rising tractor use in sub-saharan africa: evidence …...agricultural & applied economics...

Agricultural & Applied Economics Association (AAEA) – African Section

Washington DC05 August 2018

Rising Tractor Use in sub-Saharan Africa:

Evidence from Tanzania

D. van der WesthuizenT.S. JayneF.H. Meyer

This study was made possible by a number of

funders, including the generous support of the Bill

and Melinda Gates Foundation

2Bureau of Food and Agricultural Policy

Introduction

• The drivers of rising use of mechanization services on smallholder farms remainpoorly understood

Objectives:

• To identify the factors driving recent rise of mechanization use by small-holderfarmers in Tanzania

• To explore the potential role of medium & large-scale farms in promoting amovement to more capital-intensive forms of farming, not only on larger farms buton smallholder farms as well

• To evaluate whether evolving trends in factor use between labor and capital onsmallholder farms in Tanzania is consistent with the Hayami-Ruttan InducedInnovation theory


Import Data shows an Increase in Tractor Demand

Nominal value of tractor imports into region is increasing

International Trade Center &

Trademap, 2016

$-

$100 000

$200 000

$300 000

$400 000

$500 000

$600 000

2001 2003 2005 2007 2009 2011 2013 2015

No

min

al v

alu

e o

f im

po

rts

in U

S$

th

ou

san

d

Sub-Saharan AfricaSouthern AfricaNorth Eastern AfricaWestern Africa

$-

$10 000

$20 000

$30 000

$40 000

$50 000

$60 000

$70 000

$80 000

$90 000

$100 000

2001 2003 2005 2007 2009 2011 2013 2015

Val

ue

of

Imp

ort

s in

US

$ t

hou

san

d

Ghana Nigeria

Kenya Tanzania

Zambia Linear (Ghana)

Linear (Nigeria) Linear (Kenya)

Linear (Tanzania) Linear (Zambia)

700%

Causes of Rising Tractor Use in SSA

Conceptual Framework


Causes of Increased Tractor UseConceptual Framework: Hayami & Ruttan Induced Innovation

Supply:• Cost of capital has declined in Africa since 2000, real interest rates

lower & penetration of banking into rural areas has improved (Andrianaivoand Yartey, 2009; Ojah and Odongo Kodongo, 2015)

• Many medium-scale farmers own/use tractors. As these farmersexpand, there is a growing presence of tractors in rural areas

Demand:

• Rising opportunity cost of farm labor, especially in areas experiencingeconomic dynamism (Tschirley et al., 2015; Yeboah and Jayne, 2018)

• Shifts in labor force into more diversified, off-farm activities associatedwith economic transformation (Yeboah & Jayne, 2018)

• Higher global food prices Incentives to expand area under cultivation Technologies to facilitate area expansion (AGRA, 2016; Jayne et al., 2016;

Richards et al., 2016; UN Population prospects, 2017)

Data & Methods


Data & Methods

• Tractor importation data for 40 African countries – Trademap

• Tanzanian National Panel Survey (NPS) for 2008/09, 2010/11, 2012/13 &2014/15 (TNBS & World Bank) - 9,726 observations for pooled data &1,672 for HH-level panel)

• Demand function for tractor rental services:

• Ordinary Least Squares (OLS) multiple regression to test inducedinnovation hypothesis

1) Pooled generalized linear model (GLM) probit which provides a flexible

generalization of ordinary linear regression

2) Mundlak-Chamberlain device (Mundlak 1978; Chamberlain 1984), providing an

estimator that Woolridge (2010) refers to as the Correlated Random Effects (CRE)

model which address the issue of unobserved heterogeneity at household level

where Y = % change in the number of HH using tractors: ∆ 2008-2010; ∆ 2010-2012

∆ 2012-2014 & key variable of interest: ∆ in factor price (FP) ratio: ∆ 2008-2010; ∆

2010-2012 ∆ 2012-2014 where FP ratio = wage rate divided by tractor rental cost


Data & MethodsModel specification: Demand function for tractor rental services

𝑃 𝒀𝒕𝒓𝒂𝒄𝒕𝒐𝒓𝒓𝒆𝒏𝒕 = 1 𝑋𝑘) = 𝑓 𝑋, 𝐶, 𝑅, 𝑌 + ϵi

𝑋 = ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠

𝐶 = 𝑐𝑜𝑚𝑚𝑢𝑛𝑖𝑡𝑦 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠

𝑅 = 𝑟𝑒𝑔𝑖𝑜𝑛 𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛𝑠

𝑌 = 𝑦𝑒𝑎𝑟 𝑑𝑢𝑚𝑚𝑦 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠

• X: household land cultivated, gender & age of household head, assetwealth & market access conditions

• C : local wage rates, fertilizer prices, tractor rental rates, share of MSfarms as % of total number of farms in district

• R: to regional dummy variables (30 regions)

• Y: survey year dummies (3 for pooled sample; 2 for household panelanalysis)

for panel estimation ϵit =∝𝑖 + 𝜇𝑖𝑡

• Changing tractor use in Tanzania

• Shift in rental markets, especially among small-scale producers

• Tractor rental use is concentrated in certain regions

Results:

Descriptive Statistics


Changing Tractor Use in TanzaniaMore households & area using tractors; small-scale farms increasingly using rental services

World Bank online data: Tanzania National Panel

Survey, 2008/09, 2010/11, 2012/13 & 2014/15

163 029 205 390335 906

490 349

734 136625 660

856 755

1 280 714

0.0%

1.0%

2.0%

3.0%

4.0%

5.0%

6.0%

7.0%

8.0%

-150 000

350 000

850 000

1 350 000

1 850 000

2 350 000

2008/09 2010/11 2012/13 2014/15

% o

f sm

all-

scal

e H

H u

sin

g t

ract

or

ren

tal

serv

ices

To

tal

trac

tor

use

: H

ou

seh

old

s &

Cult

ivat

ed a

rea

Number of households using tractors Hectares of cultivated land where tractors were used

% of small-scale HH using tractor rental services

0-1.99 ha group: + 135% (2009-2015)

2-4.99 ha group: + 309% (2009-2015)


Tractor rental use is concentrated in certain regionsSome regions have experienced higher growth since 2008/09

0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

30.0%

35.0%A

rush

a

Man

yar

a

Kil

iman

jaro

Ku

sin

i U

ngu

ja

Moro

goro

Mji

ni

Ungu

ja

Do

dom

a

Tan

ga

Sin

gid

a

Dar

es

Sal

aam

Mtw

ara

Pw

ani

Irin

ga

Mar

a

Shin

yan

ga

Kas

kaz

ini

Pem

pa

% o

f H

H u

tili

sing

tra

cto

r re

nta

l se

rvic

es

2014/15

2008/09

World Bank online data: Tanzania National Panel

Survey, 2008/09, 2010/11, 2012/13 & 2014/15

Estimation Results

• Pooled GLM probit

• Mundlak-Chamberlain (MC) indicator / CRE model

• Predicted Probabilities


Pooled GLM & MC-CRE Probit Results Selective output for 4 approaches

Estimation approach Pooled GLM Probits Mundlak-Chamberlain Correlated Random

Effects

Dataset 2% rental regions 2% rental regions 2% tractor rental

regions

2% tractor rental

regions

Limited to HH located

in 0-5 ha

Limited to HH located

in 0-5 ha

Cultivated Land Size Distribution = 2 - 4.99 hectares 0.44*** 0.46*** 0.33*…. 0.33*…..

Cultivated Land Size Distribution = 5 - 9.99 hectares 0.62*** . 0.41…... .

year = 2012/13 0.31*** 0.36*** 0.50*** 0.52***

year = 2014/15 0.59*** 0.64*** . .

Household head sex: Male 0.24*** 0.26*** 0.18…… 0.26……

log_market_dist 0.01…… -0.01……. -0.20*….. -0.18……..

log_wage_rate_LP 0.19*** 0.19*** 0.21*** 0.19***

log_trac_rent_cost -0.22*** -0.31*** -0.30**… -0.31*……

hh_5_10_ha 4.37*** 4.14*** 0.63…… 0.35…….

Region = Arusha 0.76*** 0.79*** 0.95*…. 1.23**…

Region = Kilimanjaro 0.95*** 0.96*** 1.00**.. 1.27**…

Region = Morogoro 0.59*** 0.74*** 1.44*** 1.72***

Region = Pwani 0.77*** 0.79*** 1.57*…. 1.77**..

Region = Manyara 0.91*** 1.00*** 1.67**.. 1.85***

log_maize_price_mean . . 0.70*** 0.63***

log_trac_rent_cost_mean . . -0.80**… -0.83*…..

hh_5_10_ha_mean . . 8.90*…. 12.21***

. . . .

Constant -3.32*** -2.47** -0.03……. 0.02…….

Observations 3,728 3,524 1,644 1,564

pval in parentheses

*** p<0.01, ** p<0.05, * p<0.1


0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

0-1

.99 h

ecta

res

2-4

.99 h

ecta

res

5-9

.99 h

ecta

res

>10 h

ecta

res

Mal

e

20

14 2

5th

per

cen

tile

20

14 m

edia

n

20

14 7

5th

per

cen

tile

20

14 2

5th

per

cen

tile

20

14 m

edia

n

20

14 7

5th

per

cen

tile

20

14 2

5th

per

cen

tile

20

14 m

edia

n

20

14 7

5th

per

centi

le

20

14 9

0th

per

cen

tile

Land size category Head

type

Wage rate Tractor rental cost Concentration of 5-20ha HH

per district

Pro

bab

ilit

y P

erce

nta

ge

Model prediction Overall model prediction

Predicted Probability ScenariosDespite overall low success rate, results change quite substantially as we control for certain variables

World Bank online data: Tanzania National Panel Survey,

2008/09, 2010/11, 2012/13 & 2014/15


26.3%

55.1%

62.6%

48.5%

55.6%

61.0%

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

Dodoma Arusha Kilimanjaro Morogoro Pwani Manyara

Predicted probabilities for regions where land size = 4-9.99 ; year =2014 ; head type = male & concentration of

0-5 ha producers = 2014 median

Pro

bab

ilit

y P

erce

nta

ge

Model prediction Overall model prediction

Tractor adoption – Regionally concentrated within specific groups

Predicted probabilities for land size group = 5-9.99; year = 2014, head type = male & concentration of medium-scale producers per district = 2014 median


2008/09, 2010/11, 2012/13 & 2014/15


0%

50%

100%

150%

200%

250%

300%

2008/09 2010/11 2012/13 2014/15

Ind

ex:

20

09

= 1

00

Rental tractor use Wage rate Tractor rental cost

Induced Innovation Hypothesis

Mean of median changes in district-level factor prices


2008/09, 2010/11, 2012/13 & 2014/15



Relative change in factor prices vs. change in share of farms renting tractors

(0.25)

(0.20)

(0.15)

(0.10)

(0.05)

-

0.05

0.10

0.15

0.20

0.25

-25% -15% -5% 5% 15% 25%

Change: FP 2008-2010

(0.25)

(0.20)

(0.15)

(0.10)

(0.05)

-

0.05

0.10

0.15

0.20

0.25

-25% -15% -5% 5% 15% 25%

Change: FP 2010 - 2012

(0.25)

(0.20)

(0.15)

(0.10)

(0.05)

-

0.05

0.10

0.15

0.20

0.25

-25% -15% -5% 5% 15% 25%


(0.25)

(0.20)

(0.15)

(0.10)

(0.05)

-

0.05

0.10

0.15

0.20

0.25

-25% -15% -5% 5% 15% 25%


Y axis: Change in factor price ratio

X axis: % change in

# of farms renting

tractors


Induced Innovation Hypothesis Testing



2008/09, 2010/11, 2012/13 & 2014/15

Model Specification to test the induced innovation hypothesis:

Y = % change in the number of HH renting tractors: ∆ 2008-2010; ∆ 2010-2012 ∆ 2012-2014

X1 = ∆ in factor price (FP) ratio: ∆ 2008-2010; ∆ 2010-2012 ∆ 2012-2014 where FP ratio = wage

rate divided by tractor rental cost

X2 = Household asset wealth (lagged)

X3 = Market distance from household to closest market (lagged)

X4 = Quantity maize harvested in kilograms per hectare (lagged)

X5 = Maize price in TZS per kilogram sold (lagged)

X6 = Fertilizer cost in TZS per kilogram of fertilizer (lagged)

X7 = Concentration of 5-10 hectares farming households per district (lagged)


Induced Innovation Hypothesis Testing



2008/09, 2010/11, 2012/13 & 2014/15

OLS Regression Statistics for Change: % change in the number of HH renting tractors: ∆ 2008-2010; ∆ 2010-2012 ∆ 2012-2014

F-test 2.620 Prob(F) 0.012 Unrestricted Model

MSE1/2 0.080 CV Regr 560.623 F-test 2.620

R2 0.050Durbin-Watson

2.065 R2 0.050

RBar2 0.031 Rho -0.036 RBar2 0.031

Akaike Information Criterion

-5.034Goldfeld-Quandt

0.354Akaike Information Criterion

-5.034

Schwarz Information Criterion

-4.958Schwarz Information Criterion

-4.958

95% InterceptChange in

Factor Price Ratio

Lag_asset_wealth

Lag_market_dist

Lag_qty_harvested

Lag_maize_price

Lag_fert_cost

Lag_hh_5_10_ha

Beta 0.00 0.08 0.00 0.00 0.00 0.00 0.00 0.10S.E. 0.01 0.03 0.00 0.00 0.00 0.00 0.00 0.12t-test -0.32 2.98 0.29 -0.97 2.56 1.02 0.50 0.88Prob(t) 0.75 0.00*** 0.77 0.33 0.01*** 0.31 0.61 0.38

Elasticity at Mean -0.026 0.023 -0.348 0.547 0.685 0.270 0.186Partial Correlation 0.157 0.015 -0.052 0.136 0.056 0.025 0.047

OLS regression indicates significant & positive signs for the change in factor price ratio &

quantity of maize harvested on the % change in the number of HH renting tractors

Conclusions


Conclusions

• Concentration of medium-scale farms per district coupled with increasedtractor rental use by smallholders

• Landholding size coupled with increased tractor rental use

• Increase in # of households using tractors not limited to larger-scaleproducers

• Largest increase in tractor rental use was observed in the 2-4.99 and 5-9.99 hectares’ land size groups

• Significant regional variation in tractor rental use & adoption

• Estimation results uphold the importance of relative changes in factorprices consistent with the induced innovation hypothesis

• Although overall tractor rentals remain low, it is rising particularly in ruralareas experiencing economic transformation

THANK YOU

www.bfap.co.za

+27 82 420 6964

[email protected]

Acknowledgement: Ayala Wineman for her supportive role &

advice in the research

Supporting Slides


Tanzania Maize Market

Area increase & increase in real maize prices

Renapri, 2018

0

100000

200000

300000

400000

500000

600000

700000

800000

900000

0

500

1000

1500

2000

2500

3000

3500

4000

4500

50002

00

0

20

01

20

02

20

03

20

04

20

05

20

06

20

07

20

08

20

09

20

10

20

11

20

12

20

13

20

14

20

15

20

16

20

17

Tsh

/to

n

Th

ou

sa

nd

ha

Area PlantedDomestic Maize Price (Arusha)Domestic Maize Price (Arusha) - Real 2005 termsLinear (Area Planted)Linear (Domestic Maize Price (Arusha) - Real 2005 terms)



Change in wage rate vs. tractor rental costs

0%

50%

100%

150%

200%

250%

300%

350%

2008/09 2010/11 2012/13 2014/15

Wage rate (index: 2008/09 = 100)

Wage rate index: 10 highest rental use regions

Kilimanjaro Arusha Morogoro

Pwani Dodoma Manyara

Mbeya Iringa Mara

Mwanza

0%

50%

100%

150%

200%

250%

300%

350%

400%

450%

2008/09 2010/11 2012/13 2014/15

Tractor rental cost (index: 2008/09 = 100)

Tractor rental cost index: 10 highest rental use regions

Kilimanjaro Arusha Morogoro

Pwani Dodoma Manyara

Mbeya Iringa Mara

Mwanza

rising tractor use in sub-saharan africa: evidence …...agricultural & applied economics...

Documents