基于MAB的衍生品资产组合的风险度量研究

金融风险度量是衡量金融资产组合风险的重要工具。由于衍生品价格波动会给持有衍生品的金融机构和实体企业带来巨大损失，因此如何高效地估计衍生品资产组合风险度量是金融风险管理的一个前沿热点问题。因为资产组合常包含复杂衍生品，所以需要利用随机仿真来估计衍生品资产组合风险度量，但现有方法共性是将衍生品资产组合损失的估计与风险度量的估计割裂开来，造成估计的准确性差且速度较低。借助近年迅速发展的机器学习理论与方法，本项目申请将金融风险度量问题抽象为多臂老虎机（Multi-armed bandit，MAB）的理论问题，主要研究内容包括：（1）MAB的均值估计问题；（2）基于MAB的金融风险度量的估计；（3）金融风险度量的实证研究和应用研究。本项目旨在以衍生品资产组合风险度量为出发点，发展随机仿真、机器学习与金融风险管理等领域的交叉研究，建立高效的金融风险估计方法，为金融风险管理的实践提供理论方法与技术支持。

Financial risk measurement is an important tool for monitoring the portfolio risk. The derivatives price volatility may lead to large losses for entity enterprises and financial institutions who hold them. Therefore, how to effectively and efficiently measure the risk of derivative portfolios has been a frontier issue in financial risk management. Because portfolios usually include the complex financial derivatives, stochastic simulation is typically applied. The existing methods in the literature typically estimate the portfolio losses and portfolio risk separately, leading to inefficiency of the estimation. Based on the rapid development of machine learning theories and methods in recent years, we transform the portfolio risk measurement to problems based on multi-armed bandit (MAB). This project mainly contains three research problems: (1) the estimation of expectations in multi-armed bandit problem; (2) MAB-based method for financial risk measurement; (3) the empirical studies and applications of financial risk measurement. The project focuses on the risk measurement of derivative portfolios, and aims to develop the multidisciplinary research fields of stochastic simulation, machine learning and financial risk measurement, and then to establish widely applicable and highly efficient methods for estimating risk measures of derivative portfolios. The project will provide theoretical and technical support for the financial risk measurement in practice.