Farmers can use weather derivatives to hedge against poor harvests caused by failing rains during the growing period, excessive rain during harvesting, high winds in case of plantations or temperature variabilities in case of greenhouse crops; theme parks may want to insure against rainy weekends during peak summer seasons; and gas and power companies may use heating degree days (HDD) or cooling degree days (CDD) contracts to smooth earnings.
Such an accumulation can be the basis for a derivative contract which might be structured as an option (call or put) or as a "swap" that is an agreement to pay or to receive payment.
The first weather derivative deal was in July 1996 when Aquila Energy structured a dual-commodity hedge for Consolidated Edison (ConEd).
As the market for these products grew, the Chicago Mercantile Exchange (CME) introduced the first exchange-traded weather futures contracts (and corresponding options), in 1999.
The CME currently lists weather derivative contracts for 24 cities in the United States, eleven in Europe, six in Canada, three in Australia and three in Japan.
The CME Hurricane Index, an innovation developed by the reinsurance industry provides contracts that are based on a formula derived from the wind speed and radius of named storms at the point of U.S. landfall.
In an Opalesque video interview, Nephila Capital's Barney Schauble described how some hedge funds treat weather derivatives as an investment class.
That is because the underlying asset of the weather derivative is non-tradeable which violates a number of key assumptions of the Black-Scholes Model.
Then the user can determine how much he/she is willing to pay in order to protect his/her business from those conditions in case they occurred based on his/her cost-benefit analysis and appetite for risk.
This approach requires building a model of the underlying index, i.e. the one upon which the derivative value is determined (for example monthly/seasonal cumulative heating degree days).
However, individual members of the ensemble need to be 'dressed' (for example, with Gaussian kernels estimated from historical performance) before a reasonable probabilistic forecast can be obtained.