THE WORLD BANKTHE WORLD BANK

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

February 2007

India Country Management UnitSouth Asia Finance and Private Sector Unit

The World Bank

INDIA – National Agriculture Insurance Scheme:Market-based solutions for better risk sharing

Disclaimer

“The report has been discussed with the Government of India but does not necessarily bear their approval for all its contents, especially where the Bank has stated its judgements/opinions/policy recommendations”.

Cover Photo Courtesy: The World Bank – Rajasthan Report

Designed & Printed by: Macro Graphics Pvt. Ltd., www.macrographics.com

Abbreviations

AAFC Agriculture and Agri-Food Canada

AFSC Agriculture Financial Services Corporation

AICI Agriculture Insurance Company of India

APF Agricultural Policy Framework

APH Actual Production History

APR Assumed Pure Rate

BBF Balance Back Factor

BP Base Pure Rate

CCE Crop Cutting Experiment

CCIS Comprehensive Crop Insurance Scheme

CHU Corn Heat Unit

CIF Crop Insurance Fund

CP Catastrophic Pure Rate

CV Coefficient of Variation

FCIC Federal Re-insurance Fund

FP Final Pure Rate

FRF Federal Re-insurance Fund

GIS Geographic Information Systems

GOI Government of India

GDP Gross Domestic Product

GRP Group Risk Plan

HE Hail Endorsement

IQR Inter-Quartile Range

IU Insurance Unit

LOM Lack of Moisture

MPCI Multi-Peril Crop Insurance

MPPI Multi-Peril Production Insurance

NAIS National Agricultural Insurance Scheme

NASS National Agricultural Statistics Service

NDVI Normalized Difference Vegetative Index

NOAA National Oceanic and Atmospheric Adminis-tration

NTM Normal Theory Method

PACS Primary Agricultural Cooperative Societies

PRF Provincial Re-insurance Fund

PVI Pasture Vegetative Index

PY Probable Yield

RAF Risk Adjustment Factors

RFP Request for Proposal

RMA Risk Management Agency

SAS Statistical Analysis Software

SCC Selected Crop Climate

TA Technical Assistance

TY Threshold Yield (also called guaranteed yield)

UP Unbalanced Pure Rate

U.S. United States of America

USDA United States Department of Agriculture

UT Union Territory

Acknowledgements

The report was prepared by a team led by Niraj Verma (Finance and Private Sector Development, South Asia Region, World Bank) and Olivier Mahul (Finance and Private Sector Development, Financial Markets for Social Safety Net, World Bank, co-task lead), with inputs from Richard McConnell

(Consultant, DYMAC Risk Management Solutions), Avery Cook (Consultant, KALA Risk Management) and Tim Watts (Consultant, Watts and Associates) and his team. This technical assistance (TA) report has been prepared as part of an on-going TA to the Agriculture Insurance Company of India (AICI) to enable them to make a transition to an actuarial regime. This report focuses on the National Agriculture Insurance Scheme and is a stand alone report, though the work also entailed the preparation of supplementary technical annexes (not published). This TA is being followed by a TA on weather indexed insurance and portfolio risk management.

The report benefits greatly from the data and information provided by Agriculture Insurance Company of India and special thanks are due to Mr. M. Parshad and Mr. K. N. Rao and also all the participants from the Agriculture Insurance Company of India, who had joined a meeting to discuss the findings of a draft version of this report.

The authors are grateful to the peer reviewers, Rodney Lester (FPDSN) and Dina Umali-Deninger (SASSD) and also to Deepak Ahluwalia (SASSD) who provided useful comments. This report has been prepared under the guidance of Sadiq Ahmed (Director, Finance and Private Sector Unit and Poverty Reduction and Economic Management, South Asia Region), Barbara Kafka (Director, Operational and Quality Services) and Simon Bell (Manager, Finance and Private Sector Unit, South Asia Region). Guidance from Priya Basu (Lead Economist, Finance and Private Sector Unit, South Asia Region) is also acknowledged. Heather Fernandes, Mansi Handa and Vinod Satpathy (South Asia Finance and Private Sector Development Unit) provided administrative support. Funding support of the Swiss Agency for Development and Cooperation is also gratefully acknowledged.

Executive Summary ixChapter 1: Introduction 1Chapter 2: Crop Cutting Experiments 5 Review of the CCE Process 5 Relationship between Sample Sizes and Yield Estimate Accuracy 10Chapter 3: Improving the National Agriculture Insurance Scheme 13 Insurance Unit 13 Guaranteed Yield 15 Indemnity Limits 17 Timely Payment of Claims 17 Operational Deadlines 18 Additional Program Feature 19 Risk-area Boundaries 20 Actuarial/Insurance Principles 22Chapter 4: Premium Ratemaking 27 Data Review, Analysis and Preparation 27 Proposed Ratemaking Methodology 30 Premium Loading Factors 34 Limitations/Factors Affecting Actuarial Soundness 36 Other Rating Issues 36 Further Improvements of the Suggested Premium Ratemaking Method 38Chapter 5: Conclusions and Suggestions 39 Findings and Suggestions 39 Suggested Action Plan 43Bibliography 45Glossary 47Annex 1: Crop Risk Maps for Selected Crops and States 49

T a b l e of Contents

TABLES

Table 1.1: Selected Crops 4Table 2.1. Summary of Yield Radii at 95 Percent Confidence for Selected States,

Crops, and Years 11Table 4.1: Current and Suggested Premium Rates for Sample Crops 36

BOxES

Box 1.1: Main Features of NAIS 1Box 3.1: Time Series and PY Estimation 16

FIGURES

Figure 1: Illustrative Crop Insurance Cycle* xiFigure 1.1: Farmers Covered under NAIS 2Figure 1.2: NAIS Loss Ratio (Indemnities/Premia) 2Figure 3.1: Estimation Error for Area-yield Estimates Versus Sample Size with Reduced

Variance for Gujarat Cotton 15Figure 3.2: Functional Flow Chart – Production Insurance in Alberta 23Figure 3.3: Flow Chart of Documentation and Information in India 25Figure 4.1: Base and Catastrophic Loss Layers 31Figure 4.2: Flow Chart of the Experience-based Ratemaking Methodology 32Figure 5.1 Illustrative Crop Insurance Cycle 40

T a b l e of Contents

Introduction |ix|

Executive Summary

Background and ObjectivesFor 60 percent of India’s total population that is dependent on agriculture for a livelihood, crop insurance forms a potentially critical element of risk mitigation. Given this strategic importance, it is no surprise that Government of India (GOI) has historically focused on crop insurance as a planned mechanism to mitigate the risks of natural perils on farm production. Encouragingly, it has recently indicated a commitment to move forward with significant enhancements to its main program, the National Agriculture Insurance Scheme (NAIS), implemented by the Agriculture Insurance Company of India (AICI) with a focus on transitioning to an actuarial regime. This is potentially a major initiative given the significant scale of NAIS – which while lower than what is required, is nevertheless very large in terms of absolute numbers with around 18 million farmers insured last year, making this the largest crop insurance program in the world in terms of insured farmers.

The broad structure of NAIS has several sound elements and GOI should be complimented on this. The NAIS is based on an indexed approach, where crop yield of a given area (insurance unit, IU) is the index used. The insurance is mandatory for all farmers that borrow from financial institutions (though insurance cover is also available to non-borrowers). The actual yield of the insured crop in the IU is compared to the threshold yield (TY) computed using actual data for previous years (between 3–5 years depending on the crop), and if the former is lower than the latter, all insured farmers in the IU become eligible for the same indemnity payout. While individual farm insurance would have minimized basis risk, it

would be well nigh impossible in a country with so many small and marginal farms. Further, the method of using an ‘area-based approach’ has several other merits: most importantly it addresses the crucial issues of moral hazard and adverse selection.

However, the program as currently implemented entails an open-ended fiscal exposure for government. The annual loss ratio (indemnity/premium) has always been higher than 100 percent, that is, the total indemnities paid to farmers exceed the premia received (including premium subsidies). This is a direct consequence of the caps imposed on the premium rates of oilseeds and food crops: less than 1.5 percent and 3.5 percent, or the actuarial assessed rates, for food crops and oilseeds respectively. This has meant that AICI bears part of the claims, and the rest is borne by the state and central governments, which leads to an open-ended fiscal exposure for government. And it is this characteristic that makes India different from similar programs elsewhere which do not entail an open-ended exposure for government and where the implementing agency, unlike AICI, is not free of the insurance risks.

Further, indemnity payments under NAIS tend to get extremely delayed. The average time for the payout is a year, implying that farmers who have suffered crop losses may not get compensated on time and as a result may not be able to service their crop loans from banks, thereby becoming defaulters. This delay in claims settlement is one reason why the off-take of the NAIS, while impressive in absolute terms, is low compared to the total number of farmers in the country. And while there are several reasons for

|x| India – NAIS: Market-based Solutions for Better Risk Sharing

this delay (see Figure 1), one key reason is derived from there being no reliable actuarial estimates. As a result, it is not possible for Government to budget ex-ante for such exposures, and hence when indemnities need to be paid, time consuming Government administrative and budgetary processes for ex-post funding of the residual claims, lead to delays in claims settlement. Another cause for the delay in indemnity settlements is the long time and process entailed in the measured yield data being consolidated and shared with AICI.

To address these issues, Government has proposed to move to an actuarial regime for crop insurance. Premia would be charged by AICI on a commercial basis and Government’s support, where necessary, would provide up-front premium subsidies (though not for commercial/horticultural crops) differentiated by the economic category of farmer. AICI would receive “up-front” premium subsidies and would be responsible for all claims. This will help address the issue of delayed indemnity payments being made to farmers, since Government contribution, which currently leads to considerable delays in settlements, would be made up-front. Such a sound financial and actuarial approach will also result in introducing more discipline to the Indian crop insurance program and more efficient targeting of subsidies for poorer farmers.

In this overall background, Government/AICI requested the World Bank for technical assistance (TA) in developing an actuarially-sound rating methodology and improving the contract design of the NAIS. To move to this proposed actuarial regime, a key element of the Bank TA, entails proposing a rating methodology that AICI could use to derive actuarial premium rates for the crops it insures.1 Further, the TA focuses on the contract design of the NAIS to address

1 This is part of an overall TA request that also includes support to: (i) develop new products (e.g., weather insurance products) and their ratemaking procedures; and (ii) develop a portfolio risk management strategy for AICI’s book of business, apart from the technical support to the area-yield based NAIS program. The additional tasks have been initiated and a separate report will be submitted later.

some of the other issues – including delays in claim settlements – that the program faces at present.�

Given the highly technical nature of the TA, the approach used was to work collaboratively with AICI and bring in the skills, technology and capacity for the development and pilot testing of the area-yield rating methodology. Selected crops in a few states were chosen with the intention that the rating methodology developed for them could then be tested and applied by AICI to other crops/states.

The main audience for this report is the Ministry of Agriculture, the Insurance Division of the Ministry of Finance, and AICI, as the report provides both policy recommendations on the overall design of the agriculture insurance scheme as well as technical and actuarial recommendations on the pricing of crop insurance.

Findings and Suggestions

The key findings build on the overall sound “area-yield” based approach of the NAIS. As stated above, the intention is to provide AICI with suggestions for its consideration that could help make its crop insurance program actuarially-sound and more efficient. These suggestions, particularly those that could be implemented in the near term, are summarized below (details of suggestions for the longer term are provided in the following chapters) under three broad headings: (a) Premium ratemaking; (b) Improving the NAIS; and (c) Crop cutting experiments (CCEs). It is expected that improvements in the NAIS and transitioning to an actuarial regime will yield benefits for farmers, Government and AICI. Figure 1 presents a snapshot of some of the potential benefits that may be derived from a successful implementation of the suggestions that follow.

� The TA has a technical emphasis and focuses on the design of the crop insurance program; implementation issues related to the crop cutting experiments for example, while covered in the discussion in the report, are not the focus of this report.

June July Aug Sep Oct Nov Dec Jan Feb March April May June July

Adverse selection possibility

- DESIGN PARAMETERS Need for process efficiency Need for process efficiency

- FARMER RELATED EVENTS

- OTHER EVENTS

Reduced adverse selection possibility

July Aug Sep Oct Nov Dec Jan Feb March April May June July

June

* Illustration is for a medium duration crop; ** For borrowing farmerKharif

CCE data toAICI

Data and claimsprocessing by AICI

Final indemnitypayment by AICI

Ex-ante Governmentfinancialcontribution

Partial indemnitypayment faciliatedthrough early trigger

Farmer sows cropFarmerharvests

Crop cuttingexperiments(CCEs)conducted

CCE data to AICI

Farmer sows crop Farmer harvests Final indemnitypayment tofarmer

New contractdesign features

Farmer receivespartial indemnity

Government contributionreceived and indemnitypayment made

Deadline forinsurance purchase

Final indemnitypayment to farmer

Crop cuttingexperiments (CCEs)conducted

Data, claimsprocessing

Deadline forinsurancepurchase**

WHAT COULD HAPPEN IN AN ACTUARIAL REGIME

Actuarial regime:Actuarial premiumrates:

Better contract design:

Help Government/AICI ascertain premium rates that reflectthe true cost of riskProvide information for better economic signalling byGovernmenton agri-policyEnable Government to provide upfront contributions topremium subsidies, thereby enabling better fiscalmanagementFacilitate farmers to get quicker final settlement ofindemnities, due to upfront government contributionEnable AICI to build adequate technical reserves, improvepossibilities for international reinsurance, assume insurancerisks and operate on amarket basis

Enables early partial settlement of indemnitiesMore transparent terms and conditions of the insurance policyTechnically sounder insurance productwith lower possibilitiesof adverse selection andmoral hazard

WHAT IS HAPPENING NOW

Actuarial rates

(a) Premium ratemaking

Determining a sound actuarial rating technique is critical to assess the true cost of risk. However, this is not always easily accomplished. One of the main reasons why multi-peril crop insurance has not succeeded universally, is the sheer complexity of risk

and the lack of adequate risk-modeling technology to understand agricultural risks, in particular, the impact of natural disasters. As crop yields greatly vary from one area to another for the same weather conditions and same crops, many attempts to use standard rating methodologies, such as the Normal

Executive Summary |xi|

Figure 1: Illustrative Crop Insurance Cycle*

|xii| India – NAIS: Market-based Solutions for Better Risk Sharing

Theory Method (NTM), have led to not fully accurate (and often under-estimated) premium rates.

It is therefore, suggested that AICI could consider using an experience-based approach for rating purposes. This would constitute a premium rating methodology consistent with international best practices and also Indian conditions. The methodology is designed to achieve actuarially-sound premium rates that are stable yet reflective of regional differences and responsive to changes in risk over time. To achieve this goal under the experience-based approach, risk is divided into a base layer and a catastrophic layer, each rated separately. Premium rates have been calculated for sample crop/states. The sample rates are, on average, 18 percent higher than those derived from the current NAIS rating method.

This premium rating methodology could benefit all key stakeholders: Government, AICI and the farmers. In particular, actuarially-sound premium rates reflect the true cost of risk, that is, the price of the underlying risk exposure of any agricultural business activity. This could help the government to (i) reduce its fiscal exposure as it can better forecast public financial support, for example, through up-front premium subsidies which would mean that all the residual risks would be borne by AICI, thereby paving the way for a more ‘market’ based mechanism for crop insurance; and (ii) develop a more cost-effective agricultural subsidy program as subsidies can be better targeted, for example, to catastrophic risks. It could also help the insurance company AICI to build up adequate technical reserves to cover their insurance risks, expand outreach amongst farmers and access re-insurance markets. Finally, it would benefit farmers because it would allow for a more timely payment system and, ultimately, a more equitable crop insurance subsidy scheme.

This approach will enable ex-ante fiscal management for Government. Up-front premium subsidies derived from the actuarially determined premium rates will allow the government to budget its financial support to AICI at the end of the insurance policy sales season. However, it is suggested that the

Government act as a re-insurer of last resort for catastrophic risks (e.g., actuarially-based loss ratio higher than �00 percent) until adequate catastrophe reserves are built up.

(b) Improving the NAIS

Risk differentiation

AICI could use the premium rating methodology to differentiate the underwriting risks. The suggested ratemaking methodology would allow AICI to develop actuarially-sound premium rates that provide a more accurate measure of the yield risk exposure. These rates could be used to differentiate the underwriting risks.

To ease administrative management while preserving risk differentiation, AICI could set the insurance premium rates by crop at the district level and adjust the indemnity limits by IU. The implementation of insurance premium rates at the IU level may turn out to be impractical and administratively difficult, given the very large number of IUs. Hence, AICI could select a single premium rate at the district level and adjust indemnity limits at the IU level. These differentiated indemnity limits will help maintain risk differentiation and thus equity among farmers.

Improving the timeliness of indemnity payments

Timely indemnity payments are vital for farmers since they provide cash at the time when it is needed most. In the context of their lack of adequate access to formal finance, this could prevent them from falling into a debt trap or having to pay high interest rates on moneylender loans. And timely payments would improve the chances of access to credit for future crop seasons. The following measures could therefore assume considerable importance.

To facilitate prompt payment of final claims, Government could contribute to up-front premium subsidies based on the suggested premium rates. The Government could pay the difference between the suggested actuarially-based rates and the capped

(subsidized) premium rates; this contribution could be released to AICI at the beginning of the crop season factoring in the expected off-take. Such up-front payments could allow AICI to facilitate prompt payment of final claims, thus making the NAIS more attractive to farmers. It is estimated that such a measure could reduce delays in claim settlement by four to five months, thereby reducing the overall settlement time by as much as about 50 percent.

Government could streamline the yield estimation process. Delays in receipt of yield estimates from the states, is another key underlying reason for delays in claim settlement. It currently takes state governments around two months from the time of the raw CCE data collection to submit the crop yield estimates to AICI. Government could institute measures to minimize these delays possibly through use of technology (e.g. computerization and use of hand held devices) and better data consolidation practices.

Advanced indemnity payments even prior to harvest could be made based on weather and/or remote sensing indices. This approach will enable reaping the benefits drawn from combining the best features of both area-yield (e.g., more accurate loss estimates) and weather-based insurance (e.g., faster claim settlement). Since AICI, with the assistance of the World Bank, is currently developing actuarially-sound weather-based insurance products, the implementation of such a measure could be possible for the Kharif �007 season.

Guaranteed yields

Guaranteed yields could be based on a long term average of at least 10 years, with a technology adjustment for yield trending. A longer time series than what is currently used, will add more stability in the insurance coverage. Such a measure will, possibly, bring more stability in the participation, as non-borrowing farmers would not be able to adversely select against the insurance program by participating when coverage is high or not insuring when coverage is low. It is suggested that AICI could consider developing an yield trending methodology, based on international best practices. However, till

this is developed, during the transition phase, the guaranteed yield could be calculated with a ten-year moving average, with no yield trending adjustment.

Size of IU

If the Government wants to reduce the size of the IU, it is suggested to first develop a credible probable yield (PY) and ratemaking methodology for smaller units. However, this may take time and requires a means to index the smaller unit to a larger region for coverage stability. Delineation of the country or state by agro-climatic zones may be useful in reducing the basis risk of area-yield-based estimates but requires a significant investment in a data management system to move forward. If Government priorities lead to lowering the unit before actuarially-sound premium rates are developed, a useful approach may be to blend market-based insurance objectives and social objectives. AICI could retain claims assessed on an actuarially-sound basis at the existing IU level, while residual claims (i.e., the difference between claims reported at the smaller IU level and the claims reported at the current IU level) would be covered by the Government as a social benefit.

Operational deadlines

The Government could institute a purchase deadline for crop insurance in advance of the crop season, for both borrowing and non-borrowing farmers. For example, for the Kharif season, it is suggested to move back the cut-off date from September to July. This early deadline for insurance purchase would reduce adverse selection among non-borrowing farmers. However, the reduced time for purchasing insurance could also result in a trade-off with the outreach of the program. AICI could address this through improved communication and this might also get addressed through the likely increased use of the Kisan Credit Card.

The Government could also encourage an earlier insurance sign-up through premium discounts. AICI could introduce a premium discount within a certain period well in advance of the growing season. This discount could extend to state and central

Executive Summary |xiii|

|xiv| India – NAIS: Market-based Solutions for Better Risk Sharing

governments in order to encourage their early premium payment to match the farmers’ premium. On the other hand, because premium discounts must be offset with a load to an actuarially-sound premium rate, governments and farmers could instead be surcharged a premium penalty for late sign-up and late payment. At some point, sign-up to the crop insurance program would have to be curtailed, even with a premium surcharge.

Additional benefits

The Government could extend coverage for planting and post-harvest risks. These benefits could be built into the basic crop insurance program or added as endorsements for additional premium. Including these types of benefits as endorsements provides some “individualization” within the area-yield based program design. However, this flexibility requires an administration system that is transparent and readily transfers information directly between the farmer and the insurance company. Building the benefit into the basic crop insurance program, with premium loading for demonstrated risk at an IU level, may be the best way to encourage widespread participation and effective exposure pooling.

On balance it may be better to postpone this benefit until a proper data management system is established. Adding additional enhancements to a scheme that is missing some of the basic support instruments may complicate the process to an extent that an actuarially-sustainable scheme and other key objectives are jeopardized over the long term in order to accomplish relatively smaller short term objectives. If GOI wants to offer these benefits before the development of a proper data management system, since these benefits would not be actuarially rated (for lack of data), the Government could consider them as social benefits and thus bear the associated costs.

Communication

An effective communication strategy is critical in implementing changes to the NAIS. All the suggestions proposed above to improve the NAIS

could potentially induce significant changes in the terms and conditions of the crop insurance policies and improve the design of the NAIS. However, the success of the intended improvements in the program design of NAIS will only be achieved if an effective communication/promotion strategy targeted at farmers is appropriately implemented. Hence appropriate effort and resources will need to be channeled towards this by AICI.

(c) Crop Cutting Experiments (CCEs)

The CCE process is technically sound, although further efforts are suggested to develop a reporting system that ensures timeliness, accuracy and consistency of yield estimates. Specific suggestions include: establish a national CCE procedures manual, ensure that yield losses that cannot be attributed to an insured peril (i.e., it is due to an “uninsurable” cause of loss) are not being recorded for insurance purposes although they can be for policy reasons; possibly consider forwarding raw CCE data results to AICI as the field work is completed to compile a central database that would allow verification of yield estimates on an ongoing basis.

Suggested Action Plan

A set of suggested actions is proposed based on the findings of the TA report which factors in the outcome of detailed meetings with AICI senior management. The short term actions for AICI’s consideration that could be initiated are:

Premium rating. AICI could produce a set of actuarially-sound insurance premium rates based on the suggested ratemaking procedure. This would require enhancing institutional capacity within AICI on actuarial techniques. AICI could also review the impact of new methodology on premium level between regions/crops.Indemnity limits. Indemnity limits could be determined in the light of the suggested actuarially-sound premium rates to create a homogeneous set of premium rates. Should the Government want to implement district

premium rates, indemnity limits could be adjusted at the IU level to ensure homogeneity. On-account payment of claims. Early compensation up to 40–50 percent of likely claims, based on weather indices, could be released during the cropping season.Insurance unit. The suggested ratemaking methodology has been developed at the existing IU level. Should the Government want to lower the IU, the reduction in the IU could be considered as a social benefit and thus the additional costs could be borne by the Government.Guaranteed yield. A ten-year moving average, with no technology adjustment for yield trending, could be used to estimate the PY. Meanwhile, AICI could develop a yield trending methodology to incorporate into TY estimate procedures.Purchase deadlines. Uniform seasonality discipline would be applicable for both borrowing and non-borrowing farmers.Additional benefits. Extending coverage for sowing/planting and post-harvest risks could be postponed until a proper data management system can be established and actuarial premium rates can be established. Should the Government want to implement these

benefits earlier, they could be considered as a social benefit and thus their costs be borne by the Government.Experience database. The Government could consider steps to create a centralized data management system to accept incoming data for all states.

These suggested actions could be implemented by AICI, and would require the support of the Government of India (Ministry of Agriculture and Insurance Division of the Ministry of Finance). The development of an experience database will also require close collaboration between the State Governments (Directorate of Statistics and Economics) and the Central Government. Overall, these suggestions that are focused on design aspects would involve significant changes in the NAIS. For these changes in the program design to yield full benefits and be really successful, issues related to the implementation of the program on the ground (including operational steps on the improvement of the quality of CCEs and on the proposed ex-ante government contribution, etc.) would need to be addressed. Given the complexities involved in this process, it is suggested to pilot-test the proposed actions and, if successful, to expand them countrywide.

Executive Summary |xv|

Introduction |�|

Introduction

Agriculture’s share in Gross Domestic Product (GDP), while declining, still remains significant at around 23 percent in 2004–05

and the sector continues to account for more than 60 percent of the labor force. India has ��6 million farms (operational farm holdings) covering �63 million hectares with a vast majority of farm holdings being small and marginal in size (approximately 80 percent of farmers operate less than 2 hectares) and a significant proportion of such households are below the poverty line. For these reasons agriculture remains an important priority for the Indian Government: the mandate from the last general elections and recent announcements reinforce government’s intention that going forward, considerable attention would be placed on this sector.

The vast majority of India’s ��6 million farms cultivate rain-fed crops and are particularly

Box �.�: Main Features of NAISIt operates on an “area-yield-based” approach: if the observed seasonal area-yield per hectare of the insured crop for the defined Insurance Unit (IU) falls below a specific threshold yield (TY), all insured farmers growing that crop in the defined area will get the same indemnity payments (per unit of sum insured);

The “seasonal area-yield” estimate is determined by harvested production measurements taken at a series of randomly chosen Crop Cutting Experiment (CCE) locations;

The probable yield (PY) is based on a three-year moving average of seasonal area-yields estimated from CCEs for rice and wheat crops and a five-year moving average for all other crops;

Three coverage levels are available and the TY can be set at 60, 80 and 90 percent of the area PY, fixed by crop at the state level, offered, based on coefficient of variation (CV) for yields in the ranges of: greater than 30 percent, �6 to 30 percent, and �5 percent or less, respectively;

The program is available to all states and UTs on a voluntary basis, but once introduced in a state/UT, it must be offered for a minimum of three years;

The scheme is intended to be compulsory for borrowing farmers and voluntary for farmers without loans; and

Farmers have the option of buying additional Rupee coverage to a maximum of �50 percent of the TY multiplied by a defined price (market price or floor price established by government).

vulnerable to the vagaries of the Indian monsoon. An international disaster database3 estimates 90 million people having been affected by drought in the five year period ending in 200�. In this context, agricultural risk management products, particularly for the small and marginal farmers, are of critical importance.

The main instrument for provision of risk management to the farming community has been the Comprehensive Crop Insurance Scheme (CCIS) introduced in �985–86. In �999, this was replaced by the National Agricultural Insurance Scheme (NAIS) which is now offered by the recently established public sector firm, Agriculture Insurance Company of India (AICI). The main features of

3 CRED, International disaster database, Université Catolique de Louvain, Belgium.

Chapter 1 Chapter 1

|2| India – NAIS: Market-based Solutions for Better Risk Sharing

the NAIS include availability to all states/union territories (UTs), coverage of food crops, oilseeds and selected commercial crops and use of an “area-yield” index (see Box �.�). This “area-yield” approach reduces the traditional problems of adverse selection and moral hazard, and lowers the administrative costs relative to traditional, individual yield-based crop insurance. AICI is the only player offering such a product to farmers.

Over the last two years, the NAIS program covered about �3 million farmers during the Kharif season (June to September) and around 4 million farmers during the Rabi season (October to December) (see Figure �.�), that is, the annual crop insurance penetration is �4.5 percent. Small and marginal farmers account for 65 percent of the farmers covered under NAIS.

Despite the large numbers of farmers covered, which makes NAIS the largest program worldwide (even though, as yet, a large proportion of farmers

are not insured), several problems need to be addressed. The demand for crop insurance is concentrated in the states where crops grow under rain-fed conditions and natural risks are greater. These states include Andhra Pradesh, Gujarat, Karnataka, Orissa, Uttar Pradesh and Rajasthan. For example, over the period �999–2004, Gujarat received 28.3 percent of the nationwide indemnity payouts while accounting for only �6.7 percent of the nationwide liability. Similarly, Karnataka captured �6.5 percent of the total claims while its liability was only 8.6 percent.

Further, since its inception, the annual loss ratio (indemnity/premium) has been always higher than �00 percent, i.e., the total indemnities paid to farmers exceed the premia received (including premium subsidies). The average loss ratio was higher than 400 percent over the period 2000–2004 (see Figure �.2). This is a direct consequence of the caps imposed on the premium rates of oilseeds and food crops: less than �.5 percent and 3.5 percent,

Figure �.�: Farmers Covered under NAIS

Source: Data from AICI

8.4 8.7 9.8 8.012.7 12.7

2.1 2.02.3

3.5 4.0

0.02.04.06.08.0

10.012.014.016.018.0

2000 2001 2002 2003 2004 2005

mil l

ion Rabi

Kharif0.6

Figure �.2: NAIS Loss Ratio (Indemnities/Premia)

Source: Data from AICI

591%

189%

560%

224%142%

214% 214%

489%408%

0%

100%200%300%400%

500%600%

2000 2001 2002 2003 2004

KharifRabi

Introduction |3|

or the actuarial assessed rates, for food crops and oilseeds respectively.

Yet another critical problem has been the long delay in payment of indemnities. This has been a result of the time for the CCE data to be collated but perhaps more importantly, on account of state and central governments’ inability to expeditiously contribute to claim settlements, since they have typically not budgeted adequately for such liabilities. This has meant that farmers not receiving claims payment on time may have defaulted on their loans from banks and become ineligible for loans for the next crop cycle. This has also contributed to the relatively low off-take of crop insurance and despite significant increase in outreach in recent years, means that a long way needs to be traversed yet.

To address these issues, the Government of India (GOI) is reviewing the NAIS with dual objectives of making the scheme more attractive to farmers (especially in terms of timely payments) so as to increase the crop insurance penetration levels, and, to place the scheme on actuarial regime. According to Government’s plans, the crop insurance penetration which was about �5 percent during 2004–05 would be increased to 25 percent by 2006–07 and 50 percent by 20��–�2. This targeted increase in coverage/outreach has largely driven the move to an actuarial regime; in the absence of such a move, increased numbers of insured farmers would have meant an unsustainable level of fiscal burden without an accurate method to budget for this. Premia would be charged on a commercial basis and the Government’s support, where necessary, would provide up-front premium subsidies (though not for commercial/horticultural crops) differentiated by the economic category of farmer. AICI would receive up-front premium subsidies and would be responsible for all claims.

Properly functioning crop insurance could not only improve access to credit for farmers through reducing risk for lenders and timely payments of indemnities, but also improve resource allocation and fiscal management. With some careful attention,

the Indian crop insurance program could more effectively contribute to the rural sector. Given a renewed focus for crop insurance in India, proposed modifications to NAIS being considered include:

Altering the yield estimate to a longer moving average time series;Reducing the IU to a village Panchayat for major crops;Instilling actuarial principles into premium rate methodology;Rationalizing the coverage levels;Extending coverage to include sowing/planting failure and post-harvest damage to crops caused by adverse weather conditions;Introducing pre-payment of indemnities based on index-based insurance approaches; Assessing local calamities on an “individual plot” basis.

The Technical assistance (TA) was requested by AICI/Government in this context. The overall objective of the study and this report is to offer technical assistance to AICI in order to assist the insurance company in moving towards a market-based approach in the design of actuarially-sound area-yield insurance products. The TA aims at improving further the contract design of insurance products and suggesting a methodology to AICI to develop insurance products designed and rated with actuarially-sound actuarial techniques using lessons from international best practices. The main audience for this report is the Ministry of Agriculture, the Insurance Division of the Ministry of Finance and AICI, as the report provides both policy recommendations on the overall design of the agriculture insurance scheme as well as technical and actuarial recommendations on the pricing of crop insurance.

This market-based approach, relying on a sound actuarial regime, could help the government to (i) reduce its fiscal exposure as it can better forecast public financial support; and (ii) develop a more cost-effective agricultural subsidy program as subsidies can be better targeted, for example, to catastrophic risks. It could also help the insurance

|4| India – NAIS: Market-based Solutions for Better Risk Sharing

company AICI to build up adequate technical reserves to cover their insurance risks, expand outreach amongst farmers, and access re-insurance markets. Finally, it would benefit farmers because it would allow for a more timely payment system and, ultimately, a more equitable crop insurance subsidy scheme.

Selected crops in a few states were chosen with the intention that the rating methodology developed for them could then be considered by AICI for application to other crops/states. The list of cases (selected crops/seasons/states) is stated in Table �.�.

The report consists of five chapters, starting with this Introduction. Chapter 2 provides a review of the CCE process. Chapter 3 reviews, based on international experience, the principal modifications of the NAIS proposed by the GOI. Chapter 4 presents the suggested premium rating methodology and illustrates it for selected crops and states. Chapter 5 contains the conclusions and suggestions. The report ends with an annex presenting crop risk maps for selected crops and states. Five supplementary technical annexes, prepared by the team of technical consultants who worked on this TA, are referred to in the report (although are not published) but may be provided on request for reference purposes

Table �.�: Selected CropsState Season CropsAndhra Pradesh Kharif RiceGujarat Kharif Cotton and groundnutMaharashtra Kharif Pigeon peaMaharashtra Rabi Sorghum (dryland)Uttar Pradesh Rabi Wheat

Introduction |�|

Crop Cutting Experiments

India develops its area-yield crop estimates from “cutting” (harvesting) a random sample of �-metre x �-metre plots in fields that are randomly

chosen within villages that have also been randomly chosen within an IU. And while originally these CCEs were not designed specifically for the crop insurance program and were instead focused on providing GOI with agriculture yield estimates to develop policy and food distribution internally as well as for export, these have also been used for the crop insurance program.

This chapter includes a review of the current CCE process and provides a quantitative analysis of the relationship between sample sizes and yield estimate accuracy.

Review of the CCE Process

Overall, the methods used to choose a sample location seem to be sound and Government needs to be complimented on the degree to which its methodology is designed for non-bias. However, its current implementation may be hampered by bureaucratic problems and moral hazard problems, which, while not the focus of this report, are perhaps as critical to address as improving the actuarial design of the overall crop insurance program. In addition, while considerable effort is placed on ensuring that the sampling locations are without bias, further effort may be needed to develop a reporting system that improves accuracy, timeliness, and efficiency of the yield estimates.

Kalavakonda et al. (2003) observed significant delays in paying claims to farmers after loss years. Part

of this delay was attributable to the completion of local CCEs and the subsequent reporting of yield records to AICI for approval and claims processing. Recent information provided by AICI shows that the timing of the CCEs and subsequent reporting has improved. However, significant delays can still occur and, if India is to expand the CCEs and ensure that appropriate field work and reporting is completed accurately and on time, a more streamlined process will be required. In doing this, the process to select villages or fields randomly should not detract from the speed of the sampling process. Presumably, local officials know which fields have been seeded by crop and are available to be part of the CCE network. Consequently, villages could be selected for sampling first and then fields randomly chosen from those available soon after planting.

Accuracy

At present, Government employs individuals at the state and local level to conduct CCEs which produce a single source of data for both policy development and crop insurance throughout the country. The utilization of the single-source yield database is efficient but, since it is designed to gather basic yield information, the process may be biased with respect to an insurance concept.

For example, insurance aims at indemnifying the insured farmer for a loss due to a specified peril. Any other yield loss should not be indemnified even though production in the field is low. The current CCE system assesses the amount of actual yield in the field for policy and planning regardless of factors that influence production. From an insurance

Chapter 2 Chapter 2

|�| India – NAIS: Market-based Solutions for Better Risk Sharing

perspective, the CCE should incorporate a yield assessment in the field that records the reason for a low yield. A yield loss that cannot be attributed to an insured peril (i.e., it is due to an “uninsurable” cause of loss) should not be recorded for insurance purposes, although it can be for policy reasons. Given the importance of this insurance principle, AICI could review this element of the CCE procedures to ensure a consistent application across states and among loss adjusters. In addition, AICI could consider an ongoing and “formalized” monitoring process to ensure consistency in interpretation and implementation. A single-source CCE process with distinctions noted between insurance and policy development data series may provide greater accuracy for the insurance program.

Another point that merits review is that only crops grown under farm management conditions that are normal for an area should be eligible to contribute to the area-yield-based estimate for insurance purposes. For example, crops seeded after an acceptable date, or using technology that is not appropriate for a crop under local conditions, should not be accepted into the pool of statistical information for the area-yield-based insurance scheme. Even though an individual farmer cannot directly control the outcome of an area-yield-based estimate, individual farm actions can influence both the annual area-yield estimate and, indirectly, the PY. This is especially true if the number of field yield estimates is small or reduced to maintain administrative efficiency. However, this recommendation is contingent on the development of a network of loss adjusters to conduct the CCE process for insurance purposes (see below).

Crop management regimes that do not conform to local conditions do not necessarily mean that low production is automatic. However, the likelihood of an insurance claim will be higher and, within an insurance context, this is an important factor to be taken into consideration. These same concerns have no bearing on the outcome of yield estimates for broad policy development. Since the CCE process is designed for this broad application, it is likely that individuals engaged in this endeavor, while

conscientious to the mandate they are given, may not be as attuned to insurance principles.

One way to increase the accuracy of the area-yield estimation from an insurance perspective, is to consider the merits of the development and training of a network of loss adjusters to conduct or oversee the CCE process. The intent here would not be to duplicate a work force but in the medium- to long-term, to train a network of individuals in insurance principles so they can conduct the CCEs with an understanding of the difference between data for broad policy objectives and the data accuracy needed for insurance purposes. This will become increasingly important as India moves toward more individualized claim settlement procedures for local calamities. In Canada, the U.S., Mexico and other countries, there is an in-country network of certified insurance loss adjusters. In some countries, like Mexico, these loss adjusters have expertise in a variety of insurance fields. In general, agriculture loss assessment requires expertise not readily available in other insurance sectors. Consequently, loss assessment firms either specialize in the agriculture field or have an agriculture division.

The Canadian experience, while contextually different from India, may provide some ideas that have relevance from a medium- to long-term perspective. In Canada, the provincial agencies responsible for crop insurance, hire and train contract loss adjusters, many of whom are farmers or have a farming background. For example, in Alberta, an annual course for loss adjusters is conducted each year in conjunction with an agriculture college. Loss adjusters participate in week-long courses that cover weed identification, meteorology, soil classification, chemical damage to crops, negotiation skills, insurance contract content and interpretation, underwriting, insurance portfolio risk management, time management and crop insurance principles. As the year progresses, “in-field” conferences are arranged to provide ongoing training for specific loss situations (e.g. hail, unseeded acreage, re-seeding benefit) that have developed in the area to ensure consistency in loss assessment technique.

The consistency of loss assessment techniques at the provincial level allows a continual transfer of loss adjusters from one region to another as required without interruption. Loss adjusters can move among regions simply by being assigned to a local crew that is managed within the region by a loss assessment supervisor. This fundamental consistency across Canada also allows the transfer of loss adjusters among provinces in the event of a disaster to expedite the claim payment process.

Loss assessment instruction in Canada and the U.S. includes considerable effort to ensure that field adjusters are familiar with causes of loss that are not insurable under individual coverage plans. Anecdotal evidence in Canada indicates that uninsured causes of loss are annually assessed on five to ten percent of claim situations. In addition, within an individual coverage format, loss adjusters visit farms to ensure that non-claim yield reporting is completed accurately. Farm visits can be randomly organized or targeted to address abnormal records in a specific year or individual situations that, in the past, indicated a need for audit or monitoring. These individual coverage loss assessment elements equate to the eligible crop selection and assessment of uninsured causes of loss within an area-yield-based program scenario. Uninsured causes of loss should reduce the annual claim payment but should not detract from insurance coverage developed from the PY estimate.

Timeliness

Individual CCE results could be provided to a central data management facility in a consistent format and on a continuous basis. This continuous reporting process would speed the data verification process and allow for a re-sampling approach in which the sample size is expanded in locations where production is low. Under the current CCE process, yield information has to be verified by AICI before insurance payments can be finalized. However, AICI does not actually see the individual CCE results. The verification process begins once a considerable portion of the CCE process is completed, and at that time, it is difficult

to verify and/or identify data anomalies and their cause. Data quality could be improved with a more timely and continuous reporting process. Improving data quality and the development of a centralized insurance database, within a national insurance approach as in India, would have a number of positive outcomes such as improved:

program performance and client satisfaction due to faster claims payment (including improved access to lending sources for clients);auditing and control to ensure data accuracy;actuarial soundness;ability to predict the impact of changes in program design;ability to provide data analysis and research for policy decisions; andgreater acceptance of the NAIS in the re-insurance community.

In addition, the number of field samples could vary by year depending on the expected production outcome. For example, if 12 fields were selected for sampling in a village, but it was apparent after sampling the first six that production results were above average, those six may be all that is required for planning, policy development and the determination of PYs. Insurance would not be making a claim in this situation so the annual area-yield estimate from the reduced sample size may suffice. In a year where production results from the first six fields appear low, the sample size could be expanded up to the 12 pre-defined and randomly selected field sites to provide a more accurate estimate for the annual area-yield. Since the insurance program would be making an indemnity to insured farmers in this case, an expanded sample would provide added validity and accuracy to the estimate that is of greater importance to an insurance program than the policy and planning effort. The concept of expanded sampling based on the results of an initial sample is known as re-sampling.

India has developed a sound methodology to randomly sample fields to estimate annual crop yield. However, if sample sizes are to be expanded for an increased number of IUs, Government may consider

Crop Cutting Experiments |�|

|�| India – NAIS: Market-based Solutions for Better Risk Sharing

a review of the current �-metre x �-metre field plot size to a more manageable size that still retains the positive attributes of the current methodology, for a single loss adjuster. If appropriate, site selection could be left to the discretion of the loss adjuster with guidance to follow a particular sample pattern in a field. In addition, not all yield sample sites may need to be analyzed for a dockage sample. A random selection of a percentage of fields may be sufficient for this part of the process. Each sample taken by the adjuster in the field could be sealed, identified and sent to a central “threshing and dockage” station which could be equipped with technology to improve the speed of that process.

One method to leverage the field observations may be to utilize remote sensing technology to estimate the relative growth and production of crops. For example, research could be conducted with satellite imagery to predict the health of crop growth during the growing season, with the predictions calibrated using the results of the “in-field” CCEs. Depending on the reliability of the satellite yield model, both the sample size and the number of samples could then be reduced to a smaller level and used solely for calibration. Over time, the satellite research could result in reliable estimates of production as a percentage of the PY for states or sub-regions. These estimates, in conjunction with the CCE process, could increase the predictive capability of the area-yield-based estimates and may eventually be used to make final payments based on satellite information alone. Another alternative would be to initially use satellite predictions to target more intensive sampling where yields appear low. This approach could be used in an ongoing fashion to reduce the cost of the CCE process or as a method to phase-in a satellite-based insurance program. In the phased approach, claims would continue to be paid based on the existing CCE process with the switch to a satellite-based payment scheme implemented once sufficient experience/reliability has been developed with the satellite model. A reliable satellite model would also allow payments to be calculated for areas smaller than the current IU (such as the village Panchayat).

As with remote sensing technology, weather-index strategies could be used to make advanced payments to farmers, to form a part of a dual-trigger insurance scheme, or to initiate target sampling within the CCE process. The Joint Group, set up by the Ministry of Agriculture, recognizes in its report the opportunity to incorporate weather-index strategies into NAIS.

Efficiency

Farmer estimates of harvested production and planted acres are a valuable source of data to establish area-yields and evaluate program performance and future program changes, particularly if a longer and more stable method to establish a PY is introduced. Incorporating farmer yield reporting into the CCE sampling methodology, with a process to audit and remove yield outliers, may be cost-effective, and introduce a direct feedback and education opportunity between farmers and the insurance network through the CCE process. By providing a means for farmers to report yields, several roles are filled, including: (i) a significantly expanded source of production data; (ii) an opportunity for building an ongoing relationship between farmers and the insurance program through local representation that allows the “education focus” to develop; (iii) an atmosphere is created to attract farmer input through local representatives of the crop insurance program and make improvements to the program and delivery mechanisms that are meaningful; and (iv) an expanded database, with farmer input, could be utilized by government agriculture personnel to provide extension services to farmers to identify best practice management techniques that can increase productivity and reduce yield variation due to natural perils.

In Canada and the U.S., farmer yield reporting is an integral component in database development. In Canada, the provinces of Manitoba and Alberta utilize crop insurance data, including farmer reported records, as an extension tool to provide farmers with an added value for participating in the crop insurance program. Each insured farmer receives

the production results of their farm relative to other farmers in the immediate area. Care is taken to ensure that the average yield of local area farmers cannot be used to determine the original source. The crop insurance agency breaks down yield results based on management characteristics reported. For example, an insured farmer may have produced an annual yield for a crop of 1,000 kilograms per hectare based on specific fertilizer use, seed variety and planting date. That result can be compared to other reported yield results in a similar soil zone where different fertilizer applications, seed varieties or seeding dates were used. This comparison provides the recipient farmer with real time research results for actual farming practices as opposed to experimental plots at a research station. Farmers in Canada consider this to be useful information and it reinforces the value of participating in the crop insurance program. Over the medium- to long- term period, India may wish to consider a similar approach when pursuing an objective of increased participation by demonstrating added value for farmers who participate in the insurance program.

Possible further improvements

While the CCE process is technically sound, further efforts could be considered to develop a reporting system that ensures timeliness, accuracy and consistency of yield estimates. In order to further improve it, Government/AICI could consider to:

a) Establish a national CCE procedures manual;b) Ensure that yield losses that cannot be

attributed to an insured peril (i.e., it is due to an “uninsurable” cause of loss) are not being recorded for insurance purposes although they can be for policy reasons. In addition, AICI should consider an ongoing and “formalized” monitoring process to ensure consistency in interpretation and implementation across states and among loss adjusters;

c) Only crops grown under farm management conditions that are normal for an area should be eligible to contribute to the area-yield-based estimate for insurance purposes. For example, crops seeded after an acceptable date or using

technology that is not acceptable for a crop under local conditions, should not be accepted into the pool of statistical information for the area-yield-based insurance scheme;

d) Review the CCE process from an insurance perspective for consistent and accurate yield reporting to ensure actuarial soundness of NAIS and equity among crops and regions. In addition, CCE allocation among regions according to their exhibited level of accuracy should be reviewed to improve efficiency;

e) Forward raw CCE data results to AICI as the field work is completed to compile a central database that would allow verification of yield estimates on an ongoing basis. This continuous reporting process would speed the data verification process and would identify geographic trends in yields through the harvest season and could be used to notify channels within the claims payment function to increase the resources if needed;

f ) Incorporate farmer yield reporting into the CCE sampling methodology and introduce a direct feedback and education opportunity between farmers and the insurance network;

g) Adopt a nationally consistent approach to handle outlier yields, potentially using a Windsorized mean of CCE yields to estimate area-yields by IU.

Should the GOI want to expand CCE sample sizes for an increased number of IUs, it is suggested to:

a) Review of the current �-metre x �-metre field plot size to a more manageable size that still retains the positive attributes of the current methodology for a single loss adjuster;

b) Utilize remote sensing technology to estimate relative growth and production of crops. For example, research could be conducted with satellite imagery to predict the health of crop growth during the growing season, with the predictions calibrated using the results of the “in-field” CCEs. Depending on the reliability of the satellite yield model, both the sample size and the number of samples could then be reduced to a smaller level and used solely for calibration;

Crop Cutting Experiments |�|

|10| India – NAIS: Market-based Solutions for Better Risk Sharing

c) Use satellite predictions to target more intensive sampling where yields appear low. This approach can be used in an ongoing fashion to reduce the cost of the CCE process or as a method to phase-in a satellite-based insurance program. In the phased approach, claims would continue to be paid based on the existing CCE process with the switch to a satellite-based payment scheme implemented once sufficient reliability has been developed with the satellite model. A reliable satellite model would also allow payments to be calculated for areas smaller than the current IU (such as the village Panchayat).

Relationship between Sample Sizes and Yield Estimate AccuracyData is the “lifeblood” of any insurance program. It is particularly important for agriculture insurance schemes since useful data is usually not available from sources outside the insurance program in either a sufficient quantity or quality to support a crop insurance scheme. This section includes an analytical review of the CCE process and comments on the level of accuracy, any bias that might exist in the process, the relationship between sample size and accuracy and the impact of moving to smaller IUs.

Bias

The area-yield estimate for each IU is determined as the average of the yields from several CCE plots which are randomly selected from all possible fields (insured or uninsured) within the area. A multi-stage stratified random sampling technique is used to select the location of experimental plots “with the Mandal as the stratum, villages within the stratum as first stage units, the field in the selected village as the secondary unit of sampling and the plot of the specified size within the field as the ultimate unit of the sampling”.� The sampling procedure described in the manual is sound and if

� See “Manual on Crop Estimation Surveys on Food and Non-Food Crops in Andhra Pradesh”.

implemented consistently provides a non-biased estimate of an area-yield.

Data management

The CCE process is managed separately by each of the states participating in NAIS, resulting in several operational issues which have a direct bearing on the accuracy of the CCE process in estimating area-yields appropriate for use in the NAIS program. Some key issues are discussed below.

Information for individual CCE yield samples is generally not provided to AICI and the area-yield information is often provided after a lengthy delay. This severely limits AICI’s ability to verify and audit information that is fundamental to the insurance program.

Since there is no national CCE procedure manual or training program for field personnel or data managers, so data consistency cannot be verified. There are differences in accuracy among states but it is difficult to determine if this is due to weather events or data management. However, different methods seem to be applied by different states, since yield outliers were excluded from the area-yield calculations in some cases and not in others.

The differences in data structures between states, the lack of a consistent data model and the non-centralized nature of the CCE data within a national program make it difficult to perform an overall assessment of the CCE process. The CCE process is crucial to the effectiveness of the NAIS program both in terms of the accuracy of payments, and in providing equitable coverage for different crops and areas. Since increased yield variability results in larger payments, inequities can easily be created by the CCE procedures.

Accuracy

The level of accuracy of the existing CCE process to estimate actual yields for the IU is analyzed based on the sample crops and states selected for this study.

A �� percent confidence interval for the “true” insurance area-yield has been calculated for each year, state, crop and IU based on the available CCE data (based on an assumption of normality). The radii of these confidence intervals represent a measure of the level of accuracy achieved by the CCE process. Radii are expressed as a percent of the associated yield to facilitate comparison between crops and areas. The explanation for a 22 percent radius value can be expressed as “we are �� percent confident that the true yield for the IU is within 22 percent of the area-yield estimated by the CCE process”. Table 2.1 shows the mean radii across all IUs in a state, for a specific crop and year (Technical Annex 1 for CCE analyses for additional states has more details).

As shown in Table 2.1, the CCE process, on average, results in yield estimates that are within about 33 percent of the true yield for the insurance area (with a �� percent confidence interval). Maharashtra sorghum and Gujarat cotton have the lowest accuracy level (highest radius values, about �0 percent) with Andhra Pradesh rice and Uttar Pradesh Rabi wheat having the highest level of accuracy (lowest radius values, about 20 percent). It is difficult to say whether the differences in Gujarat cotton, Andhra Pradesh rice and Uttar Pradesh Rabi wheat are meaningful or occur simply due to differing weather conditions between states during the data periods involved. However, given that sorghum’s accuracy seems to be less than that of pigeon pea during the same time period in Maharashtra, it is possible that area-yield estimates are less accurate for sorghum. Further investigation

is required to determine if the cause of this difference relates to a greater variability in CCE estimates for this crop or to the geographic location where sorghum is more prevalent.

The mean radius with outlier control displays the mean radius, after outliers are removed, using the standard Inter-Quartile Range (IQR) process.� Assuming that the data removed by this process truly represents outliers, the results in Table 2.1 indicate that accuracy of estimation improves by approximately 20 percent. These results demonstrate that there could be significant benefits to better data management and that developing a consistent approach to handling outliers would likely improve the accuracy of area-yield estimates. The estimation of area-yields could be improved through the use of a trimmed mean (a mean calculated after dropping the high and low value) or a Windsorized mean (a mean calculated after replacing the lowest value with the second lowest value and the highest value with the second highest value.

Integrated data management system

At present, data is held by states and provided to AICI in a summary format to initiate claim payments. The Government could develop a national integrated data management system and a formalized training and

� The Inter-Quartile Range process identifies outliers as outside the range using the formula: Median Yield (+or-) 1.� times (3rd Quartile Yield – 1st Quartile Yield)

Table 2.1. Summary of Yield Radii at �� Percent Confidence for Selected States, Crops, and Years

State Crop Year

Mean 95% radius for

all IUs(%)

Mean 95% Radius with outlier control

for all IUs(%)

Reduction in radius

(%)

Number of IUs

Andhra Pradesh Rice 2003–0� 1�.� 13.� 20.� ���Gujarat Cotton 2000–03 3�.3 2�.� 22.� �1Gujarat Groundnut 1��2–03 33.2 2�.� 20.2 �1Uttar Pradesh Wheat 2003–0� 1�.� 1�.3 13.1 20Maharashtra Pigeon pea 2000–0� 30.� 2�.� 1�.� ��Maharashtra Sorgum 2000–0� �1.2 3�.� 1�.2 ��Simple Average All Average 32.7 25.9 20.7 1,260

Source : Data from AICI

Crop Cutting Experiments |11|

|12| India – NAIS: Market-based Solutions for Better Risk Sharing

monitoring network to support the yield estimation, management, operations and program design functions within NAIS.

An integrated data management network is crucial to the development of new initiatives under the NAIS. At the same time, a consistent understanding of insurance principles and program objectives is integral among insurance functions and personnel to compile an appropriate data set. A consistent and national data set could help to:

assess agro-climatic risk-areas and build credibility for PYs and premium rates to recognize smaller IUs with precision;correlate yield data with weather and/or remote sensing information when introducing advanced payments to the area-yield-based NAIS;

correlate yield data and remote sensing information to incorporate operating efficiencies in the CCE network for an increased number of IUs;accurately assess the impact of yield trending methods across states and crops;introduce an accurate pooling of premium or an equitable assessment of risk attachment by state to a national Corpus Fund; present an accurate assessment of India’s current risk within NAIS to the private re-insurance community in order to access re-insurance; effectively research and design enhancements to NAIS that will be attractive to farmers and increase participation; andutilize the crop insurance program as an efficient base on which to build social transfer policies to the rural poor in catastrophic situations.

Introduction |13|

Improving the National Agriculture Insurance Scheme

Chapter 3

The GOI has reviewed the NAIS with dual objectives of making the scheme more attractive to farmers so as to increase the

crop insurance penetration and to place the scheme on actuarial regime supported by up-front subsidies. This chapter reviews the principal modifications proposed by the GOI and makes suggestions based on this review and on lessons drawn from relevant international best practices on areas including: reduction of the IU; guaranteed yields; indemnity limits; on-account payment of claims; additional operational deadlines; additional benefits; and risk-area boundaries. It also discusses operational features that have a direct impact on the integrity of the insurance program.

Insurance Unit

The Joint Group of GOI suggested reducing the size of the IU to the village Panchayat level for major crops in order to have the area-yield estimates more closely reflect actual farm production and yield variability. Decreasing the size of the IU will increase the number of CCEs required to maintain the same level of accuracy in area-yield estimates. Increasing the number of IUs will also increase the indemnity payments under NAIS.

The IU is the geographical or physical entity at which coverage, premium and indemnity payments are focused. NAIS operates on an “area-yield-based” approach and indemnifies all insured farmers in an area relative to their elected coverage level when the “annual area-yield” estimate, determined by harvested production measurements taken at a series of randomly selected CCE locations, falls below an

area TY. The larger the IU the more likelihood that it will contain a variety of soil types, local weather patterns and farm management systems. Averaging the yields from sample sites to determine an area-yield estimate that accurately reflects individual farmers in the area, becomes more problematic as the IU becomes larger.

On the other hand, reducing the size of the IU means that more yield samples are required to estimate production, especially when harvested production is not stored and easily measured as in India. Many farms in India are small and operate at a subsistence level, meaning that harvesting is a manual process and the resulting crop may be consumed on the farm or sold locally with limited ability to track production records. In the U.S. and Canada, individual crops on a farm are the IU and coverage is determined at the farm level along with indemnity loss adjusting. However, in these countries, farms are large compared to India and production, either sold, stored or fed to livestock on the farm, can be fairly readily tracked. And data management within the crop insurance system provides insight to situations where inaccurate yield reporting may occur.

Review of methods used in other countries

In Canada, while the crop insurance program is on an individual basis, there are parallels to the India situation. Risk-area boundaries are established based on agro-climatic conditions and historical yield information. PYs or “risk-area normal yields” are established for each crop by risk-area over a minimum 20-year time series and adjusted for technology. Farmers report their annual yields

Chapter 3

|14| India – NAIS: Market-based Solutions for Better Risk Sharing

by crop, which are recorded in a provincial data management system. A percentage of farmers who have reported yields without losses are audited by experienced loss adjusters to ensure accuracy of yield reporting. All farms reporting an insurance claim are visited by a loss adjuster to measure production and determine the cause of the insurance loss. Farm yields in the data management system are analyzed relative to others reported in the same area throughout the harvesting period. This analysis will identify obvious discrepancies that can be dealt with through a verification of yield data reported and entered in the system or via a loss adjuster visit to the farm. All farmer yields reported and/or measured are used to develop the risk-area normal estimates. Annual crop yields for the farm are compared to the annual risk area-yield for the same year in a process known as ‘indexing’ to determine a coverage adjustment factor for the farm. That coverage adjustment develops over time and is multiplied by the risk-area normal to determine the farm’s coverage.

In the U.S., each farm establishes a production history through a combination of certifying its own total production and acreage planted in previous years. A proxy yield from the immediate area (county) is used if four years of actual production history cannot or is not certified. Once four yields from the farm are available, coverage for the farm is based on that farm’s own production records. Farmers who participate in the Actual Production History (APH) program in the U.S. must report their harvested production and planted acreage by management practice and unit for each crop insured. Units are not identified using geographic coordinates. Loss adjusters visit farms and measure the production records for the current year in each indemnity situation. They must also determine whether the yields that have been reported for prior years are reasonable. In addition, a random audit is conducted on farms that have not had an insurance claim but which have provided yield information to verify that the records to support that information is available. These audit procedures ensure the integrity of production data submitted to the insurance providers.

Additional information on the Canadian and U.S. crop insurance programs is provided in Technical Annex 2.

Estimated cost impact of reducing the size of the IU

There are two main impacts of moving to a smaller IU. First, the number of CCE experiments required to support a reduction in the IU will increase. Second, the cost of providing benefits will increase since there will be no offsetting of yields remaining between different village Panchayats within an IU since the village Panchayat will be the IU.

The estimation of the increase in benefits and costs is likely to be significant. While the sample data available for this study does not support a detailed analysis of the impact of the CCE requirements associated with a move to a village Panchayat IU, it is possible however to approximate the increase in cost assuming that the number of IUs will increase by a factor of seven times and the standard deviation between village yields within a village Panchayat is approximately 20 to 30 percent less than the standard deviation in yields at the IU level.

Yield variation within an IU at the village level occurs in two ways. First, the variation between CCE plots within the same village and secondly, variation between average yields of villages within the IU. Based on an examination of the components of variance for Maharashtra CCE yields, the variation between CCE plots within villages accounted for 10 to 50 percent of the variation in CCE plots within the IU. This portion of the variation in yield estimates will continue to exist if IUs are set at the village Panchayat level. The variation between average village yields accounted for 50 to 90 percent of total variation. It is reasonable that for smaller insurance areas, the variation between villages might be reduced, potentially by 20 to 30 percent, which implies that the reduction in total variation would be 10 to 30 percent.

By scaling the accuracy curves for Gujarat cotton based on a range of percentage reductions in standard

deviation, we can estimate the required sample size to maintain the current accuracy level of approximately 30 percent. Figure 3.1 shows that if variation is reduced by 10 to 30 percent, to maintain 30 percent accuracy level would require six to nine samples at a village Panchayat level.

This would result in a requirement for 42 to 63 CCE plots for village Panchayats within each of the current IUs. Based on the data provided, the current average number of CCE plots per IU is about 14. This implies an increase in the number of CCEs by three to four and one-half times (mean of 3.75 times current cost). There are currently approximately 700,000 CCEs conducted with an average cost of 300 rupees per CCE. Based on the above analysis, the CCE costs would increase from approximately 210 million rupees to 788 million rupees though this estimate may be high since there are likely some economies of scale associated with conducting a larger number of CCEs in a smaller area. In addition, lowering the insurance to village Panchayat will also generate an increase in liabilities which, according to the Joint Group, is estimated at 35 percent.7

7 Report of Joint Group on Crop Insurance, Department of Agriculture & Cooperation, Ministry of Agriculture, Government of India. December 2004.

Guaranteed YieldNAIS deploys a three-year moving average for rice and wheat and a five-year average for all other crops to calculate the area-average PY upon which the TYs for insurance policies are based. The GOI is exploring the use of a longer time series (e.g. five out of the previous seven) presumably to increase the stability of coverage for the program.

In theory, the length of the time series comprising the area PY does not create any difficulties to incorporate into a premium rating methodology. However, the shorter the time series, the more variability in the coverage offered and the greater the likelihood that “spikes” in coverage may attract participation while troughs may reverse that influence. In these situations, farmers may adversely select against the insurance program by participating when coverage is high or not insuring when coverage is low. When coverage is low their lack of participation would remove their premium contribution (and any corresponding government subsidy) from the insurance pool that otherwise would have helped to pay losses back to the program or build a fund balance for the future. In addition to improving actuarial soundness, a more stable underwriting system may also serve to improve program efficiency by removing spurious payments.

Improving the National Agriculture Insurance Scheme |15|

Figure 3.1: Estimation Error for Area-yield Estimates Versus Sample Size with Reduced Variance for Gujarat Cotton

Source: Data from AICI

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

80.00%

90.00%

0 5 10 15 20 25Error as a % of Mean Yield

Sam

pleS

i ze

Mean Est Error10% Reduction20% Reduction30% Reduction40% Reduction50% Reduction

|16| India – NAIS: Market-based Solutions for Better Risk Sharing

The area-yield estimate within NAIS, like the APH in the U.S. or the indexing method used in Alberta, Canada is intended to reflect what farmers in the area can normally be expected to produce. There are trade-offs between a short time series that is intended to be responsive to yield technology and climate change versus a longer time series that is intended to provide coverage stability. Exceptionally high yields due to “good luck” or exceptionally low yield estimates due to “misfortune” do little to indicate normal yield expectation. Farmers often want to get recognition for “managing the high yields” but consider low yields beyond their control. Governments, reacting to farmer concerns, are often susceptible to these assertions.

Two notions are helpful when considering a solution. First, the purpose of insurance is to provide coverage when an event reduces yield below what is normally expected. Second, the inclusion of one or two recent years of production records, good or bad, is not likely to be a good indicator of expected normal production in the future.

Based on the observations above, a review of the impact of setting PYs using four different methods (three-year moving average, five-year moving average, ten-year moving average and ten percent exponential smoothing) was conducted on the

sample data available for this study (Technical Annex 3 and Box 3.1).

Based on this quantitative analysis, establishing a ten-year PY in the short term is suggested. In the medium term, however, a more effective alternative would be to use a longer time series, where data is available, with the incorporation of a yield trending mechanism to more accurately reflect the normal production for an insurance area. This would serve to reduce yearly coverage fluctuations, reduce the potential for adverse selection and avoid declines in client satisfaction and/or participation relating to inadequate coverage.

Longer time series of yield data sets provide greater stability in coverage for an insurance scheme. However, as the time series increases, there is greater likelihood that crop management practices, which improve over time, will increase the production of a crop and possibly reduce the yield variability due to natural perils. For this reason, estimates of normal expected production, using longer term data series, should be adjusted to reflect current technology; otherwise, records from the past will underestimate the current production reality. The U.S. and provinces in Canada have adopted yield trending approaches to account for technology and this section includes a brief description of some selected methods.

Box 3.1: Time Series and PY Estimation A statistical analysis on the impact of the length of the time series on the PY was performed for a sample of states and crops (Technical Annex 3 has more details). This analysis shows that by moving from a three-year moving average to a ten-year moving average the following could be achieved:

When using a three-year moving average as a basis for PYs, they tend to change on average by approximately 19 percent each year. This annual change is approximately 12 percent with a five-year moving average. Coverage changes of this size may discourage farmers from participating in NAIS and reduce the effectiveness of NAIS as a risk management tool. Using a ten-year moving average reduces the average year over year change in PYs to approximately five percent – a much more reasonable number;

Using a three-year moving average can result in PYs that are anywhere from 52 percent below the long-term average yield for the area to 68 percent above the long-term average yield for the area, while using a five-year moving average results in PYs which are between 40 percent below and 50 percent above the long-term average yield. Under-coverage of 52 percent or even 40 percent provides coverage that is worthless. Using a ten-year moving average results in PYs which tend to be between 22 percent below and 19 percent above the long-term average yield for the area, which represents a significant improvement; and

The cost of providing benefits using a three-year (resp. five-year) moving average PYs is approximately eight (resp. five) percent higher than the cost of providing benefits using a ten-year year moving average.

Stable coverage is valued by farmers. For example, in Canada as well as in other countries farmers consistently rank stable coverage high in survey responses, more so than inexpensive premium or program enhancements. Stable coverage also increases the likelihood of consistent participation as opposed to selecting to purchase insurance based on the “highs and lows” of insurance coverage. Finally, it helps the insurance manager to better understand the risk their portfolio faces, since program payments are related to insured perils and not influenced by swings in participation reacting to coverage.

Although stability in the insurance coverage is desirable, responsiveness is also necessary. Failing to respond to current conditions can create under-coverage situations due to the technological advancements which occur over time. Consequently, there are trade-offs between a longer time series that increases coverage stability and a shorter time series that is more responsive. Adjusting for yield trends allows a longer time series to be used while ensuring that coverage responds to advancements in technology.

Designing a yield trending methodology that is appropriate in the Indian context will first necessitate an extensive review of the India data set. Lessons may be drawn from the method used in Canada where a seven-step method is used to adjust risk-area “normal yields” for technology trend (Technical Annex 3).

Indemnity Limits

At present, NAIS offers indemnity limits (i.e., coverage) at production triggers of 60, 80, and 90 percent of PY. The level of coverage is restricted by crop and IU within the insurance program to provide a level of equity in government support among regions.8 During Kharif 2004, 58 percent of all crops/areas were in the 60 percent indemnity zone,

8 The coefficient of variation (CV) of the area-yield is used to determine which indemnity limit will be used in the risk-area:

90 percent indemnity zone if CV is less than 15 percent 80 percent indemnity zone if CV is 15 to 30 percent 60 percent indemnity zone if CV is greater than 30 percent.

32 percent in the 80 percent indemnity zone and 10 percent in the 90 percent indemnity zone. Significant increases in premium costs brought on by a move to actuarial rates may necessitate the ability to purchase lower levels of coverage in order for many farmers (and governments) to be able to afford the premium costs and still attain a level of protection. Eliminating the 60 percent coverage may therefore be premature. In fact, lower coverage may be important to ensure continued participation and premium subsidy from governments may be better directed at lower coverage levels, with the provision for farmers to purchase higher coverage with low or no premium subsidy.

Based on international experience, one suggested possibility is to offer farmers more than one indemnity limit and let them choose. Under the U.S area-yield program Group Risk Plan (GRP), for example, farmers can choose a coverage level ranging from the Catastrophic Protection level (50 percent) to 90 percent, with 5 percent increment.

Risk-areas used to define the indemnity limit should be reconsidered based on the actuarially-sound premium rates in order to generate a homogeneous set of premium rates. Actuarially-sound premium rates associated with a 90 percent indemnity limit should be compared among crops and IUs and then adjusted through changes in the indemnity limit. This method would ensure more equity among farmers, as subsidized premium rates of food crops and oilseeds are capped.9

Timely Payment of Claims

NAIS has had difficulty adhering to timely claims payments in the past and this perhaps, is one of the most critical issues that it faces. Several factors are responsible for claim delays, including: delays in receipt of crop yield estimates from the states, verification of yield data by AICI, access to government funds to cover losses in excess of premium

9 The premium rates are 3.5 percent for oilseeds and bajra and 2.5 percent for cereals, millets and pulses during Kharif; 1.5 percent for wheat and 2 percent for other food crops and oilseeds during Rabi.

Improving the National Agriculture Insurance Scheme |17|

|18| India – NAIS: Market-based Solutions for Better Risk Sharing

and payment of claims to farmers through the banking network. Each of these components of the claim payment channel should be reviewed for expediency.

Many farmers incurring a loss of income may not have the financial strength to withstand a delay in payment and may as a result become defaulters on any loans they may have taken from banks. In a competitive environment, accurate and responsive claims payment can increase market share even if product design is no better than the competition. This reinforces the notion that sound program design on its own does not necessarily translate into a valued insurance product.

Timely indemnity payments are vital for farmers since they provide cash at the time when it is needed most. Given the lack of adequate access to formal finance, this could prevent farmers from falling into a debt trap or having to pay high interest rates on moneylender loans. The following measures could therefore assume considerable importance.

The Government could facilitate prompt payment of final claims by contributing to up-front premia subsidies based on the suggested premium rates. The Government could pay the difference between the suggested actuarially-based rates and the capped (subsidized) premium rates, to be released to AICI at the beginning of the crop season. These up-front payments would allow AICI to facilitate prompt payment of final claims, thus making the NAIS more attractive to Indian farmers. By doing that, delays in claim settlement are expected to be reduced by four to five months, thereby reducing the settlement time by as much as about 50 percent.

Delays in receipt of yield estimates from the states, is one of the key underlying reasons for delays in claim settlement. It currently takes state government around two months from the raw CCE data collection to submit the crop yield estimates to AICI. Government could institute measures to minimize these delays and the role of technology and data consolidation techniques may be critical in this regard.

Advanced indemnity payments could be made based on weather and/or remote sensing indices. This approach will enable reaping the benefits drawn from combining the best features of both area-yield (e.g., more accurate loss estimates) and weather-based insurance (e.g., faster claim settlement). Since AICI, with the assistance of the World Bank, is currently developing actuarially-sound weather-based insurance products, the implementation of such a measure could be possible for the Kharif 2007 season.

Operational Deadlines

Insurance purchase deadline

Under the current program in India, farmers can purchase crop insurance well into the growing season. Banks can provide loans to farmers in a similar time-frame and with the mandatory crop insurance intention, may automatically sign a farmer up for crop insurance when production results are already known. If weather conditions appear to be impacting crop development, farmers could select adversely against crop insurance by participating in crop insurance or increasing the insured coverage. In order to protect loan positions, banks could actively promote this behavior within the current system.

Evidence of adverse selection appears in the data as well. For example, insurance is mandatory for farmers with loans and yet, in the past, Kalavakonda et al10 recognized that only 20 percent of farmers with loans in Karnataka actually insured their crops. In these instances, program rules have not been followed and it is equally plausible that other deviations from protocol have occurred. For example, actual seeded crops could be recorded as a crop with a higher coverage level to take advantage of a spike in coverage. Given that loans, and therefore crop insurance, can be purchased well after the growing season is underway, inaccurate reporting of seeded crops to take advantage of higher coverage is a definite possibility. If more stringent eligibility criteria are implemented, in the

10 Kalavakonda, V. et al., “Karnataka Crop Insurance Study”, South Asia Region, World Bank Document, September 2003.

medium term it may also be advisable to consider an increased level of audit of participating banks either to provide incentives for good performance or penalties for failure to consistently and accurately implement NAIS policies.

To reduce the impact of adverse selection, farmers should make their insurance decisions well in advance of the growing season. Farmers should know their cropping intentions prior to the growing season and should also know if they require an operating loan to seed. Insurance deadlines should be well known in the farming community and, if banks require insurance to provide an operating loan, they could make that decision known to the farmer in advance of the growing season.

A purchase deadline for crop insurance could be instituted in advance of the crop season. To encourage an earlier insurance sign-up, AICI could introduce a premium discount within a certain period well in advance of the growing season. This discount could extend to state and central governments in order to encourage their early premium payment to match the farmers’ premia. On the other hand, because premium discounts must be offset with a load to an actuarially-sound premium rate, governments and farmers could instead be surcharged a premium penalty for late sign-up and late payment. At some point, sign-up to the crop insurance program would have to be curtailed even with a premium surcharge. These initiatives could be used to ensure “up-front” premium contributions by all parties.

Seeding deadline

In an individual coverage program, farmers are required to seed crops within a recognized “time window” to ensure that the likelihood of losses at harvest are not increased. The normal growing season is known by farmers and local agriculture experts. Seeding crops beyond this window places an adverse selection risk on the insurance program. These same concerns exist in an area-yield design like India’s and CCE procedures should ensure that only crops

that are sown within an acceptable time window are included in the yield estimation process. The Government may want to investigate the feasibility of seeding deadlines by risk-area and crop.

Additional Program Feature

Replanting/re-seeding benefit

The existing scheme covers risk only from sowing to harvesting. When sowing/planting is prevented due to adverse seasonal conditions, farmers usually lose their initial capital but also the opportunity value of the crop they will not be able to sell.

In Canada, as an example, since the basic insurance scheme to which these benefits attach are individual coverage designs, the farmer is the one who initiates the insurance claim. In these countries farms are larger and a loss adjuster visits the farm to determine if the acreage under claim is damaged and will be covered under the program. Once that is determined, the farmer can re-seed the crop and will receive an indemnity once the re-seeding operation has been completed. If a large district is known to have incurred a significant loss, individual farm loss adjustments may be overlooked and replaced with random audits.

Under an area-yield insurance scheme some means of defining land to qualify for a replanting/re-seeding benefit would need to be established. This would likely be done at the local level with a pre-defined indemnity provided to compensate for the re-seeding operation. The compensation scheme could include technical assistance through extension services, seeds (in cash or in kind), etc. However, since NAIS operates at the national level, a consistent means of determining acreage eligibility for compensation would have to be established as well as a means to track losses effectively over time to monitor the use of the provision. In addition to loss assessment, decisions would need to be made concerning:

The extent of the timing within the crop year that a re-seeding would be allowed. If a crop is damaged and can be readily re-seeded within a set time that would ensure a successful

Improving the National Agriculture Insurance Scheme |19|

|20| India – NAIS: Market-based Solutions for Better Risk Sharing

maturation of the crop, a re-seeding claim could be paid. However, if the damage occurs later in the year, perhaps a percent of coverage (staged indemnity) could be paid under the insurance scheme in the local area. This would only be done if a portion of the IU was damaged and that amount would likely not impact the outcome of the area-yield estimate;The amount to be paid for different crops to cover the re-seeding function. Different crops have different seed costs and the amounts of compensation would have to be determined in advance of the crop year;Whether the re-seeding benefit (or any other special feature) would be automatically included within the basic NAIS or if it would be offered as an option that farmers could choose to purchase for extra (subsidized) premium;Whether weather-index designs could be used as a proxy for a special feature event (crops that cannot be seeded, crops damaged after harvest, etc);Building a means to record the cause of loss, indemnity payment, acreage damaged and location into the national database format to be able to monitor results and allow analysis to support further enhancements.

There may be an administrative problem with this benefit. This benefit is associated with an early season risk and is therefore subject to adverse selection. As a result it should either only apply to farmers who have signed up before the loss event (for example, based on historical onset of the monsoon season for Kharif crops). Coverage could be used as an incentive for early sign-up. Alternately, coverage could be contingent upon participation in the previous year.

On balance it may be better to postpone this benefit until a proper data management system can be established. Adding additional enhancements to a scheme that is missing some of the basic support instruments may complicate the process to an extent that an actuarially-sustainable scheme and other key objectives are jeopardized over the long term in order to accomplish more minor short term objectives.

Should the GOI want to offer this benefit to farmers before the development of a fully integrated data management system, it is suggested to add this benefit as a compulsory add-on cover, to assess losses on an area base or on a parametric weather base. Given that this benefit cannot be actuarially rated (because of lack of data), it is also suggested to consider it as a social benefit (as opposed to insurance) and thus to transfer the claims to the Government’s account. As part of larger agricultural policy, it is also suggested to further develop extension services to farmers to help them to better mitigate these potential losses.

Post-harvest benefit

In some states crops like rice are left in the field for drying after the harvest. Quite often, especially in the coastal areas, these crops are damaged by cyclones or floods. The current NAIS program covers risk only up to harvesting.

It is suggested that post-harvest indemnity payments could be calculated on post-harvest yield adjustment factors applied to reduce the area-yield. Assessment of these adjustment factors would be accomplished through re-sampling CCE farms or separate sampling process. The area for re-sampling would be defined administratively at a sub-IU level but not at the individual level.

As in the case of re-seeding/replanting benefit, the GOI should postpone this benefit until a proper data management system can be established and actuarially-sound rates can be calculated. Should the Government want to offer this benefit to the farmers for Rabi 2007, it could be offered on a social basis (as opposed to insurance) with the claims transferred to the Government’s account.

Risk-area Boundaries

IUs in India currently follow “political boundaries” as is true for area-yield insurance programs in the U.S. This is not a deterrent to designing actuarial rating methodologies. IUs based on political delineations

will result in premium rates that reflect coverage, yield variability and indemnity losses for the area, provided yield assessment methods are non-biased and are applied consistently across all districts/states over time. However, if IU boundaries were drawn to encompass homogeneous risk-areas with natural parameters, IU production estimates would correlate better to actual production within the area.

Agro-climatic conditions are reflected over time in soil types and these factors have a bearing on yield capability. Developing IUs along these delineations will improve yield estimation accuracy to reflect farmers’ actual production within the IU. Reconfiguring boundaries to encompass homogeneous risk-areas based on natural parameters may provide a better correlation between area production estimates and actual production for the area and will result in premium rates which more accurately reflect differences in risk between areas.

In 1989, Alberta (Canada) sub-divided the province from the original 14 risk-areas into 22 risk-areas. These risk-areas crossed political boundaries and were primarily based on known production capacity from historically reported crop insurance records at the “farm crop level” and soil class information within the province. In addition, each county and municipality in the province sub-divides their respective jurisdictions according to in-field soil assessments for tax purposes. Combining yield information from the crop insurance database with municipal soil ratings allowed the further refinement of risk-areas for the crop insurance program. At present, with another 17 years of yield information available at the farm crop level since the last risk-area refinement, Alberta utilizes a further demarcation within the risk-area process. For major crops, when farmers first enter the crop insurance program, their coverage is assessed at an average of reported production results in their immediate township. As time progresses, they develop their own yield history. The Alberta experience demonstrates how program design can be altered and refined over time as data becomes available and insurance knowledge is advanced.

In the medium-term India could consider adopting a process to develop homogenous risk-area boundaries using their historical area-yield data. Agro-climatic areas, already identified in the country, could be used as a base for the review process. Historical area-yield information at the IU level could be assigned to agro-climatic zones. Analysis of the production data could determine both PYs for coverage and the variability in yields as a basis for premium. Combinations of IUs with statistically similar production parameters (average yield and variability) could be identified through a correlation matrix/geographic analysis and aggregated into homogeneous risk-areas using agro-climatic zones as an initial starting point. As in the Alberta experience, refinements can be made to the risk-areas over time as more yield data is acquired and captured in a “user-friendly” data management system that fosters research and analysis on a broad scale.

An indexing method (Technical Annex 2) could be developed to leverage CCE data to provide more accurate and stable estimates of production capability. Area-yield estimates for individual IUs could be established by determining the relative yield performance of IUs relative to a larger area such as a state, district, or homogeneous risk-area once defined. That comparison would define a relativity factor (based on credibility principles) for the IU and area-yields could then be determined by applying the relativity factor for each IU to the yield estimate for the larger area. These factors could be updated once every three years. An indexing process based on homogeneous risk-areas would significantly improve the accuracy of the process.

The development of homogeneous risk-area boundaries would be a medium-term initiative. The level of accuracy of a project to construct homogenous risk-areas is directly correlated to the availability of good quality data in a consistent format across the country. However, as a national database of the caliber required does not exist in India, its development would have to precede a project to define homogenous risk-area boundaries.

Improving the National Agriculture Insurance Scheme |21|

|22| India – NAIS: Market-based Solutions for Better Risk Sharing

Establishing risk-area boundaries will also contribute to the improvement of estimated PYs. A credibility approach to estimating PYs can help to achieve estimates that are both stable and representative of long-term production potential for an area. Using this approach, PYs would be based on yields estimated at the IU level and an alternate yield estimate that is usually based on a larger more stable area such as the state. In the context of agriculture insurance, credibility weighted estimates tend to work best if the alternate yield estimate is selected based on a larger area that has yields that relate to the IU. Once homogeneous risk-areas are established, alternate estimates based on risk-area level data will perform better than alternate estimates based on state level data. Mean yields for the IUs within risk-areas should be correlated, since these areas are agronomically related. The current IUs within a state are not likely to be agronomically related since state boundaries are set politically.

Actuarial/Insurance Principles

The implementation of actuarial premium rates for NAIS would be a key step forward for India’s crop insurance system. It is obvious from the premium-to-loss ratios exhibited in the past that premium rates for NAIS are considerably lower than the actual portfolio risk. However, moving to an actuarial premium rate regime will be difficult politically and “on the ground”. For many, the term “actuarial soundness” implies a non-biased premium rating methodology that, once initiated in the program, eliminates further concerns with actuarial integrity. This is a false assumption.

The notion of actuarial soundness and insurance principles extends beyond the premium ratemaking process. The structure of the insurance delivery mechanism, methods to assess coverage and losses, program design and its operational characteristics all have a direct bearing on the integrity of the insurance program. The India crop insurance system relies on many “entities” with varying formal insurance backgrounds or intimate knowledge of fundamental insurance principles. In addition, entities within the

system have other roles and perhaps vested interests in the outcomes of the insurance. For example, banks are required to structure their portfolios with a percentage of rural clients. Crop insurance is mandatory for loans but is not the main business line for the bank. In many cases, the mandatory condition for loans, which is apparently not stringently enforced, may simply inject a distraction to the banking system and increase the potential for adverse selection through administrative expediency. Similarly, banking systems are not designed with insurance in mind and may be more onerous than they would have otherwise been had they been designed specifically for insurance purposes.

As mentioned previously, the single-source CCE data set was originally developed for input to public policy. NAIS utilizes this network of production gathering information but some aspects of the process have not been adapted for an insurance concept. In the Canadian system, for example, insurance expertise is located within one managing agency that has the responsibility and authority to manage all aspects of the crop insurance program. While these entities are provincial crown agencies or within a government department, they operate with defined insurance mandates, principles and legal frameworks. They issue public annual reports and are, for the most part, treated as “arm’s-length from government” with respect to the design and delivery of the crop insurance program. Governments and farmers who pay the premium for the program have input to design but once that is established, the responsibility for sales, underwriting, loss assessment, data management, actuarial methods, etc. rests with the managing entity. Figure 3.2 demonstrates how a single entity, in this case Alberta, can be responsible for a crop insurance program while maintaining a distinction between political and management environments. The federal government employs an IU that works closely with provincial counterparts to ensure equity in federal support to programs and adherence to actuarial guidelines.

Premium contributions by all parties are held in trust by the managing agency and there are strict

controls to ensure that premia can only be used to pay claims to farmers or purchase re-insurance. External reviews of all actuarial methodologies by accredited actuaries occur for each province every five years and any interim methodological changes are reviewed the year in which they occur. Failure to comply with actuarial certification guidelines results in financial holdbacks by the federal government. Provincial and federal governments audit the results of each provincial insurance entity directly or through third-party firms with the results made public in government audit reports and in financial statements signed by the provincial auditing firm in the annual reports. Audits of quarterly statements and the annual reports include reviews to ensure that written policies are consistently and accurately applied within the managing agency.

While the system in the U.S. differs from that of Canada, the same principles apply. Management of the insurance program is “arm’s-length from government” with a shared responsibility between private insurance companies and federal oversight through the Risk Management Agency (RMA). The RMA approves program designs for government funding, coordinates research and establishes or approves premium rates for the program. The RMA and insurance companies contain the expertise for insurance within the U.S. system. State governments are not involved in premium funding but can be a source of input to the program.

Within both the Canadian and U.S. systems, the premium must pay for the entire transfer of the risk portfolio within the program over time. Administration may be included or excluded,

Improving the National Agriculture Insurance Scheme |23|

Figure 3.2: Functional Flow Chart – Production Insurance in Alberta

FederalGovernment

National coordination

and oversight

– Legislation and

regulations

Actuarial guidelines

Provides funding

–

–

– Publishes national

statistics

ProvincialGovernment

Legislation and regulations

Provides funding

Publishes annual reports

Policy input

Farmer

Pays premium

Reports acres/yields

Follows program guidelines

Provides farm access for

adjusting

Feedback to program

design

AFSCBoard of Directors

Interaction with

farmers

Marketing and

information

Call Centre

Sales staff training

Loss Adjuster

training

Monitoring

Claim assessment/

verification

Claim payment

Research

Gather farmer input

Collect yields

Set coverage and rates

Sales Claims andLoss Adjusting

ProgramDevelopment

FarmerAppeals

Manage appeal

process

Gather appeal

information

Appeal presentation

Political Political

Management Management

Source: Report prepared by Watts & Associates, Inc.

|24| India – NAIS: Market-based Solutions for Better Risk Sharing

however, those who pay premia or have a direct financial relationship to clients are generally perceived to have a “conflict of interest” when it comes to operating roles within the system. Consequently, systems in Canada and the U.S. utilize a managing agency with authority and responsibility to oversee the entire insurance process. Governments are influential in design and there are processes to receive political direction and input from all stakeholders in a transparent manner. However, governments and farmers are excluded from direct involvement in the fundamental operating roles of the insurance program.

Within the Indian context, premium for NAIS comes from three sources, namely farmers, the state government and the central government. The system operates as if premium is paid to AICI who assumes an “ownership role” as opposed to a management role for the accumulated premium funds. This seems to remove AICI from direct involvement in the delivery of the program due to conflict of interest interpretations. The practical effect is to remove the insurance knowledge within the India system from the actual delivery and “business end” of the insurance program.

AICI is the implementing agency for NAIS and reports through the central Ministry of Finance and the Ministry of Agriculture. The government budget for premium is within the central Ministry of Agriculture and the Ministry of Finance at the respective state levels. However, there does not appear to be a clear demarcation of responsibility for the “management” of the insurance program. According to the operational flow diagram presented in Figure 3.3, loss assessment and the determination of area-yields on which to base coverage is the responsibility of state officials. The selling role and primary contact with farmers is the mandate of banks, presumably to efficiently provide local access to rural areas and because crop insurance is intended to be mandatory for a farmer’s loan. In this system, it is difficult to determine which entity is ultimately responsible for the management and delivery of NAIS, aspects of the program that have a direct

bearing on the actuarial integrity of the program. This lack of clear authority and responsibility has likely contributed to the historical performance of crop insurance in India. In addition, with several ‘agents’ involved in program delivery, direct communication between insurance representatives and farmers may not be efficiently achieved.

A key element to success is assigning personnel specifically to the crop insurance initiative who would receive in-depth training in insurance principles, program design and operational procedures. For example, if local banks remain a delivery contact with farmers, in the medium-term it may be useful to consider assigning a trained individual with broad insurance knowledge to the bank during a sales season limited to a pre-seeding time-frame. This individual would have in-depth training in all aspects of the insurance program and receive compensation based on their ability to effectively explain the program and to complete administrative processes and the selling function. This is opposed to a primary focus on insurance sales or enhancing a loan portfolio.

AICI is considering the establishment of local insurance field offices and this may be an effective way to increase participation. The local representative could fulfill a local “train the trainer” role and be responsible to coordinate efforts among individuals assigned to the banking network as well as work with farmer associations to increase awareness of crop insurance in the farming community at large. These local representatives could also fully participate in the CCE process and ensure that data is being forwarded to a central insurance data management system to complement future research and bolster the CCE sampling methods. This would require considerable coordination and discussion with state governments. The role of the Ministry of Finance and the Ministry of Agriculture in facilitating this is likely to be critical.

The India structure could be organized to address the key functions within a crop insurance context including insurance sales, operational

Improving the National Agriculture Insurance Scheme |25|

delivery, loss assessment and research and program design. AICI proposes to move forward with several significant initiatives including assessing local calamities on an “individual” plot basis, introducing advanced indemnity payments using weather-based and index-based approaches,

accessing private sector re-insurance and expanding participation with enhancements to the program. These efforts will introduce significant complexities to the insurance program, and require a competent and well-organized resource network to be successful.

Figure 3.3: Flow Chart of Documentation and Information in India

Source : Information from AICI

Note: Claims when paid to Nodal bank, will reach A/c of farmer through bank branch/PACS

Borrowing Non-Borrowing Farmer

Avails crop loan for a notified crop(compulsory coverage)

Also seeks extended coverageand the Bank records the same

Approaches designated Bankfor insurance and submits aProposal form and premium

Designated Bank Branch /PACS

Consolidates details of coverageparticulars (crop-wise, area-wise &

month-wise)

Consolidates details of coverageparticulars (crop-wise & area-wise)

NODAL BANK

Consolidates into BFdeclarations with premium

Consolidates into NBFeclarations with remiud p m

Incorrect & incompletedeclarations for resubmission

CENTRALSTATE

1. Policy matters & overall direction2. Monitoring & review3. Sharing premium subsidy4. Providing funds for claims5. Providing funds to Corpus

1. Notification of crops & areas2. Submission of yield data3. Sharing premium subsidy4. Providing funds for claims5. Providing funds to Corpus

Optional

Acknowledgement/receipt/claims

(if & when payable)

AICI(Implementing Agency)

Premium Ratemaking |27|

Premium RatemakingChapter 4 Chapter 4

One of the main reasons why multi-peril crop insurance has not succeeded world-wide is the sheer complexity of risk and

the lack of adequate risk modeling technology to understand agricultural risks, and, in particular, natural disasters. As crop yields greatly vary from one area to another for the same weather conditions and same crops, many attempts to use standard rating methodologies about losses in crop yields, such as the Normal Theory Method (NTM), have led to not fully accurate (and often under-estimated) premium rates.

Computing actuarially-sound premium rates is critical not only for the insurance company to properly price its products, but also for governments and farmers, as actuarially-sound premium rates identify the true cost of risk, i.e., price, the underlying risk exposure of any agricultural business activity. It can thus help farmers to test the economic viability of their business (see also Chapter 1).

This market test will have major economic implications. For example, Government, as part of its agriculture policy will be able to provide more informed and better signals to farmers on the economic viability of cropping practices. Such information would also help Government to ascertain their policy with regard to different crops.

The premium methodology section is made up of three inter-related components, namely: data review in support of other documented issues; data analysis and preparation to support a robust premium ratemaking review; and the proposed

premium rating methodology (Technical Annex 4 provides a detailed presentation of the underlying work involved in proposing the suggested methodology; Technical Annex 5 which is a peer review by a Fellow of the Casualty Actuarial Society {FCAS} actuary).

Data Review, Analysis and Preparation

Quality data is a pre-requisite to perform an actuarially-sound ratemaking methodology. In particular it is necessary to conduct a review of TYs/PYs; to develop yield trending methodologies to complement the longer PY time series; to develop smaller sized IUs and/or homogeneous risk-area boundaries to reduce the inherent basis risk in the area-yield estimates relative to actual farm production; to verify the accuracy of the CCE process; to assess alternative production measurement strategies like weather- or satellite-based information; and to access private sector re-insurance to backstop the crop insurance program.

Considerable effort was spent to clean, collate and reformat the data set to make it useful for this TA. Many of the data anomalies need to be addressed internally on an ongoing basis and include different spellings of the IU identifiers on different data sets; lack of a complete set of CCE yields; the undocumented exclusion of CCE data during the area-yield calculations (presumably due to outlier identification); data “mismatches” when an IU was either split or combined; an inability to assign some IUs to the CCE data; outlier CCE and IU yields; and inconsistent IU yield data through time.

|28| India – NAIS: Market-based Solutions for Better Risk Sharing

The current NAIS data system, which is facing pressures to expand with smaller IUs, additional benefits and a desire for increased participation, is clearly unwieldy from both an efficiency and effectiveness perspective. Given that India is recognized worldwide for its expertise in the information technology field, a review of the current data management system for NAIS to address these challenges may have high pay-offs.

Adverse selection

A number of specific issues were identified in the data review process and details of the statistical analysis in support of these concerns are identified in Technical Annex 4. One of them is potential adverse selection problems.

Adverse selection in NAIS could occur in three areas. First, the flexibility in the sign-up deadline could allow clients to purchase insurance after they become aware that a claim is likely. While the area-based nature of the program mitigates this risk somewhat, in the event that a widespread problem affecting yields develops in an IU, it is likely that farmers will be aware of the problem. If a farmer chooses to sign-up for insurance based on this knowledge, the NAIS program will be subjected to additional cost without the potential for a long-term premium base to help recover from this cost.

Second, adverse selection can become an issue when PYs are subject to wide swings (as in the case of three-year or five-year averages), creating an opportunity for non-borrowing farmers to choose to purchase insurance when PYs are at high levels and, conversely, to refuse to purchase insurance when PYs are at low levels.

In exploring whether adverse selection is present in the NAIS program, the cumulative loss-to-risk was shown to be greater than the simple average loss-to-risk from 1999 to 2004 (10.34 percent versus 8.92 percent). This could be evidence that adverse selection is occurring in the NAIS program. Based on this observation, the same comparison was made

using a longer series of data (including the CCIS data). Using this data series, there is no evidence to suggest that adverse selection is occurring at the aggregate level. However, at the state level, several states stand out as possibly having significant adverse selection including Bihar, Jharkhand, Karnataka, Jammu and Kashmir.11

Potential solutions to the insurance purchasing dilemma include: initiating a sign-up deadline that is in advance of the crop season for which the crop is being insured and requiring that premium be paid at sign-up; offering separate insurance for early- and late-seeded crops would allow two separate sign-up deadlines, each set so as to avoid knowledge of the respective crop outcomes; or initiating a sign-up deadline that is in advance of the crop season for which the crop is being insured but not requiring payment of premium until the crop is seeded (timing agreed to by the farmer, with potential for field audits by local representative). In order for this solution to work effectively, farmers must be held to their commitment to insure and cannot be allowed to back out of the insurance if it appears that a claim is unlikely. Any solution would have to be adaptable to the administrative processes within NAIS.

Finally, adverse selection seems to be appearing in the differing performance of loanee and non-loanee farmers. Non-loanee farmers exhibit three and one-half times greater risk in loss-cost results than their loanee counterparts. The difference in risk between these two groups may relate to the mandatory nature of participation for loanee farmers, which causes them to participate in NAIS regardless of the level of their underlying risk. Conversely, non-loanee farmers’ participation in NAIS is derived primarily from high risk areas since the implicit subsidy makes insurance in these areas relatively attractive. This problem could be corrected with a move to an actuarially-sound rating regime, where each insured

11 It should be noted that data for Jharkhand and Jammu and Kashmir is fairly limited. The possibility of adverse selection could be investigated further when setting rates for these states.

Premium Ratemaking |29|

will be charged rates appropriate to the risk they are transferring to the program.

Outlier yield records

An extensive review of the data was conducted, including a review of IUs with yield records meeting one of the following criteria directly: IUs with an yield which exceeded the yield cap identified at the state and district level; IUs with a mean yield above the 90th percentile mean yield for the district-crop; IUs with a maximum yield above the 90th percentile maximum yield for the district-crop; IUs with a standard deviation in yield above the 90th percentile standard deviation for the district-crop; and IUs with a raw pure rate above the 90th percentile pure rate for the district-crop.

This review led to the conclusion that most abnormal results are caused by one or two exceptionally high yield records, suggesting that an outlier methodology was required. A process was developed, to handle outlier yields consisting of two main elements: a set of yield caps that represent the maximum reasonable yield records by state was developed and yields higher than the caps were excluded from the database. If this were to be used by AICI, it is suggested that these caps be reviewed periodically with agronomic experts. Further, yields outside of the range (Median – IQR * 2.5, Median + IQR * 2.5) could be excluded from the yield database.

Consistency of yield levels and yield variation

For most of the pilot crops, yield levels and yield variation seem fairly consistent through time and the average yield at the IU level has a similar relationship with the state yield. Gujarat cotton was a notable exception. The average yield level in the 1995 to 2004 period is almost six times that of the pre-1988 period and the standard deviation in yields is almost five times greater. In addition, the relationship between the average IU yield and the state yield also seems to be changing dramatically through time. For this reason, it was decided to exclude this early period from the rating analysis.

A similar analysis led to the decision to exclude pre-1995 data for Gujarat groundnut.

For Gujarat cotton there also appears to be a significant difference between the 1988 to 1994 and the 1995 to 2005 periods, both in yield levels and yield variation; however, when examining the yield data at the IU level, many IUs seem to have consistent data from 1988 to 2004, while others seem to have very inconsistent data during this period. It was therefore decided to exclude yield data from 1988 to 1994 only for those IUs with yield levels for this period that were either more than 150 percent or less than 50 percent of the yield levels for the most recent period.

Correlation, analysis and the estimation of pure rates

The methods for determining coverage and premium under an insurance plan are a balance between a desire for stability and responsiveness. For example, the desire for coverage that accurately reflects actual production results tends to encourage yield estimation at the smallest geographical IU possible within practical operational considerations. However, the move to smaller IUs will increase yield variability and the stability of coverage, which can both increase program benefits and premium costs, as well as increase the impact of localized good and bad weather events.

The same consideration also affects the rating methodology. On the one hand, there is the tendency to expand the geographical search for data outwards to create a larger insurance “pool” to help spread losses and keep premia from varying widely. On the other hand, there are farmers in areas that have historically good experience arguing that their good experience be reflected through lower premium rates. In the case of rating, the balance between stability and responsiveness is determined explicitly by calculating rates based on a weighted average between program costs at the IU level that are responsive to local conditions and, at the state or district level, that are stable and can be estimated more accurately.

|30| India – NAIS: Market-based Solutions for Better Risk Sharing

In order to determine whether or not to use the district level experience or state level experience to achieve the required stability, the correlation is reviewed between yields at the IU level and yields for their respective district and state level to determine the relative fit between these yield groupings. These correlations indicate that yields at the IU level are generally best related to yields at the district level and that where practical, program costs estimated at the district level could be used in the rating process.

Low data and district grouping

The number of unique CCEs conducted in some districts is quite limited in early years, which creates accuracy concerns regarding the use of program cost estimates in the rating process. The analysis supports the notion that IUs (districts) with at least 20 unique CCE yields provide a sufficiently robust data set to support a premium methodology which will ensure that premium rates for this geographical area will reflect regional differences but still provide appropriate accuracy and stability (Technical Annex 4). Since many districts do not meet the minimum sample size (20 unique CCE yields), the following method was developed to combine districts into areas of similar risk:

Unique CCE counts for each district were calculated by year and a representative year was selected for each state. The districts requiring combination were identified based on the unique CCE counts associated with the year selected;Correlations between district yields within each state were calculated (note that correlations in district level pure rates were also calculated but they tend to follow a similar pattern); Districts requiring combination were then combined manually with the neighboring district or districts (based on the Geographic Information Systems (GIS) maps provided) having the highest correlation until the target number of unique CCEs was reached for the district group;When correlations were similar, reference was made to the meteorological and agro-climatic zone maps provided; and

In some cases it was clear that there was insufficient data to perform a rating analysis at the district or district group level. In these cases, state level pure rates were used.

This approach may be useful when developing homogenous risk zones, although it is suggested grouping neighboring unique CCE IUs based on their respective correlations as opposed to neighboring districts. The current district level groupings are based in some part on previous decisions to group the IUs as represented by assigning one of the unique CCE yields to each IU.

Proposed Ratemaking Methodology

Agriculture risk layering

An experience-based approach is proposed for the suggested premium rating methodology for the NAIS program. This choice is consistent with international best practices, where many agricultural insurance programs are rated using an experience-based approach, including: the Canadian crop insurance, barley proxy and satellite pasture programs, as well as U.S. crop insurance Multi-Peril Crop Insurance (MPCI) and GRP. India’s agricultural sector is characterized by a large variation in risks due to the wide variety of crops and regional diversity in both soils and weather. India has a long history of providing crop insurance and has compiled a significant pool of yield and experience data. The direct use of this data is therefore, possibly the most effective means to accurately estimate program costs and reflect the diversity of risk.

The proposed rating methodology is designed to achieve actuarially-sound premium rates that are stable yet reflective of regional differences and responsive to changes in risk over time. To achieve this goal, risk is divided into a base layer and a catastrophic layer, each rated separately. The cost of providing benefits for the base layer is estimated as the credibility-weighted average of pure rates at the IU and district group levels. The cost of providing benefits for the catastrophic layer was set based on

Premium Ratemaking |31|

the weighted average cost of providing catastrophic benefits at the state level. This approach is very similar to the catastrophic risk pooling used in rating the U.S. crop insurance program and has a similar effect to the deficit loading approach applied for the Canadian crop insurance program. It is also consistent with the agricultural risk financing model promoted by the World Bank among its client countries.12

In order to select the catastrophic layer several analyses were performed (Technical Annex 4). The justification for treating the catastrophic layer differently than the base layer in the rating model is that catastrophes have the following properties: (i) catastrophes have a large impact on the rates estimated at the IU level; and (ii) the true probability of a catastrophe for a particular IU is very difficult to determine since catastrophes are relatively rare events. Consequently, it is difficult to justify charging vastly different premium rates to IUs in the same district where the basis for the difference in the rates is primarily the occurrence (or non-occurrence) of disaster events.

As depicted in Figure 4.1, the catastrophic layer, based on coverage for area-yields less than 50 percent of PY, was selected in favor of defining this layer

12 The Financing of Agricultural Production Risks: Revisiting the Role of Agriculture Insurance. World Bank, 2005.

based on capping loss-costs at a state level.13 This approach would facilitate offering differing levels of subsidy for the base and catastrophic layers of risk or even free coverage for the catastrophic level.

The main steps of the proposed ratemaking methodology are described below. A flow chart of the successive steps is depicted in Figure 4.2 on the next page.

Threshold yields and loss-costs

In order to calculate loss-costs for use in the premium rate methodology, a set of historical PYs is required. PYs for this purpose are calculated as a ten-year moving average of IU yields. There is a one-year lag involved in this process to reflect when IU yield information becomes available; for instance, if IU yield data is available from 1991 to 2005, then IU yields from 1991 to 2000 would be used to calculate a PY for use in 2001. This means that, for this example, it would be possible to calculate PYs for the years 2001 to 2006 and loss-costs for years 2001 to 2005 (since the IU yield for 2006 is not available). To make the best use of the available IU

13 The frequency of occurrence of 50 percent payments is relatively infrequent for most state-crops (see Technical Annex 4). The 50 percent level was selected for purposes of simplicity. It should be noted that the frequency of occurrence of payment by coverage level differs significantly by state-crop, indicating that it might be reasonable to select a catastrophic layer which differs by state-crop.

Figure 4.1: Base and Catastrophic Loss Layers

Yield% of PY

100%

90%

50%

Area has experienced a small yield loss

Area has above average yield

Area has experienced a loss in the base layer

Area has experienced a loss in the catastrophic layer

Layer

No Loss

Deductible

Base Layer

Catastrophic Layer

Description

|32| India – NAIS: Market-based Solutions for Better Risk Sharing

IU yields as reported from the states

Data analysisOutlier yield records consistency of yield levels and yield variation correlation Low

data and district grouping

Cleaned yield database

PY: probable yield = 10-year moving average

UP90: unbalanced pure rate @ 90% coverage: BP+CP

Rx: relativity factor @ X% coverage (using NTM)UPx: unbalanced pure rate @ X% coverage = UP90 * Rx

WLC: weighted average RL @ maximum coverage over the past 15 yearsWUPR: weighted average UP (state-crop)

BBF: balance-back factor (state-crop) = WLC/WUPR

FPRx: Final pure rate @ X% coverage = UPx * BBF

Rx: loaded rate @ X% coverage = (FPRx + F)/(1.-V-Q)With

Q: contingent load = 0.25V: variable administrative load = 0.025

F: fixed administrative load = 0.04

CL: catastrophe loss-cost@ 50% coverage= max (0.5*PY-Y,0)(0.9*PY)

IUCP: All year simple average of CL (with 0.1% minimum)CP: weighted average of IUCP within the state

IUBP: most recent 15 year simple average of BL (with minimum)DBP: weighted average of IUPB within the district group: Z: credibility factor = N (N+K) where N is #years used inBP: base pure rate = IUPB*Z+DBP*(1-Z)

RL: raw loss-cost@90% coverage = max (0.9*PY-Y,0) (0.9*PY)BL: base loss-cost @90% coverage: BL=RL-CL

Note: Weights are the average coverage amounts from 2002 to 2004. Parameters to be calculated for each IU and each crop.

Figure 4.2: Flow Chart of the Experience-Based Ratemaking Methodology

Premium Ratemaking |33|

yield data, the PY for the first ten years is filled in (i.e., 1991 to 2000) using the PY to calculate the first ten years of records (i.e., the PY for 2001). This method creates a PY to correspond to each annual IU yield so that the IU yield can be compared to a PY to generate an annual loss-cost that forms the basis for the premium rates. Although this practice will tend to slightly under-estimate the cost of providing benefit during the first ten years, it is preferable to the alternative of setting premium rates based on loss-costs for only six or seven years in many cases. PYs will be calculated in this fashion for all 13,672 IUs in the sample.

90 percent loss-cost calculations

Raw loss-costs (percent of insurance indemnities relative to coverage) are calculated at the 90 percent coverage level to reduce the likelihood of having a zero loss-cost generated since, within an insurance context, zero risk is not a practical outcome. Loss-costs represent the value of production that falls below 90 percent of the PY and would generate an insurance indemnity. Some of that production shortfall, if significant, will fall below the catastrophic level or 50 percent of the PY.

To segregate the loss-cost into catastrophic and non-catastrophic, or base loss-cost, the entire loss-cost is calculated for any yield falling below the 90 percent of PY known as the raw loss-cost and then calculate the catastrophic loss-cost portion. The difference in these two loss-cost values is the base loss-cost. In mathematical terms, this gives:

Raw loss-cost:

RL = Max (0, 0.9*PY –Y)/(0.9*PY)

Catastrophic loss-cost:

CL = Max (0, 0.5*PY –Y)/(0.9*PY)

Base loss-costs (BL) is simply the difference between raw loss-cost and the catastrophic loss-cost:

BL = RL – CL

Where:

RL = the raw loss-cost (or raw percent payment);

Y = the area-yield estimate for the insured crop and IU;

PY = the probable yield for the insured crop and IU;

CL = the catastrophic loss-cost (or catastrophic percent payment); and

BL = the base loss-cost (or base percent payment).

Note: that all loss-co sts are calculated for the 90 percent coverage level.

Calculating premium rates for the IU

The basic building block of the premium rates is the pure rate. The pure rate is an estimate of the long-term average cost of providing insurance benefits. Separate pure rates will be calculated by layer: the Base Pure Rate (BP) which estimates the long-term average cost of providing benefits for the base layer; and the Catastrophic Pure Rate (CP) which estimates the long-term average cost of providing benefits for the Catastrophic Layer. These two pure rates will then be combined in order to estimate the Final Pure Rate (FP).

Step 1. Calculate the 90 Percent Base Layer Pure Rate

The BP for the 90 percent coverage level is the weighted average of two 15-year simple averages. The first (IUBP) is the average of annual loss-costs calculated at the IU level which serves to estimate the program costs for the IU. The second (DBP) is the average of annual loss-costs calculated at the district group level for the district group in which the IU is located. This average estimates program costs for the district group.14 The weight applied between these

14 One possible simplification would be to compute DBP at the state level.

|34| India – NAIS: Market-based Solutions for Better Risk Sharing

two averages is calculated using a process called “Least Squares Credibility”. This process is designed to achieve the best statistical balance between responsiveness to local conditions and stability and accuracy.

Some limits are placed on the Least Squares Credibility weights to ensure that at least 20 percent and no more than 50 percent of the BP is derived from information reflecting program cost at the IU level (IUBP). This means that anywhere from 50 to 80 percent of the BP will be derived from information reflecting program cost at the district group level (DBP).15 It is suggested that these caps be reviewed to ensure that the desired level of stability is achieved.

Step 2. Calculate the 90 Percent Catastrophic Layer Pure Rate

The CP for the 90 percent coverage level is the simple average of the annual catastrophic loss-costs for all years calculated at the state level.

Step 3. Calculate the 90 percent Unbalanced Pure Rate (UP)

The UP for the 90 percent coverage level is calculated by adding the BP and CP together.

Step 4. Calculating the Unbalanced Pure Rate for Other Coverage Levels

Coverage is offered under NAIS at 60, 80 and 90 percent coverage and may in the future be offered at other coverage levels. So a UP must be established for each of the coverage levels. This is accomplished using coverage level relativity factors determined using the NTM. This model is currently used to set the actuarial rates in India and is a widely-used exposure-based rating model for crops. The coverage level relativity factors for a particular coverage level is simply an estimate of the relative costs of providing benefits at that coverage level versus those same costs at 90 percent.

15 It is highly recommended to compute this credibility factor Z. However, should AICI be unable to compute it, it is suggested to take Z=0.3.

Step 5. The Final Pure Rate (FP) or Balanced Pure Rate

Program costs can be most accurately estimated at the state-crop level. It is therefore standard practice to require that the IU rates “balance back” to the total program costs estimated at the state level. To accomplish this goal, a state-crop “balance back factor” (BBF) is applied to the UPs for all IUs and coverage levels. The BBF is calculated to ensure that the FPs will be exactly sufficient to cover the state estimate of the cost of providing benefits at the maximum coverage level. It ensures that if clients had purchased insurance similar to what they purchased recently (from 2002 to 2004) the FPs would be exactly adequate to recover the cost of providing benefits at the maximum coverage level. The maximum coverage level has been selected for the BBF because it is the most common coverage level selected.

Premium Loading Factors

The FP rate is loaded for administration costs, profits and contingency. The loaded rate is calculated using the standard pure rate process according to the following formula:

Rx = (FPx + F)/(1-V-Q)

Where:

Rx = the loaded rate at coverage level x for a particular crop and IU;

FPx = the final pure rate at coverage level x for a particular crop and IU;

F = 0.4 percent point factor related to fixed administration costs;

V = 2.5 percentage point factor related to variable administration costs; and

Q = 25 percentage point related to risk/contingency. This load is estimated using a fitted loss-distribution and Monte Carlo simulation.

Premium Ratemaking |35|

Administrative loads are designed to recover the administrative costs associated with operating an insurance program. For the purposes of the rating process, the fixed administrative costs are those costs which do not vary directly with premium, and the variable administrative costs are expressed as a percentage of premium. ‘F’ accounts for the fixed administrative costs or those administrative costs which do not vary with premium, such as salaries, utility costs, etc. Fixed costs have been estimated at 0.4 percentage points. ‘V’ accounts for administrative costs which vary directly with premium such as taxes, some types of service charges and sales commissions. Variable administrative costs have been estimated at 2.5 percentage points based on the sales commission/service charges provided to the cooperating financial institutions. It has been assumed that no taxes are applied to NAIS premia.

Risk loads (or contingency loads) form an important component of the rating process as without some level of contingency loading the NAIS program would be subject to wide swings in performance and very slow recovery from any deficits incurred. It is also important to note that because the rating model in large part determines its parameters directly from the historical data, the estimation of these parameters is subject to sampling error. Contingency loads are also required in order to attempt to ensure that parameter uncertainty does not result in rates which are insufficient. Since the performance of NAIS is, in large part, dependent on both the frequency and severity of disasters, risk loads have been estimated using Monte Carlo simulation (Technical Annex 4). They have been estimated at 25 percentage points.16

Based on international experience, it is suggested that year over year rate changes be capped at (+ or –) ten percent to prevent large swings in premium

16 A ‘Q’ value of 25 percent was selected such that should a large deficit occur for a five-year period, the NAIS program should recover from that deficit in approximately 10 years (see Annex 4).

rates that can affect producers’ insuring decisions. A stable, consistent level of participation is conducive to actuarial soundness, since this provides stability in the risk of the aggregate portfolio and allows an appropriate risk load to be set. If participation fluctuates significantly, the basic assumption of the rating model, “that past risk is a good predictor of future risk”, may be violated.

Suggested premium rates for crops/IUs/indemnity levels selected in this study are available in a separate document. Almost 12,000 premium rates were calculated, providing a very detailed risk assessment of the sample crops/states.

Table 4.1 compares the current state level rates to the average suggested premium rates for all IUs in the state.

As expected, suggested rates for selected crops and states are well above the subsidized (capped) NAIS premium rates. The difference between suggested (actuarially-sound) premium rates and subsidized NAIS rates is particularly high for groundnut in Gujarat (37.87 percent vs 3.5 percent).

In the case of these sample crops, the current NAIS rating method (NTM) tends to under-estimate the pure costs (i.e., pure premium rates), as compared to the suggested rating method using the experience-based approach, by an average factor of 18 percent.

Suggested commercial rates are approximately 33 percent higher than the suggested pure cost estimates, including a 2.5 percentage point variable operating load and a 25 percentage point contingency load.

Suggested premium rates for sample crops greatly vary within the same state or even the same district. For example, suggested pure premium rates (at 90 percent coverage level) for rice in Andhra Pradesh range from 2.7 percent (in Karimnagar district) to 12.0 percent (in Srikakulam district). This heterogeneity is illustrated through crop risk maps (see Annex 1) where suggested premium rates at

|36| India – NAIS: Market-based Solutions for Better Risk Sharing

90 percent coverage level are depicted. This offers a powerful tool to visualize the true price of risk exposure for selected crops and selected states.

Suggested pure premium rates are calculated at the district level as the weighted average of the pure premium rates of all the IUs located in the district, weighted by the average IU coverage from 2002 to 2004 (see Annex 1).

Limitations/Factors Affecting Actuarial SoundnessThe proposed rating methodology will provide actuarially-sound estimates of the costs of the NAIS program. However, there are several issues which could have a significant impact on the accuracy and actuarial soundness of the rates estimated through the suggested process.

Reliable, good quality data is the cornerstone of any rating process. Without good data that has been adjusted to reflect any factors affecting risk, it is unlikely that the correct premium rates will be determined. National auditing, training and good data management will go a long way to achieving this goal as well as addressing moral hazard concerns.

The historical data represents the costs of providing benefits relative to the program design and processes that existed at that time. In the event that changes

are made to the insurance program or processes, the rating model would need to be adjusted to be reflective of the costs for the insurance program and process that will be offered in the year in which the rates will be applied.

In the event that certain activities can become non-viable economically, these changes can be dramatic. It is important that the historical data be adjusted to reflect today’s management practices and industry structure and that the agricultural sector be constantly monitored in order to allow the detection of these changes and their incorporation into the rating model.

Other Rating Issues

This section discusses the issues related to setting rates for new crops, particularly the role of the NTM, in setting rates for new crops and adjusting premium rates to reflect a change in the variability of yields (caused by a change in the size of IUs, for example). It should be noted that due to the lack of data associated with these, it is impossible to ensure that the premium rates resulting from the processes discussed below will be actuarially-sound. For this reason, the performance of the resulting premium rates would need to be reviewed frequently until such time that sufficient data exists to set rates without the use of the techniques discussed below. However, the rating techniques discussed below are consistent with international best practices and represent a

Table 4.1: Current and Suggested Premium Rates for Sample Crops

Rate

Andhra Pradesh

RiceKharif(%)

Gujarat Maharashtra Uttar Pradesh Wheat Rabi(%)

Cotton Kharif(%)

Groundnut Kharif(%)

Pigeon pea Kharif(%)

Sorghum (Dryland)

Rabi(%)

2005 Subsidized NAIS Rate 2.50 15.45 3.50 2.50 2.50 1.502005 NAIS Pure Rate Estimate 3.15 10.85 23.38 9.33 8.57 2.46Suggested Pure Rate Estimate 4.44 12.82 27.05 12.02 11.91 2.51Suggested Rate (admin, no subsidy) 6.58 18.23 37.87 17.13 16.78 4.02

Note: 2005 NAIS pure rate estimates are derived from the NTM. Suggested pure rate estimated are derived from the Experience-based Approach.

Premium Ratemaking |37|

reasonable approach to estimating program costs in the absence of adequate data.

Rating new crops

The main challenge of rating new crops is generally characterized by lack of data, either for the new crop or for the new area. There are several ways to estimate and fill in missing loss-cost data in order to set premium rates for new crops. Once the missing loss-costs are filled in, the suggested rating methodology is applied exactly as if these loss-costs had been calculated from IU yields in the normal fashion.

Should 15 years of reliable yield data exist for the areas which are to be used as IUs, the suggested rating methodology can be applied without alteration. However, the most probable case is when less than 15 years of reliable data is available. In this case, the available yield data should be used to estimate loss-costs to the extent possible. An estimate of the long-term cost of providing benefits must be determined – this estimate is often referred to as the assumed pure rate (APR). The APR will be used to fill in the missing loss-costs to create a 15-year series of loss-costs to be used in the suggested rating process. In the context of the suggested rating methodology, an APR must be estimated both for the base and the catastrophic layers. Both of these APRs can be determined using the methods described for the following situations.

If AICI wants to offer insurance for a new crop in an area where insurance for an agronomically similar proxy crop with good data is already offered, then the pure rates for the proxy crop could be used directly to estimate the APRs for the new crop. However, this approach is appropriate only if the new crop and the proxy crop are very closely related.

If AICI wants to expand insurance for a crop which is currently insurable to a new area where insurance for an agronomically related proxy crop with good data is already offered, then a risk

adjusted pure rate based on the proxy crop could be used to estimate the APR. The Risk Adjustment Factors (RAF) used in this approach are calculated using ratios of the pure rates for the new crop and the proxy crop in another, preferably agronomically similar, area which has good data. This factor estimates the relative risk of target and proxy crop for the proxy area. It is then assumed that the relative risk between the target and proxy crops will be the same in the new area so that:

APR = RAF * Pure Rate for the Proxy Crop from New Area

Where: RAF = (Pure Rate New Crop in Proxy Area)/(Pure Rate Proxy Crop in Proxy Area).

Suppose AICI wants to offer insurance for a new crop where data for this crop is only available for a more aggregate level than that for which insurance is to be offered, but data for an agronomically related crop is available at both the new IU and aggregate level. In this case, a risk-adjusted pure rate, based on the new crop for the aggregate area can be used to estimate the APR for the desired area. The RAF used in this approach is based on a ratio of pure rates calculated for the proxy crop for the aggregate and the new IU areas. It is then assumed that the relationship for risk between the larger area and the new IU area will be the same for the new crop and the proxy crop:

APR = RAF * Pure Rate for the Target Crop from Aggregate Area

Where RAF = (Pure Rate Proxy Crop New IU Area)/(Pure Rate Proxy Crop Aggregate Area).

In case AICI wants to offer insurance for a new crop that is not insurable anywhere else, but some crops are insurable in the new area, the CV in yield for the new crop relative to crops already insurable in the new area should be estimated by agronomic experts. Based on this CV estimate, the NTM can then be used to estimate the risk of the new crop relative to the most similar crops insurable in the new area.

|38| India – NAIS: Market-based Solutions for Better Risk Sharing

Often more than one proxy crop is used in this process in order to bracket the risk of the new crop:

APR = RAF * Pure Rate for the Proxy Crop from New Area

Where: RAF = (NTM Pure Rate New Crop)/ (NTM Pure Rate for Proxy Crop).

Suppose AICI wants to offer insurance for a new crop that is insurable elsewhere but no other crops are insurable in the new area. In this case, the CV of yields for the new crop could be estimated based on the opinion of agronomic experts for all crops to be insured in the new area. This is accomplished by comparing the CVs for these crops in neighboring or areas otherwise assumed to have similar risk, or by bracketing the risk. Once CV estimates are established, the NTM could then be used as follows:

APR = RAF * Pure Rate for New Crop Proxy Area.

Suppose AICI want to offer insurance for a new crop that is not insurable elsewhere and the new area has no insurable crops. A CV would need to be estimated for each of the new crops to be insured in the new area. This is generally accomplished by first estimating the CVs for the new crops relative to insurable crops to determine CV estimates for an area neighboring the new area and then estimating the relative CVs for the new area relative to that neighboring area:

APR = RAF * Pure Rate for the Proxy Crop from Proxy Area

Where: RAF = (NTM Pure Rate New Crop and Area) / (NTM Pure Rate for Proxy Crop and Area).

Adjusting premium rated for a change in the variability of yields

The introduction of smaller IUs is expected to increase the variability of IU yields. It will also reduce

the amount of within-IU offsetting that occurs, creating increased cost for NAIS. The exact increase in cost will depend on the amount of correlation between the yields calculated at the current IU level and yields calculated for the new smaller IUs.

In order to adjust rates to reflect this increased cost, yield data for the smaller IUs is required. This type of data can be gathered through a pilot study or from another data source. The impact can also be estimated by extrapolating the relationship between the variance of yields at the new smaller IU level and the more aggregate level from an area where this data is available to another (preferably agronomically similar) area. In the absence of any data it is necessary to make an assumption about the amount of the increase in the variability in yields.

The NTM can provide an estimate of the impact on program costs of a change in the variability of yields. This can be used to estimate the impact of reducing the size of IUs and also to estimate the changes in program cost associated with any change which impacts on the variability. Some examples of changes which are likely to affect the yield variability are: a change in the varieties which are typically grown; a change in the management practices such as harvesting a crop earlier in the season; the introduction or removal of a driage factor. An example is provided in Technical Annex 4.

Further Improvements of the Suggested Premium Ratemaking MethodWhile the suggested premium rating method is actuarially-sound, makes effective use of the data available, and is in line with the best practices for crop insurance ratemaking (Technical Annex 5), there are still a few areas in which the rating model could be potentially improved in the medium- term: introducing homogenous risk areas; exploring the possibility of proxying/pooling data between crops, especially in regard to the estimation of catastrophic risk; and defining different coverage levels for the catastrophic layer by crop, and possibly by state.

Conclusions and Suggestions |39|

Conclusions and Suggestions

India has historically focused on crop insurance as a planned mechanism to mitigate the risks of natural perils in farm production and has been

evolving and improving its crop insurance program over the years. Its recent policy commitment to move forward with significant enhancements to NAIS is a step in the right direction. This chapter summarizes the key outcomes and conclusions on the review of the operations processes and actuarial rating methods implemented under the existing area-yield crop insurance program of NAIS. It builds on the sound, overall area-based approach of the NAIS and aims at providing suggestions, based on international best practices, that AICI could consider to make the scheme more attractive to farmers, so as to increase the crop insurance penetration levels and to place the scheme on actuarial regime.

Findings and Suggestions

The key findings build on the overall sound “area-yield” based approach of the NAIS. The intention is to provide AICI with suggestions for its consideration that could help make its crop insurance program actuarially-sound and more efficient. These suggestions, particularly those that could be implemented in the near-term, are summarized under three broad headings: (a) Premium ratemaking; (b) Improving the NAIS; and (c) CCEs. As discussed above, it is expected that improvements in the NAIS and transitioning to an actuarial regime will yield benefits for farmers, Government and AICI. Figure 5.1 below presents a snapshot of some of the potential benefits that may be derived from a successful implementation of the suggestions that follow.

(a) Premium ratemaking

Determining a sound actuarial rating technique is critical for assessing the true cost of risk. However, this is not always easily accomplished. One of the main reasons why multi-peril crop insurance has not succeeded universally is the sheer complexity of risk and the lack of adequate risk-modeling technology to understand agricultural risks, in particular, the impact of natural disasters. As crop yields greatly vary from one area to another for the same weather conditions and same crops, many attempts to use standard rating methodologies, such as the NTM, have led to not fully accurate (and often under-estimated) premium rates.

It is therefore, suggested that AICI could consider using an experience-based approach for rating purposes. This would constitute a premium rating methodology consistent with international best practices and with Indian conditions. The methodology is designed to achieve actuarially-sound premium rates that are stable yet reflective of regional differences and responsive to changes in risk over time. To achieve this goal under the experience-based approach, risk is divided into a base layer and a catastrophic layer, each rated separately (see Figure 4.2). Premium rates have been calculated for sample crop/states. The sample rates are, on average, 18 percent higher than those derived from the current NAIS rating method.

This premium rating methodology could benefit all key stakeholders: Government, AICI and the farmers. In particular, actuarially-sound premium rates reflect the true cost of risk, that is, the price of the underlying risk exposure of any agricultural business

Chapter 5 Chapter 5

|40| India – NAIS: Market-based Solutions for Better Risk Sharing

activity. This could help the government to (i) reduce its fiscal exposure as it can better forecast public financial support, for example, through up-front premium subsidies which would mean that all the residual risks would be borne by AICI, thereby paving the way for a more ‘market’ based mechanism for crop insurance; and (ii) develop a more cost-

effective agricultural subsidy program as subsidies can be better targeted, for example to catastrophic risks. It could also help the insurance company AICI to build up adequate technical reserves to cover their insurance risks, expand outreach amongst farmers and access re-insurance markets. Finally, it would benefit farmers because it would allow for a more

Figure 5.1 Illustrative Crop Insurance Cycle

Source: Data from AICI

June July Aug Sep Oct Nov Dec Jan Feb March April May June July

Adverse selection possibility

- DESIGN PARAMETERSNeed for process efficiency

- FARMER RELATED EVENTS- OTHER EVENTS

Reduced adverse selection possibility

AprilJuly Aug Sep Oct Nov Dec Jan Feb March May June July

June

* Illustration is for a medium duration crop; ** For borrowing farmerKharif

Deadline forinsurancepurchase**

CCE data toAICI

Data and claimsprocessing byAICI

Final indemnitypayment by AICI

Ex-ante Governmentfinancial contribution

Partial indemnitypaymentfacilitated throughearly trigger

Farmer sows crop Farmerharvests

Crop cuttingexperiments (CCEs)conducted

CCE data to AICI

Farmer sowscrop

Farmer harvests Final indemnitypayment tofarmer

Farmer receivespartial indemnity

Government contributionreceived and indemnitypayment made

Deadline forinsurancepurchase

Final indemnitypayment tofarmer

Actuarial regimeActuarial premiumrates:

Better contract design:

Help Government/AICI ascertain premium rates thatreflect the true cost of riskProvide information for better economic signalling byGovernmenton agri-policyEnable Government to provide up-front contributions topremium subsidies, thereby enabling better fiscalmanagementFacilitate farmers to get quicker final settlement ofindemnities, due to up-front government contributionEnable AICI to build adequate technical reserves, improvepossibilities for international reinsurance, assume insurancerisks and operate on amarket basis

Enables early partial settlement of indemnitiesMore transparent terms and conditions of the insurancepolicyTechnically sounder insurance product with lowerpossibilities of adverse selection andmoral hazard

:

Crop cuttingexperiments (CCEs)conducted

Data, claimsprocessing

WHAT IS HAPPENING NOW

Need for process efficiency

WHAT COULD HAPPEN IN AN ACTUARIAL REGIME

Actuarial rates

New contractdesign features

Conclusions and Suggestions |41|

timely payment system and, ultimately, a more equitable crop insurance subsidy scheme.

This approach will also enable ex-ante fiscal management for Government. Up-front premium subsidies derived from the actuarially determined premium rates will allow the government to budget its financial support to AICI at the end of the insurance policy sales season. However, it is suggested that the Government act as a re-insurer of last resort for catastrophic risks (e.g., actuarially-based loss ratio higher than 200 percent) until adequate catastrophe reserves are built up.

(b) Improving the NAIS

Risk differentiation

AICI could use the premium rating methodology to differentiate the underwriting risks. The suggested ratemaking methodology would allow AICI to develop actuarially-sound premium rates that provide a more accurate measure of the yield risk exposure. These rates could be used to differentiate the underwriting risks.

To ease administrative management while preserving risk differentiation, AICI could set the insurance premium rates by crop at the district level and adjust the indemnity limits by insurance unit. The implementation of insurance premium rates at the IU level may turn out to be impractical and administratively difficult, given the very large number of IUs. Hence, AICI could select a single premium rate at the district level and adjust indemnity limits at the IU level. These differentiated indemnity limits will help maintain risk differentiation and thus equity among farmers.

Improving the timeliness of indemnity payments

Timely indemnity payments are vital for farmers since they provide cash at the time when it is needed most. In the context of the lack of adequate access to formal finance, this could prevent them from falling into a debt trap or having to pay high interest rates on moneylender loans. And timely payments would

improve the chances of access to credit for future crop seasons. The following measures could therefore assume considerable importance.

To facilitate prompt payment of final claims, Government could contribute to up-front premium subsidies based on the suggested premium rates. The Government could pay the difference between the suggested actuarially-based rates and the capped (subsidized) premium rates; this contribution could be released to AICI at the beginning of the crop season. Such up-front payments could allow AICI to facilitate prompt payment of final claims, thus making the NAIS more attractive to farmers. It is estimated that such a measure could reduce delays in claim settlement by four to five months, thereby reducing the overall settlement time by as much as 50 percent.

Government could streamline the yield estimation process. Delays in receipt of yield estimates from the states, is another key underlying reason for delays in claim settlement. It currently takes state governments around two months from the time of the raw CCE data collection to submit the crop yield estimates to AICI. Government could institute measures to minimize these delays possibly through use of technology (e.g. computerization and use of hand held devices) and better data consolidation practices.

Advanced indemnity payments even prior to harvest could be made based on weather and/or remote sensing indices. This approach will enable reaping the benefits drawn from combining the best features of both area-yield (e.g., more accurate loss estimates) and weather-based insurance (e.g., faster claim settlement). Since AICI, with the assistance of the World Bank, is currently developing actuarially-sound weather-based insurance products, the implementation of such a measure could be possible for the Kharif 2007 season.

Guaranteed yields

Guaranteed yields could be based on a long term average of at least 10 years, with a technology adjustment for yield trending. A longer time series than what is currently used, will add more stability

|42| India – NAIS: Market-based Solutions for Better Risk Sharing

in the insurance coverage. Such a measure will, possibly, bring more stability in the participation, as non-borrowing farmers would not be able to adversely select against the insurance program by participating when coverage is high or not insuring when coverage is low. It is suggested that AICI could consider developing a yield trending methodology, based on international best practices. However, till this is developed, during the transition phase, the guaranteed yield could be calculated with a ten-year moving average, with no yield trending adjustment.

Size of IU

If the Government wants to reduce the size of the IU, it is suggested to first develop a credible PY and ratemaking methodology for smaller units. However, this may take time and requires a means to index the smaller unit to a larger region for coverage stability. Delineation of the country or state by agro-climatic zones may be useful in reducing the basis risk of area-yield-based estimates but requires a significant investment in a data management system to move forward. If Government priorities lead to lowering the unit before actuarially-sound premium rates are developed, a useful approach may be to blend market-based insurance objectives and social objectives. AICI could retain claims assessed on an actuarially-sound basis at the existing IU level, while residual claims (i.e., the difference between claims reported at the smaller IU level and the claims reported at the current IU level) would be covered by the Government as a social benefit.

Operational deadlines

The Government could institute a purchase deadline for crop insurance in advance of the crop season, for both borrowing and non-borrowing farmers. For example for the Kharif season, it is suggested to move back the cut-off date from September to July. This early deadline for insurance purchase would reduce adverse selection among non-borrowing farmers. However, the reduced time for purchasing insurance could also result in a trade-off with the outreach

of the program. AICI could address this through improved communication and this might also get addressed through the likely increased use of the Kisan Credit Card.

The Government could also encourage an earlier insurance sign-up through premium discounts. AICI could introduce a premium discount within a certain period well in advance of the growing season. This discount could extend to state and central governments in order to encourage their early premium payment to match the farmers’ premium. On the other hand, because premium discounts must be offset with a load to an actuarially-sound premium rate, governments and farmers could instead be surcharged a premium penalty for late sign-up and late payment. At some point, sign-up to the crop insurance program would have to be curtailed even with a premium surcharge.

Additional benefits

The Government could extend coverage for planting and post-harvest risks. These benefits could be built into the basic crop insurance program or added as endorsements for additional premium. Including these types of benefits as endorsements provides some “individualization” within the area-yield based program design. However, this flexibility requires an administration system that is transparent and readily transfers information directly between the farmer and the insurance company. Building the benefit into the basic crop insurance program, with premium loading for demonstrated risk at an IU level, may be the best way to encourage widespread participation and effective exposure pooling.

On balance it may be better to postpone this benefit until a proper data management system is established. Adding additional enhancements to a scheme that is missing some of the basic support instruments may complicate the process to an extent that an actuarially-sustainable scheme and other key objectives are jeopardized over the long

Conclusions and Suggestions |43|

term in order to accomplish relatively smaller short term objectives. If GOI wants to offer these benefits before the development of a proper data management system, since these benefits would not be actuarially rated (for lack of data), the Government could consider them as social benefits and thus bear the associated costs.

Communication

An effective communication strategy is critical in implementing changes to the NAIS. All the proposed suggestions above to improve the NAIS could potentially induce significant changes in the terms and conditions of the crop insurance policies and improve the design of the NAIS. However, the success of the intended improvements in the program design of NAIS will only be achieved if an effective communication/promotion strategy targeted at farmers is appropriately implemented. Hence appropriate effort and resources will need to be channelised towards this by AICI.

(c) CCEs

The CCE process is technically sound, although further efforts are suggested so as to develop a reporting system that ensures timeliness, accuracy and consistency of yield estimates. Specific suggestions include: establish a national CCE procedures manual, ensure that yield losses that cannot be attributed to an insured peril (i.e., it is due to an “uninsurable” cause of loss) are not being recorded for insurance purposes although they can be for policy reasons; possibly consider forwarding raw CCE data results to AICI as the field work is completed to compile a central database that would allow verification of yield estimates on an ongoing basis.

Suggested Action Plan

A set of suggested actions is proposed based on the findings of the TA report which factors in the outcome of detailed meetings with AICI senior management. The short term actions for AICI’s consideration that could be initiated are:

Premium rating. AICI could produce a set of actuarially-sound insurance premium rates based on the suggested ratemaking procedure. This would require enhancing institutional capacity within AICI on actuarial techniques. AICI could also review the impact of new methodology on premium level between regions/crops.

Indemnity limits. Indemnity limits could be determined in the light of the suggested actuarially-sound premium rates to create a homogeneous set of premium rates. Should the Government want to implement district premium rates, indemnity limits could be adjusted at the IU level to ensure homogeneity.

On-account payment of claims. Early compensation up to 40–50 percent of likely claims, based on weather indices, could be released during the cropping season.

Insurance unit. The suggested ratemaking methodology has been developed at the existing IU level. Should the Government want to lower the IU, the reduction in the IU could be considered as a social benefit and thus the additional costs could be borne by the Government.

Guaranteed yield. A ten-year moving average, with no technology adjustment for yield trending, could be used to estimate the probable yield. Meanwhile, AICI could develop a yield trending methodology to incorporate into TY estimate procedures.

Purchase deadlines. Uniform seasonality discipline would be applicable for both borrowing and non-borrowing farmers.

Additional benefits. Extending coverage for sowing/planting and post-harvest risks could be postponed until a proper data management system and actuarial premium rates can be established. Should the Government want to implement these benefits earlier, they could be considered as a social benefit and thus their costs be borne by the Government.

|44| India – NAIS: Market-based Solutions for Better Risk Sharing

Experience database. The Government could consider steps to create a centralized data management system to accept incoming data for all states.

These suggested actions could be implemented by AICI, and would require the support of the Government of India (Ministry of Agriculture and the Insurance Division of the Ministry of Finance). The development of an experience database will also require close collaboration between the State Governments (Directorate of Statistics and Economics) and the Central

Government. Overall, these suggestions that are focused on design aspects would involve significant changes in the NAIS, but for these changes in the program design to yield full benefits and be really successful, issues related to the implementation of the program on the ground (including operational steps to improve the quality of CCEs and to implement the proposed ex-ante government contribution, etc.) would need to be addressed. Given the complexities involved in this process, it is suggested to pilot-test the proposed actions and, if successful, to expand them countrywide.

Bibliography |45|

Bibliography

AICI (Agriculture Insurance Company of India). “Manual on Crop Estimation Surveys on Food and Non-Food Crops in Andhra Pradesh.” New Delhi: Agriculture Insurance Company of India.

Government of India, Ministry of Agriculture. 2004. “Report of the Joint Group on Crop Insurance.” (Department of Agriculture & Cooperation). New Delhi: Government of India.

Watts & Associates, Inc., DYMAC Risk Management Solutions Ltd, KALA Risk Management Corp. 2006. “Report prepared for

the World Bank on Agriculture Insurance Technical Assistance”.

World Bank. 2004. “India: Scaling-up Access to Finance for India’s Rural Poor”. Washington D.C.: World Bank.

World Bank. 2003. “Karnataka Crop Insurance Study”. Washington D.C.: World Bank.

World Bank. 2005. “The Financing of Agricultural Production Risks: Revisiting the Role of Agriculture Insurance”. Washington D.C.: World Bank.

Glossary |47|

Glossary

Actuarially pure premium rate Premium rate worked out considering the frequency and severity of past events.Adverse selection Selective participation observed under voluntary crop insurance program:

increasing participation during potential adverse seasons and decreasing participation during potential normal seasons.

Annual crop A crop that generally has a life cycle of upto one year.Area-yield based insurance Insurance scheme under which insurance payments are based on an area-yield

estimate determined by harvest production measurements taken at a series of randomly chosen Crop Cutting Experiments locations.

Basis risk The incidence that the yield loss observed at the insurance unit does not exactly match an individual’s loss experience.

Block/mandal Administrative sub-division of the district, which in turn is a sub-division of the state.

Claim ratio Claims expressed as percentage of premia collected.Crop cutting experiment Sampling process by which crop yields are statistically estimated for each

insurance unit.Guaranteed yield Probable yield multiplied by indemnity limit.Indemnity level Limits, in percentage, applied on probable yield to produce threshold yield.

Indemnity limits available under NAIS are 60 percent, 80 percent and 90 percent.

Insurance unit Administrative level (e.g., Block, Tehsil) where the crop yields are estimated through the crop cutting experiment process.

Loaded premium rate Actuarially pure premium rate loaded for administrative costs, profit and contingency.

Loss adjuster A representative of the insurer or an independent person employed by the insurer to assess and determine the extent of the insurer’s liability for loss or damage claimed by the insurer.

Loss-cost Claims expressed as percentage of sum insured.Moral hazard Moral hazard arises with traditional insurance when insured parties alter their

behavior so as to increase the potential likelihood or magnitude of a lossProbable yield Moving average of seasonal area-yields.Sum insured Amount of risk coverage on which the premium is paid. It is also the maximum

value of the claim.Threshold yield See Guaranteed yield.

Annex |49|

Crop Risk Maps for Selected Crops and States

Annex 1Annex 1

Suggested premium rates (for a 90 percent coverage level), derived from the experience-based approach, are depicted onto the crop risk maps below.

Note: Tehsil premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure premium rate at 90 percent coverage level of wheat crop in UP

|50| India – NAIS: Market-based Solutions for Better Risk Sharing

Note: Mandal-level premium rate are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure premium rate at 90 percent coverage level of rice crop in Andhra Pradesh

Annex |51|

Note: Tehsil-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure premium rate at 90 percent coverage level of cotton crop in Gujarat

|52| India – NAIS: Market-based Solutions for Better Risk Sharing

Note: Tehsil-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure premium rate at 90 percent coverage level of groundnut crop in Gujarat

Annex |53|

Note: Tehsil-level premium rates are defined as the weighted average of IU (Circle) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure premium rate at 90 percent coverage level of pigeon pea crop in Maharashtra

|54| India – NAIS: Market-based Solutions for Better Risk Sharing

Note: Tehsil-level premium rates are defined as the weighted average of IU (Circle) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure premium rate at 90 percent coverage level of Rabi sorghum crop in Maharashtra

Annex |55|

Note: District-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure district premium rate at 90 percent coverage level of wheat crop in UP

District level map

|56| India – NAIS: Market-based Solutions for Better Risk Sharing

Note: District-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure district premium rate at 90 percent coverage level of rice crop in Andhra Pradesh

District level map

Annex |57|

Note: District-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure district premium rate at 90 percent coverage level of cotton crop in Gujarat

District level map

|58| India – NAIS: Market-based Solutions for Better Risk Sharing

Note: District-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure district premium rate at 90 percent coverage level of groundnut crop in Gujarat

District level map

Annex |59|

Note: District-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure district premium rate at 90 percent coverage level of pigeon pea crop in Maharashtra

District level map

|60| India – NAIS: Market-based Solutions for Better Risk Sharing

Note: District-level premium rates are defined as the weighted average of IU (Panchayat) premium rates. The average coverage amounts from 2002 to 2004, for each IU, are used as weights.

Source: Risk maps prepared by RMSI (Consultant) using AICI data and Survey of India maps.

Pure district premium rate at 90 percent coverage level of Rabi soghum crop in Maharashtra

District level map

THE WORLD BANKTHE WORLD BANK