savings, subsidies, and technology adoption: field experimental evidence from mozambique
TRANSCRIPT
Savings, Subsidies, and Technology Adoption: Field Experimental Evidence
from Mozambique
Michael Carter, UC DavisRachid Laajaj, Universidad de los Andes
Dean Yang, University of Michigan
• Technology adoption has a central role in theoretical and empirical work on economic growth and development
• For the majority of the world’s poor, key technologies are those related to agricultural production– In particular, adoption of fertilizer and improved seeds have
dramatically reduced rural poverty in developing countries
• But sub-Saharan Africa is an outlier in the context of the Green Revolution
Technology adoption in developing countries
2
Source: Morris et al (2007)
Africa lags in food production…
3
… and in fertilizer use
• Low fertilizer utilization in sub-Saharan Africa has motivated decades of policies seeking to stimulate adoption
– E.g., direct subsides, but also price controls, subsidized credit, and direct provision in form of aid
4
Input subsidy programs (ISPs)
• Perhaps the most significant recent development in agricultural policy in Sub-Saharan Africa
• Large-scale subsidization of modern inputs (fertilizer, improved seeds)
5
• Across 10 countries implementing ISPs, 2011 expenditures totaled $1.05 billion, or 28.6% of public agricultural spending
• Substantial budgetary support by World Bank, other donors– Represents an about-face for many development agencies,
which for decades opposed subsidies
Burkina Faso Ethiopia Mali Kenya Nigeria Senegal Ghana Zambia Tanzania Malawi0%
10%
20%
30%
40%
50%
60%
70%
8.4%10.4%
18.1%
25.7% 26.0% 26.1%
29.9%
39.9%
46.0%
58.3%
Expe
nditu
res a
s % o
f pub
lic a
gricu
ltura
l spe
ndin
gISP expenditures in 10 SSA countries, 2011
6Source: Jayne and Rashid (2013)
• Subsidies for technology adoption may be justified on the basis of a variety of market failures– Credit constraints (Lloyd-Ellis et al 2000, Banerjee 2000) – Imperfect information, need for learning-by-doing (Conley and Udry
2010, Foster and Rosenzweig 1995, Munshi 2006)
• Of great policy interest: “graduation” from subsidies– Continuation of large-scale ISPs depends on donor funding– Under what circumstances can subsidies be phased out, leaving
households to self-finance future investments? – Answer may depend on existence of financial services
• Improved financial intermediation facilitates accumulation of investment capital, and ability to cope with risk, leading to higher investment (Gine et al 2004, Greenwood et al 1990, Townsend et al 2006)
• We are interested in particular in constraints on formal savings, and how these interact with provision of subsidies
Subsidies and financial services
7
Savings
• Several recent RCTs on savings in developing countries– Provide formal savings facilities to the poor, to complement
informal savings (e.g., Dupas and Robinson 2013, Brune et al 2016)
– Savings match programs have been attempted, mostly in developed countries (e.g., Sherraden et al 2010, Schaner 2015)
• Experimental studies of savings interventions have not examined their interaction with subsidies, or other development programs
8
• Can combining temporary subsidies with a formal savings-facilitation program enhance impacts on technology adoption over time?
• Two possibilities…
• Dynamic enhancement of subsidies– Buffer stocks/precautionary savings may increase household
willingness to take on risky fertilizer use– Savings may also facilitate carry-over of (fertilizer-boosted)
harvest income to next planting, for sustained fertilizer investment
• Dynamic substitution of subsidies– Savings can serve many purposes
• Buffer stocks / precautionary savings • Asset accumulation for other investments
– These uses may compete with continued fertilizer investment
Savings and subsidies
9
• In our study, we ask:
– Over time, does facilitation of formal savings lead to enhancement or substitution of subsidies for technology adoption?
• In other words, how does the impact of technology adoption subsidies change in the presence of savings facilitation programs?
– Are households better or worse off with the combination of programs?
– If there is an interaction, what mechanisms are operative?
Key questions
10
• U.S. Agency for International Development
• Provincial Government of Manica
• Banco Oportunidade de Mocambique (BOM)
• Food and Agriculture Organization (FAO)
• European Commission (EC)
• International Fertilizer Development Corporation
Key collaborators
11
The study
• ~1,500 rural maize farmers in 94 localities in Manica province, Mozambique– A locality is a grouping of nearby
villages
• Study participants are “progressive” farmers willing to use modern agricultural inputs– Lists generated by government
agricultural extension workers in each village
12
• Maputo
• Beira
Randomization of treatments
• Each locality randomly assigned to one of three savings treatment groups (control, basic savings, matched savings)– After stratification into groups of 3 nearby localities
• Subsidy vouchers assigned by random lottery at participant level within localities
13
No savings program
(32 localities)
Basic savings program
(30 localities)
Matched savings program
(32 localities)
SubsidyNo subsidy
prob. 1/2
prob. 1/3prob. 1/3
prob. 1/3
prob. 1/2 prob. 1/2 prob. 1/2 prob. 1/2 prob. 1/2
94 study localities
N=247N=267 N=278 N=303 N=246N=248
SubsidyNo subsidy
SubsidyNo subsidy
C: “Pure control group”
T1: “Subsidy only”
T2: “Basic savings only”
T4: “Matched savings only”
T5: “Matched savings + subsidy”
T3: “Basic savings + subsidy”
Subsidy vouchers
• 50% of study participants within each locality randomly assigned to voucher receipt in late 2010 (at start of season)
• Provided 73% discount on MZN 3,160 (~US$113) package of fertilizer, improved seeds
• Redeemed at local input suppliers
• Deadline: Jan. 31, 2011
14
• Subsidies preceded savings treatments
• Study participants had no knowledge that savings treatments would come later
Timing of treatments
Randomized distribution of
subsidy vouchers
within localities (Sep-
Dec 2010)
Basic and matched
savings info sessions in randomly-
chosen localities
(Apr-Jul 2011)
1st matched savings period (Aug-Oct 2011)
2nd matched savings period (Aug-Oct 2012)
2010-11 agricultural
season
2011-12 agricultural
season
2012-13 agricultural
season
Dec. Jun. Dec. Jun. Dec. Jun.
15
Savings treatments
• Both savings treatments began with village-level information sessions on formal savings– Emphasized use of savings for
both investment and self-insurance
• Over next two months, one representative per group of 5 study participants receives follow-up training, and asked to convey information to group-mates
• Participants also encouraged to open accounts at BOM, either at Bancomovil or fixed branch locations
16
BOM’s “Bancomovil”
• Savings accounts at Banco Oportunidade de Mocambique (BOM)• Access via 2 branches and scheduled visits by “Bancomovil” units
17
Savings game
18
Educational material on savings and fertilizer
19
Basic vs. matched savings
• Accounts offered in “basic savings” treatment were standard savings accounts
• In “matched savings” treatment:– Match is 50% of minimum balance over match period– Matching funds capped at MZN 1500 (~$54)– Match period: August 1 – October 31– Designed with agricultural cycle in mind
• Match period ends just before next planting season• If save full amount (MZN 3000), savings + match can
purchase input package sufficient for 3/4 hectare plot– Two years of match promised: 2011 and 2012
20
Sussundenga- Bancomovil (BOM)- Barclays Bank
Manica- Bancomovil (BOM)- Barclays Bank- BOM- BIM- BCI
Catandica- Bancomovil (BOM)- Caixa Financeira- BIM
Chimoio- Tchuma- Standard Bank- Barclays Bank- BOM- BIM- BCI- Socremo- Banco Terra
Study localities, by savings treatment
Bancomovil
21
Timing of surveys
Randomized distribution of
subsidy vouchers within localities (Sep-Dec 2010)
Basic and matched savings info sessions in
randomly-chosen localities
(Apr-Jul 2011)
1st matched savings period (Aug-Oct 2011)
2nd matched savings period (Aug-Oct 2012)
Interim survey(Apr 2011)
1st follow-up survey
(Sep 2011)
3rd follow-up survey
(Jul-Aug 2013)
2nd follow-up survey
(Sep 2011)
2010-11 agricultural
season
2011-12 agricultural
season
2012-13 agricultural
season
Dec. Jun. Dec. Jun. Dec. Jun.
22
Take up of subsidies and savings
Note: Means presented in top row for each variable, with standard deviations in parentheses. Voucher use data are from April 2011 interim survey, prior to savings treatments but after subsidy treatment. Savings account ownership are from 2011, 2012, and 2013 follow-up surveys. Savings match data are from BOM administrative records. In brackets: p-values of test of equality of mean in a given treatment group with mean in pure control group, after partialling-out fixed effects for 32 stratification cells (groups of three nearby localities, within which savings treatments were randomly assigned). Standard errors clustered at level of 94 localities. MZN = Mozambican meticais (27 MZN/US$).
26-33 pp increase due to subsidy
17-23 pp increase due savings treatments
23
• What is the impact of subsidies on fertilizer adoption?– Compare subsidy voucher winners and losers within
localities– Should not find differences between localities with
differing savings treatment status in first (subsidized year)
• Does the dynamic impact of subsidies vary in localities getting the savings treatments?– Is fertilizer use more or less persistent after the end of
subsidies in savings localities?
First questions
24
*
2011 2012 2013-0.02
-3.46944695195361E-18
0.02
0.04
0.06
0.08
0.1
0.12
0.14 0.138
0.036
0.011Trea
tmen
t effe
ct o
n ex
tens
ive
mar
gin
of fe
rtiliz
er u
se
• Across all study localities, impact of subsidy declines dramatically in subsequent years
Impact of subsidies
(subsidized season) (post subsidy) (post subsidy)
***
Significance levels: 1%***, 5%**, and 10%*. Fertilizer use in pure control group is 21.7% in 2011, 16.5% in 2012, and 15.7% in 2013. 25
2011 2012 2013-0.02
-3.46944695195361E-18
0.02
0.04
0.06
0.08
0.1
0.12
0.14 0.139
0.054
0.062
0.138
0.027
-0.013
Trea
tmen
t effe
ct o
n ex
tens
ive
mar
gin
of fe
rtiliz
er u
se
• Dynamic impact of subsidy is heterogeneous
• Persistence of a good fraction of subsidy’s effect in no-savings localities
• In savings localities, effect disappears by two seasons later
Heterogeneity by savings treatment status
(subsidized season)
(post subsidy) (post subsidy)
***
*
Significance levels: 1%***, 5%**, and 10%*. Fertilizer use in pure control group is 21.7% in 2011, 16.5% in 2012, and 15.7% in 2013.
***
**
No savings Savings No
savings Savings No savings Savings
26
Regression equation
27
For respondent i in locality j, stratification cell k, period t:
yijkt = z + aVijk + gb(Bjk*Vijk) + gm(Mjk*Vijk) + bbBjk + bmMjk + qk + εijkt
– yijkt = outcome variable– Vijk = indicator for subsidy treatment (individual level)– Locality-level indicators for savings treatments
• Bjk – Basic savings• Mjk – Matched savings
– qk = fixed effect for stratification cell (group of 3 localities)
• OLS, standard errors clustered at level of 94 localities
• Dynamic enhancement: gb > 0, gm > 0– Alternately, gb < 0, gm < 0 indicates dynamic substitutability
Regression results: impact of subsidy
28
Notes: Conditional distribution functions for log(1 + MZN value of fertilizer used in maize production), for no-savings, basic savings, and matched savings localities. Fertilizer use data refers to use during subsidized 2010-11 season, reported in April 2011 interim survey.
• Subsidy raises fraction using fertilizer on maize by 13.9pp (over base of 21.7%)
• No difference in impacts across localities by savings treatment status
First (subsidized) season, 2010-11
29
Notes: Conditional distribution functions for log(1 + MZN value of fertilizer used in maize production), for no-savings, basic savings, and matched savings localities. Fertilizer use data refers to use during post-subsidy 2011-12 season, reported in September 2012 follow-up survey.
• A positive impact remains in no-savings and basic savings localities, ~5-7pp
• Impact of subsidy disappears in matched savings localities
Second season (post-subsidy), 2011-12
30
Notes: Conditional distribution functions for log(1 + MZN value of fertilizer used in maize production), for no-savings, basic savings, and matched savings localities. Fertilizer use data refers to use during post-subsidy 2012-13 season, reported in September 2013 follow-up survey.
• A positive impact remains only in no-savings localities (6.2 pp, on top of base 15.7%)
• Zero impact of subsidy in both types of savings localities
Third season (post-subsidy), 2012-13
31
Alternate specifications
32
• Savings can play multiple roles in household intertemporal optimization– Risk-management: holding buffer stocks / precautionary
savings to cope with shocks (self-insurance)– Investment: funds accumulated and then used productively
• Savings opens up possibilities for alternate uses of household resources, potentially competing with fertilizer
Savings as a general, multi-purpose technology
33
Notes: Conditional distribution functions for log(1 + MZN of formal savings). Formal savings balances reported in follow-up surveys of September 2011, September 2012, and July-August 2013.
• Savings treatments lead to substantial increases in formal savings balances in all years
• Even those not receiving subsidy have resources to save
Impacts on formal savings
34
Regression equation highlighting each treatment
35
For respondent i in locality j, stratification cell k:
yijkt = z + aVijk + bbBjk + bbvBVijk + bmMjk + bmMVijk + qk + εijkt
– Replace interaction terms with dummies for each sub-treatment separately
Impact of treatments on formal savings
36
• In savings localities, resources appear to be diverted from fertilizer to savings in the post-subsidy years
• Magnitudes saved are more than large enough to “compete” with fertilizer as a destination for household resources
• Are households in savings localities any better or worse off as a result?
Household well-being
37
Impacts on consumption per capita
38
• All treatments lead to similar consumption gains, ~8%
• Cannot reject that consumption impacts are similar across treatments T1-T5
• But what about coping with risk?
Impacts on consumption, post-subsidy years
Notes: Conditional distribution functions of average of log(daily consumption per capita in household) across September 2012 and July-August 2013 follow-up surveys. 39
• Do savings treatments help households insulate consumption from negative agricultural shocks?
• We use panel data from 4 surveys on consumption and agricultural shocks– Households report whether past year was a “bad year” for
agriculture– Bad years do not appear to be influenced by treatments
Coping with risk
40
• Regression equation for hh i, locality j, period t:
Consijt = z + g Badyearijt + a [Vij * Badyearijt] + b [Savingsjt * Badyearijt]+ b Savingsjt + qi + wt + εijt
– Vij = indicator for subsidy treatment – Savingsjt = indicator for any savings treatment active in period t– HH and time fixed effects
– Four periods (t= 1, 2, 3, or 4)– Subsidy treatment time-invariant (so main effect absorbed by qi)– Savings treatment active in periods 2, 3, and 4
• Hypotheses: g < 0 : bad years reduce consumptiona < 0 : subsidy makes consumption more sensitive to bad yearsb > 0 : savings makes consumption less sensitive to bad years
Coping with risk: regression equation
41
Coping with risk: regression results
42
0.2
.4.6
.8de
nsity
2 3 4 5 6 7log(daily consumption per capita)
C: Pure control T1: Subsidy only
T2-T5: Any savings treatment
Impacts on consumption variance, 2013
• Subsidy leads to increase in variance of consumption
• Savings treatments have lower consumption variance
Notes: Probability density functions of log(daily consumption per capita in household) in July-August 2013 follow-up survey. 43
T1: Subsidy T2: Basic savings T3: Basic savings + subsidy
T4: Matched savings T5: Matched savings + subsidy
-0.02
-3.46944695195361E-18
0.02
0.04
0.06
0.08
0.1
0.105
0.018
0.051
0.037
0.009
Trea
tmen
t effe
ct o
n st
d. d
ev. o
f log
cons
umpti
on p
er ca
pita
Significance levels: 1%***, 5%**, and 10%*. Data are from 2013 survey. Standard deviation of log consumption in control group is 0.493.
Impact of treatments on consumption variance
***
*
44
T1: Subsidy T2: Basic savings T3: Basic savings + subsidy T4: Matched savings T5: Matched savings + subsidy-0.02
-3.46944695195361E-18
0.02
0.04
0.06
0.08
0.1
0.105
0.018
0.051
0.037
0.009
Trea
tmen
t effe
ct o
n st
d. d
ev. o
f log
cons
umpti
on p
er ca
pita
Impact of treatments on consumption variance
**
*
***
Only T1 raises consumption variance Increase in variance is statistically significantly lower in 3 out of 4 savings treatments
Significance levels: 1%***, 5%**, and 10%*. Data are from 2013 survey. Standard deviation of log consumption in control group is 0.493.
***
45
• Estimates imprecise, but possible increase in total investment in response to savings treatments
• Fertilizer on other crops responds positively as well
• Increase in non-agricultural investments in savings localities
What else are savings households doing?
46
• Subsidies have positive, persistent effects on fertilizer adoption – But dynamic effects disappear when combined with programs
facilitating formal savings
• But households receiving savings treatments are just as well off as subsidy recipients in no-savings localities– In terms of mean consumption– And better risk-coping
• Savings treatments help households pursue objectives beyond fertilizer– In particular, self-insurance via buffer stocks
• Households place high value on the risk-coping gains from formal financial access
• Also: rare evidence of interactions between development programs
In sum
47
Millennium Villages Project
48
• Many anti-poverty programs are “bundled”, in that they consist of multiple components
• Millennium Villages implements interventions in food, education, environment, health, etc.
• Programs to help the “ultrapoor” (Banerjee et al 2015)– Resource transfers, skills training, savings, health, etc.
• But how do the components interact with one another? Are all necessary? Do components complement one another?
• We contribute to knowledge on this front, but more such research is needed
Fighting poverty with multiple interventions
49
Extra slides
50
• Risk averse farmers face the following market failures:– Imperfect (downward-biased) info on fertilizer returns– Credit constraints (no borrowing)– High-cost savings (negative interest rate)
• Modern technology (fertilizer) riskier than traditional technology
• Precautionary savings motive
• Periods:0: post-harvest. Consume, save for planting time.1: planting. Consume, invest in planting, save buffer stocks.2: next year’s post-harvest…
• Subsidies reduce cost of fertilizer investment in period 1
• Savings programs raise interest rate on savings in both periods 0 and 1– Savings from period 0 to 1 can go to either investment or continued
buffer stock savings in period 1
Theory
51
• One-time fertilizer subsidy can raise fertilizer use in subsequent (unsubsidized) seasons by raising beliefs about returns to fertilizer
• Savings program raises interest rate on savings, but has ambiguous effects on fertilizer investment– Dynamic enhancement
• Stimulates saving from period 0 to 1, raising fertilizer investment in period 1
• Buffer stocks held in period 1 also promote risk-taking (fertilizer investment) in period 1
– Dynamic substitution• Buffer stocks compete with fertilizer in period 1; for poorly-
insured, risk-averse hhs, this can dominate• Also: savings held from periods 0 to 1 could be put in other
(non-fertilizer) investments
• Risk-aversion makes substitution more likely, by raising attractiveness of buffer stock savings in period 1
Predictions from theory
52