基于神经网络和随机波动模型的农产品期货价格预测

Agricultural Futures Price Forecasting Based on Neural Networks and Stochastic Volatility Models

摘要: 本文研究了棉花期货价格的预测问题，采用了BP神经网络模型和Heston随机波动模型两种模型并进行比较。首先使用BP神经网络模型对样本数据进行训练、验证和测试，并利用该模型对未来五天的棉花期货价格进行预测。其次，使用极大似然估计法估计Heston随机波动模型中的参数，并基于估计后的模型采用滚动时间窗法对样本外五天的棉花期货价格进行预测。最后，引入五种损失函数对两种方法的预测精度进行比较。实证结果表明，Heston随机波动模型的预测效果优于BP神经网络模型。

Abstract: This paper examines the problem of forecasting cotton futures prices, in which we utilize the BP neural network model and the Heston stochastic volatility model, and the prediction accuracy of the two models is compared. Firstly, this paper utilizes the BP neural network model to train, validate and test sample data. And then using this model predict the cotton futures price for five days outside the sample. After that, we use the maximum likelihood estimation method to estimate the parameters of the Heston stochastic volatility model. Based on the estimated model, the scrolling time window method was used to forecast cotton futures prices five days out of the sample. Finally, introducing five loss functions to compare the prediction accuracy of the two methods, the empirical results show that the Heston stochastic volatility model is better than the BP neural network model in prediction.