1. A Market for Lemons:
Maize in Kenya
Vivian Hoffmann
Agricultural and Resource Economics
University of Maryland
Samuel Mutiga Michael Milgroom Rebecca Nelson
Plant and Microbe-Biology
Cornell University
Jagger Harvey
Biosciences East and Central Africa -
International Livestock Research Institute Hub
With thanks to the Atkinson Center for a Sustainable Future
2. The Market for Lemons
“Business in underdeveloped countries is difficult”
- Akerlof (1970)
• When sellers have information about quality of
goods not available to buyers:
– Quality is not reflected in the price
– Low quality goods are traded, high quality retained
– Volume of trade is reduced
• Institutions (guarantees, branding) may arise to
solve this problem
3. Previous empirical tests
• Used vehicles
Bond, 1982, 1984; Pratt and Hoffer, 1984; Lacko,
1986; Genesove, 1993; Sultan, 2008; Emons &
Sheldon, 2009
• Workers
Gibbons and Katz, 1991
• Slaves
Greenwald & Glasspiegel, 1983; Pritchett &
Chamberlain, 1993
• Cattle
Anagol, 2011
• We apply the lemons model to a developing country food
market
4. Maize in Kenya
• Main staple food: estimated 400 grams / day
per capita consumption (Muriuki and
Siboe, 1995)
• Grown by 92% of farm households (KIHBS)
• Not well suited to growing conditions
– Vulnerable to drought
– Other crops generally more profitable (Tegemeo
Institute, 1996)
– Prone to toxic fungal contamination
5. “It’s not the same as maize from my farm”
– respondent who refused to sell sample
Maize for sale in typical roadside market Less typical maize shop in Western Kenya
7. Motivating welfare concerns
Human health
• Absence of price incentive for quality leads to
underinvestment in good storage practices /
over-supply of toxic contamination
Efficiency
• Poor quality of food available for purchase could
lead to self-provisioning, inhibit specialization in
more profitable activities (among other factors
including price risk, transaction costs)
8. Aflatoxin
• Produced by fungus Aspergillus flavus aflatoxin
• Can colonize crop in field or post-harvest
• Pre-harvest risk factors: drought, heat stress, pest attack
• Post-harvest: high moisture content during storage
Health effects (Strosnider et al. 2006):
• One of the most potent known carcinogens
• Acute exposure: liver failure, death
• Chronic exposure: liver cancer, suppressed immune
response, growth faltering in children
Observability:
• A. flavus not always visible; other, non-toxigenic molds may be
visible visible attributes poor proxy
• Mold generally tastes bad aflatoxin likely correlated with taste
(also a noisy signal)
9. This is your chicken
This is your chicken on aflatoxin
10. Necessary condition for
lemons market: asymmetric information
• Farmers have private information about the
quality of food in their possession
– Knowledge of growing and storage conditions
(own-produced only)
– Observe taste (own produced and purchased)
– This information is not available to buyers
11. Necessary condition for over-supply of
contaminant:
• Farmers (and / or others in value chain) can
influence the dimension of quality on which
information is asymmetric
12. A model of food marketing behavior
Utility is derived from food quality
and consumption of numeraire
good
Budget constraint: Income
generated by sale of food, cash crop
Exogenous food requirement
Food quality identity
Cash crop production function
Food crop production function
Food crop quality function
15. Testable predictions
1. The quality of retained food is increasing in
the quantity of maize harvested, controlling for
the quality of harvested food
16.
17. 3. The proportion of food consumed that is self-
produced is increasing household wealth.
18. Data: Posho mill survey
• When need flour, take kernels to small-scale
mill (posho mill)
• Interviewed clients & collected maize samples
19. Data: Posho mill survey
• 176 mills in 138 market centers
• Western, Rift Valley, Nyanza Provinces: 2009
• Eastern Province: 2010
• Diversity of agro-ecological zones
• Collected samples and survey data from at least
10 individuals per mill
• Maize samples sent to Nairobi for laboratory
analysis as collected
• Observations for which both survey & lab data
available: 2124
20. Respondent characteristics &
maize sorting practices
Nationally Representative Data
Sample
(means)
Urban Rural Overall Mean N
demographic & assets
female 0.75 2082
age 36.3 2088
completed primary 0.83 0.56 0.63 0.70 2077
completed secondary 0.61 0.28 0.35 0.19 2077
house: electricity 0.58 0.07 0.19 0.13 1441
house: permanent roof 0.92 0.78 0.81 0.95 1429
house: permanent walls 0.88 0.51 0.60 0.52 1369
own cell phone 0.86 0.53 0.62 0.81 1441
sorting practices
sort at miller? 0.66 2107
sorted for health 0.68 1833
sorted for taste 0.25 1833
21. Maize origin, use, and observable
properties
Mean N
maize origin
own farm 0.52 1983
posho miller 0.04 1983
purchased elsewhere 0.38 1983
gift 0.06 1983
food aid 0.00 1983
intended use
household food 0.73 2124
brewing 0.23 2124
livestock feed 0.01 2124
sell 0.03 2124
maize characteristics
1-10% broken kernels 0.28 1019
> 10% 0.05 1019
1-10% discolored 0.32 1019
> 10% 0.08 1019
price per kg if purchased (KSH) 14.1 402
22. Analysis of maize samples
• Enzyme-linked immunosorbent assay (ELISA)
for aflatoxin contamination
• Test is sensitive up to 20 ppb – above
this, sample must be diluted, precision is lost
• Key thresholds:
– 10 ppb (Kenyan regulatory standard)
– 20 ppb (test limit and FDA standard)
23. Aflatoxin by region
Vertical lines indicate limit of test accuracy (20 parts per billion) and Kenyan
regulatory standard (10 parts per billion).
24. Empirical Approach
1) Test assumption of asymmetric information
2) Test predictions of the model
3) Investigate impact of farmer practices on
quality
25. 1) Test assumption of asymmetric
information
• Qualities observable to buyers should affect price
• Qualities observable to owners should affect use
• Are there any attributes that do not affect
price, but do affect use?
26. Effect of maize characteristics on price
Observable Unobservable Both
1-10% discolored -1.077** -1.102**
(0.502) (0.514)
>10% discolored -1.547** -1.592**
(0.736) (0.733)
1-10% broken -0.030 -0.119
(0.382) (0.395)
>10% broken -1.959* -2.374*
(1.112) (1.259)
0 < ppb afla < 10 -0.426 0.042
(0.379) (0.394)
10-20 ppb afla -0.630 0.237
(0.573) (0.604)
afla ppb > 20 -0.282 0.989
(0.588) (0.969)
Constant 14.540*** 14.388*** 14.324***
(0.208) (0.303) (0.345)
Observations 294 390 294
Communities 94 94 94
R-squared 0.044 0.006 0.057
Notes: Linear model with community fixed effects, clustered errors
shown in parentheses. Data are from the Eastern sample only. *
p<0.10, ** p<0.05, *** p<0.01
27. To what extent do observables predict
unobservables?
ppb = 0 0 > ppb > 10 10 > ppb > 20 ppb > 20
1-10% broken kernels -0.043 -0.004 0.015 0.032
(0.035) (0.005) (0.012) (0.027)
>10% broken kernels -0.139*** -0.041 0.043*** 0.137*
(0.050) (0.033) (0.012) (0.071)
1-10% discolored kernels 0.026 0.001 -0.009 -0.018
(0.039) (0.002) (0.014) (0.026)
>10% discolored kernels -0.027 -0.003 0.010 0.020
(0.051) (0.007) (0.018) (0.040)
Proportion with no discolored
0.331 0.317 0.176 0.176
or broken in aflatoxin category
Observations 1014
Communities 107
Pseudo R-squared 0.067
Notes: Marginal effects on the likelihood of observing each outcome, derived from an
ordered logit regression, with standard errors clustered at the community level. Data are
from Eastern Province only. * p<0.10, ** p<0.05, *** p<0.01
28. Aflatoxin by intended use
25th 75th 95th
mean min percentile median percentile percentile max N
Food for HH 28.7 0.0 0.0 1.4 9.3 83.6 4839.3 1492
Brewing 41.5 0.0 0.0 7.2 18.9 172.4 1658.1 494
Livestock Feed 48.1 0.0 1.1 2.8 17.6 201.3 288.8 19
Sale 64.6 0.0 0.0 7.0 33.9 476.3 806.7 69
29. CDF of contamination by use
1
.8
.6
.4
.2
0
0 20 40 60 80 100
aflatoxin
consume brewing
livestock sell
31. 2) Test predictions of the model
• Impact of quantity grown on quality of
retained maize
• Compare quality of purchased vs. retained
maize
• Is self-produced maize a normal good?
32. Effect of harvest quality (yield) &
quantity (area planted) on aflatoxin
contamination of retained maize
ppb = 0 0 <ppb < 10 10 < ppb < 20 ppb > 20
(1) (2) (3) (4)
Quality and quantity of harvest
Hectares under maize 0.028*** 0.001 -0.009*** -0.020***
(0.009) (0.002) (0.003) (0.006)
Yield (100 kg / ha) 0.001** 0.000 -0.000** -0.001**
(0.001) (0.000) (0.000) (0.000)
Observations 761
Communities 106
(Pseudo) R-squared 0.228
Notes: Marginal effects from an ordered probit regression, with clustered standard errors
shown in parentheses. Controls for post-harvest practices not shown. * p<0.10, **
p<0.05, *** p<0.01
36. Alternative explanation
Perhaps those who purchase are less concerned
with bad taste / toxic contamination, purchase
even though fully informed
– Compare quality-usage patterns of own vs.
purchased maize
37. Quality-use relationship similar for own-grown
and purchased maize
Panel A: Maize grown on own farm
ppb = 0 0 > ppb > 10 10 > ppb > 20 ppb > 20
N 296 262 84 99
Food for HH
% 77.9 76.8 60.9 57.6
N 68 66 47 51
Brewing
% 17.9 19.4 34.1 29.7
N 2 4 3 4
Livestock Feed
% 0.5 1.2 2.2 2.3
N 14 9 4 18
Sale
% 3.7 2.6 2.9 10.5
% total 100 100 100 100
Panel B: Purchased maize
ppb = 0 0 > ppb > 10 10 > ppb > 20 ppb > 20
N 241 206 75 46
Food for HH
% 86.1 79.2 66.4 46.5
N 35 45 36 47
Brewing
% 12.5 17.3 31.9 47.5
N 1 4 1 1
Livestock Feed
% 0.4 1.5 0.9 1
N 3 5 1 5
Sale
% 1.1 1.9 0.9 5.1
% total 100 100 100 100
38. Quality-use relationship similar for own-grown
and purchased maize
Panel A: Maize grown on own farm
ppb = 0 0 > ppb > 10 10 > ppb > 20 ppb > 20
N 296 262 84 99
Food for HH
% 77.9 76.8 60.9 57.6
N 68 66 47 51
Brewing
% 17.9 19.4 34.1 29.7
N 2 4 3 4
Livestock Feed
% 0.5 1.2 2.2 2.3
N 14 9 4 18
Sale
% 3.7 2.6 2.9 10.5
% total 100 100 100 100
Panel B: Purchased maize
ppb = 0 0 > ppb > 10 10 > ppb > 20 ppb > 20
N 241 206 75 46
Food for HH
% 86.1 79.2 66.4 46.5
N 35 45 36 47
Brewing
% 12.5 17.3 31.9 47.5
N 1 4 1 1
Livestock Feed
% 0.4 1.5 0.9 1
N 3 5 1 5
Sale
% 1.1 1.9 0.9 5.1
% total 100 100 100 100
39. Is self-produced maize a normal good?
(REPEAT data)
Harvested
Consumed maize produced
maize
on own farm past 7 days
(100 kg)
(1) (2) (3)
Log asset value 2.502*** 0.025** 0.021*
(0.479) (0.011) (0.012)
Agricultural land (ha) 1.873** 0.013 0.013
(0.892) (0.009) (0.009)
Land squared (ha sqr) -0.034 -0.000 -0.000
(0.043) (0.000) (0.000)
Harvested maize (100kg) 0.001
(0.001)
Obsevations 679 695 677
Communities 86 86 86
R squared 0.144 0.017 0.018
Notes: Results are from linear regressions with community fixed effects.
Standard errors, shown in parentheses, are clustered at the community
level . * p<0.10, ** p<0.05, *** p<0.01
41. Determinants of visible mold & afla
Aflatoxin Discolored kernels
ppb = 0 0 <ppb < 10 10 < ppb < 20 ppb > 20 0 kernels 0-10% kernels >10% kernels
(1) (2) (3) (4) (5) (6) (7)
Quality and quantity of harvest
Hectares under maize 0.056** 0.007 -0.017** -0.046** 0.026 -0.018 -0.007
(0.025) (0.005) (0.008) (0.022) (0.026) (0.019) (0.007)
Yield (100 kg / ha) 0.002** 0.000 -0.000** -0.001** 0.001 -0.001 -0.000
(0.001) (0.000) (0.000) (0.001) (0.001) (0.001) (0.000)
Post-harvest practices
Dry maize in field 0.042 0.007 -0.012 -0.037 -0.295** 0.244** 0.051**
(0.132) (0.005) (0.035) (0.128) (0.126) (0.114) (0.024)
Improved drying 0.020 0.000 -0.006 -0.016 -0.236** 0.175** 0.061**
(0.069) (0.000) (0.020) (0.058) (0.092) (0.071) (0.031)
Months since harvest -0.048*** 0.067 0.014*** 0.040*** -0.022 0.016 0.006
(0.013) (0.045) (0.005) (0.010) (0.020) (0.014) (0.006)
Observations 361 361
Communities 74 74
(Pseudo) R-squared 0.225 0.144
Notes: Marginal effects calculated from ordered probit regressions of the categorical proportion of discolored
kernels (1 through 3), and aflatoxin contamination (4 through 7). Yellow shading indicates differences
between coefficient values across models at p<0.1. Standard errors, shown in parentheses, are clustered at the
community level for all models and tests. * p<0.10, ** p<0.05, *** p<0.01
42. Aflatoxin and farmer incentives
• Visibly moldy maize is less likely to be sold:
farmers appear to respond to price incentives
• Impact of farm practices on visible molds and
aflatoxin are not well correlated
– Inhibits learning about best practices, implies weak
incentives for aflatoxin control
– Technologies to reduce aflatoxin unlikely to be
profitable unless asymmetric information problem
can be solved
43. Overcoming the information
asymmetry: prospects & implications
• Technologies in development for rapid and
affordable screening, e.g. “e-nose”
• Efforts to develop AflaSafe certification
– Would better align farmer, trader incentives with
public health goals
– May also encourage more profitable specialization
– Could adversely impact health of poorest
consumers
44. Summary of Key Findings
• Evidence consistent with asymmetric information
– Aflatoxin (or correlate) determines how maize is used
observed by maize holders
– No impact on market price not observed by buyers
• Evidence consistent with lemons market
– Controlling for quality of harvest (yield), quality of
retained maize is increasing in harvested amount
(area planted)
– Quality of sold and purchased maize is lower than
quality of retained maize
– Consumption of self-produced maize is increasing in
household wealth