Bayes混合模型在我国农业旱灾风险度量中的运用

Application of Bayes Mixture Model in Measuring Agricultural Drought Risk in China

旱灾一直是对我国农业经济生产威胁最大的灾害之一。因此,如何深度剖析旱灾风险特征,强化风险管理,科学、积极地进行防灾减灾来大幅度减轻灾害风险负担、弱化旱灾风险负面影响,有效助力经济平稳、快速发展一直是政府和学术界长期关注的焦点问题之一。本文借鉴金融风险度量的方法经验,用VaR作为旱灾风险度量工具。通过构建混合分布模型的方式来拟合旱灾损失率分布,并根据该模型估计VaR以刻画旱灾风险大小。混合分布模型由常规分布和GPD组成,其中,GPD用来刻画风险尾部特征。同时,该模型采用Bayes计算方法来回避GPD分布阈值选择和参数估计难题。结果表明:与Gamma-GPD混合分布相比,Norm-GPD混合分布模型在拟合农业旱灾损失率分布上表现更佳,其精度较高、VaR估计值更可信。Norm-GPD混合分布的VaR估计结果表明,我国在遭受10年一遇,20年一遇和100年一遇旱灾的情况下,农业旱灾损失率不超过9.45%, 11.03%和30.21%。该结果对于我国农业旱灾风险管理以及农业灾害保险、债券的定价具有重要意义。

Drought has always been one of the biggest threats to our country’s agricultural economy.Therefore, how to deeply analyze the characteristics of drought risks, strengthen risk management, scientifically and proactively carry out disaster prevention and reduction so as to drastically reduce the burden of disaster risk, weaken the negative impact of drought risks and effectively help the economy develop steadily and rapidly has been a long-term concern of government and academia government and academia. Based on the experience of financial risk measurement, VaR was used as a measure tool for drought risk measurement. The distribution of drought loss rate was fitted by constructing a mixture distribution model. Based on this model, the VaR was estimated to describe the risk of drought. The mixture distribution model consists of a regular distribution and a GPD, where the GPD was used to characterize the tail characteristics of the risk. At the same time, this model uses Bayes method to avoid the problem of GPD’s threshold selection and parameter estimation. The results show that compared with the Gamma-GPD mixture distribution, the Norm-GPD mixture distribution model has better performance in fitting the distribution of drought loss rate, its precision is higher and the VaR estimator is more believable. The result of VaR’s estimator of Norm-GPD mixture distribution shows that the loss rate of agricultural drought does not exceed 9.45%, 11.03% and 30.21% in our country in once in 10 years, once in 20 years, once in 100 years of drought cases. The result is of great significance to the management of agricultural drought risk and the pricing of agricultural disaster insurance and bonds.