Magazine

Nov 2017

Read the latest edition of AIR and MEIR as an Interactive e-book

Embrace probabilistic risk models for the world's biggest agricultural producers

Source: Asia Insurance Review | Sep 2017

Agriculture

Agricultural risks are very complex and academia and the insurance industry need to work hand-in-hand to meet the challenges. Dr Laurent Marescot from RMS urges the insurance sector to look towards comprehensive probabilistic risk models to underwrite this class of insurance. 
 
 
China and India are the world’s largest agricultural nations in terms of output. For China, with 10% of the world’s total arable land, it produces food for 20% of the world’s population. For India, it has become the largest exporter of rice in the world in 2015, according to the Thai Rice Exporters Association.
 
   In terms of annual agriculture insurance premiums, these two nations generate roughly US$10 billion under heavily state-subsidised schemes, all with international private-sector (re)insurance involvement. 
 
   BBut despite this, agricultural insurance penetration, defined as the insurance premium as a percentage of agriculture GDP, is much less than 1%, compared to above 6% for the US. Thus, the agricultural protection gap is large and growing.
 
Need to adopt a new, more comprehensive and scientific underwriting approach
Considering the value and importance of these markets, it is surprising that most insurers still largely base their decisions on the intuition and judgement of experienced individuals backed with past insurance loss data, or past climate data sets, together with the input of some new technologies such as satellite imagery. 
 
   Adopting a new, more comprehensive and scientific underwriting approach will be essential to the growth and profitability of agricultural risk insurance markets in both countries. Traditional methods of risk assessment will begin to show their limitations, whether due to variations in trend data, limited insurance claims history or the ability to keep ahead of the changes in agriculture management practices and climate evolution.
 
Comprehensive probabilistic agricultural risk modelling solutions now available
The availability of comprehensive probabilistic agricultural risk modelling solutions for these huge producing countries is now increasing, though model development and adoption is linked to how models can manage the complex range of factors affecting agricultural risk in China and India. 
 
   The two nations have different market structures, as China’s insurance is mainly indemnity-based, while India runs index-based coverages. Both systems are in a near-constant state of flux and adjustments. 
 
Data availability is challenging though
The challenges with the availability of data to build probabilistic models also needs to be recognised, as data sets are often incomplete, inconsistent, and typically need de-trending to be relevant to the present day, as technology is changing with time (eg different types of crops, changes in health monitoring of livestock). 
 
   In addition, for model calibration, the correlation between crop yield and loss is not obvious as local negotiations concerning losses may skew results. These complexities underline that a “one-size-fits-all” modelling approach will not work.
 
Encourage insurance uptake in India
Agricultural insurance claims in India are based on yield and weather indices, and consider every stage of production and the seasons. 
 
   The new PMFBY (Pradhan Mantri Fasal Bima Yojana) agricultural insurance scheme was introduced by the Ministry of Agriculture and Farmers Welfare in 2016. The scheme covers prevented sowing and planting risk, standing crops, post-harvest losses, and “localised calamities”, and claims cover up to 90% of a crop’s value, based on yield reduction estimates from a predefined threshold. 
 
   The insurance plan is intended to encourage insurance uptake and close the agricultural protection gap for Indian farmers, who largely rely on rainfall to water their fields, and therefore suffer potentially dramatic consequences in years when monsoon rains are delayed. 
 
Modelling for a longer view of risk in India
As mentioned previously, risk management is traditionally based on past insurance loss experience, and past climate information, and consequently, the impact of potential severe events that may affect crops is under-represented using limited historical data sets. A complete agricultural probabilistic model can significantly extend the view in tail risk, for reinsurance purchase, as well as in terms of loss cost estimation for primary insurance underwriting. 
 
   Science has evolved dramatically in the past decade to enable the development of a range of crop models that take into account the impact of various threats for a large variety of crops, and at the various phases of growing stage. 
 
   In India, yield data exists since the late 1990s and can be used to build such a model, though after careful data cleansing. Climate metrics (eg minimum/maximum temperature, evapotranspiration, drought and excess rainfall, etc) and other static variables (irrigation, slope, soil water holding capacity) can be considered in the model as potentially explanatory variables of annual crop yield variations.
 
   In addition, the view into climate risk in India can be extended by developing a climate model, based on climate datasets (precipitation and temperature), to generate thousands of years of weather time-series. The crop yield models can then be run with the simulated weather data from the climate model to provide an extended and more comprehensive view of potential yield reduction in India. 
 
How models can help further
Probabilistic models also have the flexibility to be refined further. For example, as insurance contracts can be specified as “irrigated” or “rain-fed”, the simulated yields can be modified to reflect the impact on yield given the user’s assumptions. Stress testing, such as modelling specific scenarios for El Niño and La Niña years, can equally increase our understanding of the risk, ahead of the season.
 
   To illustrate how models can provide greater insight into loss distributions by extending beyond the observed loss record, consider in graph 1 how the Loss Cost (LC) estimated on observed data (11 years) changes when using modelled yield view based on an extended view on climate, eg 44 years. The observed yield is in red, modelled yield is in black.
 
Graph 1: Example of how a limited historical yield data set can bias Loss Cost (LC) estimates, for different indemnity levels (Rice Kharif, district level). LC values provided for an indemnity level of 0.8.
 
Increasing the resolution for analysing agricultural risk in China
China also has complexities in assessing agricultural risks. Like India, the historical record is limited, the perils multiple and the available data varies across the three lines of business to be modelled. 
 
   For example, crop risk in China may be affected by drought, flood, typhoon, frost, wind, and hail, with similar challenges around multiple perils for forestry and livestock lines. 
 
   In this case, as with India, probabilistic modelling can be applied to extend the view of risk. But to be meaningful, an agricultural model for China has to closely reflect the complex local market specificities. 
 
   For an international company willing to write agricultural risk in China, one challenge is the requirement to understand all of the relentless changes made to primary insurance policies. In crop, forestry, and livestock lines, insurance rates are set each year, changing the coverage landscape and hindering the process of informed decision-making. Models must ensure accurate coverage information is reflected and is constantly updated to reflect changes to terms and conditions.
 
   In addition, some international players may only have insurance information and exposure at province level, which is not granular enough for a meaningful agricultural risk assessment. Models can offer a solution, by proposing an exposure disaggregation down to county level for example, based on available planted area data (for crop), forest area or livestock types. 
 
   Understanding the exact users’ needs for a specific market when designing a modelling solution is perhaps even more critical for agricultural risk than it is when modelling more traditional property risks.
 
Embracing agricultural risk models
Risk modelling companies including RMS, have made great strides in developing innovative probabilistic agricultural models for India and China. 
 
   The products are designed to help insurers, reinsurers, and brokers better understand agricultural risk in the world’s largest farming nations. They are especially useful to the large international reinsurers who tend to lead such national risk programmes, helping them to optimise programme structures, premium levels, and capital allocations. In addition, they provide a means of sharing data in a consistent manner for better exposure at risk identification and accumulation that would lead to more effective use of underwriting techniques.
 
   They can support different kinds of decisions, such as the design of new products for new markets. Eventually, models will provide unique capabilities that would allow risk carriers and intermediaries to distil the insights they need to differentiate themselves from competitors in this large and complex market, which like others across the international commercial insurance sector, hovers at the margins of profitability. 
 
Insurers slow to follow but are catching up
As agricultural techniques evolve in China and India, and as state efforts to insure production develop, the insurance industry has sometimes been slow to follow on the modelling side, but that is changing. 
 
   As in other lines of business, acceptance of the need for reliable, credible probabilistic risk modelling is increasing, driven by severe unexpected events that take the industry by surprise. 
 
   Appreciation of the need for risk modelling in major agricultural markets is perhaps over 10 years behind that for property catastrophe modelling. Thus, even if agricultural insurance penetration in India and China multiplies – premium spending in China increased from about $100 million in 2006 to more than $6 billion in 2016 – model penetration remains low.  
 
   This too will change in time, as more insurers, reinsurers and brokers appreciate the value of extending their understanding of the factors affecting yields, and adopt the powerful tools now available to analyse the vast compendia of data to develop a unique view of risk. 
 
   But model credibility still remains to be built as well. As agricultural risk is very complex, academia and industry needs to work hand-in-hand to solve this challenge. Ultimately, in the world’s largest, most complex agricultural markets, embracing models is the only way to underwrite with both eyes open. A 
 
Dr Laurent Marescot is Senior Director, Model Product Management at RMS. 
 
Based in Zurich, Dr Marescot initially joined RMS in 2008 as part of the Zurich account management team, servicing the European (re)insurance and ILS market. He then moved to the model product management group, leading the product management group for all EMEA and APAC climatic perils, such as windstorm, typhoon, severe convective storm and flood.
 
| Print | Share
CAPTCHA image
Enter the code shown above in the box below.

Note that your comment may be edited or removed in the future, and that your comment may appear alongside the original article on websites other than this one.

 

Recent Comments

There are no comments submitted yet. Do you have an interesting opinion? Then be the first to post a comment.