基于混合密度网络模型拟合收益统计分布与计算Expected Shortfall

Describing the Distribution of Returns and Calculating Expected Shortfall Based on Mixture Density Networks

基于混合密度网络模型估计金融时间序列的时变条件密件,提出数值模拟方法计算ExpectedShortfall。对香港恒生指数的实证研究表明,混合密度网络可以有效地描述收益的经验分布统计特征和波动规律,模型评估指标反映出预测效果良好,Value-at-R isk的预测精度在高端分位点表现较好,且可有效计算Expected shortfall指标,是金融市场风险测量的有效方法。

This paper applies mixture density networks （MDNs） to forecast the time-varying conditional density value of financial time series, and puts forth a new numerical algorithm to calculate Expected Shortfall, The application in Hong Kong Hangseng index approves that MDNs effectively describe the empirical distribution of returns and the volatility mechanism and have good forecasting ability. As a new model for financial market risk measure, MDNs can precisely calculate both Value-at-Risk at high probability levels and Expected Shortfall.