The case of agriculture crop insurance

Consider the challenges faced by a weather index insurance agency. It has to cover for an episode which has a high frequency but is completely uncertain, ensure the data is credible and minimises basis risks (affected farmer not eligible for claim because the index trigger was not hit), and pay out an amount which is commensurate to the damage suffered. It also has to keep the premiums affordable for the poor farmers, ensure that pay outs are done within a reasonable time, and genuine victims are not excluded. And it has to do all this with poor quality of index data and very patchy actuarial data, not to mention very unreliable crop damage models built on the index data.

A viable insurance model assumes reasonable premiums, diversified risk pool, and low frequency of insured episode incidence. But in case of crop insurance for the poor, the actuarial model has to support ultra-low premiums even with the constraints of high (and increasing) frequency of index triggers being hit, highly correlated insured pool (weather is the same over reasonably large areas or regions), and significant enough reimbursements required to make this meaningful enough for the farmers.

If we try doing the math, we will soon realise in no time that self-financed micro-insurance for the poor is an impossible proposition. In fact, it is no surprise that even the most efficient and largest crop-insurance scheme in the developing world enjoy over 80% premium subsidy support. India’s new massive nation-wide crop insurance program, Pradhan Mantri Fasal Bima Yojana (PMFBY), has premium subsidy of a whopping 97%! Even by squeezing out all the efficiency gains from financial engineering and technology, the commercial viability frontier will still remain very distant. In any case, whatever the insurer will pay out has to come from what is remaining after covering their costs – a claim ratio above 100% is not sustainable.

In simple terms, index insurance tries to do both financial engineering and weather modelling to address a complex development challenge. This is a double challenge. One, the actuarial models have to support affordable premiums. Two, the index data model that underpins the premium calculation is robust enough to minimise basis risks and ensure that the development objectives are met. The first suffers historical data deficiencies and the second is an emerging area of research fraught with deep uncertainties. 

In contrast, a direct payment to identified victims does not involve any of the risk mitigation and transaction costs associated with managing an insurance. However, it does involve the challenge of assessing damages and their validation, whereas an index insurance only requires more easily verifiable (though less directly linked to the desired outcome) index triggers. But it eliminates the significant basis risk faced by farmers and ensures that the desired development objectives are realised. Besides, it also captures the true cost of crop damage mitigation in a clean, direct, and efficient manner. 

In light of the above, the most efficient crop risk mitigation strategy would be direct income payments and not heavily subsidised insurance.

If we are engaging on a truly evidence-based policy making mode, the focus of innovation should be on helping governments make accurate assessments of crop damages in quick time. The one area where significant efficiency gains can be realised is from optimising the process of data collection and its validation. And this could be outsourced to a competent agency. But is anyone even talking about this?

Update 1 (19.12.2018)
Underlining the point made above, crop insurance subsidies are massive even in the US

The National Agriculture Insurance Scheme (NAIS) under implementation in India since 1999 and covering over 25 million farmers had an average claims multiple of over 3.5 times the premium payment for the 2000-08 period. See also this, this and this on crop insurance.

Update 2 (21.02.2020)

Good explainer about India's PMFBY crop insurance scheme here.