Mobile Phones and Farmers’ Welfare in Niger
Jenny C. Aker and Marcel Fafchamps
Center for the Study of African Economies (CSAE) Conference March 22, 2010
Motivation
Costly information can make it difficult for market agents to engage in optimal arbitrage
Excess price dispersion for homogeneous goods is a common occurrence in developed and developing countries (Stigler, JPE 1961, Brown and Goolsbee, JPE 2002, Jensen, QJE 2007)
Motivation
Motivation
Motivation Alankoss 10 km~2 hours
15 km~3 hour
Bakin Birgi (Monday) Hawkin Sara
Motivation Alankoss
10 km~2 min
15 km~2 minutes 6
Zinder (Sunday)
Bakin Birgi (Monday) Hawkin Sara
Motivation “[With a mobile phone], no dishonest trader can cheat me when I buy and sell….” Farmer in Maradi, Niger
Motivation Goal: Assess the impact of mobile phones on farm-gate price dispersion in Niger Exploit
the quasi-experimental rollout of mobile phone
towers Investigate the impact upon farm-gate price levels
Three Datasets Market-level
time series (monthly) farm-gate price panel
1999-2008 Mobile phone tower rollout between 1999 and 2008 Unique farmer panel collected between 2005-2007
Preview of Findings The introduction of mobile phones leads to a decrease in farm-gate price dispersion The
effect is stronger for cowpeas as compared to millet The effect is stronger for markets that are in closer proximity, especially for millet
Initial evidence suggests that this has not resulted in an increase in farm-gate price levels
Mobile Phone Rollout Between 2001-2008, cell phone towers were phased-in throughout the country Mobile phone companies (Celtel/Zain, Sahelcom, Telecel) intended to provide universal coverage by 2009 There were two criteria to prioritize the rollout: o o
Whether the town was an urban center Whether the town was located near a border (Benin, Burkina Faso, Mali and Nigeria)
Widespread coverage into rural areas between 2009 and 2010
Mobile Phone Coverage by Market and Year, 2001-2008
Number of Mobile Phone Subscribers and Landlines in Niger, 2001-08 600000 550000
Number of Landlines
Number of cell phone subscribers
500000 450000 400000
584,286
Number
350000 300000 250000 200000 150000 100000 50000 0 2001/2002
2002/2003
2003/2004
2004/2005
2005/2006
2006/2007
2007/2008
How will Mobile Phones affect Farmers’ Behavior?
Direct channels (Aker and Mbiti 2010, Jensen 2010) o Reduce search costs and improve arbitrage o Reduce monsopsonistic market power
Indirect channels o Increased supply o Reduced transportation o Reduced price variability
How will Mobile Phones affect Farmers’ Behavior?
Direct channels (Aker and Mbiti 2010, Jensen 2010) o Reduce search costs and improve arbitrage o Reduce monopsonistic market power
Indirect channels o Increased supply o Reduced transportation o Reduced price variability
Linking the Model to the Data Mobile phones reduce the per-search cost as compared to personal travel Most farmers live within 10 km of their nearest principal market o 35 percent reduction in farmers’ (marginal) search costs Therefore, the introduction of mobile phones will: • #1. Increase farmers’ reservation prices (unobserved) • #2. Increase the number of markets over which farmers’ search • #3. Reduce farm-gate price dispersion across markets
Linking the Model to the Data Mobile phones reduce the per-search cost as compared to personal travel Most farmers live within 10 km of their nearest principal market o 35 percent reduction in farmers’ (marginal) search costs Therefore, the introduction of mobile phones will: • #1. Increase farmers’ reservation prices (unobserved) • #2. Increase the number of markets over which farmers’ search • #3. Reduce farm-gate price dispersion across markets
Related Literature on the Impact of Mobile Phones in Agriculture • Fisheries in India (Abraham 2007, Jensen 2007) • Consumer price dispersion and traders’ behavior in Niger (Aker 2008) • Farmer participation in Uganda (Muto and Yamano 2009) • Internet kiosks and soybean prices in India (Goyal 2009)
First Dataset: Market-Level Panel
Monthly millet and cowpea prices in 42 domestic and cross-border markets State-level rainfall and agricultural production Monthly gasoline prices Estimated transport costs between markets State-level population and urban status Road distances, road quality and estimated travel Criteria used by mobile phone companies for cell phone rollout Date of mobile phone entry in each market
Second Dataset: Farmer Panel Panel survey of traders and farmers collected between 2005-2007 395 traders and 205 farmers across 35 markets in 6 regions of Niger Census of grain markets and grain traders on each market Detailed data on farmers’ operations in 2005/2006 and 2006/2007, with retrospective questions for 2004/2005
Table 1. Description of Key Variables: Grain Trader and Market Baseline Characteristics Variable Name Sample Mean (s.d.) Panel A: Trader-Level Characteristics Socio-Demographic Characteristics Ethnicity Hausa Zarma Other Age
0.65 0.17 0.18 45.71(12.2)
Gender(male=0, female=1)
0.11(.32)
Education (0=elementary or above, 1=no education)
0.62(.48)
Trader type Wholesaler Semi-wholesaler Intermediary Retailer Years' of Experience
0.17 0.15 0.15 0.53 16.0(10.2)
Commercial Characteristics Engage in trading activities all year round
.94(.22)
Trade in agricultural output products only Engage in activities outside of trade Co-ownership of commerce
0.98(.02) 0.92(.28) .19(.40)
Changed "principal market" since he/she became a trader
.10(.31)
Number of markets where trade goods
4.42(2.84)
Number of markets where follow prices
3.87(3.0)
Number of days of storage
7.14( 9.8)
Own cell phone
.29(.45)
Own means of transport (donkey cart, light transport)
.11(.32)
Table 2. Description of Key Variables: Farmers Variable Name
Sample Mean (s.d.)
Panel A: Farmer-Level Characteristics Socio-Demographic Characteristics Household head
.915(.279)
Member of hausa ethnic group
.675(.469)
Age
49(16)
Gender(male=0, female=1)
.01(.09)
Education (0=elementary or above, 1=no education)
.85(.35)
Household size
12.6(7.92)
Own mobile phone Panel B. Agricultural Marketing Activities Sold millet in the past year
0.25
Sold cowpea in the past year
0.56
Purchased millet since the previous harvast
0.91
Number of hours walking to principal market
1.53
Access to a paved road
.269(.444)
Number of purchase and sales markets
1.46(.670)
Member of a producers' association
0.22
Sold to intermediary since the last harvest
0.45
Bought agricultural products on credit in the past year
0.41
Received payment in advance for harvest
0.16
Responsible for transport if sell product
0.64
Low Infrastructure Investment
Empirical Strategy Assess the impact of the introduction of mobile phones on agricultural price dispersion across markets “Treatment” defined as a mobile phone tower, Use market-level time-series panel dataset
not adoption
Exploit the quasi-experimental nature of the rollout of mobile phone towers Pooled difference-in-differences estimation Measure treatment effect heterogeneity over
time and space
Estimating the Impact of Mobile Phones at the Market Level Y jki ,t = β 0 + β1mobile jk ,t + X 'jk ,t γ + α jk + θ t + µ jk ,t absolute value of the log farm-gate price difference between market i and market j at time t* Mobile jk,t variable =1 if the market pair received cell phones in period t, 0 otherwise Mobileon ejk,t variable=1 if one market in pair receive mobile phone coverage, 0 otherwise Transportijt per unit/per km fuel * distance Droughtijt markets i and j at time t θt time effects (monthly or yearly) aij market-pair specific effects u ijt error with 0 conditional mean, E [u ijt|mobile ijt,Xijt, a ij, θt]=0 t time in months, t=1…111 N number of market pairs Yjk,t
Xij,t
*Alternative measures of price dispersion (correlation, covariance) and the treatment variable are also used **Include market-specific fixed effects and cluster by month - -in future will also correct for serial autocorrelation
Effects of Mobile Phones: Cowpeas Table 3. Impact of Mobile Phones on Farm-Gate Price Dispersion for Cowpea Dependent variable: |ln(Pit-)-ln(Pjt)| Mobile coverage both markets Mobile coverage one market
(1) (2) (3) -0.022*** -0.018*** -0.096*** (-1.704) (-5.05) (-18.88) -.011 (1.143)
-0.012*** (-3.02)
-0.030 (-6.92)
(4) -0.020* (-1.86)
(5) -0.026*** (-2.94)
-0.012 (-1.196)
0.002 (.265)
.0413*** (6.021)
Distance between markets
0.027*** (10.8)
Distance*mobile two markets
0.0093*** 0.011*** (2.072) (11.76)
0.010*** (11.36)
0.011 (1.070)
0.005 (.802)
Distance*mobile one market
0.0172*** 0.009*** (4.095) (8.812)
0.007*** (6.82)
0.011 (1.20)
-0.010 (-1.40)
0.131 (17.564)
0.217*** (170)
0.326*** (44)
0.124*** (15.83)
0.197*** (17.5)
Other covariates
Yes
Yes
Yes
Yes
Yes
Monthly fixed effects
No
No
Yes
No
Yes
Market pair fixed effects
No
Yes
Yes
No
Yes
Number of observations
41,070
41,070
41,070
13,646
13,646
Constant
Effects of Mobile Phones: Millet Table 4. Impact of Mobile Phones on Farm-Gate Price Dispersion for Millet Dependent variable: |ln(Pit-)-ln(Pjt)| Mobile coverage both markets Mobile coverage one market
(1) -0.022* (-1.70) -0.010 (-1.14)
(2) 0.000 (.016)
(3) 0.008** (1.98)
-0.019*** -0.017*** (-6.35) (5.18)
(4) -0.03*** (-3.16)
(5) -0.025*** (-4.17)
-0.008 (-.818)
-0.010 (1.56)
Distance between markets
.027*** (10.839)
0.046*** (8.76)
Distance*mobile two markets
0.009 (2.072)
-0.003*** (-4.25)
Distance*mobile one market
.017*** (4.09)
0.003*** (4.67)
Other covariates
Yes
Yes
Yes
Yes
Yes
Monthly fixed effects
No
No
Yes
No
Yes
Market pair fixed effects
No
Yes
Yes
No
Yes
Constant
.120*** (21.9)
.163*** (182)
.169*** (28.4)
.123*** (91)
.142*** (15.6)
Number of observations
41,070
41,070
41,070
13,646
13,646
-0.001 (-1.48)
.018*** (3.20)
0.004*** -0.021*** (5.59) (-2.71)
0.017*** (3.20) -0.006 (-.925)
Mobile Phones, Producer Price Dispersion and Distance Mobileij=0 |ln(Pit)-ln(Pjt) Mobileij=1
Distanceij
Threats to Identification Selection bias Hidden bias (conditional independence assumption) Collusive behavior and entry and exit
Balance of Pre-Treatment Variables Table 5. Comparison of Observables by Mobile Phone and non-Mobile Phone Groups (1999-2001) Difference in Difference in Unconditional Mean Means Distributions
Pre-Treatment Observables
Mobile Phone
No Mobile Phone
Unconditional
Unconditional Kolmogorov-Smirnov Test
Mean (s.d.)
Obs
Mean (s.d.)
Obs
s.e.
D-statistic
p-value
|Pit-Pjt| of Millet Producer Prices (CFA/kg)
15.04(16.05)
4566
15.99(14.8)
3044
-.949(.787)
0.0672***
0
|Pit-Pjt| of Cowpea Producer Prices (CFA/kg)
29.46(28.5)
4566
29.57(27.44)
3044
-.115(1.452)
0.0423***
0.006
Distance between markets (km)
438.75(275)
561
413.86(247)
105
24.89(26.87)
0.0647
0.852
Road Quality between markets
.338(.47)
561
.390(.49)
105
-.052(.052)
0.0518
0.972
Drought in 1999 or 2000
.050(.22)
13464
.052(.22)
2520
-.002(.008)
0.002
1
Urban center(>=35,000)
.346(.476)
561
.305(.46)
105
.041(.049)
0.041
0.998
Transport Costs between Markets (CFA/kg)
12.57(7.3)
13464
11.91(6.6)
2520
.656(.708)
0.0513
0
Panel A. Market Pair Level Data
Balance of Pre-Treatment Variables Table 5. Comparison of Observables by Mobile Phone and non-Mobile Phone Groups (1999-2001) Unconditional Mean
Pre-Treatment Observables
Mobile Phone
No Mobile Phone
Difference in Means
Difference in Distributions
Unconditional
Unconditional Kolmogorov-Smirnov Test
Mean (s.d.)
Obs
Mean (s.d.)
Obs
s.e.
Dstatistic
p-value
99.47(27.8)
327
98.62(31.5)
93
.849(4.91)
0.1579*
0.054
150.91(43.5 9)
327
139.69(39.75)
93
11.22(7.32)
0.2324*
0.063
.588(.49)
34
.4(.49)
5
.188(.24)
0.1882
0.988
Market Size
88.58(80)
34
124(77)
5
-35.41(41.3)
0.4853
0.368
Drought in 1999 or 2000
.052(.22)
816
.025(.16)
120
.027(.026)
0.0277
1
Landline service 1999-2001 (1=Yes, 0=No)
.865(.34)
816
.5(.51)
120
.364(.362)
0.3649
0.962
Urban center(>=35,000)
.382(.486)
34
0(.00)
5
.382***(.084)
0.3824
0.55
Panel B. Market Level Data Millet Producer Price
Cowpea Producer Price Road Quality to Market (1=Paved, 0=Unpaved)
Tests of Conditional Independence Table 6. Tests of the Conditional Independence Assumption Dependent Variable: Price Dispersion in 1999-2001 (Pre-Treatment) Cowpea
Millet
Coeff(s.e.) Coeff(s.e.)
Unconditional difference in means
-.116 (1.45)
-.948 (.788)
Conditional difference in means
-.124 (1.43)
-.890 (.793)
Estimating the Impact of Mobile Phones on Farmers’ Welfare ln( priceijt ) = β 0 + β1mobile jt + X 'jt γ + α j + θ t + µ jt
Ln(price) j,t Mobile jt Xijt θt
u ijt t N
log of farm-gate price of commodity i in market j at time t variable =1 if the market received mobile phone coverage in period t, 0 otherwise vector of exogenous regressors of market j at time t time effects (year) error with 0 conditional mean, E[uijt|Zijt, Xijt, ai, aj, θt]=0 t=0,…11 number of markets, N=42
Table 7. Impact of Mobile Phones on Farm-Gate Price Levels for Cowpea
Dependent variable: ln(Pit)
(1)
(2)
(3)
.036 (.029)
.001 (.0178)
.036 (.030)
Lagged dependent variable
No
No
No
Other covariates
Yes
Yes
Yes
Monthly fixed effects
Yes
Yes
Yes
Market fixed effects
No
Yes
Yes
Distance to Nigerian border
No
No
Yes
41,070
41,070
41,070
Mobile coverage both markets
Constant Number of observations
Welfare Estimates Less than 10 percent of villages had mobile phone access in 2007/2008 Mobile phone coverage and adoption increased during the period after this study (2008-2010) While no evidence of collusion in consumer prices, perhaps market power differs in consumer and producer markets