Policy

Research

Working

Paper10683e

Welf

are

C

ost

of

D

rou

ghtin

Sub-Saharan

AfricaJon

GascoigneSandra

BaquieKatja

VinhaEmmanuelSkou?asEvie

CalcuttVarun

KshirsagarConorMeenanRuthHillPov

ert

yand

Equity

Global

PracticeJanu

ary

2024Policy

Research

Working

Paper

10683Abstractis

p

ap

er

quanti?es

the

imp

act

of

d

rou

ght

on

householdconsumption

for

?ve

main

agroecological

z

ones

in

A

f

rica,develop

ing

vulnerability

(or

damage)

functions

of

therelat

ionship

between

rainfall

de?cits

and

p

overty.

Damagefunctions

a

re

a

key

elem

ent

in

models

that

quantify

therisk

of

extreme

weatherand

the

imp

acts

of

climate

change.A

lthou

gh

these

functions

are

com

m

only

estimated

forst

orm

or

?ood

damages

tobuildings,

they

are

less

oftenavailab

le

for

income

losses

f

rom

d

rou

ght

s.e

p

ap

er

takesa

regionalap

p

roach

to

the

analysis,

develop

ing

standardizedhaz

ard

de?nitions

and

methods

for

matching

haz

ard

andhousehold

data,

allowing

survey

data

from

close

to

100,000hou

sehold

s

tobe

used

in

the

analy

sis.

e

damage

func-tions

are

used

to

quantify

the

imp

act

of

historical

weatherconditionson

p

overty

for

eight

countries,

highlighting

therisk

to

p

overty

outcomes

that

weather

variability

causes.National

p

overty

rat

es

are

1–12

p

ercent

hi

gher,

dependingon

the

country,

underthe

worst

weather

conditions

relativeto

the

best

conditions

observed

in

the

p

ast

13

y

ears.

isamounts

to

an

increase

in

the

total

p

overty

gap

that

rangesf

rom

US$4

m

illion

toUS$2.4

b

illion

(2011

p

urchasingp

ower

p

arity).is

paper

is

a

p

roduct

of

the

Pov

ert

y

and

Equity

Global

Practice.

It

is

p

art

of

a

larger

e?

ort

by

the

World

Bank

top

rovide

op

en

access

to

its

research

and

make

a

contributionto

develop

ment

p

olicy

discussions

arou

nd

the

w

orld

.

Poli

cyResearch

Working

Pap

ers

a

re

also

p

osted

on

the

Web

at

http

:///p

rwp

.e

au

t

hors

may

be

contactedat

rhill@

w

orld

b

ank.

org.e

Policy

Research

Working

Paper

Series

disseminates

the

?ndings

of

work

in

progress

to

encourage

the

exchange

of

ideas

about

developmentissues.

An

objective

of

the

series

is

to

get

the

?ndings

out

quickly,

even

if

the

presentations

are

less

than

fully

polished.

e

papers

carry

thenames

of

the

authors

and

should

be

cited

accordingly.

e

?ndings,

interpretations,

and

conclusions

expressed

in

this

paper

are

entirely

thoseoftheauthors.eydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsa?liatedorganizations,orthoseoftheExecutiveDirectorsofthe

World

Bankorthegovernmentstheyrepresent.ProducedbytheResearchSupport

TeamThe

Welfare

Cost

of

Drought

in

Sub-Saharan

AfricaJonGascoigne(Centre

forDisasterProtection)?

SandraBaquie(WorldBank)?

KatjaVinha(Consultant)?

EmmanuelSkoufias(NationalUniversityof

SingaporeandWorldBank)?

EvieCalcutt(WorldBank)?

VarunKshirsagarConsultant)?

ConorMeenan(Centre

forDisasterProtection)?

RuthHill(WorldBankandCentreforDisasterProtection)1JELCodes:Q54,I32ProjectCode:P177432Keywords/topics:Shocksandvulnerabilitytopoverty,drought1

?indicates

randomizedauthororderusingtheAmericanEconomicAssociationauthor

randomizationtool,/journals/policies/random-author-order/search?RandomAuthorsSearch%5Bsearch%5D=XxSZtJmQOh5k.ThispaperstartedasaresultofaconversationwithJonGascoigneinearly2020.Jon

understoodbetterthananyofustheneedforamultidisciplinaryteamtounderstandhowtomodelthewelfareimpactsofdrought.Collaboratingacrossdisciplinescanbeachallenge,andnegotiatingthosediscussionsrequiresgoodgrace,humility,astrongfocusonrelationship,andagenuineinterestinunderstandinganother’sviewandexpertise.Jonexemplifiedthesecharacteristicsinallofhiswork,andtruetoform,broughtthemtothisprojecttoo.Wethinkitisveryfittingthattherandomizationtoolplacedhimasfirstauthoronthispaperearlyinitsdrafting.Wearesaddenedhedidn’tgetto

seethefinalproductandto

knowjusthowgratefulweall

weretohimandtheexamplehesetus.WethankStefanieBrunelinandOdyssiaNgforbeingpartoftheearlyconversationonthispaperandwethankBhavinThakrarandPaulWilsonforworkingonpartsofthe

simulationsdataexercise.1.

IntroductionClimatedisastersreversegainsandlimitprogressineffortstoreducepoverty.GlobalextremepovertyisincreasinglyconcentratedinSub-SaharanAfrica(SSA)wheredroughtisamajorrisk,ariskthatisincreasinginmanypartsofthecontinentbecauseofclimatechange.Investmentstoreducetheimpactofdroughtonwelfarearethusessentialtopovertyreductioneffortsintheregion.

Quantifyingthewelfareneedsthatarisewhenclimateshocksoccurisanimportantpartoftargetinginvestmentstoreducetheirimpact(ClarkeandDercon2016).

Thisrequiresvulnerabilityordamagefunctionsthatrelatelossesinwelfaretoclimateconditions(Auffhammer2018).Whilethesearewidelyavailablefortheimpactoffloodsandcyclonesonbuildings,thisisnotthecasewhenitcomestoquantifyingtheimpactofdroughtonhouseholdconsumptioninSub-SaharanAfrica.Thispaperpilotsaregionalapproachtoestimatingvulnerabilityfunctionsfordroughtandconsumptionusingmethodswidelyemployedinthemicroeconomicliterature(Dell,Jones,andOlken2014).

Theanalysisusessurveydatafromcloseto100,000householdsacrossninecountriesandfivelivelihood

zones,allowingconsiderablevariationinclimaticconditionstobeutilizedinestimation.

Theresultisstableestimatesthatarenotdrivenby

anyoneclimate

event.Wefindthatconsumptionisreducedby

10-20percentintheworstweatherobservationsinoursurveys(correspondingto

abouta1in10-yeardroughtevent).Weusetheestimatedvulnerability

functionsandthefullhazarddistributionineachhouseholdlocationtocalculatetheriskdroughtposestowelfareoutcomes.

Theresultsindicatethatriskislarge.Povertyis1–12percenthigherundertheworstweather

conditionsrelativetothebestconditionsobservedinthepast13years.ThisamountstoanincreaseinthetotalnationalpovertygapthatrangesfromUS$4milliontoUS$2.4billion(2011PPP)acrosscountries.To

ourknowledge,thesearethefirstcross-countryestimatesofthepovertyimpactofdroughtinAfrica.There

are

several

ways

in

which

damage

functions

for

rainfall

deficits

have

been

estimated

in

the

climate

economicsliterature.

Auffhammer

(2018)

provides

a

reviewand

examples

oftheapproaches

applied

to

US

farmincomes.

Theone

used

for

estimating

the

impact

of

extreme

weather

events

uses

observed

data

to

identify

the

causal

impact

ofweatherchangesonoutcomesofinterest(Dell,Jones,and

Olken2014).Dataonhazards

arematchedtohouseholdobservations

in

survey

data

using

information

on

the

geographic

location

of

a

household,

and

a

fixed

effectsregressionisruntoidentifytheimpactofweatherindeterminingeconomicoutcomes.Although

not

undertaken

with

the

purpose

of

estimating

vulnerability

functions,

several

studies

have

used

thismethod

to

estimate

the

impact

of

drought

on

welfare.

For

example,

in

Africa,

Hill

and

Porter

(2017)

look

at

theimpact

of

a

water

requirement

satisfaction

index

(WRSI)

on

consumption

and

poverty,

and

Hill

and

Mejia-Mantilla(2017)

look

at

the

impact

of

WRSI

on

household

income

and

consumption

in

Uganda.

In

Malawi,

Baquie

and

Fuje(2020)

look

at

the

impact

of

rainfall

on

consumption

and

poverty.

Wineman,

Mason,

Ochieng

and

Kirimi

(2019)examine

therelationship

between

rainfall

and

income,

calorieconsumption

and

poverty

in

Kenya.

Baez,

Kshirsagarand

Skoufias

(2019)

look

at

the

impact

of

rainfall

defined

dry

spells

on

nutrition

outcomes.

Outside

Africa,

Kochharand

Knippenberg

(2023)

look

at

the

impact

of

NDVI,

rainfall

and

temperature

on

consumption

and

poverty

inAfghanistan.

Studies

that

use

estimated

impacts

of

drought

to

predict

the

welfare

cost

of

other

weather

eventsincludeHillandPorter(2017),

PorterandWhite(2016),KochharandKnippenberg(2023).2Thisstudyusesthe

samemethodsas

thesepapersbuttakes

aregionalapproach(asinBaez,KshirsagarandSkoufias2019)withtheobjectiveofdevelopingrobustestimatesoftherelationshipbetweendroughtconditionsandconsumptionthatisnotspecifictoonecontextor

event.Thisrelationshipcanthenbeusedtopredicttheimpactofdroughtforothereventsorlocations.Theestimatespresentedinthispaperarewithintherangeofresultsfoundincountry-specificpapers.Forexample,amoderate(abouta1in10year)droughtispredictedtoreduceconsumptionby15percentand9percentinUgandaandEthiopia,akinto

the10percentlossweestimateforcropcultivatinghouseholdsinmaizeandhighlandzonesinthispaper.2Anttila-HughesandSharma(2015)takethesameapproachforstorms.2Takingaregionalapproachrequirestheuseofstandardizedhazarddefinitionsandmethodsformatchinghazardandhouseholddata.Thisisdonebyconsideringthemainhazardmeasuresandtechniquesusedincomparablecountrystudies,andsystematicallyassessingthechoiceofhazard,thehazard-householdmatchingapproach,andthefunctionalformusedinanalysis.Specifically,weusegriddedhistoricaldataonprecipitation,vegetation,evapotranspiration,andsoil

moisture,usingmeasuressuchastheNormalizedDifferenceVegetationIndex(NDVI),theStandardizedPrecipitationEvapotranspirationIndex(SPEI),andWRSI.Thesemeasuresreflectthehazardtolivelihoodsfromlackofrainindifferentways.Somearedirectmeasuresofclimateconditions(e.g.,precipitation,soilmoisture,andSPEI),othersaggregateclimatedatathroughcropmodels(WRSI),andothersmeasurevegetationoutcomes(NDVI).Thegriddedhazardmeasuresarethenmergedtoharmonizedhouseholdsurveysfromninecountries

(Ethiopia,Lesotho,Malawi,Mauritania,Mozambique,Niger,Nigeria,

Zambia,andZimbabwe)withlocationcoordinatesforeitherthehouseholdorthecommunity.Thesurveysusedaretheonesthatarealsousedforofficialpovertyestimates.Aconsistentapproachto

temporalmatchingwasusedtotakeintoaccountthetimingofthesurveyandagriculturalseasons.Differentapproachestospatial

mergingweretested.Thereareimportantmethodologicalinsightsfromtheanalysisthatemerge.

First,we

findthatsoilmoistureandgreennessmeasuresprovidemoreconsistentlyreliableresultsthanevapotranspirationorrainfallmeasures.Thisisanimportantfindinggivenmostoftheextantliteratureisbasedonevapotranspirationorrainfallmetrics.Second,wefindthatalinearspecificationperformsadequatelyforthetypesofmoderateshocksthatarewell-representedinthesurveydataused.Third,wefindthatahazardmeasurebasedon20kmor50kmradiusaroundeachhouseholdperformsbetterthanthosethat

usesmallerradii,perhapsbecausealargerradiusallows

localmarketeffectstoalsobeconsidered.Therelationshipbetweenhazardmeasuresandwelfareoutcomesisestimatedseparately

fordifferentlivelihoodzonesacrossSSA.Separateestimationprovidesneededflexibility,giventhatdroughtconditionsmanifestthemselvesdifferentlyacrosslivelihoodzones.Importantdifferencesinimpactsbetween

agroecologicalzonesandbetweenlandandlivestockowners

areobserved.However,thedownsideofaregionalapproachisalimitedabilitytoexploretheheterogeneity

ofimpactsacrosshouseholdsindetailwithoutsignificantinvestmentsindataharmonization,andweleaveadditionalworkontheheterogeneityofimpacts

assomethingtoexplorefurtherinfutureanalysis.Therearelimitationstotheanalysisundertakenthat

shouldbeborneinmindwheninterpretingresults.First,estimates

capturethedirectimpactoflocaldroughtconditionsandnot

impactsbeyondthelocalarea,suchasthosethatmightarisefromdisruptionofmarketsorpriceimpacts.Thedegreetowhich

thisisaconcerndependsontherelativeimportanceofdirectandindirectimpacts.Theevidenceismixedonthis.Hallegatteetal.(2016)suggestfoodpricesareanimportantchannelofthewelfareimpactsofclimatechange.

However,Artucetal.(2023)

modeltradebetweenlocations

andfindthemainimpactisthedirectproductioneffect,nottheimpactonprices.To

theextentthisisthecasethelocalestimatespresentedinthispapermay

bequiteclosetoaggregateimpacts.Secondly,theempirical

estimatesaresubjectto“survivorbias”

inthat

consumptionisonlymeasuredforhouseholdsstillinexistenceaftertheshockhasoccurred.Ifhouseholdsmoveordisintegrate(perhapsduetothedeathofoneormorehouseholdmembers)becauseofdroughtconditions,theestimatedimpactscouldbe

overorunderestimated.Thisismuchmorelikelytobethecaseforseveredroughtconditions,whereas

thosethatarecapturedinoursurveystendto

belessextremeeventsthat

areunlikelytocausemovementordisintegrationofhouseholds.Thisbringsusto

thethirdpotentialsourceofbiasintheresults.Althoughtheanalysisuses

surveydata

acrossmanyweatherconditionsintheanalysis,

thereareveryfewobservationsinthesurveydatathatexperiencedroughtconditionsbelowtwo

standarddeviationsfromthe

mean.Thislimitsthedegreeto

whichnon-linearitiescanbeestimated.Italsomeansthereislimitedempiricalsupportforpredictingthewelfarecostsofextremeevents.Afinalsourceofbiascomes

frommeasurementerrorin

thehazarddata.

Althoughcarewastakentousethebestavailabledataandmatchhazardandhouseholddata

carefully,anymeasurementerrorinthehazardvariablewillattenuatetheestimatedcoefficients,therebyunderestimatingthewelfareimpacts.Additionaluncertaintyis3presentgiventhedistributionofhazardsischangingovertimewithclimate

changeandhistoricaldataformsthebasisofallsimulationspresentedin

thispaper.Themodelsbuiltinthispaperareanimportantfirststeptowardconstructingvulnerabilityordamagefunctionsasusedinthecatastropheriskmodelingliterature,andcaninformdisasterresponseplanningandfinancing.However,theselimitationsmeanthatcareisneededininterpretingtheresultsoftheanalysis.Weenvisagethatthesetoftoolsdevelopedhereinwillprovidethebasisforrefiningtheanalysisasmoredatabecomeavailable,andexpandingcoveragetootherregions.Accompanyingthispaperiscodedesignedtoextractalargesetofpotentialshockmeasuresanddeterminewhichoftheseareconsistentwithchangesinthewelfaremeasures(out-of-sample).The

modularcodeusedisintendedto

betransparent,toaidinreplicability,andtobeavailableforotherstoimproveupon.Thegoalistoencouragebothlearningandcommonstandards,whichinturncontributetotheimplementationofdrought-responsivepoliciesthat

areobjectiveandtransparent.Thepaperisstructuredasfollows.Section2presentstheempiricalframeworkforassessingtherelationshipbetweenwelfareandhazardsinruralareaswhererainfall-dependentagriculturalproductionisdominant,andwhererainfalldeficits(droughts)thusmayhaveveryseriouswelfare

consequences.Itsetsouttheapproachusedtoestimatethevulnerabilityfunctionandan

exceedanceprobabilitycurvethat

describesthewelfarelosseslikelytooccurbecauseofrainfalldeficitsofincreasingseverity.

Section3discussesindetailthehistoricalhazarddatabaseandthehouseholdsurveydataused,themergingofthehazarddatato

thehouseholdsurveydata,andthespecificationsusedforthehazardmeasures.Section4presentsestimatesofvulnerabilityfunctions.Section5showshowthesevulnerabilityfunctionscanbeusedtoprojectwelfarelossesandpresentssomeexceedanceprobabilitycurves.Section6offers

someconclusions.2.

An

EmpiricalFrameworkfor

an

Ex

AnteAssessmentoftheWelfareCostsofDroughtAcatastropheriskmodelingframeworkusesthreecomponentstogenerateanex

anteassessmentofthecostsofacatastrophe:hazard,exposure,andvulnerabilityfunction.

Hazardscapturedata

onthepossibleweatheroutcomesthatmayoccurina

givenplace,suchaspossibletropicalcyclones.Exposurecaptureswhatcouldbeaffectedbythehazardinthatlocation,suchas

thenumberandtypeofbuildings.A

vulnerabilityfunctionprovidesanestimateofthecostofdamagethatwilloccurasaresultofagivenhazardandagivenexposure.Intheexampleoftropicalcyclonesandbuildings,itwouldindicatethecostofdamageforagiven

windspeedaffectingagiventypeofbuilding.Combiningthethreecomponentsallows

developmentofaprobabilisticimpactcurve,whichindicates

howthecostofhazardvarieswiththeprobabilityofthehazardforaspecificplace,giventheexpectedhazardsandtheexposedassets.Thissectionofthepapershows

howthisframeworkcanbe

appliedtogenerateaprobabilisticimpactcurvefordroughtinSSA.FordroughtinSSA,thehazardisrainfalldeficitsduringthegrowingseason.Asnotedintheintroduction,therearedifferentwaysofmeasuringthis

hazard,andthenextsectiondiscussestherelevantdata

inmoredetail,includingthetimeperiodforwhichhistoricaldataareavailableanduseofthedatatogenerateanestimateoftheprobabilitydistributionofhazardsinthenearfuture.Exposureinthecurrentcaseis

notaboutphysicalinfrastructureanditslocationandcharacteristics,butabouthouseholdsthatmaybeimpacteddifferentlybyrainfalldeficitsdependingontheirlocationandcharacteristics.Thedataonhouseholdsarealsodescribedinmoredetailinthenextsection.Thevulnerabilityfunctioninthiscaseistheincomeorwelfarelossesamong

households

engagedindifferentlivelihoods,whichresultfromloweryields,lowerdemandforlabor,andincreased

pricesthatmayresultfromthat.Therestofthissectionfocusesontheestimationofthevulnerabilityfunctionandhowit

isusedtoderiveaprobabilisticimpactcurve.Theestimationofthevulnerabilityfunctioniswidelypresentinthemicroeconometricliterature,eventhoughitisnotreferredtoassuchor

used

forgenerating

estimatesofprobabilisticimpact.42.1.

EstimatingavulnerabilityfunctionHouseholdwelfareatanygiventimemaybesummarizedingeneralbythefunction

below:=(,,)

+

,?(1)denotesthemeasureofwelfare(inthecurrentanalysis,logofpercapitaexpenditureshousehold?

inlocalityorcluster

,),

isthemeasureofthehazardattime

atthelevelofthegeographiccluster

denoteshouseholdobservablecharacteristics,

denotesanobservablevariablesummarizingtheenvironmentofhouseholds,and???whereof??,??isanerrortermsummarizingtheinfluenceofallunobservablefactorson?welfare.3Forthepurposesofthis

study,welfareismeasuredbythemonetaryvalueoftotalconsumptionexpendituresorfoodandnon-foodconsumption.4Themainobjectiveof

avulnerabilityfunctionisto

summarizetherelationshipbetweenthe

levelof

andthelevelof

.Ifthefunction

were

knownaprioriitwouldbesimpletopluginvaluesfor,,and

todeterminethevalueofthewelfareoutcome

.Intheabsenceofinformationonthefunction

,methodsneedtobeemployedtoestimateit.Thisnecessitygivesrisetoafundamentaltrade-offbetweenthepredictionaccuracyandtheinterpretabilityofthemodelused(Jamesetal.2013).Forexample,machinelearningmethods,suchas

random

forests,bagging,boosting,andsupportvector

machines,treat

asablackbox.Noneofthesemethodsisespeciallyconcernedabouttheexactformandinterpretabilityofthefunction

,providedthatityieldsaccuratepredictionsforthewelfareoutcome.Ontheotherhand,approximatingthefunction

withtheaveragevalueofthewelfareoutcomeonspecificvaluesoftheexplanatoryvariables

,and

—i.e.,

(

,

,

)

=conditional?,,(|==1????),

ortheconditionalexpectedfunction(CEF)—is

typicallyeasiertointerpretandisusefulfordeeper,=?insightsintothecausalrelationshipbetweenhazardandwelfare(AngristandPischke2009).Theseadvantages,however,maycomeattheexpenseofpredictiveaccuracy(in-sampleorout-of-sample).5Withtheseconsiderationsinmind,weadopttheCEFapproachforestimatingvulnerabilityfunctions.To

keeptheanalysissimpleandtractable,theapproachbeginswithanad

hocspecificationofthecontrolsusedintheregressionmodel,usingalinearapproximationtotheCEFthatisextendedtoanonlinear

specificationforthehazardandallowingforthepotentialheterogeneityofthehazard’simpactson

welfarebasedonsomeeasilyandcommonlyobservablehouseholdcharacteristicsandenvironmentalfeatures.Foranygivenwelfaremeasure

?,aloglinearapproximationtoequation(1)aboveyields=++++++,?(2)?01?21?where

denotesregion-ordistrict-levelfixedeffectsand

denotescountry-surveyyeareffects.Thedistrict-levelfixedeffectscontrolforfixedspatialcharacteristics,whetherobservedorunobserved,andthusdisentanglethehazardmeasurefrommanypossiblesourcesofomittedvariablebias.Theyeareffects

furthercontrolforanycommontrendsandthushelpensurethattherelationshipsofinterestareidentified

fromidiosyncraticlocalX3

Theelementsof

canincludetheageandeducationlevel

ofthehouseholdhead;theethnicity,religion,totalnumber,age,andgendercompositionofhouseholdmembers;ownershipofassetssuchas

livestock,phone,radio,TV,

etc.;andcharacteristicsofthehouseholdresidence,includingtypeandqualityofwaterandsanitationfacilities,mainsourceofenergy,andtypeoffloorandmaterialofconstructionforwallsandroof.4Focusingonnon-monetarydimensionsofwelfaresuchastheheightforagezscoreofchildren

(HAZ)ortheweightforheightz-score(WHZ),Skoufiasetal.(2023)

exploreanalternativespecification

ofthevulnerabilityfunctionthatallowsconsiderationoftheeffectsofhazardsthroughtheincomeaswellastheenvironmentalchannels.Inthisspecification,exposuretoweather-related

hazardsmayaffectthevalueofagriculturalincomeearned(theincomechannel),

aswellasthehealthstatusof

householdmembers

(environmentalchannel).5Indata-constrainedenvironmentsliketheoneswestudy,complex(black-box)machinelearningmodelsareunlikelytoprovidethegainsin

accuracythatjustifythelossininterpretability.5shocks.Thustheempiricalapproachisverymuchinlinewiththestrongidentificationpropertiesoftheweather-shockapproachsummarizedbyDell,Jones,andOlken(2014).Itshouldbenotedthatwithdistrictfixedeffects,,includedinthespecification,thecoefficients

canbeestimatedonlyifthevaluesofOtherwise,theuseofthedistrictdummies

willabsorbtheeffectsoftheseenvironmentalfactors.varywithindistricts.1?Theimpactofdroughtonconsumptionexpendituresisthe

cumulativeeffectof

ahousehold’scroploss

andaseriesofentitlementfailuresinalocaleconomythataretriggeredbythisinitialloss(Sen,1981;Deveraux,2007).Devereux(2007)characterizesasequenceoffourentitlementfailures:first,productionfailsasaresultoftherainsfailing;thenlabormarketsfail

ashouseholdsarelessandlessabletofindworkopportunitiesonotherfarmsorinoff-farmactivities;thencommoditymarketsfailasgrainpricesincreaseandpricesofliquidassetsdecrease.Finally,transfersfailashouseholdscannotrelyon

thesupportofothersintheirnetworkwhoface

thesameconstraintsinmeetingeverydaybasicneeds.Theformulationoutlinedhereisagnosticas

tohowrainfallisimpactingconsumption.However,thespatialmeasureofdroughtusedand

thenecessityofincorporatingdistrictfixedeffectsfocusestheanalysismoreonpickingupproductionfailuresandfailuresintheimmediatelaborandcommoditylocalmarketinwhichthedroughtismeasured.

Theimpactofdroughtthroughcommoditymarketsthatspanalargergeographicareaortransfernetworksoveralargergeographicspacewillnotbepickedup.Ifhigherlevelsofindicaterainfallsufficiency,itisanticipatedthat

?

>

0suchthatlackofrainfalland2droughtshaveanegativeeffectonwelfare(especiallywhenamonetarywelfaremeasureisused,suchasconsumptionexpendituresperadultequivalent).Theextenttowhichhazardsexperiencedbyhouseholdsinthesampleareassociatedwithsufficientlysignificantdeclinesintheirwelfaredependsinpartonthedegreetowhichthereisvariabilityintheexposuretoandintensityoftheshocksexperienced.Forexample,if

thevalueofwithinagivencluster,thenthereisnowaytoidentifytheparameter

2.Hence,itiscritical

toderiveestimatesof?

basedondatathatincludesignificantvariabilityin

.Thistaskisa

particularchallengewhenanalyzingtheimpactofdroughtbecausetheseeventsarespatiallycorrelatedandnotlocalized.Increasingthenumberofyearsa