Income Diversity and Poverty Transitions: Evidence from Vietnam. Van Q. Tran 1

Transcription

1 Income Dversty and Poverty Transtons: Evdence from Vetnam Van Q. Tran 1 Abstract The large share of the populaton n rural areas of the developng world has been dversfyng ther lvelhood to nonagrcultural actvtes. However, the amount of the lterature that dscusses the possble effects of the dversty on a household s well-beng s stll lmted. Ths study contrbutes to ths strand of the lterature by nvestgatng the effects of ncome dversty on poverty transtons. The analyss s based on household panel data collected n the 2000s from Vetnam and appled to a multnomal logt model. The results show that households wth better access to markets are more able to dversfy ther ncome sources to nonagrcultural actvtes and the dversty s helpful for a household to escape poverty or to avod fallng nto poverty. Key words: ncome dversty, poverty transtons, nonagrcultural ncome sources, Vetnam. JEL classfcaton: I32, O13, P36, R11 Acknowledgements: The author would lke to thank anonymous referees and the edtor of ths journal, and partcpants at a workshop n Bangkok, partcularly Peter Tmmer and Jkun Huang, for helpful comments on earler versons of ths paper. Fundng from the German Research Foundaton (DFG) for supportng the household surveys s gratefully acknowledged. Introducton The dynamcs of poverty have been one of the central ssues n development economcs. There has been a great deal of theoretcal studes (Cappellar and Jenkns 2004; Carter and Barret 2006) and emprcal studes (McCulloch and Baulch 1999; Glewwe, Gragnolat, and Zaman 2000; Woolard and Klasen 2005; Justno, Ltchfeld, and Pham 2008) that dscuss the transtons nto and out of poverty usng dfferent approaches and country cases. They have dentfed the characterstcs of a household, the prvate and publc assets a household possesses, the changes n macroeconomc condton such as trade reform, nflaton, and economc crss on the dynamcs of poverty. 1 Unversty of Economcs and Law, Vetnam Natonal Unversty 1

2 In fact, the majorty of the poor lves n rural areas and engages n agrcultural actvtes. Also, the large share of the rural populaton s dversfyng ther ncome sources to nonagrcultural actvtes that are usually of hgher returns and consequently makng them better off. Studes by De Janvry, Fafchamps, and Sadoulet (1991) and Knsey, Burger, and Gunnng (1998) ndcate that ncome dversfcaton s not only postvely correlated wth wealth but also wth an ncreased ablty to cope wth shocks. Dversfcaton s a way through whch rural households nsure themselves aganst the occurrence of such shocks, or, n other words, dversfcaton reduces lvelhood vulnerablty. Ths selfnsurance can also be seen as a negatve functon of the avalablty of socal nsurance, provded, for example, by the communty or famly. The better access to socal networks and nsttutons, the less lkely a household needs to apply self-nsurance systems as the dversfcaton of ncome portfolos. In contrast, socal captal can also foster the ablty to partcpate n many dfferent ncome actvtes. Ths study ams to contrbute to the lterature of vulnerablty to poverty by examnng the relatonshp between a household s dversty of ncome sources to nonagrcultural actvtes and ts transtons nto and out of poverty. The man goal s to dentfy whch households are more able to dversfy ncome sources and f such ncome dversty makes the household better off or prevents t from fallng nto poverty. Ths study examnes these research questons n the context of Vetnam, although the approach can be appled to other developng countres. Vetnam has been one of the most successful countres n the developng world n terms of economc growth and poverty reducton. The rapd economc growth, together wth market lberalzaton and trade openness that took place durng the last two decades, has lfted a large share of the populaton out of poverty (see Tran, Alkre, and Klasen 2015). Nonetheless, poverty s stll a central ssue n the country as nearly 43 percent of the populaton stll lves on less than $2 a day (World Bank 2013), and many people earn ther lvng by engagng n agrcultural actvtes. Varous populaton subgroups have benefted less from such development; households n rural areas have made slower progress than those n urban areas (see GSO 2011). Ths study uses three waves of a panel surveys from 2007, 2008, and 2010 of more than 2000 rural and per-urban households from three provnces n Vetnam. The drvers of poverty transtons are nvestgated va descrptve statstcs and emprcal results from multnomal logt (MNL) models. The analyses are based on the hypothess that a 2

3 household that dversfes ts ncome sources to nonagrcultural actvtes fnds t easer to escape poverty than a household that reles ts ncome only on agrcultural actvtes. The fndngs confrm that an ncrease n the share of nonagrcultural ncome to total household ncome s correlated wth the advancement of a household s well beng. Ths study s organzed as follows. Secton 2 descrbes the household panel data used n the analyss and presents the estmaton strategy. Secton 3 dscusses the results of the MNL models that hghlght the relatonshp between ncome dversfcaton and household well-beng. It also dscusses the robustness of the estmaton results. Fnally, Secton 4 concludes wth the key messages of ths paper. Emprcal Strategy Data Ths study s based on panel household surveys from 2007, 2008, and 2010 from the provnces of Hà Tĩnh (HT), Thừa Thên Huế (TTH), and Đắk Lắk (DL) n Vetnam for the purpose of the research project Vulnerablty n Southeast Asa beng run by a consortum of German unverstes and local research nsttutes (see Klasen and Wabel 2012). The survey covers more than 2000 households located n rural and per-urban areas n these three provnces. The three provnces have a dversty of agrcultural and ecologcal condtons wth mountanous, hghland, lowland, and coastal zones. The surveys collect nformaton on household demographcs, health, educaton, economc actvtes, employment, access to fnancal markets, publc transfers, household expendtures, and assets, and partcularly on shocks and rsks. 1.1 The Drvers of Poverty Transtons Ths study apples an MNL model presented n Wooldrdge (2002). Changes n household poverty statuses over a perod can be classfed nto several mutually exclusve outcomes. The MNL model determnes the probablty that household experences one of the mutually exclusve outcomes j. The probablty s expressed as p j P Y k 1 x j e j J e x k for j 0,1,2,, J, (1) where Y s the outcome experenced by household, βk are the set of coeffcents to be estmated, and x ncludes a household s covarates and ther changes. The model s, 3

4 however, undentfed snce there s more than one soluton for,, 0 J that leads to the same probabltes Y = 0, Y = 1, Y = 2,..., Y = J. To dentfy the model, one of the βj must be set to zero, and all other sets are estmated n relaton to that base category. For convenence, β0 s set to zero; therefore, the above probablty functon can be wrtten as p j P Y x j e j J 1 e k 1 x k, for j = 1, 2,, J and p PY J 1 e k 1 x k (2) From the panel years of 2007, 2008, and 2010, poverty dynamcs can be classfed nto eght categores of (1) beng nonpoor n all perods, (2a) poorpoornonpoor, (2b) poornonpoornonpoor, (3a) nonpoorpoorpoor, (3b) nonpoornonpoorpoor, (4a) nonpoorpoornonpoor, (4b) poornonpoorpoor, and (5) beng poor n all perods. These eght categores can be grouped nto fve mutually exclusve outcomes: (1) J = 4 and P(Y = 0) s the household s probablty of beng nonpoor n all perods, P(Y = 1) s the probablty of rsng (ncludes categores (2a) and (2b)), P(Y =2) s the probablty of fallng (ncludes categores (3a) and (3b)), and P(Y =3) s the probablty of churnng (ncludes categores (4a) and (4b)), and P(Y =4) s the probablty of beng poor n all perods. Thus, the specfc model appled n ths study when standardzng β0 = 0 s expressed as p j P Y x j e j 4 1 e k 1 x k, for j = 1, 2, 3, 4 and p PY kx 1 e k 1 (3) The (MNL) model wll estmate coeffcents for four categores relatve to the omtted category (beng nonpoor n all perods). In order the relatonshp to make more sense, the results of the (MNL) model are used to predct margnal effects, whch measure the condtonal probabltes of a change n the regressors on the outcome and are estmated as p 4 j pj j pk k x k 1. (4) A margnal effect shows the mpact of a change n an explanatory varable on the probablty of a household beng n each of the fve categores. 4

5 To nvestgate the dynamcs of poverty n Vetnam, I hypothesze that a household s poverty transtons depend on the dversty of ncome sources, ts characterstcs, and ts head s characterstcs. The measurement of poverty dynamcs refers to equvalence scaled 2 ncome and a Vetnam natonal poverty lne estmated by the Mnstry of Labour, Invalds and Socal Affars (MOLISA), whch was approxmately $ a day by Smlarly, a household s ncome s used to measure the dversty from agrcultural sources to nonagrcultural sources. Explanatory varables nclude ncome dversty and household asset levels. Income dversty s measured by an ncrease n the share of ncome from nonagrcultural actvtes to the household s total ncome over a perod. A household s asset s measured by household and ndvdual characterstcs as proxes for human captal; land use and asset ndex represent physcal assets. Also, a household s locaton wll be a proxy for the household s access to markets. Household characterstcs are measured by a household s sze that counts the number of the household s member. The household s head characterstcs nclude gender, age, ethncty, and educaton attanment. In addton, a household s physcal assets nclude quanttatve and qualtatve tems. The quanttatve assessment concerns whether the household has: a motorbke, a bke, a televson, a rado, a CD player, an electrc fan, an electrc rce cooker, a frdge, and a mattress. The qualty assessment ncludes: havng mproved floorng condton, havng mproved housng condton, havng access to mproved santaton faclty, and usng mproved cookng fuel. 4 House sze s also ncluded, measured n square meters. These tems are ncluded n the estmaton of the asset ndex va prncpal component analyss. Among the tems, motorbke plays an mportant role (wth a weght of 24 percent), and then comes televson (10 percent), whle the other tems are less mportant, each of whch contrbutes less than 10 percent to the asset ndex (see Table A.1). The locaton of a household ncludes a dummy varable ndcatng provncal locaton. DL s located n the hghlands wth basalt sol, whch s sutable for hgh-value crops such as 2 Equvalence scaled ncome s calculated by dvdng a household s total ncome by ts equvalence sze usng OECD (1982) scale. 3 Ths poverty thresshold s measured n purchasng power party (PPP) prce estmated n Reference categores: The floor s made of cement or ceramc. The man walls are made of concrete and the roof s made of slates or concrete. The household uses flushed tolet. The household cooks wth gas or electrcty. 5

6 coffee, pepper, cashew, and rubber. The populaton densty n the provnce s also low, allowng households to possess more land than households n the other two provnces. In contrast, HT and TTH are n the coastal area and are frequently ht by storms and floods. These dfferences make t reasonable to treat DL as a reference. Income Dversty and Poverty Transtons n Vetnam The Patterns of Income Dversfcaton The man ncome sources of many households were those from agrcultural actvtes. On average, agrcultural actvtes accounted for 40 percent of a household s total ncome; ths share was 50 percent n DL as many households there grew hgh-value crops such as coffee and pepper. Agrcultural actvtes usually ncluded the crop producton of rce, coffee, pepper, and corn, peanut, vegetable, and lvestock producton of cow, pg, and chcken. On average, a household had slghtly more than fve agrcultural actvtes that yelded close to Vetnamese Dong (VND) 10 mllon by 2007,.e., each actvty yelded an average of VND 2 mllon per year. Over the three-year perod, there was an ncrease of 0.5 actvtes and ncome from these actvtes ncreased by closed to 10 percent. Nonagrcultural actvtes were usually off-farm employment, self-employment, nterest from lendng, remttance, and publc transfer, etc. On average, a nonagrcultural actvty yelded closed to VND 18 mllon by Over the three years, there was only a lttle ncrease of 0.1 actvtes, but ncome from ths actvty rose by 22 percent (see Table 1). These mply that the transton to nonagrcultural actvtes affects postvely on a household s ncome; yet, the transton was relatvely slow. The pattern of ncome dversty dffered from one populaton subgroup to another. Households n DL had less agrcultural actvtes than ther counterparts n HT and TTH, but agrcultural actvtes n the former were of hgher value. On average, a household n DL had only less than four agrcultural actvtes and earned closed to VND 17 mllon from coffee, pepper, and cashew crops, whle a household n TTH and HT had more than sx agrcultural actvtes by 2007 and earned only approxmately VND 5 mllon. Nonagrcultural sectors were not developed n HT, makng households there rely more on crops and lvestock to generate ncome. Among 54 ethnc groups n Vetnam, the Knh s the majorty that accounts for nearly 86 percent of the entre populaton. They usually lve n lowlands, whch allows them to 6

7 have better access to markets and publc servces, ergo they beneft more from economc growth. On average, a Knh household had less agrcultural ncome sources and earned more from these than a household of ethnc mnorty groups dd; the earler also made a faster ncrease n the ncome. A Knh household was also more lkely to dverse ts ncome sources to nonagrcultural actvtes. Table 1 Levels and changes n ncome sources by populaton subgroups Populaton subgroups Agrcultural actvtes Nonagrcultural actvtes Number of Total ncome (ml. Number of Total ncome (ml. actvtes VND) actvtes VND) Level 2007 Absolute change Level 2007 Change (%) Level 2007 Absolute change Level 2007 Change (%) Total Head has no schoolng Head attans prmary school Mddle school and beyond Ethnc mnorty groups Knh (majorty) Hà Tĩnh Thừa Thên Huế Đắk Lắk Notes: Agrcultural actvtes nclude crops and lvestock producton. Nonagrcultural actvtes nclude all other ncome-generatng actvtes. Income values refer to prce level n Aprl ml. VND refers to mllon Vetnamese Dong. Source: Author's calculatons from Vulnerablty Surveys n Vetnam. Generally, the results show that households wth better human captal usually have better access to markets. They are, therefore, not only able to have more proftable crops and lvestocks but also more able to shft ther ncome portfolo to nonagrculture sources. In addton, households n TTH and DL have more opportuntes to swtch to nonfarmng ncome sources owng to the better performance of the economy n the two provnces as 7

8 compared wth HT. Moreover, DL households partcularly have advantages n growng hgh-value crops, such as coffee, pepper, and cashew, because of the weather and sol condton. Trends n Poverty and Inequalty The overall poverty headcount rato n Vetnam (as measured by per capta expendture) contnued to decrease from closed to 16 percent n 2006 to 14.5 percent n 2008 (GSO 2011). The poverty headcount ratos n the three provnces (as measured by equvalence scaled ncome) were lower than the ratos (as measured by per capta expendture) found by Le, Nguyen, and Sngh (LNS) (2014) (see Table 2). The three provnces not only made good progress n poverty reducton, but were also successful n keepng the equty of the development as well. The gap between the frst and the ffth ncome quntles ncreased slghtly from 4.8 to 4.8 and 5.2 over the years, respectvely, and the Gn ndex also ncreased only margnally from to and over the perod. Table 2 Poverty headcount rato at natonal poverty lne by provnce, percent Estmated from Vulnerablty Surveys Estmated by LNS Hà Tĩnh Thừa Thên Huế Đắk Lắk Total Source: Author s calculatons from vulnerablty surveys n Vetnam and Le, Nguyen, and Sngh (LNS) (2014). The Patterns of Poverty Transtons Across Groups Over the three-year perod, the majorty of households stayed nonpoor (slghtly over 62 percent) and the other 38 percent was vulnerable to poverty at some level. Ths pattern shows good progress n poverty reducton n whch a large share of the populaton rose up, approxmately 15 percent, and a small share of the populaton fell down at slghtly more than 9 percent. Addtonally, a tenth of the populaton moved around the poverty lne (closed to 11 percent) and only a small share stayed poor n all perods (2.3 percent) (see Table 3). 8

9 The changes n poverty statuses dffer across the subgroups of the populaton. Poverty s usually assocated wth a large-szed famly and a hgher burden of dependency. Nonpoor households tend to have fewer members and a lower dependency rato, 4.1 and 0.4 respectvely, whle those who are poor n at least one perod have nearly fve members and a hgher dependency rato of 0.5. In fact, the poor have low ncomes and low asset levels so they tend to lve together and share ther lmted resources (see Table 3). Table 3 Household and head characterstcs by poverty trajectory, percent Nonpoor Rsng Fallng Churnng Poor Populato n share Total Household sze Dependency rato Head s less than 36 years old Head s 3650 years old Head s 5165 years old Head s 66 years and beyond Head has no schoolng Head attans prmary school Mddle school and beyond Ethnc mnorty groups Knh (majorty) Asset ndex Lowlands Mountanous and hghlands Hà Tĩnh Thừa Thên Huế Đắk Lắk Notes: Populaton shares of the same category sum to 100. Source: Author s calculatons from Vulnerablty Surveys. There s a tendency that young and old households, headed by young or old persons, are more vulnerable to poverty than mddle-aged ones. They are less lkely to stay nonpoor 9

10 and are more lkely to fall nto poverty, fluctuate around the poverty lne, or stay poor. Young households are usually newly formed ones, whch mean that they also have to nvest n bearng and carng for chldren. Older households are usually wealther because they have experence n agrculture and lvestock producton and have accumulated more savngs and assets. However, older heads are assocated wth havng lower sklls and beng less healthy subsequently makng them more vulnerable to poverty, whch s confrmed by the result of a t test. The educaton attanment of a household s head tends to have a negatve relatonshp wth the vulnerablty to poverty. Slghtly hgher than 56 percent of households headed by men or women wthout any school are vulnerable to poverty. The share of vulnerable households decreases to 46 percent and 31 percent across the hgher educaton attanment of the head. In addton, only 10 percent of the heads who are from the Knh are llterate, whle 32 percent of the other heads cannot read or wrte. Moreover, the Knh usually lve n lowlands, whch enables them to have better access to markets and allows them to have a lower rsk of beng poor. The asset ndex s also beleved to be a good proxy for household wealth (see Flmer and Prtchett 2001). It dffers sgnfcantly across populaton subgroups; nonpoor households are agan owners of hgher asset levels, whle stay-poor households are the least, beng 0.54 and 0.24, 5 respectvely. In addton, the locaton of the household can be used as a proxy for publc physcal asset such as nfrastructure and some regonal dfferences. More than half of the households are n mountanous and hghland areas where nfrastructure such as roads, electrcty, schools, and health clncs are n poorer condton and, thus, result n worse market access. Among the chroncally poor households, the majorty of them are n the mountanous areas n HT where nfrastructure s usually of poor condtons, the natural condton s hard for agrcultural producton, and people are usually from ethnc mnorty groups who have less access to markets. 5 The asset ndex s scaled to the range of [0,1] 10

11 Fgure 1 Changes n the patterns of ncome sources by poverty trajectory, percent Source: Author s calculatons from Vulnerablty Surveys. Notes: Agrculture refers to ncome from agrcultural actvtes; nonagrcultural refers to ncome from nonagrcultural actvtes. Fnally, yet mportantly, the patterns of and the changes n ncome sources also dffer across populaton subgroups. On average, a nonpoor household had an ncome level of closed to VND 35 mllon by 2007 and t rose to slghtly hgher than VND 40 mllon by However, ts ncome from agrcultural actvtes stayed at almost the same level of VND 11 mllon over the same perod makng a decrease n the share of agrcultural actvtes n the total ncome. Apparently, households of other groups had a lower ncome level than the nonpoor ones. By 2007, the ncomes of a rsng household, a fallng one, a churnng one, and a poor one were VND 15 mllon, VND 15 mllon, VND 14 mllon, and VND 11 mllon, respectvely. The share of ncome from agrcultural actvtes to the total household ncome of household n these groups are sgnfcantly hgher than ths share for nonpoor households, and nterestngly the share for the former ones rose over the perod, whch mples that ncome from nonagrcultural actvtes of these households ncreased slower than ncome from agrcultural actvtes. Ths was the result of the economc recesson n Vetnam started late 2008, causng loss n job, wage cut, etc. and hence affects negatvely on remttance, and ncome from off-farm employment. Drvers of Poverty Transtons The margnal effects from the MLN regresson show that the ncrease n the share of ncome from nonagrcultural actvtes to total household ncome has an effect on the transtons of poverty. It had a postve relatonshp wth the probablty of rsng out of poverty, negatve relatonshp wth the probablty of fallng nto poverty as well as wth 11

12 the probablty of stayng poor (see Table 4). These relatonshps mply that the transtons toward nonagrcultural ncome sources are good for the household s wellbeng. Yet, not all the margnal effects are of hgh levels of sgnfcance because the dversty of ncome sources s not always successful. Table 4 Margnal effects from multnomal logt model Nonpoor Rsng Fallng Churnng Poor Increase n the share of nonagrculture ncome Increase n the share of nonagrculture ncome *** * (0.0295) (0.0175) (0.0167) (0.0158) ( ) *** *** ** (0.0364) (0.0224) (0.0197) (0.0189) ( ) Household sze *** *** *** *** ** ( ) ( ) ( ) ( ) ( ) Head s male ** ** * (0.0334) (0.0176) (0.0207) (0.0191) ( ) Head age e-06 ( ) ( ) ( ) ( ) (1.84e-05) Head s from the Knh (2007) *** ** (0.0350) (0.0238) (0.0186) (0.0185) ( ) Attans prmary school (0.0368) (0.0199) (0.0239) (0.0161) ( ) Attans mddle school and beyond (0.0347) (0.0205) (0.0189) (0.0179) ( ) Asset ndex 0.125*** 0.345*** 0.192*** 0.265*** * (0.0729) (0.0446) (0.0391) (0.0376) ( ) Hà Tĩnh (0.0300) (0.0195) (0.0149) (0.0164) ( ) Thừa Thên Huế * ** (0.0277) (0.0185) (0.0131) (0.0143) ( ) Notes: Omtted categores: head s female, s from ethnc mnorty groups, has no schoolng, Đắk Lắk, poverty dynamcs are referred to natonal poverty lne. All control varables refer to the base year level (2007). Nonagrculture ncome refers to nonagrcultural ncome. Pseudo R 2 = 0.226, Observatons= 1,858. The MNL regresson passes tests of IIA assumpton. Standard errors n parentheses. *** p < 0.01, ** p < 0.05, * p <

13 Households n Vetnam have a tendency to have smaller szes owng to the lower brth rate, the ncreasng mgraton, and the nclnaton of lvng n two generaton households. Nevertheless, poor households usually have a larger sze because they have more chldren, less chances to mgrate, and havng lmted resources that prevent them from separatng nto smaller households. The margnal effects show that households of a larger sze and hgher dependency rato have a lower probablty of stayng nonpoor and a hgher probablty of beng poor n at least one perod (see Table 4). Female-headed households are typcally home to more members than male-headed households, and the formers usually have less bread wnners than the later, whch consequently makes them have hgher levels of vulnerablty to poverty than ther counterparts. In addton, there was no evdence of the dfference between the vulnerablty to poverty across household sungroups as classfed by the head s age. In fact, the poverty dynamcs are determned more by the change n the household s lvelhood rather than the characterstcs of the head (Carter and Barrett 2006), and fndngs from prevous studes do not show the mportance of the head s age (Kedr and McKay 2005; Bhde and Mehta 2005). In addton, the educaton attanment of the household head does not contrbute to the dfferences n the probablty of beng n one or another poverty trajectory. In fact, the more the head s educated, the better hs/her access to producton resources, labour, and output markets s, and the more effcent he/she s n managng household resources. However, ths type of human captal s more lkely to have a long-term effect on a household s well-beng rather than on the change n shorter perod of tme. As dscussed earler, the Knh are usually more able to access to market and, hence, take advantage of publc servce and the development process, whch allows them to have a hgher probablty of beng nonpoor, and lower probabltes of beng poor n one or more perods than ther households of ethnc mnorty groups (see Table 4). Household wealth as measured by the asset ndex shows qute strong effects on poverty dynamcs. It prevents households from beng poor and s negatvely correlated wth fallng nto poverty, churnng around poverty lne, or beng poor. It s also postvely correlated wth stayng nonpoor and rsng out of poverty (see Table 4). These fndngs are n lne wth the dscusson of the role of assets n the poverty transtons (Carter and Barrett 2006) as well as wth emprcal fndngs from Bhde and Mehta (2005), and Ima, Gaha, and Kang (2011). 13

14 There was lttle evdence of the dfference among households n the three provnces n the vulnerablty to poverty. TTH s more dynamc n terms of economc actvtes owng to the development of the toursm sector, and the convenence of transportaton. Therefore, households n the provnce have a hgher probablty of movng out of poverty and a lower probablty of fallng nto poverty than ther counterparts n the other two provnces (see Table 4). Robustness Check In order to check the robustness of the MNL model for poverty dynamcs, the study apples to the transtons of poverty as referred to the poverty lne of $2.5 a day (see Table A.2). The MNL regresson, the results of whch are shown n Table 4, and Table A.2 pass the Hausman tests or suest tests of ndependence of rrelevant alternatves (IIA), whch means that assumptons of IIA could not be rejected; hence, estmates from MNL models are effcent. The reference model, n general, shows smlar effects to those n the basc one. However, there are dfferences n the sze of the effects n these models compared to the basc model because poverty dynamcs n the addtonal model refer to a hgher poverty lne. Addtonally, the results from Table 4 are n lne wth those from prevous studes. The results from the MNL regresson n ths study are, therefore, realzable. Concluson Ths study uses panel data on households from regons n Vetnam and a multnomal logt model to estmate drvers of poverty transtons. The results show a large share of the populaton s vulnerable to poverty where 38 percent of households have a rsk of beng ether transent or chroncally poor. Ths rsk vares substantally across household groups; households of a large sze, ethnc mnorty groups, and has lmted physcal assets have a hgher rsk of beng poor snce they typcally have less access to markets than the other groups, whch consequently prevents them from greatly beneftng from the economc growth. These fndngs are n lne wth most prevous studes by Carter and May (1999), Glewwe, Gragnolat, and Zaman (2000), and Woolard and Klasen (2005). A household wth better access to producton nputs and markets s more able to dversfy ts ncome sources to nonagrcultural actvtes. The transton toward nonagrcultural ncome sources usually results n hgher ncome and, therefore, an mprovement n a 14

15 household s well beng. Partcularly, the dversty to nonagrcultural ncome sources s helpful for lftng a household out of poverty, but t s not very helpful for those who churn around the poverty lne and the chroncally poor. Addtonally, nonagrcultural actvtes are subject to shocks and rsks, partcularly from macroeconomc condtons and market performance. Ths mples that household developng countres stll face challenges n the early stages of ncome dversty. References Bhde, Shashanka, and Aasha K. Mehta Trackng Poverty Through Panel Data: Rural Poverty n Inda CPRC Workng Paper 28. Chronc Poverty Research Centre, Unversty of Manchester. Cappellar, Lorenzo, and Stephen P. Jenkns Modelng Low Income Transtons. Journal of Appled Econometrcs 19: do: /jae.778. Carter, Mchael R., and Chrstopher B. Barrett The Economcs of Poverty Traps and Persstent Poverty: An Asset-Based Approach. Journal of Development Studes 42: do: / Carter, Mchael R., and Julan May Poverty, Lvelhood and Class n Rural South Afrca. World Development 27: do: /S X(98) De Janvry, A., M. Fafchamps, and E. Sadoulet Peasant Household Behavor wth Mssng Markets: Some Paradoxes Explaned. Economc Journal 101: Flmer, Deon, and Lant H. Prtchett Estmatng Wealth Effects Wthout Expendture Data or Tears: An Applcaton to Educatonal Enrollments n States of Inda. Demography 38: General Statstcs Offce (GSO) Result of the Vetnam Lvng Standard Survey Hano: Statstcal Publshng House. Glewwe, Paul, Mchele Gragnolat, and Hassan Zaman Who Ganed from Vetnam's Boom n the 1990s?. Polcy Research Workng Paper 2275, The World Bank. Ima, Katsush S., Raghav Gaha, and Woojn Kang Vulnerablty and Poverty Dynamcs n Vetnam. Appled Economcs 43: do: /

16 Justno, Patrca, Jule Ltchfeld, and Hung Tha Pham Poverty Dynamcs durng Trade Reform Evdence from Rural Vetnam. Revew of Income and Wealth 54: do: /j x. Kedr, Abb M., and Andrew McKay Chronc Poverty n Urban Ethopa: Panel Data Evdence. Internatonal Plannng Studes 10: do: / Knsey, B., K. Burger, and J.W. Gunnng Copng wth Drought n Zmbabwe: Survey Evdence on Responses of Rural Households to Rsks. World Development 26 (1): Klasen, Stephan, and Hermann Wabel, eds Vulnerablty to Poverty: Theory, Measurement and Determnants, wth Case Studes from Thaland and Vetnam. Basngstoke: Palgrave Macmllan. Le, M. Son, Duc Tho Nguyen, and Tarlok Sngh Economc Growth and Poverty n Vetnam: Evdence from Elastcty Approach. Dscusson Paper , Grffth Busness School. McCulloch, Nel, and Bob Baulch Dshtngushng the Chroncally from the Transtory Poor: Evdence from Pakstan. IDS Workng Paper 97. Insttute of Development Studes, Unversty of Sussex. Organsaton for Economc Co-operaton and Development (OECD) The OECD Lst of Socal Indcators. Pars: Organsaton for Economc Co-operaton and Development. Tran, Q. Van, Sabna Alkre, and Stephan Klasen Statc and Dynamc Dspartes between Monetary and Multdmensonal Poverty Measurement: Evdence from Vetnam. Research on Economc Inequalty 23: /S Woolard, Ingrd, and Stephan Klasen Determnants of Income Moblty and Household Poverty Dynamcs n South Afrca. The Journal of Development Studes 41: Wooldrdge, Jeffrey M Econometrc Analyss of Cross Secton and Panel Data. London: The MIT Press. World Bank. 2013, January. Data: Indcators. 16

17 1.2 Appendx Table A.1 Components of asset ndex and ther weghts Assets Egenvalue Proporton Household has a motobke Household has a televson Household has an electrc rce cooker Household has a mattress Household has a vdeo player Household cooks wth electrcty/gas Household uses mproved santaton faclty Household has an electrc fan Household has a frdge Household has mproved floorng House sze House (wall and roof) s made of mproved materals Household has rado Household has a bke Note: Proportons sum to one. 17

18 Table A.2 Margnal effects from MNL for poverty dynamcs as referred to $2.5 Nonpoor Rsng Fallng Churnng Poor Increase n the share of nonagrcultural ncome Increase n the share of nonagrcultural ncome (0.0418) (0.0281) (0.0241) (0.0250) ( ) ** ** (0.0529) (0.0353) (0.0296) (0.0311) ( ) Household sze *** *** ** * *** (0.0137) ( ) ( ) ( ) ( ) Head s male ** * (0.0439) (0.0245) (0.0276) (0.0273) ( ) Head age ** *** ** ( ) ( ) ( ) ( ) ( ) Head s from the Knh (2007) 0.187*** *** *** (0.0434) (0.0359) (0.0257) (0.0264) (0.0175) Attans prmary school (0.0520) (0.0327) (0.0274) (0.0249) ( ) Attans mddle school 0.124** ** and beyond (0.0516) (0.0338) (0.0286) (0.0309) ( ) Asset ndex 0.509*** 0.495*** *** *** *** (0.113) (0.0719) (0.0592) (0.0617) (0.0273) Hà Tĩnh *** (0.0415) (0.0296) (0.0216) (0.0270) (0.0149) Thừa Thên Huế ** *** (0.0378) (0.0254) (0.0190) (0.0229) ( ) Notes: Omtted categores: head s female, head s from ethnc mnorty groups, head has no schoolng, Đắk Lắk, poverty dynamcs are referred to natonal poverty lne. All control varables refer to the base year level (2007). Pseudo R2 = 0.257, Observatons= 1,858. The MNL regresson passes tests of IIA assumpton. Standard errors n parentheses. *** p<0.01, ** p<0.05, * p<0.1 18