Panchayats and Household Vulnerability in Rural India*

Transcription

1 Panchayats and Household Vulnerablty n Rural Inda* Raghbendra Jha, Har K. Nagarajan, Woojn Kang & Kalash C. Pradhan Address all correspondence to: Prof. Raghbendra Jha, Arndt-Corden Department of Economcs, College of Asa and the Pacfc, H C Coombs Buldng (09), Australan Natonal Unversty, Acton, ACT, 2601 Emal: r.jha@anu.edu.au; Phone: Fax: *Jha ANU; Nagarajan IRMA; Kang KDI; Pradhan NCAER Australa South Asa Research Centre,

2 I. Introducton Inda, partcularly rural Inda, has been home to entrenched poverty for long. It s, therefore, mperatve to understand people s vulnerablty to poverty. Further, snce the Government of Inda has nvested consderable human and fnancal resources n Gram Panchayats t s mportant to understand to what extent these Panchayats have been successful n redressng vulnerablty, especally snce we establsh that 1 ex-ante vulnerablty has a sgnfcant effect on ex post poverty dynamcs. We then dentfy the components of vulnerablty to better focus polcy. Fnally the mpact of copng strateges on household welfare s assessed. These are the broad themes of ths paper. We focus on sx questons. a) Does gender of the elected representatves matter; specfcally do poltcal reservatons help n reducng vulnerablty? b) What wll be the mpact of regme changes on resource re-allocaton and on household consumpton? Do all types of regme changes (based on Jat or gender) have smlar mpacts on household vulnerablty? c) Do households fall nto poverty traps,.e. experence chronc poverty, due to neffcences assocated wth governance? d) Has local governance contrbuted to households undertakng less rsky copng strateges n response to adverse shocks? (e) Durng perods of dstress, are local governments able to provde nsurance through welfare programs? (f) After experencng adverse shocks, s the nsurance that Panchayats provde suffcent to elmnate the need for households to cut ther food consumpton to less than two meals, and to avod sellng of productve assets, and thereby reduce both vulnerablty and poverty? Households experence two types of shocks covarate (e.g. natural dsasters, pest attacks on crops n the vllage) and, dosyncratc (e.g. llness, job-loss). Both types of rsk could render even non-poor members of vllages vulnerable, partcularly when household level copng mechansms are ether neffectve and/or constraned: e.g., f a preferred copng strategy of households s accessng welfare programs and partcpaton s restrcted or prevented ether due to mproper targetng or because of program capture. Hence copng mechansms are constraned by neffcences assocated wth the Panchayat Raj nsttutons (PRI) or other nsttutons. If household ncome (consumpton) s sgnfcantly covarate then nformal sources of nsurance such as famly networks are lkely to become neffectve and f these are the only sources of nsurance, then such households are prone to become vulnerable due to covarate shocks. By dentfyng vulnerablty as a sum of underlyng poverty, covarate or aggregate rsks, dosyncratc rsks and unexplaned rsks, we can prescrbe polces that wll affect household ncomes, and reduce such rsks. We wsh to examne whether n response to covarate shocks households tend to sell assets, 1 In Inda the head-count rato of poverty refers to a state n whch consumpton expendture s less than some acceptable norm. Hence, poverty s an ex post concept. 2

3 or wthdraw chldren from school. Do they attempt access the centrally sponsored welfare programs? Answers to such questons are mportant n the context of the PRI framework for such nsttutons have been desgned to mprove access by households to welfare programs and hence not resort to second best copng mechansms that could have long term adverse consequences. Wth respect to dosyncratc rsks t s pertnent to ask whether households are unable to access credt, health facltes, or apply for a scheme lke the MGNREGS n whose admnstraton the panchayats have a strong role. Copng strateges and ther mpact on economc welfare of Households We examne the entre menu of copng strateges adopted by households, rrespectve of whether such strateges were adopted as a reacton to dosyncratc or covarate rsks. The average probablty of any one of the copng strateges adopted n response to an average shock s examned. We predct the probablty of such copng strateges after controllng for varous household, vllage and governance varables. Snce copng strateges could be endogenous to consumpton, we use predcted values to explan consumpton. We compute dfference between household consumpton and certanty equvalent consumpton (whch quantfes exposure to rsk) and estmate the ablty of copng strateges to reduce ths gap. We expect that wth the progress n the reforms assocated wth PRI, households wll resort less to strateges such as reducng consumpton, not sendng chldren to schools, or sellng assets and land. II. Lterature We now brefly revew some of the papers relevant for the analyss of ex-ante vulnerablty defned as Vulnerablty as Expected Poverty (VEP e.g. Chaudhur et al., 2002; Prtchett et al., 2000), and s defned n secton 4. For the precse assessment of vulnerablty, one needs to use panel data whch s seldom avalable n developng countres. A VEP measure has the advantage that t serves as a measure of expected poverty. Further, f the expected poor actually turn out to be poor n a subsequent tme perod they can be dentfed as chroncally poor. Thus, we explore the lnkage between chronc poverty and vulnerablty. Our approach s a generalzaton of the lterature (e.g. Barrentos, 2007; CPRC 2008; McCulloch and Calandrno, 2003). 2 Furthermore, the lnk between chronc poverty and vulnerablty has rarely been emprcally explored. Several papers examne Vulnerablty as Expected Utlty (VEU) as arsng both from exposures to and management of rsks (Chambers, 1989; Alwang et al., 2001; Hetzmann et al., 2002; Barrentos, 2007). Based on Chambers dstncton between external (manly referrng to rsks or shocks) and nternal (referrng to the lack of means to manage rsks) vulnerablty, Alwang et al (2001) and 2 Two broad methods of chronc poverty n lterature are () the spells approach focusng on poverty transton (Gaha and Deolalkar, 1993) and () the component approach dstngushng permanent component of ncome from fluctuatng component (Ravallon, 1988; Jalan and Ravallon, 2001) 3

4 Hetzmann et al (2002) use the noton of a rsk chan n that those two components lead to an outcome gven by a welfare loss. There s also an extensve lterature on measurng household vulnerablty n developng countres (e.g. Kochar, 1999; Dercon and Krshnan, 2000; Chaudhur et al., 2002; McCulloch and Calandrno, 2003; Lgon and Schechter, 2003; Gaha and Ima, 2009; Jha et al., 2010; Kurosak, 2010). However, these do not address vulnerablty, rsk sharng and copng strateges n an ntegrated manner. As represented by the rsk chan hypothess, each element s related wth the other elements. Therefore, to systemcally analyze household welfare under uncertanty, one needs to nvestgate not only household vulnerablty, but also the household s rsk copng strategy and the effect of the strategy on household consumpton. The lterature on copng strateges has two characterstcs vz., a) t represents countres that are prone to shocks, b) all these countres have attempted to decentralze and devolve powers to local governments. Cameron and Worswck (2003) and Gabrella and Francesca (2009) usng Indonesan data have found that that non-poor farmers are able to smooth consumpton whle nonfarm, and poor households try to fnd alternate employment durng perods of dstress n order to smooth consumpton and also resort to dssavng. Rosenzweg and Bnswanger, (1993) fnd for vllages n sem-ard Inda that better-off farmers can have proft maxmzng portfolos even n the face of covarate rsk, because they are able to self-nsure and rely on socal networks, whereas poor farmers can t. Rosenzweg and Wolpn (1993) show for the same vllages that savng n the form of bullock pars and subsequent sale s one of the man self-nsurance mechansms. Tongruksawattana et al. (2010) use data from vllages n Northern Thaland and show that the domnant copng strateges durng dstress perods nclude use of remttances, nformal borrowng, use of savngs, and sale of assets. They also fnd that the ablty of households to access publcly sponsored programs s lmted. Castellanos and Rahut (2006) found for rural Bolva that 48% of households attempt to work more or ncrease ther workng days as a copng mechansm aganst harvest falures; 38% spend savngs and engage n barter. Also 61% of respondents from the poorest three quntles of expendture dstrbuton ndcated that they spend savngs durng crses. Dercon (2002) shows that the type of shocks matter whle tryng to understand the ablty of households to cope wth these dstress events. Usng data from Ethopa, he has shown that most households use nformal borrowng to cope wth covarate shocks whch renders them more vulnerable than before. Okamoto (2011) fnds that n Myanmar households that are able to antcpate dstress events tend to save n knd (n partcular n the form of gold and cattle). These are then sold durng dstress events. Wth nsuffcent accumulated assets households resort to nformal borrowng. For Bangladesh, Rashd et al. (2006) have found that the nature of copng strateges reveal nformaton about household characterstcs such as dversty and stablty of ncomes, asset holdngs, and educaton levels. 4

5 III. Data and Descrptve Analyss of Rsk and Copng Strateges Our analyss s based on household and vllage level data from the NCAER REDS data from the 1999 and 2006 rounds. The number of sample households n 1999 and 2006 surveys s 7474 and 8659 respectvely, of whch 5885 households were ntervewed n both rounds. We use data on these 5885 households. We use data on per capta consumpton expendture of the household, age and gender of the household head, dependency rato, land owned, percentage of land owned that s rrgated, predcted household splts, share of household members who attended Gram Sabha (GS) meetngs and, share of members who voted based on dentty. At the vllage level we use nformaton on whether the gender of the Pradhan has changed, whether the households n the vllage have access () to school, () transport () publc hosptal, (v) publc fnancal nsttuton, and, (v) publc tap for drnkng water. The followng shock varables are used: (Covarate) whether the household had been affected by drought, floods, epdemcs or anmal epdemcs; (Idosyncratc) whether the household had been affected by death of member of household, sudden health problems or crop falures. The followng copng strateges that households use are predcted usng a number of household and vllage level controls: savng; government programs; transfers from frends and relatves; depleton of physcal and/or human captal; reduced consumpton and altered crop choce. These varables are then used for analyzng VEU. Certanty equvalent consumpton (CEC) s computed usng Aufret (2000). Ths represents consumpton n a rsk free state. The dfference between actual consumpton and the CEC measures the exposure to rsk. Ths dfference s regressed on the predcted copng strateges wth varous varables assocated wth governance. The profle of covarate and dosyncratc shocks n the vllages s reported n Table 1. All types of shocks occur wth regularty each year. 52% of all households have experenced the effects of covarate shocks or have been affected by dosyncratc shocks durng (the year of the communty survey). Prce fluctuatons were hgh durng (approxmately two Panchayat perods n most vllages) mplyng sgnfcant seasonal varatons n consumpton expendtures. Although the cumulatve losses are Rs 5194, the latest epsode of general prce ncreases accounts for 25% of these losses suggestng that the prce ncreases have become more pronounced over tme. Strateges for consumpton smoothng therefore are mportant f households are to avod poverty traps. Covarate shocks have a relatve hgh ncdence. The losses due to such shocks for the households are large. Ths suggests that local government must put n place mechansms whereby households do not report to rsky copng strateges such as sellng assets or reducng savngs. 5

6 Cross country evdence on copng mechansms provded here also ponts to the fact that local governments have played a mnmal or no role durng perods of dstress. Tables 2, 3, and.4 provde nsghts nto the copng strateges adopted durng the prevous Panchayat perod by households. Table 2 deals wth a frst shock. It shows that up to 60% of all households used ther own savng durng dstress perods whle only 9% accessed any government programs. Access to welfare programs s hgher for covarate shocks than for dosyncratc shocks. Up to 6% of all households resorted to reducng meals (not havng two meals per day) Thus dssavng s the most mportant copng mechansms, followed by access to government assstance, wth reductons n number of meals the thrd mportant mechansms. Asset sales are uncommon wth frst shocks, wth only 2.4% of households resortng to them. Wth repeat shocks n the next year, the role of the local governments s even less, wth only 8% of all households havng been able to access programs through Panchayats, and more so for covarate shocks than dosyncratc shocks (table 3). For repeat shocks, only 4% of all households are able to use savngs (ther savngs havng been depleted by the prevous epsode), and nstead 15% of households resort to reducng meals, whle 23% attempt borrow and a full 19% have to sell assets. When shocks occurred after a gap of 5 years, the relance on Panchayats for dstress allevaton goes up to 12%, agan mostly for covarant shocks. These suggest that Panchayats have on average not been a major source of support n response to dstress events. Repeat events cannot be managed by households easly wthout resortng to asset sales and reducng the number of meals, suggestng that they are not adequately self-nsured. However, when shocks are repeated fve years later, the use of savngs rebounds to 52% households, whle sale of assets and reducng meals drop back to 8% and 2.3% respectvely, ndcatng sgnfcant recovery of copng ablty of the households. Thus, household copng strateges are constraned, especally when shocks repeat wthn a short perod of tme. For the prvate responses the nabltes are lkely to be assocated wth ncomplete credt and labor markets. The lmted ablty of Panchayats to help could be attrbuted to lack of resources, or neffcences of the Panchayats terms of targetng, plannng ahead for dstress events, or n terms of pathologes such as elte and program capture. Our dscusson of VEU noted that vulnerablty s enhanced by such constrants. Lmted ablty to cushon especally aganst rapdly repeatng shocks would make t more lkely that households wll fall nto poverty traps after repeat shocks. Tables 1, 2, 3 and 4 here. IV. Methodology Hoddnott and Qusumbng (2003) use three operatonal defntons of vulnerablty: () VEP, () VEU and () vulnerablty as unnsured exposure to rsk (VER). We focus on the frst two as ths wll 6

7 allow us to dscuss polcy germane to the determnants of these two types of vulnerablty as well as artculate the mpact of the local governments on reducng vulnerablty and poverty. Vulnerablty measures Ex-ante vulnerablty s defned as the probablty that a currently non-poor household wll fall below the poverty-lne, or a currently poor household wll reman n poverty n the near future. We show that computed vulnerablty can be a relable predctor of future poverty. Specfcally, we () quantfy household vulnerablty n rural Inda durng two tme perods usng the 1999 and the 2006 rounds of the REDS data of NCAER, () nvestgate the determnants of ex post poverty as well as ex-ante vulnerablty, () assess the role of ex-ante vulnerablty on poverty shft durng the sample perods (.e. movement nto/out of poverty) and fnally, (v) examne how the effects of the determnants of vulnerablty vary at dfferent ponts across the vulnerablty dstrbuton. We analyze whether greater partcpaton n GS leads to hgher household level consumpton. By partcpatng n GS households can nfluence fnancal allocatons, hold elected offcals to account and access nformaton on welfare programs. Partcpaton n governance could affect both current and the expected growth n consumpton. Electoral poltcs do much to allocate resources and nfluence consumpton and growth n consumpton. If parochal poltcs help, as Munsh and Rosenzweg (2008) suggest they do, then that could have postve effects on the vllages. They show that even though the Pradhan elected n ths manner reflects the preferences of a specfc group or allance, the spllover effects from ths has a postve effect on all households. Further, Dennger et al (2013) show that poltcal reservatons for women can nfluence certan key determnants of the qualty of governance, e.g., mprovements n the adherence to rules and targetng of centrally sponsored welfare programs both of whch can affect household level consumpton. VEU s defned as the dfference between the utlty derved from some level of certanty-equvalent consumpton (.e., consumpton n a rsk free state) and the expected utlty derved from consumpton. Households ncome and consumpton fluctuate due to exposure to shocks. To protect themselves from these shocks, households attempt to manage rsks ex-ante(e.g. dversfyng employment, crop dversfcaton, etc.) or cope wth the consequences of the rsks ex post (e.g. precautonary savngs, usng transfer ncomes, enterng the labor market, etc.). Exposure to shocks n tself need not render a household vulnerable. A combnaton of exposure to rsks and ncomplete or lack of credt, factor, and nsurance markets and other rsk dffuson mechansms make such households vulnerable. Addtonally, the uncertanty assocated wth antcpatng the tmng of rsks could lead to households undertakng copng strateges that are not optmal. For example nadequate forecast of monsoons 7

8 could lead to farmers postponng ther nursery operatons for rce cultvaton wth serous consequences for yeld. Consumpton Usng consstent household data for both tme perods we estmate the per capta consumpton for the panel households. Wth cross-sectonal data, a household consumpton functon s: parameters. ln c = β + ε (1) X Where, ε ( 0, X ϑ) ~ (2) ndexes households, X represents control varables and β and ϑ are vectors of estmable Assumng that the structure of the economy s relatvely stable over tme, future consumpton stems solely from the uncertanty about the dosyncratc shocks and unobservable characterstcs, captured byε, whch contrbute to dfferent per capta consumpton levels. Varance of the dsturbance s gven as: 2 σ ε, = X ϑ (3) β and ϑ can be estmated usng a three-step feasble generalzed least squares (FGLS) procedure.. 3 Usng βˆ andϑˆ, we can estmate the expected log consumpton and the varance of log consumpton for each household as follows: E ˆ[ln c X ] = X βˆ (4) V ˆ[ln c X ] = X ϑˆ (5) By assumng ln c s normally dstrbuted and usng the estmates above, the probablty of fallng nto (for the currently non-poor), or remanng (for the currently poor), poverty n the future s: ˆ ln z X ˆ β V = Pˆr (6) ( ) lnc < = Φ ln z X ˆ X ϑ The rght hand sde s the measure of VEP and reflects the presumpton that hgh volatlty of consumpton reduces vulnerablty for those wth expected consumpton below poverty lne whereas t 3 See Chaudhur et al. (2002) and Chaudhur (2003) for techncal detals 8

9 ncreases vulnerablty for those whose expected consumpton s above poverty lne. Hence, f we reasonably assume that the poor are rsk-averse (or credt constraned or both), they mght have lttle chance to escape from poverty. Probablty of beng poor We use a Probt model to estmate whether, condtoned on a vector of household and vllage characterstcs ( poverty lne, X ). a household s monthly per capta consumpton expendture was below the ( X ψ ) Pr ( P = 1) =Φ (7) where P = 1 f lnc < lnz and P = 0 otherwse. When 2006 data s used we address the assocaton between households vulnerablty n 1999 and the probablty of beng poor n 2006 by addng VEP n 1999 as a covarate.. 4 Key hypotheses relatng to poverty transtons tested are () whether the vulnerable poor n 1999 were more lkely to stay n poverty n 2006 (.e. poverty traps) and () whether vulnerablty was lkely to ncrease the lkelhood that the non-poor n 1999 slppng nto poverty n We consder the followng 4 unordered categores of poverty transton. P 1= those who were poor n both 1999 and 2006; P 2 = those who were poor n 1999, but non-poor n 2006; P 3 = those who were non-poor n 1999, but poor n 2006; P 0 =those who were non-poor n both 1999 and Ths s the reference case. The multnomal logt model s wrtten as e Pr( P = j) = 3 1+ ( X λ + τ VEP ˆ ) k= 1 Pr( P = 0) = 3 1+ k= 1 e e j k ( X λ + τ VEP ˆ ) 1 k ( X λ + τ VEP ˆ ) k k k,, j = 1, j = 0 2, 3 (8) (9) We need to capture the dfferent assocaton of factors on the households vulnerablty whch mght vary dependng on the poston of the household n overall dstrbuton of estmated VEP. Hence, we use quantle regresson technques ( Koenker and Basset, 1978). VEP ˆ = X µ τ + ε Wth τ Q τ ( VEP ˆ X ) = Xµ (10) τ 4 See also Ima, Gaha and Kang (2011). 9

10 For any 0 < τ < 1, the coeffcent µ τ of theτ th regresson quantle s estmated as a soluton to: mn τ y xµ τ + ( 1 τ ) y x µ τ. (11) µ τ y xµ τ y < xµ τ In contrast to OLS where the parameters are estmated at the condtonal sample mean of the dependent varable, the quantle regresson enables determnaton of whether a household s locaton n the estmated vulnerablty dstrbuton dfferentally affects the relatonshp between VEP and household characterstcs. We estmate the model for the 10 th, 25 th, 50 th (medan), 75 th and 90 th percentles. The quantle regresson approach has several advantages. Snce t s estmated by mnmzng the sum of (weghted) absolute values of the resduals as shown n (11), t s robust to outlers or asymmetrc error as well as the presence of heteroskedastcty (Deaton, 1997; Koenker & Hallock, 2001). Furthermore, by usng the entre sample to estmate each quantle, t avods sample selecton bas arsng from OLS approach dvdng the vulnerablty dstrbuton nto several subsets (Hammarstedt & Shukur, 2007). VEU VEU s defned as the dfference between the utlty derved from some level of certanty-equvalent consumpton and the expected utlty derved from consumpton. Consder two possble states. In the frst state, a rsk-averse household s certan that ts expected consumpton n the next perod wll be just below the poverty lne. Therefore, the probablty of poverty (.e. vulnerablty) n that state s one. In the second state, whle ts mean expected consumpton remans unchanged, there s probablty of 0.5 that the household wll have consumpton just above the poverty lne (and above the mean) and probablty of 0.5 that the household s consumpton wll be below the mean. Beng rsk averse, the household would prefer the frst state wth the certan consumpton to the second state wth the expected consumpton despte lower vulnerablty n the second state. Ths perverse feature of VEP s overcome by VEU measure whch can be wrtten as: V = U z) EU ( c ) (12) ( t Where z s certanty-equvalent consumpton at and above whch a household would not be consdered vulnerable, analogous to the poverty lne where, ncreasng functon. Eq. (12) can be further decomposed as follows: U s a weakly concave and, strctly V = U z) U ( Ec ) (Poverty) ( t + { U Ec ) EU [ E( c X )] } (Covarate or aggregate rsk) (13) ( t t t 10

11 + { EU E( c X )] EU [ E( c X, X )] } (Idosyncratc rsk) [ t t t t t + { EU E( c X, X )] EU ( c ) } (Unexplaned rsk & Measurement error) [ t t t t As n Lgon and Schechter (2003), the followng form of utlty functon s assumed: 1 κ c U ( c) = (14) 1 κ Where κ denotes the household coeffcent on relatve rsk averson and s set to 2. We normalze the expendture and per capta ncome so that the average expendture and per capta ncome over all households would be unty and set the poverty lne to be one by choosng z to equal average consumpton. 5 Therefore, household vulnerablty wll be zero f resources are allocated so that households receve the expected consumpton expendture wth certanty (Lgon and Schechter, 2003). E ( c t X t ) s the aggregate rsk component and E ( ct X t, X t ) the dosyncratc rsk component are estmated by: E( c t X t ) = α + ηt (15) E( c X, X ) = α + η + βx (16) t t t t t In (16), f ncome s used as the explanatory varable for consumpton as n our case, t mght be endogenous (Gaha and Ima, 2009). Hence, we use nstrumental varable (IV) estmaton for (16) n whch ncome s endogenous. Household responses to shocks To nvestgate households response to shocks, a multvarate Probt model s appled because households often choose dfferent copng nstrument at the same tme n the face of rsks and the model allows correlaton between such choces. The latent decson varable for copng nstruments characterstcs, X m and an error termε m. * Rm s assumed to be a functon of household 5 Household consumpton expendture and ncome n 2006 data were adjusted at 1999 value usng state wse CPI for agrculture and rural labourers. 11

12 R * m = ϑ X + ε m m m (17) The error term ε m has a multvarate normal dstrbuton, each wth a zero mean and varancecovarance matrxv, where V has values of 1 on the prncpal dagonal and correlatons ρ = ρ as off-dagonal elements (Cappellar and Jenkns, 2003). * R = 1 If R > 0, and 0 otherwse (18) m m mn nm Snce a household s decson on copng strateges reles not only on household characterstcs but also the nature of the rsks faced, a vector of shock varables (dummy) s ncluded for both covarate and dosyncratc rsk. The estmaton s carred out on the 2006 data only because of lack of nformaton on copng mechansms n Household response to shocks nclude () Savng, () partcpatng n government program, () Borrowng/Transfer, (v) captal depleton and, (v) other responses. Fnally, we estmate the mpact of copng strateges on the economc welfare of households. We estmate the growth of household consumpton expendture, certanty equvalent consumpton and dfference between these two as mpacted by the varous copng strateges. A two stage IV strategy s used. In the frst stage varous copng strateges e.g. savngs, help from government, wage employment, borrowng from relatves, technologcal mprovement, sellng assets, starvaton and borrowng from formal and nformal sources are regressed on measures of the number of dosyncratc and covarate shocks and a number of household, vllage and vllage level revenue and expendture characterstcs. In the second stage the predcted values of these copng strateges are used as explanatory varables along wth other household and vllage characterstcs to explan consumpton growth, certanty equvalent consumpton growth and the dfference between the two. V. Results The results of the estmaton (wth vllage fxed effects) of ex-ante per capta consumpton, (4) and (5), show: Households wth older heads have lower ex-ante per capta consumpton, however ths effect becomes nsgnfcant n When the share of members wth more than prmary educaton rses, expected consumpton s hgher. Coeffcent estmates are strongly sgnfcant n both years. In both years the coeffcents of the share of female members n the household are postve and sgnfcant and of ts square negatve and sgnfcant. Whle per capta consumpton wth more female members s postve, the varance nduced by more number of female members on household consumpton s also larger. confrmng that havng more female members substantally decreased per capta consumpton, but the relatonshp was nonlnear. Havng more land ncreased per capta consumpton but margnal change gets smaller wth land sze. Hgher share of rrgated land tended to ncrease household consumpton n both years. 12

13 Varables assocated wth votng, partcpaton n the process of governance, as well as regme changes (male to female through electons) are sgnfcant n explanng per capta consumpton. Greater partcpaton n GS and IBV ncrease per capta consumpton n both perods. IBV sgnfcantly nfluences the process of governance, as well as provdng access to welfare programs. Per capta consumpton expendtures grew as a consequence of IBV as dd expected consumpton. Both regme changes as well as poltcal reservatons had a postve mpact, but only n Estmates of consumpton and varance of dsturbance term from ths estmaton, allow us to calculate each household s vulnerablty usng (6). As the result s senstve to the choce of poverty lne, we appled 100%, 120% and 80% of the poverty lne of each year. For 100% poverty lne we obtan mean vulnerablty of 28.3% n 1999 and 16.5% n 2006,.e. a rural Indan household, on average, had 28.3% probablty of fallng nto poverty n 1999 but ths declned to 16.5% n Select results of estmatng (7) are shown n Table 5. The respectve margnal effects reveal that exante vulnerablty n 1999 translated nto ex post poverty n 2006,.e. 1% ncrease n the ex-ante probablty of fallng nto poverty would ncrease ex post probablty of poverty by 0.32%. Regardless of the poverty lne used, the coeffcents of vulnerablty are strongly sgnfcant n all cases. Female headed households had relatvely lower probablty of poverty n 2006 than male-headed households. The coeffcents of age of household head are postve (.e. ncrease probablty of poverty) and sgnfcant n 1999 but not n Hgher share of more educated members, larger sze of land and more rrgated land tend to reduce household probablty of poverty Dennger et al (2012) pont out that women nhertng land has postve effects on some of the consttuents of expendture. Our results of the estmaton of (4) and (5) shows that land nhertance had a postve mpact on expected per capta consumpton of the household (by as much as 4%), as well as reducng poverty whatever be the poverty lne chosen (probablty of the household beng poor n 2006 reduces by 3.2%). But nhertance may not have an nstantaneous effect. Thus, women who nherted land n 1999 provded externaltes to households n Table 5 here Results pertanng to the varables assocated wth governance are sgnfcant. IBV reduced probablty of beng vulnerable by up to 19.8%. The ex post mpact of IBV vared between 6.4% and 8%. The long term mpact of poltcal reservatons and regme change (male to female) on poverty was lmted and became nsgnfcant n The mpact of partcpaton n GS was the largest when we consder 120% poverty lne. Durng perods of economc downturn when more people are lkely to be poor, partcpaton n GS meetngs by households helped access resources, welfare programs and mproved qualty of vllage governance. The ex post probablty of poverty declned by 13.6%. 13

14 Access to publc servces lke school and hosptals was more mportant for ex-ante vulnerablty as opposed to ex post poverty. However access to transportaton affected both. Ths s consstent wth the mpact of access to transportaton as opposed to schools and hosptals on per capta consumpton (9.6% as compared to 8.6% and 3.2% for schools and hosptals). Panchayats have been mandated to mprove transportaton servces by ncreasng access (e.g. bus stops n vllages). Dong so reduced the ex post probablty of poverty by 6.1%. Table 6 presents the results of estmatng (8) and (9) for poverty transtons. Our man nterests are () the coeffcent of the vulnerablty n the thrd column (P 3 ) of each of cases: 100% poverty lne, 120% poverty lne and 80% poverty lne. () The dfference between the second and the frst columns (P 2 - P 1 ). The log odds of former to the base group (the chroncally non poor) represents relatve probablty of poverty for those who were not poor n 1999 but poor n 2006 to the non-poor over whereas the latter gves relatve probablty of movng out of poverty to the probablty of stayng n poverty. Therefore, the postve sgn n the thrd columns ndcates hgher probablty of fallng nto poverty whle the negatve sgn of the dfference (P 2 -P 1 ) mples less lkelhood of movng out of poverty. The coeffcent of vulnerablty n the thrd column,.e., dfferences between the coeffcents n the thrd column and the base group, range from 1.93 to 4.81, and are strongly sgnfcant, mplyng that a rse n vulnerablty tends to ncrease relatve probablty of slppng nto poverty. Besdes, the dfference n coeffcents between the second and frst columns, whch s negatve, n each case suggests that relatve probablty of escapng poverty decreased as vulnerablty ncreased. 6 Hence, reducng vulnerablty does not only prevent a household from fallng nto poverty (.e. protectve effect), but also promotes a household s escape from poverty (.e. promotonal effect). Inhertance of land by women on poverty transtons retards movng nto poverty by up to 30.7 % but does not do much for escapng from poverty. The same s true for partcpatng n GS. Ths maybe because those that attend GS are usually the non-poor and not all the vulnerable groups. 7 IBV allows households to move out of poverty but the protectve effect of IBV s nsgnfcant (although sgns are correct ). All publc servces help n transtng out of poverty. It s mportant to understand the exstence of vulnerablty dstrbutons. Polces ntended to reduce vulnerablty mght be found to be nsgnfcant on an average but are sgnfcant for specfc groups of households. A household that s most 6 We carred out statstcal test and confrmed that the dfferences n the coeffcents are statstcally sgnfcant at 1%. Also the Hausman tests for the IIA assumpton wth respect to our multnomal logt model support the hypothess that omttng one of the categores wll not change the coeffcent estmates systematcally. 7 In Jha et al (2012d) nvestgates the role of IBV votng on program partcpaton. One of the fndngs of that paper s that f the Jat of household s the same as that of the elected representatve (the Pradhan n partcular) then such a household s lkely to attend 3.2 more gram Sabha meetngs compared to all others. 14

15 vulnerable s lkely to beneft more by access to health compared to less vulnerable (upper 50% n the dstrbuton). These fndngs are of sgnfcant mportance to the Panchayats especally gven that both vulnerablty status and locaton n the dstrbuton are unobservable to the Panchayats. Table 7 reports results of quantle regressons descrbed n (10) and (11). We focus on a few of the results related to publc servces and ssues related to ndvdual empowerment and governance. Table 6 here. Access to schools and health care are mportant determnants of vulnerablty for households n the bottom half of the dstrbuton. Ther mportance to reducng vulnerablty ncreases as we move from 1999 to The same s true of access to transportaton. However changes n magntudes are not as sgnfcant as for schools and hosptals. Some of the dentfers of governance such as regme change and partcpaton n GS are also mportant. Partcpaton n GS benefts some of the more vulnerable groups to a greater extent, e.g., t reduces vulnerablty of households at the 50 percentle of the dstrbuton by as much as 10 % n 1999 and the benefts ncrease by 3% between 1999 and For elected representatves at the level of the Panchayat ths fndng s mportant snce t s qute possble that certan groups (not necessarly the less vulnerable groups) partcpate more actvely n GS and consequently ether crowd out partcpaton by other groups or capture benefts va access to nformaton. Land nhertance by women and the mpact of women headed households provde some nterestng nsghts. In both cases, such households are more vulnerable n 1999 across all vulnerablty classes. Table 6 shows that the mpact of nhertance takes place wth a lagged effect. Households are sgnfcantly less vulnerable n 2006 as a consequence of nhertance n Smlarly, women headed households whose transton nto poverty s retarded the most (Table 6) are less vulnerable n Thus, legslatons that empower women through ether poltcal reservatons or by enablng nhertance affect the ex-ante vulnerablty wth a lagged effect. Table 7 here. Turnng now to VEU we estmate (15) and (16). 8 The four components of VEU are shown n Table 8. The estmate of average VEU (0.3016) mples that the average rural Indan household has experenced about 30% of utlty loss due to poverty and varous rsks durng The negatve sgn of covarate rsk component suggests that the utlty loss would have been hgher had there been no economc growth n rural Inda durng the sample perod. Table 8 here. 8 Defntons of all varables used and summary statstcs there on are gven n Jha et al (2012b) 15

16 Estmates of covarate rsk and dosyncratc rsk usng Panel IV methods are reported n Tables 9 and 10. The coeffcent estmate, negatve and strongly sgnfcant, of tme dummy for year 1999 ndcates that aggregate rsk s mostly explaned by a postve covarate shock. Poltcal reservatons ncrease per capta consumpton over tme. The dummy varables of agro clmatc condton (ard regon s the reference) show that humd and sem-ard tropcs are postvely assocated wth household consumpton. 9 Thus, f there are poor qualty weather forecasts then households could resort to reducng consumpton. Evdence ponts to the complete lack of capacty on the part of Panchayats to ntervene n ths process. Estmaton of (15) and (16) shows up two nterestng results. 10 Poltcal reservatons for women have a delayed mpact on ncome. Second, effectve provson of publc servces affects consumpton. Snce we try to understand consumpton va changes n ncome, f dosyncratc shocks have a sgnfcant nfluence on ncome then strateges to mnmze these effects on ncome must be explored. Tables 9 and 10 here. In the frst stage estmaton, female headshp s postvely assocated wth a household s ncome and ts coeffcent s sgnfcant at 5%. The coeffcent of female share n the household s postve and sgnfcant. The negatve and sgnfcant coeffcent estmate of square term suggests a non-lnear assocaton. We cannot fnd any evdence of lfe cycle effects. The negatve coeffcent of share of members wth hgh secondary or hgher educaton, and the postve coeffcent of ts square term suggests that educaton has a postve but convex mpact on household ncome. There s evdence of a hump-shaped relatonshp between a household s productve assets and household ncome. Hgher share of rrgated land to total land ncreased household per capta ncome. In the second stage estmaton, the estmate of predcted per capta ncome s sgnfcant suggestng that a household s consumpton s determned largely by ts ncome. The supply of publc servces s postvely assocated wth household consumpton. Ths, together wth the negatve assocaton, though nsgnfcant, between publc servces and household ncome, mples that the provson of publc good tends to promote household consumpton but not necessarly household ncome. Turnng to copng strateges Table 11 provdes results of multvarate Probt estmaton usng (17) and (18) usng 2006 data. Panel (A) uses dosyncratc and covarate shocks whereas panel (B) uses shock varables and other household and vllage level varables as n (15) and (16). 11 We used Huber-Whte 9 We use the ICRISAT defnton of agro clmatc zones. 10 Income s nstrumented n the frst stage by the per capta number of productve assets (e.g. tractors) and the share of the area of rrgated land to total land a household owned. Although the possble effects of these varables on household consumpton cannot be dened, t mght be reasonable that they frst affect household ncome (Gaha and Ima, 2009). 11 For example, f a household has experenced any shock among drought, floods, lvestock epdemc and epdemc, then the dummy of covarate shock n panel (A) takes one. Smlarly, dosyncratc shock takes one f the household has n the past faced ether death of ts members, sudden health problem or crop falure. 16

17 sandwch estmator to overcome heteroskedastcty. Panels A and B reveal that the man copng mechansm s runnng down ts own savng. Panel (A) shows that the household s lkely to choose borrowng or transfer from ts relatves or frends n the face of dosyncratc shocks whereas t reles on captal depleton when t faces covarate shocks. 12 Ths choce of copng strategy seems to be sensble snce covarate shocks affect other resdents wthn the vllage as well as the household under consderaton; t s less lkely that ths household wll use transfers as a copng strategy. The household would be forced to sell ts assets, reduce consumpton or wthdraw chldren from school. Smlar fndngs are observed from panel (B) where dsaggregated shock varables are ncluded. Thrd, although the household s less lkely to choose a government provded copng program when t faces dosyncratc shock, the probablty sharply ncreases n the case of covarate shocks. The fact that governments programs are not reled upon as much as other measures concernng.. 13 Table 11 here A useful way to nvestgate the effectveness of mutual nsurance s to estmate the followng model: ln c = φ + γ ln y + ψ (ln y ) + δx + ε (19) vt vt vt vt vt Where ln cvt and ln yvt denote the growth rate of household consumpton and ncome respectvely ln y ) represents the growth rate of average vllage (or state) ncome and s treated as ( vt a proxy of aggregate shocks (Townsend, 1994) The estmate of γ,on whch much of the emprcal lterature, focuses ndcates whether the nsurance mechansm s effectve wthn vllages (states). If complete nsurance or rsk sharng exsts, γ s expected to be zero and sgnfcant. Thus, the hgher values of γ would be evdence of weak consumpton nsurance aganst ncome rsks Captal depleton s defned f a household chooses, as a copng strategy, ether sellng assets, wthdrawal of chldren educaton, reducng consumpton or change crop choce. The elements of captal depleton all would affect household future ncome. For example, reducng consumpton would lower labour productvty through nadequate nutrton ntake. Change n crop choce would also result n low rsk and low expected ncome. Therefore, these strateges are aggregated as captal depleton as ts consequence s lkely to dfferent from other copng strateges. 13 Some mght argue that ths s not the rght nterpretaton. Households rarely have to reduce food consumpton n the case of dosyncratc shocks (Rosenzweg and Bnswanger (1993). Whle t s true that households have some access to access to borrowng and transfers from neghbors, what s of concern s the propensty to reduce meals. Ths has serous consequences for households becomng vulnerable to other shocks. Whle t may be deemed approprate to let them self nsure, and the Panchayats should focus on covarant shocks- whch s exactly what t seems to be dong albet ncrementally-nevertheless gven that some households reduce cooked food consumpton pants Panchayats n a rather adverse lght. If Panchayats actually reflect household preferences then t seems llogcal that programs actually avalable wth the Panchayats to provde ncomes and food durng tmes of dstress are not beng utlzed. Is ths a symptom of Bureaucratc Dysfuncton? It certanly seems so. 14 We nterpret γ as a measure of consumpton nsurance rather than a measure of vulnerablty done n VER measure. VER model assumes that postve and negatve shocks have symmetrc effects. However, ablty dealng wth postve shocks (e.g. accumulatng assets or savng etc.) compared to negatve shocks (sellng 17

18 Food consumpton s frequently used as a measure of welfare n regons where a substantal fracton of the populaton devotes more than three quarters of ther expendture to food (Deaton, 1997). Households are expected to be conservatve n mantanng the level of consumpton expendture and hence, expendture devoted to food s lkely to be less covarate wth ncome than other components of expendture as often reflected by the Engel curve n developng countres. Further, shares of food consumpton to total consumpton expendture are qute hgh (57.7% n 1999 and 53.8% n 2006) suggestng the possblty that households mght be keen to nsure ther food consumpton aganst negatve shocks. Therefore, we use three dfferent dependent varables (total consumpton, food consumpton, non-food consumpton expendture). Results n Table 12 show that growth of consumpton s postvely related to growth of ncome. Although households rely manly on nformal rsk-copng nstruments such rsk-copng s not able suffcently smoothen the mpact of shocks. Furthermore, the negatve and sgnfcant coeffcent estmate for the change of vllage mean of log ncome suggests that there s no rsk sharng mechansm wthn vllages. 15 The constraned access to effcent rsk management mechansms among households mples that there s substantal rsk of ncrease n ural transent poor n the face of shocks. Besdes, our fndng from Table 11 that one of the sgnfcant households copng strateges s captal depleton mght nduce prolonged consequences for household welfare, and may trap them n poverty. Comparng the coeffcents of estmates for food consumpton and non-food consumpton estmatons, the smaller coeffcent estmates of the change of log ncome n food consumpton, suggests that households are keener to nsure food consumpton than non-food consumpton from ncome shocks. Further partcpaton n GS meetngs, engagng n IBV, and, regme changes (gender and Jat based) lead to nsurng food consumpton more than overall consumpton. Gven that Panchayats are responsble for the management of PDS ths result s unsurprsng. Table 12 here. Parochal poltcs at the level of the Panchayats provde many benefts ncludng accessng food artcles through the PDS. We fnd that partcpaton n government programs admnstered through the Panchayats s asssted by there beng a female Pradhan, votng n local electons, and partcpaton n GS. If governments programs are the preferred opton for certan classes of households (such as asset poor, or those wth assets or recevng transfers etc.) s lkely to dfferent among households. Therefore, n order to nterpret γ as a measure of vulnerablty t s suggested to replace ln y wth two covarates denotng absolute values of the vt sze of postve and negatve ncome changes (Hoddnott and Qusumbng, 2003). 15 As our results are based on rural representatve data, t s not necessarly contradct from the fndng of partal rsk sharng n other studes of whch many draws upon the selected sample vllage data wth mostly annual consumpton and ncome data. Our results are smlar wth the fndngs observed n Shrapur and Kanzara vllages n Gaha and Ima (2009). 18

19 low savngs) then such households are lkely to transt nto poverty f access s ether dened or provded wth condtons. Of all copng strateges, adoptng newer technologes has the largest mpact on consumpton growth. Panchayats should therefore pay attenton to agrcultural extensons programs. Depleton of savngs (adopted by nearly 60% of all households affected by shocks) leads to a 74% growth n consumpton compared to all other copng strateges. But such growth cannot be sustaned snce depleton of savngs cannot take place over a lengthy tme perod. The growth mpact of government programs s 64% on consumpton growth. Often government programs such as the MGNREGS have been poorly targeted wth partcpants not beng pad the guaranteed mnmum wage. In that case even after accessng such programs ncome and consumpton growth wll be smaller compared to other sometmes rsker strateges of copng. VI. Conclusons and some lessons for polcy Durng , 52% of households have been affected by covarate or dosyncratc shocks, wth covarant shocks much more frequent and leadng to much hgher losses than dosyncratc shocks (except for ndvdual-specfc crop losses or dryng up of own wells). Estmate of average VEU mples that the average rural Indan household has experenced about 30% of utlty loss due to poverty or nequalty, and varous rsks durng Wth respect to frst tme shocks, usng up savngs s by far the most mportant copng mechansm, wth around 60% of households usng t, followed by help from local government (9%) and reducng meals (2.4%). If the shock repeats, only 4% of households can use remanng savngs, whle 23% borrow n cash or n knd or rely on transfers; fully 15% have to reduce meals; and a full 19% have to sell assets. When shocks are repeated fve years later, the use of savngs rebounds to 52% of households, whle sale of assets and reducng meals drop back to 8% and 2.3% respectvely, ndcatng a sgnfcant, but not full, recovery of copng ablty of the households. It s therefore clear that the lmted ablty to cushon especally aganst rapdly repeatng shocks would make t more lkely that households wll fall nto poverty traps after repeat shocks. Households are lkely to choose borrowng or transfer from relatves or frends n the face of dosyncratc shocks whereas they have to resort to captal depleton when they face covarate shocks. Thus, covarate shocks are not only more frequent and cause hgher losses but also more dffcult to nsure aganst. It has long been recognzed that t s approprate to let households prmarly rely on self-nsurance for dosyncratc shocks, whle the government should focus on covarate shocks. However, ths strategy s lkely to be suboptmal gven that households resort to two rather negatve copng strateges reducng consumpton of meals, and sellng assets both wth sgnfcant long term consequences. Wthn ther lmted resources, Panchayats appear to be provdng only lmted cover for both covarate and dosyncratc shocks. 19

20 Thus copng strateges are constraned, especally when shocks repeat wthn a short perod of tme, and/or when the shocks are covarant. For the prvate responses, the nabltes are lkely to be assocated wth ncomplete credt and labor markets. The lmted ablty of Panchayats to help could be attrbuted to lack of resources, or neffcences of the Panchayats terms of targetng, plannng ahead for dstress events, or n terms of pathologes such as elte and program capture. Negatve ncome shocks reduce food consumpton less than non-food consumpton, though the dfference s not large. We also fnd that partcpaton n GS meetngs, engagng n IBV, and, regme changes (gender based as well as Jat based) lead to nsurng food consumpton more than overall consumpton. We estmate a mean vulnerablty of 28.3% n 1999 and 16.5% n A rural Indan household, on average, had 28.3% probablty of fallng nto poverty n 1999 but ths declned to 16.5% n Ths s prmarly assocated wth the farly rapd ncome growth n the vllages. We also fnd that exante vulnerablty n 1999 translates nto ex post poverty n 2006,.e. 1% ncrease n the ex-ante probablty of fallng nto poverty would ncrease ex post probablty of poverty by 0.32%. Clearly a rse n vulnerablty tends to ncrease relatve probablty of slppng nto poverty consderably. We also show that t reduces the probablty of escapng from poverty. Local governments are able to assst an average of 9% of households n the face of an ntal shock, 8% of households when a shock repeats n the next year, and 12% when shocks reoccur after 5 years. Whle households have a low probablty to access a government provded copng program when they faces dosyncratc shock, the probablty sharply ncreases for covarate shocks Durng perods of economc downturn when more people are lkely to be poor, partcpaton n GS meetngs by households wll help access resources, welfare programs and mprove qualty of governance at the level of the vllages. The ex post probablty of poverty declnes by 13.6%. Partcpaton n GS meetngs benefts some of the more vulnerable groups to a greater extent. Less vulnerable groups also seem to partcpate actvely n GS meetngs to ensure that they are adequately served. IBV reduces probablty of vulnerablty by up to 19.8%. The ex post mpact of IBV on poverty vares between 6.4% and 8%. 16 The long term mpact of poltcal reservatons and regme change (male to female) on poverty s lmted and becomes nsgnfcant n A household headed by a female has a relatvely lower probablty of poverty n 2006 than a household headed by a male: We also fnd that female headshp s postvely assocated wth a household s ncome. Ths remarkable result shows that a woman as head of the household reduces both ex-antev ulnerablty and ex post poverty. It also helps n sgnfcantly retardng the transtons 16 IBV allows households to move out of poverty but the protectve effect s nsgnfcant. 20