Index Insurance: Financial Innovations for Agricultural Risk Management and Development Sommarat Chantarat Arndt-Corden Department of Economics Australian National University PSEKP Seminar Series, Gadjah Mada University 3 January 22
Outline of my talk Motivations Development and implement index insurance program Satellite based livestock insurance in Kenya Prospects for Indonesia: Interesting research questions 2
Motivations Index Insurance Kenya Experience Prospects for Indonesia Insurance and Development Economic costs of uninsured (weather and natural disaster) risk, especially w/ threshold-based poverty traps Insurance protect rural livelihoods and escape poverty Provide safety net to prevent collapse of vulnerable populations Encourage investment and asset accumulation by the poor Induce financial deepening by crowding in credit market and social insurance Insurance pre-finance effective emergency response and recovery Timely response enhances resilience to shocks and reduce costs of humanitarian responses/social protection programs 3
Density Density 2 4 6 8 Motivations Index Insurance Kenya Experience Prospects for Indonesia Insurance and Agricultural Risk Management Kernel density estimate High frequency/ Low loss/ Idiosyncratic Self and informal insurance effective Low frequency/ Extreme loss/ Covariate Need to transfer to formal risk market L..2.3.4 Kargi kernel = epanechnikov, bandwidth =.29 Agricultural Livelihood/Income/Asset Loss L T Two types of formal agricultural insurance Conventional crop insurance Compensate actual loss, multi-peril or named coverage High costs of verifying losses Moral hazard and adverse selection Existing programs are very costly and largely subsidized Index insurance Compensate specific loss based on objectively measured index NOT actual loss Low costs no farm-level loss verification Low incentive problems insured cannot influence payout probability Challenges in minimizing basis risk No successful crop insurance in the world, Correspondence not should likely be addressed work to Sommarat.Chantarat@anu.edu.au in rural areas Promise as a market viable instruments, more suitable for rural areas in DCs 4
Indemnity Payout Motivations Index Insurance Kenya Experience Prospects for Indonesia Developing Index Insurance Program. Identify loss to be insured (L lt ) Identify uninsured loss by testing simple consumption risk sharing hypothesis (e.g., Townsend 994), L lt is uninsured if H : c = is rejected C lt = a + a l + a t + bx lt + cl lt + ε lt 2. Select objectively measured index (θ lt ) Highly correlated with loss, available reliably in near-real time, non-manipulable by insured parties, high spatial distribution, at least 2 years historical profiles 3. Quantify insurable loss from index (L θ lt ) θ lt needs to explain most of the loss variations: L lt = L θ lt + ε lt L θ lt Use micro data of L lt to minimize basis risk Π lt L θ lt L 4. Identify optimal contract structure L* L θ lt Payoff based on L θ lt : Π lt L θ lt L = max L θ lt L, sum insured Stand-alone contract, group-based contract, interlinked insurance-loan 5
Motivations Index Insurance Kenya Experience Prospects for Indonesia Developing Index Insurance Program 5. Actuarial pricing Actuarial fair premium: burn rate and/or Monte Carlo simulation based on f(θ lt ) p l L θ lt L = E Π lt L θ lt L = Π lt L θ lt L df( θ lt ) 6. Ex-ante contract evaluation Simulated welfare and behavior response impacts using dynamic model/data Field experiments to elicit willingness to pay among targeted clients 7. Develop education and extension tools for pilot sale Simplified products, financial educational tools, targeted learning network 8. Identify cost effective delivery mechanisms Delivery through mobile technology, local financial institutions, network groups 9. Long-term micro-level impact assessment Randomized survey and experiments to elicit demand, impacts on welfare, induced behavior responses from control and treatment groups 6
Motivations Index Insurance Kenya Experience Prospects for Indonesia Marsabit () Identify loss to be insured: Catastrophic livestock losses from drought as key uninsured risk in this area Observed household welfare co-move with livestock losses % 8% 6% 4% 2% % Seasonal Livestock Loss (%) 2 2 22 23 24 25 26 27 28 29 2 Balesa Maikona Kalacha Northhorr Karare Korr $. $.8 $.6 $.4 $.2 $. Average Household Consumption ($/day) 2 2 22 23 24 25 26 27 28 29 2 Maikona Kalacha Northhorr Karare Korr Logologo Ngurunit 7
98 982 983 984 985 986 987 987 988 989 99 99 992 993 994 995 996 997 998 998 999 2 2 22 23 24 25 26 27 28 29 29 2 98 982 983 984 985 986 987 987 988 989 99 99 992 993 994 995 996 997 998 998 999 2 2 22 23 24 25 26 27 28 29 29 2.8.6.4.2 5 4 3 2 - -2-3 Normal year (May 27) Drought year (May 29) NDVI (98-2) - Lower Marsabit ZNDVI (982-29) - Lower Marsabit (2) Selecting index: NASA MODIS Normalized difference vegetation index (NDVI) as index Indication of availability of vegetation over rangeland Spatiotemporal rich ( km 2 ) Available in near-real time every 5 day (982-present) 984 987 99 994 997 2 26 29 Motivations Index Insurance Kenya Experience Prospects for Indonesia NDVI (98-2) Lower Marsabit ZNDVI (98-2) Lower Marsabit Karare Logologo Ngurunit Korr Karare Logologo Ngurunit Historical Korr Droughts 8
Motivations Index Insurance Kenya Experience Prospects for Indonesia (3) Quantify insurable loss from index: construct predicted livestock loss from the empirical model: M lt = M ZNDVI lt + ε lt ZNDVI lt Oct Nov Dec Jan Feb Mar Apr May June July Aug Sep Cumulative ZNDVI lt 9
Motivations Index Insurance Kenya Experience Prospects for Indonesia (3) Quantify insurable loss from index: construct predicted livestock loss from the empirical model: M lt = M ZNDVI lt + ε lt ZNDVI lt Oct Nov Dec Jan Feb Mar Apr May June July Aug Sep Cumulative ZNDVI lt Construct X ndvi ls variables from cum. ZNDVI process Czndvi_ pre s zndvi ds s d T pre Czndvi_ pre s zndvi ds s d T pre Czndvi_ pos s zndvi ds s d T pos Czndvi_ pos s zndvi ds s d T pos
Motivations Index Insurance Kenya Experience Prospects for Indonesia (3) Quantify insurable loss from index: construct predicted livestock loss from the empirical model: M lt = M ZNDVI lt + ε lt ZNDVI lt Oct Nov Dec Jan Feb Mar Apr May June July Aug Sep Cumulative ZNDVI lt Regime switching model for zone-specific, seasonal mortality prediction: M lt = M X ndvi lt + ε lt if Czndvi_pos lt γ (good climate regime) M 2 X ndvi lt + ε 2lt if Czndvi_pos lt < γ (bad climate regime) M ZNDVI lt
Motivations Index Insurance Kenya Experience Prospects for Indonesia (3) Quantify insurable loss from index: construct predicted livestock loss from the empirical model: M lt = M ZNDVI lt + ε lt Predictive Performance of predicted livestock loss, M ZNDVI lt Out-of-sample prediction errors within +-% (especially in the bad year) Predict historical droughts well 2
Density of predicted mortality index Motivations Index Insurance Kenya Experience Prospects for Indonesia (4) Identify optimal contract structure Insurable loss: Area average livestock loss indicated by M ZNDVI lt Seasonal indemnity payment: Π lt M θ lt M, TLU, P TLU = max M ZNDVI lt M, TLU P TLU Predicted Seasonal Mortality Index Laisamis Contract Predicted mortality index Predicted mortality index Coverage: Division level, annual contract (covers two seasonal payouts) (5) Actuarial fair premium: (% of sum insured) 3
.4.8.6.8.2.4.6.8.2.2.4.4.6.6.8.8 Cumulative Cumulative Probability Probability.2.2.4.4.2.6.4.8.6.8.2.4.6.8 Cumulative Cumulative Probability Probability Probability.2.2.2.4.4.4.2.2.6.6.4.8.4.8.6.6.8.8 Cumulative Probability.2.2.4.4.6.6.8.8 Cumulative Cumulative Probability Probability.2.4.2.6.4.8.6.8 Motivations Index Insurance Kenya Experience Prospects for Indonesia (6) Ex-ante contract evaluation: simulations of stochastic dynamic model based on observed household dynamic Minimal IBLI data Performance for Very Small Herd Minimal IBLI Performance for Very Small Herd Pastoral production function: IBLI Stabilizes Pathway toward Growth for Herd Around Critical Threshold Minimal IBLI Performance for Very Small Herd Minimal IBLI Performance for Very Small Herd Herd dynamics with stochastic environment: Household budget No Insurance constraint: % IBLI 5% IBLI 2345 2 4 6 8 TLU (Beginning Herd = 5 TLU, Beta = ) Minimal IBLI Performance for Very Small Herd IBLI Stabilizes Impedes Asset Pathway Accumulation toward Growth for Herd for Herd Around Around Critical Critical Threshold Threshold No Insurance % IBLI Household No Insurance % Intertemporal IBLI 5% IBLI problem: No Insurance % IBLI Minimal IBLI Performance for Very Small Herd 5% IBLI No Insurance % IBLI 5% IBLI Minimal IBLI Performance for Very Small Herd 2345 2 4 5% IBLI 6 8 2 4 6 8 Minimal IBLI Performance for Very Small Herd 2345 TLU 2 (Beginning Herd 4 = 5 TLU, Beta = 6 ) TLU (Beginning Herd = 5 TLU, Beta = ) 8 2345 2 4 6 8 TLU (Beginning Herd = 5 TLU, Beta = ) TLU (Beginning Herd = 5 TLU, Beta = ) IBLI Eliminates Probability Kargi ( Minimal Beginning of Falling IBLI into Performance Destitution herd = 2 for for Herd TLU, Very Around Beta Small Critical = Herd ) Threshold IBLI Eliminates Probability of Falling into Destitution for Herd Around Critical Threshold Kargi ( Minimal Beginning IBLI Performance herd = 2 for TLU, Very Beta Small = Herd ) No Insurance % IBLI 5% IBLI Kargi (Beginning Herd = 3 TLU, Beta = ) IBLI Reduces Probability of Extreme Herd Loss for Very Large Herd No Insurance % IBLI No 5% Insurance IBLI % IBLI 5% IBLI No Insurance 5% IBLI % IBLI No Insurance No 5% Insurance IBLI 5% IBLI % IBLI % IBLI No No Insurance % % IBLI IBLI 5% 5% No IBLI Insurance IBLI % IBLI 5% IBLI Minimal IBLI Performance for Very Small Herd 2 2 4 4 6 6 8 8 2345 TLU 2 (Beginning Herd 4 = = 5 5 TLU, Beta = = ) 6 ) 8 TLU (Beginning Herd = 5 TLU, Beta = ) 2345 2 4 No Insurance 6 % IBLI 8 5% IBLI TLU (Beginning Herd = 5 TLU, Beta = ) 2345 2 4 6 8 Herd (TLU) TLU TLU (Beginning (Beginning Herd Herd = 2 = TLU, 5 TLU, Beta Beta = ) = ) 2345 2 4 6 8 No Insurance % IBLI 2345 2 4 6 8 5% IBLI TLU TLU (Beginning (Beginning Herd Herd = 5 TLU, TLU, Beta Beta = ) ) 2345 2 3 4 5 6 7 8 Herd (TLU) TLU TLU (Beginning (Beginning Herd Herd = 3 2 = TLU, 5 TLU, Beta Beta = ) = ) Kargi In most ( Minimal Beginning IBLI case, Performance herd = 2 insured for TLU, Very Beta Small = Herd ) herd SOSD uninsured herd: insurance reduces prob. of extreme loss IBLI Eliminates Probability of Falling into Destitution for Herd Around Critical Threshold No Insurance % IBLI 5% IBLI Contract seems to be effective despite the existence of basis risk! 4
Premium (% of insured value) Price Elasticity of Insurance Demand Motivations Index Insurance Kenya Experience Prospects for Indonesia (6) Ex-ante contract evaluation: Willingness to pay experiments (2 hhs) Demand determinants (+) familiarity with fn. product (+) with interacting financial experience with risk aversion (+) perceived loss profile (+) expected loss (+) wealth (wealth eff.) (-) perceived basis risk (+) credit constraint (buffer stock) Premium Vs. Chosen Insurance Coverage 9% All sample (E =.75) Poor group (E=.77) Vulnerable group (E =.9) Rich group (E =.6) 8% 7% 6% 5% % 25% 5% 75% % Chosen Coverage (% of household's total herd) Modest demand exists at 2%+fair Less elastic among the rich 5
Motivations Index Insurance Kenya Experience Prospects for Indonesia (7) Develop education and extension tools: using experimental games with real incentives Replicate the pastoral livelihood in the community Teach how this insurance work and how it will affect herd dynamics The game also allows us to study hh s behavior responses from insurance! 6
Motivations Index Insurance Kenya Experience Prospects for Indonesia (8) Identify cost effective delivery mechanisms to remote clients using mobile technology The contract has been commercialized in northern Kenya since 2 Contracts sold to among % of populations in the first year Local insurance company underwrites the contract with Swiss Re 7
(9) Long-term micro-level impact assessment 4-year panel household survey, baseline (29) with annual repeat Challenges: (i) cannot randomize eligibility for insurance (ii) low uptake reduces power of estimating avg. treatment effects Hence quasi-experiment with encouragement design: use IV approach with multiple instruments (to generate variation in insurance purchase) We randomize 3 instruments: () Insurance education (e it ) (2) Eligibility for cash transfer (t it ) (3) Discount coupon at -6% (d it ) Survey instruments: welfare, Induced behavior responses, formal/informal access to credit, social insurance, environmental impacts Empirical estimations of demand determinants and impacts of insurance: First stage: Second stage: Motivations Index Insurance Kenya Experience Prospects for Indonesia D it = γ + γ e it + γ 2 t it + γ 3 d it + ε it Cash transfer Educated 4 sites 4 sites Y it = ρ + ρ D it + ρ 2 X it + D it X it ρ 3 + δ i + ε it No cash transfer Not educated 4 sites 4 control sites Stay tuned! 8
Motivations Index Insurance Kenya Experience Prospects for Indonesia References Barrett, C.B. et al. (28) Altering Poverty Dynamics with Index Insurance, BASIS Brief 28-8. Barrett, C.B. and M.R. Carter (27) Asset Thresholds and Social Protection, IDS Bulletin 38(3):34-38. Carter, M.R. et al. (28). Insuring the Never-before Insured: Explaining Index Insurance through Financial Education Games, BASIS Brief 28-7. Chantarat, S. et al. (forthcoming) Designing Index-based Livestock Insurance for Managing Asset Risk in Northern Kenya. Journal of Risk and Insurance Chantarat, S. et al. (2) The Performance of Index-based Livestock Insurance: Ex Ante Assessment in the Presence of a Poverty Trap. Mimeo Chantarat, S. et al. (2) Willingness to Pay for Index-based Livestock Insurance: Results from a Field Experiment. Mimeo IBLI official site: http://livestockinsurance.wordpress.com/ 9
Prospects for Index Insurance in Indonesia Interesting research questions Motivations Index Insurance Kenya Experience Prospects for Indonesia The optimal contract design as part of existing risk management system (complementarities with self-, informal-insurance, government programs) Impact assessment on welfare, productive investments, existing risk management mechanisms Designs of financial educational tools Viability of flood index insurance (e.g., using satellite imagery?) as part of overall flood management system 2