only. The coefficient estimates suggest that compared with households that are fully specialized in agricultural production, the fully diversified households do the best with consumption levels 4 We exclude households that report having no economic activities. 9

10 per capita that are almost 23% higher than households specialized in agriculture, followed by households that are engaged in agriculture and enterprise activities with consumption levels per capita that are almost 17% higher, while households engaged in agriculture and waged employment have consumption levels per capita that are 8% higher. Households with no agricultural activities are no different in welfare terms to those that are specialized in agriculture. Table 6: Impact of diversification on household welfare (1) (2) (3) (4) Ag + Lab (0.038) 0.084** (0.038) 0.083** (0.038) 0.083** (0.038) Ag + Ent 0.162*** (0.053) 0.173*** (0.050) 0.168*** (0.051) 0.168*** (0.051) Ag + Lab + Ent 0.182** (0.053) 0.227*** (0.052) 0.227*** (0.052) 0.227*** (0.052) Lab Only (0.107) (0.098) (0.095) (0.095) Ent Only (0.139) (0.126) (0.125) (0.124) Lab + Ent (0.115) (0.110) (0.111) (0.111) HH characteristics No Yes Yes Yes Current income controls No No Yes Yes Selection control No No No Yes Time dummies Yes Yes Yes Yes R-squared Number of households 1,835 1,835 1,835 1,835 Number of observations 3,588 3,588 3,588 3,588 Note: Household fixed effects model estimated in each case. Standard errors clustered at the household level are presented in parentheses. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. The results in Table 6 represent our baseline findings on the benefits of diversification on welfare as measured by real per-capita food consumption. However, because household consumption is an autoregressive process, we ensure the robustness of our results on diversification and welfare by directly estimating a model of the form: C it β e (2) 1Sit β2x it Cit 1 i t it A challenge in estimating this model is that including household fixed effects requires the assumption that the within-transformed lagged dependent variable is orthogonal with respect to the within-transformed errors, an assumption that is mechanically violated. 5 We take this bias into consideration by taking advantage of the fact that the effect of diversification on per-capita consumption is bracketed by a model that includes household fixed effects only and a model that includes lagged consumption only (see Angrist and Pischke (2008) for a detailed discussion of this property). Our objective in estimating this model is to ensure the point estimates reported in Table 6 of the effect of the various modes of diversification on 5 It is possible to estimate the effect of a lagged dependent variable on an outcome of interest with fixed effects if we adopt alternative approaches such as using prior lags of the dependent variable as instruments, but these approaches (like that pioneered by Arellano and Bond (1991)) are more appropriate in panels with longer time dimensions or if we are able to accept restrictive assumptions about the exogeneity of earlier observations of per-capita consumption. 10

11 welfare are robust to directly including lagged consumption in our conditioning information set. 6 The results are presented in Table 7 (Table A3 of the Appendix shows the full set of fitted coefficients and associated test statistics) and support our initial conclusion: the effect of agriculture combined with other income-earning activities remains significantly and positively associated with real consumption (see columns (2) and (3)). Table 7: Impact of diversification on household welfare, LDV and FE estimates Ag + Lab Ag + Ent Ag + Lab + Ent Lab Only Ent Only Lab + Ent (1) (2) (3) Original model LDV FE 0.083** 0.054** 0.084** (0.038) (0.024) (0.038) 0.168*** 0.115*** 0.173*** (0.051) (0.031) (0.051) 0.227*** 0.163*** 0.227*** (0.052) (0.031) (0.052) *** (0.095) (0.048) (0.098) *** (0.124) (0.059) (0.127) (0.111) (0.059) (0.111) HH Fixed Effects Yes No Yes Lagged consumption No Yes No Current income controls Yes No No Selection control Yes No No R-squared Number of households 1,835 1,835 1,835 Number of observations 3,588 3,588 3,588 Note: LDV indicates lagged dependent variable. FE indicates the household fixed effects model. The model estimated in column (3) excludes current income and the selection control so that the set of control variables are comparable to those included in column (2). Standard errors clustered at the household level are presented in parentheses. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. In Table 8 we disaggregate the diversification of economic activities further separating out households that moved out of specialized agriculture between survey rounds from other types of diversified households. 7 We find that the transition out of specialized agriculture is welfare enhancing. The per capita consumption of households that move from being engaged in agricultural production only into other types of production activities is 10% higher than those who remain specialized (column (1)). When this is disaggregated by type of activity we find that this result is driven by those households that diversify by adding an enterprise activity or adding both an enterprise activity and waged employment to their portfolio of production activities. Of the non-transition households those that are diversified also perform better, particularly those that are involved in both labor and enterprise activities. 6 Food consumption recorded by VARHS is based on recall of the month before the interview. As such, each observations on consumption for a particular household is far apart in time. 7 The full set of results are presented in Table A4 of the Appendix. 11

12 Table 8: Impact of diversification out of agriculture on household welfare (1) (2) Transition out of Ag 0.101*** (0.038) Of which: Into Ag+Lab (0.043) Into Ag+ Ent 0.204*** (0.077) Into Ag+Lab+Ent 0.180** (0.083) Into Other (0.072) Activities of non-transition hhs: Ag + Lab (0.114) Ag + Ent 0.193* (0.116) Ag + Lab + Ent 0.313** (0.118) Lab Only (0.154) Ent Only (0.171) Lab + Ent (0.167) (0.114) 0.200* (0.116) 0.319*** (0.118) (0.154) (0.171) (0.167) HH characteristics Yes Yes Current income controls Yes Yes Selection control Yes Yes Time dummies Yes Yes R-squared Number of households 1,835 1,835 Number of observations 3,586 3,586 Note: Household fixed effects model estimated in each case. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level. 4.2 The determinants of diversification We now turn our attention to exploring the characteristics of households that transition out of agriculture. To do so, we estimate a probit model of the probability that a household transitions out of agriculture in a given time period as described in equation (3). P Trans 1 Φ lagx α lagincome α lagwealth α shock π i 1 i 2 i 3 i 4 α (3) Where Trans i takes a value of one if a household moved from specialized agricultural production to some other combination of economic activities and zero otherwise. The matrix X is as before, a matrix of time varying household characteristics, and is included at a lag. Income and wealth are also included at a lag. The variable shock i is an indicator for whether the household suffered a natural shock in the previous two years and π p are province dummies to capture regional variations in the transition out of specialized agriculture. The function Φ. is the standard normal cumulative distribution function, a standard assumption of the probit model, and is used to ensure that the predicted values of the model lie within a 0-1 interval and so are consistent with interpreting the left hand side variable as a probability. Here we estimate the model separately for the cross-section of households in 2010 and 2012 respectively. We do not include household fixed effects as we are interested in the variation 12 i p

13 between households in any given time period rather than the variation within households overtime. We estimate the model conditioning on households that were specialized in agricultural production in the previous period. The coefficients of the probit model do not have an obvious interpretation and so instead we present marginal effects which can be interpreted as the impact of a one unit change in the independent variable on the probability of a household transitioning from specialized agriculture. The results are presented in Table 9. The characteristics that are significant in determining the probability that a households transitions out of specialized agriculture vary somewhat across the years. Of particular note is the difference in the impact of income; in 2012, higher income households are less likely to leave specialized agriculture while in 2010 income is not a significant determinant. This suggests that diversification in the Vietnamese case is not driven by higher income households. In 2010, female headed households are more likely to transition while in 2012 the gender of the household head is not a significant determinant. Natural shocks, which include landslides, typhoons, storms, drought, pest infestation, crop disease and avian flu, have a positive impact on the probability of transitioning out of specialized agriculture in 2010 but have no significant impact in This suggests that, at least in 2010, diversification into other activities might be a mechanism that households use to cope with shocks that affect agricultural production (see Wainwright et al. (2011) for a full analysis of the role of diversification in helping households to manage risks using the VARHS data). Provincial differences are also evident. 8 We find that households in Ha Tay (the base category) are most likely to move out of specialized agriculture compared to households in other in Lao Cai, Phu Tho, Dien Bien, Quang Nam and Dak Lak in 2010 and Dien Bien, Nghe An, Dak Lak and Dak Nong in 2012, where transition is much less likely. A number of common household characteristics emerge as determinants of the transition out of agriculture in both years. For example, larger households are more likely to diversify. This is not surprising given that in larger households there is more scope for individual members to engage in different types of activities. This effect could partially be because in larger households with limited arable land holdings, more household members will be forced to seek non-agricultural employment. While we indirectly control for land holdings by including number of plots, area owned, and irrigated area owned in our wealth index measure, we do not directly control for these factors in our vector of household-level information. There is strong evidence that more educated households are more likely to move from specialized agriculture, particularly in 2010 but also to a certain extent in It is also the case that households classified as poor by the authorities are more likely to move from specialized agricultural production. Overall, this analysis suggests that there is significant heterogeneity in the types of households that are diversifying within and across the years of our dataset. For the most part it appears that it is poorer, lower income households that are choosing to diversify and those that are more vulnerable (e.g. those exposed to shocks and female-headed households). There is some evidence, however, that more educated households are more likely to diversify suggesting that, as has been found in case studies of other developing countries, there may also be different types of diversification taking place in the Vietnamese case: i) households that are pushed into diversification; and ii) those that are availing of more productive opportunities. 8 It is important to note that provincial differences are absorbed by the time invariant household fixed effect in the previous models and so are implicitly controlled for in estimating the impact of diversification on welfare. 13

14 Table 9: Determinants of transition out of specialized agriculture (marginal effects) (1) (2) (3) (4) (5) (6) (7) (8) Dep Variable: Trans Trans Lab Ent Lab + Ent Lab Ent Lab + Ent Lag Log income (0.026) *** (0.043) 0.047* (0.027) (0.019) *** (0.018) (0.027) (0.027) Lag wealth * ** *** 0.108*** (0.025) (0.042) (0.024) (0.017) (0.016) (0.036) (0.036) Lag Size 0.129*** 0.162*** *** *** (0.034) (0.059) (0.036) (0.026) (0.025) (0.047) (0.047) Lag Size ** ** *** 0.011*** squared (0.003) (0.005) (0.003) (0.002) (0.002) (0.004) (0.004) Lag Female 0.179*** * * * * Head (0.035) (0.077) (0.036) (0.026) (0.027) (0.026) (0.026) Lag Age Head ** ** (0.008) (0.013) (0.007) (0.005) (0.005) (0.007) (0.007) Lag Age Head ** ** squared (0.0001) (0.0001) (0.000) (0.000) (0.000) (0.0005) ( ) Lag Ed Head 0.106** * ** cat 2 (0.047) (0.096) (0.070) (0.035) (0.061) (0.030) (0.030) Lag Ed Head 0.215*** cat 3 (0.047) (0.090) (0.059) (0.031) (0.045) (0.035) (0.035) Lag Ed Head 0.174*** 0.222** cat 4 (0.049) (0.097) (0.065) (0.037) (0.052) (0.043) (0.043) Lag Ed Head 0.186*** *** ** *** cat 5 (0.051) (0.126) (0.115) (0.025) (0.113) (0.030) (0.027) Lag Ed Head 0.273*** *** *** cat 6 (0.043) (0.273) (0.111) (0.050) (0.098) (0.024) (0.024) Lag Number of * *** 0.073*** children (0.018) (0.035) (0.015) (0.011) (0.010) (0.022) (0.022) Lag Kinh Head ** ** (0.048) (0.101) (0.040) (0.030) (0.027) (0.066) (0.066) Lag Head born in commune (0.049) (0.079) (0.044) (0.027) (0.037) (0.052) (0.052) Lag Poor 0.078** 0.134* 0.058* ** (0.037) (0.075) (0.033) (0.021) (0.025) (0.040) (0.040) Natural shock 0.086*** ** ** (0.033) (0.063) (0.032) (0.022) (0.022) (0.045) (0.045) Lao Cai *** * 0.436** (0.090) (0.148) (0.176) (0.195) (0.054) (0.118) (0.118) Phu Tho *** * 0.170* *** *** (0.065) (0.102) (0.100) (0.103) (0.034) (0.028) (0.028) Lai Chau ** 0.692*** (0.094) (0.164) (0.070) (0.133) (0.090) (0.206) (0.207) Dien Bien *** *** ** 0.391** (0.079) (0.108) (0.178) (0.202) (0.067) (0.086) (0.086) Nghe An * ** 0.179* * * (0.069) (0.110) (0.103) (0.107) (0.042) (0.033) (0.033) Quang Nam ** *** (0.066) (0.105) (0.065) (0.063) (0.021) (0.035) (0.035) Khanh Hoa (0.118) (0.160) (0.161) (0.095) Dak Lak *** *** *** 0.071*** *** (0.084) (0.109) (0.129) (0.129) (0.024) (0.024) (0.023) Dak Nong ** (0.097) (0.132) (0.148) (0.154) (0.062) (0.062) (0.062) Lam dong * 0.060** ** (0.107) (0.147) (0.136) (0.136) (0.033) (0.030) (0.030) Long An ** ** (0.111) (0.028) (0.027) (0.000) (0.0004) (0.001) (0.000) (0.001) (0.0001) (0.000) (0.003) (0.002) (0.001) (0.011) (0.002) (0.001) (0.001) ( (0.001) (0.001) (0.001) (0.001) (0.007) (0.004) (0.001) * (0.0004) (0.003) ( (0.002) (0.003) Households 1, Note: The category lab includes all waged employment with and without agricultural production. This is also the case for the category ent and lab+ent. Robust standard errors are presented in parentheses. *** indicates significance at the 1% level, ** indicates significance at the 5% level, * indicates significance at the 10% level.

15 To explore this further we consider whether there are certain household characteristics associated with moving into waged employment, enterprise activities or a combination of the two. Table 9 also presents the results of a multinomial logit model which we estimate separately for each year and estimate conditional on transitioning out of specialized agriculture. We consider three different categories of households: those that move into waged activities, Lab, combining households that continue agricultural production with those that completely switch out of agriculture; those that move into enterprise activities, Ent, again with or without continuing agriculture; and those that move into both waged and enterprise activities, Lab+Ent, with and without agriculture. The multinomial logit specification allows us to determine the factors that make households more or less likely to move into each of these respective categories. The marginal effects presented in columns (3) to (5) relate to 2010 while those presented in columns (6) to (8) relate to They can be interpreted as the probability that a household is in a particular occupational grouping relative to the other two groupings and so the marginal effects for each variable will sum to zero. In other words if a variable has a positive impact on the probability that a household is in one category it must have a corresponding negative impact in one or both of the other categories. In 2010, higher income transition households are more likely to move into waged activities while lower income households are more likely to diversify into multiple activities, i.e. both labor and enterprise activities. While those that do diversify into multiple activities have lower incomes they score better on the wealth index suggesting that these households tend to have an abundance of illiquid assets that may motivate them to earn additional income from other activities. Female headed households are more likely to transition into waged activities as compared with enterprise activities as are households classified as poor by the authorities. Households of Kinh ethnicity are less likely to move into waged employment. Provincial variations in the choice of activities are also evident. Households in the base category, Ha Tay and Long An, are more likely to move into waged employment as compared with households in Lao Cai, Phu Tho, Lai Chau, Dien Bien and Nghe An, while in these provinces transition into enterprise activities is much more likely. This may reflect a lack of labor market opportunities in these provinces compared to Ha Tay and Long An requiring that firms wishing to transition out of agriculture must engage in their own non-farm non-wage household enterprise activities. The first point of note in relation to the 2012 results is that the group of households that transition from specialized agriculture into both waged employment and enterprise activities (Lab+Ent) is very small and as a result there are no statistically significant differences between these households and those that transition into waged employment or enterprise activities only. Also of note is that there are distinct differences in the factors that determine the choice of activities as compared with Wealth appears to be an important predictor of whether households diversify by starting an enterprise and has a negative effect on moving into waged employment. Larger households are more likely to choose waged employment while smaller households diversify into enterprise activities. As in 2010, female headed households are more likely to diversify into waged employment and less likely to start household enterprises. We find a strong education effect in 2012 that is not present in Educated heads of household are much more likely to diversify into waged employment while less educated households engage in enterprise activities. This can be explained by the fact that a higher level of education may be required to join the labor market while no education is needed to start an enterprise. In 2012 we also find that households that diversify in response to natural shocks are more likely to enter waged employment. A very different 15