Comparative Study Of Weather Derivative Temperature Prediction Models Based On Eight Regions In China

With the increasing influence of global warming and urban thermal effects on temperature,the economic losses caused by weather changes are also increasing,and the introduction of weather derivatives could be a good way to ameliorate such problems.A weather derivative is a financial instrument whose returns depend on the value of some underlying weather index.China is still in the preliminary stage of exploring the research and development of weather derivatives.In China’s weather market,most insurance contracts are used to avoid weather risks,but they are not applicable to non-catastrophic weather events.Therefore,as a new weather risk management tool,weather derivatives show great application value.The pricing accuracy of weather derivatives mainly depends on the prediction accuracy of climatic data,and China has meteorological monitoring stations in all provinces and cities,which provides strong data support for the introduction and research of weather derivatives.In the preparation of the corresponding temperature index and the pricing of derivatives,the high-precision temperature prediction model plays a particularly important role.In order to explore an effective and high-precision temperature prediction model,the main work and achievements of this paper are as follows:Based on the 42-year average daily temperature data of 360 meteorological stations in China,this paper combined rotated empirical orthogonal function and K-means to divide the temperature change in China,selected 8 representative meteorological stations,took the weather derivative contract promoted by Chicago Mercantile Exchange as an example,and adopted the parametric method and non-parametric method.Three temperature prediction models,discrete,continuous and temporal convolutional network,were established respectively.Through comparative study,it is found that the prediction effect of temporal convolutional network model,Monte Carlo simulation value of temperature index and reasonable pricing of options are significantly better than other models,and the model shows higher accuracy than the parameter estimation method,while compared with other non-parameter estimation methods,temporal convolutional network does not have gradient explosion or gradient disappearance and other problems.It is a simple and effective method of temperature prediction.The remarkable effectiveness of temporal convolutional network in temperature prediction provides a new way to price weather derivatives.In addition,weather derivatives also involve rainfall,wind energy and other aspects.According to the characteristics of different meteorological changes,reasonable models and methods are selected for research.Because the meteorological change is affected by many factors,in order to simulate the meteorological change more accurately,it should be combined with various related factors for comprehensive analysis.