The Impact of Rainfall Index insurance in Amhara, Ethiopia

The Impact of Rainfall Index Insurance in Amhara, Ethiopia

Shukri Ahmed, FAO Craig McIntosh, UC San DiegoAlexander Sarris, University of Athens

2

The promise of weather index insurance:

From the perspective of Townsend (1994), the risk that farming communities cannot manage themselves is covariate, primary source of this is weather.

Weather is an outcome that allows insurance contracts to be written with no moral hazard (Gine & Yang 2009).

Consequently, WII appears to be an effective way to protect farmers against unavoidable risks.

Interlinking credit with insurance may enhance the willingness of farmers to borrow to invest in inputs, generating a first-order expansion in productivity (Carter et al. 2015).

3

The pitfalls of weather index insurance:

Despite this promise, WII has struggled to generate demand at market prices (Cole et al. 2012).

Highly risk averse farmers may dislike possibility of ‘contract non-performance’ (Clarke 2011).

Ambiguity aversion may depress demand for complex products with unknown probability distributions (Bryan 2010, Carter & Elabed 2015, McIntosh et al. 2015).

4

Our project had two key purposes:

1. To attempt to ‘interlink’ index insurance with credit for smallholder Ethiopian farmers. 2. To work entirely with private-sector providers of insurance (Nyala) and credit (Dashen) to see if a market-driven approach to WII using initial subsidies could generate durable, sustainable demand at market prices.

Question of this project: can the right combination of individual price subsidies and interlinking with credit unlock demand for a private market product?

5

Ethiopia may be an ideal environment for weather index insurance because:

1. Vast majority of agriculture is rain-fed.2. Rainfall variability is among the highest in the world.3. Risk has been demonstrated to be a major factor

constraining farmers away from using the optimal level of inputs (Dercon and Christiaensen, 2011).

4. The presence of strong intermediary institutions such as village cooperatives and cooperative unions provide structure to offer insurance.

5. Presence of strong private-sector insurance company, Nyala.

6

However, Ethiopia is also a very challenging environment for a private-sector intervention:1. Very strong state provides entire input and output chain

for cooperative farming sector.2. Weak history of private sector involvement in

agriculture.3. Three decade history of major food relief efforts to

famine-struck areas.4. Large government safety-net program (PSNP) may

serve as a substitute for private-sector insurance (Duru 2015).

Raises the question: is it possible for the state to be too credible at providing disaster relief, thereby undermining private-sector demand for insurance?

7

Ethiopian Project on Interlinking Insurance & Credit for Agriculture (EPIICA):Project is a collaboration between researchers and:

Nyala Insurance Company (largest insurer in country)Dashen Bank (largest private-sector bank in country)Ethiopian Economics Association (fieldwork/analysis).

Purpose of project is:to test impact of rainfall insurance in one of the most drought-exposed farming populations in world.to understand the extent to which interlinking credit and insurance (rainfall-contingent loans) can unlock demand for inputs in smallholder agriculture

8

Standalone Insurance: Sold through primary (village-level) cooperatives to

members at time of purchasing inputs. Framed as input insurance, meaning that it would cover

cost of inputs if rain fails. Payoffs with trigger/exit for each of three crop phases,

optimized separately for maize, sorghum, teff, and wheat for each insured station.

Only households in villages whose center is less than 15km from an insured station offered insurance.

9

Interlinked Insurance: Cooperative Unions (collectives of village-level

cooperatives) are used as credit intermediaries. Each CU signs single loan contract with Dashen, is

made beneficiary of Nyala insurance policy. Pushes the CUs into new role, asking them to take

collateralized loans with collective assets. Premium must be paid up front for either product. Can only get the interlinked loan if insurance purchased,

but can choose standalone product also in interlinked arm.

10

Problems in the Interlinked Arm: Cooperative Unions reluctant to take on risk of loans,

particularly as government has typically provided credit to their members.

Heavy state involvement in credit sector, negative real interest rates.

Unpredictable role of government in smallholder input financing: ‘the game of chicken’.

Bureaucratic delays in screening of collateral, account opening, etc.

Interlinked credit could not be executed in either the first or the second year’s sales, only in the third.

Interlinked arm is standalone from an impact perspective.

11

Research design, intended and actual:Original sample:120 kebeles: 40 control, 40 standalone, 40 interlinked.

However, not all turn out to be deficit-rainfall threatened.

Drought-threatened sample:84 kebeles: 27 control, 29 standalone, 28 interlinked

However, Swiss Re refuses all but 7 stations.

Drought-threatened insurable sample ‘Experimental’:49 kebeles: 15 control, 17 standalone, 17 interlinked

‘Experimental’ sample: 15 control vs. 34 treatment

12

Survey Design:We ran a four round panel survey.Two baseline surveys prior to implementation.One survey in each of the years following the first two sales windows.

The household survey sampled 20 households per village:18 households that were randomly sampled members of the cooperatives.2 households that were randomly sampled from the non-cooperative members in the village.

Our analysis uses only the cooperative members, since they were the only ones with easy access to purchase insurance and inputs.

13

Organization of the Panel Analysis:The study features four rounds of household surveys, and two rounds of insurance sales for which we have post-sales outcome data:

SURVEY TIMING: SALES WINDOW TIMING: 2011: Jan – Mar: R1 Survey 2012: Jan – Mar: R2 Survey July-Aug: S1 sales, standalone only 2013: Jan – Mar: R3 Survey Apr: S1 sales payouts. May-Jul: S2 sales, standalone only 2014: Jan – Mar: R4 Survey Apr: S2 sales payouts. Apr-Jun: S3 sales, interlinked only

14

The Individual-level Voucher Experiment:

In order to improve power of the village-level experiment:We randomized the provision of insurance purchase vouchers at the individual level.

In the first two sales years, these vouchers enabled farmers to acquire up to that amount of insurance for free.

The large majority of insurance coverage issued in the project comes from these vouchers rather than from private demand.

That means that the study is largely measuring the impact of providing small amounts of insurance cover for free.

Quantity of coverage ~ directly randomized at individual level.

15

Uptake across three years:Year 1 sales window:offered subsidies only to the study sample; uptake among those offered subsidies was 34%Uptake rate <.5% among the broader population not offered subsidies.

Year 2 sales window: subsidy experiment in whole membership of coop, with vouchers of 0, $6, and $12, more than 5,000 contracts written by Aug 2013.Uptake rate in subsidized sample ~ 41%Uptake rate in unsubsidized sample ~ .5%.

Year 3 sales window:vouchers changed to cover a given fraction of purchase (no free lunch); sales almost completely shut down.Strong interlinked sales to a single cooperative, Feres Wega.

16

The Individual-level Voucher Experiment: Only 21% of farmers put any of their own money into purchase;

most took the voucher and purchased only that much coverage.

This is an experiment in giving away insurance coverage. Quantity of coverage ~ directly randomized at individual level.

050

100

Val

ue o

f Inp

uts

Insu

red

0 20 40 60Subsidy Voucher Amount, US$

All Treatments Kebeles with Any UptakeAll, fitted Uptake, fitted

Circle size proportional to number of observations at each subsidy amount

Sum Insured by Subsidy Voucher

17

Credit Constraints & Interlinked Demand:Uptake depending of type of Credit Constraint faced by the household:

For villages with any sales and for individuals who received vouchers, the Interlinked product appears particularly appropriate for the 17% of individuals who are ‘Risk Constrained’ (Bouchet et al. 2011). Otherwise, no evidence of stronger overall demand for the Interlinked product, even for those who report being Price or Quantity constrained in access to credit.

0 .2 .4 .6 0 .2 .4 .6

Standalone

Interlinked

Standalone

Interlinked

Standalone

Interlinked

Standalone

Interlinked

None Price

Quantity Risk

mean of Insurance_DemandGraphs by cred_cons

18

Balance in the Experimental Sample:

Despite reduced sample size, balance decent across villages and excellent in individual voucher experiments.

Preliminary results: please do not circulate.

Balance Test using Outcomes in Rounds 1 and 2, all treatment terms:

Uses Chemical Fertilizer

Fertilizer used per Hectare

(KG)

Uses Improved

Seeds

Number of Parcels

Cultivated

Uses Agricultural

Credit

Interlinked Treatment 0.246* 43.09 150.7 0.391 0.0335 (0.143) (29.100) (90.830) (0.352) (0.054)Standalone Treatment 0.182 30.25 116 0.199 0.0303 (0.155) (30.640) (79.670) (0.375) (0.060)Voucher S1 -0.0657 -6.359 -72.53 0.113 -0.0298 (0.088) (17.660) (53.550) (0.331) (0.043)Voucher Amount S1 0.00012 0.0164 0.253 -0.000814 -0.00000726 (0.000) (0.039) (0.161) (0.001) (0.000)Voucher S2 0.0373 0.31 -111.4 0.254 0.00134 (0.092) (17.540) (74.480) (0.252) (0.041)Voucher Amount S2 -0.000416 -0.0684 0.252 -0.000482 -0.0000728 (0.000) (0.045) (0.237) (0.001) (0.000)Baseline Outcome in Control 0.450*** 58.17*** 78.46*** 3.388*** 0.157*** (0.104) (20.610) (21.510) (0.216) (0.041)

Observations 818 809 1,636 818 1,636R-squared 0.028 0.019 0.014 0.012 0.001

19

Using the Experimental Design to analyze impact:

The core regression specifications take the form: ikt i t T kt V ikt iktfert T V

where i is individual (household), k is kebele, t is survey wave (1-4), ktT is kebele-level treatment status and iktV is individual-level voucher status, randomized per round.

Most outcomes are not observed in R2, so only the round dummies 3 and 4 are identified.

20

Panel Impacts on Fertilizer Use

Very little uptake outside of voucher group within the study sample.No indication that provision of small amounts of free insurance leads to an improvement in fertilizer use.

Preliminary results: please do not circulate.

Panel Impact on Fertilizers:

Covered by Insurance

Sum InsuredUses Chemical

Fertilizer

Number of Plots on which Chemical Ferts

Used

Urea Used per Hectare

DAP Used per Hectare

Total Fertilizer Used per Hectare

Household Received Voucher this season 0.331*** 125.8 0.0161 -0.144 -0.902 3.853 3.688 (0.047) (92.000) (0.038) (0.093) (4.472) (4.808) (8.309)Amount of Household Voucher this season 0.000218 1.887*** -0.000105 0.000382 0.00741 -0.00938 -0.00483 (0.000) (0.442) (0.000) (0.000) (0.016) (0.017) (0.029)Round 3 -0.0262** -36.09* 0.201*** 0.460*** 7.549*** 4.310** 11.57***

(0.012) (20.680) (0.024) (0.056) (2.061) (2.093) (3.782)Round 4 0.0253*** 14.15 0.155*** 0.342*** 9.051*** 8.047*** 16.92***

(0.009) (18.710) (0.022) (0.048) (2.023) (2.082) (3.742)Constant 0.000263 0.486 0.551*** 1.196*** 34.27*** 40.69*** 75.72***

(0.005) (10.820) (0.011) (0.023) (1.041) (1.032) (1.927)

Observations 2,571 2,571 2,571 2,571 2,428 2,428 2,428R-squared 0.272 0.198 0.078 0.07 0.02 0.016 0.021Number of quest_id 882 882 882 882 876 876 876

21

Instrumenting for sum insured w/ Voucher Amount

Study provides a very large degree of experimental intensive margin variation in the sum insured, yet . . .No evidence that increasing the voucher amount (and hence sum insured) leads to changes in the use of fertilizers.

Preliminary results: please do not circulate.

Impact of Sum Insured, Instrumenting for Sum Insured with Voucher Amount

Uses Chemical Fertilizer

Number of Plots on which Chemical Ferts

Used

Urea Used per Hectare

DAP Used per Hectare

Total Fertilizer Used per Hectare

Sum Insured (instrumented w voucher amt) -0.0000339 -0.0000342 0.00226 0.000481 0.00238 (0.000) (0.000) (0.004) (0.004) (0.007)R3 0.210*** 0.459*** 7.609*** 5.148** 12.59*** (0.022) (0.050) (2.279) (2.257) (4.126)R4 0.164*** 0.330*** 8.865*** 8.921*** 17.77*** (0.020) (0.047) (2.126) (2.106) (3.850)Constant 0.559*** 1.215*** 35.03*** 41.40*** 77.23*** (0.012) (0.029) (1.286) (1.274) (2.329)

Observations 2,454 2,454 2,323 2,323 2,323Number of Observations 818 818 813 813 813

22

Do past payouts drive future uptake? Contrary to others

in literature, we find that ‘payouts’ are actually a negative predictor of uptake.

First year payouts were late, may have depressed demand for insurance in second season for those who were supposed to be paid.

The Impact of Receiving a Payout on Sales in the Subsequent Season.

Purchased Insurance in Sales Season 2

Received Payout in Sales season 1 -0.104*(0.057)

Would have received payout if bought in S1 0.0807(0.060)

Any voucher S1 0.0271(0.045)

Voucher amount S1 -0.000096(0.000)

Any voucher S2 0.402***(0.082)

Voucher S2 0.000102(0.000)

Constant -0.0105(0.010)

Observations 818R-squared 0.301

23

Impacts: Seeds

Slight improvement in value of seeds, but this is local seeds not improved.

Preliminary results: please do not circulate.

Panel Impact on Seeds:

VARIABLES

Uses Any Improved

Seeds

Value of Local Seeds Used

Value of Improved

Seeds Used

Household Received Voucher this season -0.019 262.5* -13.39 (0.042) (142.300) (55.520)Amount of Household Voucher this season 0.0001 -0.594 -0.0104 (0.000) (0.431) (0.200)Round 3 0.00973 -560.7*** 56.44**

(0.023) (104.800) (26.450)Round 4 -0.00135 -442.4*** 39.49

(0.021) (104.600) (26.600)Constant 0.370*** 1,072*** 161.0***

(0.010) (48.160) (13.100)

Observations 2,571 2,571 2,571R-squared 0.001 0.032 0.003Number of quest_id 882 882 882

24

Impacts: Other Inputs

Some evidence of an increased use of Input Credit when insured (point estimate is large; 50% of baseline credit usage rate).

Preliminary results: please do not circulate.

Panel Impact on Other Inputs:

VARIABLES

Total Hectares of Land Farmed

Total Number of Parcels Cultivated

Used any Input Credit

Used any Chemical

Pesticides or Herbicides

Used Hired Labor

Household Received Voucher this season -0.189 -0.095 0.0743* 0.0256 -0.00134 (0.260) (0.117) (0.040) (0.049) (0.042)Amount of Household Voucher this season 0.000815 -0.000214 -0.000195 0.000118 9.35E-06 (0.001) (0.000) (0.000) (0.000) (0.000)Round 3 -0.484* -0.357*** 0.0821*** 0.130*** 0.0368*

(0.270) (0.068) (0.022) (0.025) (0.022)Round 4 -0.516** -0.441*** -0.016 0.127*** 0.0728***

(0.245) (0.065) (0.019) (0.024) (0.019)Constant 1.621*** 3.607*** 0.153*** 0.279*** 0.215***

(0.165) (0.035) (0.009) (0.012) (0.010)

Observations 2,571 2,571 2,571 2,571 2,571R-squared 0.005 0.066 0.029 0.049 0.011Number of quest_id 882 882 882 882 882

25

Impacts: Yields

Large secular decrease in yields for major crops despite increase in fertilizer use. Fertilizers compensating for declining soil fertility?No improvements in yields from insurance.

Preliminary results: please do not circulate.

Panel Impact on Yields:

VARIABLES

Wheat Maize Teff Sorghum

Household Received Voucher this season 188.9 34.21 50.17 -169.8 (341.700) (139.400) (72.960) (177.700)Amount of Household Voucher this season 0.413 0.258 -0.22 0.23 (0.911) (0.521) (0.249) (0.565)Round 3 -803.1*** -300.3*** -140.9*** 84.07

(243.800) (88.770) (50.730) (93.790)Round 4 -832.7*** -428.0*** -158.7*** 205.3**

(215.100) (77.360) (45.440) (83.560)Constant 1,160*** 1,395*** 740.1*** 981.4***

(90.690) (40.320) (21.360) (34.460)

Observations 360 1,090 1,774 896R-squared 0.184 0.058 0.027 0.016Number of quest_id 220 497 751 423

26

Conclusions: the successes of the projects. Succeeded in building collaborative relationship with

Nyala, Dashen. Fielded the Interlinked product in the third year. Worked with Dashen to acquire support for the

Interlinked product from USAID’s Development Credit Authority.

Issued a large number of insurance policies as a part of the project: 728 farmers in the first sales year Over 5,000 farmers in the second sales year 254 Interlinked policies sold at full market price in most recent sales

year.

27

Conclusions: however . . . . Projects results are disappointing on two levels:1. Without a 100% subsidy rate, it appears that there was no viable

demand for rainfall index insurance in this case.2. Even when free insurance was distributed, this appears to have had

no effect on the farming behavior of covered households. We have substantial individually-randomized variation in the extent of insurance

coverage, and no evidence that this generated meaningful changes in agricultural behavior.

Did this occur because the product was not marketed correctly to the field?

88% of study households in the treatment area said that they had received information about the product, 57% received a brochure describing product.

70% said they understood the product ‘well’ or ‘partially’. However, only 2% of households correctly specified deficit rainfall at the

closest station as the event that triggers payouts, but most believed it was actually an indemnity policy (which should have been more attractive).

28

Conclusions: Moving forwardBased on results of summary of literature of Randomized Controlled Trials from ATAI/JPAL:No evidence from anywhere in the world that the current type of weather index insurance products can move to scale at commercial prices.

And yet, risk remains a dominant concern in agriculture!

So what are the promising areas moving forward?1.Embrace subsidized Weather Index Insurance.2.Shift focus to ag technology that protects farmers from risk.3.Improve the design of insurance products (better indexes, group insurance).4.Pursue Meso-level insurance (government safety net programs, insuring agricultural lenders).