Stochastic Multi-factor Weather Model And Weather Derivatives Pricing Using Neural Network

The underlying weather indexes are not tradable and the weather derivatives market is an incomplete and illiquid market.Therefore,the classical Black-Scholes pricing approach cannot be directly applied to the pricing of weather derivatives.Through incorporating forward-looking information,we construct a consistent two-factor temperature model analyzed by neural networks for Zhengzhou and discusses the valuation of weather derivatives on the basis of this model.First,we use wavelet transform to analyze the historical temperature data of Zhengzhou and explain the rationality of the consistent two-factor temperature model;Then,according to the collected temperature data of Zhengzhou,we construct a consistent two-factor temperature model and use neural networks to optimize the model parameters.So the optimal consistent two-factor temperature model is obtained and the residual of this model is analyzed on this basis.We again use wavelet transform and neural networks to analyze the seasonal component of the residual and use the Irrelevant Connection Elimination scheme to eliminate irrelevant harmonics.We determine the harmonic terms contained in the residual.Finally,we study the option contract price based on the consistent two-factor temperature model and implement Monte Carlo simulation to calculate the price of two-dimensional curved surface of temperature options.And we randomly selected two sets of temperature data of Zhengzhou to simulate the option price based on consistent two-factor temperature model and the Ornstein-Uhlenbeck(O-U)model.This paper presents the process of weather derivatives pricing based on a consistent two-factor temperature model that is based on neural networks,which provides new ideas to study temperature model of Zhengzhou.