PowerPoint-presentatie

Tracking ‘value’ in agricultural value chains:
The case of ‘washed’ coffee in Ethiopia
Seneshaw Tamru1 , Bart Minten2
1KU
Leuven, 2IFPRI
[email protected]
University of California, Berkeley: May 31 – June 2, 2017
1. Introduction
Developing countries are often recommended to add ‘value’ to
their product and ‘move up’ along global value chains
(Swinnen, 2007; UNIDO, 2013; G20 Leaders, 2014):
• However, there are few empirical evidences that looked at
o
o
o
what happens to that ‘value’ when developing countries
move into value addition,
how premiums from that ‘value’ are distributed along
different levels in global value chains,
and constraints to value addition.
• We study this in the case of coffee in Ethiopia
•
2
2. Background on coffee in Ethiopia
• By far the most important export item.
o
More than 30% of exports over the last decade
• Close to 10% of GDP
• More than 4 million households directly involved
3
• Coffee quality depends importantly on the type of processing:
i.e. ‘wet-processing’ or ‘dry-processing’.
4
Producers:
Red berries
Dry berries
3kg
1kg
Wet-Processors
•
Exporters: Wet processed
(washed coffee)
Dry-Processors
Exporters: Dry
processed
(unwashed coffee)
5
• We study this in the case of coffee in Ethiopia:
o
o
Look at “high-value” washed coffee compared to natural
coffee
Track the “value” added from export level to producers
6
3. Data
• Both primary and secondary data sources will be used
1. Household Survey –conducted by ESSP
o Covered 1,600 coffee farming households in the largest
coffee producing zones of the country
o The zones were stratified based on the coffee variety
produced, as defined in the classification for export
markets :Sidama, Jimma, Nekempte, Harar, Yirgacheffe
7
3. Data
2. Producer price data collected from record books of
cooperatives and private traders
o Collected by ESSP in 2013
o 2004-2013
o 106,137 observations
3. Export level data
o From Ministry of Trade
o 2006-2014
• Detailed information on coffee type, quantity, price, sellers,
individual buyers, destination etc
• Close to 31,000 observations
8
4. Methodology:
Model 1
• The Hedonic Price Model – Fixed Effects set up
• 𝐿𝑃𝑖𝑡 = 𝛽𝑘 𝑋𝑖𝑡𝑘 + 𝑉𝑖 + 𝑈𝑖𝑡
• where 𝐿𝑃𝑖𝑡 is the logarithm of coffee price. 𝑋𝑖𝑡𝑘 is a Kdimensional row vector of time-varying different attributes of
coffee. 𝛽𝑘 is a K-dimensional column vector of parameters. 𝑉𝑖
is firm -specific effect while 𝑈𝑖𝑝 is an idiosyncratic error term.
9
Model 2
• Double Hurdle Model
•
1. Red berry sell or not, D is not observed
•
𝐷_𝑖=1 𝑖𝑓 𝑍_𝑖 𝛿+𝑢_𝑖>0
•
𝐷_𝑖=0 𝑖𝑓 𝑍_𝑖 𝛿+𝑢_𝑖≤0
•
2. 〖𝑌_𝑖〗^∗=𝑋_𝑖 𝛽+𝜀_𝑖
•
𝑌_𝑖=〖𝑌_𝑖〗^∗
•
𝑌_𝑖=0
•
𝑢_𝑖≈𝑁(0,1 )
•
𝜀_𝑖≈𝑁(0,𝜎^2)
•
𝑐𝑜𝑟𝑟(𝑢_𝑖, 𝜀_𝑖)=𝜌 unobserved elements effecting redberry seller/or not red-berry seller may affect amount of
red-berry sell
𝑖𝑓 𝐷_𝑖=1 𝑎𝑛𝑑 〖𝑌_𝑖〗^∗>0
𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 (or 𝐷_𝑖=0 or (〖𝑌_𝑖〗^∗≤0 &𝐷_𝑖=1) )
Coffee Producing
Households
Decision
1
Sell in Red
Berries or
Not?
Sell Coffee in
Red Berries
Decision 2
Farmer make decisions in two steps
• Li(θ)=1[yi=0]log[1-
Do not Sell
Coffee in Red
Berries
(xiγ)]+1[yi>0]log[ (xiγ)]
+1[yi>0]{-log [ (xiβ/σ)] +log{φ[(yi – xiβ)/σ]} –log(σ)}
• Conditional: E(y|x, y>0)= xiβ+ σλ(xiβ/σ)
• Unconditional: E(y|x)= (xiγ)[xiβ+ σλ(xiβ/σ)]
10
How much
coffee in red
berries farmers
sell
Amount of
Sales
Model 3
• Threshold Autoregressive (TAR) Model as is used in Van
Campenhout (2007)
• ∆dt =
ρout dt−1 + ρout t dt−1 + εt
εt
ρout dt−1 + ρout t dt−1 + εt
∶
dt−1 > θt
∶ − θt ≤ dt−1 ≤ θt
:
dt−1 < θt
• where 𝑑𝑡 is the difference between the price in export and the
producer level of interest- 𝑑𝑡 = 𝑃𝑡 , 𝐸 − 𝑃𝑡 , 𝐹,
11
5. Quality premiums at export level
Variables
Dependent variable: unit price
Washed
Additional Controls:
Quantity
Organic
Destination indicator
Source of origin
Year of export
Month of export
_cons
sigma_u
sigma_e
Specification 1
Specification 2
Coef.
t-value Coef.
t-value
Unit
log(usc/lb)
yes=1
0.243***
x
x
x
√
√
√
4.542***
0.207
0.161
rho (fraction of variance due to u_i)
F(29,5235)
corr(u_i, Xb)
Number of obs
Number of groups
0.623
630.35***
0.130
5,562
294
• >23 % premium for washed coffee
12
35.32
299.14
0.232***
√
√
√
√
√
√
4.640***
0.177
0.155
0.565
546.68***
0.193
5,558
294
34.12
252.46
However:
Share of washed coffee in national exports is not going up…
[Country is losing 200 million USD per year because of that]
13
So, why?
14
2. When farmers have
access to wet mills, a large
number of them do not sell to
them (or they only sell part of
their coffee)
Percent
1.Not all farmers have
access to red market
50
45
40
35
30
25
20
15
10
5
0
43
19
Farmers with option
to sell red berries
15
Coffee sold as red
6. Premium at producer level
Variables
Dependent variable: unit price
Red berries
Additional Controls:
Quantity
Year
Month
_cons
sigma_u
sigma_e
rho (fraction of variance due to u_i)
F(22,1048)/Wald chi2(26)
corr(u_i, Xb)
Number of obs
Number of groups
Unit
log(usc/lb)
yes=1
Specification 1
Specification 2
Coef.
t-value Coef.
t-value
-0.181*** -4.83
x
x
x
4.678***
0.379
0.333
0.565
23.35***
-0.122
1,167
54
√
√
√
174.78 3.748***
0.304
0.205
0.686
85.89***
0.032
1,167
54
• Dry berries might even have larger premium…
16
0.022
0.70
74.55
Higher margins (gap) for washed-red than unwasheddry
17
Price transmission
Price level pairs
Dry processed (unwashed)
Export (unwashed) and
Producer (dry)
Wet-processed (washed)
Export (washed) and
Producer (red)
Average speed Trend in speed Half life
of adj.
of adj.
(in months)
Est. transac. cost
(as a %ge of
average price)
-0.2232**
0.0004
2.744
20.23
-0.0338
-0.0015
20.107
55.13
• Poor (no) transmission between washed-red
• Fair transmission between unwashed-dry
18
Possible explanations for higher margins and worse
price
transmission
.
1. Larger handling and sunk costs involved for wet-processing - at
the processors level
o Wet-processing also requires more time
2. Washing requires more processing time and cost - e.g. removal
of the parchment etc. - at the export level
3. By the time red berries are being bought up, international prices
for washed coffee may not be known (leading to worse
transmission)
19
7. Associates of sales of red berries
Decision to sell in red
(mfx)
Variables
Unit
Coffee sales in red berries form
percent
Daily wage rate
birr/person days
Distance to nearest Bank
log(km)
Time to nearest wet mill
log(minutes)
Time to nearest asphalt road
log(minutes)
Time to nearest cooperative
log(minutes)
Time preference (default=Time neutral)
Time patient
yes=1
Time impatient
yes=1
Membership coffee cooperative
yes=1
Total active labor in HH
log(number)
Total land owned
log(ha)
Additional controls:
HH Characterisitcs
Mobile ownership
Coffee regions
_cons
sigma
_cons
Log pseudolikelihood
Wald chi2()
No of obs
1A
2A
-0.709***
-0.792***
0.138
-0.042
-0.158***
-0.297***
Quantity of red
berry sales (mfx)
1B
-5.879***
0.132
0.288**
0.968***
-0.044
-0.217***
x
√
x
x
√
x
x
√
x
2.711***
5.773***
66.901***
32.192***
20.938***
-6628.82
-3579.05
142.43*** 8040.30***
1725
1097
2B
Average
Partial Effect
(Cragg)
3
-5.208**
-1.026
2.471
-2.457***
-4.225***
-2.897**
-0.571
1.375
-1.367**
-2.351***
0.919
8.935***
7.147***
6.400**
-2.400**
0.511
4.971***
3.976***
3.561**
-1.335**
√
√
√
92.611***
√
√
√
• Cost of Labor/access labor/impatience come out as important determinants
20
(a) Understanding the role of wages/access to
labor
• Lower productivity of labor for red berry sellers:
o
o
1. More person hours per hectare deployed for red
2. Lesser bean-equivalent per
hours worked for red
21
Conclusion
• More than 23% premium for washed coffee at the export level
• However:
• share of washed coffee in coffee exports not going up
Why?
- No quality premium (on average) at producer level
- Linked to poor (or no) vertical price transmission between
washed-red cherries and export markets
22
Conclusion
• Factors explaining higher margins and price transmissions:
1-Longer processing time, larger handling and sunk costs for wetprocessing- at the processors and export levels
2. Red perishable- less sales window and possibly lower
bargaining power
3. By the time red berries are being bought up, international
prices for washed coffee may not be known
• Other factors that determine farmers to prefer dried cherries:
1/ Labor productivity significantly lower in the case of red berries
2/ Dried berries preferred for savings (and we find is a more
rewarding way (on average) than saving on the bank)
23
Implications
•
Not all “value” addition is good for the farmer; heterogenous
effects; only farmers with specific characteristics might
benefit from it
•
To improve higher share for farmers, design ways to improve
vertical price transmission for better incentives
•
Improve supplementary institutions (Access to improved
savings mechanisms might affect participation in commodity
markets)
24
Thank You!
25
Coffee quality depends importantly on the type of
processing: i.e. ‘wet-processing’ or ‘dry-processing’.
• Dry processing - berries are dried,
mostly traditional (on mats).
o processed in dry processors (hulling
machines)
• produces unwashed beans
• Wet processing - fresh red berries are
de-pulped, fermented and washed
o using wet-mill machines
• produces washed beans.
o
Red cherries delivered to washing
stations within 10 -12 hours of picking
• KEY: Farmers need to supply their coffee
in red-berries
26
Objective
• We test for quality premiums between :
o
o
washed and unwashed coffee at export level
red berries and dries berries at producer level
• Focusing more on the supply side of the issue (at the
producers level), we explore the major factors that determine
selling coffee in red berries form - the prerequisite for wetprocessing coffee.
27
Coffee quality depends importantly on the type of
processing: i.e. ‘wet-processing’ or ‘dry-processing’.
• Dry processing - berries are dried,
mostly traditional (on mats).
o processed in dry processors (hulling
machines)
• produces unwashed beans
• Wet processing - fresh red berries are
de-pulped, fermented and washed
o using wet-mill machines
• produces washed beans.
o
Red cherries delivered to washing
stations within 10 -12 hours of picking
• KEY: Farmers need to supply their coffee
in red-berries
28
• Washed coffee preserves the intrinsic quality of the
bean better than unwashed beans; the process leads
to homogenous coffee with fewer defective beans.
29
Large premiums have seemingly led to large
investments in wet mills over time
Number of wet mills in the Sidama area
30
However:
8.5
Utilized capacity
1. Overcapacity
2. When farmers have
43
40
Percent
access to wet mills, a
large number of them
do not sell to them (or
they only sell part of
their coffee)
50
19
30
20
10
0
Farmers with option
to sell red berries
31
Coffee sold as red
(b) Understanding the role of time preferences
• Patience correlated with proportion of red berry sales
• Farmers that are more patient prefer dried cherries because it
is form of savings for them
32
May-November comparison
• On average, about 20% larger premium for dry because of
advantage of later sales
33
Model 3
• Propensity Score Matchings:
A). Nearest neighbor
B). Kernel-based weighting
C). Inverse Probability of Treatment Weighting (IPTW)
D). Regression adjustment matching
34
PSM results: Labor comparison
Dependent Variables
Treatment [coffee sold in]:
1=red berries ; 0=dry berries
Matching Regression
Labor hours per hactare used for:
Harvesting
activities
Post-harvest
activities
Overall activity
Marketing
[transport] costs
per kg of coffee
Coeff.
z-value
Coeff.
z-value
Coeff.
ATET- Nearest-neighbor matching
0.321***
4.41
0.039
0.47
0.486***
8.19
1.478***
12.98
ATET- Kernel matching
ATET - Regression adjustment
matching
ATET - IPTW matching
0.169**
2.40
-0.121*
-1.65
0.335***
4.80 1.278***
11.64
0.341***
5.80
0.048
0.74
0.480***
9.28 1.531***
17.20
0.356***
5.85
0.082
1.29
0.493***
9.70
17.05
Balancing (tebalance overid test): chi2(14) 11.943
Prob>chi2 0.450
9.134
0.691
• Red: more harvesting and overall labor
o
Also more marketing (transport) cost
35
14.988
0.242
z-value Coeff.
1.477***
16.953
0.109
z-value
Lower labor productivity for red berry sellers
Dependent Variables
Treatment [coffee sold in]:
1=red berries ; 0=dry berries
Matching Regression
Labor Productivity [as measured by]:
Production [bean Eq.]
per Labor hours
Income
per Labor hours
Marketing costs [as measured by]:
Production [bean Eq.]
Income
per Marketing costs per Marketing costs
Coeff.
z-value
Coeff.
z-value
Coeff.
z-value
Coeff.
z-value
ATET- Nearest-neighbor matching
-0.101**
-2.26
-0.398***
-5.08
-1.205***
-8.69
-1.426***
-9.99
ATET- Kernel matching
ATET - Regression adjustment
matching
ATET - IPTW matching
-0.076**
-1.97
-0.105*
-1.76
-0.968***
-6.03
-1.139***
-8.78
-0.083**
-2.32
-0.371***
-6.01
-1.155***
-9.00
-1.364***
-10.08
-0.089***
-2.58
-0.346***
-5.83
-1.674***
-11.87
-1.906***
-12.06
Balancing (tebalance overid test): chi2(14)
Prob>chi2
15.247
0.228
16.6387
0.164
17.334
0.299
16.888
0.326
• Significantly lower labor productivity for red:
o
o
A. Bean eq. per labor hours used
B. Coffee income per labor hours used
• Larger marketing costs for red berry sellers in both measures as well
36
Time preferences
Farmers that are more patient prefer dried cherries because it is
form of savings for them
“I prefer selling coffee in dried form instead of red berries because
I can spread out my income that way (it is a way of saving)”
No, I disagree,
19
Yes, I agree, 76
Moreover, they are right as it is more rewarding than putting
37
money on the bank (see next slide)