Gender Methods Seminar, Dec 13, 2016
Berber Kramer, Research Fellow, Markets, Trade, and Institutions Division (IFPRI)
Abstract:
This paper analyzes implications of cash constraints for collective marketing, using the case of the Kenyan dairy sector. Collective marketing through for instance cooperatives can improve smallholder farmer income but relies on informal, non-enforceable agreements to sell outputs collectively. Sideselling of output in the local market occurs frequently and is typically attributed to price differences between the market and cooperative. This paper provides an alternative explanation, namely that farmers sell in the local market when they are cash-constrained, since cooperatives defer payments while buyers in local markets pay cash immediately. Building on semi-parametric estimation techniques for panel data, we find robust evidence of this theory. High-frequency high-detail panel data show that farmers sell more in the local market, in particular to buyers who pay cash immediately, in weeks with low cash at hand. Moreover, households cope with health shocks by selling more milk in the local market and less to the cooperative, but only in weeks they are not covered by health insurance. Effects are concentrated among female dairy farmers. For them, increased flexibility in payment and the provision of insurance through agricultural cooperatives can potentially reduce side-selling and improve the performance of collective marketing arrangements.
Screencast available here:
https://www.dropbox.com/s/d48bte3yzsd5iwz/2016-12-13%2012.03%2012_13%2C%207AB%2C%2012-1pm%2C%20Gender%20Methods%20Seminar%20with%20Berber%20Kramer%20.wmv?dl=0
[IFPRI Gender Methods Seminar] Liquid milk: Cash Constraints and the Timing of Income
1. Liquid Milk: Cash Constraints and the Timing of Income
Xin Geng, Berber Kramer and Wendy Janssens
IFPRI Gender Methods Brown Bag Seminar, December 13, 2016
Geng, Kramer and Janssens (2016) Liquid Milk 1 / 37
2. Background and Motivation
Financial planning is difficult, especially when facing cash constraints,
unpredictable incomes and expenditures (Collins et al., 2009)
Rural women affected most (Demirg¨uc¸-Kunt and Klapper, 2012)
Cash constraints affect intertemporal allocations of experimental gifts
(Dean and Sautmann, 2016; Janssens et al., 2016; Carvalho et al., 2016)
Do cash constraints affect preferences over timing of ‘real’ income?
We address this question by studying where farmers sell agricultural output:
Cooperatives defer payments at potentially higher prices, and provide
extra services (Reardon et al., 2009; Minot and Sawyer, 2014)
Local traders are trusted less to save one’s money
(Casaburi and Macchiavello, 2015)
Geng, Kramer and Janssens (2016) Liquid Milk 2 / 37
3. Preview of the Presentation
Does cash at hand affect the choice where to sell milk?
Market vs. cooperative: Sooner-smaller vs. later-larger trade-off
The share of milk sold to the cooperative increases in cash-at-hand
Corner solutions create treshold effects and nonlinearities
We estimate effects of cash at hand on milk marketing decisions
High-frequency panel data for dairy farmers in Kenya, measuring net
inflows of cash from dairy vs. non-dairy activities
Semiparametric techniques provide parameter-free estimates of how
these two variables affect marketing decisions
We find evidence that the market provides informal insurance:
Farmers often sell milk in the market, despite a lower milk price
They do so especially when they are more cash-constrained
In those weeks, the local market may pay them a higher price
Geng, Kramer and Janssens (2016) Liquid Milk 3 / 37
4. Conceptual Framework: Basic set-up
Every period, a household produces mt and decides how much to sell
outside the cooperative, st, such that it optimizes
max
0≤st ≤mt
∞
t=0
βt
u(ct) (1)
subject to the following budget constraint:
ct = yt + ptst + mt−1 − st−1 (2)
where ct represents (food) consumption and pt the market milk price.
Farmers are paid immediately for milk sold in the market
The cooperative defers payments for mt − st by one period
Non-dairy net income, yt, is assumed to be predetermined
No savings and borrowing outside the cooperative
Geng, Kramer and Janssens (2016) Liquid Milk 4 / 37
5. Conceptual Framework: Predictions
Relatively low market price (p < β): farmers sell all milk to the
cooperative
Increase in cash at hand (yt + mt−1): No effects
(Sufficiently large) decrease: Sell some milk in local market
Relatively high market price (p > β): farmers sell all milk in the
market
Decrease in cash at hand (yt + mt−1): No effects
(Sufficiently large) decrease: Sell some milk to the cooperative
Threshold effects are absent only when p = β
Geng, Kramer and Janssens (2016) Liquid Milk 5 / 37
6. Context: Dairy cooperative
Tanykina Dairies Limited in western Kenya:
Farmer-owned dairy company in the highlands near Eldoret,
operational since 2005, processing approx. 30,000 liters per day
Milk collectors pick up the milk, take it to a nearby center, weigh it,
and farmers receive a fixed price per kg of milk
Seven collection centers in total (we focus on three)
Milk payments deposited the next month in a village bank account
after deducting service and input costs
At baseline, 50% of suppliers have health insurance, monthly premium
deducted from milk payment
Study farmers never deliver to other coops but Tanykina does
compete with traders, vendors and neighbors (local market)
Geng, Kramer and Janssens (2016) Liquid Milk 6 / 37
7. Saving and Credit Cooperative (SACCO)
Geng, Kramer and Janssens (2016) Liquid Milk 7 / 37
10. Data sources
Weekly interviews with 120 Tanykina members from Oct ‘12-Oct ‘13
Individual level: Financial transactions (amount, with whom, how)
Total value of milk sold to Tanykina vs. others (not Q or P)
Non-dairy income, non-food and food expenditures
Data collected weekly at the household level:
Incidence of health problems and insurance coverage
Production and consumption of agricultural output
Only two households dropped out. Sample construction:
Omit last month, Christmas and elections
We focus on weeks in which households sell milk (85%)
Sample with variation over time: 88 households, avg. 34 weeks
Other data sources: Baseline survey and monthly market surveys
Geng, Kramer and Janssens (2016) Liquid Milk 10 / 37
11. Table 1: Household characteristics
Variation in share of No variation in share of
income from Tanykina income from Tanykina
Mean s.e. Mean s.e.
(1) (2) (3) (4)
Household head is male 0.705 0.459 0.500 0.509
Age of the household head 52.38 14.15 51.03 19.13
Number of HH members selling milk 1.489 0.547 1.300 0.466
Number of cows at baseline 4.227 2.509 3.200 1.669
Main dairy farmer:
Is male 0.216 0.414 0.300 0.466
Age 47.57 14.34 46.40 17.38
Is household head 0.477 0.502 0.733 0.450
Is spouse of household head 0.500 0.503 0.200 0.407
Can keep part of cattle income 0.659 0.477 0.793 0.412
Decides how to spend cattle income 0.655 0.478 0.793 0.412
Number of households 88 30
Notes: The sample excludes two households with high attrition and strong outliers. Further, in assessing whether there is variation
over time in the share of income received from Tanykina, we omit weeks with high attrition due to Christmas (2 weeks), elections
(1 week) and the last fieldwork month (4 weeks). The main dairy farmer is the household member reporting the highest value of
dairy income received throughout the year.
Geng, Kramer and Janssens (2016) Liquid Milk 11 / 37
12. Table 1: Household characteristics
Variation in share of No variation in share of
income from Tanykina income from Tanykina
Mean s.e. Mean s.e.
(1) (2) (3) (4)
Household head is male 0.705 0.459 0.500 0.509
Age of the household head 52.38 14.15 51.03 19.13
Number of HH members selling milk 1.489 0.547 1.300 0.466
Number of cows at baseline 4.227 2.509 3.200 1.669
Main dairy farmer:
Is male 0.216 0.414 0.300 0.466
Age 47.57 14.34 46.40 17.38
Is household head 0.477 0.502 0.733 0.450
Is spouse of household head 0.500 0.503 0.200 0.407
Can keep part of cattle income 0.659 0.477 0.793 0.412
Decides how to spend cattle income 0.655 0.478 0.793 0.412
Number of households 88 30
Notes: The sample excludes two households with high attrition and strong outliers. Further, in assessing whether there is variation
over time in the share of income received from Tanykina, we omit weeks with high attrition due to Christmas (2 weeks), elections
(1 week) and the last fieldwork month (4 weeks). The main dairy farmer is the household member reporting the highest value of
dairy income received throughout the year.
Geng, Kramer and Janssens (2016) Liquid Milk 11 / 37
13. Table 1: Household characteristics
Variation in share of No variation in share of
income from Tanykina income from Tanykina
Mean s.e. Mean s.e.
(1) (2) (3) (4)
Household head is male 0.705 0.459 0.500 0.509
Age of the household head 52.38 14.15 51.03 19.13
Number of HH members selling milk 1.489 0.547 1.300 0.466
Number of cows at baseline 4.227 2.509 3.200 1.669
Main dairy farmer:
Is male 0.216 0.414 0.300 0.466
Age 47.57 14.34 46.40 17.38
Is household head 0.477 0.502 0.733 0.450
Is spouse of household head 0.500 0.503 0.200 0.407
Can keep part of cattle income 0.659 0.477 0.793 0.412
Decides how to spend cattle income 0.655 0.478 0.793 0.412
Number of households 88 30
Notes: The sample excludes two households with high attrition and strong outliers. Further, in assessing whether there is variation
over time in the share of income received from Tanykina, we omit weeks with high attrition due to Christmas (2 weeks), elections
(1 week) and the last fieldwork month (4 weeks). The main dairy farmer is the household member reporting the highest value of
dairy income received throughout the year.
Geng, Kramer and Janssens (2016) Liquid Milk 11 / 37
14. Table 1: Household characteristics
Variation in share of No variation in share of
income from Tanykina income from Tanykina
Mean s.e. Mean s.e.
(1) (2) (3) (4)
Household head is male 0.705 0.459 0.500 0.509
Age of the household head 52.38 14.15 51.03 19.13
Number of HH members selling milk 1.489 0.547 1.300 0.466
Number of cows at baseline 4.227 2.509 3.200 1.669
Main dairy farmer:
Is male 0.216 0.414 0.300 0.466
Age 47.57 14.34 46.40 17.38
Is household head 0.477 0.502 0.733 0.450
Is spouse of household head 0.500 0.503 0.200 0.407
Can keep part of cattle income 0.659 0.477 0.793 0.412
Decides how to spend cattle income 0.655 0.478 0.793 0.412
Number of households 88 30
Notes: The sample excludes two households with high attrition and strong outliers. Further, in assessing whether there is variation
over time in the share of income received from Tanykina, we omit weeks with high attrition due to Christmas (2 weeks), elections
(1 week) and the last fieldwork month (4 weeks). The main dairy farmer is the household member reporting the highest value of
dairy income received throughout the year.
Geng, Kramer and Janssens (2016) Liquid Milk 11 / 37
15. Table 2: Summary statistics of time-varying characteristics
Variation in share of No variation in share of
income from Tanykina income from Tanykina
Mean s.e. within Mean s.e.
(1) (2) (3) (4) (5)
Liters of milk produced 71.50 37.49 19.16 52.97 34.44
Non-dairy cash income in 1,000 Sh 2.009 4.065 2.924 2.031 4.907
Non-food expenditures in 1,000 Sh 2.493 4.457 3.754 1.907 6.030
Food expenditures in 1,000 Sh 0.537 0.912 0.844 0.549 1.372
Health problem 0.263 0.440 0.391 0.271 0.445
Has insurance coverage 0.344 0.475 0.245 0.390 0.488
Sells milk 0.847 0.360 0.276 0.697 0.460
Conditional on selling milk...
Total dairy income in 1,000 Sh 1.572 0.936 0.523 1.325 1.025
Share received from Tanykina 0.503 0.413 0.232 0.629 0.483
Share sold to Tanykina∗ 0.300 0.309 0.227 0.395 0.419
Share sold in local market∗ 0.329 0.290 0.169 0.228 0.312
Share consumed by the household 0.274 0.116 0.074 0.292 0.127
Number of households (total N) 88 (3997) 30 (1381)
Notes: Sample excludes two households who dropped out. In assessing variation in the share of income received from Tanykina,
we omit Christmas (2 weeks), elections (1 week) and the last fieldwork month (4 weeks). ∗
Estimated from dividing total sales
value by the Tanykina and other buyers’ milk prices, respectively.
Geng, Kramer and Janssens (2016) Liquid Milk 12 / 37
16. Table 2: Summary statistics of time-varying characteristics
Variation in share of No variation in share of
income from Tanykina income from Tanykina
Mean s.e. within Mean s.e.
(1) (2) (3) (4) (5)
Liters of milk produced 71.50 37.49 19.16 52.97 34.44
Non-dairy cash income in 1,000 Sh 2.009 4.065 2.924 2.031 4.907
Non-food expenditures in 1,000 Sh 2.493 4.457 3.754 1.907 6.030
Food expenditures in 1,000 Sh 0.537 0.912 0.844 0.549 1.372
Health problem 0.263 0.440 0.391 0.271 0.445
Has insurance coverage 0.344 0.475 0.245 0.390 0.488
Sells milk 0.847 0.360 0.276 0.697 0.460
Conditional on selling milk...
Total dairy income in 1,000 Sh 1.572 0.936 0.523 1.325 1.025
Share received from Tanykina 0.503 0.413 0.232 0.629 0.483
Share sold to Tanykina∗ 0.300 0.309 0.227 0.395 0.419
Share sold in local market∗ 0.329 0.290 0.169 0.228 0.312
Share consumed by the household 0.274 0.116 0.074 0.292 0.127
Number of households (total N) 88 (3997) 30 (1381)
Notes: Sample excludes two households who dropped out. In assessing variation in the share of income received from Tanykina,
we omit Christmas (2 weeks), elections (1 week) and the last fieldwork month (4 weeks). ∗
Estimated from dividing total sales
value by the Tanykina and other buyers’ milk prices, respectively.
Geng, Kramer and Janssens (2016) Liquid Milk 12 / 37
17. Figure 1: Price difference between Tanykina and other outlets across time
Geng, Kramer and Janssens (2016) Liquid Milk 13 / 37
18. Figure 2: Distribution of the income share received from Tanykina
0 0.2 0.4 0.6 0.8 1
Log Milk Income Share from Tanykina
0
5
10
15
20
25
30
35
Percentage[%]
Geng, Kramer and Janssens (2016) Liquid Milk 14 / 37
19. Econometric strategy: Equation of interest
Sit = αi + f (mit−1, yit) + xitβ + it
Sit is the milk selling decision for household i in week t:
Share of milk sold to Tanykina and average milk price
Share of dairy income received from Tanykina
f (·) is an unknown smooth function of two variables:
Milk production in the last month (mit−1)
Non-dairy income net of (non-food) expenditures (yit)
Linear part: Household fixed effect (αi ) and others (xit)
Health problems, insurance coverage, and interaction
Production, median milk price (current/lag), food/milk consumption
Geng, Kramer and Janssens (2016) Liquid Milk 15 / 37
20. Econometric strategy: Semi-parametric estimation
Su and Ullah (2006) propose consistent estimators for semi-linear model,
Sit = αi + f (mit−1, yit) + xitβ + it,
using profile least squares, which goes as follows:
1. Express estimator of f (·) assuming that Sit − αi − xitβ is observed as
dependent variable
2. Substitute f (·) for the expression of this explicit but unfeasible
non-parametric estimator
3. Rearrange again such that we obtain the parametric estimators using
traditional ordinary least squares
4. Now, f (·) can be estimated given the parametric estimator
Geng, Kramer and Janssens (2016) Liquid Milk 16 / 37
21. Results: Outline
1. Semi-parametric estimates of the model for
Share of milk sold to Tanykina (estimated)
Average milk price (estimated)
Share of dairy income received from Tanykina (observed)
2. Comparison with a fully linear model
3. Additional analyses:
Do we observe effects on the extensive or intensive margin?
Does cash at hand influence milk consumption?
Heterogeneity by household type and time of the year
Geng, Kramer and Janssens (2016) Liquid Milk 17 / 37
27. Figure 8: Fitted slope of share received from Tanykina w.r.t. past production and net
income
3.6 3.8 4 4.2 4.4 4.6 4.8
Log Milk Production (L2)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
FittedSlopeofShareTanw.r.t.L2MilkProd
25% log income-expense ratio
95% confidence interval
3.81 4.22 4.67
3.6 3.8 4 4.2 4.4 4.6 4.8
Log Milk Production (L2)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
FittedSlopeofShareTanw.r.t.L2MilkProd
50% log income-expense ratio
95% confidence interval
3.81 4.22 4.67
3.6 3.8 4 4.2 4.4 4.6 4.8
Log Milk Production (L2)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
FittedSlopeofShareTanw.r.t.L2MilkProd
75% log income-expense ratio
95% confidence interval
3.81 4.22 4.67
3.6 3.8 4 4.2 4.4 4.6 4.8
Log Milk Production (L2)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
FittedSlopeofShareTanw.r.t.NetInc
25% log income-expense ratio
95% confidence interval
3.81 4.22 4.67
3.6 3.8 4 4.2 4.4 4.6 4.8
Log Milk Production (L2)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
FittedSlopeofShareTanw.r.t.NetInc
50% log income-expense ratio
95% confidence interval
3.81 4.22 4.67
3.6 3.8 4 4.2 4.4 4.6 4.8
Log Milk Production (L2)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
FittedSlopeofShareTanw.r.t.NetInc
75% log income-expense ratio
95% confidence interval
3.81 4.22 4.67
28. Results: Overview
Findings thus far:
1. Share of milk production sold to Tanykina is increasing in cash at
hand, but not across the entire distribution
2. At median levels of cash at hand, local market prices appear to
decrease in cash at hand
3. Combined, this implies that the share of dairy income received from
Tanykina increases in cash at hand
Next, explore health shocks as alternative measure of cash constraints.
Uninsured households will need cash to pay medical bills
Insured households may not need as much cash
Geng, Kramer and Janssens (2016) Liquid Milk 24 / 37
29. Table 3: Estimates of the linear part
Log average Share of Share of
price of milk sold dairy income
milk sold to Tanykina from Tanykina
(1) (2) (3)
Log food expenditures in 1,000 Sh 0.124∗∗ 0.037 0.015
(0.061) (0.037) (0.023)
HH member has health symptoms -0.007 -0.069∗∗∗ -0.058∗∗
(0.024) (0.024) (0.025)
HH has insurance coverage 0.080∗∗ -0.016 -0.009
(0.037) (0.031) (0.029)
... X HH member has health symptoms 0.009 0.067∗∗ 0.049
(0.042) (0.031) (0.031)
R-squared within households 0.002 0.106 0.147
Mean dependent variable 3.309 0.410 0.502
Number of observations 3231 3231 3231
Number of households 88 88 88
Notes: Standard errors in parentheses. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01. Controls: Milk production, milk consumption,
and district-month effects.
Geng, Kramer and Janssens (2016) Liquid Milk 25 / 37
30. Table 3: Estimates of the linear part
Log average Share of Share of
price of milk sold dairy income
milk sold to Tanykina from Tanykina
(1) (2) (3)
Log food expenditures in 1,000 Sh 0.124∗∗ 0.037 0.015
(0.061) (0.037) (0.023)
HH member has health symptoms -0.007 -0.069∗∗∗ -0.058∗∗
(0.024) (0.024) (0.025)
HH has insurance coverage 0.080∗∗ -0.016 -0.009
(0.037) (0.031) (0.029)
... X HH member has health symptoms 0.009 0.067∗∗ 0.049
(0.042) (0.031) (0.031)
R-squared within households 0.002 0.106 0.147
Mean dependent variable 3.309 0.410 0.502
Number of observations 3231 3231 3231
Number of households 88 88 88
Notes: Standard errors in parentheses. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01. Controls: Milk production, milk consumption,
and district-month effects.
Geng, Kramer and Janssens (2016) Liquid Milk 25 / 37
31. Table 3: Estimates of the linear part
Log average Share of Share of
price of milk sold dairy income
milk sold to Tanykina from Tanykina
(1) (2) (3)
Log food expenditures in 1,000 Sh 0.124∗∗ 0.037 0.015
(0.061) (0.037) (0.023)
HH member has health symptoms -0.007 -0.069∗∗∗ -0.058∗∗
(0.024) (0.024) (0.025)
HH has insurance coverage 0.080∗∗ -0.016 -0.009
(0.037) (0.031) (0.029)
... X HH member has health symptoms 0.009 0.067∗∗ 0.049
(0.042) (0.031) (0.031)
R-squared within households 0.002 0.106 0.147
Mean dependent variable 3.309 0.410 0.502
Number of observations 3231 3231 3231
Number of households 88 88 88
Notes: Standard errors in parentheses. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01. Controls: Milk production, milk consumption,
and district-month effects.
Geng, Kramer and Janssens (2016) Liquid Milk 25 / 37
32. Results: Overview
Findings thus far:
1. Share of milk production sold to Tanykina is increasing in cash at
hand, but not across the entire distribution
2. At median levels of cash at hand, local market prices are decreasing in
cash at hand
3. Combined, this implies that the share of dairy income received from
Tanykina increases in cash at hand
4. Health shocks - as alternative measure - reduce share of milk sold to
Tanykina
Estimated using a semi-parametric model: Contribution of this approach?
Geng, Kramer and Janssens (2016) Liquid Milk 26 / 37
33. Figure 9: Fitted share of dairy income from Tanykina: Semi-parametric vs. Linear
34. Results: Overview
Findings thus far:
Cash constraints appear to influence the decision where to sell, and at
what price.
Semi-parametric estimates provide richer description in context of
threshold effects and nonlinearities
Linear model provides an average approximation
Next set of analyses, using the fully linear model:
1. Are our findings strongest at the extensive versus intensive margin?
2. Do cash constraints influence milk consumption decisions?
3. Is there heterogeneity by household type and time of the month?
Geng, Kramer and Janssens (2016) Liquid Milk 28 / 37
35. Table 4: Extensive vs. intensive margin: selling no, some or all milk to Tanykina
No milk Some milk All milk
(1) (2) (3)
Panel A. Centered at 25% quantile
Log production last month -0.057∗∗ -0.038∗∗ 0.095∗∗∗
(0.025) (0.018) (0.025)
Log income-expense ratio -0.014 -0.014 0.028∗
(0.016) (0.012) (0.016)
Panel B. Centered at 50% quantile
Log production last month -0.053∗∗ -0.033∗ 0.086∗∗∗
(0.025) (0.018) (0.024)
Log income-expense ratio -0.009 -0.007 0.016
(0.011) (0.008) (0.011)
Panel C. Centered at 75% quantile
Log production last month -0.048∗ -0.027 0.075∗∗∗
(0.026) (0.019) (0.025)
Log income-expense ratio -0.000 0.004 -0.004
(0.009) (0.006) (0.009)
Interaction term 0.014 0.018 -0.032∗∗
(0.017) (0.012) (0.016)
Mean dependent variable 0.319 0.347 0.335
Number of observations 2962 2962 2962
Notes: Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP1##c.Ins1 Commu-
nity#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
36. Table 4: Extensive vs. intensive margin: selling no, some or all milk to Tanykina
No milk Some milk All milk
(1) (2) (3)
Panel A. Centered at 25% quantile
Log production last month -0.057∗∗ -0.038∗∗ 0.095∗∗∗
(0.025) (0.018) (0.025)
Log income-expense ratio -0.014 -0.014 0.028∗
(0.016) (0.012) (0.016)
Panel B. Centered at 50% quantile
Log production last month -0.053∗∗ -0.033∗ 0.086∗∗∗
(0.025) (0.018) (0.024)
Log income-expense ratio -0.009 -0.007 0.016
(0.011) (0.008) (0.011)
Panel C. Centered at 75% quantile
Log production last month -0.048∗ -0.027 0.075∗∗∗
(0.026) (0.019) (0.025)
Log income-expense ratio -0.000 0.004 -0.004
(0.009) (0.006) (0.009)
Interaction term 0.014 0.018 -0.032∗∗
(0.017) (0.012) (0.016)
Mean dependent variable 0.319 0.347 0.335
Number of observations 2962 2962 2962
Notes: Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP1##c.Ins1 Commu-
nity#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
37. Table 4: Extensive vs. intensive margin: selling no, some or all milk to Tanykina
No milk Some milk All milk
(1) (2) (3)
Panel A. Centered at 25% quantile
Log production last month -0.057∗∗ -0.038∗∗ 0.095∗∗∗
(0.025) (0.018) (0.025)
Log income-expense ratio -0.014 -0.014 0.028∗
(0.016) (0.012) (0.016)
Panel B. Centered at 50% quantile
Log production last month -0.053∗∗ -0.033∗ 0.086∗∗∗
(0.025) (0.018) (0.024)
Log income-expense ratio -0.009 -0.007 0.016
(0.011) (0.008) (0.011)
Panel C. Centered at 75% quantile
Log production last month -0.048∗ -0.027 0.075∗∗∗
(0.026) (0.019) (0.025)
Log income-expense ratio -0.000 0.004 -0.004
(0.009) (0.006) (0.009)
Interaction term 0.014 0.018 -0.032∗∗
(0.017) (0.012) (0.016)
Mean dependent variable 0.319 0.347 0.335
Number of observations 2962 2962 2962
Notes: Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP1##c.Ins1 Commu-
nity#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
38. Table 5: Home consumption versus commercialization
Sold any milk Share of milk sold (conditional)
(1) (2)
Panel A. Centered at 25% quantile
Log production last month -0.010 0.002
(0.022) (0.006)
Log income-expense ratio -0.026∗∗ -0.007∗
(0.012) (0.004)
Panel B. Centered at 50% quantile
Log production last month -0.006 0.004
(0.022) (0.006)
Log income-expense ratio -0.021∗∗ -0.003
(0.009) (0.003)
Panel C. Centered at 75% quantile
Log production last month -0.001 0.008
(0.022) (0.006)
Log income-expense ratio -0.011 0.003
(0.009) (0.002)
Interaction term 0.015 0.010∗∗
(0.014) (0.004)
Mean dependent variable 0.851 0.732
Number of observations 3480 2962
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP1##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
39. Table 5: Home consumption versus commercialization
Sold any milk Share of milk sold (conditional)
(1) (2)
Panel A. Centered at 25% quantile
Log production last month -0.010 0.002
(0.022) (0.006)
Log income-expense ratio -0.026∗∗ -0.007∗
(0.012) (0.004)
Panel B. Centered at 50% quantile
Log production last month -0.006 0.004
(0.022) (0.006)
Log income-expense ratio -0.021∗∗ -0.003
(0.009) (0.003)
Panel C. Centered at 75% quantile
Log production last month -0.001 0.008
(0.022) (0.006)
Log income-expense ratio -0.011 0.003
(0.009) (0.002)
Interaction term 0.015 0.010∗∗
(0.014) (0.004)
Mean dependent variable 0.851 0.732
Number of observations 3480 2962
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP1##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
40. Table 6: Estimates by household type (variables centered at 50% quantile)
Female head Female farmer Male farmer
(1) (2) (3)
Panel A. Share of milk sold to Tanykina
Log production last month 0.137∗∗∗ 0.051 0.347∗∗
(0.032) (0.048) (0.143)
Log income-expense ratio -0.008 0.051∗∗ 0.011
(0.014) (0.023) (0.070)
... X Log production last month 0.031 -0.069∗∗ 0.163
(0.021) (0.032) (0.179)
Mean dependent variable 0.386 0.330 0.572
Panel B. Log price per liter of milk sold
Log production last month 0.181∗∗∗ -0.029 0.111
(0.049) (0.047) (0.078)
Log income-expense ratio -0.051∗∗ -0.068∗∗∗ -0.057
(0.021) (0.022) (0.038)
... X Log production last month 0.072∗∗ 0.045 0.131
(0.032) (0.031) (0.097)
Mean dependent variable 3.323 3.268 3.333
Number of observations 909 1466 587
Number of household 26 44 18
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
41. Table 6: Estimates by household type (variables centered at 50% quantile)
Female head Female farmer Male farmer
(1) (2) (3)
Panel A. Share of milk sold to Tanykina
Log production last month 0.137∗∗∗ 0.051 0.347∗∗
(0.032) (0.048) (0.143)
Log income-expense ratio -0.008 0.051∗∗ 0.011
(0.014) (0.023) (0.070)
... X Log production last month 0.031 -0.069∗∗ 0.163
(0.021) (0.032) (0.179)
Mean dependent variable 0.386 0.330 0.572
Panel B. Log price per liter of milk sold
Log production last month 0.181∗∗∗ -0.029 0.111
(0.049) (0.047) (0.078)
Log income-expense ratio -0.051∗∗ -0.068∗∗∗ -0.057
(0.021) (0.022) (0.038)
... X Log production last month 0.072∗∗ 0.045 0.131
(0.032) (0.031) (0.097)
Mean dependent variable 3.323 3.268 3.333
Number of observations 909 1466 587
Number of household 26 44 18
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
42. Table 6: Estimates by household type (variables centered at 50% quantile)
Female head Female farmer Male farmer
(1) (2) (3)
Panel A. Share of milk sold to Tanykina
Log production last month 0.137∗∗∗ 0.051 0.347∗∗
(0.032) (0.048) (0.143)
Log income-expense ratio -0.008 0.051∗∗ 0.011
(0.014) (0.023) (0.070)
... X Log production last month 0.031 -0.069∗∗ 0.163
(0.021) (0.032) (0.179)
Mean dependent variable 0.386 0.330 0.572
Panel B. Log price per liter of milk sold
Log production last month 0.181∗∗∗ -0.029 0.111
(0.049) (0.047) (0.078)
Log income-expense ratio -0.051∗∗ -0.068∗∗∗ -0.057
(0.021) (0.022) (0.038)
... X Log production last month 0.072∗∗ 0.045 0.131
(0.032) (0.031) (0.097)
Mean dependent variable 3.323 3.268 3.333
Number of observations 909 1466 587
Number of household 26 44 18
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
43. Table 7: Estimates by week (variables centered at 50% quantile)
Week 1 Week 2 Week 3 Week 4
(1) (2) (3) (4)
Panel A. Share of milk sold to Tanykina
Log production last month 0.122 0.240∗∗∗ 0.060 -0.013
(0.079) (0.077) (0.075) (0.039)
Log income-expense ratio 0.029 -0.090∗∗∗ 0.042 0.072∗∗∗
(0.030) (0.035) (0.036) (0.023)
... X Log production last month -0.055 0.136∗∗∗ -0.029 -0.085∗∗
(0.044) (0.049) (0.054) (0.036)
Mean dependent variable 0.404 0.407 0.389 0.380
Panel B. Log price per liter of milk sold
Log production last month -0.162∗∗ 0.153∗∗ -0.020 -0.017
(0.081) (0.063) (0.054) (0.048)
Log income-expense ratio 0.038 -0.155∗∗∗ -0.070∗∗∗ -0.053∗
(0.031) (0.028) (0.026) (0.029)
... X Log production last month -0.063 0.181∗∗∗ 0.037 0.008
(0.045) (0.040) (0.039) (0.045)
Mean dependent variable 3.303 3.298 3.286 3.306
Number of observations 627 914 732 689
Number of household 88 88 88 88
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
44. Table 7: Estimates by week (variables centered at 50% quantile)
Week 1 Week 2 Week 3 Week 4
(1) (2) (3) (4)
Panel A. Share of milk sold to Tanykina
Log production last month 0.122 0.240∗∗∗ 0.060 -0.013
(0.079) (0.077) (0.075) (0.039)
Log income-expense ratio 0.029 -0.090∗∗∗ 0.042 0.072∗∗∗
(0.030) (0.035) (0.036) (0.023)
... X Log production last month -0.055 0.136∗∗∗ -0.029 -0.085∗∗
(0.044) (0.049) (0.054) (0.036)
Mean dependent variable 0.404 0.407 0.389 0.380
Panel B. Log price per liter of milk sold
Log production last month -0.162∗∗ 0.153∗∗ -0.020 -0.017
(0.081) (0.063) (0.054) (0.048)
Log income-expense ratio 0.038 -0.155∗∗∗ -0.070∗∗∗ -0.053∗
(0.031) (0.028) (0.026) (0.029)
... X Log production last month -0.063 0.181∗∗∗ 0.037 0.008
(0.045) (0.040) (0.039) (0.045)
Mean dependent variable 3.303 3.298 3.286 3.306
Number of observations 627 914 732 689
Number of household 88 88 88 88
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
45. Table 7: Estimates by week (variables centered at 50% quantile)
Week 1 Week 2 Week 3 Week 4
(1) (2) (3) (4)
Panel A. Share of milk sold to Tanykina
Log production last month 0.122 0.240∗∗∗ 0.060 -0.013
(0.079) (0.077) (0.075) (0.039)
Log income-expense ratio 0.029 -0.090∗∗∗ 0.042 0.072∗∗∗
(0.030) (0.035) (0.036) (0.023)
... X Log production last month -0.055 0.136∗∗∗ -0.029 -0.085∗∗
(0.044) (0.049) (0.054) (0.036)
Mean dependent variable 0.404 0.407 0.389 0.380
Panel B. Log price per liter of milk sold
Log production last month -0.162∗∗ 0.153∗∗ -0.020 -0.017
(0.081) (0.063) (0.054) (0.048)
Log income-expense ratio 0.038 -0.155∗∗∗ -0.070∗∗∗ -0.053∗
(0.031) (0.028) (0.026) (0.029)
... X Log production last month -0.063 0.181∗∗∗ 0.037 0.008
(0.045) (0.040) (0.039) (0.045)
Mean dependent variable 3.303 3.298 3.286 3.306
Number of observations 627 914 732 689
Number of household 88 88 88 88
Standard errors in parentheses. Controls: MilkProd L1MilkProd PriceMed L1PriceMed c.HP##c.Ins1 Community#i.Month. ∗
p < 0.10, ∗∗
p < 0.05, ∗∗∗
p < 0.01
46. Additional analyses: Summary
1. Are our findings strongest at the extensive versus intensive margin?
Cash at hand increases ⇒ Switch from selling none/some to selling all
milk
2. Do cash constraints influence milk consumption decisions?
Only non-dairy income at below-median levels of milk production
3. Is there heterogeneity by household type and time of the month?
Milk production last month affects marketing decisions mainly:
When farmer is the household head (male or female)
Around the time that the milk payment is due (second week)
Non-dairy income increases share of milk sold to Tanykina mainly:
Among female farmers who are not the household head
In the last week of the month
Geng, Kramer and Janssens (2016) Liquid Milk 33 / 37
47. Conclusion
Do cash constraints affect preferences over the timing of income?
Evidence so far focuses on experimental gifts
(Dean and Sautmann, 2016; Janssens et al., 2016; Carvalho et al.,
2016)
Cash constraints influence choice when to receive milk payments
Local traders raise prices when in need, providing informal insurance
Policy implications for cooperatives:
Farmers can benefit from collective marketing
However, cash constraints hinder farmers’ loyalty to cooperatives
Potential benefits from relaxing farmers’ cash constraints
However, low demand for weekly payments (Kramer and Kunst, 2016)
Increase access to savings devices and low-cost advance payments?
Provide insurance through cooperative (potentially as incentive)?
Geng, Kramer and Janssens (2016) Liquid Milk 34 / 37
49. References
Carvalho, L. S., Meier, S., Wang, S. W., 2016. Poverty and economic decision-making:
Evidence from changes in financial resources at payday. The American Economic
Review 106 (2), 260–284.
Casaburi, L., Macchiavello, R., 2015. Firm and Market Response to Saving Constraints:
Evidence from the Kenyan Dairy Industry. CEPR Discussion Paper No. DP10952.
Collins, D., Morduch, J., Rutherford, S., Ruthven, O., 2009. Portfolios of the poor: how
the world’s poor live on $2 a day. Princeton University Press.
Dean, M., Sautmann, A., 2016. Credit constraints and the measurement of time
preferences. Working paper.
Demirg¨uc¸-Kunt, A., Klapper, L. F., 2012. Measuring financial inclusion: The global
findex database. World Bank Policy Research Working Paper (6025).
Janssens, W., Kramer, B., Swart, L., 2016. Be patient when measuring hyperbolic
discounting: Stationarity, time consistency and time invariance in a field experiment.
Working paper.
Minot, N., Sawyer, B., 2014. Contract Farming in Developing Countries: Review of the
Evidence. Prepared for the Investment Climate Unit of the International Finance
Corporation as a longer version of the IFC Viewpoints policy note on the same topic.
Reardon, T., Barrett, C. B., Berdegu´e, J. A., Swinnen, J. F. M., 2009. Agrifood Industry
Transformation and Small Farmers in Developing Countries. World Development
37 (11), 1717–1727.
Geng, Kramer and Janssens (2016) Liquid Milk 36 / 37
50. Figure 10: Milk production and income-expenditure ratio (in logs)