market access, marketing behavior and efficiency among farming households
TRANSCRIPT
MARKET ACCESS, MARKETING BEHAVIOR AND EFFICIENCY
AMONG FARMING HOUSEHOLDS IN MOZAMBIQUE
Knowledge Mupanda
Food insecurity and poverty in sub-Saharan Africa are widespread
Farming is the main economic activity among the rural poor
Market participation and efficiency are low
Introduction
How do participation and efficiency influence each other?
What factors affect market participation and efficiency?
Research Questions
Production EfficiencyAigner, Lovell and Schmidt (1977)Bravo-Ureta et al. (2007)Kumbhakar, Ghosh and McGuckin 1991Coelli, T., D. S. P. Rao, , and G. E. Battese. 1998
Market ParticipationGoetz (1992)Key,N.,E. Sadoulet, and A. de Janvry. 2000Bellemare and Barrett (2006)
Recently in the DepartmentRios (2008)Uaiene (2008)
Literature Review
Nationally representative agricultural household survey, Trabalho do Inquerito Agricola (TIA)
total sample 4,900 (in 2002) and 6,149 (in 2005)used panel of 3,696 households surveyed in both years
All values were in MZM converted to US$
The Data
ProvinceNo. of
households
Pct. of sample
Niassa 209 5.65C. Delgado 382 10.34Nampula 473 12.80Zambezia 574 15.53Tete 436 11.80Manica 370 10.01Sofala 283 7.66Inhambane 309 8.36Gaza 430 11.63Maputo 230 6.22Total 3,696 100.00
Distribution of Households by Province
Measuring Market Participation
Market Participation (MP )Categorical {0,1}: whether or not the household sells any
cropPercent Sold (PS ):
Continuous (0,100]: if selling, percent of total crop output sold
where household i produces quantity Xij of crop j, of which it sells quantity xij , at provincial average price Pj
100*
1
1
XP
xPPS
j
j
jj
j
j
jj
Distribution of Market Participation
Less than half of the sample sold some output; MP=1 for 42.7% of households in 2002, 41.2% in 2005)Of those selling, percent sold (PS) is generally low
0%
20%
40%
60%
80%
100%
120%
0%
5%
10%
15%
20%
25%
2 5 10 20 30 40 50 60 70 80 90 More
Cum
ulat
ive
Perc
ent
Perc
ent S
ellin
g
Percent Sold
Yr_2005 Yr_2002 Cum 2005 (Secondary Axis) Cum 2002 (Secondary Axis)
Modeling Market Participation
Each household’s probability of participating (MP ) and their percent sold (PS ) is modeled by a two-stage Heckman regression, following Goetz (1992):
where Xi are household characteristics that affect both participation and percent sold; zi are fixed costs of marketing that affect only probability of participation; the error terms ui and εi are correlated so (1) and (2) are estimated simultaneously.
(1)
(2)
Measuring Farm Efficiency
where xijt is ith farm’s quantity of j = {land, labor} in year t={2002, 2005}, the βs are coefficients to be estimated, vit is a normally distributed error term and uit is half-normally distributed (>0).
Each household’s relative efficiency is measured using Stochastic Frontier Analysis, with a translog production function:
ititj jk
T
tttiktijtjk
jijtjj
jtij0it
uvdxx
xxY j
0
2
lnln
ln21lnln
Modeling Farm Efficiency
Where Xi is a vector of farm and household
characteristics, crops grown, location and market participation (which is endogenous, and instrumented by market access); γ
i are
coefficients to be estimated and wi is a normally-distributed error term
Each household’s relative inefficiency (ui) is modeled as:
i
n
iii0i wXu
1
Each household’s market access is captured by four variables, that are assumed to influence productivity only through market participation:
Distance to the nearest tarred road District has tarred road District on international border District on the sea coast
Instruments for Market Participation
Result I:
Market Participation and Market
Access
Greater market access, as measured by local infrastructure, is associated with:
(1) a higher probability of market participation
(2) a larger percent sold among participantsThe thesis also includes results on market access
and input purchase, not reported here
Hypotheses
Heckman Regressions for Market Participation
Note: All models include controls for crop grown and agro-ecological region.
Whole Panel 2002 2005Percent
SoldPartici-pation
Percent Sold
Partici-pation
Percent Sold
Partici-pation
Distance to tarred road 0.014* 0.001 0.016 0.001 0.016 0.001 Tarred road in district 3.740*** 0.104* 4.891** 0.091 2.983 0.166* District is Coastal 0.416 -0.163** -1.220 -0.172* 2.126 -0.129 District on borders -1.761 -0.067 3.182* 0.033 -6.227*** -0.166* Log of output/ha -0.857 0.422*** 6.998*** 0.540*** -2.521* 0.377***Dependents per ha -0.582* -0.049*** -1.157** -0.068*** -0.602 -0.028 Workers per ha -0.833 -0.128*** -2.738*** -0.130*** -0.342 -0.140***Education of head 0.245* -0.005 0.069 -0.007 0.671** 0.002 Number of crops 0.575 0.103*** 0.571 0.057* 0.628 0.170***Year 2005 8.261*** 0.103**Sex of head 0.140*** 0.021 0.229***Age of head -0.005*** -0.007*** -0.002 Own cattle 0.002 0.108 -0.114 Constant 22.360* -3.296*** -53.103** -4.050*** 48.381*** -2.99***Chi Square 2701.452 1059.882 1362.115 No. of Cases 7139 3667 3472
Link from market access to market participation is significant but varies by year◦ tarred roads have a significant effect on both
participation and percent sold◦ output level and household composition also matter◦ number of crops grown affects mainly participation
Control variables (not shown) are important◦ Households market cash crops much more than food◦ Agro-ecological zones differ in both probability of
participation and percent sold
Result I: Summary of Results
Result II:Market Participation
and Efficiency
Does market participation influence productivity? We can use land and labor to estimate Production
Possibility FrontiersEach farmer’s distance to the frontier is their relative
efficiencyAre market participants closer to the frontier?
But higher-output farmers are more likely to be market participants, so we must instrument market participation using market access.
Result II: Motivation
Estimate a translog production function using labor and land as inputs
Estimate inefficiency coefficients using the SPF Develop productivity model using efficiency
coefficients as dependent variable
Result II: Methodology
◦ From previous Market Participation models predict percent sold
◦ Use predicted percent sold as one of the regressors of efficiency
◦ Include all other explanatory variables in Percent Sold models except market access
Result II:Methodology (cont’d)
The Translog Production Frontier
Panel Model 2002 2005
Log of area 0.559*** 0.569*** 0.578***
Log of adults 0.441*** 0.431*** 0.422***
Log of area squared -0.124*** -0.036 -0.211***
Log of adults squared -0.347*** -0.299*** -0.356***
Log of area×log of labor 0.171*** 0.091 0.254***
Year 2005 0.145***
Constant 8.003*** 8.020*** 8.054***
Efficiency Relative to the Frontier
Note: All models include controls for crop grown and agro-ecological region.
Panel Model
2002 2005
Log of area -1.096*** -0.949*** -1.143***Log of adults 1.170*** 1.083*** 1.375***Own cattle 0.019 -0.042 0.005Sex of head -0.025 -0.080 0.027Age of head -0.003 -0.003 -0.005Education of head -0.168*** -0.151*** -0.202***Number of crops produced -0.443*** -0.312*** -0.618***Predicted percent sold 0.575*** 0.503*** 0.654***Year 2005 -4.262***Constant -9.172*** -7.469*** -15.990***Chi Square 1377.278 724.505 627.518Number of Cases 7139 3667 3472
From the SPF, labor and land are complements and show diminishing returns
Farm households closer to the frontier tend to have:◦ smaller land area and more adult members◦ less educated household heads◦ more specialization (fewer crops produced)◦ larger percent sold, when instrumented by market access
Result II: Summary of Results
Conclusions
From the nationally-representative TIA survey of about 3,500 farm households in 2002 and 2005:
Market access (tarred roads) is closely linked to market participation, which in turn is closely linked to farm productivity, when controlling for other factors
Questions?????????????
1) Market participation depends on market access
2) Efficiency depends on market participation… but market participation is endogenous to
productivity, so will be instrumented by market access, following (1)
Hypotheses
Participation may drive productivity◦ e.g. due to incentives, information and working capital
Productivity may drive participation ◦ if low-output farms remain self-sufficient
Participation and productivity are influenced by similar unobserved or immeasurable variables
Using participation to explain productivity fraught with endogeneity
Have market access instrument participation
Topic I:Motivation
This two-stage model is estimated using a quasi-maximum likelihood procedure◦ simultaneously estimates the parameters of the
participation and percentage sold models◦ Market access affects both decisions◦ Three models developed (Panel, 2002 and 2005)
Topic I:Methodology
Heckman Regressions for Market Participation
Whole Panel 2002 2005Percent
SoldParticipatio
nPercent
SoldParticipati
onPercent
SoldParticipation
Distance to tarred road 0.014* 0.001 0.016 0.001 0.016 0.001 Tarred road in district 3.740*** 0.104* 4.891** 0.091 2.983 0.166* District is Coastal 0.416 -0.163** -1.220 -0.172* 2.126 -0.129 District on borders -1.761 -0.067 3.182* 0.033 -6.227*** -0.166* Log of output/ha -0.857 0.422*** 6.998*** 0.540*** -2.521* 0.377***Dependents per ha -0.582* -0.049*** -1.157** -0.068*** -0.602 -0.028 Education of head 0.245* -0.005 0.069 -0.007 0.671** 0.002 Workers per ha -0.833 -0.128*** -2.738*** -0.130*** -0.342 -0.140***Number of crops 0.575 0.103*** 0.571 0.057* 0.628 0.170***Year 2005 8.261*** 0.103**Sex of head 0.140*** 0.021 0.229***Age of head -0.005*** -0.007*** -0.002 Own cattle 0.002 0.108 -0.114 Constant 22.360* -3.296*** -53.103** -4.050*** 48.381*** -2.99***Chi Square 2701.452 1059.882 1362.115 No. of Cases 7139 3667 3472
Note: All models include controls for crop grown and agro-ecological region.
Translog Production Function Panel Model 2002 2005
Log of area 0.559*** 0.569*** 0.578***
Log of adults 0.441*** 0.431*** 0.422***
Log of area squared -0.124*** -0.036 -0.211***
Log of adults squared -0.347*** -0.299*** -0.356***
Log of area×log of labor 0.171*** 0.091 0.254***
Year 2005 0.145***
Constant 8.003*** 8.020*** 8.054***
Inefficiency Model
Note: All models include controls for crop grown and agro-ecological region.
Panel Model
2002 2005
Log of area -1.096*** -0.949*** -1.143***Log of adults 1.170*** 1.083*** 1.375***Own cattle 0.019 -0.042 0.005Sex of head -0.025 -0.080 0.027Age of head -0.003 -0.003 -0.005Education of head -0.168*** -0.151*** -0.202***Number of crops produced -0.443*** -0.312*** -0.618***Estimated percent sold 0.575*** 0.503*** 0.654***Year 2005 -4.262***Constant -9.172*** -7.469*** -15.990***Chi Square 1377.278 724.505 627.518Number of Cases 7139 3667 3472