ABSTRACT Forecasting volatility has been widely addressed in the fields of finance, environmetrics, and other areas involving massive time series. The important part of addressing this problem is how to specify the error term's distribution. With a weaker distribution assumption, we achieve greater model flexibility. In this paper, we present a flexible semiparametric Bayesian framework to address the problem of forecasting volatility in time series data by introducing the weighted Dirichlet process mixture (WDPM). We illustrate the advantages of WDPM using simulation data and stock return data.

中文翻译：

---

解决收益分布预测问题的灵活加权狄利克雷过程混合建模与评估

摘要 预测波动性已在金融、环境计量学和其他涉及大量时间序列的领域中得到广泛解决。解决这个问题的重要部分是如何指定误差项的分布。使用较弱的分布假设，我们实现了更大的模型灵活性。在本文中，我们提出了一个灵活的半参数贝叶斯框架，通过引入加权狄利克雷过程混合 (WDPM) 来解决预测时间序列数据波动性的问题。我们使用模拟数据和股票收益数据来说明 WDPM 的优势。